File: | ufs/ffs/ffs_softdep.c |
Warning: | line 4529, column 7 Value stored to 'error' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | /* $OpenBSD: ffs_softdep.c,v 1.150 2021/04/28 09:53:53 claudio Exp $ */ |
2 | |
3 | /* |
4 | * Copyright 1998, 2000 Marshall Kirk McKusick. All Rights Reserved. |
5 | * |
6 | * The soft updates code is derived from the appendix of a University |
7 | * of Michigan technical report (Gregory R. Ganger and Yale N. Patt, |
8 | * "Soft Updates: A Solution to the Metadata Update Problem in File |
9 | * Systems", CSE-TR-254-95, August 1995). |
10 | * |
11 | * Further information about soft updates can be obtained from: |
12 | * |
13 | * Marshall Kirk McKusick http://www.mckusick.com/softdep/ |
14 | * 1614 Oxford Street mckusick@mckusick.com |
15 | * Berkeley, CA 94709-1608 +1-510-843-9542 |
16 | * USA |
17 | * |
18 | * Redistribution and use in source and binary forms, with or without |
19 | * modification, are permitted provided that the following conditions |
20 | * are met: |
21 | * |
22 | * 1. Redistributions of source code must retain the above copyright |
23 | * notice, this list of conditions and the following disclaimer. |
24 | * 2. Redistributions in binary form must reproduce the above copyright |
25 | * notice, this list of conditions and the following disclaimer in the |
26 | * documentation and/or other materials provided with the distribution. |
27 | * |
28 | * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY |
29 | * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
30 | * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
31 | * DISCLAIMED. IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR |
32 | * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
33 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
34 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
35 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
37 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
38 | * SUCH DAMAGE. |
39 | * |
40 | * from: @(#)ffs_softdep.c 9.59 (McKusick) 6/21/00 |
41 | * $FreeBSD: src/sys/ufs/ffs/ffs_softdep.c,v 1.86 2001/02/04 16:08:18 phk Exp $ |
42 | */ |
43 | |
44 | #include <sys/param.h> |
45 | #include <sys/buf.h> |
46 | #include <sys/kernel.h> |
47 | #include <sys/malloc.h> |
48 | #include <sys/mount.h> |
49 | #include <sys/proc.h> |
50 | #include <sys/pool.h> |
51 | #include <sys/syslog.h> |
52 | #include <sys/systm.h> |
53 | #include <sys/vnode.h> |
54 | #include <sys/specdev.h> |
55 | #include <crypto/siphash.h> |
56 | #include <ufs/ufs/dir.h> |
57 | #include <ufs/ufs/quota.h> |
58 | #include <ufs/ufs/inode.h> |
59 | #include <ufs/ufs/ufsmount.h> |
60 | #include <ufs/ffs/fs.h> |
61 | #include <ufs/ffs/softdep.h> |
62 | #include <ufs/ffs/ffs_extern.h> |
63 | #include <ufs/ufs/ufs_extern.h> |
64 | |
65 | #define STATIC |
66 | |
67 | /* |
68 | * Mapping of dependency structure types to malloc types. |
69 | */ |
70 | #define D_PAGEDEP0 0 |
71 | #define D_INODEDEP1 1 |
72 | #define D_NEWBLK2 2 |
73 | #define D_BMSAFEMAP3 3 |
74 | #define D_ALLOCDIRECT4 4 |
75 | #define D_INDIRDEP5 5 |
76 | #define D_ALLOCINDIR6 6 |
77 | #define D_FREEFRAG7 7 |
78 | #define D_FREEBLKS8 8 |
79 | #define D_FREEFILE9 9 |
80 | #define D_DIRADD10 10 |
81 | #define D_MKDIR11 11 |
82 | #define D_DIRREM12 12 |
83 | #define D_NEWDIRBLK13 13 |
84 | #define D_LAST13 13 |
85 | /* |
86 | * Names of softdep types. |
87 | */ |
88 | const char *softdep_typenames[] = { |
89 | "pagedep", |
90 | "inodedep", |
91 | "newblk", |
92 | "bmsafemap", |
93 | "allocdirect", |
94 | "indirdep", |
95 | "allocindir", |
96 | "freefrag", |
97 | "freeblks", |
98 | "freefile", |
99 | "diradd", |
100 | "mkdir", |
101 | "dirrem", |
102 | "newdirblk", |
103 | }; |
104 | #define TYPENAME(type)((unsigned)(type) <= 13 ? softdep_typenames[type] : "???") \ |
105 | ((unsigned)(type) <= D_LAST13 ? softdep_typenames[type] : "???") |
106 | /* |
107 | * Finding the current process. |
108 | */ |
109 | #define CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc |
110 | /* |
111 | * End system adaptation definitions. |
112 | */ |
113 | |
114 | /* |
115 | * Internal function prototypes. |
116 | */ |
117 | STATIC void softdep_error(char *, int); |
118 | STATIC void drain_output(struct vnode *, int); |
119 | STATIC int getdirtybuf(struct buf *, int); |
120 | STATIC void clear_remove(struct proc *); |
121 | STATIC void clear_inodedeps(struct proc *); |
122 | STATIC int flush_pagedep_deps(struct vnode *, struct mount *, |
123 | struct diraddhd *); |
124 | STATIC int flush_inodedep_deps(struct fs *, ufsino_t); |
125 | STATIC int handle_written_filepage(struct pagedep *, struct buf *); |
126 | STATIC void diradd_inode_written(struct diradd *, struct inodedep *); |
127 | STATIC int handle_written_inodeblock(struct inodedep *, struct buf *); |
128 | STATIC void handle_allocdirect_partdone(struct allocdirect *); |
129 | STATIC void handle_allocindir_partdone(struct allocindir *); |
130 | STATIC void initiate_write_filepage(struct pagedep *, struct buf *); |
131 | STATIC void handle_written_mkdir(struct mkdir *, int); |
132 | STATIC void initiate_write_inodeblock_ufs1(struct inodedep *, struct buf *); |
133 | #ifdef FFS21 |
134 | STATIC void initiate_write_inodeblock_ufs2(struct inodedep *, struct buf *); |
135 | #endif |
136 | STATIC void handle_workitem_freefile(struct freefile *); |
137 | STATIC void handle_workitem_remove(struct dirrem *); |
138 | STATIC struct dirrem *newdirrem(struct buf *, struct inode *, |
139 | struct inode *, int, struct dirrem **); |
140 | STATIC void free_diradd(struct diradd *); |
141 | STATIC void free_allocindir(struct allocindir *, struct inodedep *); |
142 | STATIC void free_newdirblk(struct newdirblk *); |
143 | STATIC int indir_trunc(struct inode *, daddr_t, int, daddr_t, long *); |
144 | STATIC void deallocate_dependencies(struct buf *, struct inodedep *); |
145 | STATIC void free_allocdirect(struct allocdirectlst *, |
146 | struct allocdirect *, int); |
147 | STATIC int check_inode_unwritten(struct inodedep *); |
148 | STATIC int free_inodedep(struct inodedep *); |
149 | STATIC void handle_workitem_freeblocks(struct freeblks *); |
150 | STATIC void merge_inode_lists(struct inodedep *); |
151 | STATIC void setup_allocindir_phase2(struct buf *, struct inode *, |
152 | struct allocindir *); |
153 | STATIC struct allocindir *newallocindir(struct inode *, int, daddr_t, |
154 | daddr_t); |
155 | STATIC void handle_workitem_freefrag(struct freefrag *); |
156 | STATIC struct freefrag *newfreefrag(struct inode *, daddr_t, long); |
157 | STATIC void allocdirect_merge(struct allocdirectlst *, |
158 | struct allocdirect *, struct allocdirect *); |
159 | STATIC struct bmsafemap *bmsafemap_lookup(struct buf *); |
160 | STATIC int newblk_lookup(struct fs *, daddr_t, int, |
161 | struct newblk **); |
162 | STATIC int inodedep_lookup(struct fs *, ufsino_t, int, struct inodedep **); |
163 | STATIC int pagedep_lookup(struct inode *, daddr_t, int, struct pagedep **); |
164 | STATIC void pause_timer(void *); |
165 | STATIC int request_cleanup(int, int); |
166 | STATIC int process_worklist_item(struct mount *, int *, int); |
167 | STATIC void add_to_worklist(struct worklist *); |
168 | |
169 | /* |
170 | * Exported softdep operations. |
171 | */ |
172 | void softdep_disk_io_initiation(struct buf *); |
173 | void softdep_disk_write_complete(struct buf *); |
174 | void softdep_deallocate_dependencies(struct buf *); |
175 | void softdep_move_dependencies(struct buf *, struct buf *); |
176 | int softdep_count_dependencies(struct buf *bp, int, int); |
177 | |
178 | /* |
179 | * Locking primitives. |
180 | * |
181 | * For a uniprocessor, all we need to do is protect against disk |
182 | * interrupts. For a multiprocessor, this lock would have to be |
183 | * a mutex. A single mutex is used throughout this file, though |
184 | * finer grain locking could be used if contention warranted it. |
185 | * |
186 | * For a multiprocessor, the sleep call would accept a lock and |
187 | * release it after the sleep processing was complete. In a uniprocessor |
188 | * implementation there is no such interlock, so we simple mark |
189 | * the places where it needs to be done with the `interlocked' form |
190 | * of the lock calls. Since the uniprocessor sleep already interlocks |
191 | * the spl, there is nothing that really needs to be done. |
192 | */ |
193 | #ifndef /* NOT */ DEBUG |
194 | STATIC struct lockit { |
195 | int lkt_spl; |
196 | } lk = { 0 }; |
197 | #define ACQUIRE_LOCK(lk)(lk)->lkt_spl = splraise(0x6) (lk)->lkt_spl = splbio()splraise(0x6) |
198 | #define FREE_LOCK(lk)spllower((lk)->lkt_spl) splx((lk)->lkt_spl)spllower((lk)->lkt_spl) |
199 | #define ACQUIRE_LOCK_INTERLOCKED(lk,s)(lk)->lkt_spl = (s) (lk)->lkt_spl = (s) |
200 | #define FREE_LOCK_INTERLOCKED(lk)((lk)->lkt_spl) ((lk)->lkt_spl) |
201 | |
202 | #else /* DEBUG */ |
203 | STATIC struct lockit { |
204 | int lkt_spl; |
205 | pid_t lkt_held; |
206 | int lkt_line; |
207 | } lk = { 0, -1 }; |
208 | STATIC int lockcnt; |
209 | |
210 | STATIC void acquire_lock(struct lockit *, int); |
211 | STATIC void free_lock(struct lockit *, int); |
212 | STATIC void acquire_lock_interlocked(struct lockit *, int, int); |
213 | STATIC int free_lock_interlocked(struct lockit *, int); |
214 | |
215 | #define ACQUIRE_LOCK(lk)(lk)->lkt_spl = splraise(0x6) acquire_lock(lk, __LINE__215) |
216 | #define FREE_LOCK(lk)spllower((lk)->lkt_spl) free_lock(lk, __LINE__216) |
217 | #define ACQUIRE_LOCK_INTERLOCKED(lk,s)(lk)->lkt_spl = (s) acquire_lock_interlocked(lk, (s), __LINE__217) |
218 | #define FREE_LOCK_INTERLOCKED(lk)((lk)->lkt_spl) free_lock_interlocked(lk, __LINE__218) |
219 | |
220 | STATIC void |
221 | acquire_lock(struct lockit *lk, int line) |
222 | { |
223 | pid_t holder; |
224 | int original_line; |
225 | |
226 | if (lk->lkt_held != -1) { |
227 | holder = lk->lkt_held; |
228 | original_line = lk->lkt_line; |
229 | FREE_LOCK(lk)spllower((lk)->lkt_spl); |
230 | if (holder == CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_tid) |
231 | panic("softdep_lock: locking against myself, acquired at line %d, relocked at line %d", original_line, line); |
232 | else |
233 | panic("softdep_lock: lock held by %d, acquired at line %d, relocked at line %d", holder, original_line, line); |
234 | } |
235 | lk->lkt_spl = splbio()splraise(0x6); |
236 | lk->lkt_held = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_tid; |
237 | lk->lkt_line = line; |
238 | lockcnt++; |
239 | } |
240 | |
241 | STATIC void |
242 | free_lock(struct lockit *lk, int line) |
243 | { |
244 | |
245 | if (lk->lkt_held == -1) |
246 | panic("softdep_unlock: lock not held at line %d", line); |
247 | lk->lkt_held = -1; |
248 | splx(lk->lkt_spl)spllower(lk->lkt_spl); |
249 | } |
250 | |
251 | STATIC void |
252 | acquire_lock_interlocked(struct lockit *lk, int s, int line) |
253 | { |
254 | pid_t holder; |
255 | int original_line; |
256 | |
257 | if (lk->lkt_held != -1) { |
258 | holder = lk->lkt_held; |
259 | original_line = lk->lkt_line; |
260 | FREE_LOCK_INTERLOCKED(lk)((lk)->lkt_spl); |
261 | if (holder == CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_tid) |
262 | panic("softdep_lock: locking against myself, acquired at line %d, relocked at line %d", original_line, line); |
263 | else |
264 | panic("softdep_lock: lock held by %d, acquired at line %d, relocked at line %d", holder, original_line, line); |
265 | } |
266 | lk->lkt_held = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_tid; |
267 | lk->lkt_line = line; |
268 | lk->lkt_spl = s; |
269 | lockcnt++; |
270 | } |
271 | |
272 | STATIC int |
273 | free_lock_interlocked(struct lockit *lk, int line) |
274 | { |
275 | |
276 | if (lk->lkt_held == -1) |
277 | panic("softdep_unlock_interlocked: lock not held at line %d", line); |
278 | lk->lkt_held = -1; |
279 | |
280 | return (lk->lkt_spl); |
281 | } |
282 | #endif /* DEBUG */ |
283 | |
284 | /* |
285 | * Place holder for real semaphores. |
286 | */ |
287 | struct sema { |
288 | int value; |
289 | pid_t holder; |
290 | char *name; |
291 | int prio; |
292 | }; |
293 | STATIC void sema_init(struct sema *, char *, int); |
294 | STATIC int sema_get(struct sema *, struct lockit *); |
295 | STATIC void sema_release(struct sema *); |
296 | |
297 | STATIC void |
298 | sema_init(struct sema *semap, char *name, int prio) |
299 | { |
300 | |
301 | semap->holder = -1; |
302 | semap->value = 0; |
303 | semap->name = name; |
304 | semap->prio = prio; |
305 | } |
306 | |
307 | STATIC int |
308 | sema_get(struct sema *semap, struct lockit *interlock) |
309 | { |
310 | int s; |
311 | |
312 | if (semap->value++ > 0) { |
313 | if (interlock != NULL((void *)0)) |
314 | s = FREE_LOCK_INTERLOCKED(interlock)((interlock)->lkt_spl); |
315 | tsleep_nsec(semap, semap->prio, semap->name, INFSLP0xffffffffffffffffULL); |
316 | if (interlock != NULL((void *)0)) { |
317 | ACQUIRE_LOCK_INTERLOCKED(interlock, s)(interlock)->lkt_spl = (s); |
318 | FREE_LOCK(interlock)spllower((interlock)->lkt_spl); |
319 | } |
320 | return (0); |
321 | } |
322 | semap->holder = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_tid; |
323 | if (interlock != NULL((void *)0)) |
324 | FREE_LOCK(interlock)spllower((interlock)->lkt_spl); |
325 | return (1); |
326 | } |
327 | |
328 | STATIC void |
329 | sema_release(struct sema *semap) |
330 | { |
331 | |
332 | if (semap->value <= 0 || semap->holder != CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_tid) { |
333 | #ifdef DEBUG |
334 | if (lk.lkt_held != -1) |
335 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
336 | #endif |
337 | panic("sema_release: not held"); |
338 | } |
339 | if (--semap->value > 0) { |
340 | semap->value = 0; |
341 | wakeup(semap); |
342 | } |
343 | semap->holder = -1; |
344 | } |
345 | |
346 | /* |
347 | * Memory management. |
348 | */ |
349 | STATIC struct pool pagedep_pool; |
350 | STATIC struct pool inodedep_pool; |
351 | STATIC struct pool newblk_pool; |
352 | STATIC struct pool bmsafemap_pool; |
353 | STATIC struct pool allocdirect_pool; |
354 | STATIC struct pool indirdep_pool; |
355 | STATIC struct pool allocindir_pool; |
356 | STATIC struct pool freefrag_pool; |
357 | STATIC struct pool freeblks_pool; |
358 | STATIC struct pool freefile_pool; |
359 | STATIC struct pool diradd_pool; |
360 | STATIC struct pool mkdir_pool; |
361 | STATIC struct pool dirrem_pool; |
362 | STATIC struct pool newdirblk_pool; |
363 | |
364 | static __inline void |
365 | softdep_free(struct worklist *item, int type) |
366 | { |
367 | |
368 | switch (type) { |
369 | case D_PAGEDEP0: |
370 | pool_put(&pagedep_pool, item); |
371 | break; |
372 | |
373 | case D_INODEDEP1: |
374 | pool_put(&inodedep_pool, item); |
375 | break; |
376 | |
377 | case D_BMSAFEMAP3: |
378 | pool_put(&bmsafemap_pool, item); |
379 | break; |
380 | |
381 | case D_ALLOCDIRECT4: |
382 | pool_put(&allocdirect_pool, item); |
383 | break; |
384 | |
385 | case D_INDIRDEP5: |
386 | pool_put(&indirdep_pool, item); |
387 | break; |
388 | |
389 | case D_ALLOCINDIR6: |
390 | pool_put(&allocindir_pool, item); |
391 | break; |
392 | |
393 | case D_FREEFRAG7: |
394 | pool_put(&freefrag_pool, item); |
395 | break; |
396 | |
397 | case D_FREEBLKS8: |
398 | pool_put(&freeblks_pool, item); |
399 | break; |
400 | |
401 | case D_FREEFILE9: |
402 | pool_put(&freefile_pool, item); |
403 | break; |
404 | |
405 | case D_DIRADD10: |
406 | pool_put(&diradd_pool, item); |
407 | break; |
408 | |
409 | case D_MKDIR11: |
410 | pool_put(&mkdir_pool, item); |
411 | break; |
412 | |
413 | case D_DIRREM12: |
414 | pool_put(&dirrem_pool, item); |
415 | break; |
416 | |
417 | case D_NEWDIRBLK13: |
418 | pool_put(&newdirblk_pool, item); |
419 | break; |
420 | |
421 | default: |
422 | #ifdef DEBUG |
423 | if (lk.lkt_held != -1) |
424 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
425 | #endif |
426 | panic("softdep_free: unknown type %d", type); |
427 | } |
428 | } |
429 | |
430 | struct workhead softdep_freequeue; |
431 | |
432 | static __inline void |
433 | softdep_freequeue_add(struct worklist *item) |
434 | { |
435 | int s; |
436 | |
437 | s = splbio()splraise(0x6); |
438 | LIST_INSERT_HEAD(&softdep_freequeue, item, wk_list)do { if (((item)->wk_list.le_next = (&softdep_freequeue )->lh_first) != ((void *)0)) (&softdep_freequeue)-> lh_first->wk_list.le_prev = &(item)->wk_list.le_next ; (&softdep_freequeue)->lh_first = (item); (item)-> wk_list.le_prev = &(&softdep_freequeue)->lh_first; } while (0); |
439 | splx(s)spllower(s); |
440 | } |
441 | |
442 | static __inline void |
443 | softdep_freequeue_process(void) |
444 | { |
445 | struct worklist *wk; |
446 | |
447 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
448 | |
449 | while ((wk = LIST_FIRST(&softdep_freequeue)((&softdep_freequeue)->lh_first)) != NULL((void *)0)) { |
450 | LIST_REMOVE(wk, wk_list)do { if ((wk)->wk_list.le_next != ((void *)0)) (wk)->wk_list .le_next->wk_list.le_prev = (wk)->wk_list.le_prev; *(wk )->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list .le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0); |
451 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
452 | softdep_free(wk, wk->wk_type); |
453 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
454 | } |
455 | } |
456 | |
457 | /* |
458 | * Worklist queue management. |
459 | * These routines require that the lock be held. |
460 | */ |
461 | #ifndef /* NOT */ DEBUG |
462 | #define WORKLIST_INSERT(head, item)do { (item)->wk_state |= 0x8000; do { if (((item)->wk_list .le_next = (head)->lh_first) != ((void *)0)) (head)->lh_first ->wk_list.le_prev = &(item)->wk_list.le_next; (head )->lh_first = (item); (item)->wk_list.le_prev = &(head )->lh_first; } while (0); } while (0) do { \ |
463 | (item)->wk_state |= ONWORKLIST0x8000; \ |
464 | LIST_INSERT_HEAD(head, item, wk_list)do { if (((item)->wk_list.le_next = (head)->lh_first) != ((void *)0)) (head)->lh_first->wk_list.le_prev = & (item)->wk_list.le_next; (head)->lh_first = (item); (item )->wk_list.le_prev = &(head)->lh_first; } while (0); \ |
465 | } while (0) |
466 | #define WORKLIST_REMOVE(item)do { (item)->wk_state &= ~0x8000; do { if ((item)-> wk_list.le_next != ((void *)0)) (item)->wk_list.le_next-> wk_list.le_prev = (item)->wk_list.le_prev; *(item)->wk_list .le_prev = (item)->wk_list.le_next; ((item)->wk_list.le_prev ) = ((void *)-1); ((item)->wk_list.le_next) = ((void *)-1) ; } while (0); } while (0) do { \ |
467 | (item)->wk_state &= ~ONWORKLIST0x8000; \ |
468 | LIST_REMOVE(item, wk_list)do { if ((item)->wk_list.le_next != ((void *)0)) (item)-> wk_list.le_next->wk_list.le_prev = (item)->wk_list.le_prev ; *(item)->wk_list.le_prev = (item)->wk_list.le_next; ( (item)->wk_list.le_prev) = ((void *)-1); ((item)->wk_list .le_next) = ((void *)-1); } while (0); \ |
469 | } while (0) |
470 | #define WORKITEM_FREE(item, type)softdep_freequeue_add((struct worklist *)item) softdep_freequeue_add((struct worklist *)item) |
471 | |
472 | #else /* DEBUG */ |
473 | STATIC void worklist_insert(struct workhead *, struct worklist *); |
474 | STATIC void worklist_remove(struct worklist *); |
475 | STATIC void workitem_free(struct worklist *); |
476 | |
477 | #define WORKLIST_INSERT(head, item)do { (item)->wk_state |= 0x8000; do { if (((item)->wk_list .le_next = (head)->lh_first) != ((void *)0)) (head)->lh_first ->wk_list.le_prev = &(item)->wk_list.le_next; (head )->lh_first = (item); (item)->wk_list.le_prev = &(head )->lh_first; } while (0); } while (0) worklist_insert(head, item) |
478 | #define WORKLIST_REMOVE(item)do { (item)->wk_state &= ~0x8000; do { if ((item)-> wk_list.le_next != ((void *)0)) (item)->wk_list.le_next-> wk_list.le_prev = (item)->wk_list.le_prev; *(item)->wk_list .le_prev = (item)->wk_list.le_next; ((item)->wk_list.le_prev ) = ((void *)-1); ((item)->wk_list.le_next) = ((void *)-1) ; } while (0); } while (0) worklist_remove(item) |
479 | #define WORKITEM_FREE(item, type)softdep_freequeue_add((struct worklist *)item) workitem_free((struct worklist *)item) |
480 | |
481 | STATIC void |
482 | worklist_insert(struct workhead *head, struct worklist *item) |
483 | { |
484 | |
485 | if (lk.lkt_held == -1) |
486 | panic("worklist_insert: lock not held"); |
487 | if (item->wk_state & ONWORKLIST0x8000) { |
488 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
489 | panic("worklist_insert: already on list"); |
490 | } |
491 | item->wk_state |= ONWORKLIST0x8000; |
492 | LIST_INSERT_HEAD(head, item, wk_list)do { if (((item)->wk_list.le_next = (head)->lh_first) != ((void *)0)) (head)->lh_first->wk_list.le_prev = & (item)->wk_list.le_next; (head)->lh_first = (item); (item )->wk_list.le_prev = &(head)->lh_first; } while (0); |
493 | } |
494 | |
495 | STATIC void |
496 | worklist_remove(struct worklist *item) |
497 | { |
498 | |
499 | if (lk.lkt_held == -1) |
500 | panic("worklist_remove: lock not held"); |
501 | if ((item->wk_state & ONWORKLIST0x8000) == 0) { |
502 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
503 | panic("worklist_remove: not on list"); |
504 | } |
505 | item->wk_state &= ~ONWORKLIST0x8000; |
506 | LIST_REMOVE(item, wk_list)do { if ((item)->wk_list.le_next != ((void *)0)) (item)-> wk_list.le_next->wk_list.le_prev = (item)->wk_list.le_prev ; *(item)->wk_list.le_prev = (item)->wk_list.le_next; ( (item)->wk_list.le_prev) = ((void *)-1); ((item)->wk_list .le_next) = ((void *)-1); } while (0); |
507 | } |
508 | |
509 | STATIC void |
510 | workitem_free(struct worklist *item) |
511 | { |
512 | |
513 | if (item->wk_state & ONWORKLIST0x8000) { |
514 | if (lk.lkt_held != -1) |
515 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
516 | panic("workitem_free: still on list"); |
517 | } |
518 | softdep_freequeue_add(item); |
519 | } |
520 | #endif /* DEBUG */ |
521 | |
522 | /* |
523 | * Workitem queue management |
524 | */ |
525 | STATIC struct workhead softdep_workitem_pending; |
526 | STATIC struct worklist *worklist_tail; |
527 | STATIC int num_on_worklist; /* number of worklist items to be processed */ |
528 | STATIC int softdep_worklist_busy; /* 1 => trying to do unmount */ |
529 | STATIC int softdep_worklist_req; /* serialized waiters */ |
530 | STATIC int max_softdeps; /* maximum number of structs before slowdown */ |
531 | STATIC int tickdelay = 2; /* number of ticks to pause during slowdown */ |
532 | STATIC int proc_waiting; /* tracks whether we have a timeout posted */ |
533 | STATIC int *stat_countp; /* statistic to count in proc_waiting timeout */ |
534 | STATIC struct timeout proc_waiting_timeout; |
535 | STATIC struct proc *filesys_syncer; /* proc of filesystem syncer process */ |
536 | STATIC int req_clear_inodedeps; /* syncer process flush some inodedeps */ |
537 | #define FLUSH_INODES1 1 |
538 | STATIC int req_clear_remove; /* syncer process flush some freeblks */ |
539 | #define FLUSH_REMOVE2 2 |
540 | /* |
541 | * runtime statistics |
542 | */ |
543 | STATIC int stat_worklist_push; /* number of worklist cleanups */ |
544 | STATIC int stat_blk_limit_push; /* number of times block limit neared */ |
545 | STATIC int stat_ino_limit_push; /* number of times inode limit neared */ |
546 | STATIC int stat_blk_limit_hit; /* number of times block slowdown imposed */ |
547 | STATIC int stat_ino_limit_hit; /* number of times inode slowdown imposed */ |
548 | STATIC int stat_sync_limit_hit; /* number of synchronous slowdowns imposed */ |
549 | STATIC int stat_indir_blk_ptrs; /* bufs redirtied as indir ptrs not written */ |
550 | STATIC int stat_inode_bitmap; /* bufs redirtied as inode bitmap not written */ |
551 | STATIC int stat_direct_blk_ptrs;/* bufs redirtied as direct ptrs not written */ |
552 | STATIC int stat_dir_entry; /* bufs redirtied as dir entry cannot write */ |
553 | |
554 | /* |
555 | * Add an item to the end of the work queue. |
556 | * This routine requires that the lock be held. |
557 | * This is the only routine that adds items to the list. |
558 | * The following routine is the only one that removes items |
559 | * and does so in order from first to last. |
560 | */ |
561 | STATIC void |
562 | add_to_worklist(struct worklist *wk) |
563 | { |
564 | |
565 | if (wk->wk_state & ONWORKLIST0x8000) { |
566 | #ifdef DEBUG |
567 | if (lk.lkt_held != -1) |
568 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
569 | #endif |
570 | panic("add_to_worklist: already on list"); |
571 | } |
572 | wk->wk_state |= ONWORKLIST0x8000; |
573 | if (LIST_FIRST(&softdep_workitem_pending)((&softdep_workitem_pending)->lh_first) == NULL((void *)0)) |
574 | LIST_INSERT_HEAD(&softdep_workitem_pending, wk, wk_list)do { if (((wk)->wk_list.le_next = (&softdep_workitem_pending )->lh_first) != ((void *)0)) (&softdep_workitem_pending )->lh_first->wk_list.le_prev = &(wk)->wk_list.le_next ; (&softdep_workitem_pending)->lh_first = (wk); (wk)-> wk_list.le_prev = &(&softdep_workitem_pending)->lh_first ; } while (0); |
575 | else |
576 | LIST_INSERT_AFTER(worklist_tail, wk, wk_list)do { if (((wk)->wk_list.le_next = (worklist_tail)->wk_list .le_next) != ((void *)0)) (worklist_tail)->wk_list.le_next ->wk_list.le_prev = &(wk)->wk_list.le_next; (worklist_tail )->wk_list.le_next = (wk); (wk)->wk_list.le_prev = & (worklist_tail)->wk_list.le_next; } while (0); |
577 | worklist_tail = wk; |
578 | num_on_worklist += 1; |
579 | } |
580 | |
581 | /* |
582 | * Process that runs once per second to handle items in the background queue. |
583 | * |
584 | * Note that we ensure that everything is done in the order in which they |
585 | * appear in the queue. The code below depends on this property to ensure |
586 | * that blocks of a file are freed before the inode itself is freed. This |
587 | * ordering ensures that no new <vfsid, inum, lbn> triples will be generated |
588 | * until all the old ones have been purged from the dependency lists. |
589 | */ |
590 | int |
591 | softdep_process_worklist(struct mount *matchmnt) |
592 | { |
593 | struct proc *p = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc; |
594 | int matchcnt, loopcount; |
595 | struct timeval starttime; |
596 | |
597 | /* |
598 | * First process any items on the delayed-free queue. |
599 | */ |
600 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
601 | softdep_freequeue_process(); |
602 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
603 | |
604 | /* |
605 | * Record the process identifier of our caller so that we can give |
606 | * this process preferential treatment in request_cleanup below. |
607 | * We can't do this in softdep_initialize, because the syncer doesn't |
608 | * have to run then. |
609 | * NOTE! This function _could_ be called with a curproc != syncerproc. |
610 | */ |
611 | filesys_syncer = syncerproc; |
612 | matchcnt = 0; |
613 | |
614 | /* |
615 | * There is no danger of having multiple processes run this |
616 | * code, but we have to single-thread it when softdep_flushfiles() |
617 | * is in operation to get an accurate count of the number of items |
618 | * related to its mount point that are in the list. |
619 | */ |
620 | if (matchmnt == NULL((void *)0)) { |
621 | if (softdep_worklist_busy < 0) |
622 | return(-1); |
623 | softdep_worklist_busy += 1; |
624 | } |
625 | |
626 | /* |
627 | * If requested, try removing inode or removal dependencies. |
628 | */ |
629 | if (req_clear_inodedeps) { |
630 | clear_inodedeps(p); |
631 | req_clear_inodedeps -= 1; |
632 | wakeup_one(&proc_waiting)wakeup_n((&proc_waiting), 1); |
633 | } |
634 | if (req_clear_remove) { |
635 | clear_remove(p); |
636 | req_clear_remove -= 1; |
637 | wakeup_one(&proc_waiting)wakeup_n((&proc_waiting), 1); |
638 | } |
639 | loopcount = 1; |
640 | getmicrouptime(&starttime); |
641 | while (num_on_worklist > 0) { |
642 | if (process_worklist_item(matchmnt, &matchcnt, LK_NOWAIT0x0040UL) == 0) |
643 | break; |
644 | |
645 | /* |
646 | * If a umount operation wants to run the worklist |
647 | * accurately, abort. |
648 | */ |
649 | if (softdep_worklist_req && matchmnt == NULL((void *)0)) { |
650 | matchcnt = -1; |
651 | break; |
652 | } |
653 | |
654 | /* |
655 | * If requested, try removing inode or removal dependencies. |
656 | */ |
657 | if (req_clear_inodedeps) { |
658 | clear_inodedeps(p); |
659 | req_clear_inodedeps -= 1; |
660 | wakeup_one(&proc_waiting)wakeup_n((&proc_waiting), 1); |
661 | } |
662 | if (req_clear_remove) { |
663 | clear_remove(p); |
664 | req_clear_remove -= 1; |
665 | wakeup_one(&proc_waiting)wakeup_n((&proc_waiting), 1); |
666 | } |
667 | /* |
668 | * We do not generally want to stop for buffer space, but if |
669 | * we are really being a buffer hog, we will stop and wait. |
670 | */ |
671 | #if 0 |
672 | if (loopcount++ % 128 == 0) |
673 | bwillwrite(); |
674 | #endif |
675 | /* |
676 | * Never allow processing to run for more than one |
677 | * second. Otherwise the other syncer tasks may get |
678 | * excessively backlogged. |
679 | */ |
680 | { |
681 | struct timeval diff; |
682 | struct timeval tv; |
683 | |
684 | getmicrouptime(&tv); |
685 | timersub(&tv, &starttime, &diff)do { (&diff)->tv_sec = (&tv)->tv_sec - (&starttime )->tv_sec; (&diff)->tv_usec = (&tv)->tv_usec - (&starttime)->tv_usec; if ((&diff)->tv_usec < 0) { (&diff)->tv_sec--; (&diff)->tv_usec += 1000000 ; } } while (0); |
686 | if (diff.tv_sec != 0 && matchmnt == NULL((void *)0)) { |
687 | matchcnt = -1; |
688 | break; |
689 | } |
690 | } |
691 | |
692 | /* |
693 | * Process any new items on the delayed-free queue. |
694 | */ |
695 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
696 | softdep_freequeue_process(); |
697 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
698 | } |
699 | if (matchmnt == NULL((void *)0)) { |
700 | softdep_worklist_busy -= 1; |
701 | if (softdep_worklist_req && softdep_worklist_busy == 0) |
702 | wakeup(&softdep_worklist_req); |
703 | } |
704 | return (matchcnt); |
705 | } |
706 | |
707 | /* |
708 | * Process one item on the worklist. |
709 | */ |
710 | STATIC int |
711 | process_worklist_item(struct mount *matchmnt, int *matchcnt, int flags) |
712 | { |
713 | struct worklist *wk, *wkend; |
714 | struct dirrem *dirrem; |
715 | struct mount *mp; |
716 | struct vnode *vp; |
717 | |
718 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
719 | /* |
720 | * Normally we just process each item on the worklist in order. |
721 | * However, if we are in a situation where we cannot lock any |
722 | * inodes, we have to skip over any dirrem requests whose |
723 | * vnodes are resident and locked. |
724 | */ |
725 | LIST_FOREACH(wk, &softdep_workitem_pending, wk_list)for((wk) = ((&softdep_workitem_pending)->lh_first); (wk )!= ((void *)0); (wk) = ((wk)->wk_list.le_next)) { |
726 | if ((flags & LK_NOWAIT0x0040UL) == 0 || wk->wk_type != D_DIRREM12) |
727 | break; |
728 | dirrem = WK_DIRREM(wk)((struct dirrem *)(wk)); |
729 | vp = ufs_ihashlookup(VFSTOUFS(dirrem->dm_mnt)((struct ufsmount *)((dirrem->dm_mnt)->mnt_data))->um_dev, |
730 | dirrem->dm_oldinum); |
731 | if (vp == NULL((void *)0) || !VOP_ISLOCKED(vp)) |
732 | break; |
733 | } |
734 | if (wk == NULL((void *)0)) { |
735 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
736 | return (0); |
737 | } |
738 | /* |
739 | * Remove the item to be processed. If we are removing the last |
740 | * item on the list, we need to recalculate the tail pointer. |
741 | * As this happens rarely and usually when the list is short, |
742 | * we just run down the list to find it rather than tracking it |
743 | * in the above loop. |
744 | */ |
745 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
746 | if (wk == worklist_tail) { |
747 | LIST_FOREACH(wkend, &softdep_workitem_pending, wk_list)for((wkend) = ((&softdep_workitem_pending)->lh_first); (wkend)!= ((void *)0); (wkend) = ((wkend)->wk_list.le_next )) |
748 | if (LIST_NEXT(wkend, wk_list)((wkend)->wk_list.le_next) == NULL((void *)0)) |
749 | break; |
750 | worklist_tail = wkend; |
751 | } |
752 | num_on_worklist -= 1; |
753 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
754 | switch (wk->wk_type) { |
755 | |
756 | case D_DIRREM12: |
757 | /* removal of a directory entry */ |
758 | mp = WK_DIRREM(wk)((struct dirrem *)(wk))->dm_mnt; |
759 | #if 0 |
760 | if (vn_write_suspend_wait(NULL((void *)0), mp, V_NOWAIT)) |
761 | panic("%s: dirrem on suspended filesystem", |
762 | "process_worklist_item"); |
763 | #endif |
764 | if (matchmnt != NULL((void *)0) && mp == matchmnt) |
765 | *matchcnt += 1; |
766 | handle_workitem_remove(WK_DIRREM(wk)((struct dirrem *)(wk))); |
767 | break; |
768 | |
769 | case D_FREEBLKS8: |
770 | /* releasing blocks and/or fragments from a file */ |
771 | mp = WK_FREEBLKS(wk)((struct freeblks *)(wk))->fb_mnt; |
772 | #if 0 |
773 | if (vn_write_suspend_wait(NULL((void *)0), mp, V_NOWAIT)) |
774 | panic("%s: freeblks on suspended filesystem", |
775 | "process_worklist_item"); |
776 | #endif |
777 | if (matchmnt != NULL((void *)0) && mp == matchmnt) |
778 | *matchcnt += 1; |
779 | handle_workitem_freeblocks(WK_FREEBLKS(wk)((struct freeblks *)(wk))); |
780 | break; |
781 | |
782 | case D_FREEFRAG7: |
783 | /* releasing a fragment when replaced as a file grows */ |
784 | mp = WK_FREEFRAG(wk)((struct freefrag *)(wk))->ff_mnt; |
785 | #if 0 |
786 | if (vn_write_suspend_wait(NULL((void *)0), mp, V_NOWAIT)) |
787 | panic("%s: freefrag on suspended filesystem", |
788 | "process_worklist_item"); |
789 | #endif |
790 | if (matchmnt != NULL((void *)0) && mp == matchmnt) |
791 | *matchcnt += 1; |
792 | handle_workitem_freefrag(WK_FREEFRAG(wk)((struct freefrag *)(wk))); |
793 | break; |
794 | |
795 | case D_FREEFILE9: |
796 | /* releasing an inode when its link count drops to 0 */ |
797 | mp = WK_FREEFILE(wk)((struct freefile *)(wk))->fx_mnt; |
798 | #if 0 |
799 | if (vn_write_suspend_wait(NULL((void *)0), mp, V_NOWAIT)) |
800 | panic("%s: freefile on suspended filesystem", |
801 | "process_worklist_item"); |
802 | #endif |
803 | if (matchmnt != NULL((void *)0) && mp == matchmnt) |
804 | *matchcnt += 1; |
