]>
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" |
239880ef DC |
21 | #include "xfs_log_format.h" |
22 | #include "xfs_trans_resv.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" | |
239880ef DC |
27 | #include "xfs_trans.h" |
28 | #include "xfs_trans_priv.h" | |
29 | #include "xfs_log.h" | |
1da177e4 LT |
30 | #include "xfs_log_priv.h" |
31 | #include "xfs_buf_item.h" | |
a844f451 | 32 | #include "xfs_bmap_btree.h" |
1da177e4 | 33 | #include "xfs_alloc_btree.h" |
a844f451 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 | 35 | #include "xfs_log_recover.h" |
a844f451 NS |
36 | #include "xfs_dinode.h" |
37 | #include "xfs_inode.h" | |
0b1b213f | 38 | #include "xfs_trace.h" |
f661f1e0 | 39 | #include "xfs_fsops.h" |
0e446be4 | 40 | #include "xfs_cksum.h" |
1da177e4 | 41 | |
eb01c9cd | 42 | kmem_zone_t *xfs_log_ticket_zone; |
1da177e4 | 43 | |
1da177e4 | 44 | /* Local miscellaneous function prototypes */ |
ad223e60 MT |
45 | STATIC int |
46 | xlog_commit_record( | |
47 | struct xlog *log, | |
48 | struct xlog_ticket *ticket, | |
49 | struct xlog_in_core **iclog, | |
50 | xfs_lsn_t *commitlsnp); | |
51 | ||
9a8d2fdb MT |
52 | STATIC struct xlog * |
53 | xlog_alloc_log( | |
54 | struct xfs_mount *mp, | |
55 | struct xfs_buftarg *log_target, | |
56 | xfs_daddr_t blk_offset, | |
57 | int num_bblks); | |
ad223e60 MT |
58 | STATIC int |
59 | xlog_space_left( | |
60 | struct xlog *log, | |
61 | atomic64_t *head); | |
9a8d2fdb MT |
62 | STATIC int |
63 | xlog_sync( | |
64 | struct xlog *log, | |
65 | struct xlog_in_core *iclog); | |
66 | STATIC void | |
67 | xlog_dealloc_log( | |
68 | struct xlog *log); | |
1da177e4 LT |
69 | |
70 | /* local state machine functions */ | |
71 | STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int); | |
9a8d2fdb MT |
72 | STATIC void |
73 | xlog_state_do_callback( | |
74 | struct xlog *log, | |
75 | int aborted, | |
76 | struct xlog_in_core *iclog); | |
77 | STATIC int | |
78 | xlog_state_get_iclog_space( | |
79 | struct xlog *log, | |
80 | int len, | |
81 | struct xlog_in_core **iclog, | |
82 | struct xlog_ticket *ticket, | |
83 | int *continued_write, | |
84 | int *logoffsetp); | |
85 | STATIC int | |
86 | xlog_state_release_iclog( | |
87 | struct xlog *log, | |
88 | struct xlog_in_core *iclog); | |
89 | STATIC void | |
90 | xlog_state_switch_iclogs( | |
91 | struct xlog *log, | |
92 | struct xlog_in_core *iclog, | |
93 | int eventual_size); | |
94 | STATIC void | |
95 | xlog_state_want_sync( | |
96 | struct xlog *log, | |
97 | struct xlog_in_core *iclog); | |
1da177e4 | 98 | |
ad223e60 MT |
99 | STATIC void |
100 | xlog_grant_push_ail( | |
9a8d2fdb MT |
101 | struct xlog *log, |
102 | int need_bytes); | |
103 | STATIC void | |
104 | xlog_regrant_reserve_log_space( | |
105 | struct xlog *log, | |
106 | struct xlog_ticket *ticket); | |
107 | STATIC void | |
108 | xlog_ungrant_log_space( | |
109 | struct xlog *log, | |
110 | struct xlog_ticket *ticket); | |
1da177e4 | 111 | |
cfcbbbd0 | 112 | #if defined(DEBUG) |
9a8d2fdb MT |
113 | STATIC void |
114 | xlog_verify_dest_ptr( | |
115 | struct xlog *log, | |
116 | char *ptr); | |
ad223e60 MT |
117 | STATIC void |
118 | xlog_verify_grant_tail( | |
9a8d2fdb MT |
119 | struct xlog *log); |
120 | STATIC void | |
121 | xlog_verify_iclog( | |
122 | struct xlog *log, | |
123 | struct xlog_in_core *iclog, | |
124 | int count, | |
667a9291 | 125 | bool syncing); |
9a8d2fdb MT |
126 | STATIC void |
127 | xlog_verify_tail_lsn( | |
128 | struct xlog *log, | |
129 | struct xlog_in_core *iclog, | |
130 | xfs_lsn_t tail_lsn); | |
1da177e4 LT |
131 | #else |
132 | #define xlog_verify_dest_ptr(a,b) | |
3f336c6f | 133 | #define xlog_verify_grant_tail(a) |
1da177e4 LT |
134 | #define xlog_verify_iclog(a,b,c,d) |
135 | #define xlog_verify_tail_lsn(a,b,c) | |
136 | #endif | |
137 | ||
9a8d2fdb MT |
138 | STATIC int |
139 | xlog_iclogs_empty( | |
140 | struct xlog *log); | |
1da177e4 | 141 | |
dd954c69 | 142 | static void |
663e496a | 143 | xlog_grant_sub_space( |
ad223e60 MT |
144 | struct xlog *log, |
145 | atomic64_t *head, | |
146 | int bytes) | |
dd954c69 | 147 | { |
d0eb2f38 DC |
148 | int64_t head_val = atomic64_read(head); |
149 | int64_t new, old; | |
a69ed03c | 150 | |
d0eb2f38 DC |
151 | do { |
152 | int cycle, space; | |
a69ed03c | 153 | |
d0eb2f38 | 154 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 155 | |
d0eb2f38 DC |
156 | space -= bytes; |
157 | if (space < 0) { | |
158 | space += log->l_logsize; | |
159 | cycle--; | |
160 | } | |
161 | ||
162 | old = head_val; | |
163 | new = xlog_assign_grant_head_val(cycle, space); | |
164 | head_val = atomic64_cmpxchg(head, old, new); | |
165 | } while (head_val != old); | |
dd954c69 CH |
166 | } |
167 | ||
168 | static void | |
663e496a | 169 | xlog_grant_add_space( |
ad223e60 MT |
170 | struct xlog *log, |
171 | atomic64_t *head, | |
172 | int bytes) | |
dd954c69 | 173 | { |
d0eb2f38 DC |
174 | int64_t head_val = atomic64_read(head); |
175 | int64_t new, old; | |
a69ed03c | 176 | |
d0eb2f38 DC |
177 | do { |
178 | int tmp; | |
179 | int cycle, space; | |
a69ed03c | 180 | |
d0eb2f38 | 181 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 182 | |
d0eb2f38 DC |
183 | tmp = log->l_logsize - space; |
184 | if (tmp > bytes) | |
185 | space += bytes; | |
186 | else { | |
187 | space = bytes - tmp; | |
188 | cycle++; | |
189 | } | |
190 | ||
191 | old = head_val; | |
192 | new = xlog_assign_grant_head_val(cycle, space); | |
193 | head_val = atomic64_cmpxchg(head, old, new); | |
194 | } while (head_val != old); | |
dd954c69 | 195 | } |
a69ed03c | 196 | |
c303c5b8 CH |
197 | STATIC void |
198 | xlog_grant_head_init( | |
199 | struct xlog_grant_head *head) | |
200 | { | |
201 | xlog_assign_grant_head(&head->grant, 1, 0); | |
202 | INIT_LIST_HEAD(&head->waiters); | |
203 | spin_lock_init(&head->lock); | |
204 | } | |
205 | ||
a79bf2d7 CH |
206 | STATIC void |
207 | xlog_grant_head_wake_all( | |
208 | struct xlog_grant_head *head) | |
209 | { | |
210 | struct xlog_ticket *tic; | |
211 | ||
212 | spin_lock(&head->lock); | |
213 | list_for_each_entry(tic, &head->waiters, t_queue) | |
214 | wake_up_process(tic->t_task); | |
215 | spin_unlock(&head->lock); | |
216 | } | |
217 | ||
e179840d CH |
218 | static inline int |
219 | xlog_ticket_reservation( | |
ad223e60 | 220 | struct xlog *log, |
e179840d CH |
221 | struct xlog_grant_head *head, |
222 | struct xlog_ticket *tic) | |
9f9c19ec | 223 | { |
e179840d CH |
224 | if (head == &log->l_write_head) { |
225 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
226 | return tic->t_unit_res; | |
227 | } else { | |
9f9c19ec | 228 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) |
e179840d | 229 | return tic->t_unit_res * tic->t_cnt; |
9f9c19ec | 230 | else |
e179840d | 231 | return tic->t_unit_res; |
9f9c19ec | 232 | } |
9f9c19ec CH |
233 | } |
234 | ||
235 | STATIC bool | |
e179840d | 236 | xlog_grant_head_wake( |
ad223e60 | 237 | struct xlog *log, |
e179840d | 238 | struct xlog_grant_head *head, |
9f9c19ec CH |
239 | int *free_bytes) |
240 | { | |
241 | struct xlog_ticket *tic; | |
242 | int need_bytes; | |
243 | ||
e179840d CH |
244 | list_for_each_entry(tic, &head->waiters, t_queue) { |
245 | need_bytes = xlog_ticket_reservation(log, head, tic); | |
9f9c19ec CH |
246 | if (*free_bytes < need_bytes) |
247 | return false; | |
9f9c19ec | 248 | |
e179840d CH |
249 | *free_bytes -= need_bytes; |
250 | trace_xfs_log_grant_wake_up(log, tic); | |
14a7235f | 251 | wake_up_process(tic->t_task); |
9f9c19ec CH |
252 | } |
253 | ||
254 | return true; | |
255 | } | |
256 | ||
257 | STATIC int | |
23ee3df3 | 258 | xlog_grant_head_wait( |
ad223e60 | 259 | struct xlog *log, |
23ee3df3 | 260 | struct xlog_grant_head *head, |
9f9c19ec | 261 | struct xlog_ticket *tic, |
a30b0367 DC |
262 | int need_bytes) __releases(&head->lock) |
263 | __acquires(&head->lock) | |
9f9c19ec | 264 | { |
23ee3df3 | 265 | list_add_tail(&tic->t_queue, &head->waiters); |
9f9c19ec CH |
266 | |
267 | do { | |
268 | if (XLOG_FORCED_SHUTDOWN(log)) | |
269 | goto shutdown; | |
270 | xlog_grant_push_ail(log, need_bytes); | |
271 | ||
14a7235f | 272 | __set_current_state(TASK_UNINTERRUPTIBLE); |
23ee3df3 | 273 | spin_unlock(&head->lock); |
14a7235f | 274 | |
9f9c19ec | 275 | XFS_STATS_INC(xs_sleep_logspace); |
9f9c19ec | 276 | |
14a7235f CH |
277 | trace_xfs_log_grant_sleep(log, tic); |
278 | schedule(); | |
9f9c19ec CH |
279 | trace_xfs_log_grant_wake(log, tic); |
280 | ||
23ee3df3 | 281 | spin_lock(&head->lock); |
9f9c19ec CH |
282 | if (XLOG_FORCED_SHUTDOWN(log)) |
283 | goto shutdown; | |
23ee3df3 | 284 | } while (xlog_space_left(log, &head->grant) < need_bytes); |
9f9c19ec CH |
285 | |
286 | list_del_init(&tic->t_queue); | |
287 | return 0; | |
288 | shutdown: | |
289 | list_del_init(&tic->t_queue); | |
290 | return XFS_ERROR(EIO); | |
291 | } | |
292 | ||
42ceedb3 CH |
293 | /* |
294 | * Atomically get the log space required for a log ticket. | |
295 | * | |
296 | * Once a ticket gets put onto head->waiters, it will only return after the | |
297 | * needed reservation is satisfied. | |
298 | * | |
299 | * This function is structured so that it has a lock free fast path. This is | |
300 | * necessary because every new transaction reservation will come through this | |
301 | * path. Hence any lock will be globally hot if we take it unconditionally on | |
302 | * every pass. | |
303 | * | |
304 | * As tickets are only ever moved on and off head->waiters under head->lock, we | |
305 | * only need to take that lock if we are going to add the ticket to the queue | |
306 | * and sleep. We can avoid taking the lock if the ticket was never added to | |
307 | * head->waiters because the t_queue list head will be empty and we hold the | |
308 | * only reference to it so it can safely be checked unlocked. | |
309 | */ | |
310 | STATIC int | |
311 | xlog_grant_head_check( | |
ad223e60 | 312 | struct xlog *log, |
42ceedb3 CH |
313 | struct xlog_grant_head *head, |
314 | struct xlog_ticket *tic, | |
315 | int *need_bytes) | |
316 | { | |
317 | int free_bytes; | |
318 | int error = 0; | |
319 | ||
320 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); | |
321 | ||
322 | /* | |
323 | * If there are other waiters on the queue then give them a chance at | |
324 | * logspace before us. Wake up the first waiters, if we do not wake | |
325 | * up all the waiters then go to sleep waiting for more free space, | |
326 | * otherwise try to get some space for this transaction. | |
327 | */ | |
328 | *need_bytes = xlog_ticket_reservation(log, head, tic); | |
329 | free_bytes = xlog_space_left(log, &head->grant); | |
330 | if (!list_empty_careful(&head->waiters)) { | |
331 | spin_lock(&head->lock); | |
332 | if (!xlog_grant_head_wake(log, head, &free_bytes) || | |
333 | free_bytes < *need_bytes) { | |
334 | error = xlog_grant_head_wait(log, head, tic, | |
335 | *need_bytes); | |
336 | } | |
337 | spin_unlock(&head->lock); | |
338 | } else if (free_bytes < *need_bytes) { | |
339 | spin_lock(&head->lock); | |
340 | error = xlog_grant_head_wait(log, head, tic, *need_bytes); | |
341 | spin_unlock(&head->lock); | |
342 | } | |
343 | ||
344 | return error; | |
345 | } | |
346 | ||
0adba536 CH |
347 | static void |
348 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
349 | { | |
350 | tic->t_res_num = 0; | |
351 | tic->t_res_arr_sum = 0; | |
352 | tic->t_res_num_ophdrs = 0; | |
353 | } | |
354 | ||
355 | static void | |
356 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
