Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
1da177e4 LT |
21 | #include "xfs_log.h" |
22 | #include "xfs_trans.h" | |
a844f451 NS |
23 | #include "xfs_sb.h" |
24 | #include "xfs_ag.h" | |
1da177e4 LT |
25 | #include "xfs_mount.h" |
26 | #include "xfs_error.h" | |
27 | #include "xfs_log_priv.h" | |
28 | #include "xfs_buf_item.h" | |
a844f451 | 29 | #include "xfs_bmap_btree.h" |
1da177e4 | 30 | #include "xfs_alloc_btree.h" |
a844f451 | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 | 32 | #include "xfs_log_recover.h" |
1da177e4 | 33 | #include "xfs_trans_priv.h" |
a844f451 NS |
34 | #include "xfs_dinode.h" |
35 | #include "xfs_inode.h" | |
0b1b213f | 36 | #include "xfs_trace.h" |
f661f1e0 | 37 | #include "xfs_fsops.h" |
0e446be4 | 38 | #include "xfs_cksum.h" |
1da177e4 | 39 | |
eb01c9cd | 40 | kmem_zone_t *xfs_log_ticket_zone; |
1da177e4 | 41 | |
1da177e4 | 42 | /* Local miscellaneous function prototypes */ |
ad223e60 MT |
43 | STATIC int |
44 | xlog_commit_record( | |
45 | struct xlog *log, | |
46 | struct xlog_ticket *ticket, | |
47 | struct xlog_in_core **iclog, | |
48 | xfs_lsn_t *commitlsnp); | |
49 | ||
9a8d2fdb MT |
50 | STATIC struct xlog * |
51 | xlog_alloc_log( | |
52 | struct xfs_mount *mp, | |
53 | struct xfs_buftarg *log_target, | |
54 | xfs_daddr_t blk_offset, | |
55 | int num_bblks); | |
ad223e60 MT |
56 | STATIC int |
57 | xlog_space_left( | |
58 | struct xlog *log, | |
59 | atomic64_t *head); | |
9a8d2fdb MT |
60 | STATIC int |
61 | xlog_sync( | |
62 | struct xlog *log, | |
63 | struct xlog_in_core *iclog); | |
64 | STATIC void | |
65 | xlog_dealloc_log( | |
66 | struct xlog *log); | |
1da177e4 LT |
67 | |
68 | /* local state machine functions */ | |
69 | STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int); | |
9a8d2fdb MT |
70 | STATIC void |
71 | xlog_state_do_callback( | |
72 | struct xlog *log, | |
73 | int aborted, | |
74 | struct xlog_in_core *iclog); | |
75 | STATIC int | |
76 | xlog_state_get_iclog_space( | |
77 | struct xlog *log, | |
78 | int len, | |
79 | struct xlog_in_core **iclog, | |
80 | struct xlog_ticket *ticket, | |
81 | int *continued_write, | |
82 | int *logoffsetp); | |
83 | STATIC int | |
84 | xlog_state_release_iclog( | |
85 | struct xlog *log, | |
86 | struct xlog_in_core *iclog); | |
87 | STATIC void | |
88 | xlog_state_switch_iclogs( | |
89 | struct xlog *log, | |
90 | struct xlog_in_core *iclog, | |
91 | int eventual_size); | |
92 | STATIC void | |
93 | xlog_state_want_sync( | |
94 | struct xlog *log, | |
95 | struct xlog_in_core *iclog); | |
1da177e4 | 96 | |
ad223e60 MT |
97 | STATIC void |
98 | xlog_grant_push_ail( | |
9a8d2fdb MT |
99 | struct xlog *log, |
100 | int need_bytes); | |
101 | STATIC void | |
102 | xlog_regrant_reserve_log_space( | |
103 | struct xlog *log, | |
104 | struct xlog_ticket *ticket); | |
105 | STATIC void | |
106 | xlog_ungrant_log_space( | |
107 | struct xlog *log, | |
108 | struct xlog_ticket *ticket); | |
1da177e4 | 109 | |
cfcbbbd0 | 110 | #if defined(DEBUG) |
9a8d2fdb MT |
111 | STATIC void |
112 | xlog_verify_dest_ptr( | |
113 | struct xlog *log, | |
114 | char *ptr); | |
ad223e60 MT |
115 | STATIC void |
116 | xlog_verify_grant_tail( | |
9a8d2fdb MT |
117 | struct xlog *log); |
118 | STATIC void | |
119 | xlog_verify_iclog( | |
120 | struct xlog *log, | |
121 | struct xlog_in_core *iclog, | |
122 | int count, | |
667a9291 | 123 | bool syncing); |
9a8d2fdb MT |
124 | STATIC void |
125 | xlog_verify_tail_lsn( | |
126 | struct xlog *log, | |
127 | struct xlog_in_core *iclog, | |
128 | xfs_lsn_t tail_lsn); | |
1da177e4 LT |
129 | #else |
130 | #define xlog_verify_dest_ptr(a,b) | |
3f336c6f | 131 | #define xlog_verify_grant_tail(a) |
1da177e4 LT |
132 | #define xlog_verify_iclog(a,b,c,d) |
133 | #define xlog_verify_tail_lsn(a,b,c) | |
134 | #endif | |
135 | ||
9a8d2fdb MT |
136 | STATIC int |
137 | xlog_iclogs_empty( | |
138 | struct xlog *log); | |
1da177e4 | 139 | |
dd954c69 | 140 | static void |
663e496a | 141 | xlog_grant_sub_space( |
ad223e60 MT |
142 | struct xlog *log, |
143 | atomic64_t *head, | |
144 | int bytes) | |
dd954c69 | 145 | { |
d0eb2f38 DC |
146 | int64_t head_val = atomic64_read(head); |
147 | int64_t new, old; | |
a69ed03c | 148 | |
d0eb2f38 DC |
149 | do { |
150 | int cycle, space; | |
a69ed03c | 151 | |
d0eb2f38 | 152 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 153 | |
d0eb2f38 DC |
154 | space -= bytes; |
155 | if (space < 0) { | |
156 | space += log->l_logsize; | |
157 | cycle--; | |
158 | } | |
159 | ||
160 | old = head_val; | |
161 | new = xlog_assign_grant_head_val(cycle, space); | |
162 | head_val = atomic64_cmpxchg(head, old, new); | |
163 | } while (head_val != old); | |
dd954c69 CH |
164 | } |
165 | ||
166 | static void | |
663e496a | 167 | xlog_grant_add_space( |
ad223e60 MT |
168 | struct xlog *log, |
169 | atomic64_t *head, | |
170 | int bytes) | |
dd954c69 | 171 | { |
d0eb2f38 DC |
172 | int64_t head_val = atomic64_read(head); |
173 | int64_t new, old; | |
a69ed03c | 174 | |
d0eb2f38 DC |
175 | do { |
176 | int tmp; | |
177 | int cycle, space; | |
a69ed03c | 178 | |
d0eb2f38 | 179 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 180 | |
d0eb2f38 DC |
181 | tmp = log->l_logsize - space; |
182 | if (tmp > bytes) | |
183 | space += bytes; | |
184 | else { | |
185 | space = bytes - tmp; | |
186 | cycle++; | |
187 | } | |
188 | ||
189 | old = head_val; | |
190 | new = xlog_assign_grant_head_val(cycle, space); | |
191 | head_val = atomic64_cmpxchg(head, old, new); | |
192 | } while (head_val != old); | |
dd954c69 | 193 | } |
a69ed03c | 194 | |
c303c5b8 CH |
195 | STATIC void |
196 | xlog_grant_head_init( | |
197 | struct xlog_grant_head *head) | |
198 | { | |
199 | xlog_assign_grant_head(&head->grant, 1, 0); | |
200 | INIT_LIST_HEAD(&head->waiters); | |
201 | spin_lock_init(&head->lock); | |
202 | } | |
203 | ||
a79bf2d7 CH |
204 | STATIC void |
205 | xlog_grant_head_wake_all( | |
206 | struct xlog_grant_head *head) | |
207 | { | |
208 | struct xlog_ticket *tic; | |
209 | ||
210 | spin_lock(&head->lock); | |
211 | list_for_each_entry(tic, &head->waiters, t_queue) | |
212 | wake_up_process(tic->t_task); | |
213 | spin_unlock(&head->lock); | |
214 | } | |
215 | ||
e179840d CH |
216 | static inline int |
217 | xlog_ticket_reservation( | |
ad223e60 | 218 | struct xlog *log, |
e179840d CH |
219 | struct xlog_grant_head *head, |
220 | struct xlog_ticket *tic) | |
9f9c19ec | 221 | { |
e179840d CH |
222 | if (head == &log->l_write_head) { |
223 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
224 | return tic->t_unit_res; | |
225 | } else { | |
9f9c19ec | 226 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) |
e179840d | 227 | return tic->t_unit_res * tic->t_cnt; |
9f9c19ec | 228 | else |
e179840d | 229 | return tic->t_unit_res; |
9f9c19ec | 230 | } |
9f9c19ec CH |
231 | } |
232 | ||
233 | STATIC bool | |
e179840d | 234 | xlog_grant_head_wake( |
ad223e60 | 235 | struct xlog *log, |
e179840d | 236 | struct xlog_grant_head *head, |
9f9c19ec CH |
237 | int *free_bytes) |
238 | { | |
239 | struct xlog_ticket *tic; | |
240 | int need_bytes; | |
241 | ||
e179840d CH |
242 | list_for_each_entry(tic, &head->waiters, t_queue) { |
243 | need_bytes = xlog_ticket_reservation(log, head, tic); | |
9f9c19ec CH |
244 | if (*free_bytes < need_bytes) |
245 | return false; | |
9f9c19ec | 246 | |
e179840d CH |
247 | *free_bytes -= need_bytes; |
248 | trace_xfs_log_grant_wake_up(log, tic); | |
14a7235f | 249 | wake_up_process(tic->t_task); |
9f9c19ec CH |
250 | } |
251 | ||
252 | return true; | |
253 | } | |
254 | ||
255 | STATIC int | |
23ee3df3 | 256 | xlog_grant_head_wait( |
ad223e60 | 257 | struct xlog *log, |
23ee3df3 | 258 | struct xlog_grant_head *head, |
9f9c19ec | 259 | struct xlog_ticket *tic, |
a30b0367 DC |
260 | int need_bytes) __releases(&head->lock) |
261 | __acquires(&head->lock) | |
9f9c19ec | 262 | { |
23ee3df3 | 263 | list_add_tail(&tic->t_queue, &head->waiters); |
9f9c19ec CH |
264 | |
265 | do { | |
266 | if (XLOG_FORCED_SHUTDOWN(log)) | |
267 | goto shutdown; | |
268 | xlog_grant_push_ail(log, need_bytes); | |
269 | ||
14a7235f | 270 | __set_current_state(TASK_UNINTERRUPTIBLE); |
23ee3df3 | 271 | spin_unlock(&head->lock); |
14a7235f | 272 | |
9f9c19ec | 273 | XFS_STATS_INC(xs_sleep_logspace); |
9f9c19ec | 274 | |
14a7235f CH |
275 | trace_xfs_log_grant_sleep(log, tic); |
276 | schedule(); | |
9f9c19ec CH |
277 | trace_xfs_log_grant_wake(log, tic); |
278 | ||
23ee3df3 | 279 | spin_lock(&head->lock); |
9f9c19ec CH |
280 | if (XLOG_FORCED_SHUTDOWN(log)) |
281 | goto shutdown; | |
23ee3df3 | 282 | } while (xlog_space_left(log, &head->grant) < need_bytes); |
9f9c19ec CH |
283 | |
284 | list_del_init(&tic->t_queue); | |
285 | return 0; | |
286 | shutdown: | |
287 | list_del_init(&tic->t_queue); | |
288 | return XFS_ERROR(EIO); | |
289 | } | |
290 | ||
42ceedb3 CH |
291 | /* |
292 | * Atomically get the log space required for a log ticket. | |
293 | * | |
294 | * Once a ticket gets put onto head->waiters, it will only return after the | |
295 | * needed reservation is satisfied. | |
296 | * | |
297 | * This function is structured so that it has a lock free fast path. This is | |
298 | * necessary because every new transaction reservation will come through this | |
299 | * path. Hence any lock will be globally hot if we take it unconditionally on | |
300 | * every pass. | |
301 | * | |
302 | * As tickets are only ever moved on and off head->waiters under head->lock, we | |
303 | * only need to take that lock if we are going to add the ticket to the queue | |
304 | * and sleep. We can avoid taking the lock if the ticket was never added to | |
305 | * head->waiters because the t_queue list head will be empty and we hold the | |
306 | * only reference to it so it can safely be checked unlocked. | |
307 | */ | |
308 | STATIC int | |
309 | xlog_grant_head_check( | |
ad223e60 | 310 | struct xlog *log, |
42ceedb3 CH |
311 | struct xlog_grant_head *head, |
312 | struct xlog_ticket *tic, | |
313 | int *need_bytes) | |
314 | { | |
315 | int free_bytes; | |
316 | int error = 0; | |
317 | ||
318 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); | |
319 | ||
320 | /* | |
321 | * If there are other waiters on the queue then give them a chance at | |
322 | * logspace before us. Wake up the first waiters, if we do not wake | |
323 | * up all the waiters then go to sleep waiting for more free space, | |
324 | * otherwise try to get some space for this transaction. | |
325 | */ | |
326 | *need_bytes = xlog_ticket_reservation(log, head, tic); | |
327 | free_bytes = xlog_space_left(log, &head->grant); | |
328 | if (!list_empty_careful(&head->waiters)) { | |
329 | spin_lock(&head->lock); | |
330 | if (!xlog_grant_head_wake(log, head, &free_bytes) || | |
331 | free_bytes < *need_bytes) { | |
332 | error = xlog_grant_head_wait(log, head, tic, | |
333 | *need_bytes); | |
334 | } | |
335 | spin_unlock(&head->lock); | |
336 | } else if (free_bytes < *need_bytes) { | |
337 | spin_lock(&head->lock); | |
338 | error = xlog_grant_head_wait(log, head, tic, *need_bytes); | |
339 | spin_unlock(&head->lock); | |
340 | } | |
341 | ||
342 | return error; | |
343 | } | |
344 | ||
0adba536 CH |
345 | static void |
346 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
347 | { | |
348 | tic->t_res_num = 0; | |
349 | tic->t_res_arr_sum = 0; | |
350 | tic->t_res_num_ophdrs = 0; | |
351 | } | |
352 | ||
353 | static void | |
354 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
355 | { | |
356 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
357 | /* add to overflow and start again */ | |
358 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
359 | tic->t_res_num = 0; | |
360 | tic->t_res_arr_sum = 0; | |
361 | } | |
362 | ||
363 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
364 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
365 | tic->t_res_arr_sum += len; | |
366 | tic->t_res_num++; | |
367 | } | |
dd954c69 | 368 | |
9006fb91 CH |
369 | /* |
370 | * Replenish the byte reservation required by moving the grant write head. | |
371 | */ | |
372 | int | |
373 | xfs_log_regrant( | |
374 | struct xfs_mount *mp, | |
375 | struct xlog_ticket *tic) | |
376 | { | |
ad223e60 | 377 | struct xlog *log = mp->m_log; |
9006fb91 CH |
378 | int need_bytes; |
379 | int error = 0; | |
380 | ||
381 | if (XLOG_FORCED_SHUTDOWN(log)) | |
382 | return XFS_ERROR(EIO); | |
383 | ||
384 | XFS_STATS_INC(xs_try_logspace); | |
385 | ||
386 | /* | |
387 | * This is a new transaction on the ticket, so we need to change the | |
388 | * transaction ID so that the next transaction has a different TID in | |
389 | * the log. Just add one to the existing tid so that we can see chains | |
390 | * of rolling transactions in the log easily. | |
391 | */ | |
392 | tic->t_tid++; | |
393 | ||
394 | xlog_grant_push_ail(log, tic->t_unit_res); | |
395 | ||
396 | tic->t_curr_res = tic->t_unit_res; | |
397 | xlog_tic_reset_res(tic); | |
398 | ||
399 | if (tic->t_cnt > 0) | |
400 | return 0; | |
401 | ||
402 | trace_xfs_log_regrant(log, tic); | |
403 | ||
404 | error = xlog_grant_head_check(log, &log->l_write_head, tic, | |
405 | &need_bytes); | |
406 | if (error) | |
407 | goto out_error; | |
408 | ||
409 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
410 | trace_xfs_log_regrant_exit(log, tic); | |
411 | xlog_verify_grant_tail(log); | |
412 | return 0; | |
413 | ||
414 | out_error: | |
415 | /* | |
416 | * If we are failing, make sure the ticket doesn't have any current | |
417 | * reservations. We don't want to add this back when the ticket/ | |
418 | * transaction gets cancelled. | |
419 | */ | |
420 | tic->t_curr_res = 0; | |
421 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
422 | return error; | |
423 | } | |
424 | ||
425 | /* | |
426 | * Reserve log space and return a ticket corresponding the reservation. | |
427 | * | |
428 | * Each reservation is going to reserve extra space for a log record header. | |
429 | * When writes happen to the on-disk log, we don't subtract the length of the | |
430 | * log record header from any reservation. By wasting space in each | |
431 | * reservation, we prevent over allocation problems. | |
432 | */ | |
433 | int | |
434 | xfs_log_reserve( | |
435 | struct xfs_mount *mp, | |
436 | int unit_bytes, | |
437 | int cnt, | |
438 | struct xlog_ticket **ticp, | |
439 | __uint8_t client, | |
440 | bool permanent, | |
441 | uint t_type) | |
442 | { | |
ad223e60 | 443 | struct xlog *log = mp->m_log; |
9006fb91 CH |
444 | struct xlog_ticket *tic; |
445 | int need_bytes; | |
446 | int error = 0; | |
447 | ||
448 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); | |
449 | ||
450 | if (XLOG_FORCED_SHUTDOWN(log)) | |
451 | return XFS_ERROR(EIO); | |
452 | ||
453 | XFS_STATS_INC(xs_try_logspace); | |
454 | ||
455 | ASSERT(*ticp == NULL); | |
456 | tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, | |
457 | KM_SLEEP | KM_MAYFAIL); | |
458 | if (!tic) | |
459 | return XFS_ERROR(ENOMEM); | |
460 | ||
461 | tic->t_trans_type = t_type; | |
462 | *ticp = tic; | |
463 | ||
437a255a DC |
464 | xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt |
465 | : tic->t_unit_res); | |
9006fb91 CH |
466 | |
467 | trace_xfs_log_reserve(log, tic); | |
468 | ||
469 | error = xlog_grant_head_check(log, &log->l_reserve_head, tic, | |
470 | &need_bytes); | |
471 | if (error) | |
472 | goto out_error; | |
473 | ||
474 | xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); | |
475 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
476 | trace_xfs_log_reserve_exit(log, tic); | |
477 | xlog_verify_grant_tail(log); | |
478 | return 0; | |
479 | ||
480 | out_error: | |
481 | /* | |
482 | * If we are failing, make sure the ticket doesn't have any current | |
483 | * reservations. We don't want to add this back when the ticket/ | |
484 | * transaction gets cancelled. | |
485 | */ | |
486 | tic->t_curr_res = 0; | |
487 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
488 | return error; | |
489 | } | |
490 | ||
491 | ||
1da177e4 LT |
492 | /* |
493 | * NOTES: | |
494 | * | |
495 | * 1. currblock field gets updated at startup and after in-core logs | |
496 | * marked as with WANT_SYNC. | |
497 | */ | |
498 | ||
499 | /* | |
500 | * This routine is called when a user of a log manager ticket is done with | |
501 | * the reservation. If the ticket was ever used, then a commit record for | |
502 | * the associated transaction is written out as a log operation header with | |
503 | * no data. The flag XLOG_TIC_INITED is set when the first write occurs with | |
504 | * a given ticket. If the ticket was one with a permanent reservation, then | |
505 | * a few operations are done differently. Permanent reservation tickets by | |
506 | * default don't release the reservation. They just commit the current | |
507 | * transaction with the belief that the reservation is still needed. A flag | |
508 | * must be passed in before permanent reservations are actually released. | |
509 | * When these type of tickets are not released, they need to be set into | |
510 | * the inited state again. By doing this, a start record will be written | |
511 | * out when the next write occurs. | |
512 | */ | |
513 | xfs_lsn_t | |
35a8a72f CH |
514 | xfs_log_done( |
515 | struct xfs_mount *mp, | |
516 | struct xlog_ticket *ticket, | |
517 | struct xlog_in_core **iclog, | |
518 | uint flags) | |
1da177e4 | 519 | { |
ad223e60 | 520 | struct xlog *log = mp->m_log; |
35a8a72f | 521 | xfs_lsn_t lsn = 0; |
1da177e4 | 522 | |
1da177e4 LT |
523 | if (XLOG_FORCED_SHUTDOWN(log) || |
524 | /* | |
525 | * If nothing was ever written, don't write out commit record. | |
526 | * If we get an error, just continue and give back the log ticket. | |
527 | */ | |
528 | (((ticket->t_flags & XLOG_TIC_INITED) == 0) && | |
55b66332 | 529 | (xlog_commit_record(log, ticket, iclog, &lsn)))) { |
1da177e4 LT |
530 | lsn = (xfs_lsn_t) -1; |
531 | if (ticket->t_flags & XLOG_TIC_PERM_RESERV) { | |
532 | flags |= XFS_LOG_REL_PERM_RESERV; | |
533 | } | |
534 | } | |
535 | ||
536 | ||
537 | if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 || | |
538 | (flags & XFS_LOG_REL_PERM_RESERV)) { | |
0b1b213f CH |
539 | trace_xfs_log_done_nonperm(log, ticket); |
540 | ||
1da177e4 | 541 | /* |
c41564b5 | 542 | * Release ticket if not permanent reservation or a specific |
1da177e4 LT |
543 | * request has been made to release a permanent reservation. |
544 | */ | |
545 | xlog_ungrant_log_space(log, ticket); | |
cc09c0dc | 546 | xfs_log_ticket_put(ticket); |
1da177e4 | 547 | } else { |
0b1b213f CH |
548 | trace_xfs_log_done_perm(log, ticket); |
549 | ||
1da177e4 | 550 | xlog_regrant_reserve_log_space(log, ticket); |
c6a7b0f8 LM |
551 | /* If this ticket was a permanent reservation and we aren't |
552 | * trying to release it, reset the inited flags; so next time | |
553 | * we write, a start record will be written out. | |
554 | */ | |
1da177e4 | 555 | ticket->t_flags |= XLOG_TIC_INITED; |
c6a7b0f8 | 556 | } |
1da177e4 LT |
557 | |
558 | return lsn; | |
35a8a72f | 559 | } |
1da177e4 | 560 | |
1da177e4 LT |
561 | /* |