805 | handle_workitem_freefile(WK_FREEFILE(wk)((struct freefile *)(wk))); |
806 | break; |
807 | |
808 | default: |
809 | panic("%s_process_worklist: Unknown type %s", |
810 | "softdep", TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
811 | /* NOTREACHED */ |
812 | } |
813 | return (1); |
814 | } |
815 | |
816 | /* |
817 | * Move dependencies from one buffer to another. |
818 | */ |
819 | void |
820 | softdep_move_dependencies(struct buf *oldbp, struct buf *newbp) |
821 | { |
822 | struct worklist *wk, *wktail; |
823 | |
824 | if (LIST_FIRST(&newbp->b_dep)((&newbp->b_dep)->lh_first) != NULL((void *)0)) |
825 | panic("softdep_move_dependencies: need merge code"); |
826 | wktail = NULL((void *)0); |
827 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
828 | while ((wk = LIST_FIRST(&oldbp->b_dep)((&oldbp->b_dep)->lh_first)) != NULL((void *)0)) { |
829 | LIST_REMOVE(wk, wk_list)do { if ((wk)->wk_list.le_next != ((void *)0)) (wk)->wk_list .le_next->wk_list.le_prev = (wk)->wk_list.le_prev; *(wk )->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list .le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0); |
830 | if (wktail == NULL((void *)0)) |
831 | LIST_INSERT_HEAD(&newbp->b_dep, wk, wk_list)do { if (((wk)->wk_list.le_next = (&newbp->b_dep)-> lh_first) != ((void *)0)) (&newbp->b_dep)->lh_first ->wk_list.le_prev = &(wk)->wk_list.le_next; (&newbp ->b_dep)->lh_first = (wk); (wk)->wk_list.le_prev = & (&newbp->b_dep)->lh_first; } while (0); |
832 | else |
833 | LIST_INSERT_AFTER(wktail, wk, wk_list)do { if (((wk)->wk_list.le_next = (wktail)->wk_list.le_next ) != ((void *)0)) (wktail)->wk_list.le_next->wk_list.le_prev = &(wk)->wk_list.le_next; (wktail)->wk_list.le_next = (wk); (wk)->wk_list.le_prev = &(wktail)->wk_list .le_next; } while (0); |
834 | wktail = wk; |
835 | } |
836 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
837 | } |
838 | |
839 | /* |
840 | * Purge the work list of all items associated with a particular mount point. |
841 | */ |
842 | int |
843 | softdep_flushworklist(struct mount *oldmnt, int *countp, struct proc *p) |
844 | { |
845 | struct vnode *devvp; |
846 | int count, error = 0; |
847 | |
848 | /* |
849 | * Await our turn to clear out the queue, then serialize access. |
850 | */ |
851 | while (softdep_worklist_busy) { |
852 | softdep_worklist_req += 1; |
853 | tsleep_nsec(&softdep_worklist_req, PRIBIO16, "softflush", INFSLP0xffffffffffffffffULL); |
854 | softdep_worklist_req -= 1; |
855 | } |
856 | softdep_worklist_busy = -1; |
857 | /* |
858 | * Alternately flush the block device associated with the mount |
859 | * point and process any dependencies that the flushing |
860 | * creates. We continue until no more worklist dependencies |
861 | * are found. |
862 | */ |
863 | *countp = 0; |
864 | devvp = VFSTOUFS(oldmnt)((struct ufsmount *)((oldmnt)->mnt_data))->um_devvp; |
865 | while ((count = softdep_process_worklist(oldmnt)) > 0) { |
866 | *countp += count; |
867 | vn_lock(devvp, LK_EXCLUSIVE0x0001UL | LK_RETRY0x2000UL); |
868 | error = VOP_FSYNC(devvp, p->p_ucred, MNT_WAIT1, p); |
869 | VOP_UNLOCK(devvp); |
870 | if (error) |
871 | break; |
872 | } |
873 | softdep_worklist_busy = 0; |
874 | if (softdep_worklist_req) |
875 | wakeup(&softdep_worklist_req); |
876 | return (error); |
877 | } |
878 | |
879 | /* |
880 | * Flush all vnodes and worklist items associated with a specified mount point. |
881 | */ |
882 | int |
883 | softdep_flushfiles(struct mount *oldmnt, int flags, struct proc *p) |
884 | { |
885 | int error, count, loopcnt; |
886 | |
887 | /* |
888 | * Alternately flush the vnodes associated with the mount |
889 | * point and process any dependencies that the flushing |
890 | * creates. In theory, this loop can happen at most twice, |
891 | * but we give it a few extra just to be sure. |
892 | */ |
893 | for (loopcnt = 10; loopcnt > 0; loopcnt--) { |
894 | /* |
895 | * Do another flush in case any vnodes were brought in |
896 | * as part of the cleanup operations. |
897 | */ |
898 | if ((error = ffs_flushfiles(oldmnt, flags, p)) != 0) |
899 | break; |
900 | if ((error = softdep_flushworklist(oldmnt, &count, p)) != 0 || |
901 | count == 0) |
902 | break; |
903 | } |
904 | /* |
905 | * If the reboot process sleeps during the loop, the update |
906 | * process may call softdep_process_worklist() and create |
907 | * new dirty vnodes at the mount point. Call ffs_flushfiles() |
908 | * again after the loop has flushed all soft dependencies. |
909 | */ |
910 | if (error == 0) |
911 | error = ffs_flushfiles(oldmnt, flags, p); |
912 | /* |
913 | * If we are unmounting then it is an error to fail. If we |
914 | * are simply trying to downgrade to read-only, then filesystem |
915 | * activity can keep us busy forever, so we just fail with EBUSY. |
916 | */ |
917 | if (loopcnt == 0) { |
918 | error = EBUSY16; |
919 | } |
920 | return (error); |
921 | } |
922 | |
923 | /* |
924 | * Structure hashing. |
925 | * |
926 | * There are three types of structures that can be looked up: |
927 | * 1) pagedep structures identified by mount point, inode number, |
928 | * and logical block. |
929 | * 2) inodedep structures identified by mount point and inode number. |
930 | * 3) newblk structures identified by mount point and |
931 | * physical block number. |
932 | * |
933 | * The "pagedep" and "inodedep" dependency structures are hashed |
934 | * separately from the file blocks and inodes to which they correspond. |
935 | * This separation helps when the in-memory copy of an inode or |
936 | * file block must be replaced. It also obviates the need to access |
937 | * an inode or file page when simply updating (or de-allocating) |
938 | * dependency structures. Lookup of newblk structures is needed to |
939 | * find newly allocated blocks when trying to associate them with |
940 | * their allocdirect or allocindir structure. |
941 | * |
942 | * The lookup routines optionally create and hash a new instance when |
943 | * an existing entry is not found. |
944 | */ |
945 | #define DEPALLOC0x0001 0x0001 /* allocate structure if lookup fails */ |
946 | #define NODELAY0x0002 0x0002 /* cannot do background work */ |
947 | |
948 | SIPHASH_KEY softdep_hashkey; |
949 | |
950 | /* |
951 | * Structures and routines associated with pagedep caching. |
952 | */ |
953 | LIST_HEAD(pagedep_hashhead, pagedep)struct pagedep_hashhead { struct pagedep *lh_first; } *pagedep_hashtbl; |
954 | u_long pagedep_hash; /* size of hash table - 1 */ |
955 | STATIC struct sema pagedep_in_progress; |
956 | |
957 | /* |
958 | * Look up a pagedep. Return 1 if found, 0 if not found or found |
959 | * when asked to allocate but not associated with any buffer. |
960 | * If not found, allocate if DEPALLOC flag is passed. |
961 | * Found or allocated entry is returned in pagedeppp. |
962 | * This routine must be called with splbio interrupts blocked. |
963 | */ |
964 | STATIC int |
965 | pagedep_lookup(struct inode *ip, daddr_t lbn, int flags, |
966 | struct pagedep **pagedeppp) |
967 | { |
968 | SIPHASH_CTX ctx; |
969 | struct pagedep *pagedep; |
970 | struct pagedep_hashhead *pagedephd; |
971 | struct mount *mp; |
972 | int i; |
973 | |
974 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
975 | |
976 | #ifdef DEBUG |
977 | if (lk.lkt_held == -1) |
978 | panic("pagedep_lookup: lock not held"); |
979 | #endif |
980 | mp = ITOV(ip)((ip)->i_vnode)->v_mount; |
981 | |
982 | SipHash24_Init(&ctx, &softdep_hashkey)SipHash_Init((&ctx), (&softdep_hashkey)); |
983 | SipHash24_Update(&ctx, &mp, sizeof(mp))SipHash_Update((&ctx), 2, 4, (&mp), (sizeof(mp))); |
984 | SipHash24_Update(&ctx, &ip->i_number, sizeof(ip->i_number))SipHash_Update((&ctx), 2, 4, (&ip->i_number), (sizeof (ip->i_number))); |
985 | SipHash24_Update(&ctx, &lbn, sizeof(lbn))SipHash_Update((&ctx), 2, 4, (&lbn), (sizeof(lbn))); |
986 | pagedephd = &pagedep_hashtbl[SipHash24_End(&ctx)SipHash_End((&ctx), 2, 4) & pagedep_hash]; |
987 | top: |
988 | LIST_FOREACH(pagedep, pagedephd, pd_hash)for((pagedep) = ((pagedephd)->lh_first); (pagedep)!= ((void *)0); (pagedep) = ((pagedep)->pd_hash.le_next)) |
989 | if (ip->i_number == pagedep->pd_ino && |
990 | lbn == pagedep->pd_lbn && |
991 | mp == pagedep->pd_mnt) |
992 | break; |
993 | if (pagedep) { |
994 | *pagedeppp = pagedep; |
995 | if ((flags & DEPALLOC0x0001) != 0 && |
996 | (pagedep->pd_statepd_list.wk_state & ONWORKLIST0x8000) == 0) |
997 | return (0); |
998 | return (1); |
999 | } |
1000 | if ((flags & DEPALLOC0x0001) == 0) { |
1001 | *pagedeppp = NULL((void *)0); |
1002 | return (0); |
1003 | } |
1004 | if (sema_get(&pagedep_in_progress, &lk) == 0) { |
1005 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1006 | goto top; |
1007 | } |
1008 | pagedep = pool_get(&pagedep_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
1009 | pagedep->pd_list.wk_type = D_PAGEDEP0; |
1010 | pagedep->pd_mnt = mp; |
1011 | pagedep->pd_ino = ip->i_number; |
1012 | pagedep->pd_lbn = lbn; |
1013 | LIST_INIT(&pagedep->pd_dirremhd)do { ((&pagedep->pd_dirremhd)->lh_first) = ((void * )0); } while (0); |
1014 | LIST_INIT(&pagedep->pd_pendinghd)do { ((&pagedep->pd_pendinghd)->lh_first) = ((void * )0); } while (0); |
1015 | for (i = 0; i < DAHASHSZ6; i++) |
1016 | LIST_INIT(&pagedep->pd_diraddhd[i])do { ((&pagedep->pd_diraddhd[i])->lh_first) = ((void *)0); } while (0); |
1017 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1018 | LIST_INSERT_HEAD(pagedephd, pagedep, pd_hash)do { if (((pagedep)->pd_hash.le_next = (pagedephd)->lh_first ) != ((void *)0)) (pagedephd)->lh_first->pd_hash.le_prev = &(pagedep)->pd_hash.le_next; (pagedephd)->lh_first = (pagedep); (pagedep)->pd_hash.le_prev = &(pagedephd )->lh_first; } while (0); |
1019 | sema_release(&pagedep_in_progress); |
1020 | *pagedeppp = pagedep; |
1021 | return (0); |
1022 | } |
1023 | |
1024 | /* |
1025 | * Structures and routines associated with inodedep caching. |
1026 | */ |
1027 | LIST_HEAD(inodedep_hashhead, inodedep)struct inodedep_hashhead { struct inodedep *lh_first; } *inodedep_hashtbl; |
1028 | STATIC u_long inodedep_hash; /* size of hash table - 1 */ |
1029 | STATIC long num_inodedep; /* number of inodedep allocated */ |
1030 | STATIC struct sema inodedep_in_progress; |
1031 | |
1032 | /* |
1033 | * Look up a inodedep. Return 1 if found, 0 if not found. |
1034 | * If not found, allocate if DEPALLOC flag is passed. |
1035 | * Found or allocated entry is returned in inodedeppp. |
1036 | * This routine must be called with splbio interrupts blocked. |
1037 | */ |
1038 | STATIC int |
1039 | inodedep_lookup(struct fs *fs, ufsino_t inum, int flags, |
1040 | struct inodedep **inodedeppp) |
1041 | { |
1042 | SIPHASH_CTX ctx; |
1043 | struct inodedep *inodedep; |
1044 | struct inodedep_hashhead *inodedephd; |
1045 | int firsttry; |
1046 | |
1047 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
1048 | |
1049 | #ifdef DEBUG |
1050 | if (lk.lkt_held == -1) |
1051 | panic("inodedep_lookup: lock not held"); |
1052 | #endif |
1053 | firsttry = 1; |
1054 | SipHash24_Init(&ctx, &softdep_hashkey)SipHash_Init((&ctx), (&softdep_hashkey)); |
1055 | SipHash24_Update(&ctx, &fs, sizeof(fs))SipHash_Update((&ctx), 2, 4, (&fs), (sizeof(fs))); |
1056 | SipHash24_Update(&ctx, &inum, sizeof(inum))SipHash_Update((&ctx), 2, 4, (&inum), (sizeof(inum))); |
1057 | inodedephd = &inodedep_hashtbl[SipHash24_End(&ctx)SipHash_End((&ctx), 2, 4) & inodedep_hash]; |
1058 | top: |
1059 | LIST_FOREACH(inodedep, inodedephd, id_hash)for((inodedep) = ((inodedephd)->lh_first); (inodedep)!= (( void *)0); (inodedep) = ((inodedep)->id_hash.le_next)) |
1060 | if (inum == inodedep->id_ino && fs == inodedep->id_fs) |
1061 | break; |
1062 | if (inodedep) { |
1063 | *inodedeppp = inodedep; |
1064 | return (1); |
1065 | } |
1066 | if ((flags & DEPALLOC0x0001) == 0) { |
1067 | *inodedeppp = NULL((void *)0); |
1068 | return (0); |
1069 | } |
1070 | /* |
1071 | * If we are over our limit, try to improve the situation. |
1072 | */ |
1073 | if (num_inodedep > max_softdeps && firsttry && (flags & NODELAY0x0002) == 0 && |
1074 | request_cleanup(FLUSH_INODES1, 1)) { |
1075 | firsttry = 0; |
1076 | goto top; |
1077 | } |
1078 | if (sema_get(&inodedep_in_progress, &lk) == 0) { |
1079 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1080 | goto top; |
1081 | } |
1082 | num_inodedep += 1; |
1083 | inodedep = pool_get(&inodedep_pool, PR_WAITOK0x0001); |
1084 | inodedep->id_list.wk_type = D_INODEDEP1; |
1085 | inodedep->id_fs = fs; |
1086 | inodedep->id_ino = inum; |
1087 | inodedep->id_stateid_list.wk_state = ALLCOMPLETE(0x0001 | 0x0004 | 0x0008); |
1088 | inodedep->id_nlinkdelta = 0; |
1089 | inodedep->id_savedino1id_un.idu_savedino1 = NULL((void *)0); |
1090 | inodedep->id_savedsize = -1; |
1091 | inodedep->id_buf = NULL((void *)0); |
1092 | LIST_INIT(&inodedep->id_pendinghd)do { ((&inodedep->id_pendinghd)->lh_first) = ((void *)0); } while (0); |
1093 | LIST_INIT(&inodedep->id_inowait)do { ((&inodedep->id_inowait)->lh_first) = ((void * )0); } while (0); |
1094 | LIST_INIT(&inodedep->id_bufwait)do { ((&inodedep->id_bufwait)->lh_first) = ((void * )0); } while (0); |
1095 | TAILQ_INIT(&inodedep->id_inoupdt)do { (&inodedep->id_inoupdt)->tqh_first = ((void *) 0); (&inodedep->id_inoupdt)->tqh_last = &(& inodedep->id_inoupdt)->tqh_first; } while (0); |
1096 | TAILQ_INIT(&inodedep->id_newinoupdt)do { (&inodedep->id_newinoupdt)->tqh_first = ((void *)0); (&inodedep->id_newinoupdt)->tqh_last = & (&inodedep->id_newinoupdt)->tqh_first; } while (0); |
1097 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1098 | LIST_INSERT_HEAD(inodedephd, inodedep, id_hash)do { if (((inodedep)->id_hash.le_next = (inodedephd)->lh_first ) != ((void *)0)) (inodedephd)->lh_first->id_hash.le_prev = &(inodedep)->id_hash.le_next; (inodedephd)->lh_first = (inodedep); (inodedep)->id_hash.le_prev = &(inodedephd )->lh_first; } while (0); |
1099 | sema_release(&inodedep_in_progress); |
1100 | *inodedeppp = inodedep; |
1101 | return (0); |
1102 | } |
1103 | |
1104 | /* |
1105 | * Structures and routines associated with newblk caching. |
1106 | */ |
1107 | LIST_HEAD(newblk_hashhead, newblk)struct newblk_hashhead { struct newblk *lh_first; } *newblk_hashtbl; |
1108 | u_long newblk_hash; /* size of hash table - 1 */ |
1109 | STATIC struct sema newblk_in_progress; |
1110 | |
1111 | /* |
1112 | * Look up a newblk. Return 1 if found, 0 if not found. |
1113 | * If not found, allocate if DEPALLOC flag is passed. |
1114 | * Found or allocated entry is returned in newblkpp. |
1115 | */ |
1116 | STATIC int |
1117 | newblk_lookup(struct fs *fs, daddr_t newblkno, int flags, |
1118 | struct newblk **newblkpp) |
1119 | { |
1120 | SIPHASH_CTX ctx; |
1121 | struct newblk *newblk; |
1122 | struct newblk_hashhead *newblkhd; |
1123 | |
1124 | SipHash24_Init(&ctx, &softdep_hashkey)SipHash_Init((&ctx), (&softdep_hashkey)); |
1125 | SipHash24_Update(&ctx, &fs, sizeof(fs))SipHash_Update((&ctx), 2, 4, (&fs), (sizeof(fs))); |
1126 | SipHash24_Update(&ctx, &newblkno, sizeof(newblkno))SipHash_Update((&ctx), 2, 4, (&newblkno), (sizeof(newblkno ))); |
1127 | newblkhd = &newblk_hashtbl[SipHash24_End(&ctx)SipHash_End((&ctx), 2, 4) & newblk_hash]; |
1128 | top: |
1129 | LIST_FOREACH(newblk, newblkhd, nb_hash)for((newblk) = ((newblkhd)->lh_first); (newblk)!= ((void * )0); (newblk) = ((newblk)->nb_hash.le_next)) |
1130 | if (newblkno == newblk->nb_newblkno && fs == newblk->nb_fs) |
1131 | break; |
1132 | if (newblk) { |
1133 | *newblkpp = newblk; |
1134 | return (1); |
1135 | } |
1136 | if ((flags & DEPALLOC0x0001) == 0) { |
1137 | *newblkpp = NULL((void *)0); |
1138 | return (0); |
1139 | } |
1140 | if (sema_get(&newblk_in_progress, NULL((void *)0)) == 0) |
1141 | goto top; |
1142 | newblk = pool_get(&newblk_pool, PR_WAITOK0x0001); |
1143 | newblk->nb_state = 0; |
1144 | newblk->nb_fs = fs; |
1145 | newblk->nb_newblkno = newblkno; |
1146 | LIST_INSERT_HEAD(newblkhd, newblk, nb_hash)do { if (((newblk)->nb_hash.le_next = (newblkhd)->lh_first ) != ((void *)0)) (newblkhd)->lh_first->nb_hash.le_prev = &(newblk)->nb_hash.le_next; (newblkhd)->lh_first = (newblk); (newblk)->nb_hash.le_prev = &(newblkhd)-> lh_first; } while (0); |
1147 | sema_release(&newblk_in_progress); |
1148 | *newblkpp = newblk; |
1149 | return (0); |
1150 | } |
1151 | |
1152 | /* |
1153 | * Executed during filesystem system initialization before |
1154 | * mounting any file systems. |
1155 | */ |
1156 | void |
1157 | softdep_initialize(void) |
1158 | { |
1159 | |
1160 | bioops.io_start = softdep_disk_io_initiation; |
1161 | bioops.io_complete = softdep_disk_write_complete; |
1162 | bioops.io_deallocate = softdep_deallocate_dependencies; |
1163 | bioops.io_movedeps = softdep_move_dependencies; |
1164 | bioops.io_countdeps = softdep_count_dependencies; |
1165 | |
1166 | LIST_INIT(&mkdirlisthd)do { ((&mkdirlisthd)->lh_first) = ((void *)0); } while (0); |
1167 | LIST_INIT(&softdep_workitem_pending)do { ((&softdep_workitem_pending)->lh_first) = ((void * )0); } while (0); |
1168 | #ifdef KMEMSTATS1 |
1169 | max_softdeps = min (initialvnodes * 8, |
1170 | kmemstats[M_INODEDEP79].ks_limit / (2 * sizeof(struct inodedep))); |
1171 | #else |
1172 | max_softdeps = initialvnodes * 4; |
1173 | #endif |
1174 | arc4random_buf(&softdep_hashkey, sizeof(softdep_hashkey)); |
1175 | pagedep_hashtbl = hashinit(initialvnodes / 5, M_PAGEDEP78, M_WAITOK0x0001, |
1176 | &pagedep_hash); |
1177 | sema_init(&pagedep_in_progress, "pagedep", PRIBIO16); |
1178 | inodedep_hashtbl = hashinit(initialvnodes, M_INODEDEP79, M_WAITOK0x0001, |
1179 | &inodedep_hash); |
1180 | sema_init(&inodedep_in_progress, "inodedep", PRIBIO16); |
1181 | newblk_hashtbl = hashinit(64, M_NEWBLK80, M_WAITOK0x0001, &newblk_hash); |
1182 | sema_init(&newblk_in_progress, "newblk", PRIBIO16); |
1183 | timeout_set(&proc_waiting_timeout, pause_timer, NULL((void *)0)); |
1184 | pool_init(&pagedep_pool, sizeof(struct pagedep), 0, IPL_NONE0x0, |
1185 | PR_WAITOK0x0001, "pagedep", NULL((void *)0)); |
1186 | pool_init(&inodedep_pool, sizeof(struct inodedep), 0, IPL_NONE0x0, |
1187 | PR_WAITOK0x0001, "inodedep", NULL((void *)0)); |
1188 | pool_init(&newblk_pool, sizeof(struct newblk), 0, IPL_NONE0x0, |
1189 | PR_WAITOK0x0001, "newblk", NULL((void *)0)); |
1190 | pool_init(&bmsafemap_pool, sizeof(struct bmsafemap), 0, IPL_NONE0x0, |
1191 | PR_WAITOK0x0001, "bmsafemap", NULL((void *)0)); |
1192 | pool_init(&allocdirect_pool, sizeof(struct allocdirect), 0, IPL_NONE0x0, |
1193 | PR_WAITOK0x0001, "allocdir", NULL((void *)0)); |
1194 | pool_init(&indirdep_pool, sizeof(struct indirdep), 0, IPL_NONE0x0, |
1195 | PR_WAITOK0x0001, "indirdep", NULL((void *)0)); |
1196 | pool_init(&allocindir_pool, sizeof(struct allocindir), 0, IPL_NONE0x0, |
1197 | PR_WAITOK0x0001, "allocindir", NULL((void *)0)); |
1198 | pool_init(&freefrag_pool, sizeof(struct freefrag), 0, IPL_NONE0x0, |
1199 | PR_WAITOK0x0001, "freefrag", NULL((void *)0)); |
1200 | pool_init(&freeblks_pool, sizeof(struct freeblks), 0, IPL_NONE0x0, |
1201 | PR_WAITOK0x0001, "freeblks", NULL((void *)0)); |
1202 | pool_init(&freefile_pool, sizeof(struct freefile), 0, IPL_NONE0x0, |
1203 | PR_WAITOK0x0001, "freefile", NULL((void *)0)); |
1204 | pool_init(&diradd_pool, sizeof(struct diradd), 0, IPL_NONE0x0, |
1205 | PR_WAITOK0x0001, "diradd", NULL((void *)0)); |
1206 | pool_init(&mkdir_pool, sizeof(struct mkdir), 0, IPL_NONE0x0, |
1207 | PR_WAITOK0x0001, "mkdir", NULL((void *)0)); |
1208 | pool_init(&dirrem_pool, sizeof(struct dirrem), 0, IPL_NONE0x0, |
1209 | PR_WAITOK0x0001, "dirrem", NULL((void *)0)); |
1210 | pool_init(&newdirblk_pool, sizeof(struct newdirblk), 0, IPL_NONE0x0, |
1211 | PR_WAITOK0x0001, "newdirblk", NULL((void *)0)); |
1212 | } |
1213 | |
1214 | /* |
1215 | * Called at mount time to notify the dependency code that a |
1216 | * filesystem wishes to use it. |
1217 | */ |
1218 | int |
1219 | softdep_mount(struct vnode *devvp, struct mount *mp, struct fs *fs, |
1220 | struct ucred *cred) |
1221 | { |
1222 | struct csum_total cstotal; |
1223 | struct cg *cgp; |
1224 | struct buf *bp; |
1225 | int error, cyl; |
1226 | |
1227 | /* |
1228 | * When doing soft updates, the counters in the |
1229 | * superblock may have gotten out of sync, so we have |
1230 | * to scan the cylinder groups and recalculate them. |
1231 | */ |
1232 | if ((fs->fs_flags & FS_UNCLEAN0x01) == 0) |
1233 | return (0); |
1234 | memset(&cstotal, 0, sizeof(cstotal))__builtin_memset((&cstotal), (0), (sizeof(cstotal))); |
1235 | for (cyl = 0; cyl < fs->fs_ncg; cyl++) { |
1236 | if ((error = bread(devvp, fsbtodb(fs, cgtod(fs, cyl))((((((daddr_t)(fs)->fs_fpg * (cyl)) + (fs)->fs_cgoffset * ((cyl) & ~((fs)->fs_cgmask))) + (fs)->fs_cblkno) ) << (fs)->fs_fsbtodb), |
1237 | fs->fs_cgsize, &bp)) != 0) { |
1238 | brelse(bp); |
1239 | return (error); |
1240 | } |
1241 | cgp = (struct cg *)bp->b_data; |
1242 | cstotal.cs_nffree += cgp->cg_cs.cs_nffree; |
1243 | cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree; |
1244 | cstotal.cs_nifree += cgp->cg_cs.cs_nifree; |
1245 | cstotal.cs_ndir += cgp->cg_cs.cs_ndir; |
1246 | fs->fs_cs(fs, cyl)fs_csp[cyl] = cgp->cg_cs; |
1247 | brelse(bp); |
1248 | } |
1249 | #ifdef DEBUG |
1250 | if (memcmp(&cstotal, &fs->fs_cstotal, sizeof(cstotal))__builtin_memcmp((&cstotal), (&fs->fs_cstotal), (sizeof (cstotal)))) |
1251 | printf("ffs_mountfs: superblock updated for soft updates\n"); |
1252 | #endif |
1253 | memcpy(&fs->fs_cstotal, &cstotal, sizeof(cstotal))__builtin_memcpy((&fs->fs_cstotal), (&cstotal), (sizeof (cstotal))); |
1254 | return (0); |
1255 | } |
1256 | |
1257 | /* |
1258 | * Protecting the freemaps (or bitmaps). |
1259 | * |
1260 | * To eliminate the need to execute fsck before mounting a file system |
1261 | * after a power failure, one must (conservatively) guarantee that the |
1262 | * on-disk copy of the bitmaps never indicate that a live inode or block is |
1263 | * free. So, when a block or inode is allocated, the bitmap should be |
1264 | * updated (on disk) before any new pointers. When a block or inode is |
1265 | * freed, the bitmap should not be updated until all pointers have been |
1266 | * reset. The latter dependency is handled by the delayed de-allocation |
1267 | * approach described below for block and inode de-allocation. The former |
1268 | * dependency is handled by calling the following procedure when a block or |
1269 | * inode is allocated. When an inode is allocated an "inodedep" is created |
1270 | * with its DEPCOMPLETE flag cleared until its bitmap is written to disk. |
1271 | * Each "inodedep" is also inserted into the hash indexing structure so |
1272 | * that any additional link additions can be made dependent on the inode |
1273 | * allocation. |
1274 | * |
1275 | * The ufs file system maintains a number of free block counts (e.g., per |
1276 | * cylinder group, per cylinder and per <cylinder, rotational position> pair) |
1277 | * in addition to the bitmaps. These counts are used to improve efficiency |
1278 | * during allocation and therefore must be consistent with the bitmaps. |
1279 | * There is no convenient way to guarantee post-crash consistency of these |
1280 | * counts with simple update ordering, for two main reasons: (1) The counts |
1281 | * and bitmaps for a single cylinder group block are not in the same disk |
1282 | * sector. If a disk write is interrupted (e.g., by power failure), one may |
1283 | * be written and the other not. (2) Some of the counts are located in the |
1284 | * superblock rather than the cylinder group block. So, we focus our soft |
1285 | * updates implementation on protecting the bitmaps. When mounting a |
1286 | * filesystem, we recompute the auxiliary counts from the bitmaps. |
1287 | */ |
1288 | |
1289 | /* |
1290 | * Called just after updating the cylinder group block to allocate an inode. |
1291 | */ |
1292 | /* buffer for cylgroup block with inode map */ |
1293 | /* inode related to allocation */ |
1294 | /* new inode number being allocated */ |
1295 | void |
1296 | softdep_setup_inomapdep(struct buf *bp, struct inode *ip, ufsino_t newinum) |
1297 | { |
1298 | struct inodedep *inodedep; |
1299 | struct bmsafemap *bmsafemap; |
1300 | |
1301 | /* |
1302 | * Create a dependency for the newly allocated inode. |
1303 | * Panic if it already exists as something is seriously wrong. |
1304 | * Otherwise add it to the dependency list for the buffer holding |
1305 | * the cylinder group map from which it was allocated. |
1306 | */ |
1307 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1308 | if (inodedep_lookup(ip->i_fsinode_u.fs, newinum, DEPALLOC0x0001 | NODELAY0x0002, &inodedep) |
1309 | != 0) { |
1310 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1311 | panic("softdep_setup_inomapdep: found inode"); |
1312 | } |
1313 | inodedep->id_buf = bp; |
1314 | inodedep->id_stateid_list.wk_state &= ~DEPCOMPLETE0x0008; |
1315 | bmsafemap = bmsafemap_lookup(bp); |
1316 | LIST_INSERT_HEAD(&bmsafemap->sm_inodedephd, inodedep, id_deps)do { if (((inodedep)->id_deps.le_next = (&bmsafemap-> sm_inodedephd)->lh_first) != ((void *)0)) (&bmsafemap-> sm_inodedephd)->lh_first->id_deps.le_prev = &(inodedep )->id_deps.le_next; (&bmsafemap->sm_inodedephd)-> lh_first = (inodedep); (inodedep)->id_deps.le_prev = & (&bmsafemap->sm_inodedephd)->lh_first; } while (0); |
1317 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1318 | } |
1319 | |
1320 | /* |
1321 | * Called just after updating the cylinder group block to |
1322 | * allocate block or fragment. |
1323 | */ |
1324 | /* buffer for cylgroup block with block map */ |
1325 | /* filesystem doing allocation */ |
1326 | /* number of newly allocated block */ |
1327 | void |
1328 | softdep_setup_blkmapdep(struct buf *bp, struct fs *fs, daddr_t newblkno) |
1329 | { |
1330 | struct newblk *newblk; |
1331 | struct bmsafemap *bmsafemap; |
1332 | |
1333 | /* |
1334 | * Create a dependency for the newly allocated block. |
1335 | * Add it to the dependency list for the buffer holding |
1336 | * the cylinder group map from which it was allocated. |
1337 | */ |
1338 | if (newblk_lookup(fs, newblkno, DEPALLOC0x0001, &newblk) != 0) |
1339 | panic("softdep_setup_blkmapdep: found block"); |
1340 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1341 | newblk->nb_bmsafemap = bmsafemap = bmsafemap_lookup(bp); |
1342 | LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk, nb_deps)do { if (((newblk)->nb_deps.le_next = (&bmsafemap-> sm_newblkhd)->lh_first) != ((void *)0)) (&bmsafemap-> sm_newblkhd)->lh_first->nb_deps.le_prev = &(newblk) ->nb_deps.le_next; (&bmsafemap->sm_newblkhd)->lh_first = (newblk); (newblk)->nb_deps.le_prev = &(&bmsafemap ->sm_newblkhd)->lh_first; } while (0); |
1343 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1344 | } |
1345 | |
1346 | /* |
1347 | * Find the bmsafemap associated with a cylinder group buffer. |
1348 | * If none exists, create one. The buffer must be locked when |
1349 | * this routine is called and this routine must be called with |
1350 | * splbio interrupts blocked. |
1351 | */ |
1352 | STATIC struct bmsafemap * |
1353 | bmsafemap_lookup(struct buf *bp) |
1354 | { |
1355 | struct bmsafemap *bmsafemap; |
1356 | struct worklist *wk; |
1357 | |
1358 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
1359 | |
1360 | #ifdef DEBUG |
1361 | if (lk.lkt_held == -1) |
1362 | panic("bmsafemap_lookup: lock not held"); |
1363 | #endif |
1364 | LIST_FOREACH(wk, &bp->b_dep, wk_list)for((wk) = ((&bp->b_dep)->lh_first); (wk)!= ((void * )0); (wk) = ((wk)->wk_list.le_next)) |
1365 | if (wk->wk_type == D_BMSAFEMAP3) |
1366 | return (WK_BMSAFEMAP(wk)((struct bmsafemap *)(wk))); |
1367 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1368 | bmsafemap = pool_get(&bmsafemap_pool, PR_WAITOK0x0001); |
1369 | bmsafemap->sm_list.wk_type = D_BMSAFEMAP3; |
1370 | bmsafemap->sm_list.wk_state = 0; |
1371 | bmsafemap->sm_buf = bp; |
1372 | LIST_INIT(&bmsafemap->sm_allocdirecthd)do { ((&bmsafemap->sm_allocdirecthd)->lh_first) = ( (void *)0); } while (0); |
1373 | LIST_INIT(&bmsafemap->sm_allocindirhd)do { ((&bmsafemap->sm_allocindirhd)->lh_first) = (( void *)0); } while (0); |
1374 | LIST_INIT(&bmsafemap->sm_inodedephd)do { ((&bmsafemap->sm_inodedephd)->lh_first) = ((void *)0); } while (0); |
1375 | LIST_INIT(&bmsafemap->sm_newblkhd)do { ((&bmsafemap->sm_newblkhd)->lh_first) = ((void *)0); } while (0); |
1376 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1377 | WORKLIST_INSERT(&bp->b_dep, &bmsafemap->sm_list)do { (&bmsafemap->sm_list)->wk_state |= 0x8000; do { if (((&bmsafemap->sm_list)->wk_list.le_next = (& bp->b_dep)->lh_first) != ((void *)0)) (&bp->b_dep )->lh_first->wk_list.le_prev = &(&bmsafemap-> sm_list)->wk_list.le_next; (&bp->b_dep)->lh_first = (&bmsafemap->sm_list); (&bmsafemap->sm_list) ->wk_list.le_prev = &(&bp->b_dep)->lh_first; } while (0); } while (0); |
1378 | return (bmsafemap); |
1379 | } |
1380 | |
1381 | /* |
1382 | * Direct block allocation dependencies. |
1383 | * |
1384 | * When a new block is allocated, the corresponding disk locations must be |
1385 | * initialized (with zeros or new data) before the on-disk inode points to |
1386 | * them. Also, the freemap from which the block was allocated must be |
1387 | * updated (on disk) before the inode's pointer. These two dependencies are |
1388 | * independent of each other and are needed for all file blocks and indirect |
1389 | * blocks that are pointed to directly by the inode. Just before the |
1390 | * "in-core" version of the inode is updated with a newly allocated block |
1391 | * number, a procedure (below) is called to setup allocation dependency |
1392 | * structures. These structures are removed when the corresponding |
1393 | * dependencies are satisfied or when the block allocation becomes obsolete |
1394 | * (i.e., the file is deleted, the block is de-allocated, or the block is a |
1395 | * fragment that gets upgraded). All of these cases are handled in |
1396 | * procedures described later. |
1397 | * |
1398 | * When a file extension causes a fragment to be upgraded, either to a larger |
1399 | * fragment or to a full block, the on-disk location may change (if the |
1400 | * previous fragment could not simply be extended). In this case, the old |
1401 | * fragment must be de-allocated, but not until after the inode's pointer has |
1402 | * been updated. In most cases, this is handled by later procedures, which |
1403 | * will construct a "freefrag" structure to be added to the workitem queue |
1404 | * when the inode update is complete (or obsolete). The main exception to |
1405 | * this is when an allocation occurs while a pending allocation dependency |
1406 | * (for the same block pointer) remains. This case is handled in the main |
1407 | * allocation dependency setup procedure by immediately freeing the |
1408 | * unreferenced fragments. |
1409 | */ |
1410 | /* inode to which block is being added */ |
1411 | /* block pointer within inode */ |
1412 | /* disk block number being added */ |
1413 | /* previous block number, 0 unless frag */ |
1414 | /* size of new block */ |
1415 | /* size of new block */ |
1416 | /* bp for allocated block */ |
1417 | void |
1418 | softdep_setup_allocdirect(struct inode *ip, daddr_t lbn, daddr_t newblkno, |
1419 | daddr_t oldblkno, long newsize, long oldsize, struct buf *bp) |
1420 | { |
1421 | struct allocdirect *adp, *oldadp; |
1422 | struct allocdirectlst *adphead; |
1423 | struct bmsafemap *bmsafemap; |
1424 | struct inodedep *inodedep; |
1425 | struct pagedep *pagedep; |
1426 | struct newblk *newblk; |
1427 | |
1428 | adp = pool_get(&allocdirect_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
1429 | adp->ad_list.wk_type = D_ALLOCDIRECT4; |
1430 | adp->ad_lbn = lbn; |
1431 | adp->ad_newblkno = newblkno; |
1432 | adp->ad_oldblkno = oldblkno; |
1433 | adp->ad_newsize = newsize; |
1434 | adp->ad_oldsize = oldsize; |
1435 | adp->ad_statead_list.wk_state = ATTACHED0x0001; |
1436 | LIST_INIT(&adp->ad_newdirblk)do { ((&adp->ad_newdirblk)->lh_first) = ((void *)0) ; } while (0); |
1437 | if (newblkno == oldblkno) |
1438 | adp->ad_freefrag = NULL((void *)0); |
1439 | else |
1440 | adp->ad_freefrag = newfreefrag(ip, oldblkno, oldsize); |
1441 | |
1442 | if (newblk_lookup(ip->i_fsinode_u.fs, newblkno, 0, &newblk) == 0) |
1443 | panic("softdep_setup_allocdirect: lost block"); |
1444 | |
1445 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1446 | inodedep_lookup(ip->i_fsinode_u.fs, ip->i_number, DEPALLOC0x0001 | NODELAY0x0002, &inodedep); |
1447 | adp->ad_inodedep = inodedep; |
1448 | |
1449 | if (newblk->nb_state == DEPCOMPLETE0x0008) { |
1450 | adp->ad_statead_list.wk_state |= DEPCOMPLETE0x0008; |
1451 | adp->ad_buf = NULL((void *)0); |
1452 | } else { |
1453 | bmsafemap = newblk->nb_bmsafemap; |
1454 | adp->ad_buf = bmsafemap->sm_buf; |
1455 | LIST_REMOVE(newblk, nb_deps)do { if ((newblk)->nb_deps.le_next != ((void *)0)) (newblk )->nb_deps.le_next->nb_deps.le_prev = (newblk)->nb_deps .le_prev; *(newblk)->nb_deps.le_prev = (newblk)->nb_deps .le_next; ((newblk)->nb_deps.le_prev) = ((void *)-1); ((newblk )->nb_deps.le_next) = ((void *)-1); } while (0); |
1456 | LIST_INSERT_HEAD(&bmsafemap->sm_allocdirecthd, adp, ad_deps)do { if (((adp)->ad_deps.le_next = (&bmsafemap->sm_allocdirecthd )->lh_first) != ((void *)0)) (&bmsafemap->sm_allocdirecthd )->lh_first->ad_deps.le_prev = &(adp)->ad_deps.le_next ; (&bmsafemap->sm_allocdirecthd)->lh_first = (adp); (adp)->ad_deps.le_prev = &(&bmsafemap->sm_allocdirecthd )->lh_first; } while (0); |
1457 | } |
1458 | LIST_REMOVE(newblk, nb_hash)do { if ((newblk)->nb_hash.le_next != ((void *)0)) (newblk )->nb_hash.le_next->nb_hash.le_prev = (newblk)->nb_hash .le_prev; *(newblk)->nb_hash.le_prev = (newblk)->nb_hash .le_next; ((newblk)->nb_hash.le_prev) = ((void *)-1); ((newblk )->nb_hash.le_next) = ((void *)-1); } while (0); |
1459 | pool_put(&newblk_pool, newblk); |
1460 | |
1461 | if (bp == NULL((void *)0)) { |
1462 | /* |
1463 | * XXXUBC - Yes, I know how to fix this, but not right now. |