357 | { | |
358 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
359 | /* add to overflow and start again */ | |
360 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
361 | tic->t_res_num = 0; | |
362 | tic->t_res_arr_sum = 0; | |
363 | } | |
364 | ||
365 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
366 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
367 | tic->t_res_arr_sum += len; | |
368 | tic->t_res_num++; | |
369 | } | |
dd954c69 | 370 | |
9006fb91 CH |
371 | /* |
372 | * Replenish the byte reservation required by moving the grant write head. | |
373 | */ | |
374 | int | |
375 | xfs_log_regrant( | |
376 | struct xfs_mount *mp, | |
377 | struct xlog_ticket *tic) | |
378 | { | |
ad223e60 | 379 | struct xlog *log = mp->m_log; |
9006fb91 CH |
380 | int need_bytes; |
381 | int error = 0; | |
382 | ||
383 | if (XLOG_FORCED_SHUTDOWN(log)) | |
384 | return XFS_ERROR(EIO); | |
385 | ||
386 | XFS_STATS_INC(xs_try_logspace); | |
387 | ||
388 | /* | |
389 | * This is a new transaction on the ticket, so we need to change the | |
390 | * transaction ID so that the next transaction has a different TID in | |
391 | * the log. Just add one to the existing tid so that we can see chains | |
392 | * of rolling transactions in the log easily. | |
393 | */ | |
394 | tic->t_tid++; | |
395 | ||
396 | xlog_grant_push_ail(log, tic->t_unit_res); | |
397 | ||
398 | tic->t_curr_res = tic->t_unit_res; | |
399 | xlog_tic_reset_res(tic); | |
400 | ||
401 | if (tic->t_cnt > 0) | |
402 | return 0; | |
403 | ||
404 | trace_xfs_log_regrant(log, tic); | |
405 | ||
406 | error = xlog_grant_head_check(log, &log->l_write_head, tic, | |
407 | &need_bytes); | |
408 | if (error) | |
409 | goto out_error; | |
410 | ||
411 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
412 | trace_xfs_log_regrant_exit(log, tic); | |
413 | xlog_verify_grant_tail(log); | |
414 | return 0; | |
415 | ||
416 | out_error: | |
417 | /* | |
418 | * If we are failing, make sure the ticket doesn't have any current | |
419 | * reservations. We don't want to add this back when the ticket/ | |
420 | * transaction gets cancelled. | |
421 | */ | |
422 | tic->t_curr_res = 0; | |
423 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
424 | return error; | |
425 | } | |
426 | ||
427 | /* | |
428 | * Reserve log space and return a ticket corresponding the reservation. | |
429 | * | |
430 | * Each reservation is going to reserve extra space for a log record header. | |
431 | * When writes happen to the on-disk log, we don't subtract the length of the | |
432 | * log record header from any reservation. By wasting space in each | |
433 | * reservation, we prevent over allocation problems. | |
434 | */ | |
435 | int | |
436 | xfs_log_reserve( | |
437 | struct xfs_mount *mp, | |
438 | int unit_bytes, | |
439 | int cnt, | |
440 | struct xlog_ticket **ticp, | |
441 | __uint8_t client, | |
442 | bool permanent, | |
443 | uint t_type) | |
444 | { | |
ad223e60 | 445 | struct xlog *log = mp->m_log; |
9006fb91 CH |
446 | struct xlog_ticket *tic; |
447 | int need_bytes; | |
448 | int error = 0; | |
449 | ||
450 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); | |
451 | ||
452 | if (XLOG_FORCED_SHUTDOWN(log)) | |
453 | return XFS_ERROR(EIO); | |
454 | ||
455 | XFS_STATS_INC(xs_try_logspace); | |
456 | ||
457 | ASSERT(*ticp == NULL); | |
458 | tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, | |
459 | KM_SLEEP | KM_MAYFAIL); | |
460 | if (!tic) | |
461 | return XFS_ERROR(ENOMEM); | |
462 | ||
463 | tic->t_trans_type = t_type; | |
464 | *ticp = tic; | |
465 | ||
437a255a DC |
466 | xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt |
467 | : tic->t_unit_res); | |
9006fb91 CH |
468 | |
469 | trace_xfs_log_reserve(log, tic); | |
470 | ||
471 | error = xlog_grant_head_check(log, &log->l_reserve_head, tic, | |
472 | &need_bytes); | |
473 | if (error) | |
474 | goto out_error; | |
475 | ||
476 | xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); | |
477 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
478 | trace_xfs_log_reserve_exit(log, tic); | |
479 | xlog_verify_grant_tail(log); | |
480 | return 0; | |
481 | ||
482 | out_error: | |
483 | /* | |
484 | * If we are failing, make sure the ticket doesn't have any current | |
485 | * reservations. We don't want to add this back when the ticket/ | |
486 | * transaction gets cancelled. | |
487 | */ | |
488 | tic->t_curr_res = 0; | |
489 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
490 | return error; | |
491 | } | |
492 | ||
493 | ||
1da177e4 LT |
494 | /* |
495 | * NOTES: | |
496 | * | |
497 | * 1. currblock field gets updated at startup and after in-core logs | |
498 | * marked as with WANT_SYNC. | |
499 | */ | |
500 | ||
501 | /* | |
502 | * This routine is called when a user of a log manager ticket is done with | |
503 | * the reservation. If the ticket was ever used, then a commit record for | |
504 | * the associated transaction is written out as a log operation header with | |
505 | * no data. The flag XLOG_TIC_INITED is set when the first write occurs with | |
506 | * a given ticket. If the ticket was one with a permanent reservation, then | |
507 | * a few operations are done differently. Permanent reservation tickets by | |
508 | * default don't release the reservation. They just commit the current | |
509 | * transaction with the belief that the reservation is still needed. A flag | |
510 | * must be passed in before permanent reservations are actually released. | |
511 | * When these type of tickets are not released, they need to be set into | |
512 | * the inited state again. By doing this, a start record will be written | |
513 | * out when the next write occurs. | |
514 | */ | |
515 | xfs_lsn_t | |
35a8a72f CH |
516 | xfs_log_done( |
517 | struct xfs_mount *mp, | |
518 | struct xlog_ticket *ticket, | |
519 | struct xlog_in_core **iclog, | |
520 | uint flags) | |
1da177e4 | 521 | { |
ad223e60 | 522 | struct xlog *log = mp->m_log; |
35a8a72f | 523 | xfs_lsn_t lsn = 0; |
1da177e4 | 524 | |
1da177e4 LT |
525 | if (XLOG_FORCED_SHUTDOWN(log) || |
526 | /* | |
527 | * If nothing was ever written, don't write out commit record. | |
528 | * If we get an error, just continue and give back the log ticket. | |
529 | */ | |
530 | (((ticket->t_flags & XLOG_TIC_INITED) == 0) && | |
55b66332 | 531 | (xlog_commit_record(log, ticket, iclog, &lsn)))) { |
1da177e4 LT |
532 | lsn = (xfs_lsn_t) -1; |
533 | if (ticket->t_flags & XLOG_TIC_PERM_RESERV) { | |
534 | flags |= XFS_LOG_REL_PERM_RESERV; | |
535 | } | |
536 | } | |
537 | ||
538 | ||
539 | if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 || | |
540 | (flags & XFS_LOG_REL_PERM_RESERV)) { | |
0b1b213f CH |
541 | trace_xfs_log_done_nonperm(log, ticket); |
542 | ||
1da177e4 | 543 | /* |
c41564b5 | 544 | * Release ticket if not permanent reservation or a specific |
1da177e4 LT |
545 | * request has been made to release a permanent reservation. |
546 | */ | |
547 | xlog_ungrant_log_space(log, ticket); | |
cc09c0dc | 548 | xfs_log_ticket_put(ticket); |
1da177e4 | 549 | } else { |
0b1b213f CH |
550 | trace_xfs_log_done_perm(log, ticket); |
551 | ||
1da177e4 | 552 | xlog_regrant_reserve_log_space(log, ticket); |
c6a7b0f8 LM |
553 | /* If this ticket was a permanent reservation and we aren't |
554 | * trying to release it, reset the inited flags; so next time | |
555 | * we write, a start record will be written out. | |
556 | */ | |
1da177e4 | 557 | ticket->t_flags |= XLOG_TIC_INITED; |
c6a7b0f8 | 558 | } |
1da177e4 LT |
559 | |
560 | return lsn; | |
35a8a72f | 561 | } |
1da177e4 | 562 | |
1da177e4 LT |
563 | /* |
564 | * Attaches a new iclog I/O completion callback routine during | |
565 | * transaction commit. If the log is in error state, a non-zero | |
566 | * return code is handed back and the caller is responsible for | |
567 | * executing the callback at an appropriate time. | |
568 | */ | |
569 | int | |
35a8a72f CH |
570 | xfs_log_notify( |
571 | struct xfs_mount *mp, | |
572 | struct xlog_in_core *iclog, | |
573 | xfs_log_callback_t *cb) | |
1da177e4 | 574 | { |
b22cd72c | 575 | int abortflg; |
1da177e4 | 576 | |
114d23aa | 577 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
578 | abortflg = (iclog->ic_state & XLOG_STATE_IOERROR); |
579 | if (!abortflg) { | |
580 | ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) || | |
581 | (iclog->ic_state == XLOG_STATE_WANT_SYNC)); | |
582 | cb->cb_next = NULL; | |
583 | *(iclog->ic_callback_tail) = cb; | |
584 | iclog->ic_callback_tail = &(cb->cb_next); | |
585 | } | |
114d23aa | 586 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 | 587 | return abortflg; |
35a8a72f | 588 | } |
1da177e4 LT |
589 | |
590 | int | |
35a8a72f CH |
591 | xfs_log_release_iclog( |
592 | struct xfs_mount *mp, | |
593 | struct xlog_in_core *iclog) | |
1da177e4 | 594 | { |
35a8a72f | 595 | if (xlog_state_release_iclog(mp->m_log, iclog)) { |
7d04a335 | 596 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 597 | return EIO; |
1da177e4 LT |
598 | } |
599 | ||
600 | return 0; | |
601 | } | |
602 | ||
1da177e4 LT |
603 | /* |
604 | * Mount a log filesystem | |
605 | * | |
606 | * mp - ubiquitous xfs mount point structure | |
607 | * log_target - buftarg of on-disk log device | |
608 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
609 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
610 | * | |
611 | * Return error or zero. | |
612 | */ | |
613 | int | |
249a8c11 DC |
614 | xfs_log_mount( |
615 | xfs_mount_t *mp, | |
616 | xfs_buftarg_t *log_target, | |
617 | xfs_daddr_t blk_offset, | |
618 | int num_bblks) | |
1da177e4 | 619 | { |
3e7b91cf JL |
620 | int error = 0; |
621 | int min_logfsbs; | |
249a8c11 | 622 | |
1da177e4 | 623 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) |
a0fa2b67 | 624 | xfs_notice(mp, "Mounting Filesystem"); |
1da177e4 | 625 | else { |
a0fa2b67 DC |
626 | xfs_notice(mp, |
627 | "Mounting filesystem in no-recovery mode. Filesystem will be inconsistent."); | |
bd186aa9 | 628 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
629 | } |
630 | ||
631 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e DC |
632 | if (IS_ERR(mp->m_log)) { |
633 | error = -PTR_ERR(mp->m_log); | |
644c3567 DC |
634 | goto out; |
635 | } | |
1da177e4 | 636 | |
3e7b91cf JL |
637 | /* |
638 | * Validate the given log space and drop a critical message via syslog | |
639 | * if the log size is too small that would lead to some unexpected | |
640 | * situations in transaction log space reservation stage. | |
641 | * | |
642 | * Note: we can't just reject the mount if the validation fails. This | |
643 | * would mean that people would have to downgrade their kernel just to | |
644 | * remedy the situation as there is no way to grow the log (short of | |
645 | * black magic surgery with xfs_db). | |
646 | * | |
647 | * We can, however, reject mounts for CRC format filesystems, as the | |
648 | * mkfs binary being used to make the filesystem should never create a | |
649 | * filesystem with a log that is too small. | |
650 | */ | |
651 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
652 | ||
653 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
654 | xfs_warn(mp, | |
655 | "Log size %d blocks too small, minimum size is %d blocks", | |
656 | mp->m_sb.sb_logblocks, min_logfsbs); | |
657 | error = EINVAL; | |
658 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { | |
659 | xfs_warn(mp, | |