562 | * Attaches a new iclog I/O completion callback routine during | |
563 | * transaction commit. If the log is in error state, a non-zero | |
564 | * return code is handed back and the caller is responsible for | |
565 | * executing the callback at an appropriate time. | |
566 | */ | |
567 | int | |
35a8a72f CH |
568 | xfs_log_notify( |
569 | struct xfs_mount *mp, | |
570 | struct xlog_in_core *iclog, | |
571 | xfs_log_callback_t *cb) | |
1da177e4 | 572 | { |
b22cd72c | 573 | int abortflg; |
1da177e4 | 574 | |
114d23aa | 575 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
576 | abortflg = (iclog->ic_state & XLOG_STATE_IOERROR); |
577 | if (!abortflg) { | |
578 | ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) || | |
579 | (iclog->ic_state == XLOG_STATE_WANT_SYNC)); | |
580 | cb->cb_next = NULL; | |
581 | *(iclog->ic_callback_tail) = cb; | |
582 | iclog->ic_callback_tail = &(cb->cb_next); | |
583 | } | |
114d23aa | 584 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 | 585 | return abortflg; |
35a8a72f | 586 | } |
1da177e4 LT |
587 | |
588 | int | |
35a8a72f CH |
589 | xfs_log_release_iclog( |
590 | struct xfs_mount *mp, | |
591 | struct xlog_in_core *iclog) | |
1da177e4 | 592 | { |
35a8a72f | 593 | if (xlog_state_release_iclog(mp->m_log, iclog)) { |
7d04a335 | 594 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 595 | return EIO; |
1da177e4 LT |
596 | } |
597 | ||
598 | return 0; | |
599 | } | |
600 | ||
1da177e4 LT |
601 | /* |
602 | * Mount a log filesystem | |
603 | * | |
604 | * mp - ubiquitous xfs mount point structure | |
605 | * log_target - buftarg of on-disk log device | |
606 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
607 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
608 | * | |
609 | * Return error or zero. | |
610 | */ | |
611 | int | |
249a8c11 DC |
612 | xfs_log_mount( |
613 | xfs_mount_t *mp, | |
614 | xfs_buftarg_t *log_target, | |
615 | xfs_daddr_t blk_offset, | |
616 | int num_bblks) | |
1da177e4 | 617 | { |
3e7b91cf JL |
618 | int error = 0; |
619 | int min_logfsbs; | |
249a8c11 | 620 | |
1da177e4 | 621 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) |
a0fa2b67 | 622 | xfs_notice(mp, "Mounting Filesystem"); |
1da177e4 | 623 | else { |
a0fa2b67 DC |
624 | xfs_notice(mp, |
625 | "Mounting filesystem in no-recovery mode. Filesystem will be inconsistent."); | |
bd186aa9 | 626 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
627 | } |
628 | ||
629 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e DC |
630 | if (IS_ERR(mp->m_log)) { |
631 | error = -PTR_ERR(mp->m_log); | |
644c3567 DC |
632 | goto out; |
633 | } | |
1da177e4 | 634 | |
3e7b91cf JL |
635 | /* |
636 | * Validate the given log space and drop a critical message via syslog | |
637 | * if the log size is too small that would lead to some unexpected | |
638 | * situations in transaction log space reservation stage. | |
639 | * | |
640 | * Note: we can't just reject the mount if the validation fails. This | |
641 | * would mean that people would have to downgrade their kernel just to | |
642 | * remedy the situation as there is no way to grow the log (short of | |
643 | * black magic surgery with xfs_db). | |
644 | * | |
645 | * We can, however, reject mounts for CRC format filesystems, as the | |
646 | * mkfs binary being used to make the filesystem should never create a | |
647 | * filesystem with a log that is too small. | |
648 | */ | |
649 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
650 | ||
651 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
652 | xfs_warn(mp, | |
653 | "Log size %d blocks too small, minimum size is %d blocks", | |
654 | mp->m_sb.sb_logblocks, min_logfsbs); | |
655 | error = EINVAL; | |
656 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { | |
657 | xfs_warn(mp, | |
658 | "Log size %d blocks too large, maximum size is %lld blocks", | |
659 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
660 | error = EINVAL; | |
661 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { | |
662 | xfs_warn(mp, | |
663 | "log size %lld bytes too large, maximum size is %lld bytes", | |
664 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
665 | XFS_MAX_LOG_BYTES); | |
666 | error = EINVAL; | |
667 | } | |
668 | if (error) { | |
669 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
670 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); | |
671 | ASSERT(0); | |
672 | goto out_free_log; | |
673 | } | |
674 | xfs_crit(mp, | |
675 | "Log size out of supported range. Continuing onwards, but if log hangs are\n" | |
676 | "experienced then please report this message in the bug report."); | |
677 | } | |
678 | ||
249a8c11 DC |
679 | /* |
680 | * Initialize the AIL now we have a log. | |
681 | */ | |
249a8c11 DC |
682 | error = xfs_trans_ail_init(mp); |
683 | if (error) { | |
a0fa2b67 | 684 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 685 | goto out_free_log; |
249a8c11 | 686 | } |
a9c21c1b | 687 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 688 | |
1da177e4 LT |
689 | /* |
690 | * skip log recovery on a norecovery mount. pretend it all | |
691 | * just worked. | |
692 | */ | |
693 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 694 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
695 | |
696 | if (readonly) | |
bd186aa9 | 697 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 698 | |
65be6054 | 699 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
700 | |
701 | if (readonly) | |
bd186aa9 | 702 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 703 | if (error) { |
a0fa2b67 DC |
704 | xfs_warn(mp, "log mount/recovery failed: error %d", |
705 | error); | |
26430752 | 706 | goto out_destroy_ail; |
1da177e4 LT |
707 | } |
708 | } | |
709 | ||
710 | /* Normal transactions can now occur */ | |
711 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
712 | ||
71e330b5 DC |
713 | /* |
714 | * Now the log has been fully initialised and we know were our | |
715 | * space grant counters are, we can initialise the permanent ticket | |
716 | * needed for delayed logging to work. | |
717 | */ | |
718 | xlog_cil_init_post_recovery(mp->m_log); | |
719 | ||
1da177e4 | 720 | return 0; |
26430752 CH |
721 | |
722 | out_destroy_ail: | |
723 | xfs_trans_ail_destroy(mp); | |
724 | out_free_log: | |
725 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 726 | out: |
249a8c11 | 727 | return error; |
26430752 | 728 | } |
1da177e4 LT |
729 | |
730 | /* | |
f661f1e0 DC |
731 | * Finish the recovery of the file system. This is separate from the |
732 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
733 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
734 | * here. | |
1da177e4 | 735 | * |
f661f1e0 DC |
736 | * If we finish recovery successfully, start the background log work. If we are |
737 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
738 | * it. | |
1da177e4 LT |
739 | */ |
740 | int | |
4249023a | 741 | xfs_log_mount_finish(xfs_mount_t *mp) |
1da177e4 | 742 | { |
f661f1e0 | 743 | int error = 0; |
1da177e4 | 744 | |
f661f1e0 | 745 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
4249023a | 746 | error = xlog_recover_finish(mp->m_log); |
f661f1e0 DC |
747 | if (!error) |
748 | xfs_log_work_queue(mp); | |
749 | } else { | |
bd186aa9 | 750 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
751 | } |
752 | ||
f661f1e0 | 753 | |
1da177e4 LT |
754 | return error; |
755 | } | |
756 | ||
1da177e4 LT |
757 | /* |
758 | * Final log writes as part of unmount. | |
759 | * | |
760 | * Mark the filesystem clean as unmount happens. Note that during relocation | |
761 | * this routine needs to be executed as part of source-bag while the | |
762 | * deallocation must not be done until source-end. | |
763 | */ | |
764 | ||
765 | /* | |
766 | * Unmount record used to have a string "Unmount filesystem--" in the | |
767 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
768 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 769 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
770 | * As far as I know, there weren't any dependencies on the old behaviour. |
771 | */ | |
772 | ||
773 | int | |
774 | xfs_log_unmount_write(xfs_mount_t *mp) | |
775 | { | |
9a8d2fdb | 776 | struct xlog *log = mp->m_log; |
1da177e4 LT |
777 | xlog_in_core_t *iclog; |
778 | #ifdef DEBUG | |
779 | xlog_in_core_t *first_iclog; | |
780 | #endif | |
35a8a72f | 781 | xlog_ticket_t *tic = NULL; |
1da177e4 LT |
782 | xfs_lsn_t lsn; |
783 | int error; | |
1da177e4 | 784 | |
1da177e4 LT |
785 | /* |
786 | * Don't write out unmount record on read-only mounts. | |
787 | * Or, if we are doing a forced umount (typically because of IO errors). | |
788 | */ | |
bd186aa9 | 789 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
790 | return 0; |
791 | ||
a14a348b | 792 | error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
b911ca04 | 793 | ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log))); |
1da177e4 LT |
794 | |
795 | #ifdef DEBUG | |
796 | first_iclog = iclog = log->l_iclog; | |
797 | do { | |
798 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
799 | ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE); | |
800 | ASSERT(iclog->ic_offset == 0); | |
801 | } | |
802 | iclog = iclog->ic_next; | |
803 | } while (iclog != first_iclog); | |
804 | #endif | |
805 | if (! (XLOG_FORCED_SHUTDOWN(log))) { | |
955e47ad TS |
806 | error = xfs_log_reserve(mp, 600, 1, &tic, |
807 | XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE); | |
1da177e4 | 808 | if (!error) { |
55b66332 DC |
809 | /* the data section must be 32 bit size aligned */ |
810 | struct { | |
811 | __uint16_t magic; | |
812 | __uint16_t pad1; | |
813 | __uint32_t pad2; /* may as well make it 64 bits */ | |
814 | } magic = { | |
815 | .magic = XLOG_UNMOUNT_TYPE, | |
816 | }; | |
817 | struct xfs_log_iovec reg = { | |
4e0d5f92 | 818 | .i_addr = &magic, |
55b66332 DC |
819 | .i_len = sizeof(magic), |
820 | .i_type = XLOG_REG_TYPE_UNMOUNT, | |
821 | }; | |
822 | struct xfs_log_vec vec = { | |
823 | .lv_niovecs = 1, | |
824 | .lv_iovecp = ®, | |
825 | }; | |
826 | ||
3948659e | 827 | /* remove inited flag, and account for space used */ |
55b66332 | 828 | tic->t_flags = 0; |
3948659e | 829 | tic->t_curr_res -= sizeof(magic); |
55b66332 | 830 | error = xlog_write(log, &vec, tic, &lsn, |
1da177e4 LT |
831 | NULL, XLOG_UNMOUNT_TRANS); |
832 | /* | |
833 | * At this point, we're umounting anyway, | |
834 | * so there's no point in transitioning log state | |
835 | * to IOERROR. Just continue... | |
836 | */ | |
837 | } | |
838 | ||
a0fa2b67 DC |
839 | if (error) |
840 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
1da177e4 LT |
841 | |
842 | ||
b22cd72c | 843 | spin_lock(&log->l_icloglock); |
1da177e4 | 844 | iclog = log->l_iclog; |
155cc6b7 | 845 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 846 | xlog_state_want_sync(log, iclog); |
39e2defe | 847 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 848 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 849 | |
b22cd72c | 850 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
851 | if (!(iclog->ic_state == XLOG_STATE_ACTIVE || |
852 | iclog->ic_state == XLOG_STATE_DIRTY)) { | |
853 | if (!XLOG_FORCED_SHUTDOWN(log)) { | |
eb40a875 DC |
854 | xlog_wait(&iclog->ic_force_wait, |
855 | &log->l_icloglock); | |
1da177e4 | 856 | } else { |
b22cd72c | 857 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
858 | } |
859 | } else { | |
b22cd72c | 860 | spin_unlock(&log->l_icloglock); |
1da177e4 | 861 | } |
955e47ad | 862 | if (tic) { |
0b1b213f | 863 | trace_xfs_log_umount_write(log, tic); |
955e47ad | 864 | xlog_ungrant_log_space(log, tic); |
cc09c0dc | 865 | xfs_log_ticket_put(tic); |
955e47ad | 866 | } |
1da177e4 LT |
867 | } else { |
868 | /* | |
869 | * We're already in forced_shutdown mode, couldn't | |
870 | * even attempt to write out the unmount transaction. | |
871 | * | |
872 | * Go through the motions of sync'ing and releasing | |
873 | * the iclog, even though no I/O will actually happen, | |
c41564b5 | 874 | * we need to wait for other log I/Os that may already |
1da177e4 LT |
875 | * be in progress. Do this as a separate section of |
876 | * code so we'll know if we ever get stuck here that | |
877 | * we're in this odd situation of trying to unmount | |
878 | * a file system that went into forced_shutdown as | |
879 | * the result of an unmount.. | |
880 | */ | |
b22cd72c | 881 | spin_lock(&log->l_icloglock); |
1da177e4 | 882 | iclog = log->l_iclog; |
155cc6b7 | 883 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 LT |
884 | |
885 | xlog_state_want_sync(log, iclog); | |
39e2defe | 886 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 887 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 888 | |
b22cd72c | 889 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
890 | |
891 | if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE | |
892 | || iclog->ic_state == XLOG_STATE_DIRTY | |
893 | || iclog->ic_state == XLOG_STATE_IOERROR) ) { | |
894 | ||
eb40a875 DC |
895 | xlog_wait(&iclog->ic_force_wait, |
896 | &log->l_icloglock); | |
1da177e4 | 897 | } else { |
b22cd72c | 898 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
899 | } |
900 | } | |
901 | ||
1bb7d6b5 | 902 | return error; |
1da177e4 LT |
903 | } /* xfs_log_unmount_write */ |
904 | ||
905 | /* | |
c75921a7 | 906 | * Empty the log for unmount/freeze. |
cf2931db DC |
907 | * |
908 | * To do this, we first need to shut down the background log work so it is not | |
909 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
910 | * the log so we can then flush them out. Once they have completed their IO and | |
911 | * run the callbacks removing themselves from the AIL, we can write the unmount | |
c75921a7 | 912 | * record. |
1da177e4 LT |
913 | */ |
914 | void | |
c75921a7 DC |
915 | xfs_log_quiesce( |
916 | struct xfs_mount *mp) | |
1da177e4 | 917 | { |
f661f1e0 | 918 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
919 | xfs_log_force(mp, XFS_LOG_SYNC); |
920 | ||
921 | /* | |
922 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
923 | * will push it, xfs_wait_buftarg() will not wait for it. Further, | |
924 | * xfs_buf_iowait() cannot be used because it was pushed with the | |
925 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
926 | * the IO to complete. | |
927 | */ | |
928 | xfs_ail_push_all_sync(mp->m_ail); | |
929 | xfs_wait_buftarg(mp->m_ddev_targp); | |
930 | xfs_buf_lock(mp->m_sb_bp); | |
931 | xfs_buf_unlock(mp->m_sb_bp); | |
932 | ||
933 | xfs_log_unmount_write(mp); | |
c75921a7 DC |
934 | } |
935 | ||
936 | /* | |
937 | * Shut down and release the AIL and Log. | |
938 | * | |
939 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
940 | * from the AIL so that the log is empty before we write the unmount record to | |
941 | * the log. Once this is done, we can tear down the AIL and the log. | |
942 | */ | |
943 | void | |
944 | xfs_log_unmount( | |
945 | struct xfs_mount *mp) | |
946 | { | |
947 | xfs_log_quiesce(mp); | |
cf2931db | 948 | |
249a8c11 | 949 | xfs_trans_ail_destroy(mp); |
c41564b5 | 950 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
951 | } |
952 | ||
43f5efc5 DC |
953 | void |
954 | xfs_log_item_init( | |
955 | struct xfs_mount *mp, | |
956 | struct xfs_log_item *item, | |
957 | int type, | |
272e42b2 | 958 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
959 | { |
960 | item->li_mountp = mp; | |
961 | item->li_ailp = mp->m_ail; | |
962 | item->li_type = type; | |
963 | item->li_ops = ops; | |
71e330b5 DC |
964 | item->li_lv = NULL; |
965 | ||
966 | INIT_LIST_HEAD(&item->li_ail); | |
967 | INIT_LIST_HEAD(&item->li_cil); | |
43f5efc5 DC |
968 | } |
969 | ||
09a423a3 CH |
970 | /* |
971 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 972 | */ |
1da177e4 | 973 | void |
09a423a3 | 974 | xfs_log_space_wake( |
cfb7cdca | 975 | struct xfs_mount *mp) |
1da177e4 | 976 | { |
ad223e60 | 977 | struct xlog *log = mp->m_log; |
cfb7cdca | 978 | int free_bytes; |
1da177e4 | 979 | |
1da177e4 LT |
980 | if (XLOG_FORCED_SHUTDOWN(log)) |
981 | return; | |
1da177e4 | 982 | |
28496968 | 983 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
984 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
985 | ||
28496968 CH |
986 | spin_lock(&log->l_write_head.lock); |
987 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 988 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 989 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 990 | } |
10547941 | 991 | |
28496968 | 992 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
993 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
994 | ||
28496968 CH |
995 | spin_lock(&log->l_reserve_head.lock); |
996 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 997 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 998 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 999 | } |
3f16b985 | 1000 | } |
1da177e4 LT |
1001 | |
1002 | /* | |
2c6e24ce DC |
1003 | * Determine if we have a transaction that has gone to disk that needs to be |
1004 | * covered. To begin the transition to the idle state firstly the log needs to | |
1005 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1006 | * we start attempting to cover the log. | |
b6f8dd49 | 1007 | * |
2c6e24ce DC |
1008 | * Only if we are then in a state where covering is needed, the caller is |
1009 | * informed that dummy transactions are required to move the log into the idle | |
1010 | * state. | |
1011 | * | |
1012 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1013 | * cover the log as we may be in a situation where there isn't log space | |
1014 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1015 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1016 | * there's no point in running a dummy transaction at this point because we | |
1017 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 LT |
1018 | */ |
1019 | int | |
1020 | xfs_log_need_covered(xfs_mount_t *mp) | |
1021 | { | |
9a8d2fdb | 1022 | struct xlog *log = mp->m_log; |
2c6e24ce | 1023 | int needed = 0; |
1da177e4 | 1024 | |
92821e2b | 1025 | if (!xfs_fs_writable(mp)) |
1da177e4 LT |
1026 | return 0; |
1027 | ||
2c6e24ce DC |
1028 | if (!xlog_cil_empty(log)) |
1029 | return 0; | |
1030 | ||
b22cd72c | 1031 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1032 | switch (log->l_covered_state) { |
1033 | case XLOG_STATE_COVER_DONE: | |
1034 | case XLOG_STATE_COVER_DONE2: | |
1035 | case XLOG_STATE_COVER_IDLE: | |
1036 | break; | |
1037 | case XLOG_STATE_COVER_NEED: | |
1038 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1039 | if (xfs_ail_min_lsn(log->l_ailp)) |
1040 | break; | |
1041 | if (!xlog_iclogs_empty(log)) | |
1042 | break; | |
1043 | ||
1044 | needed = 1; | |
1045 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) | |
1046 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1047 | else | |
1048 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1049 | break; | |
b6f8dd49 | 1050 | default: |
1da177e4 | 1051 | needed = 1; |
b6f8dd49 | 1052 | break; |
1da177e4 | 1053 | } |
b22cd72c | 1054 | spin_unlock(&log->l_icloglock); |
014c2544 | 1055 | return needed; |
1da177e4 LT |
1056 | } |