1464 | */ |
1465 | panic("softdep_setup_allocdirect: Bonk art in the head"); |
1466 | } |
1467 | WORKLIST_INSERT(&bp->b_dep, &adp->ad_list)do { (&adp->ad_list)->wk_state |= 0x8000; do { if ( ((&adp->ad_list)->wk_list.le_next = (&bp->b_dep )->lh_first) != ((void *)0)) (&bp->b_dep)->lh_first ->wk_list.le_prev = &(&adp->ad_list)->wk_list .le_next; (&bp->b_dep)->lh_first = (&adp->ad_list ); (&adp->ad_list)->wk_list.le_prev = &(&bp ->b_dep)->lh_first; } while (0); } while (0); |
1468 | if (lbn >= NDADDR12) { |
1469 | /* allocating an indirect block */ |
1470 | if (oldblkno != 0) { |
1471 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1472 | panic("softdep_setup_allocdirect: non-zero indir"); |
1473 | } |
1474 | } else { |
1475 | /* |
1476 | * Allocating a direct block. |
1477 | * |
1478 | * If we are allocating a directory block, then we must |
1479 | * allocate an associated pagedep to track additions and |
1480 | * deletions. |
1481 | */ |
1482 | if ((DIP(ip, mode)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_mode : (ip)->dinode_u.ffs2_din->di_mode) & IFMT0170000) == IFDIR0040000 && |
1483 | pagedep_lookup(ip, lbn, DEPALLOC0x0001, &pagedep) == 0) |
1484 | WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list)do { (&pagedep->pd_list)->wk_state |= 0x8000; do { if (((&pagedep->pd_list)->wk_list.le_next = (&bp-> b_dep)->lh_first) != ((void *)0)) (&bp->b_dep)-> lh_first->wk_list.le_prev = &(&pagedep->pd_list )->wk_list.le_next; (&bp->b_dep)->lh_first = (& pagedep->pd_list); (&pagedep->pd_list)->wk_list. le_prev = &(&bp->b_dep)->lh_first; } while (0); } while (0); |
1485 | } |
1486 | /* |
1487 | * The list of allocdirects must be kept in sorted and ascending |
1488 | * order so that the rollback routines can quickly determine the |
1489 | * first uncommitted block (the size of the file stored on disk |
1490 | * ends at the end of the lowest committed fragment, or if there |
1491 | * are no fragments, at the end of the highest committed block). |
1492 | * Since files generally grow, the typical case is that the new |
1493 | * block is to be added at the end of the list. We speed this |
1494 | * special case by checking against the last allocdirect in the |
1495 | * list before laboriously traversing the list looking for the |
1496 | * insertion point. |
1497 | */ |
1498 | adphead = &inodedep->id_newinoupdt; |
1499 | oldadp = TAILQ_LAST(adphead, allocdirectlst)(*(((struct allocdirectlst *)((adphead)->tqh_last))->tqh_last )); |
1500 | if (oldadp == NULL((void *)0) || oldadp->ad_lbn <= lbn) { |
1501 | /* insert at end of list */ |
1502 | TAILQ_INSERT_TAIL(adphead, adp, ad_next)do { (adp)->ad_next.tqe_next = ((void *)0); (adp)->ad_next .tqe_prev = (adphead)->tqh_last; *(adphead)->tqh_last = (adp); (adphead)->tqh_last = &(adp)->ad_next.tqe_next ; } while (0); |
1503 | if (oldadp != NULL((void *)0) && oldadp->ad_lbn == lbn) |
1504 | allocdirect_merge(adphead, adp, oldadp); |
1505 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1506 | return; |
1507 | } |
1508 | TAILQ_FOREACH(oldadp, adphead, ad_next)for((oldadp) = ((adphead)->tqh_first); (oldadp) != ((void * )0); (oldadp) = ((oldadp)->ad_next.tqe_next)) { |
1509 | if (oldadp->ad_lbn >= lbn) |
1510 | break; |
1511 | } |
1512 | if (oldadp == NULL((void *)0)) { |
1513 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1514 | panic("softdep_setup_allocdirect: lost entry"); |
1515 | } |
1516 | /* insert in middle of list */ |
1517 | TAILQ_INSERT_BEFORE(oldadp, adp, ad_next)do { (adp)->ad_next.tqe_prev = (oldadp)->ad_next.tqe_prev ; (adp)->ad_next.tqe_next = (oldadp); *(oldadp)->ad_next .tqe_prev = (adp); (oldadp)->ad_next.tqe_prev = &(adp) ->ad_next.tqe_next; } while (0); |
1518 | if (oldadp->ad_lbn == lbn) |
1519 | allocdirect_merge(adphead, adp, oldadp); |
1520 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1521 | } |
1522 | |
1523 | /* |
1524 | * Replace an old allocdirect dependency with a newer one. |
1525 | * This routine must be called with splbio interrupts blocked. |
1526 | */ |
1527 | /* head of list holding allocdirects */ |
1528 | /* allocdirect being added */ |
1529 | /* existing allocdirect being checked */ |
1530 | STATIC void |
1531 | allocdirect_merge(struct allocdirectlst *adphead, struct allocdirect *newadp, |
1532 | struct allocdirect *oldadp) |
1533 | { |
1534 | struct worklist *wk; |
1535 | struct freefrag *freefrag; |
1536 | struct newdirblk *newdirblk; |
1537 | |
1538 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
1539 | |
1540 | #ifdef DEBUG |
1541 | if (lk.lkt_held == -1) |
1542 | panic("allocdirect_merge: lock not held"); |
1543 | #endif |
1544 | if (newadp->ad_oldblkno != oldadp->ad_newblkno || |
1545 | newadp->ad_oldsize != oldadp->ad_newsize || |
1546 | newadp->ad_lbn >= NDADDR12) { |
1547 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1548 | panic("allocdirect_merge: old %lld != new %lld || lbn %lld >= " |
1549 | "%d", (long long)newadp->ad_oldblkno, |
1550 | (long long)oldadp->ad_newblkno, (long long)newadp->ad_lbn, |
1551 | NDADDR12); |
1552 | } |
1553 | newadp->ad_oldblkno = oldadp->ad_oldblkno; |
1554 | newadp->ad_oldsize = oldadp->ad_oldsize; |
1555 | /* |
1556 | * If the old dependency had a fragment to free or had never |
1557 | * previously had a block allocated, then the new dependency |
1558 | * can immediately post its freefrag and adopt the old freefrag. |
1559 | * This action is done by swapping the freefrag dependencies. |
1560 | * The new dependency gains the old one's freefrag, and the |
1561 | * old one gets the new one and then immediately puts it on |
1562 | * the worklist when it is freed by free_allocdirect. It is |
1563 | * not possible to do this swap when the old dependency had a |
1564 | * non-zero size but no previous fragment to free. This condition |
1565 | * arises when the new block is an extension of the old block. |
1566 | * Here, the first part of the fragment allocated to the new |
1567 | * dependency is part of the block currently claimed on disk by |
1568 | * the old dependency, so cannot legitimately be freed until the |
1569 | * conditions for the new dependency are fulfilled. |
1570 | */ |
1571 | if (oldadp->ad_freefrag != NULL((void *)0) || oldadp->ad_oldblkno == 0) { |
1572 | freefrag = newadp->ad_freefrag; |
1573 | newadp->ad_freefrag = oldadp->ad_freefrag; |
1574 | oldadp->ad_freefrag = freefrag; |
1575 | } |
1576 | /* |
1577 | * If we are tracking a new directory-block allocation, |
1578 | * move it from the old allocdirect to the new allocdirect. |
1579 | */ |
1580 | if ((wk = LIST_FIRST(&oldadp->ad_newdirblk)((&oldadp->ad_newdirblk)->lh_first)) != NULL((void *)0)) { |
1581 | newdirblk = WK_NEWDIRBLK(wk)((struct newdirblk *)(wk)); |
1582 | WORKLIST_REMOVE(&newdirblk->db_list)do { (&newdirblk->db_list)->wk_state &= ~0x8000 ; do { if ((&newdirblk->db_list)->wk_list.le_next != ((void *)0)) (&newdirblk->db_list)->wk_list.le_next ->wk_list.le_prev = (&newdirblk->db_list)->wk_list .le_prev; *(&newdirblk->db_list)->wk_list.le_prev = (&newdirblk->db_list)->wk_list.le_next; ((&newdirblk ->db_list)->wk_list.le_prev) = ((void *)-1); ((&newdirblk ->db_list)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
1583 | if (LIST_FIRST(&oldadp->ad_newdirblk)((&oldadp->ad_newdirblk)->lh_first) != NULL((void *)0)) |
1584 | panic("allocdirect_merge: extra newdirblk"); |
1585 | WORKLIST_INSERT(&newadp->ad_newdirblk, &newdirblk->db_list)do { (&newdirblk->db_list)->wk_state |= 0x8000; do { if (((&newdirblk->db_list)->wk_list.le_next = (& newadp->ad_newdirblk)->lh_first) != ((void *)0)) (& newadp->ad_newdirblk)->lh_first->wk_list.le_prev = & (&newdirblk->db_list)->wk_list.le_next; (&newadp ->ad_newdirblk)->lh_first = (&newdirblk->db_list ); (&newdirblk->db_list)->wk_list.le_prev = &(& newadp->ad_newdirblk)->lh_first; } while (0); } while ( 0); |
1586 | } |
1587 | free_allocdirect(adphead, oldadp, 0); |
1588 | } |
1589 | |
1590 | /* |
1591 | * Allocate a new freefrag structure if needed. |
1592 | */ |
1593 | STATIC struct freefrag * |
1594 | newfreefrag(struct inode *ip, daddr_t blkno, long size) |
1595 | { |
1596 | struct freefrag *freefrag; |
1597 | struct fs *fs; |
1598 | |
1599 | if (blkno == 0) |
1600 | return (NULL((void *)0)); |
1601 | fs = ip->i_fsinode_u.fs; |
1602 | if (fragnum(fs, blkno)((blkno) & ((fs)->fs_frag - 1)) + numfrags(fs, size)((size) >> (fs)->fs_fshift) > fs->fs_frag) |
1603 | panic("newfreefrag: frag size"); |
1604 | freefrag = pool_get(&freefrag_pool, PR_WAITOK0x0001); |
1605 | freefrag->ff_list.wk_type = D_FREEFRAG7; |
1606 | freefrag->ff_stateff_list.wk_state = DIP(ip, uid)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_uid : (ip)->dinode_u.ffs2_din->di_uid) & ~ONWORKLIST0x8000; /* used below */ |
1607 | freefrag->ff_inum = ip->i_number; |
1608 | freefrag->ff_mnt = ITOV(ip)((ip)->i_vnode)->v_mount; |
1609 | freefrag->ff_devvp = ip->i_devvpi_ump->um_devvp; |
1610 | freefrag->ff_blkno = blkno; |
1611 | freefrag->ff_fragsize = size; |
1612 | return (freefrag); |
1613 | } |
1614 | |
1615 | /* |
1616 | * This workitem de-allocates fragments that were replaced during |
1617 | * file block allocation. |
1618 | */ |
1619 | STATIC void |
1620 | handle_workitem_freefrag(struct freefrag *freefrag) |
1621 | { |
1622 | struct inode tip; |
1623 | struct ufs1_dinode dtip1; |
1624 | |
1625 | tip.i_vnode = NULL((void *)0); |
1626 | tip.i_din1dinode_u.ffs1_din = &dtip1; |
1627 | tip.i_fsinode_u.fs = VFSTOUFS(freefrag->ff_mnt)((struct ufsmount *)((freefrag->ff_mnt)->mnt_data))->um_fsufsmount_u.fs; |
1628 | tip.i_ump = VFSTOUFS(freefrag->ff_mnt)((struct ufsmount *)((freefrag->ff_mnt)->mnt_data)); |
1629 | tip.i_dev = freefrag->ff_devvp->v_rdevv_un.vu_specinfo->si_rdev; |
1630 | tip.i_number = freefrag->ff_inum; |
1631 | tip.i_ffs1_uiddinode_u.ffs1_din->di_uid = freefrag->ff_stateff_list.wk_state & ~ONWORKLIST0x8000; /* set above */ |
1632 | ffs_blkfree(&tip, freefrag->ff_blkno, freefrag->ff_fragsize); |
1633 | pool_put(&freefrag_pool, freefrag); |
1634 | } |
1635 | |
1636 | /* |
1637 | * Indirect block allocation dependencies. |
1638 | * |
1639 | * The same dependencies that exist for a direct block also exist when |
1640 | * a new block is allocated and pointed to by an entry in a block of |
1641 | * indirect pointers. The undo/redo states described above are also |
1642 | * used here. Because an indirect block contains many pointers that |
1643 | * may have dependencies, a second copy of the entire in-memory indirect |
1644 | * block is kept. The buffer cache copy is always completely up-to-date. |
1645 | * The second copy, which is used only as a source for disk writes, |
1646 | * contains only the safe pointers (i.e., those that have no remaining |
1647 | * update dependencies). The second copy is freed when all pointers |
1648 | * are safe. The cache is not allowed to replace indirect blocks with |
1649 | * pending update dependencies. If a buffer containing an indirect |
1650 | * block with dependencies is written, these routines will mark it |
1651 | * dirty again. It can only be successfully written once all the |
1652 | * dependencies are removed. The ffs_fsync routine in conjunction with |
1653 | * softdep_sync_metadata work together to get all the dependencies |
1654 | * removed so that a file can be successfully written to disk. Three |
1655 | * procedures are used when setting up indirect block pointer |
1656 | * dependencies. The division is necessary because of the organization |
1657 | * of the "balloc" routine and because of the distinction between file |
1658 | * pages and file metadata blocks. |
1659 | */ |
1660 | |
1661 | /* |
1662 | * Allocate a new allocindir structure. |
1663 | */ |
1664 | /* inode for file being extended */ |
1665 | /* offset of pointer in indirect block */ |
1666 | /* disk block number being added */ |
1667 | /* previous block number, 0 if none */ |
1668 | STATIC struct allocindir * |
1669 | newallocindir(struct inode *ip, int ptrno, daddr_t newblkno, |
1670 | daddr_t oldblkno) |
1671 | { |
1672 | struct allocindir *aip; |
1673 | |
1674 | aip = pool_get(&allocindir_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
1675 | aip->ai_list.wk_type = D_ALLOCINDIR6; |
1676 | aip->ai_stateai_list.wk_state = ATTACHED0x0001; |
1677 | aip->ai_offset = ptrno; |
1678 | aip->ai_newblkno = newblkno; |
1679 | aip->ai_oldblkno = oldblkno; |
1680 | aip->ai_freefrag = newfreefrag(ip, oldblkno, ip->i_fsinode_u.fs->fs_bsize); |
1681 | return (aip); |
1682 | } |
1683 | |
1684 | /* |
1685 | * Called just before setting an indirect block pointer |
1686 | * to a newly allocated file page. |
1687 | */ |
1688 | /* inode for file being extended */ |
1689 | /* allocated block number within file */ |
1690 | /* buffer with indirect blk referencing page */ |
1691 | /* offset of pointer in indirect block */ |
1692 | /* disk block number being added */ |
1693 | /* previous block number, 0 if none */ |
1694 | /* buffer holding allocated page */ |
1695 | void |
1696 | softdep_setup_allocindir_page(struct inode *ip, daddr_t lbn, struct buf *bp, |
1697 | int ptrno, daddr_t newblkno, daddr_t oldblkno, struct buf *nbp) |
1698 | { |
1699 | struct allocindir *aip; |
1700 | struct pagedep *pagedep; |
1701 | |
1702 | aip = newallocindir(ip, ptrno, newblkno, oldblkno); |
1703 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1704 | /* |
1705 | * If we are allocating a directory page, then we must |
1706 | * allocate an associated pagedep to track additions and |
1707 | * deletions. |
1708 | */ |
1709 | if ((DIP(ip, mode)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_mode : (ip)->dinode_u.ffs2_din->di_mode) & IFMT0170000) == IFDIR0040000 && |
1710 | pagedep_lookup(ip, lbn, DEPALLOC0x0001, &pagedep) == 0) |
1711 | WORKLIST_INSERT(&nbp->b_dep, &pagedep->pd_list)do { (&pagedep->pd_list)->wk_state |= 0x8000; do { if (((&pagedep->pd_list)->wk_list.le_next = (&nbp ->b_dep)->lh_first) != ((void *)0)) (&nbp->b_dep )->lh_first->wk_list.le_prev = &(&pagedep->pd_list )->wk_list.le_next; (&nbp->b_dep)->lh_first = (& pagedep->pd_list); (&pagedep->pd_list)->wk_list. le_prev = &(&nbp->b_dep)->lh_first; } while (0) ; } while (0); |
1712 | if (nbp == NULL((void *)0)) { |
1713 | /* |
1714 | * XXXUBC - Yes, I know how to fix this, but not right now. |
1715 | */ |
1716 | panic("softdep_setup_allocindir_page: Bonk art in the head"); |
1717 | } |
1718 | WORKLIST_INSERT(&nbp->b_dep, &aip->ai_list)do { (&aip->ai_list)->wk_state |= 0x8000; do { if ( ((&aip->ai_list)->wk_list.le_next = (&nbp->b_dep )->lh_first) != ((void *)0)) (&nbp->b_dep)->lh_first ->wk_list.le_prev = &(&aip->ai_list)->wk_list .le_next; (&nbp->b_dep)->lh_first = (&aip->ai_list ); (&aip->ai_list)->wk_list.le_prev = &(&nbp ->b_dep)->lh_first; } while (0); } while (0); |
1719 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1720 | setup_allocindir_phase2(bp, ip, aip); |
1721 | } |
1722 | |
1723 | /* |
1724 | * Called just before setting an indirect block pointer to a |
1725 | * newly allocated indirect block. |
1726 | */ |
1727 | /* newly allocated indirect block */ |
1728 | /* inode for file being extended */ |
1729 | /* indirect block referencing allocated block */ |
1730 | /* offset of pointer in indirect block */ |
1731 | /* disk block number being added */ |
1732 | void |
1733 | softdep_setup_allocindir_meta(struct buf *nbp, struct inode *ip, |
1734 | struct buf *bp, int ptrno, daddr_t newblkno) |
1735 | { |
1736 | struct allocindir *aip; |
1737 | |
1738 | aip = newallocindir(ip, ptrno, newblkno, 0); |
1739 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1740 | WORKLIST_INSERT(&nbp->b_dep, &aip->ai_list)do { (&aip->ai_list)->wk_state |= 0x8000; do { if ( ((&aip->ai_list)->wk_list.le_next = (&nbp->b_dep )->lh_first) != ((void *)0)) (&nbp->b_dep)->lh_first ->wk_list.le_prev = &(&aip->ai_list)->wk_list .le_next; (&nbp->b_dep)->lh_first = (&aip->ai_list ); (&aip->ai_list)->wk_list.le_prev = &(&nbp ->b_dep)->lh_first; } while (0); } while (0); |
1741 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1742 | setup_allocindir_phase2(bp, ip, aip); |
1743 | } |
1744 | |
1745 | /* |
1746 | * Called to finish the allocation of the "aip" allocated |
1747 | * by one of the two routines above. |
1748 | */ |
1749 | /* in-memory copy of the indirect block */ |
1750 | /* inode for file being extended */ |
1751 | /* allocindir allocated by the above routines */ |
1752 | STATIC void |
1753 | setup_allocindir_phase2(struct buf *bp, struct inode *ip, |
1754 | struct allocindir *aip) |
1755 | { |
1756 | struct worklist *wk; |
1757 | struct indirdep *indirdep, *newindirdep; |
1758 | struct bmsafemap *bmsafemap; |
1759 | struct allocindir *oldaip; |
1760 | struct freefrag *freefrag; |
1761 | struct newblk *newblk; |
1762 | |
1763 | if (bp->b_lblkno >= 0) |
1764 | panic("setup_allocindir_phase2: not indir blk"); |
1765 | for (indirdep = NULL((void *)0), newindirdep = NULL((void *)0); ; ) { |
1766 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1767 | LIST_FOREACH(wk, &bp->b_dep, wk_list)for((wk) = ((&bp->b_dep)->lh_first); (wk)!= ((void * )0); (wk) = ((wk)->wk_list.le_next)) { |
1768 | if (wk->wk_type != D_INDIRDEP5) |
1769 | continue; |
1770 | indirdep = WK_INDIRDEP(wk)((struct indirdep *)(wk)); |
1771 | break; |
1772 | } |
1773 | if (indirdep == NULL((void *)0) && newindirdep) { |
1774 | indirdep = newindirdep; |
1775 | WORKLIST_INSERT(&bp->b_dep, &indirdep->ir_list)do { (&indirdep->ir_list)->wk_state |= 0x8000; do { if (((&indirdep->ir_list)->wk_list.le_next = (& bp->b_dep)->lh_first) != ((void *)0)) (&bp->b_dep )->lh_first->wk_list.le_prev = &(&indirdep-> ir_list)->wk_list.le_next; (&bp->b_dep)->lh_first = (&indirdep->ir_list); (&indirdep->ir_list)-> wk_list.le_prev = &(&bp->b_dep)->lh_first; } while (0); } while (0); |
1776 | newindirdep = NULL((void *)0); |
1777 | } |
1778 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1779 | if (indirdep) { |
1780 | if (newblk_lookup(ip->i_fsinode_u.fs, aip->ai_newblkno, 0, |
1781 | &newblk) == 0) |
1782 | panic("setup_allocindir: lost block"); |
1783 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1784 | if (newblk->nb_state == DEPCOMPLETE0x0008) { |
1785 | aip->ai_stateai_list.wk_state |= DEPCOMPLETE0x0008; |
1786 | aip->ai_buf = NULL((void *)0); |
1787 | } else { |
1788 | bmsafemap = newblk->nb_bmsafemap; |
1789 | aip->ai_buf = bmsafemap->sm_buf; |
1790 | LIST_REMOVE(newblk, nb_deps)do { if ((newblk)->nb_deps.le_next != ((void *)0)) (newblk )->nb_deps.le_next->nb_deps.le_prev = (newblk)->nb_deps .le_prev; *(newblk)->nb_deps.le_prev = (newblk)->nb_deps .le_next; ((newblk)->nb_deps.le_prev) = ((void *)-1); ((newblk )->nb_deps.le_next) = ((void *)-1); } while (0); |
1791 | LIST_INSERT_HEAD(&bmsafemap->sm_allocindirhd,do { if (((aip)->ai_deps.le_next = (&bmsafemap->sm_allocindirhd )->lh_first) != ((void *)0)) (&bmsafemap->sm_allocindirhd )->lh_first->ai_deps.le_prev = &(aip)->ai_deps.le_next ; (&bmsafemap->sm_allocindirhd)->lh_first = (aip); ( aip)->ai_deps.le_prev = &(&bmsafemap->sm_allocindirhd )->lh_first; } while (0) |
1792 | aip, ai_deps)do { if (((aip)->ai_deps.le_next = (&bmsafemap->sm_allocindirhd )->lh_first) != ((void *)0)) (&bmsafemap->sm_allocindirhd )->lh_first->ai_deps.le_prev = &(aip)->ai_deps.le_next ; (&bmsafemap->sm_allocindirhd)->lh_first = (aip); ( aip)->ai_deps.le_prev = &(&bmsafemap->sm_allocindirhd )->lh_first; } while (0); |
1793 | } |
1794 | LIST_REMOVE(newblk, nb_hash)do { if ((newblk)->nb_hash.le_next != ((void *)0)) (newblk )->nb_hash.le_next->nb_hash.le_prev = (newblk)->nb_hash .le_prev; *(newblk)->nb_hash.le_prev = (newblk)->nb_hash .le_next; ((newblk)->nb_hash.le_prev) = ((void *)-1); ((newblk )->nb_hash.le_next) = ((void *)-1); } while (0); |
1795 | pool_put(&newblk_pool, newblk); |
1796 | aip->ai_indirdep = indirdep; |
1797 | /* |
1798 | * Check to see if there is an existing dependency |
1799 | * for this block. If there is, merge the old |
1800 | * dependency into the new one. |
1801 | */ |
1802 | if (aip->ai_oldblkno == 0) |
1803 | oldaip = NULL((void *)0); |
1804 | else |
1805 | |
1806 | LIST_FOREACH(oldaip, &indirdep->ir_deplisthd, ai_next)for((oldaip) = ((&indirdep->ir_deplisthd)->lh_first ); (oldaip)!= ((void *)0); (oldaip) = ((oldaip)->ai_next.le_next )) |
1807 | if (oldaip->ai_offset == aip->ai_offset) |
1808 | break; |
1809 | freefrag = NULL((void *)0); |
1810 | if (oldaip != NULL((void *)0)) { |
1811 | if (oldaip->ai_newblkno != aip->ai_oldblkno) { |
1812 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1813 | panic("setup_allocindir_phase2: blkno"); |
1814 | } |
1815 | aip->ai_oldblkno = oldaip->ai_oldblkno; |
1816 | freefrag = aip->ai_freefrag; |
1817 | aip->ai_freefrag = oldaip->ai_freefrag; |
1818 | oldaip->ai_freefrag = NULL((void *)0); |
1819 | free_allocindir(oldaip, NULL((void *)0)); |
1820 | } |
1821 | LIST_INSERT_HEAD(&indirdep->ir_deplisthd, aip, ai_next)do { if (((aip)->ai_next.le_next = (&indirdep->ir_deplisthd )->lh_first) != ((void *)0)) (&indirdep->ir_deplisthd )->lh_first->ai_next.le_prev = &(aip)->ai_next.le_next ; (&indirdep->ir_deplisthd)->lh_first = (aip); (aip )->ai_next.le_prev = &(&indirdep->ir_deplisthd) ->lh_first; } while (0); |
1822 | if (ip->i_ump->um_fstype == UM_UFS11) |
1823 | ((int32_t *)indirdep->ir_savebp->b_data) |
1824 | [aip->ai_offset] = aip->ai_oldblkno; |
1825 | else |
1826 | ((int64_t *)indirdep->ir_savebp->b_data) |
1827 | [aip->ai_offset] = aip->ai_oldblkno; |
1828 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1829 | if (freefrag != NULL((void *)0)) |
1830 | handle_workitem_freefrag(freefrag); |
1831 | } |
1832 | if (newindirdep) { |
1833 | if (indirdep->ir_savebp != NULL((void *)0)) |
1834 | brelse(newindirdep->ir_savebp); |
1835 | WORKITEM_FREE(newindirdep, D_INDIRDEP)softdep_freequeue_add((struct worklist *)newindirdep); |
1836 | } |
1837 | if (indirdep) |
1838 | break; |
1839 | newindirdep = pool_get(&indirdep_pool, PR_WAITOK0x0001); |
1840 | newindirdep->ir_list.wk_type = D_INDIRDEP5; |
1841 | newindirdep->ir_stateir_list.wk_state = ATTACHED0x0001; |
1842 | if (ip->i_ump->um_fstype == UM_UFS11) |
1843 | newindirdep->ir_stateir_list.wk_state |= UFS1FMT0x2000; |
1844 | LIST_INIT(&newindirdep->ir_deplisthd)do { ((&newindirdep->ir_deplisthd)->lh_first) = ((void *)0); } while (0); |
1845 | LIST_INIT(&newindirdep->ir_donehd)do { ((&newindirdep->ir_donehd)->lh_first) = ((void *)0); } while (0); |
1846 | if (bp->b_blkno == bp->b_lblkno) { |
1847 | VOP_BMAP(bp->b_vp, bp->b_lblkno, NULL((void *)0), &bp->b_blkno, |
1848 | NULL((void *)0)); |
1849 | } |
1850 | newindirdep->ir_savebp = |
1851 | getblk(ip->i_devvpi_ump->um_devvp, bp->b_blkno, bp->b_bcount, 0, INFSLP0xffffffffffffffffULL); |
1852 | #if 0 |
1853 | BUF_KERNPROC(newindirdep->ir_savebp); |
1854 | #endif |
1855 | memcpy(newindirdep->ir_savebp->b_data, bp->b_data, bp->b_bcount)__builtin_memcpy((newindirdep->ir_savebp->b_data), (bp-> b_data), (bp->b_bcount)); |
1856 | } |
1857 | } |
1858 | |
1859 | /* |
1860 | * Block de-allocation dependencies. |
1861 | * |
1862 | * When blocks are de-allocated, the on-disk pointers must be nullified before |
1863 | * the blocks are made available for use by other files. (The true |
1864 | * requirement is that old pointers must be nullified before new on-disk |
1865 | * pointers are set. We chose this slightly more stringent requirement to |
1866 | * reduce complexity.) Our implementation handles this dependency by updating |
1867 | * the inode (or indirect block) appropriately but delaying the actual block |
1868 | * de-allocation (i.e., freemap and free space count manipulation) until |
1869 | * after the updated versions reach stable storage. After the disk is |
1870 | * updated, the blocks can be safely de-allocated whenever it is convenient. |
1871 | * This implementation handles only the common case of reducing a file's |
1872 | * length to zero. Other cases are handled by the conventional synchronous |
1873 | * write approach. |
1874 | * |
1875 | * The ffs implementation with which we worked double-checks |
1876 | * the state of the block pointers and file size as it reduces |
1877 | * a file's length. Some of this code is replicated here in our |
1878 | * soft updates implementation. The freeblks->fb_chkcnt field is |
1879 | * used to transfer a part of this information to the procedure |
1880 | * that eventually de-allocates the blocks. |
1881 | * |
1882 | * This routine should be called from the routine that shortens |
1883 | * a file's length, before the inode's size or block pointers |
1884 | * are modified. It will save the block pointer information for |
1885 | * later release and zero the inode so that the calling routine |
1886 | * can release it. |
1887 | */ |
1888 | /* The inode whose length is to be reduced */ |
1889 | /* The new length for the file */ |
1890 | void |
1891 | softdep_setup_freeblocks(struct inode *ip, off_t length) |
1892 | { |
1893 | struct freeblks *freeblks; |
1894 | struct inodedep *inodedep; |
1895 | struct allocdirect *adp; |
1896 | struct vnode *vp; |
1897 | struct buf *bp; |
1898 | struct fs *fs; |
1899 | int i, delay, error; |
1900 | |
1901 | fs = ip->i_fsinode_u.fs; |
1902 | if (length != 0) |
1903 | panic("softdep_setup_freeblocks: non-zero length"); |
1904 | freeblks = pool_get(&freeblks_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
1905 | freeblks->fb_list.wk_type = D_FREEBLKS8; |
1906 | freeblks->fb_statefb_list.wk_state = ATTACHED0x0001; |
1907 | freeblks->fb_uid = DIP(ip, uid)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_uid : (ip)->dinode_u.ffs2_din->di_uid); |
1908 | freeblks->fb_previousinum = ip->i_number; |
1909 | freeblks->fb_devvp = ip->i_devvpi_ump->um_devvp; |
1910 | freeblks->fb_mnt = ITOV(ip)((ip)->i_vnode)->v_mount; |
1911 | freeblks->fb_oldsize = DIP(ip, size)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_size : (ip)->dinode_u.ffs2_din->di_size); |
1912 | freeblks->fb_newsize = length; |
1913 | freeblks->fb_chkcnt = DIP(ip, blocks)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_blocks : (ip)->dinode_u.ffs2_din->di_blocks); |
1914 | |
1915 | for (i = 0; i < NDADDR12; i++) { |
1916 | freeblks->fb_dblks[i] = DIP(ip, db[i])(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_db[i] : (ip)->dinode_u.ffs2_din->di_db[i]); |
1917 | DIP_ASSIGN(ip, db[i], 0)do { if ((ip)->i_ump->um_fstype == 1) (ip)->dinode_u .ffs1_din->di_db[i] = (0); else (ip)->dinode_u.ffs2_din ->di_db[i] = (0); } while (0); |
1918 | } |
1919 | |
1920 | for (i = 0; i < NIADDR3; i++) { |
1921 | freeblks->fb_iblks[i] = DIP(ip, ib[i])(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_ib[i] : (ip)->dinode_u.ffs2_din->di_ib[i]); |
1922 | DIP_ASSIGN(ip, ib[i], 0)do { if ((ip)->i_ump->um_fstype == 1) (ip)->dinode_u .ffs1_din->di_ib[i] = (0); else (ip)->dinode_u.ffs2_din ->di_ib[i] = (0); } while (0); |
1923 | } |
1924 | |
1925 | DIP_ASSIGN(ip, blocks, 0)do { if ((ip)->i_ump->um_fstype == 1) (ip)->dinode_u .ffs1_din->di_blocks = (0); else (ip)->dinode_u.ffs2_din ->di_blocks = (0); } while (0); |
1926 | DIP_ASSIGN(ip, size, 0)do { if ((ip)->i_ump->um_fstype == 1) (ip)->dinode_u .ffs1_din->di_size = (0); else (ip)->dinode_u.ffs2_din-> di_size = (0); } while (0); |
1927 | |
1928 | /* |
1929 | * Push the zero'ed inode to to its disk buffer so that we are free |
1930 | * to delete its dependencies below. Once the dependencies are gone |
1931 | * the buffer can be safely released. |
1932 | */ |
1933 | if ((error = bread(ip->i_devvpi_ump->um_devvp, |
1934 | fsbtodb(fs, ino_to_fsba(fs, ip->i_number))((((daddr_t)(((((daddr_t)(fs)->fs_fpg * (((ip->i_number ) / (fs)->fs_ipg))) + (fs)->fs_cgoffset * ((((ip->i_number ) / (fs)->fs_ipg)) & ~((fs)->fs_cgmask))) + (fs)-> fs_iblkno) + ((((((ip->i_number) % (fs)->fs_ipg) / ((fs )->fs_inopb))) << ((fs))->fs_fragshift))))) << (fs)->fs_fsbtodb), |
1935 | (int)fs->fs_bsize, &bp)) != 0) |
1936 | softdep_error("softdep_setup_freeblocks", error); |
1937 | |
1938 | if (ip->i_ump->um_fstype == UM_UFS11) |
1939 | *((struct ufs1_dinode *) bp->b_data + |
1940 | ino_to_fsbo(fs, ip->i_number)((ip->i_number) % ((fs)->fs_inopb))) = *ip->i_din1dinode_u.ffs1_din; |
1941 | else |
1942 | *((struct ufs2_dinode *) bp->b_data + |
1943 | ino_to_fsbo(fs, ip->i_number)((ip->i_number) % ((fs)->fs_inopb))) = *ip->i_din2dinode_u.ffs2_din; |
1944 | |
1945 | /* |
1946 | * Find and eliminate any inode dependencies. |
1947 | */ |
1948 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1949 | (void) inodedep_lookup(fs, ip->i_number, DEPALLOC0x0001, &inodedep); |
1950 | if ((inodedep->id_stateid_list.wk_state & IOSTARTED0x0200) != 0) { |
1951 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1952 | panic("softdep_setup_freeblocks: inode busy"); |
1953 | } |
1954 | /* |
1955 | * Add the freeblks structure to the list of operations that |
1956 | * must await the zero'ed inode being written to disk. If we |
1957 | * still have a bitmap dependency (delay == 0), then the inode |
1958 | * has never been written to disk, so we can process the |
1959 | * freeblks below once we have deleted the dependencies. |
1960 | */ |
1961 | delay = (inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008); |
1962 | if (delay) |
1963 | WORKLIST_INSERT(&inodedep->id_bufwait, &freeblks->fb_list)do { (&freeblks->fb_list)->wk_state |= 0x8000; do { if (((&freeblks->fb_list)->wk_list.le_next = (& inodedep->id_bufwait)->lh_first) != ((void *)0)) (& inodedep->id_bufwait)->lh_first->wk_list.le_prev = & (&freeblks->fb_list)->wk_list.le_next; (&inodedep ->id_bufwait)->lh_first = (&freeblks->fb_list); ( &freeblks->fb_list)->wk_list.le_prev = &(&inodedep ->id_bufwait)->lh_first; } while (0); } while (0); |
1964 | /* |
1965 | * Because the file length has been truncated to zero, any |
1966 | * pending block allocation dependency structures associated |
1967 | * with this inode are obsolete and can simply be de-allocated. |
1968 | * We must first merge the two dependency lists to get rid of |
1969 | * any duplicate freefrag structures, then purge the merged list. |
1970 | * If we still have a bitmap dependency, then the inode has never |
1971 | * been written to disk, so we can free any fragments without delay. |
1972 | */ |
1973 | merge_inode_lists(inodedep); |
1974 | while ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first)) != NULL((void *)0)) |
1975 | free_allocdirect(&inodedep->id_inoupdt, adp, delay); |
1976 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1977 | bdwrite(bp); |
1978 | /* |
1979 | * We must wait for any I/O in progress to finish so that |
1980 | * all potential buffers on the dirty list will be visible. |
1981 | * Once they are all there, walk the list and get rid of |
1982 | * any dependencies. |
1983 | */ |
1984 | vp = ITOV(ip)((ip)->i_vnode); |
1985 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1986 | drain_output(vp, 1); |
1987 | while ((bp = LIST_FIRST(&vp->v_dirtyblkhd)((&vp->v_dirtyblkhd)->lh_first))) { |
1988 | if (getdirtybuf(bp, MNT_WAIT1) <= 0) |
1989 | break; |
1990 | (void) inodedep_lookup(fs, ip->i_number, 0, &inodedep); |
1991 | deallocate_dependencies(bp, inodedep); |
1992 | bp->b_flags |= B_INVAL0x00000800 | B_NOCACHE0x00001000; |
1993 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
1994 | brelse(bp); |
1995 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
1996 | } |
1997 | if (inodedep_lookup(fs, ip->i_number, 0, &inodedep) != 0) |
1998 | (void) free_inodedep(inodedep); |
1999 | |
2000 | if (delay) { |
2001 | freeblks->fb_statefb_list.wk_state |= DEPCOMPLETE0x0008; |
2002 | /* |
2003 | * If the inode with zeroed block pointers is now on disk we |
2004 | * can start freeing blocks. Add freeblks to the worklist |
2005 | * instead of calling handle_workitem_freeblocks() directly as |
2006 | * it is more likely that additional IO is needed to complete |
2007 | * the request than in the !delay case. |
2008 | */ |
2009 | if ((freeblks->fb_statefb_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) |
2010 | add_to_worklist(&freeblks->fb_list); |
2011 | } |
2012 | |
2013 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2014 | /* |
2015 | * If the inode has never been written to disk (delay == 0), |
2016 | * then we can process the freeblks now that we have deleted |
2017 | * the dependencies. |
2018 | */ |
2019 | if (!delay) |
2020 | handle_workitem_freeblocks(freeblks); |
2021 | } |
2022 | |
2023 | /* |
2024 | * Reclaim any dependency structures from a buffer that is about to |
2025 | * be reallocated to a new vnode. The buffer must be locked, thus, |
2026 | * no I/O completion operations can occur while we are manipulating |
2027 | * its associated dependencies. The mutex is held so that other I/O's |
2028 | * associated with related dependencies do not occur. |
2029 | */ |
2030 | STATIC void |
2031 | deallocate_dependencies(struct buf *bp, struct inodedep *inodedep) |
2032 | { |
2033 | struct worklist *wk; |
2034 | struct indirdep *indirdep; |
2035 | struct allocindir *aip; |
2036 | struct pagedep *pagedep; |
2037 | struct dirrem *dirrem; |
2038 | struct diradd *dap; |
2039 | int i; |
2040 | |
2041 | while ((wk = LIST_FIRST(&bp->b_dep)((&bp->b_dep)->lh_first)) != NULL((void *)0)) { |
2042 | switch (wk->wk_type) { |
2043 | |
2044 | case D_INDIRDEP5: |
2045 | indirdep = WK_INDIRDEP(wk)((struct indirdep *)(wk)); |
2046 | /* |
2047 | * None of the indirect pointers will ever be visible, |
2048 | * so they can simply be tossed. GOINGAWAY ensures |
2049 | * that allocated pointers will be saved in the buffer |
2050 | * cache until they are freed. Note that they will |
2051 | * only be able to be found by their physical address |
2052 | * since the inode mapping the logical address will |
2053 | * be gone. The save buffer used for the safe copy |
2054 | * was allocated in setup_allocindir_phase2 using |
2055 | * the physical address so it could be used for this |
2056 | * purpose. Hence we swap the safe copy with the real |