660 | "Log size %d blocks too large, maximum size is %lld blocks", | |
661 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
662 | error = EINVAL; | |
663 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { | |
664 | xfs_warn(mp, | |
665 | "log size %lld bytes too large, maximum size is %lld bytes", | |
666 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
667 | XFS_MAX_LOG_BYTES); | |
668 | error = EINVAL; | |
669 | } | |
670 | if (error) { | |
671 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
672 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); | |
673 | ASSERT(0); | |
674 | goto out_free_log; | |
675 | } | |
676 | xfs_crit(mp, | |
677 | "Log size out of supported range. Continuing onwards, but if log hangs are\n" | |
678 | "experienced then please report this message in the bug report."); | |
679 | } | |
680 | ||
249a8c11 DC |
681 | /* |
682 | * Initialize the AIL now we have a log. | |
683 | */ | |
249a8c11 DC |
684 | error = xfs_trans_ail_init(mp); |
685 | if (error) { | |
a0fa2b67 | 686 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 687 | goto out_free_log; |
249a8c11 | 688 | } |
a9c21c1b | 689 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 690 | |
1da177e4 LT |
691 | /* |
692 | * skip log recovery on a norecovery mount. pretend it all | |
693 | * just worked. | |
694 | */ | |
695 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 696 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
697 | |
698 | if (readonly) | |
bd186aa9 | 699 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 700 | |
65be6054 | 701 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
702 | |
703 | if (readonly) | |
bd186aa9 | 704 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 705 | if (error) { |
a0fa2b67 DC |
706 | xfs_warn(mp, "log mount/recovery failed: error %d", |
707 | error); | |
26430752 | 708 | goto out_destroy_ail; |
1da177e4 LT |
709 | } |
710 | } | |
711 | ||
712 | /* Normal transactions can now occur */ | |
713 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
714 | ||
71e330b5 DC |
715 | /* |
716 | * Now the log has been fully initialised and we know were our | |
717 | * space grant counters are, we can initialise the permanent ticket | |
718 | * needed for delayed logging to work. | |
719 | */ | |
720 | xlog_cil_init_post_recovery(mp->m_log); | |
721 | ||
1da177e4 | 722 | return 0; |
26430752 CH |
723 | |
724 | out_destroy_ail: | |
725 | xfs_trans_ail_destroy(mp); | |
726 | out_free_log: | |
727 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 728 | out: |
249a8c11 | 729 | return error; |
26430752 | 730 | } |
1da177e4 LT |
731 | |
732 | /* | |
f661f1e0 DC |
733 | * Finish the recovery of the file system. This is separate from the |
734 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
735 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
736 | * here. | |
1da177e4 | 737 | * |
f661f1e0 DC |
738 | * If we finish recovery successfully, start the background log work. If we are |
739 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
740 | * it. | |
1da177e4 LT |
741 | */ |
742 | int | |
4249023a | 743 | xfs_log_mount_finish(xfs_mount_t *mp) |
1da177e4 | 744 | { |
f661f1e0 | 745 | int error = 0; |
1da177e4 | 746 | |
f661f1e0 | 747 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
4249023a | 748 | error = xlog_recover_finish(mp->m_log); |
f661f1e0 DC |
749 | if (!error) |
750 | xfs_log_work_queue(mp); | |
751 | } else { | |
bd186aa9 | 752 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
753 | } |
754 | ||
f661f1e0 | 755 | |
1da177e4 LT |
756 | return error; |
757 | } | |
758 | ||
1da177e4 LT |
759 | /* |
760 | * Final log writes as part of unmount. | |
761 | * | |
762 | * Mark the filesystem clean as unmount happens. Note that during relocation | |
763 | * this routine needs to be executed as part of source-bag while the | |
764 | * deallocation must not be done until source-end. | |
765 | */ | |
766 | ||
767 | /* | |
768 | * Unmount record used to have a string "Unmount filesystem--" in the | |
769 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
770 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 771 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
772 | * As far as I know, there weren't any dependencies on the old behaviour. |
773 | */ | |
774 | ||
775 | int | |
776 | xfs_log_unmount_write(xfs_mount_t *mp) | |
777 | { | |
9a8d2fdb | 778 | struct xlog *log = mp->m_log; |
1da177e4 LT |
779 | xlog_in_core_t *iclog; |
780 | #ifdef DEBUG | |
781 | xlog_in_core_t *first_iclog; | |
782 | #endif | |
35a8a72f | 783 | xlog_ticket_t *tic = NULL; |
1da177e4 LT |
784 | xfs_lsn_t lsn; |
785 | int error; | |
1da177e4 | 786 | |
1da177e4 LT |
787 | /* |
788 | * Don't write out unmount record on read-only mounts. | |
789 | * Or, if we are doing a forced umount (typically because of IO errors). | |
790 | */ | |
bd186aa9 | 791 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
792 | return 0; |
793 | ||
a14a348b | 794 | error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
b911ca04 | 795 | ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log))); |
1da177e4 LT |
796 | |
797 | #ifdef DEBUG | |
798 | first_iclog = iclog = log->l_iclog; | |
799 | do { | |
800 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
801 | ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE); | |
802 | ASSERT(iclog->ic_offset == 0); | |
803 | } | |
804 | iclog = iclog->ic_next; | |
805 | } while (iclog != first_iclog); | |
806 | #endif | |
807 | if (! (XLOG_FORCED_SHUTDOWN(log))) { | |
955e47ad TS |
808 | error = xfs_log_reserve(mp, 600, 1, &tic, |
809 | XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE); | |
1da177e4 | 810 | if (!error) { |
55b66332 DC |
811 | /* the data section must be 32 bit size aligned */ |
812 | struct { | |
813 | __uint16_t magic; | |
814 | __uint16_t pad1; | |
815 | __uint32_t pad2; /* may as well make it 64 bits */ | |
816 | } magic = { | |
817 | .magic = XLOG_UNMOUNT_TYPE, | |
818 | }; | |
819 | struct xfs_log_iovec reg = { | |
4e0d5f92 | 820 | .i_addr = &magic, |
55b66332 DC |
821 | .i_len = sizeof(magic), |
822 | .i_type = XLOG_REG_TYPE_UNMOUNT, | |
823 | }; | |
824 | struct xfs_log_vec vec = { | |
825 | .lv_niovecs = 1, | |
826 | .lv_iovecp = ®, | |
827 | }; | |
828 | ||
3948659e | 829 | /* remove inited flag, and account for space used */ |
55b66332 | 830 | tic->t_flags = 0; |
3948659e | 831 | tic->t_curr_res -= sizeof(magic); |
55b66332 | 832 | error = xlog_write(log, &vec, tic, &lsn, |
1da177e4 LT |
833 | NULL, XLOG_UNMOUNT_TRANS); |
834 | /* | |
835 | * At this point, we're umounting anyway, | |
836 | * so there's no point in transitioning log state | |
837 | * to IOERROR. Just continue... | |
838 | */ | |
839 | } | |
840 | ||
a0fa2b67 DC |
841 | if (error) |
842 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
1da177e4 LT |
843 | |
844 | ||
b22cd72c | 845 | spin_lock(&log->l_icloglock); |
1da177e4 | 846 | iclog = log->l_iclog; |
155cc6b7 | 847 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 848 | xlog_state_want_sync(log, iclog); |
39e2defe | 849 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 850 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 851 | |
b22cd72c | 852 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
853 | if (!(iclog->ic_state == XLOG_STATE_ACTIVE || |
854 | iclog->ic_state == XLOG_STATE_DIRTY)) { | |
855 | if (!XLOG_FORCED_SHUTDOWN(log)) { | |
eb40a875 DC |
856 | xlog_wait(&iclog->ic_force_wait, |
857 | &log->l_icloglock); | |
1da177e4 | 858 | } else { |
b22cd72c | 859 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
860 | } |
861 | } else { | |
b22cd72c | 862 | spin_unlock(&log->l_icloglock); |
1da177e4 | 863 | } |
955e47ad | 864 | if (tic) { |
0b1b213f | 865 | trace_xfs_log_umount_write(log, tic); |
955e47ad | 866 | xlog_ungrant_log_space(log, tic); |
cc09c0dc | 867 | xfs_log_ticket_put(tic); |
955e47ad | 868 | } |
1da177e4 LT |
869 | } else { |
870 | /* | |
871 | * We're already in forced_shutdown mode, couldn't | |
872 | * even attempt to write out the unmount transaction. | |
873 | * | |
874 | * Go through the motions of sync'ing and releasing | |
875 | * the iclog, even though no I/O will actually happen, | |
c41564b5 | 876 | * we need to wait for other log I/Os that may already |
1da177e4 LT |
877 | * be in progress. Do this as a separate section of |
878 | * code so we'll know if we ever get stuck here that | |
879 | * we're in this odd situation of trying to unmount | |
880 | * a file system that went into forced_shutdown as | |
881 | * the result of an unmount.. | |
882 | */ | |
b22cd72c | 883 | spin_lock(&log->l_icloglock); |
1da177e4 | 884 | iclog = log->l_iclog; |
155cc6b7 | 885 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 LT |
886 | |
887 | xlog_state_want_sync(log, iclog); | |
39e2defe | 888 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 889 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 890 | |
b22cd72c | 891 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
892 | |
893 | if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE | |
894 | || iclog->ic_state == XLOG_STATE_DIRTY | |
895 | || iclog->ic_state == XLOG_STATE_IOERROR) ) { | |
896 | ||
eb40a875 DC |
897 | xlog_wait(&iclog->ic_force_wait, |
898 | &log->l_icloglock); | |
1da177e4 | 899 | } else { |
b22cd72c | 900 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
901 | } |
902 | } | |
903 | ||
1bb7d6b5 | 904 | return error; |
1da177e4 LT |
905 | } /* xfs_log_unmount_write */ |
906 | ||
907 | /* | |
c75921a7 | 908 | * Empty the log for unmount/freeze. |
cf2931db DC |
909 | * |
910 | * To do this, we first need to shut down the background log work so it is not | |
911 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
912 | * the log so we can then flush them out. Once they have completed their IO and | |
913 | * run the callbacks removing themselves from the AIL, we can write the unmount | |
c75921a7 | 914 | * record. |
1da177e4 LT |
915 | */ |
916 | void | |
c75921a7 DC |
917 | xfs_log_quiesce( |
918 | struct xfs_mount *mp) | |
1da177e4 | 919 | { |
f661f1e0 | 920 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
921 | xfs_log_force(mp, XFS_LOG_SYNC); |
922 | ||
923 | /* | |
924 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
925 | * will push it, xfs_wait_buftarg() will not wait for it. Further, | |
926 | * xfs_buf_iowait() cannot be used because it was pushed with the | |
927 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
928 | * the IO to complete. | |
929 | */ | |
930 | xfs_ail_push_all_sync(mp->m_ail); | |
931 | xfs_wait_buftarg(mp->m_ddev_targp); | |
932 | xfs_buf_lock(mp->m_sb_bp); | |
933 | xfs_buf_unlock(mp->m_sb_bp); | |
934 | ||
935 | xfs_log_unmount_write(mp); | |
c75921a7 DC |
936 | } |
937 | ||
938 | /* | |
939 | * Shut down and release the AIL and Log. | |
940 | * | |
941 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
942 | * from the AIL so that the log is empty before we write the unmount record to | |
943 | * the log. Once this is done, we can tear down the AIL and the log. | |
944 | */ | |
945 | void | |
946 | xfs_log_unmount( | |
947 | struct xfs_mount *mp) | |
948 | { | |
949 | xfs_log_quiesce(mp); | |
cf2931db | 950 | |
249a8c11 | 951 | xfs_trans_ail_destroy(mp); |
c41564b5 | 952 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
953 | } |
954 | ||
43f5efc5 DC |
955 | void |
956 | xfs_log_item_init( | |
957 | struct xfs_mount *mp, | |
958 | struct xfs_log_item *item, | |