1057 | ||
09a423a3 | 1058 | /* |
1da177e4 LT |
1059 | * We may be holding the log iclog lock upon entering this routine. |
1060 | */ | |
1061 | xfs_lsn_t | |
1c304625 | 1062 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1063 | struct xfs_mount *mp) |
1da177e4 | 1064 | { |
ad223e60 | 1065 | struct xlog *log = mp->m_log; |
1c304625 CH |
1066 | struct xfs_log_item *lip; |
1067 | xfs_lsn_t tail_lsn; | |
1068 | ||
1069 | assert_spin_locked(&mp->m_ail->xa_lock); | |
1da177e4 | 1070 | |
09a423a3 CH |
1071 | /* |
1072 | * To make sure we always have a valid LSN for the log tail we keep | |
1073 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1074 | * and use that when the AIL was empty. |
09a423a3 | 1075 | */ |
1c304625 CH |
1076 | lip = xfs_ail_min(mp->m_ail); |
1077 | if (lip) | |
1078 | tail_lsn = lip->li_lsn; | |
1079 | else | |
84f3c683 | 1080 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
1c3cb9ec | 1081 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1082 | return tail_lsn; |
1c3cb9ec | 1083 | } |
1da177e4 | 1084 | |
1c304625 CH |
1085 | xfs_lsn_t |
1086 | xlog_assign_tail_lsn( | |
1087 | struct xfs_mount *mp) | |
1088 | { | |
1089 | xfs_lsn_t tail_lsn; | |
1090 | ||
1091 | spin_lock(&mp->m_ail->xa_lock); | |
1092 | tail_lsn = xlog_assign_tail_lsn_locked(mp); | |
1093 | spin_unlock(&mp->m_ail->xa_lock); | |
1094 | ||
1095 | return tail_lsn; | |
1096 | } | |
1097 | ||
1da177e4 LT |
1098 | /* |
1099 | * Return the space in the log between the tail and the head. The head | |
1100 | * is passed in the cycle/bytes formal parms. In the special case where | |
1101 | * the reserve head has wrapped passed the tail, this calculation is no | |
1102 | * longer valid. In this case, just return 0 which means there is no space | |
1103 | * in the log. This works for all places where this function is called | |
1104 | * with the reserve head. Of course, if the write head were to ever | |
1105 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1106 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1107 | * | |
1108 | * This code also handles the case where the reservation head is behind | |
1109 | * the tail. The details of this case are described below, but the end | |
1110 | * result is that we return the size of the log as the amount of space left. | |
1111 | */ | |
a8272ce0 | 1112 | STATIC int |
a69ed03c | 1113 | xlog_space_left( |
ad223e60 | 1114 | struct xlog *log, |
c8a09ff8 | 1115 | atomic64_t *head) |
1da177e4 | 1116 | { |
a69ed03c DC |
1117 | int free_bytes; |
1118 | int tail_bytes; | |
1119 | int tail_cycle; | |
1120 | int head_cycle; | |
1121 | int head_bytes; | |
1da177e4 | 1122 | |
a69ed03c | 1123 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1124 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1125 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1126 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1127 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1128 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1129 | return 0; |
a69ed03c DC |
1130 | else if (tail_cycle < head_cycle) { |
1131 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1132 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1133 | } else { |
1134 | /* | |
1135 | * The reservation head is behind the tail. | |
1136 | * In this case we just want to return the size of the | |
1137 | * log as the amount of space left. | |
1138 | */ | |
a0fa2b67 | 1139 | xfs_alert(log->l_mp, |
1da177e4 LT |
1140 | "xlog_space_left: head behind tail\n" |
1141 | " tail_cycle = %d, tail_bytes = %d\n" | |
1142 | " GH cycle = %d, GH bytes = %d", | |
a69ed03c | 1143 | tail_cycle, tail_bytes, head_cycle, head_bytes); |
1da177e4 LT |
1144 | ASSERT(0); |
1145 | free_bytes = log->l_logsize; | |
1146 | } | |
1147 | return free_bytes; | |
a69ed03c | 1148 | } |
1da177e4 LT |
1149 | |
1150 | ||
1151 | /* | |
1152 | * Log function which is called when an io completes. | |
1153 | * | |
1154 | * The log manager needs its own routine, in order to control what | |
1155 | * happens with the buffer after the write completes. | |
1156 | */ | |
1157 | void | |
1158 | xlog_iodone(xfs_buf_t *bp) | |
1159 | { | |
9a8d2fdb MT |
1160 | struct xlog_in_core *iclog = bp->b_fspriv; |
1161 | struct xlog *l = iclog->ic_log; | |
1162 | int aborted = 0; | |
1da177e4 LT |
1163 | |
1164 | /* | |
1165 | * Race to shutdown the filesystem if we see an error. | |
1166 | */ | |
5a52c2a5 | 1167 | if (XFS_TEST_ERROR((xfs_buf_geterror(bp)), l->l_mp, |
1da177e4 | 1168 | XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) { |
901796af | 1169 | xfs_buf_ioerror_alert(bp, __func__); |
c867cb61 | 1170 | xfs_buf_stale(bp); |
7d04a335 | 1171 | xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 LT |
1172 | /* |
1173 | * This flag will be propagated to the trans-committed | |
1174 | * callback routines to let them know that the log-commit | |
1175 | * didn't succeed. | |
1176 | */ | |
1177 | aborted = XFS_LI_ABORTED; | |
1178 | } else if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
1179 | aborted = XFS_LI_ABORTED; | |
1180 | } | |
3db296f3 DC |
1181 | |
1182 | /* log I/O is always issued ASYNC */ | |
1183 | ASSERT(XFS_BUF_ISASYNC(bp)); | |
1da177e4 | 1184 | xlog_state_done_syncing(iclog, aborted); |
3db296f3 DC |
1185 | /* |
1186 | * do not reference the buffer (bp) here as we could race | |
1187 | * with it being freed after writing the unmount record to the | |
1188 | * log. | |
1189 | */ | |
c3f8fc73 | 1190 | } |
1da177e4 | 1191 | |
1da177e4 LT |
1192 | /* |
1193 | * Return size of each in-core log record buffer. | |
1194 | * | |
9da096fd | 1195 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1196 | * |
1197 | * If the filesystem blocksize is too large, we may need to choose a | |
1198 | * larger size since the directory code currently logs entire blocks. | |
1199 | */ | |
1200 | ||
1201 | STATIC void | |
9a8d2fdb MT |
1202 | xlog_get_iclog_buffer_size( |
1203 | struct xfs_mount *mp, | |
1204 | struct xlog *log) | |
1da177e4 LT |
1205 | { |
1206 | int size; | |
1207 | int xhdrs; | |
1208 | ||
1cb51258 ES |
1209 | if (mp->m_logbufs <= 0) |
1210 | log->l_iclog_bufs = XLOG_MAX_ICLOGS; | |
1211 | else | |
cfcbbbd0 | 1212 | log->l_iclog_bufs = mp->m_logbufs; |
1da177e4 LT |
1213 | |
1214 | /* | |
1215 | * Buffer size passed in from mount system call. | |
1216 | */ | |
cfcbbbd0 | 1217 | if (mp->m_logbsize > 0) { |
1da177e4 LT |
1218 | size = log->l_iclog_size = mp->m_logbsize; |
1219 | log->l_iclog_size_log = 0; | |
1220 | while (size != 1) { | |
1221 | log->l_iclog_size_log++; | |
1222 | size >>= 1; | |
1223 | } | |
1224 | ||
62118709 | 1225 | if (xfs_sb_version_haslogv2(&mp->m_sb)) { |
9da096fd MP |
1226 | /* # headers = size / 32k |
1227 | * one header holds cycles from 32k of data | |
1da177e4 LT |
1228 | */ |
1229 | ||
1230 | xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE; | |
1231 | if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE) | |
1232 | xhdrs++; | |
1233 | log->l_iclog_hsize = xhdrs << BBSHIFT; | |
1234 | log->l_iclog_heads = xhdrs; | |
1235 | } else { | |
1236 | ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE); | |
1237 | log->l_iclog_hsize = BBSIZE; | |
1238 | log->l_iclog_heads = 1; | |
1239 | } | |
cfcbbbd0 | 1240 | goto done; |
1da177e4 LT |
1241 | } |
1242 | ||
9da096fd | 1243 | /* All machines use 32kB buffers by default. */ |
1cb51258 ES |
1244 | log->l_iclog_size = XLOG_BIG_RECORD_BSIZE; |
1245 | log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT; | |
1da177e4 LT |
1246 | |
1247 | /* the default log size is 16k or 32k which is one header sector */ | |
1248 | log->l_iclog_hsize = BBSIZE; | |
1249 | log->l_iclog_heads = 1; | |
1250 | ||
7153f8ba CH |
1251 | done: |
1252 | /* are we being asked to make the sizes selected above visible? */ | |
cfcbbbd0 NS |
1253 | if (mp->m_logbufs == 0) |
1254 | mp->m_logbufs = log->l_iclog_bufs; | |
1255 | if (mp->m_logbsize == 0) | |
1256 | mp->m_logbsize = log->l_iclog_size; | |
1da177e4 LT |
1257 | } /* xlog_get_iclog_buffer_size */ |
1258 | ||
1259 | ||
f661f1e0 DC |
1260 | void |
1261 | xfs_log_work_queue( | |
1262 | struct xfs_mount *mp) | |
1263 | { | |
5889608d | 1264 | queue_delayed_work(mp->m_log_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1265 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * Every sync period we need to unpin all items in the AIL and push them to | |
1270 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1271 | * indicate that the filesystem is idle. | |
1272 | */ | |
1273 | void | |
1274 | xfs_log_worker( | |
1275 | struct work_struct *work) | |
1276 | { | |
1277 | struct xlog *log = container_of(to_delayed_work(work), | |
1278 | struct xlog, l_work); | |
1279 | struct xfs_mount *mp = log->l_mp; | |
1280 | ||
1281 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
1282 | if (xfs_log_need_covered(mp)) | |
1283 | xfs_fs_log_dummy(mp); | |
1284 | else | |
1285 | xfs_log_force(mp, 0); | |
1286 | ||
1287 | /* start pushing all the metadata that is currently dirty */ | |
1288 | xfs_ail_push_all(mp->m_ail); | |
1289 | ||
1290 | /* queue us up again */ | |
1291 | xfs_log_work_queue(mp); | |
1292 | } | |
1293 | ||
1da177e4 LT |
1294 | /* |
1295 | * This routine initializes some of the log structure for a given mount point. | |
1296 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1297 | * some other stuff may be filled in too. | |
1298 | */ | |
9a8d2fdb MT |
1299 | STATIC struct xlog * |
1300 | xlog_alloc_log( | |
1301 | struct xfs_mount *mp, | |
1302 | struct xfs_buftarg *log_target, | |
1303 | xfs_daddr_t blk_offset, | |
1304 | int num_bblks) | |
1da177e4 | 1305 | { |
9a8d2fdb | 1306 | struct xlog *log; |
1da177e4 LT |
1307 | xlog_rec_header_t *head; |
1308 | xlog_in_core_t **iclogp; | |
1309 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1310 | xfs_buf_t *bp; | |
1311 | int i; | |
a6cb767e | 1312 | int error = ENOMEM; |
69ce58f0 | 1313 | uint log2_size = 0; |
1da177e4 | 1314 | |
9a8d2fdb | 1315 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1316 | if (!log) { |
a0fa2b67 | 1317 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1318 | goto out; |
1319 | } | |
1da177e4 LT |
1320 | |
1321 | log->l_mp = mp; | |
1322 | log->l_targ = log_target; | |
1323 | log->l_logsize = BBTOB(num_bblks); | |
1324 | log->l_logBBstart = blk_offset; | |
1325 | log->l_logBBsize = num_bblks; | |
1326 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1327 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1328 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1329 | |
1330 | log->l_prev_block = -1; | |
1da177e4 | 1331 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1332 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1333 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1334 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 CH |
1335 | |
1336 | xlog_grant_head_init(&log->l_reserve_head); | |
1337 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1338 | |
a6cb767e | 1339 | error = EFSCORRUPTED; |
62118709 | 1340 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1341 | log2_size = mp->m_sb.sb_logsectlog; |
1342 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1343 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1344 | log2_size, BBSHIFT); | |
a6cb767e DC |
1345 | goto out_free_log; |
1346 | } | |
1347 | ||
69ce58f0 AE |
1348 | log2_size -= BBSHIFT; |
1349 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1350 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1351 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1352 | goto out_free_log; |
1353 | } | |
69ce58f0 AE |
1354 | |
1355 | /* for larger sector sizes, must have v2 or external log */ | |
1356 | if (log2_size && log->l_logBBstart > 0 && | |
1357 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1358 | xfs_warn(mp, |
1359 | "log sector size (0x%x) invalid for configuration.", | |
1360 | log2_size); | |
a6cb767e DC |
1361 | goto out_free_log; |
1362 | } | |
1da177e4 | 1363 | } |
69ce58f0 | 1364 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1365 | |
1366 | xlog_get_iclog_buffer_size(mp, log); | |
1367 | ||
a6cb767e | 1368 | error = ENOMEM; |
e70b73f8 | 1369 | bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0); |
644c3567 DC |
1370 | if (!bp) |
1371 | goto out_free_log; | |
cb669ca5 | 1372 | bp->b_iodone = xlog_iodone; |
0c842ad4 | 1373 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
1374 | log->l_xbuf = bp; |
1375 | ||
007c61c6 | 1376 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1377 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1378 | |
1da177e4 LT |
1379 | iclogp = &log->l_iclog; |
1380 | /* | |
1381 | * The amount of memory to allocate for the iclog structure is | |
1382 | * rather funky due to the way the structure is defined. It is | |
1383 | * done this way so that we can use different sizes for machines | |
1384 | * with different amounts of memory. See the definition of | |
1385 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1386 | */ | |
1da177e4 LT |
1387 | ASSERT(log->l_iclog_size >= 4096); |
1388 | for (i=0; i < log->l_iclog_bufs; i++) { | |
644c3567 DC |
1389 | *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL); |
1390 | if (!*iclogp) | |
1391 | goto out_free_iclog; | |
1392 | ||
1da177e4 | 1393 | iclog = *iclogp; |
1da177e4 LT |
1394 | iclog->ic_prev = prev_iclog; |
1395 | prev_iclog = iclog; | |
1fa40b01 | 1396 | |
686865f7 | 1397 | bp = xfs_buf_get_uncached(mp->m_logdev_targp, |
e70b73f8 | 1398 | BTOBB(log->l_iclog_size), 0); |
644c3567 DC |
1399 | if (!bp) |
1400 | goto out_free_iclog; | |
c8da0faf | 1401 | |
cb669ca5 | 1402 | bp->b_iodone = xlog_iodone; |
1fa40b01 | 1403 | iclog->ic_bp = bp; |
b28708d6 | 1404 | iclog->ic_data = bp->b_addr; |
4679b2d3 | 1405 | #ifdef DEBUG |
1da177e4 | 1406 | log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header); |
4679b2d3 | 1407 | #endif |
1da177e4 LT |
1408 | head = &iclog->ic_header; |
1409 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1410 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1411 | head->h_version = cpu_to_be32( | |
62118709 | 1412 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1413 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1414 | /* new fields */ |
b53e675d | 1415 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1416 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1417 | ||
4e94b71b | 1418 | iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize; |
1da177e4 LT |
1419 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1420 | iclog->ic_log = log; | |
114d23aa DC |
1421 | atomic_set(&iclog->ic_refcnt, 0); |
1422 | spin_lock_init(&iclog->ic_callback_lock); | |
1da177e4 | 1423 | iclog->ic_callback_tail = &(iclog->ic_callback); |
b28708d6 | 1424 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1425 | |
0c842ad4 | 1426 | ASSERT(xfs_buf_islocked(iclog->ic_bp)); |
eb40a875 DC |
1427 | init_waitqueue_head(&iclog->ic_force_wait); |
1428 | init_waitqueue_head(&iclog->ic_write_wait); | |
1da177e4 LT |
1429 | |
1430 | iclogp = &iclog->ic_next; | |
1431 | } | |
1432 | *iclogp = log->l_iclog; /* complete ring */ | |
1433 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1434 | ||
71e330b5 DC |
1435 | error = xlog_cil_init(log); |
1436 | if (error) | |
1437 | goto out_free_iclog; | |
1da177e4 | 1438 | return log; |
644c3567 DC |
1439 | |
1440 | out_free_iclog: | |
1441 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1442 | prev_iclog = iclog->ic_next; | |
eb40a875 | 1443 | if (iclog->ic_bp) |
644c3567 | 1444 | xfs_buf_free(iclog->ic_bp); |
644c3567 DC |
1445 | kmem_free(iclog); |
1446 | } | |
1447 | spinlock_destroy(&log->l_icloglock); | |
644c3567 DC |
1448 | xfs_buf_free(log->l_xbuf); |
1449 | out_free_log: | |
1450 | kmem_free(log); | |
a6cb767e DC |
1451 | out: |
1452 | return ERR_PTR(-error); | |
1da177e4 LT |
1453 | } /* xlog_alloc_log */ |
1454 | ||
1455 | ||
1456 | /* | |
1457 | * Write out the commit record of a transaction associated with the given | |
1458 | * ticket. Return the lsn of the commit record. | |
1459 | */ | |
1460 | STATIC int | |
55b66332 | 1461 | xlog_commit_record( |
ad223e60 | 1462 | struct xlog *log, |
55b66332 DC |
1463 | struct xlog_ticket *ticket, |
1464 | struct xlog_in_core **iclog, | |
1465 | xfs_lsn_t *commitlsnp) | |
1da177e4 | 1466 | { |
55b66332 DC |
1467 | struct xfs_mount *mp = log->l_mp; |
1468 | int error; | |
1469 | struct xfs_log_iovec reg = { | |
1470 | .i_addr = NULL, | |
1471 | .i_len = 0, | |
1472 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1473 | }; | |
1474 | struct xfs_log_vec vec = { | |
1475 | .lv_niovecs = 1, | |
1476 | .lv_iovecp = ®, | |
1477 | }; | |
1da177e4 LT |
1478 | |
1479 | ASSERT_ALWAYS(iclog); | |
55b66332 DC |
1480 | error = xlog_write(log, &vec, ticket, commitlsnp, iclog, |
1481 | XLOG_COMMIT_TRANS); | |
1482 | if (error) | |
7d04a335 | 1483 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1484 | return error; |
55b66332 | 1485 | } |
1da177e4 LT |
1486 | |
1487 | /* | |
1488 | * Push on the buffer cache code if we ever use more than 75% of the on-disk | |
1489 | * log space. This code pushes on the lsn which would supposedly free up | |
1490 | * the 25% which we want to leave free. We may need to adopt a policy which | |
1491 | * pushes on an lsn which is further along in the log once we reach the high | |
1492 | * water mark. In this manner, we would be creating a low water mark. | |
1493 | */ | |
a8272ce0 | 1494 | STATIC void |
2ced19cb | 1495 | xlog_grant_push_ail( |
ad223e60 | 1496 | struct xlog *log, |
2ced19cb | 1497 | int need_bytes) |
1da177e4 | 1498 | { |
2ced19cb | 1499 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1500 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1501 | int free_blocks; |
1502 | int free_bytes; | |
1503 | int threshold_block; | |
1504 | int threshold_cycle; | |
1505 | int free_threshold; | |
1506 | ||
1507 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1508 | ||
28496968 | 1509 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1510 | free_blocks = BTOBBT(free_bytes); |
1511 | ||
1512 | /* | |
1513 | * Set the threshold for the minimum number of free blocks in the | |
1514 | * log to the maximum of what the caller needs, one quarter of the | |
1515 | * log, and 256 blocks. | |
1516 | */ | |
1517 | free_threshold = BTOBB(need_bytes); | |
1518 | free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2)); | |
1519 | free_threshold = MAX(free_threshold, 256); | |
1520 | if (free_blocks >= free_threshold) | |
1521 | return; | |
1522 | ||
1c3cb9ec DC |
1523 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1524 | &threshold_block); | |
1525 | threshold_block += free_threshold; | |
1da177e4 | 1526 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1527 | threshold_block -= log->l_logBBsize; |
1528 | threshold_cycle += 1; | |
1da177e4 | 1529 | } |
2ced19cb DC |
1530 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1531 | threshold_block); | |
1532 | /* | |
1533 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1534 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1535 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1536 | */ |