2057 | * copy, allowing the safe copy to be freed and holding |
2058 | * on to the real copy for later use in indir_trunc. |
2059 | */ |
2060 | if (indirdep->ir_stateir_list.wk_state & GOINGAWAY0x0100) { |
2061 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2062 | panic("deallocate_dependencies: already gone"); |
2063 | } |
2064 | indirdep->ir_stateir_list.wk_state |= GOINGAWAY0x0100; |
2065 | while ((aip = LIST_FIRST(&indirdep->ir_deplisthd)((&indirdep->ir_deplisthd)->lh_first))) |
2066 | free_allocindir(aip, inodedep); |
2067 | if (bp->b_lblkno >= 0 || |
2068 | bp->b_blkno != indirdep->ir_savebp->b_lblkno) { |
2069 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2070 | panic("deallocate_dependencies: not indir"); |
2071 | } |
2072 | memcpy(indirdep->ir_savebp->b_data, bp->b_data,__builtin_memcpy((indirdep->ir_savebp->b_data), (bp-> b_data), (bp->b_bcount)) |
2073 | bp->b_bcount)__builtin_memcpy((indirdep->ir_savebp->b_data), (bp-> b_data), (bp->b_bcount)); |
2074 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
2075 | WORKLIST_INSERT(&indirdep->ir_savebp->b_dep, wk)do { (wk)->wk_state |= 0x8000; do { if (((wk)->wk_list. le_next = (&indirdep->ir_savebp->b_dep)->lh_first ) != ((void *)0)) (&indirdep->ir_savebp->b_dep)-> lh_first->wk_list.le_prev = &(wk)->wk_list.le_next; (&indirdep->ir_savebp->b_dep)->lh_first = (wk); (wk)->wk_list.le_prev = &(&indirdep->ir_savebp ->b_dep)->lh_first; } while (0); } while (0); |
2076 | continue; |
2077 | |
2078 | case D_PAGEDEP0: |
2079 | pagedep = WK_PAGEDEP(wk)((struct pagedep *)(wk)); |
2080 | /* |
2081 | * None of the directory additions will ever be |
2082 | * visible, so they can simply be tossed. |
2083 | */ |
2084 | for (i = 0; i < DAHASHSZ6; i++) |
2085 | while ((dap = |
2086 | LIST_FIRST(&pagedep->pd_diraddhd[i])((&pagedep->pd_diraddhd[i])->lh_first))) |
2087 | free_diradd(dap); |
2088 | while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)((&pagedep->pd_pendinghd)->lh_first))) |
2089 | free_diradd(dap); |
2090 | /* |
2091 | * Copy any directory remove dependencies to the list |
2092 | * to be processed after the zero'ed inode is written. |
2093 | * If the inode has already been written, then they |
2094 | * can be dumped directly onto the work list. |
2095 | */ |
2096 | while ((dirrem = LIST_FIRST(&pagedep->pd_dirremhd)((&pagedep->pd_dirremhd)->lh_first))) { |
2097 | LIST_REMOVE(dirrem, dm_next)do { if ((dirrem)->dm_next.le_next != ((void *)0)) (dirrem )->dm_next.le_next->dm_next.le_prev = (dirrem)->dm_next .le_prev; *(dirrem)->dm_next.le_prev = (dirrem)->dm_next .le_next; ((dirrem)->dm_next.le_prev) = ((void *)-1); ((dirrem )->dm_next.le_next) = ((void *)-1); } while (0); |
2098 | dirrem->dm_dirinumdm_un.dmu_dirinum = pagedep->pd_ino; |
2099 | if (inodedep == NULL((void *)0) || |
2100 | (inodedep->id_stateid_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == |
2101 | ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) |
2102 | add_to_worklist(&dirrem->dm_list); |
2103 | else |
2104 | WORKLIST_INSERT(&inodedep->id_bufwait,do { (&dirrem->dm_list)->wk_state |= 0x8000; do { if (((&dirrem->dm_list)->wk_list.le_next = (&inodedep ->id_bufwait)->lh_first) != ((void *)0)) (&inodedep ->id_bufwait)->lh_first->wk_list.le_prev = &(& dirrem->dm_list)->wk_list.le_next; (&inodedep->id_bufwait )->lh_first = (&dirrem->dm_list); (&dirrem-> dm_list)->wk_list.le_prev = &(&inodedep->id_bufwait )->lh_first; } while (0); } while (0) |
2105 | &dirrem->dm_list)do { (&dirrem->dm_list)->wk_state |= 0x8000; do { if (((&dirrem->dm_list)->wk_list.le_next = (&inodedep ->id_bufwait)->lh_first) != ((void *)0)) (&inodedep ->id_bufwait)->lh_first->wk_list.le_prev = &(& dirrem->dm_list)->wk_list.le_next; (&inodedep->id_bufwait )->lh_first = (&dirrem->dm_list); (&dirrem-> dm_list)->wk_list.le_prev = &(&inodedep->id_bufwait )->lh_first; } while (0); } while (0); |
2106 | } |
2107 | if ((pagedep->pd_statepd_list.wk_state & NEWBLOCK0x0800) != 0) { |
2108 | LIST_FOREACH(wk, &inodedep->id_bufwait, wk_list)for((wk) = ((&inodedep->id_bufwait)->lh_first); (wk )!= ((void *)0); (wk) = ((wk)->wk_list.le_next)) |
2109 | if (wk->wk_type == D_NEWDIRBLK13 && |
2110 | WK_NEWDIRBLK(wk)((struct newdirblk *)(wk))->db_pagedep == |
2111 | pagedep) |
2112 | break; |
2113 | if (wk != NULL((void *)0)) { |
2114 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
2115 | free_newdirblk(WK_NEWDIRBLK(wk)((struct newdirblk *)(wk))); |
2116 | } else { |
2117 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2118 | panic("deallocate_dependencies: " |
2119 | "lost pagedep"); |
2120 | } |
2121 | } |
2122 | WORKLIST_REMOVE(&pagedep->pd_list)do { (&pagedep->pd_list)->wk_state &= ~0x8000; do { if ((&pagedep->pd_list)->wk_list.le_next != ((void *)0)) (&pagedep->pd_list)->wk_list.le_next->wk_list .le_prev = (&pagedep->pd_list)->wk_list.le_prev; *( &pagedep->pd_list)->wk_list.le_prev = (&pagedep ->pd_list)->wk_list.le_next; ((&pagedep->pd_list )->wk_list.le_prev) = ((void *)-1); ((&pagedep->pd_list )->wk_list.le_next) = ((void *)-1); } while (0); } while ( 0); |
2123 | LIST_REMOVE(pagedep, pd_hash)do { if ((pagedep)->pd_hash.le_next != ((void *)0)) (pagedep )->pd_hash.le_next->pd_hash.le_prev = (pagedep)->pd_hash .le_prev; *(pagedep)->pd_hash.le_prev = (pagedep)->pd_hash .le_next; ((pagedep)->pd_hash.le_prev) = ((void *)-1); ((pagedep )->pd_hash.le_next) = ((void *)-1); } while (0); |
2124 | WORKITEM_FREE(pagedep, D_PAGEDEP)softdep_freequeue_add((struct worklist *)pagedep); |
2125 | continue; |
2126 | |
2127 | case D_ALLOCINDIR6: |
2128 | free_allocindir(WK_ALLOCINDIR(wk)((struct allocindir *)(wk)), inodedep); |
2129 | continue; |
2130 | |
2131 | case D_ALLOCDIRECT4: |
2132 | case D_INODEDEP1: |
2133 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2134 | panic("deallocate_dependencies: Unexpected type %s", |
2135 | TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
2136 | /* NOTREACHED */ |
2137 | |
2138 | default: |
2139 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2140 | panic("deallocate_dependencies: Unknown type %s", |
2141 | TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
2142 | /* NOTREACHED */ |
2143 | } |
2144 | } |
2145 | } |
2146 | |
2147 | /* |
2148 | * Free an allocdirect. Generate a new freefrag work request if appropriate. |
2149 | * This routine must be called with splbio interrupts blocked. |
2150 | */ |
2151 | STATIC void |
2152 | free_allocdirect(struct allocdirectlst *adphead, struct allocdirect *adp, |
2153 | int delay) |
2154 | { |
2155 | struct newdirblk *newdirblk; |
2156 | struct worklist *wk; |
2157 | |
2158 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
2159 | |
2160 | #ifdef DEBUG |
2161 | if (lk.lkt_held == -1) |
2162 | panic("free_allocdirect: lock not held"); |
2163 | #endif |
2164 | if ((adp->ad_statead_list.wk_state & DEPCOMPLETE0x0008) == 0) |
2165 | LIST_REMOVE(adp, ad_deps)do { if ((adp)->ad_deps.le_next != ((void *)0)) (adp)-> ad_deps.le_next->ad_deps.le_prev = (adp)->ad_deps.le_prev ; *(adp)->ad_deps.le_prev = (adp)->ad_deps.le_next; ((adp )->ad_deps.le_prev) = ((void *)-1); ((adp)->ad_deps.le_next ) = ((void *)-1); } while (0); |
2166 | TAILQ_REMOVE(adphead, adp, ad_next)do { if (((adp)->ad_next.tqe_next) != ((void *)0)) (adp)-> ad_next.tqe_next->ad_next.tqe_prev = (adp)->ad_next.tqe_prev ; else (adphead)->tqh_last = (adp)->ad_next.tqe_prev; * (adp)->ad_next.tqe_prev = (adp)->ad_next.tqe_next; ((adp )->ad_next.tqe_prev) = ((void *)-1); ((adp)->ad_next.tqe_next ) = ((void *)-1); } while (0); |
2167 | if ((adp->ad_statead_list.wk_state & COMPLETE0x0004) == 0) |
2168 | WORKLIST_REMOVE(&adp->ad_list)do { (&adp->ad_list)->wk_state &= ~0x8000; do { if ((&adp->ad_list)->wk_list.le_next != ((void *)0 )) (&adp->ad_list)->wk_list.le_next->wk_list.le_prev = (&adp->ad_list)->wk_list.le_prev; *(&adp-> ad_list)->wk_list.le_prev = (&adp->ad_list)->wk_list .le_next; ((&adp->ad_list)->wk_list.le_prev) = ((void *)-1); ((&adp->ad_list)->wk_list.le_next) = ((void *)-1); } while (0); } while (0); |
2169 | if (adp->ad_freefrag != NULL((void *)0)) { |
2170 | if (delay) |
2171 | WORKLIST_INSERT(&adp->ad_inodedep->id_bufwait,do { (&adp->ad_freefrag->ff_list)->wk_state |= 0x8000 ; do { if (((&adp->ad_freefrag->ff_list)->wk_list .le_next = (&adp->ad_inodedep->id_bufwait)->lh_first ) != ((void *)0)) (&adp->ad_inodedep->id_bufwait)-> lh_first->wk_list.le_prev = &(&adp->ad_freefrag ->ff_list)->wk_list.le_next; (&adp->ad_inodedep-> id_bufwait)->lh_first = (&adp->ad_freefrag->ff_list ); (&adp->ad_freefrag->ff_list)->wk_list.le_prev = &(&adp->ad_inodedep->id_bufwait)->lh_first ; } while (0); } while (0) |
2172 | &adp->ad_freefrag->ff_list)do { (&adp->ad_freefrag->ff_list)->wk_state |= 0x8000 ; do { if (((&adp->ad_freefrag->ff_list)->wk_list .le_next = (&adp->ad_inodedep->id_bufwait)->lh_first ) != ((void *)0)) (&adp->ad_inodedep->id_bufwait)-> lh_first->wk_list.le_prev = &(&adp->ad_freefrag ->ff_list)->wk_list.le_next; (&adp->ad_inodedep-> id_bufwait)->lh_first = (&adp->ad_freefrag->ff_list ); (&adp->ad_freefrag->ff_list)->wk_list.le_prev = &(&adp->ad_inodedep->id_bufwait)->lh_first ; } while (0); } while (0); |
2173 | else |
2174 | add_to_worklist(&adp->ad_freefrag->ff_list); |
2175 | } |
2176 | if ((wk = LIST_FIRST(&adp->ad_newdirblk)((&adp->ad_newdirblk)->lh_first)) != NULL((void *)0)) { |
2177 | newdirblk = WK_NEWDIRBLK(wk)((struct newdirblk *)(wk)); |
2178 | WORKLIST_REMOVE(&newdirblk->db_list)do { (&newdirblk->db_list)->wk_state &= ~0x8000 ; do { if ((&newdirblk->db_list)->wk_list.le_next != ((void *)0)) (&newdirblk->db_list)->wk_list.le_next ->wk_list.le_prev = (&newdirblk->db_list)->wk_list .le_prev; *(&newdirblk->db_list)->wk_list.le_prev = (&newdirblk->db_list)->wk_list.le_next; ((&newdirblk ->db_list)->wk_list.le_prev) = ((void *)-1); ((&newdirblk ->db_list)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
2179 | if (LIST_FIRST(&adp->ad_newdirblk)((&adp->ad_newdirblk)->lh_first) != NULL((void *)0)) |
2180 | panic("free_allocdirect: extra newdirblk"); |
2181 | if (delay) |
2182 | WORKLIST_INSERT(&adp->ad_inodedep->id_bufwait,do { (&newdirblk->db_list)->wk_state |= 0x8000; do { if (((&newdirblk->db_list)->wk_list.le_next = (& adp->ad_inodedep->id_bufwait)->lh_first) != ((void * )0)) (&adp->ad_inodedep->id_bufwait)->lh_first-> wk_list.le_prev = &(&newdirblk->db_list)->wk_list .le_next; (&adp->ad_inodedep->id_bufwait)->lh_first = (&newdirblk->db_list); (&newdirblk->db_list) ->wk_list.le_prev = &(&adp->ad_inodedep->id_bufwait )->lh_first; } while (0); } while (0) |
2183 | &newdirblk->db_list)do { (&newdirblk->db_list)->wk_state |= 0x8000; do { if (((&newdirblk->db_list)->wk_list.le_next = (& adp->ad_inodedep->id_bufwait)->lh_first) != ((void * )0)) (&adp->ad_inodedep->id_bufwait)->lh_first-> wk_list.le_prev = &(&newdirblk->db_list)->wk_list .le_next; (&adp->ad_inodedep->id_bufwait)->lh_first = (&newdirblk->db_list); (&newdirblk->db_list) ->wk_list.le_prev = &(&adp->ad_inodedep->id_bufwait )->lh_first; } while (0); } while (0); |
2184 | else |
2185 | free_newdirblk(newdirblk); |
2186 | } |
2187 | WORKITEM_FREE(adp, D_ALLOCDIRECT)softdep_freequeue_add((struct worklist *)adp); |
2188 | } |
2189 | |
2190 | /* |
2191 | * Free a newdirblk. Clear the NEWBLOCK flag on its associated pagedep. |
2192 | * This routine must be called with splbio interrupts blocked. |
2193 | */ |
2194 | void |
2195 | free_newdirblk(struct newdirblk *newdirblk) |
2196 | { |
2197 | struct pagedep *pagedep; |
2198 | struct diradd *dap; |
2199 | int i; |
2200 | |
2201 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
2202 | |
2203 | #ifdef DEBUG |
2204 | if (lk.lkt_held == -1) |
2205 | panic("free_newdirblk: lock not held"); |
2206 | #endif |
2207 | /* |
2208 | * If the pagedep is still linked onto the directory buffer |
2209 | * dependency chain, then some of the entries on the |
2210 | * pd_pendinghd list may not be committed to disk yet. In |
2211 | * this case, we will simply clear the NEWBLOCK flag and |
2212 | * let the pd_pendinghd list be processed when the pagedep |
2213 | * is next written. If the pagedep is no longer on the buffer |
2214 | * dependency chain, then all the entries on the pd_pending |
2215 | * list are committed to disk and we can free them here. |
2216 | */ |
2217 | pagedep = newdirblk->db_pagedep; |
2218 | pagedep->pd_statepd_list.wk_state &= ~NEWBLOCK0x0800; |
2219 | if ((pagedep->pd_statepd_list.wk_state & ONWORKLIST0x8000) == 0) |
2220 | while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)((&pagedep->pd_pendinghd)->lh_first)) != NULL((void *)0)) |
2221 | free_diradd(dap); |
2222 | /* |
2223 | * If no dependencies remain, the pagedep will be freed. |
2224 | */ |
2225 | for (i = 0; i < DAHASHSZ6; i++) |
2226 | if (LIST_FIRST(&pagedep->pd_diraddhd[i])((&pagedep->pd_diraddhd[i])->lh_first) != NULL((void *)0)) |
2227 | break; |
2228 | if (i == DAHASHSZ6 && (pagedep->pd_statepd_list.wk_state & ONWORKLIST0x8000) == 0) { |
2229 | LIST_REMOVE(pagedep, pd_hash)do { if ((pagedep)->pd_hash.le_next != ((void *)0)) (pagedep )->pd_hash.le_next->pd_hash.le_prev = (pagedep)->pd_hash .le_prev; *(pagedep)->pd_hash.le_prev = (pagedep)->pd_hash .le_next; ((pagedep)->pd_hash.le_prev) = ((void *)-1); ((pagedep )->pd_hash.le_next) = ((void *)-1); } while (0); |
2230 | WORKITEM_FREE(pagedep, D_PAGEDEP)softdep_freequeue_add((struct worklist *)pagedep); |
2231 | } |
2232 | WORKITEM_FREE(newdirblk, D_NEWDIRBLK)softdep_freequeue_add((struct worklist *)newdirblk); |
2233 | } |
2234 | |
2235 | /* |
2236 | * Prepare an inode to be freed. The actual free operation is not |
2237 | * done until the zero'ed inode has been written to disk. |
2238 | */ |
2239 | void |
2240 | softdep_freefile(struct vnode *pvp, ufsino_t ino, mode_t mode) |
2241 | { |
2242 | struct inode *ip = VTOI(pvp)((struct inode *)(pvp)->v_data); |
2243 | struct inodedep *inodedep; |
2244 | struct freefile *freefile; |
2245 | |
2246 | /* |
2247 | * This sets up the inode de-allocation dependency. |
2248 | */ |
2249 | freefile = pool_get(&freefile_pool, PR_WAITOK0x0001); |
2250 | freefile->fx_list.wk_type = D_FREEFILE9; |
2251 | freefile->fx_list.wk_state = 0; |
2252 | freefile->fx_mode = mode; |
2253 | freefile->fx_oldinum = ino; |
2254 | freefile->fx_devvp = ip->i_devvpi_ump->um_devvp; |
2255 | freefile->fx_mnt = ITOV(ip)((ip)->i_vnode)->v_mount; |
2256 | |
2257 | /* |
2258 | * If the inodedep does not exist, then the zero'ed inode has |
2259 | * been written to disk. If the allocated inode has never been |
2260 | * written to disk, then the on-disk inode is zero'ed. In either |
2261 | * case we can free the file immediately. |
2262 | */ |
2263 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
2264 | if (inodedep_lookup(ip->i_fsinode_u.fs, ino, 0, &inodedep) == 0 || |
2265 | check_inode_unwritten(inodedep)) { |
2266 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2267 | handle_workitem_freefile(freefile); |
2268 | return; |
2269 | } |
2270 | WORKLIST_INSERT(&inodedep->id_inowait, &freefile->fx_list)do { (&freefile->fx_list)->wk_state |= 0x8000; do { if (((&freefile->fx_list)->wk_list.le_next = (& inodedep->id_inowait)->lh_first) != ((void *)0)) (& inodedep->id_inowait)->lh_first->wk_list.le_prev = & (&freefile->fx_list)->wk_list.le_next; (&inodedep ->id_inowait)->lh_first = (&freefile->fx_list); ( &freefile->fx_list)->wk_list.le_prev = &(&inodedep ->id_inowait)->lh_first; } while (0); } while (0); |
2271 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2272 | } |
2273 | |
2274 | /* |
2275 | * Check to see if an inode has never been written to disk. If |
2276 | * so free the inodedep and return success, otherwise return failure. |
2277 | * This routine must be called with splbio interrupts blocked. |
2278 | * |
2279 | * If we still have a bitmap dependency, then the inode has never |
2280 | * been written to disk. Drop the dependency as it is no longer |
2281 | * necessary since the inode is being deallocated. We set the |
2282 | * ALLCOMPLETE flags since the bitmap now properly shows that the |
2283 | * inode is not allocated. Even if the inode is actively being |
2284 | * written, it has been rolled back to its zero'ed state, so we |
2285 | * are ensured that a zero inode is what is on the disk. For short |
2286 | * lived files, this change will usually result in removing all the |
2287 | * dependencies from the inode so that it can be freed immediately. |
2288 | */ |
2289 | STATIC int |
2290 | check_inode_unwritten(struct inodedep *inodedep) |
2291 | { |
2292 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
2293 | |
2294 | if ((inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008) != 0 || |
2295 | LIST_FIRST(&inodedep->id_pendinghd)((&inodedep->id_pendinghd)->lh_first) != NULL((void *)0) || |
2296 | LIST_FIRST(&inodedep->id_bufwait)((&inodedep->id_bufwait)->lh_first) != NULL((void *)0) || |
2297 | LIST_FIRST(&inodedep->id_inowait)((&inodedep->id_inowait)->lh_first) != NULL((void *)0) || |
2298 | TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) != NULL((void *)0) || |
2299 | TAILQ_FIRST(&inodedep->id_newinoupdt)((&inodedep->id_newinoupdt)->tqh_first) != NULL((void *)0) || |
2300 | inodedep->id_nlinkdelta != 0) |
2301 | return (0); |
2302 | inodedep->id_stateid_list.wk_state |= ALLCOMPLETE(0x0001 | 0x0004 | 0x0008); |
2303 | LIST_REMOVE(inodedep, id_deps)do { if ((inodedep)->id_deps.le_next != ((void *)0)) (inodedep )->id_deps.le_next->id_deps.le_prev = (inodedep)->id_deps .le_prev; *(inodedep)->id_deps.le_prev = (inodedep)->id_deps .le_next; ((inodedep)->id_deps.le_prev) = ((void *)-1); (( inodedep)->id_deps.le_next) = ((void *)-1); } while (0); |
2304 | inodedep->id_buf = NULL((void *)0); |
2305 | if (inodedep->id_stateid_list.wk_state & ONWORKLIST0x8000) |
2306 | WORKLIST_REMOVE(&inodedep->id_list)do { (&inodedep->id_list)->wk_state &= ~0x8000; do { if ((&inodedep->id_list)->wk_list.le_next != ( (void *)0)) (&inodedep->id_list)->wk_list.le_next-> wk_list.le_prev = (&inodedep->id_list)->wk_list.le_prev ; *(&inodedep->id_list)->wk_list.le_prev = (&inodedep ->id_list)->wk_list.le_next; ((&inodedep->id_list )->wk_list.le_prev) = ((void *)-1); ((&inodedep->id_list )->wk_list.le_next) = ((void *)-1); } while (0); } while ( 0); |
2307 | if (inodedep->id_savedino1id_un.idu_savedino1 != NULL((void *)0)) { |
2308 | free(inodedep->id_savedino1id_un.idu_savedino1, M_INODEDEP79, inodedep->id_unsize); |
2309 | inodedep->id_savedino1id_un.idu_savedino1 = NULL((void *)0); |
2310 | } |
2311 | if (free_inodedep(inodedep) == 0) { |
2312 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2313 | panic("check_inode_unwritten: busy inode"); |
2314 | } |
2315 | return (1); |
2316 | } |
2317 | |
2318 | /* |
2319 | * Try to free an inodedep structure. Return 1 if it could be freed. |
2320 | */ |
2321 | STATIC int |
2322 | free_inodedep(struct inodedep *inodedep) |
2323 | { |
2324 | |
2325 | if ((inodedep->id_stateid_list.wk_state & ONWORKLIST0x8000) != 0 || |
2326 | (inodedep->id_stateid_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) != ALLCOMPLETE(0x0001 | 0x0004 | 0x0008) || |
2327 | LIST_FIRST(&inodedep->id_pendinghd)((&inodedep->id_pendinghd)->lh_first) != NULL((void *)0) || |
2328 | LIST_FIRST(&inodedep->id_bufwait)((&inodedep->id_bufwait)->lh_first) != NULL((void *)0) || |
2329 | LIST_FIRST(&inodedep->id_inowait)((&inodedep->id_inowait)->lh_first) != NULL((void *)0) || |
2330 | TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) != NULL((void *)0) || |
2331 | TAILQ_FIRST(&inodedep->id_newinoupdt)((&inodedep->id_newinoupdt)->tqh_first) != NULL((void *)0) || |
2332 | inodedep->id_nlinkdelta != 0 || inodedep->id_savedino1id_un.idu_savedino1 != NULL((void *)0)) |
2333 | return (0); |
2334 | LIST_REMOVE(inodedep, id_hash)do { if ((inodedep)->id_hash.le_next != ((void *)0)) (inodedep )->id_hash.le_next->id_hash.le_prev = (inodedep)->id_hash .le_prev; *(inodedep)->id_hash.le_prev = (inodedep)->id_hash .le_next; ((inodedep)->id_hash.le_prev) = ((void *)-1); (( inodedep)->id_hash.le_next) = ((void *)-1); } while (0); |
2335 | WORKITEM_FREE(inodedep, D_INODEDEP)softdep_freequeue_add((struct worklist *)inodedep); |
2336 | num_inodedep -= 1; |
2337 | return (1); |
2338 | } |
2339 | |
2340 | /* |
2341 | * This workitem routine performs the block de-allocation. |
2342 | * The workitem is added to the pending list after the updated |
2343 | * inode block has been written to disk. As mentioned above, |
2344 | * checks regarding the number of blocks de-allocated (compared |
2345 | * to the number of blocks allocated for the file) are also |
2346 | * performed in this function. |
2347 | */ |
2348 | STATIC void |
2349 | handle_workitem_freeblocks(struct freeblks *freeblks) |
2350 | { |
2351 | struct inode tip; |
2352 | daddr_t bn; |
2353 | union { |
2354 | struct ufs1_dinode di1; |
2355 | struct ufs2_dinode di2; |
2356 | } di; |
2357 | struct fs *fs; |
2358 | int i, level, bsize; |
2359 | long nblocks, blocksreleased = 0; |
2360 | int error, allerror = 0; |
2361 | daddr_t baselbns[NIADDR3], tmpval; |
2362 | |
2363 | if (VFSTOUFS(freeblks->fb_mnt)((struct ufsmount *)((freeblks->fb_mnt)->mnt_data))->um_fstype == UM_UFS11) |
2364 | tip.i_din1dinode_u.ffs1_din = &di.di1; |
2365 | else |
2366 | tip.i_din2dinode_u.ffs2_din = &di.di2; |
2367 | |
2368 | tip.i_fsinode_u.fs = fs = VFSTOUFS(freeblks->fb_mnt)((struct ufsmount *)((freeblks->fb_mnt)->mnt_data))->um_fsufsmount_u.fs; |
2369 | tip.i_number = freeblks->fb_previousinum; |
2370 | tip.i_ump = VFSTOUFS(freeblks->fb_mnt)((struct ufsmount *)((freeblks->fb_mnt)->mnt_data)); |
2371 | tip.i_dev = freeblks->fb_devvp->v_rdevv_un.vu_specinfo->si_rdev; |
2372 | DIP_ASSIGN(&tip, size, freeblks->fb_oldsize)do { if ((&tip)->i_ump->um_fstype == 1) (&tip)-> dinode_u.ffs1_din->di_size = (freeblks->fb_oldsize); else (&tip)->dinode_u.ffs2_din->di_size = (freeblks-> fb_oldsize); } while (0); |
2373 | DIP_ASSIGN(&tip, uid, freeblks->fb_uid)do { if ((&tip)->i_ump->um_fstype == 1) (&tip)-> dinode_u.ffs1_din->di_uid = (freeblks->fb_uid); else (& tip)->dinode_u.ffs2_din->di_uid = (freeblks->fb_uid) ; } while (0); |
2374 | tip.i_vnode = NULL((void *)0); |
2375 | tmpval = 1; |
2376 | baselbns[0] = NDADDR12; |
2377 | for (i = 1; i < NIADDR3; i++) { |
2378 | tmpval *= NINDIR(fs)((fs)->fs_nindir); |
2379 | baselbns[i] = baselbns[i - 1] + tmpval; |
2380 | } |
2381 | nblocks = btodb(fs->fs_bsize)((fs->fs_bsize) >> 9); |
2382 | blocksreleased = 0; |
2383 | /* |
2384 | * Indirect blocks first. |
2385 | */ |
2386 | for (level = (NIADDR3 - 1); level >= 0; level--) { |
2387 | if ((bn = freeblks->fb_iblks[level]) == 0) |
2388 | continue; |
2389 | if ((error = indir_trunc(&tip, fsbtodb(fs, bn)((bn) << (fs)->fs_fsbtodb), level, |
2390 | baselbns[level], &blocksreleased)) != 0) |
2391 | allerror = error; |
2392 | ffs_blkfree(&tip, bn, fs->fs_bsize); |
2393 | blocksreleased += nblocks; |
2394 | } |
2395 | /* |
2396 | * All direct blocks or frags. |
2397 | */ |
2398 | for (i = (NDADDR12 - 1); i >= 0; i--) { |
2399 | if ((bn = freeblks->fb_dblks[i]) == 0) |
2400 | continue; |
2401 | bsize = blksize(fs, &tip, i)(((i) >= 12 || ((((&tip))->i_ump->um_fstype == 1 ) ? ((&tip))->dinode_u.ffs1_din->di_size : ((&tip ))->dinode_u.ffs2_din->di_size) >= ((i) + 1) << (fs)->fs_bshift) ? (u_int64_t)(fs)->fs_bsize : ((((((( (((&tip))->i_ump->um_fstype == 1) ? ((&tip))-> dinode_u.ffs1_din->di_size : ((&tip))->dinode_u.ffs2_din ->di_size)) & (fs)->fs_qbmask)) + (fs)->fs_qfmask ) & (fs)->fs_fmask))); |
2402 | ffs_blkfree(&tip, bn, bsize); |
2403 | blocksreleased += btodb(bsize)((bsize) >> 9); |
2404 | } |
2405 | |
2406 | #ifdef DIAGNOSTIC1 |
2407 | if (freeblks->fb_chkcnt != blocksreleased) |
2408 | printf("handle_workitem_freeblocks: block count\n"); |
2409 | if (allerror) |
2410 | softdep_error("handle_workitem_freeblks", allerror); |
2411 | #endif /* DIAGNOSTIC */ |
2412 | WORKITEM_FREE(freeblks, D_FREEBLKS)softdep_freequeue_add((struct worklist *)freeblks); |
2413 | } |
2414 | |
2415 | /* |
2416 | * Release blocks associated with the inode ip and stored in the indirect |
2417 | * block dbn. If level is greater than SINGLE, the block is an indirect block |
2418 | * and recursive calls to indirtrunc must be used to cleanse other indirect |
2419 | * blocks. |
2420 | */ |
2421 | STATIC int |
2422 | indir_trunc(struct inode *ip, daddr_t dbn, int level, daddr_t lbn, |
2423 | long *countp) |
2424 | { |
2425 | struct buf *bp; |
2426 | int32_t *bap1 = NULL((void *)0); |
2427 | int64_t nb, *bap2 = NULL((void *)0); |
2428 | struct fs *fs; |
2429 | struct worklist *wk; |
2430 | struct indirdep *indirdep; |
2431 | int i, lbnadd, nblocks, ufs1fmt; |
2432 | int error, allerror = 0; |
2433 | |
2434 | fs = ip->i_fsinode_u.fs; |
2435 | lbnadd = 1; |
2436 | for (i = level; i > 0; i--) |
2437 | lbnadd *= NINDIR(fs)((fs)->fs_nindir); |
2438 | /* |
2439 | * Get buffer of block pointers to be freed. This routine is not |
2440 | * called until the zero'ed inode has been written, so it is safe |
2441 | * to free blocks as they are encountered. Because the inode has |
2442 | * been zero'ed, calls to bmap on these blocks will fail. So, we |
2443 | * have to use the on-disk address and the block device for the |
2444 | * filesystem to look them up. If the file was deleted before its |
2445 | * indirect blocks were all written to disk, the routine that set |
2446 | * us up (deallocate_dependencies) will have arranged to leave |
2447 | * a complete copy of the indirect block in memory for our use. |
2448 | * Otherwise we have to read the blocks in from the disk. |
2449 | */ |
2450 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
2451 | if ((bp = incore(ip->i_devvpi_ump->um_devvp, dbn)) != NULL((void *)0) && |
2452 | (wk = LIST_FIRST(&bp->b_dep)((&bp->b_dep)->lh_first)) != NULL((void *)0)) { |
2453 | if (wk->wk_type != D_INDIRDEP5 || |
2454 | (indirdep = WK_INDIRDEP(wk)((struct indirdep *)(wk)))->ir_savebp != bp || |
2455 | (indirdep->ir_stateir_list.wk_state & GOINGAWAY0x0100) == 0) { |
2456 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2457 | panic("indir_trunc: lost indirdep"); |
2458 | } |
2459 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
2460 | WORKITEM_FREE(indirdep, D_INDIRDEP)softdep_freequeue_add((struct worklist *)indirdep); |
2461 | if (LIST_FIRST(&bp->b_dep)((&bp->b_dep)->lh_first) != NULL((void *)0)) { |
2462 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2463 | panic("indir_trunc: dangling dep"); |
2464 | } |
2465 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2466 | } else { |
2467 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2468 | error = bread(ip->i_devvpi_ump->um_devvp, dbn, (int)fs->fs_bsize, &bp); |
2469 | if (error) |
2470 | return (error); |
2471 | } |
2472 | /* |
2473 | * Recursively free indirect blocks. |
2474 | */ |
2475 | if (ip->i_ump->um_fstype == UM_UFS11) { |
2476 | ufs1fmt = 1; |
2477 | bap1 = (int32_t *)bp->b_data; |
2478 | } else { |
2479 | ufs1fmt = 0; |
2480 | bap2 = (int64_t *)bp->b_data; |
2481 | } |
2482 | nblocks = btodb(fs->fs_bsize)((fs->fs_bsize) >> 9); |
2483 | for (i = NINDIR(fs)((fs)->fs_nindir) - 1; i >= 0; i--) { |
2484 | if (ufs1fmt) |
2485 | nb = bap1[i]; |
2486 | else |
2487 | nb = bap2[i]; |
2488 | if (nb == 0) |
2489 | continue; |
2490 | if (level != 0) { |
2491 | if ((error = indir_trunc(ip, fsbtodb(fs, nb)((nb) << (fs)->fs_fsbtodb), |
2492 | level - 1, lbn + (i * lbnadd), countp)) != 0) |
2493 | allerror = error; |
2494 | } |
2495 | ffs_blkfree(ip, nb, fs->fs_bsize); |
2496 | *countp += nblocks; |
2497 | } |
2498 | bp->b_flags |= B_INVAL0x00000800 | B_NOCACHE0x00001000; |
2499 | brelse(bp); |
2500 | return (allerror); |
2501 | } |
2502 | |
2503 | /* |
2504 | * Free an allocindir. |
2505 | * This routine must be called with splbio interrupts blocked. |
2506 | */ |
2507 | STATIC void |
2508 | free_allocindir(struct allocindir *aip, struct inodedep *inodedep) |
2509 | { |
2510 | struct freefrag *freefrag; |
2511 | |
2512 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
2513 | |
2514 | #ifdef DEBUG |
2515 | if (lk.lkt_held == -1) |
2516 | panic("free_allocindir: lock not held"); |
2517 | #endif |
2518 | if ((aip->ai_stateai_list.wk_state & DEPCOMPLETE0x0008) == 0) |
2519 | LIST_REMOVE(aip, ai_deps)do { if ((aip)->ai_deps.le_next != ((void *)0)) (aip)-> ai_deps.le_next->ai_deps.le_prev = (aip)->ai_deps.le_prev ; *(aip)->ai_deps.le_prev = (aip)->ai_deps.le_next; ((aip )->ai_deps.le_prev) = ((void *)-1); ((aip)->ai_deps.le_next ) = ((void *)-1); } while (0); |
2520 | if (aip->ai_stateai_list.wk_state & ONWORKLIST0x8000) |
2521 | WORKLIST_REMOVE(&aip->ai_list)do { (&aip->ai_list)->wk_state &= ~0x8000; do { if ((&aip->ai_list)->wk_list.le_next != ((void *)0 )) (&aip->ai_list)->wk_list.le_next->wk_list.le_prev = (&aip->ai_list)->wk_list.le_prev; *(&aip-> ai_list)->wk_list.le_prev = (&aip->ai_list)->wk_list .le_next; ((&aip->ai_list)->wk_list.le_prev) = ((void *)-1); ((&aip->ai_list)->wk_list.le_next) = ((void *)-1); } while (0); } while (0); |
2522 | LIST_REMOVE(aip, ai_next)do { if ((aip)->ai_next.le_next != ((void *)0)) (aip)-> ai_next.le_next->ai_next.le_prev = (aip)->ai_next.le_prev ; *(aip)->ai_next.le_prev = (aip)->ai_next.le_next; ((aip )->ai_next.le_prev) = ((void *)-1); ((aip)->ai_next.le_next ) = ((void *)-1); } while (0); |
2523 | if ((freefrag = aip->ai_freefrag) != NULL((void *)0)) { |
2524 | if (inodedep == NULL((void *)0)) |
2525 | add_to_worklist(&freefrag->ff_list); |
2526 | else |
2527 | WORKLIST_INSERT(&inodedep->id_bufwait,do { (&freefrag->ff_list)->wk_state |= 0x8000; do { if (((&freefrag->ff_list)->wk_list.le_next = (& inodedep->id_bufwait)->lh_first) != ((void *)0)) (& inodedep->id_bufwait)->lh_first->wk_list.le_prev = & (&freefrag->ff_list)->wk_list.le_next; (&inodedep ->id_bufwait)->lh_first = (&freefrag->ff_list); ( &freefrag->ff_list)->wk_list.le_prev = &(&inodedep ->id_bufwait)->lh_first; } while (0); } while (0) |
2528 | &freefrag->ff_list)do { (&freefrag->ff_list)->wk_state |= 0x8000; do { if (((&freefrag->ff_list)->wk_list.le_next = (& inodedep->id_bufwait)->lh_first) != ((void *)0)) (& inodedep->id_bufwait)->lh_first->wk_list.le_prev = & (&freefrag->ff_list)->wk_list.le_next; (&inodedep ->id_bufwait)->lh_first = (&freefrag->ff_list); ( &freefrag->ff_list)->wk_list.le_prev = &(&inodedep ->id_bufwait)->lh_first; } while (0); } while (0); |
2529 | } |
2530 | WORKITEM_FREE(aip, D_ALLOCINDIR)softdep_freequeue_add((struct worklist *)aip); |
2531 | } |
2532 | |
2533 | /* |
2534 | * Directory entry addition dependencies. |
2535 | * |
2536 | * When adding a new directory entry, the inode (with its incremented link |
2537 | * count) must be written to disk before the directory entry's pointer to it. |
2538 | * Also, if the inode is newly allocated, the corresponding freemap must be |
2539 | * updated (on disk) before the directory entry's pointer. These requirements |
2540 | * are met via undo/redo on the directory entry's pointer, which consists |
2541 | * simply of the inode number. |
2542 | * |
2543 | * As directory entries are added and deleted, the free space within a |
2544 | * directory block can become fragmented. The ufs file system will compact |
2545 | * a fragmented directory block to make space for a new entry. When this |
2546 | * occurs, the offsets of previously added entries change. Any "diradd" |
2547 | * dependency structures corresponding to these entries must be updated with |
2548 | * the new offsets. |
2549 | */ |
2550 | |
2551 | /* |
2552 | * This routine is called after the in-memory inode's link |
2553 | * count has been incremented, but before the directory entry's |
2554 | * pointer to the inode has been set. |
2555 | */ |
2556 | /* buffer containing directory block */ |
2557 | /* inode for directory */ |
2558 | /* offset of new entry in directory */ |
2559 | /* inode referenced by new directory entry */ |
2560 | /* non-NULL => contents of new mkdir */ |
2561 | /* entry is in a newly allocated block */ |
2562 | int |
2563 | softdep_setup_directory_add(struct buf *bp, struct inode *dp, off_t diroffset, |
2564 | long newinum, struct buf *newdirbp, int isnewblk) |
2565 | { |
2566 | int offset; /* offset of new entry within directory block */ |
2567 | daddr_t lbn; /* block in directory containing new entry */ |
2568 | struct fs *fs; |
2569 | struct diradd *dap; |
2570 | struct allocdirect *adp; |
2571 | struct pagedep *pagedep; |
2572 | struct inodedep *inodedep; |
2573 | struct newdirblk *newdirblk = NULL((void *)0); |
2574 | struct mkdir *mkdir1, *mkdir2; |
2575 | |
2576 | |
2577 | fs = dp->i_fsinode_u.fs; |
2578 | lbn = lblkno(fs, diroffset)((diroffset) >> (fs)->fs_bshift); |
2579 | offset = blkoff(fs, diroffset)((diroffset) & (fs)->fs_qbmask); |
2580 | dap = pool_get(&diradd_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
2581 | dap->da_list.wk_type = D_DIRADD10; |
2582 | dap->da_offset = offset; |
2583 | dap->da_newinum = newinum; |
2584 | dap->da_stateda_list.wk_state = ATTACHED0x0001; |
2585 | if (isnewblk && lbn < NDADDR12 && fragoff(fs, diroffset)((diroffset) & (fs)->fs_qfmask) == 0) { |
2586 | newdirblk = pool_get(&newdirblk_pool, PR_WAITOK0x0001); |
2587 | newdirblk->db_list.wk_type = D_NEWDIRBLK13; |
2588 | newdirblk->db_statedb_list.wk_state = 0; |
2589 | } |
2590 | if (newdirbp == NULL((void *)0)) { |
2591 | dap->da_stateda_list.wk_state |= DEPCOMPLETE0x0008; |
2592 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
2593 | } else { |
2594 | dap->da_stateda_list.wk_state |= MKDIR_BODY0x0020 | MKDIR_PARENT0x0010; |
2595 | mkdir1 = pool_get(&mkdir_pool, PR_WAITOK0x0001); |
2596 | mkdir1->md_list.wk_type = D_MKDIR11; |
2597 | mkdir1->md_statemd_list.wk_state = MKDIR_BODY0x0020; |
2598 | mkdir1->md_diradd = dap; |
2599 | mkdir2 = pool_get(&mkdir_pool, PR_WAITOK0x0001); |
2600 | mkdir2->md_list.wk_type = D_MKDIR11; |
2601 | mkdir2->md_statemd_list.wk_state = MKDIR_PARENT0x0010; |
2602 | mkdir2->md_diradd = dap; |
2603 | /* |
2604 | * Dependency on "." and ".." being written to disk. |
2605 | */ |