959 | int type, | |
272e42b2 | 960 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
961 | { |
962 | item->li_mountp = mp; | |
963 | item->li_ailp = mp->m_ail; | |
964 | item->li_type = type; | |
965 | item->li_ops = ops; | |
71e330b5 DC |
966 | item->li_lv = NULL; |
967 | ||
968 | INIT_LIST_HEAD(&item->li_ail); | |
969 | INIT_LIST_HEAD(&item->li_cil); | |
43f5efc5 DC |
970 | } |
971 | ||
09a423a3 CH |
972 | /* |
973 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 974 | */ |
1da177e4 | 975 | void |
09a423a3 | 976 | xfs_log_space_wake( |
cfb7cdca | 977 | struct xfs_mount *mp) |
1da177e4 | 978 | { |
ad223e60 | 979 | struct xlog *log = mp->m_log; |
cfb7cdca | 980 | int free_bytes; |
1da177e4 | 981 | |
1da177e4 LT |
982 | if (XLOG_FORCED_SHUTDOWN(log)) |
983 | return; | |
1da177e4 | 984 | |
28496968 | 985 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
986 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
987 | ||
28496968 CH |
988 | spin_lock(&log->l_write_head.lock); |
989 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 990 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 991 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 992 | } |
10547941 | 993 | |
28496968 | 994 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
995 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
996 | ||
28496968 CH |
997 | spin_lock(&log->l_reserve_head.lock); |
998 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 999 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 1000 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 1001 | } |
3f16b985 | 1002 | } |
1da177e4 LT |
1003 | |
1004 | /* | |
2c6e24ce DC |
1005 | * Determine if we have a transaction that has gone to disk that needs to be |
1006 | * covered. To begin the transition to the idle state firstly the log needs to | |
1007 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1008 | * we start attempting to cover the log. | |
b6f8dd49 | 1009 | * |
2c6e24ce DC |
1010 | * Only if we are then in a state where covering is needed, the caller is |
1011 | * informed that dummy transactions are required to move the log into the idle | |
1012 | * state. | |
1013 | * | |
1014 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1015 | * cover the log as we may be in a situation where there isn't log space | |
1016 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1017 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1018 | * there's no point in running a dummy transaction at this point because we | |
1019 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 LT |
1020 | */ |
1021 | int | |
1022 | xfs_log_need_covered(xfs_mount_t *mp) | |
1023 | { | |
9a8d2fdb | 1024 | struct xlog *log = mp->m_log; |
2c6e24ce | 1025 | int needed = 0; |
1da177e4 | 1026 | |
92821e2b | 1027 | if (!xfs_fs_writable(mp)) |
1da177e4 LT |
1028 | return 0; |
1029 | ||
2c6e24ce DC |
1030 | if (!xlog_cil_empty(log)) |
1031 | return 0; | |
1032 | ||
b22cd72c | 1033 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1034 | switch (log->l_covered_state) { |
1035 | case XLOG_STATE_COVER_DONE: | |
1036 | case XLOG_STATE_COVER_DONE2: | |
1037 | case XLOG_STATE_COVER_IDLE: | |
1038 | break; | |
1039 | case XLOG_STATE_COVER_NEED: | |
1040 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1041 | if (xfs_ail_min_lsn(log->l_ailp)) |
1042 | break; | |
1043 | if (!xlog_iclogs_empty(log)) | |
1044 | break; | |
1045 | ||
1046 | needed = 1; | |
1047 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) | |
1048 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1049 | else | |
1050 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1051 | break; | |
b6f8dd49 | 1052 | default: |
1da177e4 | 1053 | needed = 1; |
b6f8dd49 | 1054 | break; |
1da177e4 | 1055 | } |
b22cd72c | 1056 | spin_unlock(&log->l_icloglock); |
014c2544 | 1057 | return needed; |
1da177e4 LT |
1058 | } |
1059 | ||
09a423a3 | 1060 | /* |
1da177e4 LT |
1061 | * We may be holding the log iclog lock upon entering this routine. |
1062 | */ | |
1063 | xfs_lsn_t | |
1c304625 | 1064 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1065 | struct xfs_mount *mp) |
1da177e4 | 1066 | { |
ad223e60 | 1067 | struct xlog *log = mp->m_log; |
1c304625 CH |
1068 | struct xfs_log_item *lip; |
1069 | xfs_lsn_t tail_lsn; | |
1070 | ||
1071 | assert_spin_locked(&mp->m_ail->xa_lock); | |
1da177e4 | 1072 | |
09a423a3 CH |
1073 | /* |
1074 | * To make sure we always have a valid LSN for the log tail we keep | |
1075 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1076 | * and use that when the AIL was empty. |
09a423a3 | 1077 | */ |
1c304625 CH |
1078 | lip = xfs_ail_min(mp->m_ail); |
1079 | if (lip) | |
1080 | tail_lsn = lip->li_lsn; | |
1081 | else | |
84f3c683 | 1082 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
1c3cb9ec | 1083 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1084 | return tail_lsn; |
1c3cb9ec | 1085 | } |
1da177e4 | 1086 | |
1c304625 CH |
1087 | xfs_lsn_t |
1088 | xlog_assign_tail_lsn( | |
1089 | struct xfs_mount *mp) | |
1090 | { | |
1091 | xfs_lsn_t tail_lsn; | |
1092 | ||
1093 | spin_lock(&mp->m_ail->xa_lock); | |
1094 | tail_lsn = xlog_assign_tail_lsn_locked(mp); | |
1095 | spin_unlock(&mp->m_ail->xa_lock); | |
1096 | ||
1097 | return tail_lsn; | |
1098 | } | |
1099 | ||
1da177e4 LT |
1100 | /* |
1101 | * Return the space in the log between the tail and the head. The head | |
1102 | * is passed in the cycle/bytes formal parms. In the special case where | |
1103 | * the reserve head has wrapped passed the tail, this calculation is no | |
1104 | * longer valid. In this case, just return 0 which means there is no space | |
1105 | * in the log. This works for all places where this function is called | |
1106 | * with the reserve head. Of course, if the write head were to ever | |
1107 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1108 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1109 | * | |
1110 | * This code also handles the case where the reservation head is behind | |
1111 | * the tail. The details of this case are described below, but the end | |
1112 | * result is that we return the size of the log as the amount of space left. | |
1113 | */ | |
a8272ce0 | 1114 | STATIC int |
a69ed03c | 1115 | xlog_space_left( |
ad223e60 | 1116 | struct xlog *log, |
c8a09ff8 | 1117 | atomic64_t *head) |
1da177e4 | 1118 | { |
a69ed03c DC |
1119 | int free_bytes; |
1120 | int tail_bytes; | |
1121 | int tail_cycle; | |
1122 | int head_cycle; | |
1123 | int head_bytes; | |
1da177e4 | 1124 | |
a69ed03c | 1125 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1126 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1127 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1128 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1129 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1130 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1131 | return 0; |
a69ed03c DC |
1132 | else if (tail_cycle < head_cycle) { |
1133 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1134 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1135 | } else { |
1136 | /* | |
1137 | * The reservation head is behind the tail. | |
1138 | * In this case we just want to return the size of the | |
1139 | * log as the amount of space left. | |
1140 | */ | |
a0fa2b67 | 1141 | xfs_alert(log->l_mp, |
1da177e4 LT |
1142 | "xlog_space_left: head behind tail\n" |
1143 | " tail_cycle = %d, tail_bytes = %d\n" | |
1144 | " GH cycle = %d, GH bytes = %d", | |
a69ed03c | 1145 | tail_cycle, tail_bytes, head_cycle, head_bytes); |
1da177e4 LT |
1146 | ASSERT(0); |
1147 | free_bytes = log->l_logsize; | |
1148 | } | |
1149 | return free_bytes; | |
a69ed03c | 1150 | } |
1da177e4 LT |
1151 | |
1152 | ||
1153 | /* | |
1154 | * Log function which is called when an io completes. | |
1155 | * | |
1156 | * The log manager needs its own routine, in order to control what | |
1157 | * happens with the buffer after the write completes. | |
1158 | */ | |
1159 | void | |
1160 | xlog_iodone(xfs_buf_t *bp) | |
1161 | { | |
9a8d2fdb MT |
1162 | struct xlog_in_core *iclog = bp->b_fspriv; |
1163 | struct xlog *l = iclog->ic_log; | |
1164 | int aborted = 0; | |
1da177e4 LT |
1165 | |
1166 | /* | |
1167 | * Race to shutdown the filesystem if we see an error. | |
1168 | */ | |
5a52c2a5 | 1169 | if (XFS_TEST_ERROR((xfs_buf_geterror(bp)), l->l_mp, |
1da177e4 | 1170 | XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) { |
901796af | 1171 | xfs_buf_ioerror_alert(bp, __func__); |
c867cb61 | 1172 | xfs_buf_stale(bp); |
7d04a335 | 1173 | xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 LT |
1174 | /* |
1175 | * This flag will be propagated to the trans-committed | |
1176 | * callback routines to let them know that the log-commit | |
1177 | * didn't succeed. | |
1178 | */ | |
1179 | aborted = XFS_LI_ABORTED; | |
1180 | } else if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
1181 | aborted = XFS_LI_ABORTED; | |
1182 | } | |
3db296f3 DC |
1183 | |
1184 | /* log I/O is always issued ASYNC */ | |
1185 | ASSERT(XFS_BUF_ISASYNC(bp)); | |
1da177e4 | 1186 | xlog_state_done_syncing(iclog, aborted); |
3db296f3 DC |
1187 | /* |
1188 | * do not reference the buffer (bp) here as we could race | |
1189 | * with it being freed after writing the unmount record to the | |
1190 | * log. | |
1191 | */ | |
c3f8fc73 | 1192 | } |
1da177e4 | 1193 | |
1da177e4 LT |
1194 | /* |
1195 | * Return size of each in-core log record buffer. | |
1196 | * | |
9da096fd | 1197 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1198 | * |
1199 | * If the filesystem blocksize is too large, we may need to choose a | |
1200 | * larger size since the directory code currently logs entire blocks. | |
1201 | */ | |
1202 | ||
1203 | STATIC void | |
9a8d2fdb MT |
1204 | xlog_get_iclog_buffer_size( |
1205 | struct xfs_mount *mp, | |
1206 | struct xlog *log) | |
1da177e4 LT |
1207 | { |
1208 | int size; | |
1209 | int xhdrs; | |
1210 | ||
1cb51258 ES |
1211 | if (mp->m_logbufs <= 0) |
1212 | log->l_iclog_bufs = XLOG_MAX_ICLOGS; | |
1213 | else | |
cfcbbbd0 | 1214 | log->l_iclog_bufs = mp->m_logbufs; |
1da177e4 LT |
1215 | |
1216 | /* | |
1217 | * Buffer size passed in from mount system call. | |
1218 | */ | |
cfcbbbd0 | 1219 | if (mp->m_logbsize > 0) { |
1da177e4 LT |
1220 | size = log->l_iclog_size = mp->m_logbsize; |
1221 | log->l_iclog_size_log = 0; | |
1222 | while (size != 1) { | |
1223 | log->l_iclog_size_log++; | |
1224 | size >>= 1; | |
1225 | } | |
1226 | ||
62118709 | 1227 | if (xfs_sb_version_haslogv2(&mp->m_sb)) { |
9da096fd MP |
1228 | /* # headers = size / 32k |
1229 | * one header holds cycles from 32k of data | |
1da177e4 LT |
1230 | */ |
1231 | ||
1232 | xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE; | |
1233 | if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE) | |
1234 | xhdrs++; | |
1235 | log->l_iclog_hsize = xhdrs << BBSHIFT; | |
1236 | log->l_iclog_heads = xhdrs; | |
1237 | } else { | |
1238 | ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE); | |
1239 | log->l_iclog_hsize = BBSIZE; | |
1240 | log->l_iclog_heads = 1; | |
1241 | } | |
cfcbbbd0 | 1242 | goto done; |
1da177e4 LT |
1243 | } |
1244 | ||
9da096fd | 1245 | /* All machines use 32kB buffers by default. */ |
1cb51258 ES |
1246 | log->l_iclog_size = XLOG_BIG_RECORD_BSIZE; |
1247 | log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT; | |
1da177e4 LT |
1248 | |
1249 | /* the default log size is 16k or 32k which is one header sector */ | |