84f3c683 DC |
1537 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1538 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1539 | threshold_lsn = last_sync_lsn; | |
2ced19cb DC |
1540 | |
1541 | /* | |
1542 | * Get the transaction layer to kick the dirty buffers out to | |
1543 | * disk asynchronously. No point in trying to do this if | |
1544 | * the filesystem is shutting down. | |
1545 | */ | |
1546 | if (!XLOG_FORCED_SHUTDOWN(log)) | |
fd074841 | 1547 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1548 | } |
1da177e4 | 1549 | |
0e446be4 CH |
1550 | /* |
1551 | * Stamp cycle number in every block | |
1552 | */ | |
1553 | STATIC void | |
1554 | xlog_pack_data( | |
1555 | struct xlog *log, | |
1556 | struct xlog_in_core *iclog, | |
1557 | int roundoff) | |
1558 | { | |
1559 | int i, j, k; | |
1560 | int size = iclog->ic_offset + roundoff; | |
1561 | __be32 cycle_lsn; | |
1562 | xfs_caddr_t dp; | |
1563 | ||
1564 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1565 | ||
1566 | dp = iclog->ic_datap; | |
1567 | for (i = 0; i < BTOBB(size); i++) { | |
1568 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1569 | break; | |
1570 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1571 | *(__be32 *)dp = cycle_lsn; | |
1572 | dp += BBSIZE; | |
1573 | } | |
1574 | ||
1575 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1576 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1577 | ||
1578 | for ( ; i < BTOBB(size); i++) { | |
1579 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1580 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1581 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1582 | *(__be32 *)dp = cycle_lsn; | |
1583 | dp += BBSIZE; | |
1584 | } | |
1585 | ||
1586 | for (i = 1; i < log->l_iclog_heads; i++) | |
1587 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | /* | |
1592 | * Calculate the checksum for a log buffer. | |
1593 | * | |
1594 | * This is a little more complicated than it should be because the various | |
1595 | * headers and the actual data are non-contiguous. | |
1596 | */ | |
f9668a09 | 1597 | __le32 |
0e446be4 CH |
1598 | xlog_cksum( |
1599 | struct xlog *log, | |
1600 | struct xlog_rec_header *rhead, | |
1601 | char *dp, | |
1602 | int size) | |
1603 | { | |
1604 | __uint32_t crc; | |
1605 | ||
1606 | /* first generate the crc for the record header ... */ | |
1607 | crc = xfs_start_cksum((char *)rhead, | |
1608 | sizeof(struct xlog_rec_header), | |
1609 | offsetof(struct xlog_rec_header, h_crc)); | |
1610 | ||
1611 | /* ... then for additional cycle data for v2 logs ... */ | |
1612 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1613 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1614 | int i; | |
1615 | ||
1616 | for (i = 1; i < log->l_iclog_heads; i++) { | |
1617 | crc = crc32c(crc, &xhdr[i].hic_xheader, | |
1618 | sizeof(struct xlog_rec_ext_header)); | |
1619 | } | |
1620 | } | |
1621 | ||
1622 | /* ... and finally for the payload */ | |
1623 | crc = crc32c(crc, dp, size); | |
1624 | ||
1625 | return xfs_end_cksum(crc); | |
1626 | } | |
1627 | ||
873ff550 CH |
1628 | /* |
1629 | * The bdstrat callback function for log bufs. This gives us a central | |
1630 | * place to trap bufs in case we get hit by a log I/O error and need to | |
1631 | * shutdown. Actually, in practice, even when we didn't get a log error, | |
1632 | * we transition the iclogs to IOERROR state *after* flushing all existing | |
1633 | * iclogs to disk. This is because we don't want anymore new transactions to be | |
1634 | * started or completed afterwards. | |
1635 | */ | |
1636 | STATIC int | |
1637 | xlog_bdstrat( | |
1638 | struct xfs_buf *bp) | |
1639 | { | |
adadbeef | 1640 | struct xlog_in_core *iclog = bp->b_fspriv; |
873ff550 | 1641 | |
873ff550 | 1642 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
5a52c2a5 | 1643 | xfs_buf_ioerror(bp, EIO); |
c867cb61 | 1644 | xfs_buf_stale(bp); |
1a1a3e97 | 1645 | xfs_buf_ioend(bp, 0); |
873ff550 CH |
1646 | /* |
1647 | * It would seem logical to return EIO here, but we rely on | |
1648 | * the log state machine to propagate I/O errors instead of | |
1649 | * doing it here. | |
1650 | */ | |
1651 | return 0; | |
1652 | } | |
1653 | ||
873ff550 CH |
1654 | xfs_buf_iorequest(bp); |
1655 | return 0; | |
1656 | } | |
1da177e4 LT |
1657 | |
1658 | /* | |
1659 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1660 | * fashion. Previously, we should have moved the current iclog | |
1661 | * ptr in the log to point to the next available iclog. This allows further | |
1662 | * write to continue while this code syncs out an iclog ready to go. | |
1663 | * Before an in-core log can be written out, the data section must be scanned | |
1664 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1665 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1666 | * cycle count because there in an implicit assumption that drives will | |
1667 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1668 | * we can't have part of a 512 byte block written and part not written. By | |
1669 | * tagging each block, we will know which blocks are valid when recovering | |
1670 | * after an unclean shutdown. | |
1671 | * | |
1672 | * This routine is single threaded on the iclog. No other thread can be in | |
1673 | * this routine with the same iclog. Changing contents of iclog can there- | |
1674 | * fore be done without grabbing the state machine lock. Updating the global | |
1675 | * log will require grabbing the lock though. | |
1676 | * | |
1677 | * The entire log manager uses a logical block numbering scheme. Only | |
1678 | * log_sync (and then only bwrite()) know about the fact that the log may | |
1679 | * not start with block zero on a given device. The log block start offset | |
1680 | * is added immediately before calling bwrite(). | |
1681 | */ | |
1682 | ||
a8272ce0 | 1683 | STATIC int |
9a8d2fdb MT |
1684 | xlog_sync( |
1685 | struct xlog *log, | |
1686 | struct xlog_in_core *iclog) | |
1da177e4 | 1687 | { |
1da177e4 | 1688 | xfs_buf_t *bp; |
b53e675d | 1689 | int i; |
1da177e4 LT |
1690 | uint count; /* byte count of bwrite */ |
1691 | uint count_init; /* initial count before roundup */ | |
1692 | int roundoff; /* roundoff to BB or stripe */ | |
1693 | int split = 0; /* split write into two regions */ | |
1694 | int error; | |
62118709 | 1695 | int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb); |
0e446be4 | 1696 | int size; |
1da177e4 LT |
1697 | |
1698 | XFS_STATS_INC(xs_log_writes); | |
155cc6b7 | 1699 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
1700 | |
1701 | /* Add for LR header */ | |
1702 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
1703 | ||
1704 | /* Round out the log write size */ | |
1705 | if (v2 && log->l_mp->m_sb.sb_logsunit > 1) { | |
1706 | /* we have a v2 stripe unit to use */ | |
1707 | count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init)); | |
1708 | } else { | |
1709 | count = BBTOB(BTOBB(count_init)); | |
1710 | } | |
1711 | roundoff = count - count_init; | |
1712 | ASSERT(roundoff >= 0); | |
1713 | ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && | |
1714 | roundoff < log->l_mp->m_sb.sb_logsunit) | |
1715 | || | |
1716 | (log->l_mp->m_sb.sb_logsunit <= 1 && | |
1717 | roundoff < BBTOB(1))); | |
1718 | ||
1719 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1720 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1721 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1722 | |
1723 | /* put cycle number in every block */ | |
1724 | xlog_pack_data(log, iclog, roundoff); | |
1725 | ||
1726 | /* real byte length */ | |
0e446be4 CH |
1727 | size = iclog->ic_offset; |
1728 | if (v2) | |
1729 | size += roundoff; | |
1730 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 1731 | |
f5faad79 | 1732 | bp = iclog->ic_bp; |
b53e675d | 1733 | XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn))); |
1da177e4 LT |
1734 | |
1735 | XFS_STATS_ADD(xs_log_blocks, BTOBB(count)); | |
1736 | ||
1737 | /* Do we need to split this write into 2 parts? */ | |
1738 | if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) { | |
0e446be4 CH |
1739 | char *dptr; |
1740 | ||
1da177e4 LT |
1741 | split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp))); |
1742 | count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)); | |
0e446be4 CH |
1743 | iclog->ic_bwritecnt = 2; |
1744 | ||
1745 | /* | |
1746 | * Bump the cycle numbers at the start of each block in the | |
1747 | * part of the iclog that ends up in the buffer that gets | |
1748 | * written to the start of the log. | |
1749 | * | |
1750 | * Watch out for the header magic number case, though. | |
1751 | */ | |
1752 | dptr = (char *)&iclog->ic_header + count; | |
1753 | for (i = 0; i < split; i += BBSIZE) { | |
1754 | __uint32_t cycle = be32_to_cpu(*(__be32 *)dptr); | |
1755 | if (++cycle == XLOG_HEADER_MAGIC_NUM) | |
1756 | cycle++; | |
1757 | *(__be32 *)dptr = cpu_to_be32(cycle); | |
1758 | ||
1759 | dptr += BBSIZE; | |
1760 | } | |
1da177e4 LT |
1761 | } else { |
1762 | iclog->ic_bwritecnt = 1; | |
1763 | } | |
0e446be4 CH |
1764 | |
1765 | /* calculcate the checksum */ | |
1766 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
1767 | iclog->ic_datap, size); | |
1768 | ||
aa0e8833 | 1769 | bp->b_io_length = BTOBB(count); |
adadbeef | 1770 | bp->b_fspriv = iclog; |
f5faad79 | 1771 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 | 1772 | XFS_BUF_ASYNC(bp); |
1d5ae5df | 1773 | bp->b_flags |= XBF_SYNCIO; |
651701d7 | 1774 | |
a27a263b | 1775 | if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) { |
e163cbde CH |
1776 | bp->b_flags |= XBF_FUA; |
1777 | ||
a27a263b | 1778 | /* |
e163cbde CH |
1779 | * Flush the data device before flushing the log to make |
1780 | * sure all meta data written back from the AIL actually made | |
1781 | * it to disk before stamping the new log tail LSN into the | |
1782 | * log buffer. For an external log we need to issue the | |
1783 | * flush explicitly, and unfortunately synchronously here; | |
1784 | * for an internal log we can simply use the block layer | |
1785 | * state machine for preflushes. | |
a27a263b CH |
1786 | */ |
1787 | if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp) | |
1788 | xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp); | |
e163cbde CH |
1789 | else |
1790 | bp->b_flags |= XBF_FLUSH; | |
a27a263b | 1791 | } |
1da177e4 LT |
1792 | |
1793 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1794 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1795 | ||
003fd6c8 | 1796 | xlog_verify_iclog(log, iclog, count, true); |
1da177e4 LT |
1797 | |
1798 | /* account for log which doesn't start at block #0 */ | |
1799 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); | |
1800 | /* | |
1801 | * Don't call xfs_bwrite here. We do log-syncs even when the filesystem | |
1802 | * is shutting down. | |
1803 | */ | |
1804 | XFS_BUF_WRITE(bp); | |
1805 | ||
901796af CH |
1806 | error = xlog_bdstrat(bp); |
1807 | if (error) { | |
1808 | xfs_buf_ioerror_alert(bp, "xlog_sync"); | |
014c2544 | 1809 | return error; |
1da177e4 LT |
1810 | } |
1811 | if (split) { | |
f5faad79 | 1812 | bp = iclog->ic_log->l_xbuf; |
1da177e4 | 1813 | XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */ |
02fe03d9 CS |
1814 | xfs_buf_associate_memory(bp, |
1815 | (char *)&iclog->ic_header + count, split); | |
adadbeef | 1816 | bp->b_fspriv = iclog; |
f5faad79 | 1817 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 | 1818 | XFS_BUF_ASYNC(bp); |
1d5ae5df | 1819 | bp->b_flags |= XBF_SYNCIO; |
f538d4da | 1820 | if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) |
e163cbde | 1821 | bp->b_flags |= XBF_FUA; |
1da177e4 LT |
1822 | |
1823 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1824 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1825 | ||
c41564b5 | 1826 | /* account for internal log which doesn't start at block #0 */ |
1da177e4 LT |
1827 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); |
1828 | XFS_BUF_WRITE(bp); | |
901796af CH |
1829 | error = xlog_bdstrat(bp); |
1830 | if (error) { | |
1831 | xfs_buf_ioerror_alert(bp, "xlog_sync (split)"); | |
014c2544 | 1832 | return error; |
1da177e4 LT |
1833 | } |
1834 | } | |
014c2544 | 1835 | return 0; |
1da177e4 LT |
1836 | } /* xlog_sync */ |
1837 | ||
1da177e4 | 1838 | /* |
c41564b5 | 1839 | * Deallocate a log structure |
1da177e4 | 1840 | */ |
a8272ce0 | 1841 | STATIC void |
9a8d2fdb MT |
1842 | xlog_dealloc_log( |
1843 | struct xlog *log) | |
1da177e4 LT |
1844 | { |
1845 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1846 | int i; |
1847 | ||
71e330b5 DC |
1848 | xlog_cil_destroy(log); |
1849 | ||
44396476 DC |
1850 | /* |
1851 | * always need to ensure that the extra buffer does not point to memory | |
1852 | * owned by another log buffer before we free it. | |
1853 | */ | |
e70b73f8 | 1854 | xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size)); |
44396476 DC |
1855 | xfs_buf_free(log->l_xbuf); |
1856 | ||
1da177e4 LT |
1857 | iclog = log->l_iclog; |
1858 | for (i=0; i<log->l_iclog_bufs; i++) { | |
1da177e4 | 1859 | xfs_buf_free(iclog->ic_bp); |
1da177e4 | 1860 | next_iclog = iclog->ic_next; |
f0e2d93c | 1861 | kmem_free(iclog); |
1da177e4 LT |
1862 | iclog = next_iclog; |
1863 | } | |
1da177e4 | 1864 | spinlock_destroy(&log->l_icloglock); |
1da177e4 | 1865 | |
1da177e4 | 1866 | log->l_mp->m_log = NULL; |
f0e2d93c | 1867 | kmem_free(log); |
c41564b5 | 1868 | } /* xlog_dealloc_log */ |
1da177e4 LT |
1869 | |
1870 | /* | |
1871 | * Update counters atomically now that memcpy is done. | |
1872 | */ | |
1873 | /* ARGSUSED */ | |
1874 | static inline void | |
9a8d2fdb MT |
1875 | xlog_state_finish_copy( |
1876 | struct xlog *log, | |
1877 | struct xlog_in_core *iclog, | |
1878 | int record_cnt, | |
1879 | int copy_bytes) | |
1da177e4 | 1880 | { |
b22cd72c | 1881 | spin_lock(&log->l_icloglock); |
1da177e4 | 1882 | |
413d57c9 | 1883 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 LT |
1884 | iclog->ic_offset += copy_bytes; |
1885 | ||
b22cd72c | 1886 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
1887 | } /* xlog_state_finish_copy */ |
1888 | ||
1889 | ||
1890 | ||
1891 | ||
7e9c6396 TS |
1892 | /* |
1893 | * print out info relating to regions written which consume | |
1894 | * the reservation | |
1895 | */ | |
71e330b5 DC |
1896 | void |
1897 | xlog_print_tic_res( | |
1898 | struct xfs_mount *mp, | |
1899 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
1900 | { |
1901 | uint i; | |
1902 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
1903 | ||
1904 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
1905 | static char *res_type_str[XLOG_REG_TYPE_MAX] = { | |
1906 | "bformat", | |
1907 | "bchunk", | |
1908 | "efi_format", | |
1909 | "efd_format", | |
1910 | "iformat", | |
1911 | "icore", | |
1912 | "iext", | |
1913 | "ibroot", | |
1914 | "ilocal", | |
1915 | "iattr_ext", | |
1916 | "iattr_broot", | |
1917 | "iattr_local", | |
1918 | "qformat", | |
1919 | "dquot", | |
1920 | "quotaoff", | |
1921 | "LR header", | |
1922 | "unmount", | |
1923 | "commit", | |
1924 | "trans header" | |
1925 | }; | |
1926 | static char *trans_type_str[XFS_TRANS_TYPE_MAX] = { | |
1927 | "SETATTR_NOT_SIZE", | |
1928 | "SETATTR_SIZE", | |
1929 | "INACTIVE", | |
1930 | "CREATE", | |
1931 | "CREATE_TRUNC", | |
1932 | "TRUNCATE_FILE", | |
1933 | "REMOVE", | |
1934 | "LINK", | |
1935 | "RENAME", | |
1936 | "MKDIR", | |
1937 | "RMDIR", | |
1938 | "SYMLINK", | |
1939 | "SET_DMATTRS", | |
1940 | "GROWFS", | |
1941 | "STRAT_WRITE", | |
1942 | "DIOSTRAT", | |
1943 | "WRITE_SYNC", | |
1944 | "WRITEID", | |
1945 | "ADDAFORK", | |
1946 | "ATTRINVAL", | |
1947 | "ATRUNCATE", | |
1948 | "ATTR_SET", | |
1949 | "ATTR_RM", | |
1950 | "ATTR_FLAG", | |
1951 | "CLEAR_AGI_BUCKET", | |
1952 | "QM_SBCHANGE", | |
1953 | "DUMMY1", | |
1954 | "DUMMY2", | |
1955 | "QM_QUOTAOFF", | |
1956 | "QM_DQALLOC", | |
1957 | "QM_SETQLIM", | |
1958 | "QM_DQCLUSTER", | |
1959 | "QM_QINOCREATE", | |
1960 | "QM_QUOTAOFF_END", | |
1961 | "SB_UNIT", | |
1962 | "FSYNC_TS", | |
1963 | "GROWFSRT_ALLOC", | |
1964 | "GROWFSRT_ZERO", | |
1965 | "GROWFSRT_FREE", | |
1966 | "SWAPEXT" | |
1967 | }; | |
1968 | ||
a0fa2b67 | 1969 | xfs_warn(mp, |
93b8a585 | 1970 | "xlog_write: reservation summary:\n" |
a0fa2b67 DC |
1971 | " trans type = %s (%u)\n" |
1972 | " unit res = %d bytes\n" | |
1973 | " current res = %d bytes\n" | |
1974 | " total reg = %u bytes (o/flow = %u bytes)\n" | |
1975 | " ophdrs = %u (ophdr space = %u bytes)\n" | |
1976 | " ophdr + reg = %u bytes\n" | |
1977 | " num regions = %u\n", | |
1978 | ((ticket->t_trans_type <= 0 || | |
1979 | ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ? | |
1980 | "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]), | |
1981 | ticket->t_trans_type, | |
1982 | ticket->t_unit_res, | |
1983 | ticket->t_curr_res, | |
1984 | ticket->t_res_arr_sum, ticket->t_res_o_flow, | |
1985 | ticket->t_res_num_ophdrs, ophdr_spc, | |
1986 | ticket->t_res_arr_sum + | |
1987 | ticket->t_res_o_flow + ophdr_spc, | |
1988 | ticket->t_res_num); | |
7e9c6396 TS |
1989 | |
1990 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 1991 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 1992 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 TS |
1993 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
1994 | "bad-rtype" : res_type_str[r_type-1]), | |
1995 | ticket->t_res_arr[i].r_len); | |
1996 | } | |
169a7b07 | 1997 | |
a0fa2b67 | 1998 | xfs_alert_tag(mp, XFS_PTAG_LOGRES, |
93b8a585 | 1999 | "xlog_write: reservation ran out. Need to up reservation"); |
297aa637 | 2000 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
7e9c6396 | 2001 | } |
7e9c6396 | 2002 | |
b5203cd0 DC |
2003 | /* |
2004 | * Calculate the potential space needed by the log vector. Each region gets | |
2005 | * its own xlog_op_header_t and may need to be double word aligned. | |
2006 | */ | |
2007 | static int | |
2008 | xlog_write_calc_vec_length( | |
2009 | struct xlog_ticket *ticket, | |
55b66332 | 2010 | struct xfs_log_vec *log_vector) |
b5203cd0 | 2011 | { |
55b66332 | 2012 | struct xfs_log_vec *lv; |
b5203cd0 DC |
2013 | int headers = 0; |
2014 | int len = 0; | |
2015 | int i; | |
2016 | ||
2017 | /* acct for start rec of xact */ | |
2018 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2019 | headers++; | |
2020 | ||
55b66332 | 2021 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2022 | /* we don't write ordered log vectors */ |
2023 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2024 | continue; | |
2025 | ||
55b66332 DC |
2026 | headers += lv->lv_niovecs; |
2027 | ||
2028 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2029 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2030 | |
55b66332 DC |
2031 | len += vecp->i_len; |
2032 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2033 | } | |
b5203cd0 DC |
2034 | } |
2035 | ||
2036 | ticket->t_res_num_ophdrs += headers; | |
2037 | len += headers * sizeof(struct xlog_op_header); | |
2038 | ||
2039 | return len; | |
2040 | } | |
2041 | ||
2042 | /* | |