2606 | mkdir1->md_buf = newdirbp; |
2607 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
2608 | LIST_INSERT_HEAD(&mkdirlisthd, mkdir1, md_mkdirs)do { if (((mkdir1)->md_mkdirs.le_next = (&mkdirlisthd) ->lh_first) != ((void *)0)) (&mkdirlisthd)->lh_first ->md_mkdirs.le_prev = &(mkdir1)->md_mkdirs.le_next; (&mkdirlisthd)->lh_first = (mkdir1); (mkdir1)->md_mkdirs .le_prev = &(&mkdirlisthd)->lh_first; } while (0); |
2609 | WORKLIST_INSERT(&newdirbp->b_dep, &mkdir1->md_list)do { (&mkdir1->md_list)->wk_state |= 0x8000; do { if (((&mkdir1->md_list)->wk_list.le_next = (&newdirbp ->b_dep)->lh_first) != ((void *)0)) (&newdirbp-> b_dep)->lh_first->wk_list.le_prev = &(&mkdir1-> md_list)->wk_list.le_next; (&newdirbp->b_dep)->lh_first = (&mkdir1->md_list); (&mkdir1->md_list)->wk_list .le_prev = &(&newdirbp->b_dep)->lh_first; } while (0); } while (0); |
2610 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2611 | bdwrite(newdirbp); |
2612 | /* |
2613 | * Dependency on link count increase for parent directory |
2614 | */ |
2615 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
2616 | if (inodedep_lookup(fs, dp->i_number, 0, &inodedep) == 0 |
2617 | || (inodedep->id_stateid_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) { |
2618 | dap->da_stateda_list.wk_state &= ~MKDIR_PARENT0x0010; |
2619 | WORKITEM_FREE(mkdir2, D_MKDIR)softdep_freequeue_add((struct worklist *)mkdir2); |
2620 | } else { |
2621 | LIST_INSERT_HEAD(&mkdirlisthd, mkdir2, md_mkdirs)do { if (((mkdir2)->md_mkdirs.le_next = (&mkdirlisthd) ->lh_first) != ((void *)0)) (&mkdirlisthd)->lh_first ->md_mkdirs.le_prev = &(mkdir2)->md_mkdirs.le_next; (&mkdirlisthd)->lh_first = (mkdir2); (mkdir2)->md_mkdirs .le_prev = &(&mkdirlisthd)->lh_first; } while (0); |
2622 | WORKLIST_INSERT(&inodedep->id_bufwait,&mkdir2->md_list)do { (&mkdir2->md_list)->wk_state |= 0x8000; do { if (((&mkdir2->md_list)->wk_list.le_next = (&inodedep ->id_bufwait)->lh_first) != ((void *)0)) (&inodedep ->id_bufwait)->lh_first->wk_list.le_prev = &(& mkdir2->md_list)->wk_list.le_next; (&inodedep->id_bufwait )->lh_first = (&mkdir2->md_list); (&mkdir2-> md_list)->wk_list.le_prev = &(&inodedep->id_bufwait )->lh_first; } while (0); } while (0); |
2623 | } |
2624 | } |
2625 | /* |
2626 | * Link into parent directory pagedep to await its being written. |
2627 | */ |
2628 | if (pagedep_lookup(dp, lbn, DEPALLOC0x0001, &pagedep) == 0) |
2629 | WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list)do { (&pagedep->pd_list)->wk_state |= 0x8000; do { if (((&pagedep->pd_list)->wk_list.le_next = (&bp-> b_dep)->lh_first) != ((void *)0)) (&bp->b_dep)-> lh_first->wk_list.le_prev = &(&pagedep->pd_list )->wk_list.le_next; (&bp->b_dep)->lh_first = (& pagedep->pd_list); (&pagedep->pd_list)->wk_list. le_prev = &(&bp->b_dep)->lh_first; } while (0); } while (0); |
2630 | dap->da_pagedepda_un.dau_pagedep = pagedep; |
2631 | LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap,do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_diraddhd [(((offset) >> 2) % 6)])->lh_first) != ((void *)0)) ( &pagedep->pd_diraddhd[(((offset) >> 2) % 6)])-> lh_first->da_pdlist.le_prev = &(dap)->da_pdlist.le_next ; (&pagedep->pd_diraddhd[(((offset) >> 2) % 6)]) ->lh_first = (dap); (dap)->da_pdlist.le_prev = &(& pagedep->pd_diraddhd[(((offset) >> 2) % 6)])->lh_first ; } while (0) |
2632 | da_pdlist)do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_diraddhd [(((offset) >> 2) % 6)])->lh_first) != ((void *)0)) ( &pagedep->pd_diraddhd[(((offset) >> 2) % 6)])-> lh_first->da_pdlist.le_prev = &(dap)->da_pdlist.le_next ; (&pagedep->pd_diraddhd[(((offset) >> 2) % 6)]) ->lh_first = (dap); (dap)->da_pdlist.le_prev = &(& pagedep->pd_diraddhd[(((offset) >> 2) % 6)])->lh_first ; } while (0); |
2633 | /* |
2634 | * Link into its inodedep. Put it on the id_bufwait list if the inode |
2635 | * is not yet written. If it is written, do the post-inode write |
2636 | * processing to put it on the id_pendinghd list. |
2637 | */ |
2638 | (void) inodedep_lookup(fs, newinum, DEPALLOC0x0001, &inodedep); |
2639 | if ((inodedep->id_stateid_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) |
2640 | diradd_inode_written(dap, inodedep); |
2641 | else |
2642 | WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list)do { (&dap->da_list)->wk_state |= 0x8000; do { if ( ((&dap->da_list)->wk_list.le_next = (&inodedep-> id_bufwait)->lh_first) != ((void *)0)) (&inodedep-> id_bufwait)->lh_first->wk_list.le_prev = &(&dap ->da_list)->wk_list.le_next; (&inodedep->id_bufwait )->lh_first = (&dap->da_list); (&dap->da_list )->wk_list.le_prev = &(&inodedep->id_bufwait)-> lh_first; } while (0); } while (0); |
2643 | if (isnewblk) { |
2644 | /* |
2645 | * Directories growing into indirect blocks are rare |
2646 | * enough and the frequency of new block allocation |
2647 | * in those cases even more rare, that we choose not |
2648 | * to bother tracking them. Rather we simply force the |
2649 | * new directory entry to disk. |
2650 | */ |
2651 | if (lbn >= NDADDR12) { |
2652 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2653 | /* |
2654 | * We only have a new allocation when at the |
2655 | * beginning of a new block, not when we are |
2656 | * expanding into an existing block. |
2657 | */ |
2658 | if (blkoff(fs, diroffset)((diroffset) & (fs)->fs_qbmask) == 0) |
2659 | return (1); |
2660 | return (0); |
2661 | } |
2662 | /* |
2663 | * We only have a new allocation when at the beginning |
2664 | * of a new fragment, not when we are expanding into an |
2665 | * existing fragment. Also, there is nothing to do if we |
2666 | * are already tracking this block. |
2667 | */ |
2668 | if (fragoff(fs, diroffset)((diroffset) & (fs)->fs_qfmask) != 0) { |
2669 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2670 | return (0); |
2671 | } |
2672 | |
2673 | if ((pagedep->pd_statepd_list.wk_state & NEWBLOCK0x0800) != 0) { |
2674 | WORKITEM_FREE(newdirblk, D_NEWDIRBLK)softdep_freequeue_add((struct worklist *)newdirblk); |
2675 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2676 | return (0); |
2677 | } |
2678 | /* |
2679 | * Find our associated allocdirect and have it track us. |
2680 | */ |
2681 | if (inodedep_lookup(fs, dp->i_number, 0, &inodedep) == 0) |
2682 | panic("softdep_setup_directory_add: lost inodedep"); |
2683 | adp = TAILQ_LAST(&inodedep->id_newinoupdt, allocdirectlst)(*(((struct allocdirectlst *)((&inodedep->id_newinoupdt )->tqh_last))->tqh_last)); |
2684 | if (adp == NULL((void *)0) || adp->ad_lbn != lbn) { |
2685 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2686 | panic("softdep_setup_directory_add: lost entry"); |
2687 | } |
2688 | pagedep->pd_statepd_list.wk_state |= NEWBLOCK0x0800; |
2689 | newdirblk->db_pagedep = pagedep; |
2690 | WORKLIST_INSERT(&adp->ad_newdirblk, &newdirblk->db_list)do { (&newdirblk->db_list)->wk_state |= 0x8000; do { if (((&newdirblk->db_list)->wk_list.le_next = (& adp->ad_newdirblk)->lh_first) != ((void *)0)) (&adp ->ad_newdirblk)->lh_first->wk_list.le_prev = &(& newdirblk->db_list)->wk_list.le_next; (&adp->ad_newdirblk )->lh_first = (&newdirblk->db_list); (&newdirblk ->db_list)->wk_list.le_prev = &(&adp->ad_newdirblk )->lh_first; } while (0); } while (0); |
2691 | } |
2692 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2693 | return (0); |
2694 | } |
2695 | |
2696 | /* |
2697 | * This procedure is called to change the offset of a directory |
2698 | * entry when compacting a directory block which must be owned |
2699 | * exclusively by the caller. Note that the actual entry movement |
2700 | * must be done in this procedure to ensure that no I/O completions |
2701 | * occur while the move is in progress. |
2702 | */ |
2703 | /* inode for directory */ |
2704 | /* address of dp->i_offset */ |
2705 | /* address of old directory location */ |
2706 | /* address of new directory location */ |
2707 | /* size of directory entry */ |
2708 | void |
2709 | softdep_change_directoryentry_offset(struct inode *dp, caddr_t base, |
2710 | caddr_t oldloc, caddr_t newloc, int entrysize) |
2711 | { |
2712 | int offset, oldoffset, newoffset; |
2713 | struct pagedep *pagedep; |
2714 | struct diradd *dap; |
2715 | daddr_t lbn; |
2716 | |
2717 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
2718 | lbn = lblkno(dp->i_fs, dp->i_offset)((dp->i_offset) >> (dp->inode_u.fs)->fs_bshift ); |
2719 | offset = blkoff(dp->i_fs, dp->i_offset)((dp->i_offset) & (dp->inode_u.fs)->fs_qbmask); |
2720 | if (pagedep_lookup(dp, lbn, 0, &pagedep) == 0) |
2721 | goto done; |
2722 | oldoffset = offset + (oldloc - base); |
2723 | newoffset = offset + (newloc - base); |
2724 | |
2725 | LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(oldoffset)], da_pdlist)for((dap) = ((&pagedep->pd_diraddhd[(((oldoffset) >> 2) % 6)])->lh_first); (dap)!= ((void *)0); (dap) = ((dap) ->da_pdlist.le_next)) { |
2726 | if (dap->da_offset != oldoffset) |
2727 | continue; |
2728 | dap->da_offset = newoffset; |
2729 | if (DIRADDHASH(newoffset)(((newoffset) >> 2) % 6) == DIRADDHASH(oldoffset)(((oldoffset) >> 2) % 6)) |
2730 | break; |
2731 | LIST_REMOVE(dap, da_pdlist)do { if ((dap)->da_pdlist.le_next != ((void *)0)) (dap)-> da_pdlist.le_next->da_pdlist.le_prev = (dap)->da_pdlist .le_prev; *(dap)->da_pdlist.le_prev = (dap)->da_pdlist. le_next; ((dap)->da_pdlist.le_prev) = ((void *)-1); ((dap) ->da_pdlist.le_next) = ((void *)-1); } while (0); |
2732 | LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(newoffset)],do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_diraddhd [(((newoffset) >> 2) % 6)])->lh_first) != ((void *)0 )) (&pagedep->pd_diraddhd[(((newoffset) >> 2) % 6 )])->lh_first->da_pdlist.le_prev = &(dap)->da_pdlist .le_next; (&pagedep->pd_diraddhd[(((newoffset) >> 2) % 6)])->lh_first = (dap); (dap)->da_pdlist.le_prev = &(&pagedep->pd_diraddhd[(((newoffset) >> 2) % 6)])->lh_first; } while (0) |
2733 | dap, da_pdlist)do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_diraddhd [(((newoffset) >> 2) % 6)])->lh_first) != ((void *)0 )) (&pagedep->pd_diraddhd[(((newoffset) >> 2) % 6 )])->lh_first->da_pdlist.le_prev = &(dap)->da_pdlist .le_next; (&pagedep->pd_diraddhd[(((newoffset) >> 2) % 6)])->lh_first = (dap); (dap)->da_pdlist.le_prev = &(&pagedep->pd_diraddhd[(((newoffset) >> 2) % 6)])->lh_first; } while (0); |
2734 | break; |
2735 | } |
2736 | if (dap == NULL((void *)0)) { |
2737 | |
2738 | LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist)for((dap) = ((&pagedep->pd_pendinghd)->lh_first); ( dap)!= ((void *)0); (dap) = ((dap)->da_pdlist.le_next)) { |
2739 | if (dap->da_offset == oldoffset) { |
2740 | dap->da_offset = newoffset; |
2741 | break; |
2742 | } |
2743 | } |
2744 | } |
2745 | done: |
2746 | memmove(newloc, oldloc, entrysize)__builtin_memmove((newloc), (oldloc), (entrysize)); |
2747 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2748 | } |
2749 | |
2750 | /* |
2751 | * Free a diradd dependency structure. This routine must be called |
2752 | * with splbio interrupts blocked. |
2753 | */ |
2754 | STATIC void |
2755 | free_diradd(struct diradd *dap) |
2756 | { |
2757 | struct dirrem *dirrem; |
2758 | struct pagedep *pagedep; |
2759 | struct inodedep *inodedep; |
2760 | struct mkdir *mkdir, *nextmd; |
2761 | |
2762 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
2763 | |
2764 | #ifdef DEBUG |
2765 | if (lk.lkt_held == -1) |
2766 | panic("free_diradd: lock not held"); |
2767 | #endif |
2768 | WORKLIST_REMOVE(&dap->da_list)do { (&dap->da_list)->wk_state &= ~0x8000; do { if ((&dap->da_list)->wk_list.le_next != ((void *)0 )) (&dap->da_list)->wk_list.le_next->wk_list.le_prev = (&dap->da_list)->wk_list.le_prev; *(&dap-> da_list)->wk_list.le_prev = (&dap->da_list)->wk_list .le_next; ((&dap->da_list)->wk_list.le_prev) = ((void *)-1); ((&dap->da_list)->wk_list.le_next) = ((void *)-1); } while (0); } while (0); |
2769 | LIST_REMOVE(dap, da_pdlist)do { if ((dap)->da_pdlist.le_next != ((void *)0)) (dap)-> da_pdlist.le_next->da_pdlist.le_prev = (dap)->da_pdlist .le_prev; *(dap)->da_pdlist.le_prev = (dap)->da_pdlist. le_next; ((dap)->da_pdlist.le_prev) = ((void *)-1); ((dap) ->da_pdlist.le_next) = ((void *)-1); } while (0); |
2770 | if ((dap->da_stateda_list.wk_state & DIRCHG0x0080) == 0) { |
2771 | pagedep = dap->da_pagedepda_un.dau_pagedep; |
2772 | } else { |
2773 | dirrem = dap->da_previousda_un.dau_previous; |
2774 | pagedep = dirrem->dm_pagedepdm_un.dmu_pagedep; |
2775 | dirrem->dm_dirinumdm_un.dmu_dirinum = pagedep->pd_ino; |
2776 | add_to_worklist(&dirrem->dm_list); |
2777 | } |
2778 | if (inodedep_lookup(VFSTOUFS(pagedep->pd_mnt)((struct ufsmount *)((pagedep->pd_mnt)->mnt_data))->um_fsufsmount_u.fs, dap->da_newinum, |
2779 | 0, &inodedep) != 0) |
2780 | (void) free_inodedep(inodedep); |
2781 | if ((dap->da_stateda_list.wk_state & (MKDIR_PARENT0x0010 | MKDIR_BODY0x0020)) != 0) { |
2782 | for (mkdir = LIST_FIRST(&mkdirlisthd)((&mkdirlisthd)->lh_first); mkdir; mkdir = nextmd) { |
2783 | nextmd = LIST_NEXT(mkdir, md_mkdirs)((mkdir)->md_mkdirs.le_next); |
2784 | if (mkdir->md_diradd != dap) |
2785 | continue; |
2786 | dap->da_stateda_list.wk_state &= ~mkdir->md_statemd_list.wk_state; |
2787 | WORKLIST_REMOVE(&mkdir->md_list)do { (&mkdir->md_list)->wk_state &= ~0x8000; do { if ((&mkdir->md_list)->wk_list.le_next != ((void *)0)) (&mkdir->md_list)->wk_list.le_next->wk_list .le_prev = (&mkdir->md_list)->wk_list.le_prev; *(& mkdir->md_list)->wk_list.le_prev = (&mkdir->md_list )->wk_list.le_next; ((&mkdir->md_list)->wk_list. le_prev) = ((void *)-1); ((&mkdir->md_list)->wk_list .le_next) = ((void *)-1); } while (0); } while (0); |
2788 | LIST_REMOVE(mkdir, md_mkdirs)do { if ((mkdir)->md_mkdirs.le_next != ((void *)0)) (mkdir )->md_mkdirs.le_next->md_mkdirs.le_prev = (mkdir)->md_mkdirs .le_prev; *(mkdir)->md_mkdirs.le_prev = (mkdir)->md_mkdirs .le_next; ((mkdir)->md_mkdirs.le_prev) = ((void *)-1); ((mkdir )->md_mkdirs.le_next) = ((void *)-1); } while (0); |
2789 | WORKITEM_FREE(mkdir, D_MKDIR)softdep_freequeue_add((struct worklist *)mkdir); |
2790 | } |
2791 | if ((dap->da_stateda_list.wk_state & (MKDIR_PARENT0x0010 | MKDIR_BODY0x0020)) != 0) { |
2792 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2793 | panic("free_diradd: unfound ref"); |
2794 | } |
2795 | } |
2796 | WORKITEM_FREE(dap, D_DIRADD)softdep_freequeue_add((struct worklist *)dap); |
2797 | } |
2798 | |
2799 | /* |
2800 | * Directory entry removal dependencies. |
2801 | * |
2802 | * When removing a directory entry, the entry's inode pointer must be |
2803 | * zero'ed on disk before the corresponding inode's link count is decremented |
2804 | * (possibly freeing the inode for re-use). This dependency is handled by |
2805 | * updating the directory entry but delaying the inode count reduction until |
2806 | * after the directory block has been written to disk. After this point, the |
2807 | * inode count can be decremented whenever it is convenient. |
2808 | */ |
2809 | |
2810 | /* |
2811 | * This routine should be called immediately after removing |
2812 | * a directory entry. The inode's link count should not be |
2813 | * decremented by the calling procedure -- the soft updates |
2814 | * code will do this task when it is safe. |
2815 | */ |
2816 | /* buffer containing directory block */ |
2817 | /* inode for the directory being modified */ |
2818 | /* inode for directory entry being removed */ |
2819 | /* indicates if doing RMDIR */ |
2820 | void |
2821 | softdep_setup_remove(struct buf *bp, struct inode *dp, struct inode *ip, |
2822 | int isrmdir) |
2823 | { |
2824 | struct dirrem *dirrem, *prevdirrem; |
2825 | |
2826 | /* |
2827 | * Allocate a new dirrem if appropriate and ACQUIRE_LOCK. |
2828 | */ |
2829 | dirrem = newdirrem(bp, dp, ip, isrmdir, &prevdirrem); |
2830 | |
2831 | /* |
2832 | * If the COMPLETE flag is clear, then there were no active |
2833 | * entries and we want to roll back to a zeroed entry until |
2834 | * the new inode is committed to disk. If the COMPLETE flag is |
2835 | * set then we have deleted an entry that never made it to |
2836 | * disk. If the entry we deleted resulted from a name change, |
2837 | * then the old name still resides on disk. We cannot delete |
2838 | * its inode (returned to us in prevdirrem) until the zeroed |
2839 | * directory entry gets to disk. The new inode has never been |
2840 | * referenced on the disk, so can be deleted immediately. |
2841 | */ |
2842 | if ((dirrem->dm_statedm_list.wk_state & COMPLETE0x0004) == 0) { |
2843 | LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd, dirrem,do { if (((dirrem)->dm_next.le_next = (&dirrem->dm_un .dmu_pagedep->pd_dirremhd)->lh_first) != ((void *)0)) ( &dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first ->dm_next.le_prev = &(dirrem)->dm_next.le_next; (& dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first = ( dirrem); (dirrem)->dm_next.le_prev = &(&dirrem-> dm_un.dmu_pagedep->pd_dirremhd)->lh_first; } while (0) |
2844 | dm_next)do { if (((dirrem)->dm_next.le_next = (&dirrem->dm_un .dmu_pagedep->pd_dirremhd)->lh_first) != ((void *)0)) ( &dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first ->dm_next.le_prev = &(dirrem)->dm_next.le_next; (& dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first = ( dirrem); (dirrem)->dm_next.le_prev = &(&dirrem-> dm_un.dmu_pagedep->pd_dirremhd)->lh_first; } while (0); |
2845 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2846 | } else { |
2847 | if (prevdirrem != NULL((void *)0)) |
2848 | LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd,do { if (((prevdirrem)->dm_next.le_next = (&dirrem-> dm_un.dmu_pagedep->pd_dirremhd)->lh_first) != ((void *) 0)) (&dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first ->dm_next.le_prev = &(prevdirrem)->dm_next.le_next; (&dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first = (prevdirrem); (prevdirrem)->dm_next.le_prev = &(& dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first; } while (0) |
2849 | prevdirrem, dm_next)do { if (((prevdirrem)->dm_next.le_next = (&dirrem-> dm_un.dmu_pagedep->pd_dirremhd)->lh_first) != ((void *) 0)) (&dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first ->dm_next.le_prev = &(prevdirrem)->dm_next.le_next; (&dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first = (prevdirrem); (prevdirrem)->dm_next.le_prev = &(& dirrem->dm_un.dmu_pagedep->pd_dirremhd)->lh_first; } while (0); |
2850 | dirrem->dm_dirinumdm_un.dmu_dirinum = dirrem->dm_pagedepdm_un.dmu_pagedep->pd_ino; |
2851 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2852 | handle_workitem_remove(dirrem); |
2853 | } |
2854 | } |
2855 | |
2856 | STATIC long num_dirrem; /* number of dirrem allocated */ |
2857 | /* |
2858 | * Allocate a new dirrem if appropriate and return it along with |
2859 | * its associated pagedep. Called without a lock, returns with lock. |
2860 | */ |
2861 | /* buffer containing directory block */ |
2862 | /* inode for the directory being modified */ |
2863 | /* inode for directory entry being removed */ |
2864 | /* indicates if doing RMDIR */ |
2865 | /* previously referenced inode, if any */ |
2866 | STATIC struct dirrem * |
2867 | newdirrem(struct buf *bp, struct inode *dp, struct inode *ip, int isrmdir, |
2868 | struct dirrem **prevdirremp) |
2869 | { |
2870 | int offset; |
2871 | daddr_t lbn; |
2872 | struct diradd *dap; |
2873 | struct dirrem *dirrem; |
2874 | struct pagedep *pagedep; |
2875 | |
2876 | /* |
2877 | * Whiteouts have no deletion dependencies. |
2878 | */ |
2879 | if (ip == NULL((void *)0)) |
2880 | panic("newdirrem: whiteout"); |
2881 | /* |
2882 | * If we are over our limit, try to improve the situation. |
2883 | * Limiting the number of dirrem structures will also limit |
2884 | * the number of freefile and freeblks structures. |
2885 | */ |
2886 | if (num_dirrem > max_softdeps / 2) |
2887 | (void) request_cleanup(FLUSH_REMOVE2, 0); |
2888 | num_dirrem += 1; |
2889 | dirrem = pool_get(&dirrem_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
2890 | dirrem->dm_list.wk_type = D_DIRREM12; |
2891 | dirrem->dm_statedm_list.wk_state = isrmdir ? RMDIR0x0040 : 0; |
2892 | dirrem->dm_mnt = ITOV(ip)((ip)->i_vnode)->v_mount; |
2893 | dirrem->dm_oldinum = ip->i_number; |
2894 | *prevdirremp = NULL((void *)0); |
2895 | |
2896 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
2897 | lbn = lblkno(dp->i_fs, dp->i_offset)((dp->i_offset) >> (dp->inode_u.fs)->fs_bshift ); |
2898 | offset = blkoff(dp->i_fs, dp->i_offset)((dp->i_offset) & (dp->inode_u.fs)->fs_qbmask); |
2899 | if (pagedep_lookup(dp, lbn, DEPALLOC0x0001, &pagedep) == 0) |
2900 | WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list)do { (&pagedep->pd_list)->wk_state |= 0x8000; do { if (((&pagedep->pd_list)->wk_list.le_next = (&bp-> b_dep)->lh_first) != ((void *)0)) (&bp->b_dep)-> lh_first->wk_list.le_prev = &(&pagedep->pd_list )->wk_list.le_next; (&bp->b_dep)->lh_first = (& pagedep->pd_list); (&pagedep->pd_list)->wk_list. le_prev = &(&bp->b_dep)->lh_first; } while (0); } while (0); |
2901 | dirrem->dm_pagedepdm_un.dmu_pagedep = pagedep; |
2902 | /* |
2903 | * Check for a diradd dependency for the same directory entry. |
2904 | * If present, then both dependencies become obsolete and can |
2905 | * be de-allocated. Check for an entry on both the pd_dirraddhd |
2906 | * list and the pd_pendinghd list. |
2907 | */ |
2908 | |
2909 | LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(offset)], da_pdlist)for((dap) = ((&pagedep->pd_diraddhd[(((offset) >> 2) % 6)])->lh_first); (dap)!= ((void *)0); (dap) = ((dap) ->da_pdlist.le_next)) |
2910 | if (dap->da_offset == offset) |
2911 | break; |
2912 | if (dap == NULL((void *)0)) { |
2913 | |
2914 | LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist)for((dap) = ((&pagedep->pd_pendinghd)->lh_first); ( dap)!= ((void *)0); (dap) = ((dap)->da_pdlist.le_next)) |
2915 | if (dap->da_offset == offset) |
2916 | break; |
2917 | if (dap == NULL((void *)0)) |
2918 | return (dirrem); |
2919 | } |
2920 | /* |
2921 | * Must be ATTACHED at this point. |
2922 | */ |
2923 | if ((dap->da_stateda_list.wk_state & ATTACHED0x0001) == 0) { |
2924 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2925 | panic("newdirrem: not ATTACHED"); |
2926 | } |
2927 | if (dap->da_newinum != ip->i_number) { |
2928 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
2929 | panic("newdirrem: inum %u should be %u", |
2930 | ip->i_number, dap->da_newinum); |
2931 | } |
2932 | /* |
2933 | * If we are deleting a changed name that never made it to disk, |
2934 | * then return the dirrem describing the previous inode (which |
2935 | * represents the inode currently referenced from this entry on disk). |
2936 | */ |
2937 | if ((dap->da_stateda_list.wk_state & DIRCHG0x0080) != 0) { |
2938 | *prevdirremp = dap->da_previousda_un.dau_previous; |
2939 | dap->da_stateda_list.wk_state &= ~DIRCHG0x0080; |
2940 | dap->da_pagedepda_un.dau_pagedep = pagedep; |
2941 | } |
2942 | /* |
2943 | * We are deleting an entry that never made it to disk. |
2944 | * Mark it COMPLETE so we can delete its inode immediately. |
2945 | */ |
2946 | dirrem->dm_statedm_list.wk_state |= COMPLETE0x0004; |
2947 | free_diradd(dap); |
2948 | return (dirrem); |
2949 | } |
2950 | |
2951 | /* |
2952 | * Directory entry change dependencies. |
2953 | * |
2954 | * Changing an existing directory entry requires that an add operation |
2955 | * be completed first followed by a deletion. The semantics for the addition |
2956 | * are identical to the description of adding a new entry above except |
2957 | * that the rollback is to the old inode number rather than zero. Once |
2958 | * the addition dependency is completed, the removal is done as described |
2959 | * in the removal routine above. |
2960 | */ |
2961 | |
2962 | /* |
2963 | * This routine should be called immediately after changing |
2964 | * a directory entry. The inode's link count should not be |
2965 | * decremented by the calling procedure -- the soft updates |
2966 | * code will perform this task when it is safe. |
2967 | */ |
2968 | /* buffer containing directory block */ |
2969 | /* inode for the directory being modified */ |
2970 | /* inode for directory entry being removed */ |
2971 | /* new inode number for changed entry */ |
2972 | /* indicates if doing RMDIR */ |
2973 | void |
2974 | softdep_setup_directory_change(struct buf *bp, struct inode *dp, |
2975 | struct inode *ip, long newinum, int isrmdir) |
2976 | { |
2977 | int offset; |
2978 | struct diradd *dap; |
2979 | struct dirrem *dirrem, *prevdirrem; |
2980 | struct pagedep *pagedep; |
2981 | struct inodedep *inodedep; |
2982 | |
2983 | offset = blkoff(dp->i_fs, dp->i_offset)((dp->i_offset) & (dp->inode_u.fs)->fs_qbmask); |
2984 | dap = pool_get(&diradd_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
2985 | dap->da_list.wk_type = D_DIRADD10; |
2986 | dap->da_stateda_list.wk_state = DIRCHG0x0080 | ATTACHED0x0001 | DEPCOMPLETE0x0008; |
2987 | dap->da_offset = offset; |
2988 | dap->da_newinum = newinum; |
2989 | |
2990 | /* |
2991 | * Allocate a new dirrem and ACQUIRE_LOCK. |
2992 | */ |
2993 | dirrem = newdirrem(bp, dp, ip, isrmdir, &prevdirrem); |
2994 | pagedep = dirrem->dm_pagedepdm_un.dmu_pagedep; |
2995 | /* |
2996 | * The possible values for isrmdir: |
2997 | * 0 - non-directory file rename |
2998 | * 1 - directory rename within same directory |
2999 | * inum - directory rename to new directory of given inode number |
3000 | * When renaming to a new directory, we are both deleting and |
3001 | * creating a new directory entry, so the link count on the new |
3002 | * directory should not change. Thus we do not need the followup |
3003 | * dirrem which is usually done in handle_workitem_remove. We set |
3004 | * the DIRCHG flag to tell handle_workitem_remove to skip the |
3005 | * followup dirrem. |
3006 | */ |
3007 | if (isrmdir > 1) |
3008 | dirrem->dm_statedm_list.wk_state |= DIRCHG0x0080; |
3009 | |
3010 | /* |
3011 | * If the COMPLETE flag is clear, then there were no active |
3012 | * entries and we want to roll back to the previous inode until |
3013 | * the new inode is committed to disk. If the COMPLETE flag is |
3014 | * set, then we have deleted an entry that never made it to disk. |
3015 | * If the entry we deleted resulted from a name change, then the old |
3016 | * inode reference still resides on disk. Any rollback that we do |
3017 | * needs to be to that old inode (returned to us in prevdirrem). If |
3018 | * the entry we deleted resulted from a create, then there is |
3019 | * no entry on the disk, so we want to roll back to zero rather |
3020 | * than the uncommitted inode. In either of the COMPLETE cases we |
3021 | * want to immediately free the unwritten and unreferenced inode. |
3022 | */ |
3023 | if ((dirrem->dm_statedm_list.wk_state & COMPLETE0x0004) == 0) { |
3024 | dap->da_previousda_un.dau_previous = dirrem; |
3025 | } else { |
3026 | if (prevdirrem != NULL((void *)0)) { |
3027 | dap->da_previousda_un.dau_previous = prevdirrem; |
3028 | } else { |
3029 | dap->da_stateda_list.wk_state &= ~DIRCHG0x0080; |
3030 | dap->da_pagedepda_un.dau_pagedep = pagedep; |
3031 | } |
3032 | dirrem->dm_dirinumdm_un.dmu_dirinum = pagedep->pd_ino; |
3033 | add_to_worklist(&dirrem->dm_list); |
3034 | } |
3035 | /* |
3036 | * Link into its inodedep. Put it on the id_bufwait list if the inode |
3037 | * is not yet written. If it is written, do the post-inode write |
3038 | * processing to put it on the id_pendinghd list. |
3039 | */ |
3040 | if (inodedep_lookup(dp->i_fsinode_u.fs, newinum, DEPALLOC0x0001, &inodedep) == 0 || |
3041 | (inodedep->id_stateid_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) { |
3042 | dap->da_stateda_list.wk_state |= COMPLETE0x0004; |
3043 | LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist)do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_pendinghd )->lh_first) != ((void *)0)) (&pagedep->pd_pendinghd )->lh_first->da_pdlist.le_prev = &(dap)->da_pdlist .le_next; (&pagedep->pd_pendinghd)->lh_first = (dap ); (dap)->da_pdlist.le_prev = &(&pagedep->pd_pendinghd )->lh_first; } while (0); |
3044 | WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list)do { (&dap->da_list)->wk_state |= 0x8000; do { if ( ((&dap->da_list)->wk_list.le_next = (&inodedep-> id_pendinghd)->lh_first) != ((void *)0)) (&inodedep-> id_pendinghd)->lh_first->wk_list.le_prev = &(&dap ->da_list)->wk_list.le_next; (&inodedep->id_pendinghd )->lh_first = (&dap->da_list); (&dap->da_list )->wk_list.le_prev = &(&inodedep->id_pendinghd) ->lh_first; } while (0); } while (0); |
3045 | } else { |
3046 | LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)],do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_diraddhd [(((offset) >> 2) % 6)])->lh_first) != ((void *)0)) ( &pagedep->pd_diraddhd[(((offset) >> 2) % 6)])-> lh_first->da_pdlist.le_prev = &(dap)->da_pdlist.le_next ; (&pagedep->pd_diraddhd[(((offset) >> 2) % 6)]) ->lh_first = (dap); (dap)->da_pdlist.le_prev = &(& pagedep->pd_diraddhd[(((offset) >> 2) % 6)])->lh_first ; } while (0) |
3047 | dap, da_pdlist)do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_diraddhd [(((offset) >> 2) % 6)])->lh_first) != ((void *)0)) ( &pagedep->pd_diraddhd[(((offset) >> 2) % 6)])-> lh_first->da_pdlist.le_prev = &(dap)->da_pdlist.le_next ; (&pagedep->pd_diraddhd[(((offset) >> 2) % 6)]) ->lh_first = (dap); (dap)->da_pdlist.le_prev = &(& pagedep->pd_diraddhd[(((offset) >> 2) % 6)])->lh_first ; } while (0); |
3048 | WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list)do { (&dap->da_list)->wk_state |= 0x8000; do { if ( ((&dap->da_list)->wk_list.le_next = (&inodedep-> id_bufwait)->lh_first) != ((void *)0)) (&inodedep-> id_bufwait)->lh_first->wk_list.le_prev = &(&dap ->da_list)->wk_list.le_next; (&inodedep->id_bufwait )->lh_first = (&dap->da_list); (&dap->da_list )->wk_list.le_prev = &(&inodedep->id_bufwait)-> lh_first; } while (0); } while (0); |
3049 | } |
3050 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3051 | } |
3052 | |
3053 | /* |
3054 | * Called whenever the link count on an inode is changed. |
3055 | * It creates an inode dependency so that the new reference(s) |
3056 | * to the inode cannot be committed to disk until the updated |
3057 | * inode has been written. |
3058 | */ |
3059 | /* the inode with the increased link count */ |
3060 | /* do background work or not */ |
3061 | void |
3062 | softdep_change_linkcnt(struct inode *ip, int nodelay) |
3063 | { |
3064 | struct inodedep *inodedep; |
3065 | int flags; |
3066 | |
3067 | /* |
3068 | * If requested, do not allow background work to happen. |
3069 | */ |
3070 | flags = DEPALLOC0x0001; |
3071 | if (nodelay) |
3072 | flags |= NODELAY0x0002; |
3073 | |
3074 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3075 | |
3076 | (void) inodedep_lookup(ip->i_fsinode_u.fs, ip->i_number, flags, &inodedep); |
3077 | if (DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink) < ip->i_effnlink) { |
3078 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3079 | panic("softdep_change_linkcnt: bad delta"); |
3080 | } |
3081 | |
3082 | inodedep->id_nlinkdelta = DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink) - ip->i_effnlink; |
3083 | |
3084 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3085 | } |
3086 | |
3087 | /* |
3088 | * This workitem decrements the inode's link count. |
3089 | * If the link count reaches zero, the file is removed. |
3090 | */ |
3091 | STATIC void |
3092 | handle_workitem_remove(struct dirrem *dirrem) |
3093 | { |
3094 | struct proc *p = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc; /* XXX */ |
3095 | struct inodedep *inodedep; |
3096 | struct vnode *vp; |
3097 | struct inode *ip; |
3098 | ufsino_t oldinum; |
3099 | int error; |
3100 | |
3101 | if ((error = VFS_VGET(dirrem->dm_mnt, dirrem->dm_oldinum, &vp)(*(dirrem->dm_mnt)->mnt_op->vfs_vget)(dirrem->dm_mnt , dirrem->dm_oldinum, &vp)) != 0) { |
3102 | softdep_error("handle_workitem_remove: vget", error); |
3103 | return; |
3104 | } |
3105 | ip = VTOI(vp)((struct inode *)(vp)->v_data); |
3106 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3107 | if ((inodedep_lookup(ip->i_fsinode_u.fs, dirrem->dm_oldinum, 0, &inodedep)) |
3108 | == 0) { |
3109 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3110 | panic("handle_workitem_remove: lost inodedep"); |
3111 | } |
3112 | /* |
3113 | * Normal file deletion. |
3114 | */ |
3115 | if ((dirrem->dm_statedm_list.wk_state & RMDIR0x0040) == 0) { |
3116 | DIP_ADD(ip, nlink, -1)do { if ((ip)->i_ump->um_fstype == 1) (ip)->dinode_u .ffs1_din->di_nlink += (-1); else (ip)->dinode_u.ffs2_din ->di_nlink += (-1); } while (0); |
3117 | ip->i_flag |= IN_CHANGE0x0002; |
3118 | if (DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink) < ip->i_effnlink) { |
3119 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3120 | panic("handle_workitem_remove: bad file delta"); |
3121 | } |
3122 | inodedep->id_nlinkdelta = DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink) - ip->i_effnlink; |
3123 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3124 | vput(vp); |
3125 | num_dirrem -= 1; |
3126 | WORKITEM_FREE(dirrem, D_DIRREM)softdep_freequeue_add((struct worklist *)dirrem); |
3127 | return; |
3128 | } |
3129 | /* |
3130 | * Directory deletion. Decrement reference count for both the |
3131 | * just deleted parent directory entry and the reference for ".". |
3132 | * Next truncate the directory to length zero. When the |
3133 | * truncation completes, arrange to have the reference count on |
3134 | * the parent decremented to account for the loss of "..". |
3135 | */ |
3136 | DIP_ADD(ip, nlink, -2)do { if ((ip)->i_ump->um_fstype == 1) (ip)->dinode_u .ffs1_din->di_nlink += (-2); else (ip)->dinode_u.ffs2_din ->di_nlink += (-2); } while (0); |
3137 | ip->i_flag |= IN_CHANGE0x0002; |
3138 | if (DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink) < ip->i_effnlink) |
3139 | panic("handle_workitem_remove: bad dir delta"); |