1250 | log->l_iclog_hsize = BBSIZE; | |
1251 | log->l_iclog_heads = 1; | |
1252 | ||
7153f8ba CH |
1253 | done: |
1254 | /* are we being asked to make the sizes selected above visible? */ | |
cfcbbbd0 NS |
1255 | if (mp->m_logbufs == 0) |
1256 | mp->m_logbufs = log->l_iclog_bufs; | |
1257 | if (mp->m_logbsize == 0) | |
1258 | mp->m_logbsize = log->l_iclog_size; | |
1da177e4 LT |
1259 | } /* xlog_get_iclog_buffer_size */ |
1260 | ||
1261 | ||
f661f1e0 DC |
1262 | void |
1263 | xfs_log_work_queue( | |
1264 | struct xfs_mount *mp) | |
1265 | { | |
5889608d | 1266 | queue_delayed_work(mp->m_log_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1267 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * Every sync period we need to unpin all items in the AIL and push them to | |
1272 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1273 | * indicate that the filesystem is idle. | |
1274 | */ | |
1275 | void | |
1276 | xfs_log_worker( | |
1277 | struct work_struct *work) | |
1278 | { | |
1279 | struct xlog *log = container_of(to_delayed_work(work), | |
1280 | struct xlog, l_work); | |
1281 | struct xfs_mount *mp = log->l_mp; | |
1282 | ||
1283 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
1284 | if (xfs_log_need_covered(mp)) | |
1285 | xfs_fs_log_dummy(mp); | |
1286 | else | |
1287 | xfs_log_force(mp, 0); | |
1288 | ||
1289 | /* start pushing all the metadata that is currently dirty */ | |
1290 | xfs_ail_push_all(mp->m_ail); | |
1291 | ||
1292 | /* queue us up again */ | |
1293 | xfs_log_work_queue(mp); | |
1294 | } | |
1295 | ||
1da177e4 LT |
1296 | /* |
1297 | * This routine initializes some of the log structure for a given mount point. | |
1298 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1299 | * some other stuff may be filled in too. | |
1300 | */ | |
9a8d2fdb MT |
1301 | STATIC struct xlog * |
1302 | xlog_alloc_log( | |
1303 | struct xfs_mount *mp, | |
1304 | struct xfs_buftarg *log_target, | |
1305 | xfs_daddr_t blk_offset, | |
1306 | int num_bblks) | |
1da177e4 | 1307 | { |
9a8d2fdb | 1308 | struct xlog *log; |
1da177e4 LT |
1309 | xlog_rec_header_t *head; |
1310 | xlog_in_core_t **iclogp; | |
1311 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1312 | xfs_buf_t *bp; | |
1313 | int i; | |
a6cb767e | 1314 | int error = ENOMEM; |
69ce58f0 | 1315 | uint log2_size = 0; |
1da177e4 | 1316 | |
9a8d2fdb | 1317 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1318 | if (!log) { |
a0fa2b67 | 1319 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1320 | goto out; |
1321 | } | |
1da177e4 LT |
1322 | |
1323 | log->l_mp = mp; | |
1324 | log->l_targ = log_target; | |
1325 | log->l_logsize = BBTOB(num_bblks); | |
1326 | log->l_logBBstart = blk_offset; | |
1327 | log->l_logBBsize = num_bblks; | |
1328 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1329 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1330 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1331 | |
1332 | log->l_prev_block = -1; | |
1da177e4 | 1333 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1334 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1335 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1336 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 CH |
1337 | |
1338 | xlog_grant_head_init(&log->l_reserve_head); | |
1339 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1340 | |
a6cb767e | 1341 | error = EFSCORRUPTED; |
62118709 | 1342 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1343 | log2_size = mp->m_sb.sb_logsectlog; |
1344 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1345 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1346 | log2_size, BBSHIFT); | |
a6cb767e DC |
1347 | goto out_free_log; |
1348 | } | |
1349 | ||
69ce58f0 AE |
1350 | log2_size -= BBSHIFT; |
1351 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1352 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1353 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1354 | goto out_free_log; |
1355 | } | |
69ce58f0 AE |
1356 | |
1357 | /* for larger sector sizes, must have v2 or external log */ | |
1358 | if (log2_size && log->l_logBBstart > 0 && | |
1359 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1360 | xfs_warn(mp, |
1361 | "log sector size (0x%x) invalid for configuration.", | |
1362 | log2_size); | |
a6cb767e DC |
1363 | goto out_free_log; |
1364 | } | |
1da177e4 | 1365 | } |
69ce58f0 | 1366 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1367 | |
1368 | xlog_get_iclog_buffer_size(mp, log); | |
1369 | ||
a6cb767e | 1370 | error = ENOMEM; |
e70b73f8 | 1371 | bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0); |
644c3567 DC |
1372 | if (!bp) |
1373 | goto out_free_log; | |
cb669ca5 | 1374 | bp->b_iodone = xlog_iodone; |
0c842ad4 | 1375 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
1376 | log->l_xbuf = bp; |
1377 | ||
007c61c6 | 1378 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1379 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1380 | |
1da177e4 LT |
1381 | iclogp = &log->l_iclog; |
1382 | /* | |
1383 | * The amount of memory to allocate for the iclog structure is | |
1384 | * rather funky due to the way the structure is defined. It is | |
1385 | * done this way so that we can use different sizes for machines | |
1386 | * with different amounts of memory. See the definition of | |
1387 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1388 | */ | |
1da177e4 LT |
1389 | ASSERT(log->l_iclog_size >= 4096); |
1390 | for (i=0; i < log->l_iclog_bufs; i++) { | |
644c3567 DC |
1391 | *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL); |
1392 | if (!*iclogp) | |
1393 | goto out_free_iclog; | |
1394 | ||
1da177e4 | 1395 | iclog = *iclogp; |
1da177e4 LT |
1396 | iclog->ic_prev = prev_iclog; |
1397 | prev_iclog = iclog; | |
1fa40b01 | 1398 | |
686865f7 | 1399 | bp = xfs_buf_get_uncached(mp->m_logdev_targp, |
e70b73f8 | 1400 | BTOBB(log->l_iclog_size), 0); |
644c3567 DC |
1401 | if (!bp) |
1402 | goto out_free_iclog; | |
c8da0faf | 1403 | |
cb669ca5 | 1404 | bp->b_iodone = xlog_iodone; |
1fa40b01 | 1405 | iclog->ic_bp = bp; |
b28708d6 | 1406 | iclog->ic_data = bp->b_addr; |
4679b2d3 | 1407 | #ifdef DEBUG |
1da177e4 | 1408 | log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header); |
4679b2d3 | 1409 | #endif |
1da177e4 LT |
1410 | head = &iclog->ic_header; |
1411 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1412 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1413 | head->h_version = cpu_to_be32( | |
62118709 | 1414 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1415 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1416 | /* new fields */ |
b53e675d | 1417 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1418 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1419 | ||
4e94b71b | 1420 | iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize; |
1da177e4 LT |
1421 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1422 | iclog->ic_log = log; | |
114d23aa DC |
1423 | atomic_set(&iclog->ic_refcnt, 0); |
1424 | spin_lock_init(&iclog->ic_callback_lock); | |
1da177e4 | 1425 | iclog->ic_callback_tail = &(iclog->ic_callback); |
b28708d6 | 1426 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1427 | |
0c842ad4 | 1428 | ASSERT(xfs_buf_islocked(iclog->ic_bp)); |
eb40a875 DC |
1429 | init_waitqueue_head(&iclog->ic_force_wait); |
1430 | init_waitqueue_head(&iclog->ic_write_wait); | |
1da177e4 LT |
1431 | |
1432 | iclogp = &iclog->ic_next; | |
1433 | } | |
1434 | *iclogp = log->l_iclog; /* complete ring */ | |
1435 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1436 | ||
71e330b5 DC |
1437 | error = xlog_cil_init(log); |
1438 | if (error) | |
1439 | goto out_free_iclog; | |
1da177e4 | 1440 | return log; |
644c3567 DC |
1441 | |
1442 | out_free_iclog: | |
1443 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1444 | prev_iclog = iclog->ic_next; | |
eb40a875 | 1445 | if (iclog->ic_bp) |
644c3567 | 1446 | xfs_buf_free(iclog->ic_bp); |
644c3567 DC |
1447 | kmem_free(iclog); |
1448 | } | |
1449 | spinlock_destroy(&log->l_icloglock); | |
644c3567 DC |
1450 | xfs_buf_free(log->l_xbuf); |
1451 | out_free_log: | |
1452 | kmem_free(log); | |
a6cb767e DC |
1453 | out: |
1454 | return ERR_PTR(-error); | |
1da177e4 LT |
1455 | } /* xlog_alloc_log */ |
1456 | ||
1457 | ||
1458 | /* | |
1459 | * Write out the commit record of a transaction associated with the given | |
1460 | * ticket. Return the lsn of the commit record. | |
1461 | */ | |
1462 | STATIC int | |
55b66332 | 1463 | xlog_commit_record( |
ad223e60 | 1464 | struct xlog *log, |
55b66332 DC |
1465 | struct xlog_ticket *ticket, |
1466 | struct xlog_in_core **iclog, | |
1467 | xfs_lsn_t *commitlsnp) | |
1da177e4 | 1468 | { |
55b66332 DC |
1469 | struct xfs_mount *mp = log->l_mp; |
1470 | int error; | |
1471 | struct xfs_log_iovec reg = { | |
1472 | .i_addr = NULL, | |
1473 | .i_len = 0, | |
1474 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1475 | }; | |
1476 | struct xfs_log_vec vec = { | |
1477 | .lv_niovecs = 1, | |
1478 | .lv_iovecp = ®, | |
1479 | }; | |
1da177e4 LT |
1480 | |
1481 | ASSERT_ALWAYS(iclog); | |
55b66332 DC |
1482 | error = xlog_write(log, &vec, ticket, commitlsnp, iclog, |
1483 | XLOG_COMMIT_TRANS); | |
1484 | if (error) | |
7d04a335 | 1485 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1486 | return error; |
55b66332 | 1487 | } |
1da177e4 LT |
1488 | |
1489 | /* | |
1490 | * Push on the buffer cache code if we ever use more than 75% of the on-disk | |
1491 | * log space. This code pushes on the lsn which would supposedly free up | |
1492 | * the 25% which we want to leave free. We may need to adopt a policy which | |
1493 | * pushes on an lsn which is further along in the log once we reach the high | |
1494 | * water mark. In this manner, we would be creating a low water mark. | |
1495 | */ | |
a8272ce0 | 1496 | STATIC void |
2ced19cb | 1497 | xlog_grant_push_ail( |
ad223e60 | 1498 | struct xlog *log, |
2ced19cb | 1499 | int need_bytes) |
1da177e4 | 1500 | { |
2ced19cb | 1501 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1502 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1503 | int free_blocks; |
1504 | int free_bytes; | |
1505 | int threshold_block; | |
1506 | int threshold_cycle; | |
1507 | int free_threshold; | |
1508 | ||
1509 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1510 | ||
28496968 | 1511 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1512 | free_blocks = BTOBBT(free_bytes); |
1513 | ||
1514 | /* | |
1515 | * Set the threshold for the minimum number of free blocks in the | |
1516 | * log to the maximum of what the caller needs, one quarter of the | |
1517 | * log, and 256 blocks. | |
1518 | */ | |
1519 | free_threshold = BTOBB(need_bytes); | |
1520 | free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2)); | |
1521 | free_threshold = MAX(free_threshold, 256); | |
1522 | if (free_blocks >= free_threshold) | |
1523 | return; | |
1524 | ||
1c3cb9ec DC |
1525 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1526 | &threshold_block); | |
1527 | threshold_block += free_threshold; | |
1da177e4 | 1528 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1529 | threshold_block -= log->l_logBBsize; |
1530 | threshold_cycle += 1; | |
1da177e4 | 1531 | } |
2ced19cb DC |
1532 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1533 | threshold_block); | |
1534 | /* | |