2043 | * If first write for transaction, insert start record We can't be trying to | |
2044 | * commit if we are inited. We can't have any "partial_copy" if we are inited. | |
2045 | */ | |
2046 | static int | |
2047 | xlog_write_start_rec( | |
e6b1f273 | 2048 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2049 | struct xlog_ticket *ticket) |
2050 | { | |
b5203cd0 DC |
2051 | if (!(ticket->t_flags & XLOG_TIC_INITED)) |
2052 | return 0; | |
2053 | ||
2054 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); | |
2055 | ophdr->oh_clientid = ticket->t_clientid; | |
2056 | ophdr->oh_len = 0; | |
2057 | ophdr->oh_flags = XLOG_START_TRANS; | |
2058 | ophdr->oh_res2 = 0; | |
2059 | ||
2060 | ticket->t_flags &= ~XLOG_TIC_INITED; | |
2061 | ||
2062 | return sizeof(struct xlog_op_header); | |
2063 | } | |
2064 | ||
2065 | static xlog_op_header_t * | |
2066 | xlog_write_setup_ophdr( | |
ad223e60 | 2067 | struct xlog *log, |
e6b1f273 | 2068 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2069 | struct xlog_ticket *ticket, |
2070 | uint flags) | |
2071 | { | |
b5203cd0 DC |
2072 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2073 | ophdr->oh_clientid = ticket->t_clientid; | |
2074 | ophdr->oh_res2 = 0; | |
2075 | ||
2076 | /* are we copying a commit or unmount record? */ | |
2077 | ophdr->oh_flags = flags; | |
2078 | ||
2079 | /* | |
2080 | * We've seen logs corrupted with bad transaction client ids. This | |
2081 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2082 | * and shut down the filesystem. | |
2083 | */ | |
2084 | switch (ophdr->oh_clientid) { | |
2085 | case XFS_TRANSACTION: | |
2086 | case XFS_VOLUME: | |
2087 | case XFS_LOG: | |
2088 | break; | |
2089 | default: | |
a0fa2b67 | 2090 | xfs_warn(log->l_mp, |
b5203cd0 DC |
2091 | "Bad XFS transaction clientid 0x%x in ticket 0x%p", |
2092 | ophdr->oh_clientid, ticket); | |
2093 | return NULL; | |
2094 | } | |
2095 | ||
2096 | return ophdr; | |
2097 | } | |
2098 | ||
2099 | /* | |
2100 | * Set up the parameters of the region copy into the log. This has | |
2101 | * to handle region write split across multiple log buffers - this | |
2102 | * state is kept external to this function so that this code can | |
ac0e300f | 2103 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2104 | */ |
2105 | static int | |
2106 | xlog_write_setup_copy( | |
2107 | struct xlog_ticket *ticket, | |
2108 | struct xlog_op_header *ophdr, | |
2109 | int space_available, | |
2110 | int space_required, | |
2111 | int *copy_off, | |
2112 | int *copy_len, | |
2113 | int *last_was_partial_copy, | |
2114 | int *bytes_consumed) | |
2115 | { | |
2116 | int still_to_copy; | |
2117 | ||
2118 | still_to_copy = space_required - *bytes_consumed; | |
2119 | *copy_off = *bytes_consumed; | |
2120 | ||
2121 | if (still_to_copy <= space_available) { | |
2122 | /* write of region completes here */ | |
2123 | *copy_len = still_to_copy; | |
2124 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2125 | if (*last_was_partial_copy) | |
2126 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2127 | *last_was_partial_copy = 0; | |
2128 | *bytes_consumed = 0; | |
2129 | return 0; | |
2130 | } | |
2131 | ||
2132 | /* partial write of region, needs extra log op header reservation */ | |
2133 | *copy_len = space_available; | |
2134 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2135 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2136 | if (*last_was_partial_copy) | |
2137 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2138 | *bytes_consumed += *copy_len; | |
2139 | (*last_was_partial_copy)++; | |
2140 | ||
2141 | /* account for new log op header */ | |
2142 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2143 | ticket->t_res_num_ophdrs++; | |
2144 | ||
2145 | return sizeof(struct xlog_op_header); | |
2146 | } | |
2147 | ||
2148 | static int | |
2149 | xlog_write_copy_finish( | |
ad223e60 | 2150 | struct xlog *log, |
b5203cd0 DC |
2151 | struct xlog_in_core *iclog, |
2152 | uint flags, | |
2153 | int *record_cnt, | |
2154 | int *data_cnt, | |
2155 | int *partial_copy, | |
2156 | int *partial_copy_len, | |
2157 | int log_offset, | |
2158 | struct xlog_in_core **commit_iclog) | |
2159 | { | |
2160 | if (*partial_copy) { | |
2161 | /* | |
2162 | * This iclog has already been marked WANT_SYNC by | |
2163 | * xlog_state_get_iclog_space. | |
2164 | */ | |
2165 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2166 | *record_cnt = 0; | |
2167 | *data_cnt = 0; | |
2168 | return xlog_state_release_iclog(log, iclog); | |
2169 | } | |
2170 | ||
2171 | *partial_copy = 0; | |
2172 | *partial_copy_len = 0; | |
2173 | ||
2174 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2175 | /* no more space in this iclog - push it. */ | |
2176 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2177 | *record_cnt = 0; | |
2178 | *data_cnt = 0; | |
2179 | ||
2180 | spin_lock(&log->l_icloglock); | |
2181 | xlog_state_want_sync(log, iclog); | |
2182 | spin_unlock(&log->l_icloglock); | |
2183 | ||
2184 | if (!commit_iclog) | |
2185 | return xlog_state_release_iclog(log, iclog); | |
2186 | ASSERT(flags & XLOG_COMMIT_TRANS); | |
2187 | *commit_iclog = iclog; | |
2188 | } | |
2189 | ||
2190 | return 0; | |
2191 | } | |
2192 | ||
1da177e4 LT |
2193 | /* |
2194 | * Write some region out to in-core log | |
2195 | * | |
2196 | * This will be called when writing externally provided regions or when | |
2197 | * writing out a commit record for a given transaction. | |
2198 | * | |
2199 | * General algorithm: | |
2200 | * 1. Find total length of this write. This may include adding to the | |
2201 | * lengths passed in. | |
2202 | * 2. Check whether we violate the tickets reservation. | |
2203 | * 3. While writing to this iclog | |
2204 | * A. Reserve as much space in this iclog as can get | |
2205 | * B. If this is first write, save away start lsn | |
2206 | * C. While writing this region: | |
2207 | * 1. If first write of transaction, write start record | |
2208 | * 2. Write log operation header (header per region) | |
2209 | * 3. Find out if we can fit entire region into this iclog | |
2210 | * 4. Potentially, verify destination memcpy ptr | |
2211 | * 5. Memcpy (partial) region | |
2212 | * 6. If partial copy, release iclog; otherwise, continue | |
2213 | * copying more regions into current iclog | |
2214 | * 4. Mark want sync bit (in simulation mode) | |
2215 | * 5. Release iclog for potential flush to on-disk log. | |
2216 | * | |
2217 | * ERRORS: | |
2218 | * 1. Panic if reservation is overrun. This should never happen since | |
2219 | * reservation amounts are generated internal to the filesystem. | |
2220 | * NOTES: | |
2221 | * 1. Tickets are single threaded data structures. | |
2222 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2223 | * syncing routine. When a single log_write region needs to span | |
2224 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2225 | * on all log operation writes which don't contain the end of the | |
2226 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2227 | * operation which contains the end of the continued log_write region. | |
2228 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2229 | * we don't really know exactly how much space will be used. As a result, | |
2230 | * we don't update ic_offset until the end when we know exactly how many | |
2231 | * bytes have been written out. | |
2232 | */ | |
71e330b5 | 2233 | int |
35a8a72f | 2234 | xlog_write( |
ad223e60 | 2235 | struct xlog *log, |
55b66332 | 2236 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2237 | struct xlog_ticket *ticket, |
2238 | xfs_lsn_t *start_lsn, | |
2239 | struct xlog_in_core **commit_iclog, | |
2240 | uint flags) | |
1da177e4 | 2241 | { |
99428ad0 | 2242 | struct xlog_in_core *iclog = NULL; |
55b66332 DC |
2243 | struct xfs_log_iovec *vecp; |
2244 | struct xfs_log_vec *lv; | |
99428ad0 CH |
2245 | int len; |
2246 | int index; | |
2247 | int partial_copy = 0; | |
2248 | int partial_copy_len = 0; | |
2249 | int contwr = 0; | |
2250 | int record_cnt = 0; | |
2251 | int data_cnt = 0; | |
2252 | int error; | |
2253 | ||
2254 | *start_lsn = 0; | |
2255 | ||
55b66332 | 2256 | len = xlog_write_calc_vec_length(ticket, log_vector); |
71e330b5 | 2257 | |
93b8a585 CH |
2258 | /* |
2259 | * Region headers and bytes are already accounted for. | |
2260 | * We only need to take into account start records and | |
2261 | * split regions in this function. | |
2262 | */ | |
2263 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2264 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
2265 | ||
2266 | /* | |
2267 | * Commit record headers need to be accounted for. These | |
2268 | * come in as separate writes so are easy to detect. | |
2269 | */ | |
2270 | if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS)) | |
2271 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
71e330b5 DC |
2272 | |
2273 | if (ticket->t_curr_res < 0) | |
55b66332 | 2274 | xlog_print_tic_res(log->l_mp, ticket); |
1da177e4 | 2275 | |
55b66332 DC |
2276 | index = 0; |
2277 | lv = log_vector; | |
2278 | vecp = lv->lv_iovecp; | |
fd63875c | 2279 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2280 | void *ptr; |
99428ad0 | 2281 | int log_offset; |
1da177e4 | 2282 | |
99428ad0 CH |
2283 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2284 | &contwr, &log_offset); | |
2285 | if (error) | |
2286 | return error; | |
1da177e4 | 2287 | |
99428ad0 | 2288 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2289 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2290 | |
99428ad0 CH |
2291 | /* start_lsn is the first lsn written to. That's all we need. */ |
2292 | if (!*start_lsn) | |
2293 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
b5203cd0 | 2294 | |
99428ad0 CH |
2295 | /* |
2296 | * This loop writes out as many regions as can fit in the amount | |
2297 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2298 | */ | |
fd63875c DC |
2299 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2300 | struct xfs_log_iovec *reg; | |
99428ad0 CH |
2301 | struct xlog_op_header *ophdr; |
2302 | int start_rec_copy; | |
2303 | int copy_len; | |
2304 | int copy_off; | |
fd63875c DC |
2305 | bool ordered = false; |
2306 | ||
2307 | /* ordered log vectors have no regions to write */ | |
2308 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2309 | ASSERT(lv->lv_niovecs == 0); | |
2310 | ordered = true; | |
2311 | goto next_lv; | |
2312 | } | |
99428ad0 | 2313 | |
fd63875c | 2314 | reg = &vecp[index]; |
55b66332 | 2315 | ASSERT(reg->i_len % sizeof(__int32_t) == 0); |
e6b1f273 | 2316 | ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0); |
99428ad0 CH |
2317 | |
2318 | start_rec_copy = xlog_write_start_rec(ptr, ticket); | |
2319 | if (start_rec_copy) { | |
2320 | record_cnt++; | |
e6b1f273 | 2321 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2322 | start_rec_copy); |
2323 | } | |
b5203cd0 | 2324 | |
99428ad0 CH |
2325 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags); |
2326 | if (!ophdr) | |
2327 | return XFS_ERROR(EIO); | |
2328 | ||
e6b1f273 | 2329 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2330 | sizeof(struct xlog_op_header)); |
2331 | ||
2332 | len += xlog_write_setup_copy(ticket, ophdr, | |
2333 | iclog->ic_size-log_offset, | |
55b66332 | 2334 | reg->i_len, |
99428ad0 CH |
2335 | ©_off, ©_len, |
2336 | &partial_copy, | |
2337 | &partial_copy_len); | |
2338 | xlog_verify_dest_ptr(log, ptr); | |
2339 | ||
2340 | /* copy region */ | |
2341 | ASSERT(copy_len >= 0); | |
e6b1f273 CH |
2342 | memcpy(ptr, reg->i_addr + copy_off, copy_len); |
2343 | xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len); | |
99428ad0 CH |
2344 | |
2345 | copy_len += start_rec_copy + sizeof(xlog_op_header_t); | |
2346 | record_cnt++; | |
2347 | data_cnt += contwr ? copy_len : 0; | |
2348 | ||
2349 | error = xlog_write_copy_finish(log, iclog, flags, | |
2350 | &record_cnt, &data_cnt, | |
2351 | &partial_copy, | |
2352 | &partial_copy_len, | |
2353 | log_offset, | |
2354 | commit_iclog); | |
2355 | if (error) | |
2356 | return error; | |
2357 | ||
2358 | /* | |
2359 | * if we had a partial copy, we need to get more iclog | |
2360 | * space but we don't want to increment the region | |
2361 | * index because there is still more is this region to | |
2362 | * write. | |
2363 | * | |
2364 | * If we completed writing this region, and we flushed | |
2365 | * the iclog (indicated by resetting of the record | |
2366 | * count), then we also need to get more log space. If | |
2367 | * this was the last record, though, we are done and | |
2368 | * can just return. | |
2369 | */ | |
2370 | if (partial_copy) | |
2371 | break; | |
2372 | ||
55b66332 | 2373 | if (++index == lv->lv_niovecs) { |
fd63875c | 2374 | next_lv: |
55b66332 DC |
2375 | lv = lv->lv_next; |
2376 | index = 0; | |
2377 | if (lv) | |
2378 | vecp = lv->lv_iovecp; | |
2379 | } | |
fd63875c | 2380 | if (record_cnt == 0 && ordered == false) { |
55b66332 | 2381 | if (!lv) |
99428ad0 CH |
2382 | return 0; |
2383 | break; | |
2384 | } | |
2385 | } | |
2386 | } | |
2387 | ||
2388 | ASSERT(len == 0); | |
2389 | ||
2390 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); | |
2391 | if (!commit_iclog) | |
2392 | return xlog_state_release_iclog(log, iclog); | |
1da177e4 | 2393 | |
1da177e4 LT |
2394 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2395 | *commit_iclog = iclog; | |
2396 | return 0; | |
99428ad0 | 2397 | } |
1da177e4 LT |
2398 | |
2399 | ||
2400 | /***************************************************************************** | |
2401 | * | |
2402 | * State Machine functions | |
2403 | * | |
2404 | ***************************************************************************** | |
2405 | */ | |
2406 | ||
2407 | /* Clean iclogs starting from the head. This ordering must be | |
2408 | * maintained, so an iclog doesn't become ACTIVE beyond one that | |
2409 | * is SYNCING. This is also required to maintain the notion that we use | |
12017faf | 2410 | * a ordered wait queue to hold off would be writers to the log when every |
1da177e4 LT |
2411 | * iclog is trying to sync to disk. |
2412 | * | |
2413 | * State Change: DIRTY -> ACTIVE | |
2414 | */ | |
ba0f32d4 | 2415 | STATIC void |
9a8d2fdb MT |
2416 | xlog_state_clean_log( |
2417 | struct xlog *log) | |
1da177e4 LT |
2418 | { |
2419 | xlog_in_core_t *iclog; | |
2420 | int changed = 0; | |
2421 | ||
2422 | iclog = log->l_iclog; | |
2423 | do { | |
2424 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
2425 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2426 | iclog->ic_offset = 0; | |
114d23aa | 2427 | ASSERT(iclog->ic_callback == NULL); |
1da177e4 LT |
2428 | /* |
2429 | * If the number of ops in this iclog indicate it just | |
2430 | * contains the dummy transaction, we can | |
2431 | * change state into IDLE (the second time around). | |
2432 | * Otherwise we should change the state into | |
2433 | * NEED a dummy. | |
2434 | * We don't need to cover the dummy. | |
2435 | */ | |
2436 | if (!changed && | |
b53e675d CH |
2437 | (be32_to_cpu(iclog->ic_header.h_num_logops) == |
2438 | XLOG_COVER_OPS)) { | |
1da177e4 LT |
2439 | changed = 1; |
2440 | } else { | |
2441 | /* | |
2442 | * We have two dirty iclogs so start over | |
2443 | * This could also be num of ops indicates | |
2444 | * this is not the dummy going out. | |
2445 | */ | |
2446 | changed = 2; | |
2447 | } | |
2448 | iclog->ic_header.h_num_logops = 0; | |
2449 | memset(iclog->ic_header.h_cycle_data, 0, | |
2450 | sizeof(iclog->ic_header.h_cycle_data)); | |
2451 | iclog->ic_header.h_lsn = 0; | |
2452 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) | |
2453 | /* do nothing */; | |
2454 | else | |
2455 | break; /* stop cleaning */ | |
2456 | iclog = iclog->ic_next; | |
2457 | } while (iclog != log->l_iclog); | |
2458 | ||
2459 | /* log is locked when we are called */ | |
2460 | /* | |
2461 | * Change state for the dummy log recording. | |
2462 | * We usually go to NEED. But we go to NEED2 if the changed indicates | |
2463 | * we are done writing the dummy record. | |
2464 | * If we are done with the second dummy recored (DONE2), then | |
2465 | * we go to IDLE. | |
2466 | */ | |
2467 | if (changed) { | |
2468 | switch (log->l_covered_state) { | |
2469 | case XLOG_STATE_COVER_IDLE: | |
2470 | case XLOG_STATE_COVER_NEED: | |
2471 | case XLOG_STATE_COVER_NEED2: | |
2472 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2473 | break; | |
2474 | ||
2475 | case XLOG_STATE_COVER_DONE: | |
2476 | if (changed == 1) | |
2477 | log->l_covered_state = XLOG_STATE_COVER_NEED2; | |
2478 | else | |
2479 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2480 | break; | |
2481 | ||
2482 | case XLOG_STATE_COVER_DONE2: | |
2483 | if (changed == 1) | |
2484 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
2485 | else | |
2486 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2487 | break; | |
2488 | ||
2489 | default: | |
2490 | ASSERT(0); | |
2491 | } | |
2492 | } | |
2493 | } /* xlog_state_clean_log */ | |
2494 | ||
2495 | STATIC xfs_lsn_t | |
2496 | xlog_get_lowest_lsn( | |
9a8d2fdb | 2497 | struct xlog *log) |
1da177e4 LT |
2498 | { |
2499 | xlog_in_core_t *lsn_log; | |
2500 | xfs_lsn_t lowest_lsn, lsn; | |
2501 | ||
2502 | lsn_log = log->l_iclog; | |
2503 | lowest_lsn = 0; | |
2504 | do { | |
2505 | if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) { | |
b53e675d | 2506 | lsn = be64_to_cpu(lsn_log->ic_header.h_lsn); |
1da177e4 LT |
2507 | if ((lsn && !lowest_lsn) || |
2508 | (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) { | |
2509 | lowest_lsn = lsn; | |
2510 | } | |
2511 | } | |
2512 | lsn_log = lsn_log->ic_next; | |
2513 | } while (lsn_log != log->l_iclog); | |
014c2544 | 2514 | return lowest_lsn; |
1da177e4 LT |
2515 | } |
2516 | ||
2517 | ||
2518 | STATIC void | |
2519 | xlog_state_do_callback( | |
9a8d2fdb MT |
2520 | struct xlog *log, |
2521 | int aborted, | |
2522 | struct xlog_in_core *ciclog) | |
1da177e4 LT |
2523 | { |
2524 | xlog_in_core_t *iclog; | |
2525 | xlog_in_core_t *first_iclog; /* used to know when we've | |
2526 | * processed all iclogs once */ | |
2527 | xfs_log_callback_t *cb, *cb_next; | |
2528 | int flushcnt = 0; | |
2529 | xfs_lsn_t lowest_lsn; | |
2530 | int ioerrors; /* counter: iclogs with errors */ | |
2531 | int loopdidcallbacks; /* flag: inner loop did callbacks*/ | |
2532 | int funcdidcallbacks; /* flag: function did callbacks */ | |
2533 | int repeats; /* for issuing console warnings if | |
2534 | * looping too many times */ | |
d748c623 | 2535 | int wake = 0; |
1da177e4 | 2536 | |
b22cd72c | 2537 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2538 | first_iclog = iclog = log->l_iclog; |
2539 | ioerrors = 0; | |
2540 | funcdidcallbacks = 0; | |
2541 | repeats = 0; | |
2542 | ||
2543 | do { | |
2544 | /* | |
2545 | * Scan all iclogs starting with the one pointed to by the | |
2546 | * log. Reset this starting point each time the log is | |
2547 | * unlocked (during callbacks). | |
2548 | * | |
2549 | * Keep looping through iclogs until one full pass is made | |
2550 | * without running any callbacks. | |
2551 | */ | |
2552 | first_iclog = log->l_iclog; | |