3140 | inodedep->id_nlinkdelta = DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink) - ip->i_effnlink; |
3141 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3142 | if ((error = UFS_TRUNCATE(ip, (off_t)0, 0, p->p_ucred)((ip)->i_vtbl->iv_truncate)((ip), ((off_t)0), (0), (p-> p_ucred))) != 0) |
3143 | softdep_error("handle_workitem_remove: truncate", error); |
3144 | /* |
3145 | * Rename a directory to a new parent. Since, we are both deleting |
3146 | * and creating a new directory entry, the link count on the new |
3147 | * directory should not change. Thus we skip the followup dirrem. |
3148 | */ |
3149 | if (dirrem->dm_statedm_list.wk_state & DIRCHG0x0080) { |
3150 | vput(vp); |
3151 | num_dirrem -= 1; |
3152 | WORKITEM_FREE(dirrem, D_DIRREM)softdep_freequeue_add((struct worklist *)dirrem); |
3153 | return; |
3154 | } |
3155 | /* |
3156 | * If the inodedep does not exist, then the zero'ed inode has |
3157 | * been written to disk. If the allocated inode has never been |
3158 | * written to disk, then the on-disk inode is zero'ed. In either |
3159 | * case we can remove the file immediately. |
3160 | */ |
3161 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3162 | dirrem->dm_statedm_list.wk_state = 0; |
3163 | oldinum = dirrem->dm_oldinum; |
3164 | dirrem->dm_oldinum = dirrem->dm_dirinumdm_un.dmu_dirinum; |
3165 | if (inodedep_lookup(ip->i_fsinode_u.fs, oldinum, 0, &inodedep) == 0 || |
3166 | check_inode_unwritten(inodedep)) { |
3167 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3168 | vput(vp); |
3169 | handle_workitem_remove(dirrem); |
3170 | return; |
3171 | } |
3172 | WORKLIST_INSERT(&inodedep->id_inowait, &dirrem->dm_list)do { (&dirrem->dm_list)->wk_state |= 0x8000; do { if (((&dirrem->dm_list)->wk_list.le_next = (&inodedep ->id_inowait)->lh_first) != ((void *)0)) (&inodedep ->id_inowait)->lh_first->wk_list.le_prev = &(& dirrem->dm_list)->wk_list.le_next; (&inodedep->id_inowait )->lh_first = (&dirrem->dm_list); (&dirrem-> dm_list)->wk_list.le_prev = &(&inodedep->id_inowait )->lh_first; } while (0); } while (0); |
3173 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3174 | ip->i_flag |= IN_CHANGE0x0002; |
3175 | UFS_UPDATE(VTOI(vp), 0)((((struct inode *)(vp)->v_data))->i_vtbl->iv_update )((((struct inode *)(vp)->v_data)), (0)); |
3176 | vput(vp); |
3177 | } |
3178 | |
3179 | /* |
3180 | * Inode de-allocation dependencies. |
3181 | * |
3182 | * When an inode's link count is reduced to zero, it can be de-allocated. We |
3183 | * found it convenient to postpone de-allocation until after the inode is |
3184 | * written to disk with its new link count (zero). At this point, all of the |
3185 | * on-disk inode's block pointers are nullified and, with careful dependency |
3186 | * list ordering, all dependencies related to the inode will be satisfied and |
3187 | * the corresponding dependency structures de-allocated. So, if/when the |
3188 | * inode is reused, there will be no mixing of old dependencies with new |
3189 | * ones. This artificial dependency is set up by the block de-allocation |
3190 | * procedure above (softdep_setup_freeblocks) and completed by the |
3191 | * following procedure. |
3192 | */ |
3193 | STATIC void |
3194 | handle_workitem_freefile(struct freefile *freefile) |
3195 | { |
3196 | struct fs *fs; |
3197 | struct vnode vp; |
3198 | struct inode tip; |
3199 | #ifdef DEBUG |
3200 | struct inodedep *idp; |
3201 | #endif |
3202 | int error; |
3203 | |
3204 | fs = VFSTOUFS(freefile->fx_mnt)((struct ufsmount *)((freefile->fx_mnt)->mnt_data))->um_fsufsmount_u.fs; |
3205 | #ifdef DEBUG |
3206 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3207 | error = inodedep_lookup(fs, freefile->fx_oldinum, 0, &idp); |
3208 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3209 | if (error) |
3210 | panic("handle_workitem_freefile: inodedep survived"); |
3211 | #endif |
3212 | tip.i_ump = VFSTOUFS(freefile->fx_mnt)((struct ufsmount *)((freefile->fx_mnt)->mnt_data)); |
3213 | tip.i_dev = freefile->fx_devvp->v_rdevv_un.vu_specinfo->si_rdev; |
3214 | tip.i_fsinode_u.fs = fs; |
3215 | tip.i_vnode = &vp; |
3216 | vp.v_data = &tip; |
3217 | |
3218 | if ((error = ffs_freefile(&tip, freefile->fx_oldinum, |
3219 | freefile->fx_mode)) != 0) { |
3220 | softdep_error("handle_workitem_freefile", error); |
3221 | } |
3222 | WORKITEM_FREE(freefile, D_FREEFILE)softdep_freequeue_add((struct worklist *)freefile); |
3223 | } |
3224 | |
3225 | /* |
3226 | * Disk writes. |
3227 | * |
3228 | * The dependency structures constructed above are most actively used when file |
3229 | * system blocks are written to disk. No constraints are placed on when a |
3230 | * block can be written, but unsatisfied update dependencies are made safe by |
3231 | * modifying (or replacing) the source memory for the duration of the disk |
3232 | * write. When the disk write completes, the memory block is again brought |
3233 | * up-to-date. |
3234 | * |
3235 | * In-core inode structure reclamation. |
3236 | * |
3237 | * Because there are a finite number of "in-core" inode structures, they are |
3238 | * reused regularly. By transferring all inode-related dependencies to the |
3239 | * in-memory inode block and indexing them separately (via "inodedep"s), we |
3240 | * can allow "in-core" inode structures to be reused at any time and avoid |
3241 | * any increase in contention. |
3242 | * |
3243 | * Called just before entering the device driver to initiate a new disk I/O. |
3244 | * The buffer must be locked, thus, no I/O completion operations can occur |
3245 | * while we are manipulating its associated dependencies. |
3246 | */ |
3247 | /* structure describing disk write to occur */ |
3248 | void |
3249 | softdep_disk_io_initiation(struct buf *bp) |
3250 | { |
3251 | struct worklist *wk, *nextwk; |
3252 | struct indirdep *indirdep; |
3253 | struct inodedep *inodedep; |
3254 | struct buf *sbp; |
3255 | |
3256 | /* |
3257 | * We only care about write operations. There should never |
3258 | * be dependencies for reads. |
3259 | */ |
3260 | if (bp->b_flags & B_READ0x00008000) |
3261 | panic("softdep_disk_io_initiation: read"); |
3262 | |
3263 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3264 | |
3265 | /* |
3266 | * Do any necessary pre-I/O processing. |
3267 | */ |
3268 | for (wk = LIST_FIRST(&bp->b_dep)((&bp->b_dep)->lh_first); wk; wk = nextwk) { |
3269 | nextwk = LIST_NEXT(wk, wk_list)((wk)->wk_list.le_next); |
3270 | switch (wk->wk_type) { |
3271 | |
3272 | case D_PAGEDEP0: |
3273 | initiate_write_filepage(WK_PAGEDEP(wk)((struct pagedep *)(wk)), bp); |
3274 | continue; |
3275 | |
3276 | case D_INODEDEP1: |
3277 | inodedep = WK_INODEDEP(wk)((struct inodedep *)(wk)); |
3278 | if (inodedep->id_fs->fs_magic == FS_UFS1_MAGIC0x011954) |
3279 | initiate_write_inodeblock_ufs1(inodedep, bp); |
3280 | #ifdef FFS21 |
3281 | else |
3282 | initiate_write_inodeblock_ufs2(inodedep, bp); |
3283 | #endif |
3284 | continue; |
3285 | |
3286 | case D_INDIRDEP5: |
3287 | indirdep = WK_INDIRDEP(wk)((struct indirdep *)(wk)); |
3288 | if (indirdep->ir_stateir_list.wk_state & GOINGAWAY0x0100) |
3289 | panic("disk_io_initiation: indirdep gone"); |
3290 | /* |
3291 | * If there are no remaining dependencies, this |
3292 | * will be writing the real pointers, so the |
3293 | * dependency can be freed. |
3294 | */ |
3295 | if (LIST_FIRST(&indirdep->ir_deplisthd)((&indirdep->ir_deplisthd)->lh_first) == NULL((void *)0)) { |
3296 | sbp = indirdep->ir_savebp; |
3297 | sbp->b_flags |= B_INVAL0x00000800 | B_NOCACHE0x00001000; |
3298 | /* inline expand WORKLIST_REMOVE(wk); */ |
3299 | wk->wk_state &= ~ONWORKLIST0x8000; |
3300 | LIST_REMOVE(wk, wk_list)do { if ((wk)->wk_list.le_next != ((void *)0)) (wk)->wk_list .le_next->wk_list.le_prev = (wk)->wk_list.le_prev; *(wk )->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list .le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0); |
3301 | WORKITEM_FREE(indirdep, D_INDIRDEP)softdep_freequeue_add((struct worklist *)indirdep); |
3302 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3303 | brelse(sbp); |
3304 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3305 | continue; |
3306 | } |
3307 | /* |
3308 | * Replace up-to-date version with safe version. |
3309 | */ |
3310 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3311 | indirdep->ir_saveddata = malloc(bp->b_bcount, |
3312 | M_INDIRDEP83, M_WAITOK0x0001); |
3313 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3314 | indirdep->ir_stateir_list.wk_state &= ~ATTACHED0x0001; |
3315 | indirdep->ir_stateir_list.wk_state |= UNDONE0x0002; |
3316 | memcpy(indirdep->ir_saveddata, bp->b_data, bp->b_bcount)__builtin_memcpy((indirdep->ir_saveddata), (bp->b_data) , (bp->b_bcount)); |
3317 | memcpy(bp->b_data, indirdep->ir_savebp->b_data,__builtin_memcpy((bp->b_data), (indirdep->ir_savebp-> b_data), (bp->b_bcount)) |
3318 | bp->b_bcount)__builtin_memcpy((bp->b_data), (indirdep->ir_savebp-> b_data), (bp->b_bcount)); |
3319 | continue; |
3320 | |
3321 | case D_MKDIR11: |
3322 | case D_BMSAFEMAP3: |
3323 | case D_ALLOCDIRECT4: |
3324 | case D_ALLOCINDIR6: |
3325 | continue; |
3326 | |
3327 | default: |
3328 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3329 | panic("handle_disk_io_initiation: Unexpected type %s", |
3330 | TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
3331 | /* NOTREACHED */ |
3332 | } |
3333 | } |
3334 | |
3335 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3336 | } |
3337 | |
3338 | /* |
3339 | * Called from within the procedure above to deal with unsatisfied |
3340 | * allocation dependencies in a directory. The buffer must be locked, |
3341 | * thus, no I/O completion operations can occur while we are |
3342 | * manipulating its associated dependencies. |
3343 | */ |
3344 | STATIC void |
3345 | initiate_write_filepage(struct pagedep *pagedep, struct buf *bp) |
3346 | { |
3347 | struct diradd *dap; |
3348 | struct direct *ep; |
3349 | int i; |
3350 | |
3351 | if (pagedep->pd_statepd_list.wk_state & IOSTARTED0x0200) { |
3352 | /* |
3353 | * This can only happen if there is a driver that does not |
3354 | * understand chaining. Here biodone will reissue the call |
3355 | * to strategy for the incomplete buffers. |
3356 | */ |
3357 | printf("initiate_write_filepage: already started\n"); |
3358 | return; |
3359 | } |
3360 | pagedep->pd_statepd_list.wk_state |= IOSTARTED0x0200; |
3361 | for (i = 0; i < DAHASHSZ6; i++) { |
3362 | LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist)for((dap) = ((&pagedep->pd_diraddhd[i])->lh_first); (dap)!= ((void *)0); (dap) = ((dap)->da_pdlist.le_next)) { |
3363 | ep = (struct direct *) |
3364 | ((char *)bp->b_data + dap->da_offset); |
3365 | if (ep->d_ino != dap->da_newinum) { |
3366 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3367 | panic("%s: dir inum %u != new %u", |
3368 | "initiate_write_filepage", |
3369 | ep->d_ino, dap->da_newinum); |
3370 | } |
3371 | if (dap->da_stateda_list.wk_state & DIRCHG0x0080) |
3372 | ep->d_ino = dap->da_previousda_un.dau_previous->dm_oldinum; |
3373 | else |
3374 | ep->d_ino = 0; |
3375 | dap->da_stateda_list.wk_state &= ~ATTACHED0x0001; |
3376 | dap->da_stateda_list.wk_state |= UNDONE0x0002; |
3377 | } |
3378 | } |
3379 | } |
3380 | |
3381 | /* |
3382 | * Called from within the procedure above to deal with unsatisfied |
3383 | * allocation dependencies in an inodeblock. The buffer must be |
3384 | * locked, thus, no I/O completion operations can occur while we |
3385 | * are manipulating its associated dependencies. |
3386 | */ |
3387 | /* The inode block */ |
3388 | STATIC void |
3389 | initiate_write_inodeblock_ufs1(struct inodedep *inodedep, struct buf *bp) |
3390 | { |
3391 | struct allocdirect *adp, *lastadp; |
3392 | struct ufs1_dinode *dp; |
3393 | struct fs *fs; |
3394 | #ifdef DIAGNOSTIC1 |
3395 | daddr_t prevlbn = 0; |
3396 | int32_t d1, d2; |
3397 | #endif |
3398 | int i, deplist; |
3399 | |
3400 | if (inodedep->id_stateid_list.wk_state & IOSTARTED0x0200) { |
3401 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3402 | panic("initiate_write_inodeblock: already started"); |
3403 | } |
3404 | inodedep->id_stateid_list.wk_state |= IOSTARTED0x0200; |
3405 | fs = inodedep->id_fs; |
3406 | dp = (struct ufs1_dinode *)bp->b_data + |
3407 | ino_to_fsbo(fs, inodedep->id_ino)((inodedep->id_ino) % ((fs)->fs_inopb)); |
3408 | /* |
3409 | * If the bitmap is not yet written, then the allocated |
3410 | * inode cannot be written to disk. |
3411 | */ |
3412 | if ((inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008) == 0) { |
3413 | if (inodedep->id_savedino1id_un.idu_savedino1 != NULL((void *)0)) { |
3414 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3415 | panic("initiate_write_inodeblock: already doing I/O"); |
3416 | } |
3417 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3418 | inodedep->id_savedino1id_un.idu_savedino1 = malloc(sizeof(struct ufs1_dinode), |
3419 | M_INODEDEP79, M_WAITOK0x0001); |
3420 | inodedep->id_unsize = sizeof(struct ufs1_dinode); |
3421 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
3422 | *inodedep->id_savedino1id_un.idu_savedino1 = *dp; |
3423 | memset(dp, 0, sizeof(struct ufs1_dinode))__builtin_memset((dp), (0), (sizeof(struct ufs1_dinode))); |
3424 | return; |
3425 | } |
3426 | /* |
3427 | * If no dependencies, then there is nothing to roll back. |
3428 | */ |
3429 | inodedep->id_savedsize = dp->di_size; |
3430 | if (TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) == NULL((void *)0)) |
3431 | return; |
3432 | /* |
3433 | * Set the dependencies to busy. |
3434 | */ |
3435 | for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first); adp; |
3436 | adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) { |
3437 | #ifdef DIAGNOSTIC1 |
3438 | if (deplist != 0 && prevlbn >= adp->ad_lbn) { |
3439 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3440 | panic("softdep_write_inodeblock: lbn order"); |
3441 | } |
3442 | prevlbn = adp->ad_lbn; |
3443 | if (adp->ad_lbn < NDADDR12 && |
3444 | (d1 = dp->di_db[adp->ad_lbn]) != (d2 = adp->ad_newblkno)) { |
3445 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3446 | panic("%s: direct pointer #%lld mismatch %d != %d", |
3447 | "softdep_write_inodeblock", (long long)adp->ad_lbn, |
3448 | d1, d2); |
3449 | } |
3450 | if (adp->ad_lbn >= NDADDR12 && |
3451 | (d1 = dp->di_ib[adp->ad_lbn - NDADDR12]) != |
3452 | (d2 = adp->ad_newblkno)) { |
3453 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3454 | panic("%s: indirect pointer #%lld mismatch %d != %d", |
3455 | "softdep_write_inodeblock", (long long)(adp->ad_lbn - |
3456 | NDADDR12), d1, d2); |
3457 | } |
3458 | deplist |= 1 << adp->ad_lbn; |
3459 | if ((adp->ad_statead_list.wk_state & ATTACHED0x0001) == 0) { |
3460 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3461 | panic("softdep_write_inodeblock: Unknown state 0x%x", |
3462 | adp->ad_statead_list.wk_state); |
3463 | } |
3464 | #endif /* DIAGNOSTIC */ |
3465 | adp->ad_statead_list.wk_state &= ~ATTACHED0x0001; |
3466 | adp->ad_statead_list.wk_state |= UNDONE0x0002; |
3467 | } |
3468 | /* |
3469 | * The on-disk inode cannot claim to be any larger than the last |
3470 | * fragment that has been written. Otherwise, the on-disk inode |
3471 | * might have fragments that were not the last block in the file |
3472 | * which would corrupt the filesystem. |
3473 | */ |
3474 | for (lastadp = NULL((void *)0), adp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first); adp; |
3475 | lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) { |
3476 | if (adp->ad_lbn >= NDADDR12) |
3477 | break; |
3478 | dp->di_db[adp->ad_lbn] = adp->ad_oldblkno; |
3479 | /* keep going until hitting a rollback to a frag */ |
3480 | if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize) |
3481 | continue; |
3482 | dp->di_size = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize; |
3483 | for (i = adp->ad_lbn + 1; i < NDADDR12; i++) { |
3484 | #ifdef DIAGNOSTIC1 |
3485 | if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0) { |
3486 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3487 | panic("softdep_write_inodeblock: lost dep1"); |
3488 | } |
3489 | #endif /* DIAGNOSTIC */ |
3490 | dp->di_db[i] = 0; |
3491 | } |
3492 | for (i = 0; i < NIADDR3; i++) { |
3493 | #ifdef DIAGNOSTIC1 |
3494 | if (dp->di_ib[i] != 0 && |
3495 | (deplist & ((1 << NDADDR12) << i)) == 0) { |
3496 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3497 | panic("softdep_write_inodeblock: lost dep2"); |
3498 | } |
3499 | #endif /* DIAGNOSTIC */ |
3500 | dp->di_ib[i] = 0; |
3501 | } |
3502 | return; |
3503 | } |
3504 | /* |
3505 | * If we have zero'ed out the last allocated block of the file, |
3506 | * roll back the size to the last currently allocated block. |
3507 | * We know that this last allocated block is a full-sized as |
3508 | * we already checked for fragments in the loop above. |
3509 | */ |
3510 | if (lastadp != NULL((void *)0) && |
3511 | dp->di_size <= (lastadp->ad_lbn + 1) * fs->fs_bsize) { |
3512 | for (i = lastadp->ad_lbn; i >= 0; i--) |
3513 | if (dp->di_db[i] != 0) |
3514 | break; |
3515 | dp->di_size = (i + 1) * fs->fs_bsize; |
3516 | } |
3517 | /* |
3518 | * The only dependencies are for indirect blocks. |
3519 | * |
3520 | * The file size for indirect block additions is not guaranteed. |
3521 | * Such a guarantee would be non-trivial to achieve. The conventional |
3522 | * synchronous write implementation also does not make this guarantee. |
3523 | * Fsck should catch and fix discrepancies. Arguably, the file size |
3524 | * can be over-estimated without destroying integrity when the file |
3525 | * moves into the indirect blocks (i.e., is large). If we want to |
3526 | * postpone fsck, we are stuck with this argument. |
3527 | */ |
3528 | for (; adp; adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) |
3529 | dp->di_ib[adp->ad_lbn - NDADDR12] = 0; |
3530 | } |
3531 | |
3532 | #ifdef FFS21 |
3533 | /* |
3534 | * Version of initiate_write_inodeblock that handles FFS2 dinodes. |
3535 | */ |
3536 | /* The inode block */ |
3537 | STATIC void |
3538 | initiate_write_inodeblock_ufs2(struct inodedep *inodedep, struct buf *bp) |
3539 | { |
3540 | struct allocdirect *adp, *lastadp; |
3541 | struct ufs2_dinode *dp; |
3542 | struct fs *fs = inodedep->id_fs; |
3543 | #ifdef DIAGNOSTIC1 |
3544 | daddr_t prevlbn = -1, d1, d2; |
3545 | #endif |
3546 | int deplist, i; |
3547 | |
3548 | if (inodedep->id_stateid_list.wk_state & IOSTARTED0x0200) |
3549 | panic("initiate_write_inodeblock_ufs2: already started"); |
3550 | inodedep->id_stateid_list.wk_state |= IOSTARTED0x0200; |
3551 | fs = inodedep->id_fs; |
3552 | dp = (struct ufs2_dinode *)bp->b_data + |
3553 | ino_to_fsbo(fs, inodedep->id_ino)((inodedep->id_ino) % ((fs)->fs_inopb)); |
3554 | /* |
3555 | * If the bitmap is not yet written, then the allocated |
3556 | * inode cannot be written to disk. |
3557 | */ |
3558 | if ((inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008) == 0) { |
3559 | if (inodedep->id_savedino2id_un.idu_savedino2 != NULL((void *)0)) |
3560 | panic("initiate_write_inodeblock_ufs2: I/O underway"); |
3561 | inodedep->id_savedino2id_un.idu_savedino2 = malloc(sizeof(struct ufs2_dinode), |
3562 | M_INODEDEP79, M_WAITOK0x0001); |
3563 | inodedep->id_unsize = sizeof(struct ufs2_dinode); |
3564 | *inodedep->id_savedino2id_un.idu_savedino2 = *dp; |
3565 | memset(dp, 0, sizeof(struct ufs2_dinode))__builtin_memset((dp), (0), (sizeof(struct ufs2_dinode))); |
3566 | return; |
3567 | } |
3568 | /* |
3569 | * If no dependencies, then there is nothing to roll back. |
3570 | */ |
3571 | inodedep->id_savedsize = dp->di_size; |
3572 | if (TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) == NULL((void *)0)) |
3573 | return; |
3574 | |
3575 | #ifdef notyet |
3576 | inodedep->id_savedextsize = dp->di_extsize; |
3577 | if (TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) == NULL((void *)0) && |
3578 | TAILQ_FIRST(&inodedep->id_extupdt)((&inodedep->id_extupdt)->tqh_first) == NULL((void *)0)) |
3579 | return; |
3580 | /* |
3581 | * Set the ext data dependencies to busy. |
3582 | */ |
3583 | for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_extupdt)((&inodedep->id_extupdt)->tqh_first); adp; |
3584 | adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) { |
3585 | #ifdef DIAGNOSTIC1 |
3586 | if (deplist != 0 && prevlbn >= adp->ad_lbn) { |
3587 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3588 | panic("softdep_write_inodeblock: lbn order"); |
3589 | } |
3590 | prevlbn = adp->ad_lbn; |
3591 | if ((d1 = dp->di_extb[adp->ad_lbn]) != |
3592 | (d2 = adp->ad_newblkno)) { |
3593 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3594 | panic("%s: direct pointer #%lld mismatch %lld != %lld", |
3595 | "softdep_write_inodeblock", (long long)adp->ad_lbn, |
3596 | d1, d2); |
3597 | } |
3598 | deplist |= 1 << adp->ad_lbn; |
3599 | if ((adp->ad_statead_list.wk_state & ATTACHED0x0001) == 0) { |
3600 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3601 | panic("softdep_write_inodeblock: Unknown state 0x%x", |
3602 | adp->ad_statead_list.wk_state); |
3603 | } |
3604 | #endif /* DIAGNOSTIC */ |
3605 | adp->ad_statead_list.wk_state &= ~ATTACHED0x0001; |
3606 | adp->ad_statead_list.wk_state |= UNDONE0x0002; |
3607 | } |
3608 | /* |
3609 | * The on-disk inode cannot claim to be any larger than the last |
3610 | * fragment that has been written. Otherwise, the on-disk inode |
3611 | * might have fragments that were not the last block in the ext |
3612 | * data which would corrupt the filesystem. |
3613 | */ |
3614 | for (lastadp = NULL((void *)0), adp = TAILQ_FIRST(&inodedep->id_extupdt)((&inodedep->id_extupdt)->tqh_first); adp; |
3615 | lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) { |
3616 | dp->di_extb[adp->ad_lbn] = adp->ad_oldblkno; |
3617 | /* keep going until hitting a rollback to a frag */ |
3618 | if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize) |
3619 | continue; |
3620 | dp->di_extsize = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize; |
3621 | for (i = adp->ad_lbn + 1; i < NXADDR2; i++) { |
3622 | #ifdef DIAGNOSTIC1 |
3623 | if (dp->di_extb[i] != 0 && (deplist & (1 << i)) == 0) { |
3624 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3625 | panic("softdep_write_inodeblock: lost dep1"); |
3626 | } |
3627 | #endif /* DIAGNOSTIC */ |
3628 | dp->di_extb[i] = 0; |
3629 | } |
3630 | lastadp = NULL((void *)0); |
3631 | break; |
3632 | } |
3633 | /* |
3634 | * If we have zero'ed out the last allocated block of the ext |
3635 | * data, roll back the size to the last currently allocated block. |
3636 | * We know that this last allocated block is a full-sized as |
3637 | * we already checked for fragments in the loop above. |
3638 | */ |
3639 | if (lastadp != NULL((void *)0) && |
3640 | dp->di_extsize <= (lastadp->ad_lbn + 1) * fs->fs_bsize) { |
3641 | for (i = lastadp->ad_lbn; i >= 0; i--) |
3642 | if (dp->di_extb[i] != 0) |
3643 | break; |
3644 | dp->di_extsize = (i + 1) * fs->fs_bsize; |
3645 | } |
3646 | #endif /* notyet */ |
3647 | |
3648 | /* |
3649 | * Set the file data dependencies to busy. |
3650 | */ |
3651 | for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first); adp; |
3652 | adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) { |
3653 | #ifdef DIAGNOSTIC1 |
3654 | if (deplist != 0 && prevlbn >= adp->ad_lbn) { |
3655 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3656 | panic("softdep_write_inodeblock: lbn order"); |
3657 | } |
3658 | prevlbn = adp->ad_lbn; |
3659 | if (adp->ad_lbn < NDADDR12 && |
3660 | (d1 = dp->di_db[adp->ad_lbn]) != (d2 = adp->ad_newblkno)) { |
3661 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3662 | panic("%s: direct pointer #%lld mismatch %lld != %lld", |
3663 | "softdep_write_inodeblock", (long long)adp->ad_lbn, |
3664 | d1, d2); |
3665 | } |
3666 | if (adp->ad_lbn >= NDADDR12 && |
3667 | (d1 = dp->di_ib[adp->ad_lbn - NDADDR12]) != |
3668 | (d2 = adp->ad_newblkno)) { |
3669 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3670 | panic("%s: indirect pointer #%lld mismatch %lld != %lld", |
3671 | "softdep_write_inodeblock", (long long)(adp->ad_lbn - |
3672 | NDADDR12), d1, d2); |
3673 | } |
3674 | deplist |= 1 << adp->ad_lbn; |
3675 | if ((adp->ad_statead_list.wk_state & ATTACHED0x0001) == 0) { |
3676 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3677 | panic("softdep_write_inodeblock: Unknown state 0x%x", |
3678 | adp->ad_statead_list.wk_state); |
3679 | } |
3680 | #endif /* DIAGNOSTIC */ |
3681 | adp->ad_statead_list.wk_state &= ~ATTACHED0x0001; |
3682 | adp->ad_statead_list.wk_state |= UNDONE0x0002; |
3683 | } |
3684 | /* |
3685 | * The on-disk inode cannot claim to be any larger than the last |
3686 | * fragment that has been written. Otherwise, the on-disk inode |
3687 | * might have fragments that were not the last block in the file |
3688 | * which would corrupt the filesystem. |
3689 | */ |
3690 | for (lastadp = NULL((void *)0), adp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first); adp; |
3691 | lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) { |
3692 | if (adp->ad_lbn >= NDADDR12) |
3693 | break; |
3694 | dp->di_db[adp->ad_lbn] = adp->ad_oldblkno; |
3695 | /* keep going until hitting a rollback to a frag */ |
3696 | if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize) |
3697 | continue; |
3698 | dp->di_size = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize; |
3699 | for (i = adp->ad_lbn + 1; i < NDADDR12; i++) { |
3700 | #ifdef DIAGNOSTIC1 |
3701 | if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0) { |
3702 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3703 | panic("softdep_write_inodeblock: lost dep2"); |
3704 | } |
3705 | #endif /* DIAGNOSTIC */ |
3706 | dp->di_db[i] = 0; |
3707 | } |
3708 | for (i = 0; i < NIADDR3; i++) { |
3709 | #ifdef DIAGNOSTIC1 |
3710 | if (dp->di_ib[i] != 0 && |
3711 | (deplist & ((1 << NDADDR12) << i)) == 0) { |
3712 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
3713 | panic("softdep_write_inodeblock: lost dep3"); |
3714 | } |
3715 | #endif /* DIAGNOSTIC */ |
3716 | dp->di_ib[i] = 0; |
3717 | } |
3718 | return; |
3719 | } |
3720 | /* |
3721 | * If we have zero'ed out the last allocated block of the file, |
3722 | * roll back the size to the last currently allocated block. |
3723 | * We know that this last allocated block is a full-sized as |
3724 | * we already checked for fragments in the loop above. |
3725 | */ |
3726 | if (lastadp != NULL((void *)0) && |
3727 | dp->di_size <= (lastadp->ad_lbn + 1) * fs->fs_bsize) { |
3728 | for (i = lastadp->ad_lbn; i >= 0; i--) |
3729 | if (dp->di_db[i] != 0) |
3730 | break; |
3731 | dp->di_size = (i + 1) * fs->fs_bsize; |
3732 | } |
3733 | /* |
3734 | * The only dependencies are for indirect blocks. |
3735 | * |
3736 | * The file size for indirect block additions is not guaranteed. |
3737 | * Such a guarantee would be non-trivial to achieve. The conventional |
3738 | * synchronous write implementation also does not make this guarantee. |
3739 | * Fsck should catch and fix discrepancies. Arguably, the file size |
3740 | * can be over-estimated without destroying integrity when the file |
3741 | * moves into the indirect blocks (i.e., is large). If we want to |
3742 | * postpone fsck, we are stuck with this argument. |
3743 | */ |
3744 | for (; adp; adp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next)) |
3745 | dp->di_ib[adp->ad_lbn - NDADDR12] = 0; |
3746 | } |
3747 | #endif /* FFS2 */ |
3748 | |
3749 | /* |
3750 | * This routine is called during the completion interrupt |
3751 | * service routine for a disk write (from the procedure called |
3752 | * by the device driver to inform the file system caches of |
3753 | * a request completion). It should be called early in this |
3754 | * procedure, before the block is made available to other |
3755 | * processes or other routines are called. |
3756 | */ |
3757 | /* describes the completed disk write */ |
3758 | void |
3759 | softdep_disk_write_complete(struct buf *bp) |
3760 | { |
3761 | struct worklist *wk; |
3762 | struct workhead reattach; |
3763 | struct newblk *newblk; |
3764 | struct allocindir *aip; |
3765 | struct allocdirect *adp; |
3766 | struct indirdep *indirdep; |
3767 | struct inodedep *inodedep; |
3768 | struct bmsafemap *bmsafemap; |
3769 | |
3770 | /* |
3771 | * If an error occurred while doing the write, then the data |
3772 | * has not hit the disk and the dependencies cannot be unrolled. |
3773 | */ |
3774 | if ((bp->b_flags & B_ERROR0x00000400) && !(bp->b_flags & B_INVAL0x00000800)) |
3775 | return; |
3776 | |
3777 | #ifdef DEBUG |
3778 | if (lk.lkt_held != -1) |
3779 | panic("softdep_disk_write_complete: lock is held"); |
3780 | lk.lkt_held = -2; |
3781 | #endif |
3782 | LIST_INIT(&reattach)do { ((&reattach)->lh_first) = ((void *)0); } while (0 ); |
3783 | while ((wk = LIST_FIRST(&bp->b_dep)((&bp->b_dep)->lh_first)) != NULL((void *)0)) { |
3784 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
3785 | switch (wk->wk_type) { |
3786 | |
3787 | case D_PAGEDEP0: |
3788 | if (handle_written_filepage(WK_PAGEDEP(wk)((struct pagedep *)(wk)), bp)) |
3789 | WORKLIST_INSERT(&reattach, wk)do { (wk)->wk_state |= 0x8000; do { if (((wk)->wk_list. le_next = (&reattach)->lh_first) != ((void *)0)) (& reattach)->lh_first->wk_list.le_prev = &(wk)->wk_list .le_next; (&reattach)->lh_first = (wk); (wk)->wk_list .le_prev = &(&reattach)->lh_first; } while (0); } while (0); |
3790 | continue; |
3791 | |
3792 | case D_INODEDEP1: |
3793 | if (handle_written_inodeblock(WK_INODEDEP(wk)((struct inodedep *)(wk)), bp)) |
3794 | WORKLIST_INSERT(&reattach, wk)do { (wk)->wk_state |= 0x8000; do { if (((wk)->wk_list. le_next = (&reattach)->lh_first) != ((void *)0)) (& reattach)->lh_first->wk_list.le_prev = &(wk)->wk_list .le_next; (&reattach)->lh_first = (wk); (wk)->wk_list .le_prev = &(&reattach)->lh_first; } while (0); } while (0); |
3795 | continue; |
3796 | |
3797 | case D_BMSAFEMAP3: |
3798 | bmsafemap = WK_BMSAFEMAP(wk)((struct bmsafemap *)(wk)); |
3799 | while ((newblk = LIST_FIRST(&bmsafemap->sm_newblkhd)((&bmsafemap->sm_newblkhd)->lh_first))) { |
3800 | newblk->nb_state |= DEPCOMPLETE0x0008; |
3801 | newblk->nb_bmsafemap = NULL((void *)0); |
3802 | LIST_REMOVE(newblk, nb_deps)do { if ((newblk)->nb_deps.le_next != ((void *)0)) (newblk )->nb_deps.le_next->nb_deps.le_prev = (newblk)->nb_deps .le_prev; *(newblk)->nb_deps.le_prev = (newblk)->nb_deps .le_next; ((newblk)->nb_deps.le_prev) = ((void *)-1); ((newblk )->nb_deps.le_next) = ((void *)-1); } while (0); |
3803 | } |
3804 | while ((adp = |
3805 | LIST_FIRST(&bmsafemap->sm_allocdirecthd)((&bmsafemap->sm_allocdirecthd)->lh_first))) { |
3806 | adp->ad_statead_list.wk_state |= DEPCOMPLETE0x0008; |
3807 | adp->ad_buf = NULL((void *)0); |
3808 | LIST_REMOVE(adp, ad_deps)do { if ((adp)->ad_deps.le_next != ((void *)0)) (adp)-> ad_deps.le_next->ad_deps.le_prev = (adp)->ad_deps.le_prev ; *(adp)->ad_deps.le_prev = (adp)->ad_deps.le_next; ((adp )->ad_deps.le_prev) = ((void *)-1); ((adp)->ad_deps.le_next ) = ((void *)-1); } while (0); |
3809 | handle_allocdirect_partdone(adp); |
3810 | } |
3811 | while ((aip = |
3812 | LIST_FIRST(&bmsafemap->sm_allocindirhd)((&bmsafemap->sm_allocindirhd)->lh_first))) { |
3813 | aip->ai_stateai_list.wk_state |= DEPCOMPLETE0x0008; |
3814 | aip->ai_buf = NULL((void *)0); |
3815 | LIST_REMOVE(aip, ai_deps)do { if ((aip)->ai_deps.le_next != ((void *)0)) (aip)-> ai_deps.le_next->ai_deps.le_prev = (aip)->ai_deps.le_prev ; *(aip)->ai_deps.le_prev = (aip)->ai_deps.le_next; ((aip )->ai_deps.le_prev) = ((void *)-1); ((aip)->ai_deps.le_next ) = ((void *)-1); } while (0); |
3816 | handle_allocindir_partdone(aip); |
3817 | } |
3818 | while ((inodedep = |
3819 | LIST_FIRST(&bmsafemap->sm_inodedephd)((&bmsafemap->sm_inodedephd)->lh_first)) != NULL((void *)0)) { |
3820 | inodedep->id_stateid_list.wk_state |= DEPCOMPLETE0x0008; |
3821 | LIST_REMOVE(inodedep, id_deps)do { if ((inodedep)->id_deps.le_next != ((void *)0)) (inodedep )->id_deps.le_next->id_deps.le_prev = (inodedep)->id_deps .le_prev; *(inodedep)->id_deps.le_prev = (inodedep)->id_deps .le_next; ((inodedep)->id_deps.le_prev) = ((void *)-1); (( inodedep)->id_deps.le_next) = ((void *)-1); } while (0); |
3822 | inodedep->id_buf = NULL((void *)0); |
3823 | } |
3824 | WORKITEM_FREE(bmsafemap, D_BMSAFEMAP)softdep_freequeue_add((struct worklist *)bmsafemap); |
3825 | continue; |
3826 | |
3827 | case D_MKDIR11: |
3828 | handle_written_mkdir(WK_MKDIR(wk)((struct mkdir *)(wk)), MKDIR_BODY0x0020); |
3829 | continue; |
3830 | |
3831 | case D_ALLOCDIRECT4: |
3832 | adp = WK_ALLOCDIRECT(wk)((struct allocdirect *)(wk)); |
3833 | adp->ad_statead_list.wk_state |= COMPLETE0x0004; |
3834 | handle_allocdirect_partdone(adp); |
3835 | continue; |
3836 | |
3837 | case D_ALLOCINDIR6: |
3838 | aip = WK_ALLOCINDIR(wk)((struct allocindir *)(wk)); |
3839 | aip->ai_stateai_list.wk_state |= COMPLETE0x0004; |
3840 | handle_allocindir_partdone(aip); |
3841 | continue; |
3842 | |
3843 | case D_INDIRDEP5: |
3844 | indirdep = WK_INDIRDEP(wk)((struct indirdep *)(wk)); |
3845 | if (indirdep->ir_stateir_list.wk_state & GOINGAWAY0x0100) |
3846 | panic("disk_write_complete: indirdep gone"); |
3847 | memcpy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount)__builtin_memcpy((bp->b_data), (indirdep->ir_saveddata) , (bp->b_bcount)); |
3848 | free(indirdep->ir_saveddata, M_INDIRDEP83, bp->b_bcount); |
3849 | indirdep->ir_saveddata = NULL((void *)0); |
3850 | indirdep->ir_stateir_list.wk_state &= ~UNDONE0x0002; |
3851 | indirdep->ir_stateir_list.wk_state |= ATTACHED0x0001; |
3852 | while ((aip = LIST_FIRST(&indirdep->ir_donehd)((&indirdep->ir_donehd)->lh_first))) { |
3853 | handle_allocindir_partdone(aip); |
3854 | if (aip == LIST_FIRST(&indirdep->ir_donehd)((&indirdep->ir_donehd)->lh_first)) |
3855 | panic("disk_write_complete: not gone"); |
3856 | } |