1535 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1536 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1537 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1538 | */ |
84f3c683 DC |
1539 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1540 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1541 | threshold_lsn = last_sync_lsn; | |
2ced19cb DC |
1542 | |
1543 | /* | |
1544 | * Get the transaction layer to kick the dirty buffers out to | |
1545 | * disk asynchronously. No point in trying to do this if | |
1546 | * the filesystem is shutting down. | |
1547 | */ | |
1548 | if (!XLOG_FORCED_SHUTDOWN(log)) | |
fd074841 | 1549 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1550 | } |
1da177e4 | 1551 | |
0e446be4 CH |
1552 | /* |
1553 | * Stamp cycle number in every block | |
1554 | */ | |
1555 | STATIC void | |
1556 | xlog_pack_data( | |
1557 | struct xlog *log, | |
1558 | struct xlog_in_core *iclog, | |
1559 | int roundoff) | |
1560 | { | |
1561 | int i, j, k; | |
1562 | int size = iclog->ic_offset + roundoff; | |
1563 | __be32 cycle_lsn; | |
1564 | xfs_caddr_t dp; | |
1565 | ||
1566 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1567 | ||
1568 | dp = iclog->ic_datap; | |
1569 | for (i = 0; i < BTOBB(size); i++) { | |
1570 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1571 | break; | |
1572 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1573 | *(__be32 *)dp = cycle_lsn; | |
1574 | dp += BBSIZE; | |
1575 | } | |
1576 | ||
1577 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1578 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1579 | ||
1580 | for ( ; i < BTOBB(size); i++) { | |
1581 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1582 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1583 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1584 | *(__be32 *)dp = cycle_lsn; | |
1585 | dp += BBSIZE; | |
1586 | } | |
1587 | ||
1588 | for (i = 1; i < log->l_iclog_heads; i++) | |
1589 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1590 | } | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * Calculate the checksum for a log buffer. | |
1595 | * | |
1596 | * This is a little more complicated than it should be because the various | |
1597 | * headers and the actual data are non-contiguous. | |
1598 | */ | |
f9668a09 | 1599 | __le32 |
0e446be4 CH |
1600 | xlog_cksum( |
1601 | struct xlog *log, | |
1602 | struct xlog_rec_header *rhead, | |
1603 | char *dp, | |
1604 | int size) | |
1605 | { | |
1606 | __uint32_t crc; | |
1607 | ||
1608 | /* first generate the crc for the record header ... */ | |
1609 | crc = xfs_start_cksum((char *)rhead, | |
1610 | sizeof(struct xlog_rec_header), | |
1611 | offsetof(struct xlog_rec_header, h_crc)); | |
1612 | ||
1613 | /* ... then for additional cycle data for v2 logs ... */ | |
1614 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1615 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1616 | int i; | |
1617 | ||
1618 | for (i = 1; i < log->l_iclog_heads; i++) { | |
1619 | crc = crc32c(crc, &xhdr[i].hic_xheader, | |
1620 | sizeof(struct xlog_rec_ext_header)); | |
1621 | } | |
1622 | } | |
1623 | ||
1624 | /* ... and finally for the payload */ | |
1625 | crc = crc32c(crc, dp, size); | |
1626 | ||
1627 | return xfs_end_cksum(crc); | |
1628 | } | |
1629 | ||
873ff550 CH |
1630 | /* |
1631 | * The bdstrat callback function for log bufs. This gives us a central | |
1632 | * place to trap bufs in case we get hit by a log I/O error and need to | |
1633 | * shutdown. Actually, in practice, even when we didn't get a log error, | |
1634 | * we transition the iclogs to IOERROR state *after* flushing all existing | |
1635 | * iclogs to disk. This is because we don't want anymore new transactions to be | |
1636 | * started or completed afterwards. | |
1637 | */ | |
1638 | STATIC int | |
1639 | xlog_bdstrat( | |
1640 | struct xfs_buf *bp) | |
1641 | { | |
adadbeef | 1642 | struct xlog_in_core *iclog = bp->b_fspriv; |
873ff550 | 1643 | |
873ff550 | 1644 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
5a52c2a5 | 1645 | xfs_buf_ioerror(bp, EIO); |
c867cb61 | 1646 | xfs_buf_stale(bp); |
1a1a3e97 | 1647 | xfs_buf_ioend(bp, 0); |
873ff550 CH |
1648 | /* |
1649 | * It would seem logical to return EIO here, but we rely on | |
1650 | * the log state machine to propagate I/O errors instead of | |
1651 | * doing it here. | |
1652 | */ | |
1653 | return 0; | |
1654 | } | |
1655 | ||
873ff550 CH |
1656 | xfs_buf_iorequest(bp); |
1657 | return 0; | |
1658 | } | |
1da177e4 LT |
1659 | |
1660 | /* | |
1661 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1662 | * fashion. Previously, we should have moved the current iclog | |
1663 | * ptr in the log to point to the next available iclog. This allows further | |
1664 | * write to continue while this code syncs out an iclog ready to go. | |
1665 | * Before an in-core log can be written out, the data section must be scanned | |
1666 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1667 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1668 | * cycle count because there in an implicit assumption that drives will | |
1669 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1670 | * we can't have part of a 512 byte block written and part not written. By | |
1671 | * tagging each block, we will know which blocks are valid when recovering | |
1672 | * after an unclean shutdown. | |
1673 | * | |
1674 | * This routine is single threaded on the iclog. No other thread can be in | |
1675 | * this routine with the same iclog. Changing contents of iclog can there- | |
1676 | * fore be done without grabbing the state machine lock. Updating the global | |
1677 | * log will require grabbing the lock though. | |
1678 | * | |
1679 | * The entire log manager uses a logical block numbering scheme. Only | |
1680 | * log_sync (and then only bwrite()) know about the fact that the log may | |
1681 | * not start with block zero on a given device. The log block start offset | |
1682 | * is added immediately before calling bwrite(). | |
1683 | */ | |
1684 | ||
a8272ce0 | 1685 | STATIC int |
9a8d2fdb MT |
1686 | xlog_sync( |
1687 | struct xlog *log, | |
1688 | struct xlog_in_core *iclog) | |
1da177e4 | 1689 | { |
1da177e4 | 1690 | xfs_buf_t *bp; |
b53e675d | 1691 | int i; |
1da177e4 LT |
1692 | uint count; /* byte count of bwrite */ |
1693 | uint count_init; /* initial count before roundup */ | |
1694 | int roundoff; /* roundoff to BB or stripe */ | |
1695 | int split = 0; /* split write into two regions */ | |
1696 | int error; | |
62118709 | 1697 | int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb); |
0e446be4 | 1698 | int size; |
1da177e4 LT |
1699 | |
1700 | XFS_STATS_INC(xs_log_writes); | |
155cc6b7 | 1701 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
1702 | |
1703 | /* Add for LR header */ | |
1704 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
1705 | ||
1706 | /* Round out the log write size */ | |
1707 | if (v2 && log->l_mp->m_sb.sb_logsunit > 1) { | |
1708 | /* we have a v2 stripe unit to use */ | |
1709 | count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init)); | |
1710 | } else { | |
1711 | count = BBTOB(BTOBB(count_init)); | |
1712 | } | |
1713 | roundoff = count - count_init; | |
1714 | ASSERT(roundoff >= 0); | |
1715 | ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && | |
1716 | roundoff < log->l_mp->m_sb.sb_logsunit) | |
1717 | || | |
1718 | (log->l_mp->m_sb.sb_logsunit <= 1 && | |
1719 | roundoff < BBTOB(1))); | |
1720 | ||
1721 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1722 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1723 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1724 | |
1725 | /* put cycle number in every block */ | |
1726 | xlog_pack_data(log, iclog, roundoff); | |
1727 | ||
1728 | /* real byte length */ | |
0e446be4 CH |
1729 | size = iclog->ic_offset; |
1730 | if (v2) | |
1731 | size += roundoff; | |
1732 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 1733 | |
f5faad79 | 1734 | bp = iclog->ic_bp; |
b53e675d | 1735 | XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn))); |
1da177e4 LT |
1736 | |
1737 | XFS_STATS_ADD(xs_log_blocks, BTOBB(count)); | |
1738 | ||
1739 | /* Do we need to split this write into 2 parts? */ | |
1740 | if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) { | |
0e446be4 CH |
1741 | char *dptr; |
1742 | ||
1da177e4 LT |
1743 | split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp))); |
1744 | count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)); | |
0e446be4 CH |
1745 | iclog->ic_bwritecnt = 2; |
1746 | ||
1747 | /* | |
1748 | * Bump the cycle numbers at the start of each block in the | |
1749 | * part of the iclog that ends up in the buffer that gets | |
1750 | * written to the start of the log. | |
1751 | * | |
1752 | * Watch out for the header magic number case, though. | |
1753 | */ | |
1754 | dptr = (char *)&iclog->ic_header + count; | |
1755 | for (i = 0; i < split; i += BBSIZE) { | |
1756 | __uint32_t cycle = be32_to_cpu(*(__be32 *)dptr); | |
1757 | if (++cycle == XLOG_HEADER_MAGIC_NUM) | |
1758 | cycle++; | |
1759 | *(__be32 *)dptr = cpu_to_be32(cycle); | |
1760 | ||
1761 | dptr += BBSIZE; | |
1762 | } | |
1da177e4 LT |
1763 | } else { |
1764 | iclog->ic_bwritecnt = 1; | |
1765 | } | |
0e446be4 CH |
1766 | |
1767 | /* calculcate the checksum */ | |
1768 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
1769 | iclog->ic_datap, size); | |
1770 | ||
aa0e8833 | 1771 | bp->b_io_length = BTOBB(count); |
adadbeef | 1772 | bp->b_fspriv = iclog; |
f5faad79 | 1773 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 | 1774 | XFS_BUF_ASYNC(bp); |
1d5ae5df | 1775 | bp->b_flags |= XBF_SYNCIO; |
651701d7 | 1776 | |
a27a263b | 1777 | if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) { |
e163cbde CH |
1778 | bp->b_flags |= XBF_FUA; |
1779 | ||
a27a263b | 1780 | /* |
e163cbde CH |
1781 | * Flush the data device before flushing the log to make |
1782 | * sure all meta data written back from the AIL actually made | |
1783 | * it to disk before stamping the new log tail LSN into the | |
1784 | * log buffer. For an external log we need to issue the | |
1785 | * flush explicitly, and unfortunately synchronously here; | |
1786 | * for an internal log we can simply use the block layer | |
1787 | * state machine for preflushes. | |
a27a263b CH |
1788 | */ |
1789 | if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp) | |
1790 | xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp); | |
e163cbde CH |
1791 | else |
1792 | bp->b_flags |= XBF_FLUSH; | |
a27a263b | 1793 | } |
1da177e4 LT |
1794 | |
1795 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1796 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1797 | ||
003fd6c8 | 1798 | xlog_verify_iclog(log, iclog, count, true); |
1da177e4 LT |
1799 | |
1800 | /* account for log which doesn't start at block #0 */ | |
1801 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); | |
1802 | /* | |
1803 | * Don't call xfs_bwrite here. We do log-syncs even when the filesystem | |
1804 | * is shutting down. | |
1805 | */ | |
1806 | XFS_BUF_WRITE(bp); | |
1807 | ||
901796af CH |
1808 | error = xlog_bdstrat(bp); |
1809 | if (error) { | |
1810 | xfs_buf_ioerror_alert(bp, "xlog_sync"); | |
014c2544 | 1811 | return error; |
1da177e4 LT |
1812 | } |
1813 | if (split) { | |
f5faad79 | 1814 | bp = iclog->ic_log->l_xbuf; |
1da177e4 | 1815 | XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */ |
02fe03d9 CS |
1816 | xfs_buf_associate_memory(bp, |