2553 | iclog = log->l_iclog; | |
2554 | loopdidcallbacks = 0; | |
2555 | repeats++; | |
2556 | ||
2557 | do { | |
2558 | ||
2559 | /* skip all iclogs in the ACTIVE & DIRTY states */ | |
2560 | if (iclog->ic_state & | |
2561 | (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) { | |
2562 | iclog = iclog->ic_next; | |
2563 | continue; | |
2564 | } | |
2565 | ||
2566 | /* | |
2567 | * Between marking a filesystem SHUTDOWN and stopping | |
2568 | * the log, we do flush all iclogs to disk (if there | |
2569 | * wasn't a log I/O error). So, we do want things to | |
2570 | * go smoothly in case of just a SHUTDOWN w/o a | |
2571 | * LOG_IO_ERROR. | |
2572 | */ | |
2573 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
2574 | /* | |
2575 | * Can only perform callbacks in order. Since | |
2576 | * this iclog is not in the DONE_SYNC/ | |
2577 | * DO_CALLBACK state, we skip the rest and | |
2578 | * just try to clean up. If we set our iclog | |
2579 | * to DO_CALLBACK, we will not process it when | |
2580 | * we retry since a previous iclog is in the | |
2581 | * CALLBACK and the state cannot change since | |
b22cd72c | 2582 | * we are holding the l_icloglock. |
1da177e4 LT |
2583 | */ |
2584 | if (!(iclog->ic_state & | |
2585 | (XLOG_STATE_DONE_SYNC | | |
2586 | XLOG_STATE_DO_CALLBACK))) { | |
2587 | if (ciclog && (ciclog->ic_state == | |
2588 | XLOG_STATE_DONE_SYNC)) { | |
2589 | ciclog->ic_state = XLOG_STATE_DO_CALLBACK; | |
2590 | } | |
2591 | break; | |
2592 | } | |
2593 | /* | |
2594 | * We now have an iclog that is in either the | |
2595 | * DO_CALLBACK or DONE_SYNC states. The other | |
2596 | * states (WANT_SYNC, SYNCING, or CALLBACK were | |
2597 | * caught by the above if and are going to | |
2598 | * clean (i.e. we aren't doing their callbacks) | |
2599 | * see the above if. | |
2600 | */ | |
2601 | ||
2602 | /* | |
2603 | * We will do one more check here to see if we | |
2604 | * have chased our tail around. | |
2605 | */ | |
2606 | ||
2607 | lowest_lsn = xlog_get_lowest_lsn(log); | |
b53e675d CH |
2608 | if (lowest_lsn && |
2609 | XFS_LSN_CMP(lowest_lsn, | |
84f3c683 | 2610 | be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { |
1da177e4 LT |
2611 | iclog = iclog->ic_next; |
2612 | continue; /* Leave this iclog for | |
2613 | * another thread */ | |
2614 | } | |
2615 | ||
2616 | iclog->ic_state = XLOG_STATE_CALLBACK; | |
2617 | ||
1da177e4 | 2618 | |
84f3c683 | 2619 | /* |
d35e88fa DC |
2620 | * Completion of a iclog IO does not imply that |
2621 | * a transaction has completed, as transactions | |
2622 | * can be large enough to span many iclogs. We | |
2623 | * cannot change the tail of the log half way | |
2624 | * through a transaction as this may be the only | |
2625 | * transaction in the log and moving th etail to | |
2626 | * point to the middle of it will prevent | |
2627 | * recovery from finding the start of the | |
2628 | * transaction. Hence we should only update the | |
2629 | * last_sync_lsn if this iclog contains | |
2630 | * transaction completion callbacks on it. | |
2631 | * | |
2632 | * We have to do this before we drop the | |
84f3c683 DC |
2633 | * icloglock to ensure we are the only one that |
2634 | * can update it. | |
1da177e4 | 2635 | */ |
84f3c683 DC |
2636 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), |
2637 | be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); | |
d35e88fa DC |
2638 | if (iclog->ic_callback) |
2639 | atomic64_set(&log->l_last_sync_lsn, | |
2640 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1da177e4 | 2641 | |
84f3c683 | 2642 | } else |
1da177e4 | 2643 | ioerrors++; |
84f3c683 DC |
2644 | |
2645 | spin_unlock(&log->l_icloglock); | |
1da177e4 | 2646 | |
114d23aa DC |
2647 | /* |
2648 | * Keep processing entries in the callback list until | |
2649 | * we come around and it is empty. We need to | |
2650 | * atomically see that the list is empty and change the | |
2651 | * state to DIRTY so that we don't miss any more | |
2652 | * callbacks being added. | |
2653 | */ | |
2654 | spin_lock(&iclog->ic_callback_lock); | |
2655 | cb = iclog->ic_callback; | |
4b80916b | 2656 | while (cb) { |
1da177e4 LT |
2657 | iclog->ic_callback_tail = &(iclog->ic_callback); |
2658 | iclog->ic_callback = NULL; | |
114d23aa | 2659 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2660 | |
2661 | /* perform callbacks in the order given */ | |
4b80916b | 2662 | for (; cb; cb = cb_next) { |
1da177e4 LT |
2663 | cb_next = cb->cb_next; |
2664 | cb->cb_func(cb->cb_arg, aborted); | |
2665 | } | |
114d23aa | 2666 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
2667 | cb = iclog->ic_callback; |
2668 | } | |
2669 | ||
2670 | loopdidcallbacks++; | |
2671 | funcdidcallbacks++; | |
2672 | ||
114d23aa | 2673 | spin_lock(&log->l_icloglock); |
4b80916b | 2674 | ASSERT(iclog->ic_callback == NULL); |
114d23aa | 2675 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2676 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) |
2677 | iclog->ic_state = XLOG_STATE_DIRTY; | |
2678 | ||
2679 | /* | |
2680 | * Transition from DIRTY to ACTIVE if applicable. | |
2681 | * NOP if STATE_IOERROR. | |
2682 | */ | |
2683 | xlog_state_clean_log(log); | |
2684 | ||
2685 | /* wake up threads waiting in xfs_log_force() */ | |
eb40a875 | 2686 | wake_up_all(&iclog->ic_force_wait); |
1da177e4 LT |
2687 | |
2688 | iclog = iclog->ic_next; | |
2689 | } while (first_iclog != iclog); | |
a3c6685e NS |
2690 | |
2691 | if (repeats > 5000) { | |
2692 | flushcnt += repeats; | |
2693 | repeats = 0; | |
a0fa2b67 | 2694 | xfs_warn(log->l_mp, |
a3c6685e | 2695 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2696 | __func__, flushcnt); |
1da177e4 LT |
2697 | } |
2698 | } while (!ioerrors && loopdidcallbacks); | |
2699 | ||
2700 | /* | |
2701 | * make one last gasp attempt to see if iclogs are being left in | |
2702 | * limbo.. | |
2703 | */ | |
2704 | #ifdef DEBUG | |
2705 | if (funcdidcallbacks) { | |
2706 | first_iclog = iclog = log->l_iclog; | |
2707 | do { | |
2708 | ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK); | |
2709 | /* | |
2710 | * Terminate the loop if iclogs are found in states | |
2711 | * which will cause other threads to clean up iclogs. | |
2712 | * | |
2713 | * SYNCING - i/o completion will go through logs | |
2714 | * DONE_SYNC - interrupt thread should be waiting for | |
b22cd72c | 2715 | * l_icloglock |
1da177e4 LT |
2716 | * IOERROR - give up hope all ye who enter here |
2717 | */ | |
2718 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
2719 | iclog->ic_state == XLOG_STATE_SYNCING || | |
2720 | iclog->ic_state == XLOG_STATE_DONE_SYNC || | |
2721 | iclog->ic_state == XLOG_STATE_IOERROR ) | |
2722 | break; | |
2723 | iclog = iclog->ic_next; | |
2724 | } while (first_iclog != iclog); | |
2725 | } | |
2726 | #endif | |
2727 | ||
d748c623 MW |
2728 | if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) |
2729 | wake = 1; | |
b22cd72c | 2730 | spin_unlock(&log->l_icloglock); |
d748c623 MW |
2731 | |
2732 | if (wake) | |
eb40a875 | 2733 | wake_up_all(&log->l_flush_wait); |
d748c623 | 2734 | } |
1da177e4 LT |
2735 | |
2736 | ||
2737 | /* | |
2738 | * Finish transitioning this iclog to the dirty state. | |
2739 | * | |
2740 | * Make sure that we completely execute this routine only when this is | |
2741 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2742 | * when we reach the end of the physical log, get turned into 2 separate | |
2743 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2744 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2745 | * the second completion goes through. | |
2746 | * | |
2747 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2748 | * global state machine log lock. |
1da177e4 | 2749 | */ |
a8272ce0 | 2750 | STATIC void |
1da177e4 LT |
2751 | xlog_state_done_syncing( |
2752 | xlog_in_core_t *iclog, | |
2753 | int aborted) | |
2754 | { | |
9a8d2fdb | 2755 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2756 | |
b22cd72c | 2757 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2758 | |
2759 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING || | |
2760 | iclog->ic_state == XLOG_STATE_IOERROR); | |
155cc6b7 | 2761 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
2762 | ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2); |
2763 | ||
2764 | ||
2765 | /* | |
2766 | * If we got an error, either on the first buffer, or in the case of | |
2767 | * split log writes, on the second, we mark ALL iclogs STATE_IOERROR, | |
2768 | * and none should ever be attempted to be written to disk | |
2769 | * again. | |
2770 | */ | |
2771 | if (iclog->ic_state != XLOG_STATE_IOERROR) { | |
2772 | if (--iclog->ic_bwritecnt == 1) { | |
b22cd72c | 2773 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2774 | return; |
2775 | } | |
2776 | iclog->ic_state = XLOG_STATE_DONE_SYNC; | |
2777 | } | |
2778 | ||
2779 | /* | |
2780 | * Someone could be sleeping prior to writing out the next | |
2781 | * iclog buffer, we wake them all, one will get to do the | |
2782 | * I/O, the others get to wait for the result. | |
2783 | */ | |
eb40a875 | 2784 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2785 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2786 | xlog_state_do_callback(log, aborted, iclog); /* also cleans log */ |
2787 | } /* xlog_state_done_syncing */ | |
2788 | ||
2789 | ||
2790 | /* | |
2791 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2792 | * sleep. We wait on the flush queue on the head iclog as that should be |
2793 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2794 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2795 | * |
2796 | * The in-core logs are used in a circular fashion. They are not used | |
2797 | * out-of-order even when an iclog past the head is free. | |
2798 | * | |
2799 | * return: | |
2800 | * * log_offset where xlog_write() can start writing into the in-core | |
2801 | * log's data space. | |
2802 | * * in-core log pointer to which xlog_write() should write. | |
2803 | * * boolean indicating this is a continued write to an in-core log. | |
2804 | * If this is the last write, then the in-core log's offset field | |
2805 | * needs to be incremented, depending on the amount of data which | |
2806 | * is copied. | |
2807 | */ | |
a8272ce0 | 2808 | STATIC int |
9a8d2fdb MT |
2809 | xlog_state_get_iclog_space( |
2810 | struct xlog *log, | |
2811 | int len, | |
2812 | struct xlog_in_core **iclogp, | |
2813 | struct xlog_ticket *ticket, | |
2814 | int *continued_write, | |
2815 | int *logoffsetp) | |
1da177e4 | 2816 | { |
1da177e4 LT |
2817 | int log_offset; |
2818 | xlog_rec_header_t *head; | |
2819 | xlog_in_core_t *iclog; | |
2820 | int error; | |
2821 | ||
2822 | restart: | |
b22cd72c | 2823 | spin_lock(&log->l_icloglock); |
1da177e4 | 2824 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2825 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2826 | return XFS_ERROR(EIO); |
2827 | } | |
2828 | ||
2829 | iclog = log->l_iclog; | |
d748c623 | 2830 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
1da177e4 | 2831 | XFS_STATS_INC(xs_log_noiclogs); |
d748c623 MW |
2832 | |
2833 | /* Wait for log writes to have flushed */ | |
eb40a875 | 2834 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
2835 | goto restart; |
2836 | } | |
d748c623 | 2837 | |
1da177e4 LT |
2838 | head = &iclog->ic_header; |
2839 | ||
155cc6b7 | 2840 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
2841 | log_offset = iclog->ic_offset; |
2842 | ||
2843 | /* On the 1st write to an iclog, figure out lsn. This works | |
2844 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
2845 | * committing to. If the offset is set, that's how many blocks | |
2846 | * must be written. | |
2847 | */ | |
2848 | if (log_offset == 0) { | |
2849 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 2850 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
2851 | log->l_iclog_hsize, |
2852 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
2853 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
2854 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 2855 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
2856 | ASSERT(log->l_curr_block >= 0); |
2857 | } | |
2858 | ||
2859 | /* If there is enough room to write everything, then do it. Otherwise, | |
2860 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
2861 | * bit is on, so this will get flushed out. Don't update ic_offset | |
2862 | * until you know exactly how many bytes get copied. Therefore, wait | |
2863 | * until later to update ic_offset. | |
2864 | * | |
2865 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
2866 | * can fit into remaining data section. | |
2867 | */ | |
2868 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
2869 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2870 | ||
49641f1a DC |
2871 | /* |
2872 | * If I'm the only one writing to this iclog, sync it to disk. | |
2873 | * We need to do an atomic compare and decrement here to avoid | |
2874 | * racing with concurrent atomic_dec_and_lock() calls in | |
2875 | * xlog_state_release_iclog() when there is more than one | |
2876 | * reference to the iclog. | |
2877 | */ | |
2878 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) { | |
2879 | /* we are the only one */ | |
b22cd72c | 2880 | spin_unlock(&log->l_icloglock); |
49641f1a DC |
2881 | error = xlog_state_release_iclog(log, iclog); |
2882 | if (error) | |
014c2544 | 2883 | return error; |
1da177e4 | 2884 | } else { |
b22cd72c | 2885 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2886 | } |
2887 | goto restart; | |
2888 | } | |
2889 | ||
2890 | /* Do we have enough room to write the full amount in the remainder | |
2891 | * of this iclog? Or must we continue a write on the next iclog and | |
2892 | * mark this iclog as completely taken? In the case where we switch | |
2893 | * iclogs (to mark it taken), this particular iclog will release/sync | |
2894 | * to disk in xlog_write(). | |
2895 | */ | |
2896 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
2897 | *continued_write = 0; | |
2898 | iclog->ic_offset += len; | |
2899 | } else { | |
2900 | *continued_write = 1; | |
2901 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2902 | } | |
2903 | *iclogp = iclog; | |
2904 | ||
2905 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 2906 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2907 | |
2908 | *logoffsetp = log_offset; | |
2909 | return 0; | |
2910 | } /* xlog_state_get_iclog_space */ | |
2911 | ||
1da177e4 LT |
2912 | /* The first cnt-1 times through here we don't need to |
2913 | * move the grant write head because the permanent | |
2914 | * reservation has reserved cnt times the unit amount. | |
2915 | * Release part of current permanent unit reservation and | |
2916 | * reset current reservation to be one units worth. Also | |
2917 | * move grant reservation head forward. | |
2918 | */ | |
2919 | STATIC void | |
9a8d2fdb MT |
2920 | xlog_regrant_reserve_log_space( |
2921 | struct xlog *log, | |
2922 | struct xlog_ticket *ticket) | |
1da177e4 | 2923 | { |
0b1b213f CH |
2924 | trace_xfs_log_regrant_reserve_enter(log, ticket); |
2925 | ||
1da177e4 LT |
2926 | if (ticket->t_cnt > 0) |
2927 | ticket->t_cnt--; | |
2928 | ||
28496968 | 2929 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 2930 | ticket->t_curr_res); |
28496968 | 2931 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 2932 | ticket->t_curr_res); |
1da177e4 | 2933 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2934 | xlog_tic_reset_res(ticket); |
0b1b213f CH |
2935 | |
2936 | trace_xfs_log_regrant_reserve_sub(log, ticket); | |
2937 | ||
1da177e4 | 2938 | /* just return if we still have some of the pre-reserved space */ |
d0eb2f38 | 2939 | if (ticket->t_cnt > 0) |
1da177e4 | 2940 | return; |
1da177e4 | 2941 | |
28496968 | 2942 | xlog_grant_add_space(log, &log->l_reserve_head.grant, |
a69ed03c | 2943 | ticket->t_unit_res); |
0b1b213f CH |
2944 | |
2945 | trace_xfs_log_regrant_reserve_exit(log, ticket); | |
2946 | ||
1da177e4 | 2947 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2948 | xlog_tic_reset_res(ticket); |
1da177e4 LT |
2949 | } /* xlog_regrant_reserve_log_space */ |
2950 | ||
2951 | ||
2952 | /* | |
2953 | * Give back the space left from a reservation. | |
2954 | * | |
2955 | * All the information we need to make a correct determination of space left | |
2956 | * is present. For non-permanent reservations, things are quite easy. The | |
2957 | * count should have been decremented to zero. We only need to deal with the | |
2958 | * space remaining in the current reservation part of the ticket. If the | |
2959 | * ticket contains a permanent reservation, there may be left over space which | |
2960 | * needs to be released. A count of N means that N-1 refills of the current | |
2961 | * reservation can be done before we need to ask for more space. The first | |
2962 | * one goes to fill up the first current reservation. Once we run out of | |
2963 | * space, the count will stay at zero and the only space remaining will be | |
2964 | * in the current reservation field. | |
2965 | */ | |
2966 | STATIC void | |
9a8d2fdb MT |
2967 | xlog_ungrant_log_space( |
2968 | struct xlog *log, | |
2969 | struct xlog_ticket *ticket) | |
1da177e4 | 2970 | { |
663e496a DC |
2971 | int bytes; |
2972 | ||
1da177e4 LT |
2973 | if (ticket->t_cnt > 0) |
2974 | ticket->t_cnt--; | |
2975 | ||
0b1b213f | 2976 | trace_xfs_log_ungrant_enter(log, ticket); |
0b1b213f | 2977 | trace_xfs_log_ungrant_sub(log, ticket); |
1da177e4 | 2978 | |
663e496a DC |
2979 | /* |
2980 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
2981 | * up more space based on the remaining count. |
2982 | */ | |
663e496a | 2983 | bytes = ticket->t_curr_res; |
1da177e4 LT |
2984 | if (ticket->t_cnt > 0) { |
2985 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 2986 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
2987 | } |
2988 | ||
28496968 CH |
2989 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
2990 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 2991 | |
0b1b213f CH |
2992 | trace_xfs_log_ungrant_exit(log, ticket); |
2993 | ||
cfb7cdca | 2994 | xfs_log_space_wake(log->l_mp); |
09a423a3 | 2995 | } |
1da177e4 | 2996 | |
1da177e4 LT |
2997 | /* |
2998 | * Flush iclog to disk if this is the last reference to the given iclog and | |
2999 | * the WANT_SYNC bit is set. | |
3000 | * | |
3001 | * When this function is entered, the iclog is not necessarily in the | |
3002 | * WANT_SYNC state. It may be sitting around waiting to get filled. | |
3003 | * | |
3004 | * | |
3005 | */ | |
a8272ce0 | 3006 | STATIC int |
b589334c | 3007 | xlog_state_release_iclog( |
9a8d2fdb MT |
3008 | struct xlog *log, |
3009 | struct xlog_in_core *iclog) | |
1da177e4 | 3010 | { |
1da177e4 LT |
3011 | int sync = 0; /* do we sync? */ |
3012 | ||
155cc6b7 DC |
3013 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3014 | return XFS_ERROR(EIO); | |
3015 | ||
3016 | ASSERT(atomic_read(&iclog->ic_refcnt) > 0); | |
3017 | if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) | |
3018 | return 0; | |
3019 | ||