3857 | WORKLIST_INSERT(&reattach, wk)do { (wk)->wk_state |= 0x8000; do { if (((wk)->wk_list. le_next = (&reattach)->lh_first) != ((void *)0)) (& reattach)->lh_first->wk_list.le_prev = &(wk)->wk_list .le_next; (&reattach)->lh_first = (wk); (wk)->wk_list .le_prev = &(&reattach)->lh_first; } while (0); } while (0); |
3858 | if ((bp->b_flags & B_DELWRI0x00000080) == 0) |
3859 | stat_indir_blk_ptrs++; |
3860 | buf_dirty(bp); |
3861 | continue; |
3862 | |
3863 | default: |
3864 | panic("handle_disk_write_complete: Unknown type %s", |
3865 | TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
3866 | /* NOTREACHED */ |
3867 | } |
3868 | } |
3869 | /* |
3870 | * Reattach any requests that must be redone. |
3871 | */ |
3872 | while ((wk = LIST_FIRST(&reattach)((&reattach)->lh_first)) != NULL((void *)0)) { |
3873 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
3874 | WORKLIST_INSERT(&bp->b_dep, wk)do { (wk)->wk_state |= 0x8000; do { if (((wk)->wk_list. le_next = (&bp->b_dep)->lh_first) != ((void *)0)) ( &bp->b_dep)->lh_first->wk_list.le_prev = &(wk )->wk_list.le_next; (&bp->b_dep)->lh_first = (wk ); (wk)->wk_list.le_prev = &(&bp->b_dep)->lh_first ; } while (0); } while (0); |
3875 | } |
3876 | #ifdef DEBUG |
3877 | if (lk.lkt_held != -2) |
3878 | panic("softdep_disk_write_complete: lock lost"); |
3879 | lk.lkt_held = -1; |
3880 | #endif |
3881 | } |
3882 | |
3883 | /* |
3884 | * Called from within softdep_disk_write_complete above. Note that |
3885 | * this routine is always called from interrupt level with further |
3886 | * splbio interrupts blocked. |
3887 | */ |
3888 | /* the completed allocdirect */ |
3889 | STATIC void |
3890 | handle_allocdirect_partdone(struct allocdirect *adp) |
3891 | { |
3892 | struct allocdirect *listadp; |
3893 | struct inodedep *inodedep; |
3894 | long bsize, delay; |
3895 | |
3896 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
3897 | |
3898 | if ((adp->ad_statead_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) != ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) |
3899 | return; |
3900 | if (adp->ad_buf != NULL((void *)0)) |
3901 | panic("handle_allocdirect_partdone: dangling dep"); |
3902 | |
3903 | /* |
3904 | * The on-disk inode cannot claim to be any larger than the last |
3905 | * fragment that has been written. Otherwise, the on-disk inode |
3906 | * might have fragments that were not the last block in the file |
3907 | * which would corrupt the filesystem. Thus, we cannot free any |
3908 | * allocdirects after one whose ad_oldblkno claims a fragment as |
3909 | * these blocks must be rolled back to zero before writing the inode. |
3910 | * We check the currently active set of allocdirects in id_inoupdt. |
3911 | */ |
3912 | inodedep = adp->ad_inodedep; |
3913 | bsize = inodedep->id_fs->fs_bsize; |
3914 | TAILQ_FOREACH(listadp, &inodedep->id_inoupdt, ad_next)for((listadp) = ((&inodedep->id_inoupdt)->tqh_first ); (listadp) != ((void *)0); (listadp) = ((listadp)->ad_next .tqe_next)) { |
3915 | /* found our block */ |
3916 | if (listadp == adp) |
3917 | break; |
3918 | /* continue if ad_oldlbn is not a fragment */ |
3919 | if (listadp->ad_oldsize == 0 || |
3920 | listadp->ad_oldsize == bsize) |
3921 | continue; |
3922 | /* hit a fragment */ |
3923 | return; |
3924 | } |
3925 | /* |
3926 | * If we have reached the end of the current list without |
3927 | * finding the just finished dependency, then it must be |
3928 | * on the future dependency list. Future dependencies cannot |
3929 | * be freed until they are moved to the current list. |
3930 | */ |
3931 | if (listadp == NULL((void *)0)) { |
3932 | #ifdef DEBUG |
3933 | TAILQ_FOREACH(listadp, &inodedep->id_newinoupdt, ad_next)for((listadp) = ((&inodedep->id_newinoupdt)->tqh_first ); (listadp) != ((void *)0); (listadp) = ((listadp)->ad_next .tqe_next)) |
3934 | /* found our block */ |
3935 | if (listadp == adp) |
3936 | break; |
3937 | if (listadp == NULL((void *)0)) |
3938 | panic("handle_allocdirect_partdone: lost dep"); |
3939 | #endif /* DEBUG */ |
3940 | return; |
3941 | } |
3942 | /* |
3943 | * If we have found the just finished dependency, then free |
3944 | * it along with anything that follows it that is complete. |
3945 | * If the inode still has a bitmap dependency, then it has |
3946 | * never been written to disk, hence the on-disk inode cannot |
3947 | * reference the old fragment so we can free it without delay. |
3948 | */ |
3949 | delay = (inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008); |
3950 | for (; adp; adp = listadp) { |
3951 | listadp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next); |
3952 | if ((adp->ad_statead_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) != ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) |
3953 | return; |
3954 | free_allocdirect(&inodedep->id_inoupdt, adp, delay); |
3955 | } |
3956 | } |
3957 | |
3958 | /* |
3959 | * Called from within softdep_disk_write_complete above. Note that |
3960 | * this routine is always called from interrupt level with further |
3961 | * splbio interrupts blocked. |
3962 | */ |
3963 | /* the completed allocindir */ |
3964 | STATIC void |
3965 | handle_allocindir_partdone(struct allocindir *aip) |
3966 | { |
3967 | struct indirdep *indirdep; |
3968 | |
3969 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
3970 | |
3971 | if ((aip->ai_stateai_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) != ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) |
3972 | return; |
3973 | if (aip->ai_buf != NULL((void *)0)) |
3974 | panic("handle_allocindir_partdone: dangling dependency"); |
3975 | indirdep = aip->ai_indirdep; |
3976 | if (indirdep->ir_stateir_list.wk_state & UNDONE0x0002) { |
3977 | LIST_REMOVE(aip, ai_next)do { if ((aip)->ai_next.le_next != ((void *)0)) (aip)-> ai_next.le_next->ai_next.le_prev = (aip)->ai_next.le_prev ; *(aip)->ai_next.le_prev = (aip)->ai_next.le_next; ((aip )->ai_next.le_prev) = ((void *)-1); ((aip)->ai_next.le_next ) = ((void *)-1); } while (0); |
3978 | LIST_INSERT_HEAD(&indirdep->ir_donehd, aip, ai_next)do { if (((aip)->ai_next.le_next = (&indirdep->ir_donehd )->lh_first) != ((void *)0)) (&indirdep->ir_donehd) ->lh_first->ai_next.le_prev = &(aip)->ai_next.le_next ; (&indirdep->ir_donehd)->lh_first = (aip); (aip)-> ai_next.le_prev = &(&indirdep->ir_donehd)->lh_first ; } while (0); |
3979 | return; |
3980 | } |
3981 | if (indirdep->ir_stateir_list.wk_state & UFS1FMT0x2000) |
3982 | ((int32_t *)indirdep->ir_savebp->b_data)[aip->ai_offset] = |
3983 | aip->ai_newblkno; |
3984 | else |
3985 | ((int64_t *)indirdep->ir_savebp->b_data)[aip->ai_offset] = |
3986 | aip->ai_newblkno; |
3987 | LIST_REMOVE(aip, ai_next)do { if ((aip)->ai_next.le_next != ((void *)0)) (aip)-> ai_next.le_next->ai_next.le_prev = (aip)->ai_next.le_prev ; *(aip)->ai_next.le_prev = (aip)->ai_next.le_next; ((aip )->ai_next.le_prev) = ((void *)-1); ((aip)->ai_next.le_next ) = ((void *)-1); } while (0); |
3988 | if (aip->ai_freefrag != NULL((void *)0)) |
3989 | add_to_worklist(&aip->ai_freefrag->ff_list); |
3990 | WORKITEM_FREE(aip, D_ALLOCINDIR)softdep_freequeue_add((struct worklist *)aip); |
3991 | } |
3992 | |
3993 | /* |
3994 | * Called from within softdep_disk_write_complete above to restore |
3995 | * in-memory inode block contents to their most up-to-date state. Note |
3996 | * that this routine is always called from interrupt level with further |
3997 | * splbio interrupts blocked. |
3998 | */ |
3999 | /* buffer containing the inode block */ |
4000 | STATIC int |
4001 | handle_written_inodeblock(struct inodedep *inodedep, struct buf *bp) |
4002 | { |
4003 | struct worklist *wk, *filefree; |
4004 | struct allocdirect *adp, *nextadp; |
4005 | struct ufs1_dinode *dp1 = NULL((void *)0); |
4006 | struct ufs2_dinode *dp2 = NULL((void *)0); |
4007 | int hadchanges, fstype; |
4008 | |
4009 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
4010 | |
4011 | if ((inodedep->id_stateid_list.wk_state & IOSTARTED0x0200) == 0) |
4012 | panic("handle_written_inodeblock: not started"); |
4013 | inodedep->id_stateid_list.wk_state &= ~IOSTARTED0x0200; |
4014 | |
4015 | if (inodedep->id_fs->fs_magic == FS_UFS1_MAGIC0x011954) { |
4016 | fstype = UM_UFS11; |
4017 | dp1 = (struct ufs1_dinode *) bp->b_data + |
4018 | ino_to_fsbo(inodedep->id_fs, inodedep->id_ino)((inodedep->id_ino) % ((inodedep->id_fs)->fs_inopb)); |
4019 | } else { |
4020 | fstype = UM_UFS22; |
4021 | dp2 = (struct ufs2_dinode *) bp->b_data + |
4022 | ino_to_fsbo(inodedep->id_fs, inodedep->id_ino)((inodedep->id_ino) % ((inodedep->id_fs)->fs_inopb)); |
4023 | } |
4024 | |
4025 | /* |
4026 | * If we had to rollback the inode allocation because of |
4027 | * bitmaps being incomplete, then simply restore it. |
4028 | * Keep the block dirty so that it will not be reclaimed until |
4029 | * all associated dependencies have been cleared and the |
4030 | * corresponding updates written to disk. |
4031 | */ |
4032 | if (inodedep->id_savedino1id_un.idu_savedino1 != NULL((void *)0)) { |
4033 | if (fstype == UM_UFS11) |
4034 | *dp1 = *inodedep->id_savedino1id_un.idu_savedino1; |
4035 | else |
4036 | *dp2 = *inodedep->id_savedino2id_un.idu_savedino2; |
4037 | free(inodedep->id_savedino1id_un.idu_savedino1, M_INODEDEP79, inodedep->id_unsize); |
4038 | inodedep->id_savedino1id_un.idu_savedino1 = NULL((void *)0); |
4039 | if ((bp->b_flags & B_DELWRI0x00000080) == 0) |
4040 | stat_inode_bitmap++; |
4041 | buf_dirty(bp); |
4042 | return (1); |
4043 | } |
4044 | inodedep->id_stateid_list.wk_state |= COMPLETE0x0004; |
4045 | /* |
4046 | * Roll forward anything that had to be rolled back before |
4047 | * the inode could be updated. |
4048 | */ |
4049 | hadchanges = 0; |
4050 | for (adp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first); adp; adp = nextadp) { |
4051 | nextadp = TAILQ_NEXT(adp, ad_next)((adp)->ad_next.tqe_next); |
4052 | if (adp->ad_statead_list.wk_state & ATTACHED0x0001) |
4053 | panic("handle_written_inodeblock: new entry"); |
4054 | if (fstype == UM_UFS11) { |
4055 | if (adp->ad_lbn < NDADDR12) { |
4056 | if (dp1->di_db[adp->ad_lbn] != adp->ad_oldblkno) |
4057 | panic("%s: %s #%lld mismatch %d != " |
4058 | "%lld", |
4059 | "handle_written_inodeblock", |
4060 | "direct pointer", |
4061 | (long long)adp->ad_lbn, |
4062 | dp1->di_db[adp->ad_lbn], |
4063 | (long long)adp->ad_oldblkno); |
4064 | dp1->di_db[adp->ad_lbn] = adp->ad_newblkno; |
4065 | } else { |
4066 | if (dp1->di_ib[adp->ad_lbn - NDADDR12] != 0) |
4067 | panic("%s: %s #%lld allocated as %d", |
4068 | "handle_written_inodeblock", |
4069 | "indirect pointer", |
4070 | (long long)(adp->ad_lbn - NDADDR12), |
4071 | dp1->di_ib[adp->ad_lbn - NDADDR12]); |
4072 | dp1->di_ib[adp->ad_lbn - NDADDR12] = |
4073 | adp->ad_newblkno; |
4074 | } |
4075 | } else { |
4076 | if (adp->ad_lbn < NDADDR12) { |
4077 | if (dp2->di_db[adp->ad_lbn] != adp->ad_oldblkno) |
4078 | panic("%s: %s #%lld mismatch %lld != " |
4079 | "%lld", "handle_written_inodeblock", |
4080 | "direct pointer", |
4081 | (long long)adp->ad_lbn, |
4082 | dp2->di_db[adp->ad_lbn], |
4083 | (long long)adp->ad_oldblkno); |
4084 | dp2->di_db[adp->ad_lbn] = adp->ad_newblkno; |
4085 | } else { |
4086 | if (dp2->di_ib[adp->ad_lbn - NDADDR12] != 0) |
4087 | panic("%s: %s #%lld allocated as %lld", |
4088 | "handle_written_inodeblock", |
4089 | "indirect pointer", |
4090 | (long long)(adp->ad_lbn - NDADDR12), |
4091 | dp2->di_ib[adp->ad_lbn - NDADDR12]); |
4092 | dp2->di_ib[adp->ad_lbn - NDADDR12] = |
4093 | adp->ad_newblkno; |
4094 | } |
4095 | } |
4096 | adp->ad_statead_list.wk_state &= ~UNDONE0x0002; |
4097 | adp->ad_statead_list.wk_state |= ATTACHED0x0001; |
4098 | hadchanges = 1; |
4099 | } |
4100 | if (hadchanges && (bp->b_flags & B_DELWRI0x00000080) == 0) |
4101 | stat_direct_blk_ptrs++; |
4102 | /* |
4103 | * Reset the file size to its most up-to-date value. |
4104 | */ |
4105 | if (inodedep->id_savedsize == -1) |
4106 | panic("handle_written_inodeblock: bad size"); |
4107 | |
4108 | if (fstype == UM_UFS11) { |
4109 | if (dp1->di_size != inodedep->id_savedsize) { |
4110 | dp1->di_size = inodedep->id_savedsize; |
4111 | hadchanges = 1; |
4112 | } |
4113 | } else { |
4114 | if (dp2->di_size != inodedep->id_savedsize) { |
4115 | dp2->di_size = inodedep->id_savedsize; |
4116 | hadchanges = 1; |
4117 | } |
4118 | } |
4119 | inodedep->id_savedsize = -1; |
4120 | /* |
4121 | * If there were any rollbacks in the inode block, then it must be |
4122 | * marked dirty so that its will eventually get written back in |
4123 | * its correct form. |
4124 | */ |
4125 | if (hadchanges) |
4126 | buf_dirty(bp); |
4127 | /* |
4128 | * Process any allocdirects that completed during the update. |
4129 | */ |
4130 | if ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first)) != NULL((void *)0)) |
4131 | handle_allocdirect_partdone(adp); |
4132 | /* |
4133 | * Process deallocations that were held pending until the |
4134 | * inode had been written to disk. Freeing of the inode |
4135 | * is delayed until after all blocks have been freed to |
4136 | * avoid creation of new <vfsid, inum, lbn> triples |
4137 | * before the old ones have been deleted. |
4138 | */ |
4139 | filefree = NULL((void *)0); |
4140 | while ((wk = LIST_FIRST(&inodedep->id_bufwait)((&inodedep->id_bufwait)->lh_first)) != NULL((void *)0)) { |
4141 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
4142 | switch (wk->wk_type) { |
4143 | |
4144 | case D_FREEFILE9: |
4145 | /* |
4146 | * We defer adding filefree to the worklist until |
4147 | * all other additions have been made to ensure |
4148 | * that it will be done after all the old blocks |
4149 | * have been freed. |
4150 | */ |
4151 | if (filefree != NULL((void *)0)) |
4152 | panic("handle_written_inodeblock: filefree"); |
4153 | filefree = wk; |
4154 | continue; |
4155 | |
4156 | case D_MKDIR11: |
4157 | handle_written_mkdir(WK_MKDIR(wk)((struct mkdir *)(wk)), MKDIR_PARENT0x0010); |
4158 | continue; |
4159 | |
4160 | case D_DIRADD10: |
4161 | diradd_inode_written(WK_DIRADD(wk)((struct diradd *)(wk)), inodedep); |
4162 | continue; |
4163 | |
4164 | case D_FREEBLKS8: |
4165 | wk->wk_state |= COMPLETE0x0004; |
4166 | if ((wk->wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) != ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) |
4167 | continue; |
4168 | /* FALLTHROUGH */ |
4169 | case D_FREEFRAG7: |
4170 | case D_DIRREM12: |
4171 | add_to_worklist(wk); |
4172 | continue; |
4173 | |
4174 | case D_NEWDIRBLK13: |
4175 | free_newdirblk(WK_NEWDIRBLK(wk)((struct newdirblk *)(wk))); |
4176 | continue; |
4177 | |
4178 | default: |
4179 | panic("handle_written_inodeblock: Unknown type %s", |
4180 | TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
4181 | /* NOTREACHED */ |
4182 | } |
4183 | } |
4184 | if (filefree != NULL((void *)0)) { |
4185 | if (free_inodedep(inodedep) == 0) |
4186 | panic("handle_written_inodeblock: live inodedep"); |
4187 | add_to_worklist(filefree); |
4188 | return (0); |
4189 | } |
4190 | |
4191 | /* |
4192 | * If no outstanding dependencies, free it. |
4193 | */ |
4194 | if (free_inodedep(inodedep) || |
4195 | TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) == NULL((void *)0)) |
4196 | return (0); |
4197 | return (hadchanges); |
4198 | } |
4199 | |
4200 | /* |
4201 | * Process a diradd entry after its dependent inode has been written. |
4202 | * This routine must be called with splbio interrupts blocked. |
4203 | */ |
4204 | STATIC void |
4205 | diradd_inode_written(struct diradd *dap, struct inodedep *inodedep) |
4206 | { |
4207 | struct pagedep *pagedep; |
4208 | |
4209 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
4210 | |
4211 | dap->da_stateda_list.wk_state |= COMPLETE0x0004; |
4212 | if ((dap->da_stateda_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) { |
4213 | if (dap->da_stateda_list.wk_state & DIRCHG0x0080) |
4214 | pagedep = dap->da_previousda_un.dau_previous->dm_pagedepdm_un.dmu_pagedep; |
4215 | else |
4216 | pagedep = dap->da_pagedepda_un.dau_pagedep; |
4217 | LIST_REMOVE(dap, da_pdlist)do { if ((dap)->da_pdlist.le_next != ((void *)0)) (dap)-> da_pdlist.le_next->da_pdlist.le_prev = (dap)->da_pdlist .le_prev; *(dap)->da_pdlist.le_prev = (dap)->da_pdlist. le_next; ((dap)->da_pdlist.le_prev) = ((void *)-1); ((dap) ->da_pdlist.le_next) = ((void *)-1); } while (0); |
4218 | LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist)do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_pendinghd )->lh_first) != ((void *)0)) (&pagedep->pd_pendinghd )->lh_first->da_pdlist.le_prev = &(dap)->da_pdlist .le_next; (&pagedep->pd_pendinghd)->lh_first = (dap ); (dap)->da_pdlist.le_prev = &(&pagedep->pd_pendinghd )->lh_first; } while (0); |
4219 | } |
4220 | WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list)do { (&dap->da_list)->wk_state |= 0x8000; do { if ( ((&dap->da_list)->wk_list.le_next = (&inodedep-> id_pendinghd)->lh_first) != ((void *)0)) (&inodedep-> id_pendinghd)->lh_first->wk_list.le_prev = &(&dap ->da_list)->wk_list.le_next; (&inodedep->id_pendinghd )->lh_first = (&dap->da_list); (&dap->da_list )->wk_list.le_prev = &(&inodedep->id_pendinghd) ->lh_first; } while (0); } while (0); |
4221 | } |
4222 | |
4223 | /* |
4224 | * Handle the completion of a mkdir dependency. |
4225 | */ |
4226 | STATIC void |
4227 | handle_written_mkdir(struct mkdir *mkdir, int type) |
4228 | { |
4229 | struct diradd *dap; |
4230 | struct pagedep *pagedep; |
4231 | |
4232 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
4233 | |
4234 | if (mkdir->md_statemd_list.wk_state != type) |
4235 | panic("handle_written_mkdir: bad type"); |
4236 | dap = mkdir->md_diradd; |
4237 | dap->da_stateda_list.wk_state &= ~type; |
4238 | if ((dap->da_stateda_list.wk_state & (MKDIR_PARENT0x0010 | MKDIR_BODY0x0020)) == 0) |
4239 | dap->da_stateda_list.wk_state |= DEPCOMPLETE0x0008; |
4240 | if ((dap->da_stateda_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) { |
4241 | if (dap->da_stateda_list.wk_state & DIRCHG0x0080) |
4242 | pagedep = dap->da_previousda_un.dau_previous->dm_pagedepdm_un.dmu_pagedep; |
4243 | else |
4244 | pagedep = dap->da_pagedepda_un.dau_pagedep; |
4245 | LIST_REMOVE(dap, da_pdlist)do { if ((dap)->da_pdlist.le_next != ((void *)0)) (dap)-> da_pdlist.le_next->da_pdlist.le_prev = (dap)->da_pdlist .le_prev; *(dap)->da_pdlist.le_prev = (dap)->da_pdlist. le_next; ((dap)->da_pdlist.le_prev) = ((void *)-1); ((dap) ->da_pdlist.le_next) = ((void *)-1); } while (0); |
4246 | LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist)do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_pendinghd )->lh_first) != ((void *)0)) (&pagedep->pd_pendinghd )->lh_first->da_pdlist.le_prev = &(dap)->da_pdlist .le_next; (&pagedep->pd_pendinghd)->lh_first = (dap ); (dap)->da_pdlist.le_prev = &(&pagedep->pd_pendinghd )->lh_first; } while (0); |
4247 | } |
4248 | LIST_REMOVE(mkdir, md_mkdirs)do { if ((mkdir)->md_mkdirs.le_next != ((void *)0)) (mkdir )->md_mkdirs.le_next->md_mkdirs.le_prev = (mkdir)->md_mkdirs .le_prev; *(mkdir)->md_mkdirs.le_prev = (mkdir)->md_mkdirs .le_next; ((mkdir)->md_mkdirs.le_prev) = ((void *)-1); ((mkdir )->md_mkdirs.le_next) = ((void *)-1); } while (0); |
4249 | WORKITEM_FREE(mkdir, D_MKDIR)softdep_freequeue_add((struct worklist *)mkdir); |
4250 | } |
4251 | |
4252 | /* |
4253 | * Called from within softdep_disk_write_complete above. |
4254 | * A write operation was just completed. Removed inodes can |
4255 | * now be freed and associated block pointers may be committed. |
4256 | * Note that this routine is always called from interrupt level |
4257 | * with further splbio interrupts blocked. |
4258 | */ |
4259 | /* buffer containing the written page */ |
4260 | STATIC int |
4261 | handle_written_filepage(struct pagedep *pagedep, struct buf *bp) |
4262 | { |
4263 | struct dirrem *dirrem; |
4264 | struct diradd *dap, *nextdap; |
4265 | struct direct *ep; |
4266 | int i, chgs; |
4267 | |
4268 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
4269 | |
4270 | if ((pagedep->pd_statepd_list.wk_state & IOSTARTED0x0200) == 0) |
4271 | panic("handle_written_filepage: not started"); |
4272 | pagedep->pd_statepd_list.wk_state &= ~IOSTARTED0x0200; |
4273 | /* |
4274 | * Process any directory removals that have been committed. |
4275 | */ |
4276 | while ((dirrem = LIST_FIRST(&pagedep->pd_dirremhd)((&pagedep->pd_dirremhd)->lh_first)) != NULL((void *)0)) { |
4277 | LIST_REMOVE(dirrem, dm_next)do { if ((dirrem)->dm_next.le_next != ((void *)0)) (dirrem )->dm_next.le_next->dm_next.le_prev = (dirrem)->dm_next .le_prev; *(dirrem)->dm_next.le_prev = (dirrem)->dm_next .le_next; ((dirrem)->dm_next.le_prev) = ((void *)-1); ((dirrem )->dm_next.le_next) = ((void *)-1); } while (0); |
4278 | dirrem->dm_dirinumdm_un.dmu_dirinum = pagedep->pd_ino; |
4279 | add_to_worklist(&dirrem->dm_list); |
4280 | } |
4281 | /* |
4282 | * Free any directory additions that have been committed. |
4283 | * If it is a newly allocated block, we have to wait until |
4284 | * the on-disk directory inode claims the new block. |
4285 | */ |
4286 | if ((pagedep->pd_statepd_list.wk_state & NEWBLOCK0x0800) == 0) |
4287 | while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)((&pagedep->pd_pendinghd)->lh_first)) != NULL((void *)0)) |
4288 | free_diradd(dap); |
4289 | /* |
4290 | * Uncommitted directory entries must be restored. |
4291 | */ |
4292 | for (chgs = 0, i = 0; i < DAHASHSZ6; i++) { |
4293 | for (dap = LIST_FIRST(&pagedep->pd_diraddhd[i])((&pagedep->pd_diraddhd[i])->lh_first); dap; |
4294 | dap = nextdap) { |
4295 | nextdap = LIST_NEXT(dap, da_pdlist)((dap)->da_pdlist.le_next); |
4296 | if (dap->da_stateda_list.wk_state & ATTACHED0x0001) |
4297 | panic("handle_written_filepage: attached"); |
4298 | ep = (struct direct *) |
4299 | ((char *)bp->b_data + dap->da_offset); |
4300 | ep->d_ino = dap->da_newinum; |
4301 | dap->da_stateda_list.wk_state &= ~UNDONE0x0002; |
4302 | dap->da_stateda_list.wk_state |= ATTACHED0x0001; |
4303 | chgs = 1; |
4304 | /* |
4305 | * If the inode referenced by the directory has |
4306 | * been written out, then the dependency can be |
4307 | * moved to the pending list. |
4308 | */ |
4309 | if ((dap->da_stateda_list.wk_state & ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) == ALLCOMPLETE(0x0001 | 0x0004 | 0x0008)) { |
4310 | LIST_REMOVE(dap, da_pdlist)do { if ((dap)->da_pdlist.le_next != ((void *)0)) (dap)-> da_pdlist.le_next->da_pdlist.le_prev = (dap)->da_pdlist .le_prev; *(dap)->da_pdlist.le_prev = (dap)->da_pdlist. le_next; ((dap)->da_pdlist.le_prev) = ((void *)-1); ((dap) ->da_pdlist.le_next) = ((void *)-1); } while (0); |
4311 | LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap,do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_pendinghd )->lh_first) != ((void *)0)) (&pagedep->pd_pendinghd )->lh_first->da_pdlist.le_prev = &(dap)->da_pdlist .le_next; (&pagedep->pd_pendinghd)->lh_first = (dap ); (dap)->da_pdlist.le_prev = &(&pagedep->pd_pendinghd )->lh_first; } while (0) |
4312 | da_pdlist)do { if (((dap)->da_pdlist.le_next = (&pagedep->pd_pendinghd )->lh_first) != ((void *)0)) (&pagedep->pd_pendinghd )->lh_first->da_pdlist.le_prev = &(dap)->da_pdlist .le_next; (&pagedep->pd_pendinghd)->lh_first = (dap ); (dap)->da_pdlist.le_prev = &(&pagedep->pd_pendinghd )->lh_first; } while (0); |
4313 | } |
4314 | } |
4315 | } |
4316 | /* |
4317 | * If there were any rollbacks in the directory, then it must be |
4318 | * marked dirty so that its will eventually get written back in |
4319 | * its correct form. |
4320 | */ |
4321 | if (chgs) { |
4322 | if ((bp->b_flags & B_DELWRI0x00000080) == 0) |
4323 | stat_dir_entry++; |
4324 | buf_dirty(bp); |
4325 | return (1); |
4326 | } |
4327 | /* |
4328 | * If we are not waiting for a new directory block to be |
4329 | * claimed by its inode, then the pagedep will be freed. |
4330 | * Otherwise it will remain to track any new entries on |
4331 | * the page in case they are fsync'ed. |
4332 | */ |
4333 | if ((pagedep->pd_statepd_list.wk_state & NEWBLOCK0x0800) == 0) { |
4334 | LIST_REMOVE(pagedep, pd_hash)do { if ((pagedep)->pd_hash.le_next != ((void *)0)) (pagedep )->pd_hash.le_next->pd_hash.le_prev = (pagedep)->pd_hash .le_prev; *(pagedep)->pd_hash.le_prev = (pagedep)->pd_hash .le_next; ((pagedep)->pd_hash.le_prev) = ((void *)-1); ((pagedep )->pd_hash.le_next) = ((void *)-1); } while (0); |
4335 | WORKITEM_FREE(pagedep, D_PAGEDEP)softdep_freequeue_add((struct worklist *)pagedep); |
4336 | } |
4337 | return (0); |
4338 | } |
4339 | |
4340 | /* |
4341 | * Writing back in-core inode structures. |
4342 | * |
4343 | * The file system only accesses an inode's contents when it occupies an |
4344 | * "in-core" inode structure. These "in-core" structures are separate from |
4345 | * the page frames used to cache inode blocks. Only the latter are |
4346 | * transferred to/from the disk. So, when the updated contents of the |
4347 | * "in-core" inode structure are copied to the corresponding in-memory inode |
4348 | * block, the dependencies are also transferred. The following procedure is |
4349 | * called when copying a dirty "in-core" inode to a cached inode block. |
4350 | */ |
4351 | |
4352 | /* |
4353 | * Called when an inode is loaded from disk. If the effective link count |
4354 | * differed from the actual link count when it was last flushed, then we |
4355 | * need to ensure that the correct effective link count is put back. |
4356 | */ |
4357 | /* the "in_core" copy of the inode */ |
4358 | void |
4359 | softdep_load_inodeblock(struct inode *ip) |
4360 | { |
4361 | struct inodedep *inodedep; |
4362 | |
4363 | /* |
4364 | * Check for alternate nlink count. |
4365 | */ |
4366 | ip->i_effnlink = DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink); |
4367 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4368 | if (inodedep_lookup(ip->i_fsinode_u.fs, ip->i_number, 0, &inodedep) == 0) { |
4369 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4370 | return; |
4371 | } |
4372 | ip->i_effnlink -= inodedep->id_nlinkdelta; |
4373 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4374 | } |
4375 | |
4376 | /* |
4377 | * This routine is called just before the "in-core" inode |
4378 | * information is to be copied to the in-memory inode block. |
4379 | * Recall that an inode block contains several inodes. If |
4380 | * the force flag is set, then the dependencies will be |
4381 | * cleared so that the update can always be made. Note that |
4382 | * the buffer is locked when this routine is called, so we |
4383 | * will never be in the middle of writing the inode block |
4384 | * to disk. |
4385 | */ |
4386 | /* the "in_core" copy of the inode */ |
4387 | /* the buffer containing the inode block */ |
4388 | /* nonzero => update must be allowed */ |
4389 | void |
4390 | softdep_update_inodeblock(struct inode *ip, struct buf *bp, int waitfor) |
4391 | { |
4392 | struct inodedep *inodedep; |
4393 | struct worklist *wk; |
4394 | int error, gotit; |
4395 | |
4396 | /* |
4397 | * If the effective link count is not equal to the actual link |
4398 | * count, then we must track the difference in an inodedep while |
4399 | * the inode is (potentially) tossed out of the cache. Otherwise, |
4400 | * if there is no existing inodedep, then there are no dependencies |
4401 | * to track. |
4402 | */ |
4403 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4404 | if (inodedep_lookup(ip->i_fsinode_u.fs, ip->i_number, 0, &inodedep) == 0) { |
4405 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4406 | if (ip->i_effnlink != DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink)) |
4407 | panic("softdep_update_inodeblock: bad link count"); |
4408 | return; |
4409 | } |
4410 | if (inodedep->id_nlinkdelta != DIP(ip, nlink)(((ip)->i_ump->um_fstype == 1) ? (ip)->dinode_u.ffs1_din ->di_nlink : (ip)->dinode_u.ffs2_din->di_nlink) - ip->i_effnlink) { |
4411 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4412 | panic("softdep_update_inodeblock: bad delta"); |
4413 | } |
4414 | /* |
4415 | * Changes have been initiated. Anything depending on these |
4416 | * changes cannot occur until this inode has been written. |
4417 | */ |
4418 | inodedep->id_stateid_list.wk_state &= ~COMPLETE0x0004; |
4419 | if ((inodedep->id_stateid_list.wk_state & ONWORKLIST0x8000) == 0) |
4420 | WORKLIST_INSERT(&bp->b_dep, &inodedep->id_list)do { (&inodedep->id_list)->wk_state |= 0x8000; do { if (((&inodedep->id_list)->wk_list.le_next = (& bp->b_dep)->lh_first) != ((void *)0)) (&bp->b_dep )->lh_first->wk_list.le_prev = &(&inodedep-> id_list)->wk_list.le_next; (&bp->b_dep)->lh_first = (&inodedep->id_list); (&inodedep->id_list)-> wk_list.le_prev = &(&bp->b_dep)->lh_first; } while (0); } while (0); |
4421 | /* |
4422 | * Any new dependencies associated with the incore inode must |
4423 | * now be moved to the list associated with the buffer holding |
4424 | * the in-memory copy of the inode. Once merged process any |
4425 | * allocdirects that are completed by the merger. |
4426 | */ |
4427 | merge_inode_lists(inodedep); |
4428 | if (TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) != NULL((void *)0)) |
4429 | handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first)); |
4430 | /* |
4431 | * Now that the inode has been pushed into the buffer, the |
4432 | * operations dependent on the inode being written to disk |
4433 | * can be moved to the id_bufwait so that they will be |
4434 | * processed when the buffer I/O completes. |
4435 | */ |
4436 | while ((wk = LIST_FIRST(&inodedep->id_inowait)((&inodedep->id_inowait)->lh_first)) != NULL((void *)0)) { |
4437 | WORKLIST_REMOVE(wk)do { (wk)->wk_state &= ~0x8000; do { if ((wk)->wk_list .le_next != ((void *)0)) (wk)->wk_list.le_next->wk_list .le_prev = (wk)->wk_list.le_prev; *(wk)->wk_list.le_prev = (wk)->wk_list.le_next; ((wk)->wk_list.le_prev) = ((void *)-1); ((wk)->wk_list.le_next) = ((void *)-1); } while (0 ); } while (0); |
4438 | WORKLIST_INSERT(&inodedep->id_bufwait, wk)do { (wk)->wk_state |= 0x8000; do { if (((wk)->wk_list. le_next = (&inodedep->id_bufwait)->lh_first) != ((void *)0)) (&inodedep->id_bufwait)->lh_first->wk_list .le_prev = &(wk)->wk_list.le_next; (&inodedep-> id_bufwait)->lh_first = (wk); (wk)->wk_list.le_prev = & (&inodedep->id_bufwait)->lh_first; } while (0); } while (0); |
4439 | } |
4440 | /* |
4441 | * Newly allocated inodes cannot be written until the bitmap |
4442 | * that allocates them have been written (indicated by |
4443 | * DEPCOMPLETE being set in id_state). If we are doing a |
4444 | * forced sync (e.g., an fsync on a file), we force the bitmap |
4445 | * to be written so that the update can be done. |
4446 | */ |
4447 | do { |
4448 | if ((inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008) != 0 || waitfor == 0) { |
4449 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4450 | return; |
4451 | } |
4452 | bp = inodedep->id_buf; |
4453 | gotit = getdirtybuf(bp, MNT_WAIT1); |
4454 | } while (gotit == -1); |
4455 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4456 | if (gotit && (error = bwrite(bp)) != 0) |
4457 | softdep_error("softdep_update_inodeblock: bwrite", error); |
4458 | if ((inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008) == 0) |
4459 | panic("softdep_update_inodeblock: update failed"); |
4460 | } |
4461 | |
4462 | /* |
4463 | * Merge the new inode dependency list (id_newinoupdt) into the old |
4464 | * inode dependency list (id_inoupdt). This routine must be called |
4465 | * with splbio interrupts blocked. |
4466 | */ |
4467 | STATIC void |
4468 | merge_inode_lists(struct inodedep *inodedep) |
4469 | { |
4470 | struct allocdirect *listadp, *newadp; |
4471 | |
4472 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
4473 | |
4474 | newadp = TAILQ_FIRST(&inodedep->id_newinoupdt)((&inodedep->id_newinoupdt)->tqh_first); |
4475 | for (listadp = TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first); listadp && newadp;) { |
4476 | if (listadp->ad_lbn < newadp->ad_lbn) { |
4477 | listadp = TAILQ_NEXT(listadp, ad_next)((listadp)->ad_next.tqe_next); |
4478 | continue; |
4479 | } |
4480 | TAILQ_REMOVE(&inodedep->id_newinoupdt, newadp, ad_next)do { if (((newadp)->ad_next.tqe_next) != ((void *)0)) (newadp )->ad_next.tqe_next->ad_next.tqe_prev = (newadp)->ad_next .tqe_prev; else (&inodedep->id_newinoupdt)->tqh_last = (newadp)->ad_next.tqe_prev; *(newadp)->ad_next.tqe_prev = (newadp)->ad_next.tqe_next; ((newadp)->ad_next.tqe_prev ) = ((void *)-1); ((newadp)->ad_next.tqe_next) = ((void *) -1); } while (0); |
4481 | TAILQ_INSERT_BEFORE(listadp, newadp, ad_next)do { (newadp)->ad_next.tqe_prev = (listadp)->ad_next.tqe_prev ; (newadp)->ad_next.tqe_next = (listadp); *(listadp)->ad_next .tqe_prev = (newadp); (listadp)->ad_next.tqe_prev = &( newadp)->ad_next.tqe_next; } while (0); |
4482 | if (listadp->ad_lbn == newadp->ad_lbn) { |
4483 | allocdirect_merge(&inodedep->id_inoupdt, newadp, |
4484 | listadp); |
4485 | listadp = newadp; |
4486 | } |
4487 | newadp = TAILQ_FIRST(&inodedep->id_newinoupdt)((&inodedep->id_newinoupdt)->tqh_first); |
4488 | } |
4489 | TAILQ_CONCAT(&inodedep->id_inoupdt, &inodedep->id_newinoupdt, ad_next)do { if (!(((&inodedep->id_newinoupdt)->tqh_first) == ((void *)0))) { *(&inodedep->id_inoupdt)->tqh_last = (&inodedep->id_newinoupdt)->tqh_first; (&inodedep ->id_newinoupdt)->tqh_first->ad_next.tqe_prev = (& inodedep->id_inoupdt)->tqh_last; (&inodedep->id_inoupdt )->tqh_last = (&inodedep->id_newinoupdt)->tqh_last ; do { ((&inodedep->id_newinoupdt))->tqh_first = (( void *)0); ((&inodedep->id_newinoupdt))->tqh_last = &((&inodedep->id_newinoupdt))->tqh_first; } while (0); } } while (0); |
4490 | } |
4491 | |
4492 | /* |
4493 | * If we are doing an fsync, then we must ensure that any directory |
4494 | * entries for the inode have been written after the inode gets to disk. |
4495 | */ |
4496 | /* the "in_core" copy of the inode */ |
4497 | int |
4498 | softdep_fsync(struct vnode *vp) |
4499 | { |
4500 | struct inodedep *inodedep; |
4501 | struct pagedep *pagedep; |