1817 | (char *)&iclog->ic_header + count, split); | |
adadbeef | 1818 | bp->b_fspriv = iclog; |
f5faad79 | 1819 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 | 1820 | XFS_BUF_ASYNC(bp); |
1d5ae5df | 1821 | bp->b_flags |= XBF_SYNCIO; |
f538d4da | 1822 | if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) |
e163cbde | 1823 | bp->b_flags |= XBF_FUA; |
1da177e4 LT |
1824 | |
1825 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1826 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1827 | ||
c41564b5 | 1828 | /* account for internal log which doesn't start at block #0 */ |
1da177e4 LT |
1829 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); |
1830 | XFS_BUF_WRITE(bp); | |
901796af CH |
1831 | error = xlog_bdstrat(bp); |
1832 | if (error) { | |
1833 | xfs_buf_ioerror_alert(bp, "xlog_sync (split)"); | |
014c2544 | 1834 | return error; |
1da177e4 LT |
1835 | } |
1836 | } | |
014c2544 | 1837 | return 0; |
1da177e4 LT |
1838 | } /* xlog_sync */ |
1839 | ||
1da177e4 | 1840 | /* |
c41564b5 | 1841 | * Deallocate a log structure |
1da177e4 | 1842 | */ |
a8272ce0 | 1843 | STATIC void |
9a8d2fdb MT |
1844 | xlog_dealloc_log( |
1845 | struct xlog *log) | |
1da177e4 LT |
1846 | { |
1847 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1848 | int i; |
1849 | ||
71e330b5 DC |
1850 | xlog_cil_destroy(log); |
1851 | ||
44396476 DC |
1852 | /* |
1853 | * always need to ensure that the extra buffer does not point to memory | |
1854 | * owned by another log buffer before we free it. | |
1855 | */ | |
e70b73f8 | 1856 | xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size)); |
44396476 DC |
1857 | xfs_buf_free(log->l_xbuf); |
1858 | ||
1da177e4 LT |
1859 | iclog = log->l_iclog; |
1860 | for (i=0; i<log->l_iclog_bufs; i++) { | |
1da177e4 | 1861 | xfs_buf_free(iclog->ic_bp); |
1da177e4 | 1862 | next_iclog = iclog->ic_next; |
f0e2d93c | 1863 | kmem_free(iclog); |
1da177e4 LT |
1864 | iclog = next_iclog; |
1865 | } | |
1da177e4 | 1866 | spinlock_destroy(&log->l_icloglock); |
1da177e4 | 1867 | |
1da177e4 | 1868 | log->l_mp->m_log = NULL; |
f0e2d93c | 1869 | kmem_free(log); |
c41564b5 | 1870 | } /* xlog_dealloc_log */ |
1da177e4 LT |
1871 | |
1872 | /* | |
1873 | * Update counters atomically now that memcpy is done. | |
1874 | */ | |
1875 | /* ARGSUSED */ | |
1876 | static inline void | |
9a8d2fdb MT |
1877 | xlog_state_finish_copy( |
1878 | struct xlog *log, | |
1879 | struct xlog_in_core *iclog, | |
1880 | int record_cnt, | |
1881 | int copy_bytes) | |
1da177e4 | 1882 | { |
b22cd72c | 1883 | spin_lock(&log->l_icloglock); |
1da177e4 | 1884 | |
413d57c9 | 1885 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 LT |
1886 | iclog->ic_offset += copy_bytes; |
1887 | ||
b22cd72c | 1888 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
1889 | } /* xlog_state_finish_copy */ |
1890 | ||
1891 | ||
1892 | ||
1893 | ||
7e9c6396 TS |
1894 | /* |
1895 | * print out info relating to regions written which consume | |
1896 | * the reservation | |
1897 | */ | |
71e330b5 DC |
1898 | void |
1899 | xlog_print_tic_res( | |
1900 | struct xfs_mount *mp, | |
1901 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
1902 | { |
1903 | uint i; | |
1904 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
1905 | ||
1906 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
1907 | static char *res_type_str[XLOG_REG_TYPE_MAX] = { | |
1908 | "bformat", | |
1909 | "bchunk", | |
1910 | "efi_format", | |
1911 | "efd_format", | |
1912 | "iformat", | |
1913 | "icore", | |
1914 | "iext", | |
1915 | "ibroot", | |
1916 | "ilocal", | |
1917 | "iattr_ext", | |
1918 | "iattr_broot", | |
1919 | "iattr_local", | |
1920 | "qformat", | |
1921 | "dquot", | |
1922 | "quotaoff", | |
1923 | "LR header", | |
1924 | "unmount", | |
1925 | "commit", | |
1926 | "trans header" | |
1927 | }; | |
1928 | static char *trans_type_str[XFS_TRANS_TYPE_MAX] = { | |
1929 | "SETATTR_NOT_SIZE", | |
1930 | "SETATTR_SIZE", | |
1931 | "INACTIVE", | |
1932 | "CREATE", | |
1933 | "CREATE_TRUNC", | |
1934 | "TRUNCATE_FILE", | |
1935 | "REMOVE", | |
1936 | "LINK", | |
1937 | "RENAME", | |
1938 | "MKDIR", | |
1939 | "RMDIR", | |
1940 | "SYMLINK", | |
1941 | "SET_DMATTRS", | |
1942 | "GROWFS", | |
1943 | "STRAT_WRITE", | |
1944 | "DIOSTRAT", | |
1945 | "WRITE_SYNC", | |
1946 | "WRITEID", | |
1947 | "ADDAFORK", | |
1948 | "ATTRINVAL", | |
1949 | "ATRUNCATE", | |
1950 | "ATTR_SET", | |
1951 | "ATTR_RM", | |
1952 | "ATTR_FLAG", | |
1953 | "CLEAR_AGI_BUCKET", | |
1954 | "QM_SBCHANGE", | |
1955 | "DUMMY1", | |
1956 | "DUMMY2", | |
1957 | "QM_QUOTAOFF", | |
1958 | "QM_DQALLOC", | |
1959 | "QM_SETQLIM", | |
1960 | "QM_DQCLUSTER", | |
1961 | "QM_QINOCREATE", | |
1962 | "QM_QUOTAOFF_END", | |
1963 | "SB_UNIT", | |
1964 | "FSYNC_TS", | |
1965 | "GROWFSRT_ALLOC", | |
1966 | "GROWFSRT_ZERO", | |
1967 | "GROWFSRT_FREE", | |
1968 | "SWAPEXT" | |
1969 | }; | |
1970 | ||
a0fa2b67 | 1971 | xfs_warn(mp, |
93b8a585 | 1972 | "xlog_write: reservation summary:\n" |
a0fa2b67 DC |
1973 | " trans type = %s (%u)\n" |
1974 | " unit res = %d bytes\n" | |
1975 | " current res = %d bytes\n" | |
1976 | " total reg = %u bytes (o/flow = %u bytes)\n" | |
1977 | " ophdrs = %u (ophdr space = %u bytes)\n" | |
1978 | " ophdr + reg = %u bytes\n" | |
1979 | " num regions = %u\n", | |
1980 | ((ticket->t_trans_type <= 0 || | |
1981 | ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ? | |
1982 | "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]), | |
1983 | ticket->t_trans_type, | |
1984 | ticket->t_unit_res, | |
1985 | ticket->t_curr_res, | |
1986 | ticket->t_res_arr_sum, ticket->t_res_o_flow, | |
1987 | ticket->t_res_num_ophdrs, ophdr_spc, | |
1988 | ticket->t_res_arr_sum + | |
1989 | ticket->t_res_o_flow + ophdr_spc, | |
1990 | ticket->t_res_num); | |
7e9c6396 TS |
1991 | |
1992 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 1993 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 1994 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 TS |
1995 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
1996 | "bad-rtype" : res_type_str[r_type-1]), | |
1997 | ticket->t_res_arr[i].r_len); | |
1998 | } | |
169a7b07 | 1999 | |
a0fa2b67 | 2000 | xfs_alert_tag(mp, XFS_PTAG_LOGRES, |
93b8a585 | 2001 | "xlog_write: reservation ran out. Need to up reservation"); |
297aa637 | 2002 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
7e9c6396 | 2003 | } |
7e9c6396 | 2004 | |
b5203cd0 DC |
2005 | /* |
2006 | * Calculate the potential space needed by the log vector. Each region gets | |
2007 | * its own xlog_op_header_t and may need to be double word aligned. | |
2008 | */ | |
2009 | static int | |
2010 | xlog_write_calc_vec_length( | |
2011 | struct xlog_ticket *ticket, | |
55b66332 | 2012 | struct xfs_log_vec *log_vector) |
b5203cd0 | 2013 | { |
55b66332 | 2014 | struct xfs_log_vec *lv; |
b5203cd0 DC |
2015 | int headers = 0; |
2016 | int len = 0; | |
2017 | int i; | |
2018 | ||
2019 | /* acct for start rec of xact */ | |
2020 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2021 | headers++; | |
2022 | ||
55b66332 | 2023 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2024 | /* we don't write ordered log vectors */ |
2025 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2026 | continue; | |
2027 | ||
55b66332 DC |
2028 | headers += lv->lv_niovecs; |
2029 | ||
2030 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2031 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2032 | |
55b66332 DC |
2033 | len += vecp->i_len; |
2034 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2035 | } | |
b5203cd0 DC |
2036 | } |
2037 | ||
2038 | ticket->t_res_num_ophdrs += headers; | |
2039 | len += headers * sizeof(struct xlog_op_header); | |
2040 | ||
2041 | return len; | |
2042 | } | |
2043 | ||
2044 | /* | |
2045 | * If first write for transaction, insert start record We can't be trying to | |
2046 | * commit if we are inited. We can't have any "partial_copy" if we are inited. | |
2047 | */ | |
2048 | static int | |
2049 | xlog_write_start_rec( | |
e6b1f273 | 2050 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2051 | struct xlog_ticket *ticket) |
2052 | { | |
b5203cd0 DC |
2053 | if (!(ticket->t_flags & XLOG_TIC_INITED)) |
2054 | return 0; | |
2055 | ||
2056 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); | |
2057 | ophdr->oh_clientid = ticket->t_clientid; | |
2058 | ophdr->oh_len = 0; | |
2059 | ophdr->oh_flags = XLOG_START_TRANS; | |
2060 | ophdr->oh_res2 = 0; | |
2061 | ||
2062 | ticket->t_flags &= ~XLOG_TIC_INITED; | |
2063 | ||
2064 | return sizeof(struct xlog_op_header); | |
2065 | } | |
2066 | ||
2067 | static xlog_op_header_t * | |
2068 | xlog_write_setup_ophdr( | |
ad223e60 | 2069 | struct xlog *log, |
e6b1f273 | 2070 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2071 | struct xlog_ticket *ticket, |
2072 | uint flags) | |
2073 | { | |
b5203cd0 DC |
2074 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2075 | ophdr->oh_clientid = ticket->t_clientid; | |
2076 | ophdr->oh_res2 = 0; | |
2077 | ||
2078 | /* are we copying a commit or unmount record? */ | |
2079 | ophdr->oh_flags = flags; | |
2080 | ||
2081 | /* | |
2082 | * We've seen logs corrupted with bad transaction client ids. This | |
2083 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2084 | * and shut down the filesystem. | |
2085 | */ | |
2086 | switch (ophdr->oh_clientid) { | |
2087 | case XFS_TRANSACTION: | |
2088 | case XFS_VOLUME: | |
2089 | case XFS_LOG: | |
2090 | break; | |
2091 | default: | |
a0fa2b67 | 2092 | xfs_warn(log->l_mp, |
b5203cd0 DC |
2093 | "Bad XFS transaction clientid 0x%x in ticket 0x%p", |
2094 | ophdr->oh_clientid, ticket); | |
2095 | return NULL; | |
2096 | } | |
2097 | ||
2098 | return ophdr; | |
2099 | } | |
2100 | ||
2101 | /* | |
2102 | * Set up the parameters of the region copy into the log. This has | |
2103 | * to handle region write split across multiple log buffers - this | |
2104 | * state is kept external to this function so that this code can | |
ac0e300f | 2105 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2106 | */ |
2107 | static int | |
2108 | xlog_write_setup_copy( | |
2109 | struct xlog_ticket *ticket, | |
2110 | struct xlog_op_header *ophdr, | |
2111 | int space_available, | |
2112 | int space_required, | |
2113 | int *copy_off, | |
2114 | int *copy_len, | |
2115 | int *last_was_partial_copy, | |
2116 | int *bytes_consumed) | |
2117 | { | |
2118 | int still_to_copy; | |
2119 | ||
2120 | still_to_copy = space_required - *bytes_consumed; | |
2121 | *copy_off = *bytes_consumed; | |
2122 | ||
2123 | if (still_to_copy <= space_available) { | |
2124 | /* write of region completes here */ | |
2125 | *copy_len = still_to_copy; | |
2126 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2127 | if (*last_was_partial_copy) | |
2128 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2129 | *last_was_partial_copy = 0; | |
2130 | *bytes_consumed = 0; | |
2131 | return 0; | |
2132 | } | |
2133 | ||
2134 | /* partial write of region, needs extra log op header reservation */ | |
2135 | *copy_len = space_available; | |
2136 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2137 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2138 | if (*last_was_partial_copy) | |
2139 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2140 | *bytes_consumed += *copy_len; | |
2141 | (*last_was_partial_copy)++; | |
2142 | ||
2143 | /* account for new log op header */ | |
2144 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2145 | ticket->t_res_num_ophdrs++; | |