1da177e4 | 3020 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
b22cd72c | 3021 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3022 | return XFS_ERROR(EIO); |
3023 | } | |
1da177e4 LT |
3024 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE || |
3025 | iclog->ic_state == XLOG_STATE_WANT_SYNC); | |
3026 | ||
155cc6b7 | 3027 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { |
b589334c | 3028 | /* update tail before writing to iclog */ |
1c3cb9ec | 3029 | xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp); |
1da177e4 LT |
3030 | sync++; |
3031 | iclog->ic_state = XLOG_STATE_SYNCING; | |
1c3cb9ec DC |
3032 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); |
3033 | xlog_verify_tail_lsn(log, iclog, tail_lsn); | |
1da177e4 LT |
3034 | /* cycle incremented when incrementing curr_block */ |
3035 | } | |
b22cd72c | 3036 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3037 | |
3038 | /* | |
3039 | * We let the log lock go, so it's possible that we hit a log I/O | |
c41564b5 | 3040 | * error or some other SHUTDOWN condition that marks the iclog |
1da177e4 LT |
3041 | * as XLOG_STATE_IOERROR before the bwrite. However, we know that |
3042 | * this iclog has consistent data, so we ignore IOERROR | |
3043 | * flags after this point. | |
3044 | */ | |
b589334c | 3045 | if (sync) |
1da177e4 | 3046 | return xlog_sync(log, iclog); |
014c2544 | 3047 | return 0; |
1da177e4 LT |
3048 | } /* xlog_state_release_iclog */ |
3049 | ||
3050 | ||
3051 | /* | |
3052 | * This routine will mark the current iclog in the ring as WANT_SYNC | |
3053 | * and move the current iclog pointer to the next iclog in the ring. | |
3054 | * When this routine is called from xlog_state_get_iclog_space(), the | |
3055 | * exact size of the iclog has not yet been determined. All we know is | |
3056 | * that every data block. We have run out of space in this log record. | |
3057 | */ | |
3058 | STATIC void | |
9a8d2fdb MT |
3059 | xlog_state_switch_iclogs( |
3060 | struct xlog *log, | |
3061 | struct xlog_in_core *iclog, | |
3062 | int eventual_size) | |
1da177e4 LT |
3063 | { |
3064 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
3065 | if (!eventual_size) | |
3066 | eventual_size = iclog->ic_offset; | |
3067 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3068 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3069 | log->l_prev_block = log->l_curr_block; |
3070 | log->l_prev_cycle = log->l_curr_cycle; | |
3071 | ||
3072 | /* roll log?: ic_offset changed later */ | |
3073 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3074 | ||
3075 | /* Round up to next log-sunit */ | |
62118709 | 3076 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
1da177e4 LT |
3077 | log->l_mp->m_sb.sb_logsunit > 1) { |
3078 | __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit); | |
3079 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); | |
3080 | } | |
3081 | ||
3082 | if (log->l_curr_block >= log->l_logBBsize) { | |
3083 | log->l_curr_cycle++; | |
3084 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3085 | log->l_curr_cycle++; | |
3086 | log->l_curr_block -= log->l_logBBsize; | |
3087 | ASSERT(log->l_curr_block >= 0); | |
3088 | } | |
3089 | ASSERT(iclog == log->l_iclog); | |
3090 | log->l_iclog = iclog->ic_next; | |
3091 | } /* xlog_state_switch_iclogs */ | |
3092 | ||
1da177e4 LT |
3093 | /* |
3094 | * Write out all data in the in-core log as of this exact moment in time. | |
3095 | * | |
3096 | * Data may be written to the in-core log during this call. However, | |
3097 | * we don't guarantee this data will be written out. A change from past | |
3098 | * implementation means this routine will *not* write out zero length LRs. | |
3099 | * | |
3100 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3101 | * If we determine there is no flushable data, we just return. There is no | |
3102 | * flushable data if: | |
3103 | * | |
3104 | * 1. the current iclog is active and has no data; the previous iclog | |
3105 | * is in the active or dirty state. | |
3106 | * 2. the current iclog is drity, and the previous iclog is in the | |
3107 | * active or dirty state. | |
3108 | * | |
12017faf | 3109 | * We may sleep if: |
1da177e4 LT |
3110 | * |
3111 | * 1. the current iclog is not in the active nor dirty state. | |
3112 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3113 | * active nor dirty state. | |
3114 | * 3. the current iclog is active, and there is another thread writing | |
3115 | * to this particular iclog. | |
3116 | * 4. a) the current iclog is active and has no other writers | |
3117 | * b) when we return from flushing out this iclog, it is still | |
3118 | * not in the active nor dirty state. | |
3119 | */ | |
a14a348b CH |
3120 | int |
3121 | _xfs_log_force( | |
3122 | struct xfs_mount *mp, | |
3123 | uint flags, | |
3124 | int *log_flushed) | |
1da177e4 | 3125 | { |
ad223e60 | 3126 | struct xlog *log = mp->m_log; |
a14a348b CH |
3127 | struct xlog_in_core *iclog; |
3128 | xfs_lsn_t lsn; | |
3129 | ||
3130 | XFS_STATS_INC(xs_log_force); | |
1da177e4 | 3131 | |
93b8a585 | 3132 | xlog_cil_force(log); |
71e330b5 | 3133 | |
b22cd72c | 3134 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3135 | |
3136 | iclog = log->l_iclog; | |
3137 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3138 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3139 | return XFS_ERROR(EIO); |
3140 | } | |
3141 | ||
3142 | /* If the head iclog is not active nor dirty, we just attach | |
3143 | * ourselves to the head and go to sleep. | |
3144 | */ | |
3145 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3146 | iclog->ic_state == XLOG_STATE_DIRTY) { | |
3147 | /* | |
3148 | * If the head is dirty or (active and empty), then | |
3149 | * we need to look at the previous iclog. If the previous | |
3150 | * iclog is active or dirty we are done. There is nothing | |
3151 | * to sync out. Otherwise, we attach ourselves to the | |
3152 | * previous iclog and go to sleep. | |
3153 | */ | |
3154 | if (iclog->ic_state == XLOG_STATE_DIRTY || | |
155cc6b7 DC |
3155 | (atomic_read(&iclog->ic_refcnt) == 0 |
3156 | && iclog->ic_offset == 0)) { | |
1da177e4 LT |
3157 | iclog = iclog->ic_prev; |
3158 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3159 | iclog->ic_state == XLOG_STATE_DIRTY) | |
3160 | goto no_sleep; | |
3161 | else | |
3162 | goto maybe_sleep; | |
3163 | } else { | |
155cc6b7 | 3164 | if (atomic_read(&iclog->ic_refcnt) == 0) { |
1da177e4 LT |
3165 | /* We are the only one with access to this |
3166 | * iclog. Flush it out now. There should | |
3167 | * be a roundoff of zero to show that someone | |
3168 | * has already taken care of the roundoff from | |
3169 | * the previous sync. | |
3170 | */ | |
155cc6b7 | 3171 | atomic_inc(&iclog->ic_refcnt); |
b53e675d | 3172 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
1da177e4 | 3173 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 3174 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3175 | |
3176 | if (xlog_state_release_iclog(log, iclog)) | |
3177 | return XFS_ERROR(EIO); | |
a14a348b CH |
3178 | |
3179 | if (log_flushed) | |
3180 | *log_flushed = 1; | |
b22cd72c | 3181 | spin_lock(&log->l_icloglock); |
b53e675d | 3182 | if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn && |
1da177e4 LT |
3183 | iclog->ic_state != XLOG_STATE_DIRTY) |
3184 | goto maybe_sleep; | |
3185 | else | |
3186 | goto no_sleep; | |
3187 | } else { | |
3188 | /* Someone else is writing to this iclog. | |
3189 | * Use its call to flush out the data. However, | |
3190 | * the other thread may not force out this LR, | |
3191 | * so we mark it WANT_SYNC. | |
3192 | */ | |
3193 | xlog_state_switch_iclogs(log, iclog, 0); | |
3194 | goto maybe_sleep; | |
3195 | } | |
3196 | } | |
3197 | } | |
3198 | ||
3199 | /* By the time we come around again, the iclog could've been filled | |
3200 | * which would give it another lsn. If we have a new lsn, just | |
3201 | * return because the relevant data has been flushed. | |
3202 | */ | |
3203 | maybe_sleep: | |
3204 | if (flags & XFS_LOG_SYNC) { | |
3205 | /* | |
3206 | * We must check if we're shutting down here, before | |
b22cd72c | 3207 | * we wait, while we're holding the l_icloglock. |
1da177e4 LT |
3208 | * Then we check again after waking up, in case our |
3209 | * sleep was disturbed by a bad news. | |
3210 | */ | |
3211 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3212 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3213 | return XFS_ERROR(EIO); |
3214 | } | |
3215 | XFS_STATS_INC(xs_log_force_sleep); | |
eb40a875 | 3216 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); |
1da177e4 LT |
3217 | /* |
3218 | * No need to grab the log lock here since we're | |
3219 | * only deciding whether or not to return EIO | |
3220 | * and the memory read should be atomic. | |
3221 | */ | |
3222 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3223 | return XFS_ERROR(EIO); | |
a14a348b CH |
3224 | if (log_flushed) |
3225 | *log_flushed = 1; | |
1da177e4 LT |
3226 | } else { |
3227 | ||
3228 | no_sleep: | |
b22cd72c | 3229 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3230 | } |
3231 | return 0; | |
a14a348b | 3232 | } |
1da177e4 | 3233 | |
a14a348b CH |
3234 | /* |
3235 | * Wrapper for _xfs_log_force(), to be used when caller doesn't care | |
3236 | * about errors or whether the log was flushed or not. This is the normal | |
3237 | * interface to use when trying to unpin items or move the log forward. | |
3238 | */ | |
3239 | void | |
3240 | xfs_log_force( | |
3241 | xfs_mount_t *mp, | |
3242 | uint flags) | |
3243 | { | |
3244 | int error; | |
3245 | ||
14c26c6a | 3246 | trace_xfs_log_force(mp, 0); |
a14a348b | 3247 | error = _xfs_log_force(mp, flags, NULL); |
a0fa2b67 DC |
3248 | if (error) |
3249 | xfs_warn(mp, "%s: error %d returned.", __func__, error); | |
a14a348b | 3250 | } |
1da177e4 LT |
3251 | |
3252 | /* | |
a14a348b | 3253 | * Force the in-core log to disk for a specific LSN. |
1da177e4 LT |
3254 | * |
3255 | * Find in-core log with lsn. | |
3256 | * If it is in the DIRTY state, just return. | |
3257 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3258 | * state and go to sleep or return. | |
3259 | * If it is in any other state, go to sleep or return. | |
3260 | * | |
a14a348b CH |
3261 | * Synchronous forces are implemented with a signal variable. All callers |
3262 | * to force a given lsn to disk will wait on a the sv attached to the | |
3263 | * specific in-core log. When given in-core log finally completes its | |
3264 | * write to disk, that thread will wake up all threads waiting on the | |
3265 | * sv. | |
1da177e4 | 3266 | */ |
a14a348b CH |
3267 | int |
3268 | _xfs_log_force_lsn( | |
3269 | struct xfs_mount *mp, | |
3270 | xfs_lsn_t lsn, | |
3271 | uint flags, | |
3272 | int *log_flushed) | |
1da177e4 | 3273 | { |
ad223e60 | 3274 | struct xlog *log = mp->m_log; |
a14a348b CH |
3275 | struct xlog_in_core *iclog; |
3276 | int already_slept = 0; | |
1da177e4 | 3277 | |
a14a348b | 3278 | ASSERT(lsn != 0); |
1da177e4 | 3279 | |
a14a348b | 3280 | XFS_STATS_INC(xs_log_force); |
1da177e4 | 3281 | |
93b8a585 CH |
3282 | lsn = xlog_cil_force_lsn(log, lsn); |
3283 | if (lsn == NULLCOMMITLSN) | |
3284 | return 0; | |
71e330b5 | 3285 | |
a14a348b CH |
3286 | try_again: |
3287 | spin_lock(&log->l_icloglock); | |
3288 | iclog = log->l_iclog; | |
3289 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3290 | spin_unlock(&log->l_icloglock); |
a14a348b | 3291 | return XFS_ERROR(EIO); |
1da177e4 LT |
3292 | } |
3293 | ||
a14a348b CH |
3294 | do { |
3295 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { | |
3296 | iclog = iclog->ic_next; | |
3297 | continue; | |
3298 | } | |
3299 | ||
3300 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
3301 | spin_unlock(&log->l_icloglock); | |
3302 | return 0; | |
3303 | } | |
3304 | ||
3305 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3306 | /* | |
3307 | * We sleep here if we haven't already slept (e.g. | |
3308 | * this is the first time we've looked at the correct | |
3309 | * iclog buf) and the buffer before us is going to | |
3310 | * be sync'ed. The reason for this is that if we | |
3311 | * are doing sync transactions here, by waiting for | |
3312 | * the previous I/O to complete, we can allow a few | |
3313 | * more transactions into this iclog before we close | |
3314 | * it down. | |
3315 | * | |
3316 | * Otherwise, we mark the buffer WANT_SYNC, and bump | |
3317 | * up the refcnt so we can release the log (which | |
3318 | * drops the ref count). The state switch keeps new | |
3319 | * transaction commits from using this buffer. When | |
3320 | * the current commits finish writing into the buffer, | |
3321 | * the refcount will drop to zero and the buffer will | |
3322 | * go out then. | |
3323 | */ | |
3324 | if (!already_slept && | |
3325 | (iclog->ic_prev->ic_state & | |
3326 | (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) { | |
3327 | ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); | |
3328 | ||
3329 | XFS_STATS_INC(xs_log_force_sleep); | |
3330 | ||
eb40a875 DC |
3331 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3332 | &log->l_icloglock); | |
a14a348b CH |
3333 | if (log_flushed) |
3334 | *log_flushed = 1; | |
3335 | already_slept = 1; | |
3336 | goto try_again; | |
3337 | } | |
155cc6b7 | 3338 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 3339 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 3340 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3341 | if (xlog_state_release_iclog(log, iclog)) |
3342 | return XFS_ERROR(EIO); | |
a14a348b CH |
3343 | if (log_flushed) |
3344 | *log_flushed = 1; | |
b22cd72c | 3345 | spin_lock(&log->l_icloglock); |
1da177e4 | 3346 | } |
1da177e4 | 3347 | |
a14a348b CH |
3348 | if ((flags & XFS_LOG_SYNC) && /* sleep */ |
3349 | !(iclog->ic_state & | |
3350 | (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) { | |
3351 | /* | |
3352 | * Don't wait on completion if we know that we've | |
3353 | * gotten a log write error. | |
3354 | */ | |
3355 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
3356 | spin_unlock(&log->l_icloglock); | |
3357 | return XFS_ERROR(EIO); | |
3358 | } | |
3359 | XFS_STATS_INC(xs_log_force_sleep); | |
eb40a875 | 3360 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); |
a14a348b CH |
3361 | /* |
3362 | * No need to grab the log lock here since we're | |
3363 | * only deciding whether or not to return EIO | |
3364 | * and the memory read should be atomic. | |
3365 | */ | |
3366 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3367 | return XFS_ERROR(EIO); | |
1da177e4 | 3368 | |
a14a348b CH |
3369 | if (log_flushed) |
3370 | *log_flushed = 1; | |
3371 | } else { /* just return */ | |
b22cd72c | 3372 | spin_unlock(&log->l_icloglock); |
1da177e4 | 3373 | } |
1da177e4 | 3374 | |
a14a348b CH |
3375 | return 0; |
3376 | } while (iclog != log->l_iclog); | |
1da177e4 | 3377 | |
a14a348b CH |
3378 | spin_unlock(&log->l_icloglock); |
3379 | return 0; | |
3380 | } | |
3381 | ||
3382 | /* | |
3383 | * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care | |
3384 | * about errors or whether the log was flushed or not. This is the normal | |
3385 | * interface to use when trying to unpin items or move the log forward. | |
3386 | */ | |
3387 | void | |
3388 | xfs_log_force_lsn( | |
3389 | xfs_mount_t *mp, | |
3390 | xfs_lsn_t lsn, | |
3391 | uint flags) | |
3392 | { | |
3393 | int error; | |
1da177e4 | 3394 | |
14c26c6a | 3395 | trace_xfs_log_force(mp, lsn); |
a14a348b | 3396 | error = _xfs_log_force_lsn(mp, lsn, flags, NULL); |
a0fa2b67 DC |
3397 | if (error) |
3398 | xfs_warn(mp, "%s: error %d returned.", __func__, error); | |
a14a348b | 3399 | } |
1da177e4 LT |
3400 | |
3401 | /* | |
3402 | * Called when we want to mark the current iclog as being ready to sync to | |
3403 | * disk. | |
3404 | */ | |
a8272ce0 | 3405 | STATIC void |
9a8d2fdb MT |
3406 | xlog_state_want_sync( |
3407 | struct xlog *log, | |
3408 | struct xlog_in_core *iclog) | |
1da177e4 | 3409 | { |
a8914f3a | 3410 | assert_spin_locked(&log->l_icloglock); |
1da177e4 LT |
3411 | |
3412 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3413 | xlog_state_switch_iclogs(log, iclog, 0); | |
3414 | } else { | |
3415 | ASSERT(iclog->ic_state & | |
3416 | (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR)); | |
3417 | } | |
39e2defe | 3418 | } |
1da177e4 LT |
3419 | |
3420 | ||
3421 | /***************************************************************************** | |
3422 | * | |
3423 | * TICKET functions | |
3424 | * | |
3425 | ***************************************************************************** | |
3426 | */ | |
3427 | ||
3428 | /* | |
9da096fd | 3429 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3430 | */ |
cc09c0dc DC |
3431 | void |
3432 | xfs_log_ticket_put( | |
3433 | xlog_ticket_t *ticket) | |
1da177e4 | 3434 | { |
cc09c0dc | 3435 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3436 | if (atomic_dec_and_test(&ticket->t_ref)) |
cc09c0dc | 3437 | kmem_zone_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3438 | } |
1da177e4 | 3439 | |
cc09c0dc DC |
3440 | xlog_ticket_t * |
3441 | xfs_log_ticket_get( | |
3442 | xlog_ticket_t *ticket) | |
3443 | { | |
3444 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3445 | atomic_inc(&ticket->t_ref); | |
3446 | return ticket; | |
3447 | } | |
1da177e4 LT |
3448 | |
3449 | /* | |
e773fc93 JL |
3450 | * Figure out the total log space unit (in bytes) that would be |
3451 | * required for a log ticket. | |
1da177e4 | 3452 | */ |
e773fc93 JL |
3453 | int |
3454 | xfs_log_calc_unit_res( | |
3455 | struct xfs_mount *mp, | |
3456 | int unit_bytes) | |
1da177e4 | 3457 | { |
e773fc93 JL |
3458 | struct xlog *log = mp->m_log; |
3459 | int iclog_space; | |
3460 | uint num_headers; | |
1da177e4 LT |
3461 | |
3462 | /* | |
3463 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3464 | * in the log. A unit in this case is the amount of space for one | |
3465 | * of these log operations. Normal reservations have a cnt of 1 | |
3466 | * and their unit amount is the total amount of space required. | |
3467 | * | |
3468 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3469 | * which occupy space in the on-disk log. |
3470 | * | |
3471 | * Normal form of a transaction is: | |
3472 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3473 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3474 | * | |
3475 | * We need to account for all the leadup data and trailer data | |
3476 | * around the transaction data. | |
3477 | * And then we need to account for the worst case in terms of using | |
3478 | * more space. | |
3479 | * The worst case will happen if: | |
3480 | * - the placement of the transaction happens to be such that the | |
3481 | * roundoff is at its maximum | |
3482 | * - the transaction data is synced before the commit record is synced | |
3483 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3484 | * Therefore the commit record is in its own Log Record. | |
3485 | * This can happen as the commit record is called with its | |
3486 | * own region to xlog_write(). | |