4502 | struct worklist *wk; |
4503 | struct diradd *dap; |
4504 | struct mount *mnt; |
4505 | struct vnode *pvp; |
4506 | struct inode *ip; |
4507 | struct inode *pip; |
4508 | struct buf *bp; |
4509 | struct fs *fs; |
4510 | struct proc *p = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc; /* XXX */ |
4511 | int error, flushparent; |
4512 | ufsino_t parentino; |
4513 | daddr_t lbn; |
4514 | |
4515 | ip = VTOI(vp)((struct inode *)(vp)->v_data); |
4516 | fs = ip->i_fsinode_u.fs; |
4517 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4518 | if (inodedep_lookup(fs, ip->i_number, 0, &inodedep) == 0) { |
4519 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4520 | return (0); |
4521 | } |
4522 | if (LIST_FIRST(&inodedep->id_inowait)((&inodedep->id_inowait)->lh_first) != NULL((void *)0) || |
4523 | LIST_FIRST(&inodedep->id_bufwait)((&inodedep->id_bufwait)->lh_first) != NULL((void *)0) || |
4524 | TAILQ_FIRST(&inodedep->id_inoupdt)((&inodedep->id_inoupdt)->tqh_first) != NULL((void *)0) || |
4525 | TAILQ_FIRST(&inodedep->id_newinoupdt)((&inodedep->id_newinoupdt)->tqh_first) != NULL((void *)0)) { |
4526 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4527 | panic("softdep_fsync: pending ops"); |
4528 | } |
4529 | for (error = 0, flushparent = 0; ; ) { |
Value stored to 'error' is never read | |
4530 | if ((wk = LIST_FIRST(&inodedep->id_pendinghd)((&inodedep->id_pendinghd)->lh_first)) == NULL((void *)0)) |
4531 | break; |
4532 | if (wk->wk_type != D_DIRADD10) { |
4533 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4534 | panic("softdep_fsync: Unexpected type %s", |
4535 | TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
4536 | } |
4537 | dap = WK_DIRADD(wk)((struct diradd *)(wk)); |
4538 | /* |
4539 | * Flush our parent if this directory entry has a MKDIR_PARENT |
4540 | * dependency or is contained in a newly allocated block. |
4541 | */ |
4542 | if (dap->da_stateda_list.wk_state & DIRCHG0x0080) |
4543 | pagedep = dap->da_previousda_un.dau_previous->dm_pagedepdm_un.dmu_pagedep; |
4544 | else |
4545 | pagedep = dap->da_pagedepda_un.dau_pagedep; |
4546 | mnt = pagedep->pd_mnt; |
4547 | parentino = pagedep->pd_ino; |
4548 | lbn = pagedep->pd_lbn; |
4549 | if ((dap->da_stateda_list.wk_state & (MKDIR_BODY0x0020 | COMPLETE0x0004)) != COMPLETE0x0004) { |
4550 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4551 | panic("softdep_fsync: dirty"); |
4552 | } |
4553 | if ((dap->da_stateda_list.wk_state & MKDIR_PARENT0x0010) || |
4554 | (pagedep->pd_statepd_list.wk_state & NEWBLOCK0x0800)) |
4555 | flushparent = 1; |
4556 | else |
4557 | flushparent = 0; |
4558 | /* |
4559 | * If we are being fsync'ed as part of vgone'ing this vnode, |
4560 | * then we will not be able to release and recover the |
4561 | * vnode below, so we just have to give up on writing its |
4562 | * directory entry out. It will eventually be written, just |
4563 | * not now, but then the user was not asking to have it |
4564 | * written, so we are not breaking any promises. |
4565 | */ |
4566 | mtx_enter(&vnode_mtx); |
4567 | if (vp->v_lflag & VXLOCK0x0100) { |
4568 | mtx_leave(&vnode_mtx); |
4569 | break; |
4570 | } |
4571 | mtx_leave(&vnode_mtx); |
4572 | /* |
4573 | * We prevent deadlock by always fetching inodes from the |
4574 | * root, moving down the directory tree. Thus, when fetching |
4575 | * our parent directory, we must unlock ourselves before |
4576 | * requesting the lock on our parent. See the comment in |
4577 | * ufs_lookup for details on possible races. |
4578 | */ |
4579 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4580 | VOP_UNLOCK(vp); |
4581 | error = VFS_VGET(mnt, parentino, &pvp)(*(mnt)->mnt_op->vfs_vget)(mnt, parentino, &pvp); |
4582 | vn_lock(vp, LK_EXCLUSIVE0x0001UL | LK_RETRY0x2000UL); |
4583 | if (error != 0) |
4584 | return (error); |
4585 | /* |
4586 | * All MKDIR_PARENT dependencies and all the NEWBLOCK pagedeps |
4587 | * that are contained in direct blocks will be resolved by |
4588 | * doing a UFS_UPDATE. Pagedeps contained in indirect blocks |
4589 | * may require a complete sync'ing of the directory. So, we |
4590 | * try the cheap and fast UFS_UPDATE first, and if that fails, |
4591 | * then we do the slower VOP_FSYNC of the directory. |
4592 | */ |
4593 | pip = VTOI(pvp)((struct inode *)(pvp)->v_data); |
4594 | if (flushparent) { |
4595 | error = UFS_UPDATE(pip, 1)((pip)->i_vtbl->iv_update)((pip), (1)); |
4596 | if (error) { |
4597 | vput(pvp); |
4598 | return (error); |
4599 | } |
4600 | if (pagedep->pd_statepd_list.wk_state & NEWBLOCK0x0800) { |
4601 | error = VOP_FSYNC(pvp, p->p_ucred, MNT_WAIT1, p); |
4602 | if (error) { |
4603 | vput(pvp); |
4604 | return (error); |
4605 | } |
4606 | } |
4607 | } |
4608 | /* |
4609 | * Flush directory page containing the inode's name. |
4610 | */ |
4611 | error = bread(pvp, lbn, fs->fs_bsize, &bp); |
4612 | if (error == 0) { |
4613 | bp->b_bcount = blksize(fs, pip, lbn)(((lbn) >= 12 || ((((pip))->i_ump->um_fstype == 1) ? ((pip))->dinode_u.ffs1_din->di_size : ((pip))->dinode_u .ffs2_din->di_size) >= ((lbn) + 1) << (fs)->fs_bshift ) ? (u_int64_t)(fs)->fs_bsize : ((((((((((pip))->i_ump-> um_fstype == 1) ? ((pip))->dinode_u.ffs1_din->di_size : ((pip))->dinode_u.ffs2_din->di_size)) & (fs)->fs_qbmask )) + (fs)->fs_qfmask) & (fs)->fs_fmask))); |
4614 | error = bwrite(bp); |
4615 | } else |
4616 | brelse(bp); |
4617 | vput(pvp); |
4618 | if (error != 0) |
4619 | return (error); |
4620 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4621 | if (inodedep_lookup(fs, ip->i_number, 0, &inodedep) == 0) |
4622 | break; |
4623 | } |
4624 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4625 | return (0); |
4626 | } |
4627 | |
4628 | /* |
4629 | * Flush all the dirty bitmaps associated with the block device |
4630 | * before flushing the rest of the dirty blocks so as to reduce |
4631 | * the number of dependencies that will have to be rolled back. |
4632 | */ |
4633 | void |
4634 | softdep_fsync_mountdev(struct vnode *vp, int waitfor) |
4635 | { |
4636 | struct buf *bp, *nbp; |
4637 | struct worklist *wk; |
4638 | |
4639 | if (!vn_isdisk(vp, NULL((void *)0))) |
4640 | panic("softdep_fsync_mountdev: vnode not a disk"); |
4641 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4642 | LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp)for ((bp) = ((&vp->v_dirtyblkhd)->lh_first); (bp) && ((nbp) = ((bp)->b_vnbufs.le_next), 1); (bp) = (nbp)) { |
4643 | /* |
4644 | * If it is already scheduled, skip to the next buffer. |
4645 | */ |
4646 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
4647 | if (bp->b_flags & B_BUSY0x00000010) |
4648 | continue; |
4649 | |
4650 | if ((bp->b_flags & B_DELWRI0x00000080) == 0) { |
4651 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4652 | panic("softdep_fsync_mountdev: not dirty"); |
4653 | } |
4654 | /* |
4655 | * We are only interested in bitmaps with outstanding |
4656 | * dependencies. |
4657 | */ |
4658 | if ((wk = LIST_FIRST(&bp->b_dep)((&bp->b_dep)->lh_first)) == NULL((void *)0) || |
4659 | wk->wk_type != D_BMSAFEMAP3) { |
4660 | continue; |
4661 | } |
4662 | bremfreebufcache_take(bp); |
4663 | buf_acquire(bp); |
4664 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4665 | (void) bawrite(bp); |
4666 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4667 | /* |
4668 | * Since we may have slept during the I/O, we need |
4669 | * to start from a known point. |
4670 | */ |
4671 | nbp = LIST_FIRST(&vp->v_dirtyblkhd)((&vp->v_dirtyblkhd)->lh_first); |
4672 | } |
4673 | if (waitfor == MNT_WAIT1) |
4674 | drain_output(vp, 1); |
4675 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4676 | } |
4677 | |
4678 | /* |
4679 | * This routine is called when we are trying to synchronously flush a |
4680 | * file. This routine must eliminate any filesystem metadata dependencies |
4681 | * so that the syncing routine can succeed by pushing the dirty blocks |
4682 | * associated with the file. If any I/O errors occur, they are returned. |
4683 | */ |
4684 | int |
4685 | softdep_sync_metadata(struct vop_fsync_args *ap) |
4686 | { |
4687 | struct vnode *vp = ap->a_vp; |
4688 | struct pagedep *pagedep; |
4689 | struct allocdirect *adp; |
4690 | struct allocindir *aip; |
4691 | struct buf *bp, *nbp; |
4692 | struct worklist *wk; |
4693 | int i, gotit, error, waitfor; |
4694 | |
4695 | /* |
4696 | * Check whether this vnode is involved in a filesystem |
4697 | * that is doing soft dependency processing. |
4698 | */ |
4699 | if (!vn_isdisk(vp, NULL((void *)0))) { |
4700 | if (!DOINGSOFTDEP(vp)((vp)->v_mount->mnt_flag & 0x04000000)) |
4701 | return (0); |
4702 | } else |
4703 | if (vp->v_specmountpointv_un.vu_specinfo->si_mountpoint == NULL((void *)0) || |
4704 | (vp->v_specmountpointv_un.vu_specinfo->si_mountpoint->mnt_flag & MNT_SOFTDEP0x04000000) == 0) |
4705 | return (0); |
4706 | /* |
4707 | * Ensure that any direct block dependencies have been cleared. |
4708 | */ |
4709 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4710 | if ((error = flush_inodedep_deps(VTOI(vp)((struct inode *)(vp)->v_data)->i_fsinode_u.fs, VTOI(vp)((struct inode *)(vp)->v_data)->i_number))) { |
4711 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4712 | return (error); |
4713 | } |
4714 | /* |
4715 | * For most files, the only metadata dependencies are the |
4716 | * cylinder group maps that allocate their inode or blocks. |
4717 | * The block allocation dependencies can be found by traversing |
4718 | * the dependency lists for any buffers that remain on their |
4719 | * dirty buffer list. The inode allocation dependency will |
4720 | * be resolved when the inode is updated with MNT_WAIT. |
4721 | * This work is done in two passes. The first pass grabs most |
4722 | * of the buffers and begins asynchronously writing them. The |
4723 | * only way to wait for these asynchronous writes is to sleep |
4724 | * on the filesystem vnode which may stay busy for a long time |
4725 | * if the filesystem is active. So, instead, we make a second |
4726 | * pass over the dependencies blocking on each write. In the |
4727 | * usual case we will be blocking against a write that we |
4728 | * initiated, so when it is done the dependency will have been |
4729 | * resolved. Thus the second pass is expected to end quickly. |
4730 | */ |
4731 | waitfor = MNT_NOWAIT2; |
4732 | top: |
4733 | /* |
4734 | * We must wait for any I/O in progress to finish so that |
4735 | * all potential buffers on the dirty list will be visible. |
4736 | */ |
4737 | drain_output(vp, 1); |
4738 | bp = LIST_FIRST(&vp->v_dirtyblkhd)((&vp->v_dirtyblkhd)->lh_first); |
4739 | gotit = getdirtybuf(bp, MNT_WAIT1); |
4740 | if (gotit == 0) { |
4741 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4742 | return (0); |
4743 | } else if (gotit == -1) |
4744 | goto top; |
4745 | loop: |
4746 | /* |
4747 | * As we hold the buffer locked, none of its dependencies |
4748 | * will disappear. |
4749 | */ |
4750 | LIST_FOREACH(wk, &bp->b_dep, wk_list)for((wk) = ((&bp->b_dep)->lh_first); (wk)!= ((void * )0); (wk) = ((wk)->wk_list.le_next)) { |
4751 | switch (wk->wk_type) { |
4752 | |
4753 | case D_ALLOCDIRECT4: |
4754 | adp = WK_ALLOCDIRECT(wk)((struct allocdirect *)(wk)); |
4755 | if (adp->ad_statead_list.wk_state & DEPCOMPLETE0x0008) |
4756 | break; |
4757 | nbp = adp->ad_buf; |
4758 | gotit = getdirtybuf(nbp, waitfor); |
4759 | if (gotit == 0) |
4760 | break; |
4761 | else if (gotit == -1) |
4762 | goto loop; |
4763 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4764 | if (waitfor == MNT_NOWAIT2) { |
4765 | bawrite(nbp); |
4766 | } else if ((error = VOP_BWRITE(nbp)) != 0) { |
4767 | bawrite(bp); |
4768 | return (error); |
4769 | } |
4770 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4771 | break; |
4772 | |
4773 | case D_ALLOCINDIR6: |
4774 | aip = WK_ALLOCINDIR(wk)((struct allocindir *)(wk)); |
4775 | if (aip->ai_stateai_list.wk_state & DEPCOMPLETE0x0008) |
4776 | break; |
4777 | nbp = aip->ai_buf; |
4778 | gotit = getdirtybuf(nbp, waitfor); |
4779 | if (gotit == 0) |
4780 | break; |
4781 | else if (gotit == -1) |
4782 | goto loop; |
4783 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4784 | if (waitfor == MNT_NOWAIT2) { |
4785 | bawrite(nbp); |
4786 | } else if ((error = VOP_BWRITE(nbp)) != 0) { |
4787 | bawrite(bp); |
4788 | return (error); |
4789 | } |
4790 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4791 | break; |
4792 | |
4793 | case D_INDIRDEP5: |
4794 | restart: |
4795 | |
4796 | LIST_FOREACH(aip, &WK_INDIRDEP(wk)->ir_deplisthd, ai_next)for((aip) = ((&((struct indirdep *)(wk))->ir_deplisthd )->lh_first); (aip)!= ((void *)0); (aip) = ((aip)->ai_next .le_next)) { |
4797 | if (aip->ai_stateai_list.wk_state & DEPCOMPLETE0x0008) |
4798 | continue; |
4799 | nbp = aip->ai_buf; |
4800 | if (getdirtybuf(nbp, MNT_WAIT1) <= 0) |
4801 | goto restart; |
4802 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4803 | if ((error = VOP_BWRITE(nbp)) != 0) { |
4804 | bawrite(bp); |
4805 | return (error); |
4806 | } |
4807 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4808 | goto restart; |
4809 | } |
4810 | break; |
4811 | |
4812 | case D_INODEDEP1: |
4813 | if ((error = flush_inodedep_deps(WK_INODEDEP(wk)((struct inodedep *)(wk))->id_fs, |
4814 | WK_INODEDEP(wk)((struct inodedep *)(wk))->id_ino)) != 0) { |
4815 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4816 | bawrite(bp); |
4817 | return (error); |
4818 | } |
4819 | break; |
4820 | |
4821 | case D_PAGEDEP0: |
4822 | /* |
4823 | * We are trying to sync a directory that may |
4824 | * have dependencies on both its own metadata |
4825 | * and/or dependencies on the inodes of any |
4826 | * recently allocated files. We walk its diradd |
4827 | * lists pushing out the associated inode. |
4828 | */ |
4829 | pagedep = WK_PAGEDEP(wk)((struct pagedep *)(wk)); |
4830 | for (i = 0; i < DAHASHSZ6; i++) { |
4831 | if (LIST_FIRST(&pagedep->pd_diraddhd[i])((&pagedep->pd_diraddhd[i])->lh_first) == |
4832 | NULL((void *)0)) |
4833 | continue; |
4834 | if ((error = |
4835 | flush_pagedep_deps(vp, pagedep->pd_mnt, |
4836 | &pagedep->pd_diraddhd[i]))) { |
4837 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4838 | bawrite(bp); |
4839 | return (error); |
4840 | } |
4841 | } |
4842 | break; |
4843 | |
4844 | case D_MKDIR11: |
4845 | /* |
4846 | * This case should never happen if the vnode has |
4847 | * been properly sync'ed. However, if this function |
4848 | * is used at a place where the vnode has not yet |
4849 | * been sync'ed, this dependency can show up. So, |
4850 | * rather than panic, just flush it. |
4851 | */ |
4852 | nbp = WK_MKDIR(wk)((struct mkdir *)(wk))->md_buf; |
4853 | KASSERT(bp != nbp)((bp != nbp) ? (void)0 : __assert("diagnostic ", "/usr/src/sys/ufs/ffs/ffs_softdep.c" , 4853, "bp != nbp")); |
4854 | gotit = getdirtybuf(nbp, waitfor); |
4855 | if (gotit == 0) |
4856 | break; |
4857 | else if (gotit == -1) |
4858 | goto loop; |
4859 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4860 | if (waitfor == MNT_NOWAIT2) { |
4861 | bawrite(nbp); |
4862 | } else if ((error = VOP_BWRITE(nbp)) != 0) { |
4863 | bawrite(bp); |
4864 | return (error); |
4865 | } |
4866 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4867 | break; |
4868 | |
4869 | case D_BMSAFEMAP3: |
4870 | /* |
4871 | * This case should never happen if the vnode has |
4872 | * been properly sync'ed. However, if this function |
4873 | * is used at a place where the vnode has not yet |
4874 | * been sync'ed, this dependency can show up. So, |
4875 | * rather than panic, just flush it. |
4876 | */ |
4877 | nbp = WK_BMSAFEMAP(wk)((struct bmsafemap *)(wk))->sm_buf; |
4878 | if (bp == nbp) |
4879 | break; |
4880 | gotit = getdirtybuf(nbp, waitfor); |
4881 | if (gotit == 0) |
4882 | break; |
4883 | else if (gotit == -1) |
4884 | goto loop; |
4885 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4886 | if (waitfor == MNT_NOWAIT2) { |
4887 | bawrite(nbp); |
4888 | } else if ((error = VOP_BWRITE(nbp)) != 0) { |
4889 | bawrite(bp); |
4890 | return (error); |
4891 | } |
4892 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4893 | break; |
4894 | |
4895 | default: |
4896 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4897 | panic("softdep_sync_metadata: Unknown type %s", |
4898 | TYPENAME(wk->wk_type)((unsigned)(wk->wk_type) <= 13 ? softdep_typenames[wk-> wk_type] : "???")); |
4899 | /* NOTREACHED */ |
4900 | } |
4901 | } |
4902 | do { |
4903 | nbp = LIST_NEXT(bp, b_vnbufs)((bp)->b_vnbufs.le_next); |
4904 | gotit = getdirtybuf(nbp, MNT_WAIT1); |
4905 | } while (gotit == -1); |
4906 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4907 | bawrite(bp); |
4908 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4909 | if (nbp != NULL((void *)0)) { |
4910 | bp = nbp; |
4911 | goto loop; |
4912 | } |
4913 | /* |
4914 | * The brief unlock is to allow any pent up dependency |
4915 | * processing to be done. Then proceed with the second pass. |
4916 | */ |
4917 | if (waitfor == MNT_NOWAIT2) { |
4918 | waitfor = MNT_WAIT1; |
4919 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4920 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4921 | goto top; |
4922 | } |
4923 | |
4924 | /* |
4925 | * If we have managed to get rid of all the dirty buffers, |
4926 | * then we are done. For certain directories and block |
4927 | * devices, we may need to do further work. |
4928 | * |
4929 | * We must wait for any I/O in progress to finish so that |
4930 | * all potential buffers on the dirty list will be visible. |
4931 | */ |
4932 | drain_output(vp, 1); |
4933 | if (LIST_EMPTY(&vp->v_dirtyblkhd)(((&vp->v_dirtyblkhd)->lh_first) == ((void *)0))) { |
4934 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4935 | return (0); |
4936 | } |
4937 | |
4938 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4939 | /* |
4940 | * If we are trying to sync a block device, some of its buffers may |
4941 | * contain metadata that cannot be written until the contents of some |
4942 | * partially written files have been written to disk. The only easy |
4943 | * way to accomplish this is to sync the entire filesystem (luckily |
4944 | * this happens rarely). |
4945 | */ |
4946 | if (vn_isdisk(vp, NULL((void *)0)) && |
4947 | vp->v_specmountpointv_un.vu_specinfo->si_mountpoint && !VOP_ISLOCKED(vp) && |
4948 | (error = VFS_SYNC(vp->v_specmountpoint, MNT_WAIT, 0, ap->a_cred,(*(vp->v_un.vu_specinfo->si_mountpoint)->mnt_op-> vfs_sync)(vp->v_un.vu_specinfo->si_mountpoint, 1, 0, ap ->a_cred, ap->a_p) |
4949 | ap->a_p)(*(vp->v_un.vu_specinfo->si_mountpoint)->mnt_op-> vfs_sync)(vp->v_un.vu_specinfo->si_mountpoint, 1, 0, ap ->a_cred, ap->a_p)) != 0) |
4950 | return (error); |
4951 | return (0); |
4952 | } |
4953 | |
4954 | /* |
4955 | * Flush the dependencies associated with an inodedep. |
4956 | * Called with splbio blocked. |
4957 | */ |
4958 | STATIC int |
4959 | flush_inodedep_deps(struct fs *fs, ufsino_t ino) |
4960 | { |
4961 | struct inodedep *inodedep; |
4962 | struct allocdirect *adp; |
4963 | int gotit, error, waitfor; |
4964 | struct buf *bp; |
4965 | |
4966 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
4967 | |
4968 | /* |
4969 | * This work is done in two passes. The first pass grabs most |
4970 | * of the buffers and begins asynchronously writing them. The |
4971 | * only way to wait for these asynchronous writes is to sleep |
4972 | * on the filesystem vnode which may stay busy for a long time |
4973 | * if the filesystem is active. So, instead, we make a second |
4974 | * pass over the dependencies blocking on each write. In the |
4975 | * usual case we will be blocking against a write that we |
4976 | * initiated, so when it is done the dependency will have been |
4977 | * resolved. Thus the second pass is expected to end quickly. |
4978 | * We give a brief window at the top of the loop to allow |
4979 | * any pending I/O to complete. |
4980 | */ |
4981 | for (waitfor = MNT_NOWAIT2; ; ) { |
4982 | retry_ino: |
4983 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4984 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
4985 | if (inodedep_lookup(fs, ino, 0, &inodedep) == 0) |
4986 | return (0); |
4987 | TAILQ_FOREACH(adp, &inodedep->id_inoupdt, ad_next)for((adp) = ((&inodedep->id_inoupdt)->tqh_first); ( adp) != ((void *)0); (adp) = ((adp)->ad_next.tqe_next)) { |
4988 | if (adp->ad_statead_list.wk_state & DEPCOMPLETE0x0008) |
4989 | continue; |
4990 | bp = adp->ad_buf; |
4991 | gotit = getdirtybuf(bp, waitfor); |
4992 | if (gotit == 0) { |
4993 | if (waitfor == MNT_NOWAIT2) |
4994 | continue; |
4995 | break; |
4996 | } else if (gotit == -1) |
4997 | goto retry_ino; |
4998 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
4999 | if (waitfor == MNT_NOWAIT2) { |
5000 | bawrite(bp); |
5001 | } else if ((error = VOP_BWRITE(bp)) != 0) { |
5002 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5003 | return (error); |
5004 | } |
5005 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5006 | break; |
5007 | } |
5008 | if (adp != NULL((void *)0)) |
5009 | continue; |
5010 | retry_newino: |
5011 | TAILQ_FOREACH(adp, &inodedep->id_newinoupdt, ad_next)for((adp) = ((&inodedep->id_newinoupdt)->tqh_first) ; (adp) != ((void *)0); (adp) = ((adp)->ad_next.tqe_next)) { |
5012 | if (adp->ad_statead_list.wk_state & DEPCOMPLETE0x0008) |
5013 | continue; |
5014 | bp = adp->ad_buf; |
5015 | gotit = getdirtybuf(bp, waitfor); |
5016 | if (gotit == 0) { |
5017 | if (waitfor == MNT_NOWAIT2) |
5018 | continue; |
5019 | break; |
5020 | } else if (gotit == -1) |
5021 | goto retry_newino; |
5022 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5023 | if (waitfor == MNT_NOWAIT2) { |
5024 | bawrite(bp); |
5025 | } else if ((error = VOP_BWRITE(bp)) != 0) { |
5026 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5027 | return (error); |
5028 | } |
5029 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5030 | break; |
5031 | } |
5032 | if (adp != NULL((void *)0)) |
5033 | continue; |
5034 | /* |
5035 | * If pass2, we are done, otherwise do pass 2. |
5036 | */ |
5037 | if (waitfor == MNT_WAIT1) |
5038 | break; |
5039 | waitfor = MNT_WAIT1; |
5040 | } |
5041 | /* |
5042 | * Try freeing inodedep in case all dependencies have been removed. |
5043 | */ |
5044 | if (inodedep_lookup(fs, ino, 0, &inodedep) != 0) |
5045 | (void) free_inodedep(inodedep); |
5046 | return (0); |
5047 | } |
5048 | |
5049 | /* |
5050 | * Eliminate a pagedep dependency by flushing out all its diradd dependencies. |
5051 | * Called with splbio blocked. |
5052 | */ |
5053 | STATIC int |
5054 | flush_pagedep_deps(struct vnode *pvp, struct mount *mp, |
5055 | struct diraddhd *diraddhdp) |
5056 | { |
5057 | struct proc *p = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc; /* XXX */ |
5058 | struct worklist *wk; |
5059 | struct inodedep *inodedep; |
5060 | struct ufsmount *ump; |
5061 | struct diradd *dap; |
5062 | struct vnode *vp; |
5063 | int gotit, error = 0; |
5064 | struct buf *bp; |
5065 | ufsino_t inum; |
5066 | |
5067 | splassert(IPL_BIO)do { if (splassert_ctl > 0) { splassert_check(0x6, __func__ ); } } while (0); |
5068 | |
5069 | ump = VFSTOUFS(mp)((struct ufsmount *)((mp)->mnt_data)); |
5070 | while ((dap = LIST_FIRST(diraddhdp)((diraddhdp)->lh_first)) != NULL((void *)0)) { |
5071 | /* |
5072 | * Flush ourselves if this directory entry |
5073 | * has a MKDIR_PARENT dependency. |
5074 | */ |
5075 | if (dap->da_stateda_list.wk_state & MKDIR_PARENT0x0010) { |
5076 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5077 | if ((error = UFS_UPDATE(VTOI(pvp), 1)((((struct inode *)(pvp)->v_data))->i_vtbl->iv_update )((((struct inode *)(pvp)->v_data)), (1)))) |
5078 | break; |
5079 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5080 | /* |
5081 | * If that cleared dependencies, go on to next. |
5082 | */ |
5083 | if (dap != LIST_FIRST(diraddhdp)((diraddhdp)->lh_first)) |
5084 | continue; |
5085 | if (dap->da_stateda_list.wk_state & MKDIR_PARENT0x0010) { |
5086 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5087 | panic("flush_pagedep_deps: MKDIR_PARENT"); |
5088 | } |
5089 | } |
5090 | /* |
5091 | * A newly allocated directory must have its "." and |
5092 | * ".." entries written out before its name can be |
5093 | * committed in its parent. We do not want or need |
5094 | * the full semantics of a synchronous VOP_FSYNC as |
5095 | * that may end up here again, once for each directory |
5096 | * level in the filesystem. Instead, we push the blocks |
5097 | * and wait for them to clear. We have to fsync twice |
5098 | * because the first call may choose to defer blocks |
5099 | * that still have dependencies, but deferral will |
5100 | * happen at most once. |
5101 | */ |
5102 | inum = dap->da_newinum; |
5103 | if (dap->da_stateda_list.wk_state & MKDIR_BODY0x0020) { |
5104 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5105 | if ((error = VFS_VGET(mp, inum, &vp)(*(mp)->mnt_op->vfs_vget)(mp, inum, &vp)) != 0) |
5106 | break; |
5107 | if ((error=VOP_FSYNC(vp, p->p_ucred, MNT_NOWAIT2, p)) || |
5108 | (error=VOP_FSYNC(vp, p->p_ucred, MNT_NOWAIT2, p))) { |
5109 | vput(vp); |
5110 | break; |
5111 | } |
5112 | drain_output(vp, 0); |
5113 | /* |
5114 | * If first block is still dirty with a D_MKDIR |
5115 | * dependency then it needs to be written now. |
5116 | */ |
5117 | for (;;) { |
5118 | error = 0; |
5119 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5120 | bp = incore(vp, 0); |
5121 | if (bp == NULL((void *)0)) { |
5122 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5123 | break; |
5124 | } |
5125 | LIST_FOREACH(wk, &bp->b_dep, wk_list)for((wk) = ((&bp->b_dep)->lh_first); (wk)!= ((void * )0); (wk) = ((wk)->wk_list.le_next)) |
5126 | if (wk->wk_type == D_MKDIR11) |
5127 | break; |
5128 | if (wk) { |
5129 | gotit = getdirtybuf(bp, MNT_WAIT1); |
5130 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5131 | if (gotit == -1) |
5132 | continue; |
5133 | if (gotit && (error = bwrite(bp)) != 0) |
5134 | break; |
5135 | } else |
5136 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5137 | break; |
5138 | } |
5139 | vput(vp); |
5140 | /* Flushing of first block failed */ |
5141 | if (error) |
5142 | break; |
5143 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5144 | /* |
5145 | * If that cleared dependencies, go on to next. |
5146 | */ |
5147 | if (dap != LIST_FIRST(diraddhdp)((diraddhdp)->lh_first)) |
5148 | continue; |
5149 | if (dap->da_stateda_list.wk_state & MKDIR_BODY0x0020) { |
5150 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5151 | panic("flush_pagedep_deps: MKDIR_BODY"); |
5152 | } |
5153 | } |
5154 | /* |
5155 | * Flush the inode on which the directory entry depends. |
5156 | * Having accounted for MKDIR_PARENT and MKDIR_BODY above, |
5157 | * the only remaining dependency is that the updated inode |
5158 | * count must get pushed to disk. The inode has already |
5159 | * been pushed into its inode buffer (via VOP_UPDATE) at |
5160 | * the time of the reference count change. So we need only |
5161 | * locate that buffer, ensure that there will be no rollback |
5162 | * caused by a bitmap dependency, then write the inode buffer. |
5163 | */ |
5164 | if (inodedep_lookup(ump->um_fsufsmount_u.fs, inum, 0, &inodedep) == 0) { |
5165 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5166 | panic("flush_pagedep_deps: lost inode"); |
5167 | } |
5168 | /* |
5169 | * If the inode still has bitmap dependencies, |
5170 | * push them to disk. |
5171 | */ |
5172 | retry: |
5173 | if ((inodedep->id_stateid_list.wk_state & DEPCOMPLETE0x0008) == 0) { |
5174 | bp = inodedep->id_buf; |
5175 | gotit = getdirtybuf(bp, MNT_WAIT1); |
5176 | if (gotit == -1) |
5177 | goto retry; |
5178 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5179 | if (gotit && (error = bwrite(bp)) != 0) |
5180 | break; |
5181 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5182 | if (dap != LIST_FIRST(diraddhdp)((diraddhdp)->lh_first)) |
5183 | continue; |
5184 | } |
5185 | /* |
5186 | * If the inode is still sitting in a buffer waiting |
5187 | * to be written, push it to disk. |
5188 | */ |
5189 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5190 | if ((error = bread(ump->um_devvp, |
5191 | fsbtodb(ump->um_fs, ino_to_fsba(ump->um_fs, inum))((((daddr_t)(((((daddr_t)(ump->ufsmount_u.fs)->fs_fpg * (((inum) / (ump->ufsmount_u.fs)->fs_ipg))) + (ump-> ufsmount_u.fs)->fs_cgoffset * ((((inum) / (ump->ufsmount_u .fs)->fs_ipg)) & ~((ump->ufsmount_u.fs)->fs_cgmask ))) + (ump->ufsmount_u.fs)->fs_iblkno) + ((((((inum) % ( ump->ufsmount_u.fs)->fs_ipg) / ((ump->ufsmount_u.fs) ->fs_inopb))) << ((ump->ufsmount_u.fs))->fs_fragshift ))))) << (ump->ufsmount_u.fs)->fs_fsbtodb), |
5192 | (int)ump->um_fsufsmount_u.fs->fs_bsize, &bp)) != 0) { |
5193 | brelse(bp); |
5194 | break; |
5195 | } |
5196 | if ((error = bwrite(bp)) != 0) |
5197 | break; |
5198 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5199 | /* |
5200 | * If we have failed to get rid of all the dependencies |
5201 | * then something is seriously wrong. |
5202 | */ |
5203 | if (dap == LIST_FIRST(diraddhdp)((diraddhdp)->lh_first)) { |
5204 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5205 | panic("flush_pagedep_deps: flush failed"); |
5206 | } |
5207 | } |
5208 | if (error) |
5209 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5210 | return (error); |
5211 | } |
5212 | |
5213 | /* |
5214 | * A large burst of file addition or deletion activity can drive the |
5215 | * memory load excessively high. First attempt to slow things down |
5216 | * using the techniques below. If that fails, this routine requests |
5217 | * the offending operations to fall back to running synchronously |
5218 | * until the memory load returns to a reasonable level. |
5219 | */ |
5220 | int |
5221 | softdep_slowdown(struct vnode *vp) |
5222 | { |
5223 | int max_softdeps_hard; |
5224 | |
5225 | max_softdeps_hard = max_softdeps * 11 / 10; |
5226 | if (num_dirrem < max_softdeps_hard / 2 && |
5227 | num_inodedep < max_softdeps_hard) |
5228 | return (0); |
5229 | stat_sync_limit_hit += 1; |
5230 | return (1); |
5231 | } |
5232 | |
5233 | /* |
5234 | * If memory utilization has gotten too high, deliberately slow things |
5235 | * down and speed up the I/O processing. |
5236 | */ |
5237 | STATIC int |
5238 | request_cleanup(int resource, int islocked) |
5239 | { |
5240 | struct proc *p = CURPROC({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc; |
5241 | int s; |
5242 | |
5243 | /* |
5244 | * We never hold up the filesystem syncer process. |
5245 | */ |
5246 | if (p == filesys_syncer || (p->p_flag & P_SOFTDEP0x10000000)) |
5247 | return (0); |
5248 | /* |
5249 | * First check to see if the work list has gotten backlogged. |
5250 | * If it has, co-opt this process to help clean up two entries. |
5251 | * Because this process may hold inodes locked, we cannot |
5252 | * handle any remove requests that might block on a locked |
5253 | * inode as that could lead to deadlock. We set P_SOFTDEP |
5254 | * to avoid recursively processing the worklist. |
5255 | */ |
5256 | if (num_on_worklist > max_softdeps / 10) { |
5257 | atomic_setbits_intx86_atomic_setbits_u32(&p->p_flag, P_SOFTDEP0x10000000); |
5258 | if (islocked) |
5259 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5260 | process_worklist_item(NULL((void *)0), NULL((void *)0), LK_NOWAIT0x0040UL); |
5261 | process_worklist_item(NULL((void *)0), NULL((void *)0), LK_NOWAIT0x0040UL); |
5262 | atomic_clearbits_intx86_atomic_clearbits_u32(&p->p_flag, P_SOFTDEP0x10000000); |
5263 | stat_worklist_push += 2; |
5264 | if (islocked) |
5265 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5266 | return(1); |
5267 | } |
5268 | /* |
5269 | * Next, we attempt to speed up the syncer process. If that |
5270 | * is successful, then we allow the process to continue. |
5271 | */ |
5272 | if (speedup_syncer()) |
5273 | return(0); |
5274 | /* |
5275 | * If we are resource constrained on inode dependencies, try |
5276 | * flushing some dirty inodes. Otherwise, we are constrained |
5277 | * by file deletions, so try accelerating flushes of directories |
5278 | * with removal dependencies. We would like to do the cleanup |
5279 | * here, but we probably hold an inode locked at this point and |
5280 | * that might deadlock against one that we try to clean. So, |
5281 | * the best that we can do is request the syncer daemon to do |
5282 | * the cleanup for us. |
5283 | */ |
5284 | switch (resource) { |
5285 | |
5286 | case FLUSH_INODES1: |
5287 | stat_ino_limit_push += 1; |
5288 | req_clear_inodedeps += 1; |
5289 | stat_countp = &stat_ino_limit_hit; |
5290 | break; |
5291 | |
5292 | case FLUSH_REMOVE2: |
5293 | stat_blk_limit_push += 1; |
5294 | req_clear_remove += 1; |
5295 | stat_countp = &stat_blk_limit_hit; |
5296 | break; |
5297 | |
5298 | default: |
5299 | if (islocked) |
5300 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5301 | panic("request_cleanup: unknown type"); |
5302 | } |
5303 | /* |
5304 | * Hopefully the syncer daemon will catch up and awaken us. |
5305 | * We wait at most tickdelay before proceeding in any case. |
5306 | */ |
5307 | if (islocked == 0) |
5308 | ACQUIRE_LOCK(&lk)(&lk)->lkt_spl = splraise(0x6); |
5309 | proc_waiting += 1; |
5310 | if (!timeout_pending(&proc_waiting_timeout)((&proc_waiting_timeout)->to_flags & 0x02)) |
5311 | timeout_add(&proc_waiting_timeout, tickdelay > 2 ? tickdelay : 2); |
5312 | |
5313 | s = FREE_LOCK_INTERLOCKED(&lk)((&lk)->lkt_spl); |
5314 | tsleep_nsec(&proc_waiting, PPAUSE40, "softupdate", INFSLP0xffffffffffffffffULL); |
5315 | ACQUIRE_LOCK_INTERLOCKED(&lk, s)(&lk)->lkt_spl = (s); |
5316 | proc_waiting -= 1; |
5317 | if (islocked == 0) |
5318 | FREE_LOCK(&lk)spllower((&lk)->lkt_spl); |
5319 | return (1); |
5320 | } |
5321 | |
5322 | /* |
5323 | * Awaken processes pausing in request_cleanup and clear proc_waiting |
5324 | * to indicate that there is no longer a timer running. |