3487 | * This then means that in the worst case, roundoff can happen for | |
3488 | * the commit-rec as well. | |
3489 | * The commit-rec is smaller than padding in this scenario and so it is | |
3490 | * not added separately. | |
1da177e4 LT |
3491 | */ |
3492 | ||
32fb9b57 TS |
3493 | /* for trans header */ |
3494 | unit_bytes += sizeof(xlog_op_header_t); | |
3495 | unit_bytes += sizeof(xfs_trans_header_t); | |
3496 | ||
1da177e4 | 3497 | /* for start-rec */ |
32fb9b57 TS |
3498 | unit_bytes += sizeof(xlog_op_header_t); |
3499 | ||
9b9fc2b7 DC |
3500 | /* |
3501 | * for LR headers - the space for data in an iclog is the size minus | |
3502 | * the space used for the headers. If we use the iclog size, then we | |
3503 | * undercalculate the number of headers required. | |
3504 | * | |
3505 | * Furthermore - the addition of op headers for split-recs might | |
3506 | * increase the space required enough to require more log and op | |
3507 | * headers, so take that into account too. | |
3508 | * | |
3509 | * IMPORTANT: This reservation makes the assumption that if this | |
3510 | * transaction is the first in an iclog and hence has the LR headers | |
3511 | * accounted to it, then the remaining space in the iclog is | |
3512 | * exclusively for this transaction. i.e. if the transaction is larger | |
3513 | * than the iclog, it will be the only thing in that iclog. | |
3514 | * Fundamentally, this means we must pass the entire log vector to | |
3515 | * xlog_write to guarantee this. | |
3516 | */ | |
3517 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3518 | num_headers = howmany(unit_bytes, iclog_space); | |
3519 | ||
3520 | /* for split-recs - ophdrs added when data split over LRs */ | |
3521 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3522 | ||
3523 | /* add extra header reservations if we overrun */ | |
3524 | while (!num_headers || | |
3525 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3526 | unit_bytes += sizeof(xlog_op_header_t); | |
3527 | num_headers++; | |
3528 | } | |
32fb9b57 | 3529 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3530 | |
32fb9b57 TS |
3531 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3532 | unit_bytes += log->l_iclog_hsize; | |
3533 | ||
32fb9b57 | 3534 | /* for roundoff padding for transaction data and one for commit record */ |
e773fc93 | 3535 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) { |
1da177e4 | 3536 | /* log su roundoff */ |
e773fc93 | 3537 | unit_bytes += 2 * mp->m_sb.sb_logsunit; |
1da177e4 LT |
3538 | } else { |
3539 | /* BB roundoff */ | |
e773fc93 | 3540 | unit_bytes += 2 * BBSIZE; |
1da177e4 LT |
3541 | } |
3542 | ||
e773fc93 JL |
3543 | return unit_bytes; |
3544 | } | |
3545 | ||
3546 | /* | |
3547 | * Allocate and initialise a new log ticket. | |
3548 | */ | |
3549 | struct xlog_ticket * | |
3550 | xlog_ticket_alloc( | |
3551 | struct xlog *log, | |
3552 | int unit_bytes, | |
3553 | int cnt, | |
3554 | char client, | |
3555 | bool permanent, | |
3556 | xfs_km_flags_t alloc_flags) | |
3557 | { | |
3558 | struct xlog_ticket *tic; | |
3559 | int unit_res; | |
3560 | ||
3561 | tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags); | |
3562 | if (!tic) | |
3563 | return NULL; | |
3564 | ||
3565 | unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes); | |
3566 | ||
cc09c0dc | 3567 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3568 | tic->t_task = current; |
10547941 | 3569 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3570 | tic->t_unit_res = unit_res; |
3571 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3572 | tic->t_cnt = cnt; |
3573 | tic->t_ocnt = cnt; | |
ecb3403d | 3574 | tic->t_tid = prandom_u32(); |
1da177e4 LT |
3575 | tic->t_clientid = client; |
3576 | tic->t_flags = XLOG_TIC_INITED; | |
7e9c6396 | 3577 | tic->t_trans_type = 0; |
9006fb91 | 3578 | if (permanent) |
1da177e4 | 3579 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3580 | |
0adba536 | 3581 | xlog_tic_reset_res(tic); |
7e9c6396 | 3582 | |
1da177e4 | 3583 | return tic; |
cc09c0dc | 3584 | } |
1da177e4 LT |
3585 | |
3586 | ||
3587 | /****************************************************************************** | |
3588 | * | |
3589 | * Log debug routines | |
3590 | * | |
3591 | ****************************************************************************** | |
3592 | */ | |
cfcbbbd0 | 3593 | #if defined(DEBUG) |
1da177e4 LT |
3594 | /* |
3595 | * Make sure that the destination ptr is within the valid data region of | |
3596 | * one of the iclogs. This uses backup pointers stored in a different | |
3597 | * part of the log in case we trash the log structure. | |
3598 | */ | |
3599 | void | |
e6b1f273 | 3600 | xlog_verify_dest_ptr( |
ad223e60 | 3601 | struct xlog *log, |
e6b1f273 | 3602 | char *ptr) |
1da177e4 LT |
3603 | { |
3604 | int i; | |
3605 | int good_ptr = 0; | |
3606 | ||
e6b1f273 CH |
3607 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3608 | if (ptr >= log->l_iclog_bak[i] && | |
3609 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3610 | good_ptr++; |
3611 | } | |
e6b1f273 CH |
3612 | |
3613 | if (!good_ptr) | |
a0fa2b67 | 3614 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3615 | } |
1da177e4 | 3616 | |
da8a1a4a DC |
3617 | /* |
3618 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3619 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3620 | * the cycles differ by exactly one and check the byte count. | |
3621 | * | |
3622 | * This check is run unlocked, so can give false positives. Rather than assert | |
3623 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3624 | * determine if they want to panic the machine when such an error occurs. For | |
3625 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3626 | * an assert, it can be turned off at runtime. | |
3627 | */ | |
3f336c6f DC |
3628 | STATIC void |
3629 | xlog_verify_grant_tail( | |
ad223e60 | 3630 | struct xlog *log) |
3f336c6f | 3631 | { |
1c3cb9ec | 3632 | int tail_cycle, tail_blocks; |
a69ed03c | 3633 | int cycle, space; |
3f336c6f | 3634 | |
28496968 | 3635 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3636 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3637 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3638 | if (cycle - 1 != tail_cycle && |
3639 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3640 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3641 | "%s: cycle - 1 != tail_cycle", __func__); | |
3642 | log->l_flags |= XLOG_TAIL_WARN; | |
3643 | } | |
3644 | ||
3645 | if (space > BBTOB(tail_blocks) && | |
3646 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3647 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3648 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3649 | log->l_flags |= XLOG_TAIL_WARN; | |
3650 | } | |
3f336c6f DC |
3651 | } |
3652 | } | |
3653 | ||
1da177e4 LT |
3654 | /* check if it will fit */ |
3655 | STATIC void | |
9a8d2fdb MT |
3656 | xlog_verify_tail_lsn( |
3657 | struct xlog *log, | |
3658 | struct xlog_in_core *iclog, | |
3659 | xfs_lsn_t tail_lsn) | |
1da177e4 LT |
3660 | { |
3661 | int blocks; | |
3662 | ||
3663 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3664 | blocks = | |
3665 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3666 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3667 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3668 | } else { |
3669 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3670 | ||
3671 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3672 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3673 | |
3674 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3675 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3676 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3677 | } |
3678 | } /* xlog_verify_tail_lsn */ | |
3679 | ||
3680 | /* | |
3681 | * Perform a number of checks on the iclog before writing to disk. | |
3682 | * | |
3683 | * 1. Make sure the iclogs are still circular | |
3684 | * 2. Make sure we have a good magic number | |
3685 | * 3. Make sure we don't have magic numbers in the data | |
3686 | * 4. Check fields of each log operation header for: | |
3687 | * A. Valid client identifier | |
3688 | * B. tid ptr value falls in valid ptr space (user space code) | |
3689 | * C. Length in log record header is correct according to the | |
3690 | * individual operation headers within record. | |
3691 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3692 | * log, check the preceding blocks of the physical log to make sure all | |
3693 | * the cycle numbers agree with the current cycle number. | |
3694 | */ | |
3695 | STATIC void | |
9a8d2fdb MT |
3696 | xlog_verify_iclog( |
3697 | struct xlog *log, | |
3698 | struct xlog_in_core *iclog, | |
3699 | int count, | |
667a9291 | 3700 | bool syncing) |
1da177e4 LT |
3701 | { |
3702 | xlog_op_header_t *ophead; | |
3703 | xlog_in_core_t *icptr; | |
3704 | xlog_in_core_2_t *xhdr; | |
3705 | xfs_caddr_t ptr; | |
3706 | xfs_caddr_t base_ptr; | |
3707 | __psint_t field_offset; | |
3708 | __uint8_t clientid; | |
3709 | int len, i, j, k, op_len; | |
3710 | int idx; | |
1da177e4 LT |
3711 | |
3712 | /* check validity of iclog pointers */ | |
b22cd72c | 3713 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3714 | icptr = log->l_iclog; |
3715 | for (i=0; i < log->l_iclog_bufs; i++) { | |
4b80916b | 3716 | if (icptr == NULL) |
a0fa2b67 | 3717 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
1da177e4 LT |
3718 | icptr = icptr->ic_next; |
3719 | } | |
3720 | if (icptr != log->l_iclog) | |
a0fa2b67 | 3721 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3722 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3723 | |
3724 | /* check log magic numbers */ | |
69ef921b | 3725 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3726 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3727 | |
b53e675d CH |
3728 | ptr = (xfs_caddr_t) &iclog->ic_header; |
3729 | for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count; | |
1da177e4 | 3730 | ptr += BBSIZE) { |
69ef921b | 3731 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3732 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3733 | __func__); | |
1da177e4 LT |
3734 | } |
3735 | ||
3736 | /* check fields */ | |
b53e675d | 3737 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
1da177e4 LT |
3738 | ptr = iclog->ic_datap; |
3739 | base_ptr = ptr; | |
3740 | ophead = (xlog_op_header_t *)ptr; | |
b28708d6 | 3741 | xhdr = iclog->ic_data; |
1da177e4 LT |
3742 | for (i = 0; i < len; i++) { |
3743 | ophead = (xlog_op_header_t *)ptr; | |
3744 | ||
3745 | /* clientid is only 1 byte */ | |
3746 | field_offset = (__psint_t) | |
3747 | ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr); | |
667a9291 | 3748 | if (!syncing || (field_offset & 0x1ff)) { |
1da177e4 LT |
3749 | clientid = ophead->oh_clientid; |
3750 | } else { | |
3751 | idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap); | |
3752 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3753 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3754 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3755 | clientid = xlog_get_client_id( |
3756 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3757 | } else { |
03bea6fe CH |
3758 | clientid = xlog_get_client_id( |
3759 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3760 | } |
3761 | } | |
3762 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 DC |
3763 | xfs_warn(log->l_mp, |
3764 | "%s: invalid clientid %d op 0x%p offset 0x%lx", | |
3765 | __func__, clientid, ophead, | |
3766 | (unsigned long)field_offset); | |
1da177e4 LT |
3767 | |
3768 | /* check length */ | |
3769 | field_offset = (__psint_t) | |
3770 | ((xfs_caddr_t)&(ophead->oh_len) - base_ptr); | |
667a9291 | 3771 | if (!syncing || (field_offset & 0x1ff)) { |
67fcb7bf | 3772 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 LT |
3773 | } else { |
3774 | idx = BTOBBT((__psint_t)&ophead->oh_len - | |
3775 | (__psint_t)iclog->ic_datap); | |
3776 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3777 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3778 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3779 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3780 | } else { |
b53e675d | 3781 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3782 | } |
3783 | } | |
3784 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3785 | } | |
3786 | } /* xlog_verify_iclog */ | |
cfcbbbd0 | 3787 | #endif |
1da177e4 LT |
3788 | |
3789 | /* | |
b22cd72c | 3790 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3791 | */ |
3792 | STATIC int | |
3793 | xlog_state_ioerror( | |
9a8d2fdb | 3794 | struct xlog *log) |
1da177e4 LT |
3795 | { |
3796 | xlog_in_core_t *iclog, *ic; | |
3797 | ||
3798 | iclog = log->l_iclog; | |
3799 | if (! (iclog->ic_state & XLOG_STATE_IOERROR)) { | |
3800 | /* | |
3801 | * Mark all the incore logs IOERROR. | |
3802 | * From now on, no log flushes will result. | |
3803 | */ | |
3804 | ic = iclog; | |
3805 | do { | |
3806 | ic->ic_state = XLOG_STATE_IOERROR; | |
3807 | ic = ic->ic_next; | |
3808 | } while (ic != iclog); | |
014c2544 | 3809 | return 0; |
1da177e4 LT |
3810 | } |
3811 | /* | |
3812 | * Return non-zero, if state transition has already happened. | |
3813 | */ | |
014c2544 | 3814 | return 1; |
1da177e4 LT |
3815 | } |
3816 | ||
3817 | /* | |
3818 | * This is called from xfs_force_shutdown, when we're forcibly | |
3819 | * shutting down the filesystem, typically because of an IO error. | |
3820 | * Our main objectives here are to make sure that: | |
3821 | * a. the filesystem gets marked 'SHUTDOWN' for all interested | |
3822 | * parties to find out, 'atomically'. | |
3823 | * b. those who're sleeping on log reservations, pinned objects and | |
3824 | * other resources get woken up, and be told the bad news. | |
3825 | * c. nothing new gets queued up after (a) and (b) are done. | |
3826 | * d. if !logerror, flush the iclogs to disk, then seal them off | |
3827 | * for business. | |
9da1ab18 DC |
3828 | * |
3829 | * Note: for delayed logging the !logerror case needs to flush the regions | |
3830 | * held in memory out to the iclogs before flushing them to disk. This needs | |
3831 | * to be done before the log is marked as shutdown, otherwise the flush to the | |
3832 | * iclogs will fail. | |
1da177e4 LT |
3833 | */ |
3834 | int | |
3835 | xfs_log_force_umount( | |
3836 | struct xfs_mount *mp, | |
3837 | int logerror) | |
3838 | { | |
9a8d2fdb | 3839 | struct xlog *log; |
1da177e4 | 3840 | int retval; |
1da177e4 LT |
3841 | |
3842 | log = mp->m_log; | |
3843 | ||
3844 | /* | |
3845 | * If this happens during log recovery, don't worry about | |
3846 | * locking; the log isn't open for business yet. | |
3847 | */ | |
3848 | if (!log || | |
3849 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3850 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 CH |
3851 | if (mp->m_sb_bp) |
3852 | XFS_BUF_DONE(mp->m_sb_bp); | |
014c2544 | 3853 | return 0; |
1da177e4 LT |
3854 | } |
3855 | ||
3856 | /* | |
3857 | * Somebody could've already done the hard work for us. | |
3858 | * No need to get locks for this. | |
3859 | */ | |
3860 | if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) { | |
3861 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); | |
014c2544 | 3862 | return 1; |
1da177e4 LT |
3863 | } |
3864 | retval = 0; | |
9da1ab18 DC |
3865 | |
3866 | /* | |
3867 | * Flush the in memory commit item list before marking the log as | |
3868 | * being shut down. We need to do it in this order to ensure all the | |
3869 | * completed transactions are flushed to disk with the xfs_log_force() | |
3870 | * call below. | |
3871 | */ | |
93b8a585 | 3872 | if (!logerror) |
a44f13ed | 3873 | xlog_cil_force(log); |
9da1ab18 | 3874 | |
1da177e4 | 3875 | /* |
3f16b985 DC |
3876 | * mark the filesystem and the as in a shutdown state and wake |
3877 | * everybody up to tell them the bad news. | |
1da177e4 | 3878 | */ |
b22cd72c | 3879 | spin_lock(&log->l_icloglock); |
1da177e4 | 3880 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 CH |
3881 | if (mp->m_sb_bp) |
3882 | XFS_BUF_DONE(mp->m_sb_bp); | |
3883 | ||
1da177e4 LT |
3884 | /* |
3885 | * This flag is sort of redundant because of the mount flag, but | |
3886 | * it's good to maintain the separation between the log and the rest | |
3887 | * of XFS. | |
3888 | */ | |
3889 | log->l_flags |= XLOG_IO_ERROR; | |
3890 | ||
3891 | /* | |
3892 | * If we hit a log error, we want to mark all the iclogs IOERROR | |
3893 | * while we're still holding the loglock. | |
3894 | */ | |
3895 | if (logerror) | |
3896 | retval = xlog_state_ioerror(log); | |
b22cd72c | 3897 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3898 | |
3899 | /* | |
10547941 DC |
3900 | * We don't want anybody waiting for log reservations after this. That |
3901 | * means we have to wake up everybody queued up on reserveq as well as | |
3902 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3903 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3904 | * action is protected by the grant locks. |
1da177e4 | 3905 | */ |
a79bf2d7 CH |
3906 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3907 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3908 | |
a14a348b | 3909 | if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) { |
1da177e4 LT |
3910 | ASSERT(!logerror); |
3911 | /* | |
3912 | * Force the incore logs to disk before shutting the | |
3913 | * log down completely. | |
3914 | */ | |
a14a348b CH |
3915 | _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
3916 | ||
b22cd72c | 3917 | spin_lock(&log->l_icloglock); |
1da177e4 | 3918 | retval = xlog_state_ioerror(log); |
b22cd72c | 3919 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3920 | } |
3921 | /* | |
3922 | * Wake up everybody waiting on xfs_log_force. | |
3923 | * Callback all log item committed functions as if the | |
3924 | * log writes were completed. | |
3925 | */ | |
3926 | xlog_state_do_callback(log, XFS_LI_ABORTED, NULL); | |
3927 | ||
3928 | #ifdef XFSERRORDEBUG | |
3929 | { | |
3930 | xlog_in_core_t *iclog; | |
3931 | ||
b22cd72c | 3932 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3933 | iclog = log->l_iclog; |
3934 | do { | |
3935 | ASSERT(iclog->ic_callback == 0); | |
3936 | iclog = iclog->ic_next; | |
3937 | } while (iclog != log->l_iclog); | |
b22cd72c | 3938 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3939 | } |
3940 | #endif | |
3941 | /* return non-zero if log IOERROR transition had already happened */ | |
014c2544 | 3942 | return retval; |
1da177e4 LT |
3943 | } |
3944 | ||
ba0f32d4 | 3945 | STATIC int |
9a8d2fdb MT |
3946 | xlog_iclogs_empty( |
3947 | struct xlog *log) | |
1da177e4 LT |
3948 | { |
3949 | xlog_in_core_t *iclog; | |
3950 | ||
3951 | iclog = log->l_iclog; | |
3952 | do { | |
3953 | /* endianness does not matter here, zero is zero in | |
3954 | * any language. | |
3955 | */ | |
3956 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3957 | return 0; |
1da177e4 LT |
3958 | iclog = iclog->ic_next; |
3959 | } while (iclog != log->l_iclog); | |
014c2544 | 3960 | return 1; |
1da177e4 | 3961 | } |
f661f1e0 | 3962 |