2146 | ||
2147 | return sizeof(struct xlog_op_header); | |
2148 | } | |
2149 | ||
2150 | static int | |
2151 | xlog_write_copy_finish( | |
ad223e60 | 2152 | struct xlog *log, |
b5203cd0 DC |
2153 | struct xlog_in_core *iclog, |
2154 | uint flags, | |
2155 | int *record_cnt, | |
2156 | int *data_cnt, | |
2157 | int *partial_copy, | |
2158 | int *partial_copy_len, | |
2159 | int log_offset, | |
2160 | struct xlog_in_core **commit_iclog) | |
2161 | { | |
2162 | if (*partial_copy) { | |
2163 | /* | |
2164 | * This iclog has already been marked WANT_SYNC by | |
2165 | * xlog_state_get_iclog_space. | |
2166 | */ | |
2167 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2168 | *record_cnt = 0; | |
2169 | *data_cnt = 0; | |
2170 | return xlog_state_release_iclog(log, iclog); | |
2171 | } | |
2172 | ||
2173 | *partial_copy = 0; | |
2174 | *partial_copy_len = 0; | |
2175 | ||
2176 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2177 | /* no more space in this iclog - push it. */ | |
2178 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2179 | *record_cnt = 0; | |
2180 | *data_cnt = 0; | |
2181 | ||
2182 | spin_lock(&log->l_icloglock); | |
2183 | xlog_state_want_sync(log, iclog); | |
2184 | spin_unlock(&log->l_icloglock); | |
2185 | ||
2186 | if (!commit_iclog) | |
2187 | return xlog_state_release_iclog(log, iclog); | |
2188 | ASSERT(flags & XLOG_COMMIT_TRANS); | |
2189 | *commit_iclog = iclog; | |
2190 | } | |
2191 | ||
2192 | return 0; | |
2193 | } | |
2194 | ||
1da177e4 LT |
2195 | /* |
2196 | * Write some region out to in-core log | |
2197 | * | |
2198 | * This will be called when writing externally provided regions or when | |
2199 | * writing out a commit record for a given transaction. | |
2200 | * | |
2201 | * General algorithm: | |
2202 | * 1. Find total length of this write. This may include adding to the | |
2203 | * lengths passed in. | |
2204 | * 2. Check whether we violate the tickets reservation. | |
2205 | * 3. While writing to this iclog | |
2206 | * A. Reserve as much space in this iclog as can get | |
2207 | * B. If this is first write, save away start lsn | |
2208 | * C. While writing this region: | |
2209 | * 1. If first write of transaction, write start record | |
2210 | * 2. Write log operation header (header per region) | |
2211 | * 3. Find out if we can fit entire region into this iclog | |
2212 | * 4. Potentially, verify destination memcpy ptr | |
2213 | * 5. Memcpy (partial) region | |
2214 | * 6. If partial copy, release iclog; otherwise, continue | |
2215 | * copying more regions into current iclog | |
2216 | * 4. Mark want sync bit (in simulation mode) | |
2217 | * 5. Release iclog for potential flush to on-disk log. | |
2218 | * | |
2219 | * ERRORS: | |
2220 | * 1. Panic if reservation is overrun. This should never happen since | |
2221 | * reservation amounts are generated internal to the filesystem. | |
2222 | * NOTES: | |
2223 | * 1. Tickets are single threaded data structures. | |
2224 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2225 | * syncing routine. When a single log_write region needs to span | |
2226 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2227 | * on all log operation writes which don't contain the end of the | |
2228 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2229 | * operation which contains the end of the continued log_write region. | |
2230 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2231 | * we don't really know exactly how much space will be used. As a result, | |
2232 | * we don't update ic_offset until the end when we know exactly how many | |
2233 | * bytes have been written out. | |
2234 | */ | |
71e330b5 | 2235 | int |
35a8a72f | 2236 | xlog_write( |
ad223e60 | 2237 | struct xlog *log, |
55b66332 | 2238 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2239 | struct xlog_ticket *ticket, |
2240 | xfs_lsn_t *start_lsn, | |
2241 | struct xlog_in_core **commit_iclog, | |
2242 | uint flags) | |
1da177e4 | 2243 | { |
99428ad0 | 2244 | struct xlog_in_core *iclog = NULL; |
55b66332 DC |
2245 | struct xfs_log_iovec *vecp; |
2246 | struct xfs_log_vec *lv; | |
99428ad0 CH |
2247 | int len; |
2248 | int index; | |
2249 | int partial_copy = 0; | |
2250 | int partial_copy_len = 0; | |
2251 | int contwr = 0; | |
2252 | int record_cnt = 0; | |
2253 | int data_cnt = 0; | |
2254 | int error; | |
2255 | ||
2256 | *start_lsn = 0; | |
2257 | ||
55b66332 | 2258 | len = xlog_write_calc_vec_length(ticket, log_vector); |
71e330b5 | 2259 | |
93b8a585 CH |
2260 | /* |
2261 | * Region headers and bytes are already accounted for. | |
2262 | * We only need to take into account start records and | |
2263 | * split regions in this function. | |
2264 | */ | |
2265 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2266 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
2267 | ||
2268 | /* | |
2269 | * Commit record headers need to be accounted for. These | |
2270 | * come in as separate writes so are easy to detect. | |
2271 | */ | |
2272 | if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS)) | |
2273 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
71e330b5 DC |
2274 | |
2275 | if (ticket->t_curr_res < 0) | |
55b66332 | 2276 | xlog_print_tic_res(log->l_mp, ticket); |
1da177e4 | 2277 | |
55b66332 DC |
2278 | index = 0; |
2279 | lv = log_vector; | |
2280 | vecp = lv->lv_iovecp; | |
fd63875c | 2281 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2282 | void *ptr; |
99428ad0 | 2283 | int log_offset; |
1da177e4 | 2284 | |
99428ad0 CH |
2285 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2286 | &contwr, &log_offset); | |
2287 | if (error) | |
2288 | return error; | |
1da177e4 | 2289 | |
99428ad0 | 2290 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2291 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2292 | |
99428ad0 CH |
2293 | /* start_lsn is the first lsn written to. That's all we need. */ |
2294 | if (!*start_lsn) | |
2295 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
b5203cd0 | 2296 | |
99428ad0 CH |
2297 | /* |
2298 | * This loop writes out as many regions as can fit in the amount | |
2299 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2300 | */ | |
fd63875c DC |
2301 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2302 | struct xfs_log_iovec *reg; | |
99428ad0 CH |
2303 | struct xlog_op_header *ophdr; |
2304 | int start_rec_copy; | |
2305 | int copy_len; | |
2306 | int copy_off; | |
fd63875c DC |
2307 | bool ordered = false; |
2308 | ||
2309 | /* ordered log vectors have no regions to write */ | |
2310 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2311 | ASSERT(lv->lv_niovecs == 0); | |
2312 | ordered = true; | |
2313 | goto next_lv; | |
2314 | } | |
99428ad0 | 2315 | |
fd63875c | 2316 | reg = &vecp[index]; |
55b66332 | 2317 | ASSERT(reg->i_len % sizeof(__int32_t) == 0); |
e6b1f273 | 2318 | ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0); |
99428ad0 CH |
2319 | |
2320 | start_rec_copy = xlog_write_start_rec(ptr, ticket); | |
2321 | if (start_rec_copy) { | |
2322 | record_cnt++; | |
e6b1f273 | 2323 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2324 | start_rec_copy); |
2325 | } | |
b5203cd0 | 2326 | |
99428ad0 CH |
2327 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags); |
2328 | if (!ophdr) | |
2329 | return XFS_ERROR(EIO); | |
2330 | ||
e6b1f273 | 2331 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2332 | sizeof(struct xlog_op_header)); |
2333 | ||
2334 | len += xlog_write_setup_copy(ticket, ophdr, | |
2335 | iclog->ic_size-log_offset, | |
55b66332 | 2336 | reg->i_len, |
99428ad0 CH |
2337 | ©_off, ©_len, |
2338 | &partial_copy, | |
2339 | &partial_copy_len); | |
2340 | xlog_verify_dest_ptr(log, ptr); | |
2341 | ||
2342 | /* copy region */ | |
2343 | ASSERT(copy_len >= 0); | |
e6b1f273 CH |
2344 | memcpy(ptr, reg->i_addr + copy_off, copy_len); |
2345 | xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len); | |
99428ad0 CH |
2346 | |
2347 | copy_len += start_rec_copy + sizeof(xlog_op_header_t); | |
2348 | record_cnt++; | |
2349 | data_cnt += contwr ? copy_len : 0; | |
2350 | ||
2351 | error = xlog_write_copy_finish(log, iclog, flags, | |
2352 | &record_cnt, &data_cnt, | |
2353 | &partial_copy, | |
2354 | &partial_copy_len, | |
2355 | log_offset, | |
2356 | commit_iclog); | |
2357 | if (error) | |
2358 | return error; | |
2359 | ||
2360 | /* | |
2361 | * if we had a partial copy, we need to get more iclog | |
2362 | * space but we don't want to increment the region | |
2363 | * index because there is still more is this region to | |
2364 | * write. | |
2365 | * | |
2366 | * If we completed writing this region, and we flushed | |
2367 | * the iclog (indicated by resetting of the record | |
2368 | * count), then we also need to get more log space. If | |
2369 | * this was the last record, though, we are done and | |
2370 | * can just return. | |
2371 | */ | |
2372 | if (partial_copy) | |
2373 | break; | |
2374 | ||
55b66332 | 2375 | if (++index == lv->lv_niovecs) { |
fd63875c | 2376 | next_lv: |
55b66332 DC |
2377 | lv = lv->lv_next; |
2378 | index = 0; | |
2379 | if (lv) | |
2380 | vecp = lv->lv_iovecp; | |
2381 | } | |
fd63875c | 2382 | if (record_cnt == 0 && ordered == false) { |
55b66332 | 2383 | if (!lv) |
99428ad0 CH |
2384 | return 0; |
2385 | break; | |
2386 | } | |
2387 | } | |
2388 | } | |
2389 | ||
2390 | ASSERT(len == 0); | |
2391 | ||
2392 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); | |
2393 | if (!commit_iclog) | |
2394 | return xlog_state_release_iclog(log, iclog); | |
1da177e4 | 2395 | |
1da177e4 LT |
2396 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2397 | *commit_iclog = iclog; | |
2398 | return 0; | |
99428ad0 | 2399 | } |
1da177e4 LT |
2400 | |
2401 | ||
2402 | /***************************************************************************** | |
2403 | * | |
2404 | * State Machine functions | |
2405 | * | |
2406 | ***************************************************************************** | |
2407 | */ | |
2408 | ||
2409 | /* Clean iclogs starting from the head. This ordering must be | |
2410 | * maintained, so an iclog doesn't become ACTIVE beyond one that | |
2411 | * is SYNCING. This is also required to maintain the notion that we use | |
12017faf | 2412 | * a ordered wait queue to hold off would be writers to the log when every |
1da177e4 LT |
2413 | * iclog is trying to sync to disk. |
2414 | * | |
2415 | * State Change: DIRTY -> ACTIVE | |
2416 | */ | |
ba0f32d4 | 2417 | STATIC void |
9a8d2fdb MT |
2418 | xlog_state_clean_log( |
2419 | struct xlog *log) | |
1da177e4 LT |
2420 | { |
2421 | xlog_in_core_t *iclog; | |
2422 | int changed = 0; | |
2423 | ||
2424 | iclog = log->l_iclog; | |
2425 | do { | |
2426 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
2427 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2428 | iclog->ic_offset = 0; | |
114d23aa | 2429 | ASSERT(iclog->ic_callback == NULL); |
1da177e4 LT |
2430 | /* |
2431 | * If the number of ops in this iclog indicate it just | |
2432 | * contains the dummy transaction, we can | |
2433 | * change state into IDLE (the second time around). | |
2434 | * Otherwise we should change the state into | |
2435 | * NEED a dummy. | |
2436 | * We don't need to cover the dummy. | |
2437 | */ | |
2438 | if (!changed && | |
b53e675d CH |
2439 | (be32_to_cpu(iclog->ic_header.h_num_logops) == |
2440 | XLOG_COVER_OPS)) { | |
1da177e4 LT |
2441 | changed = 1; |
2442 | } else { | |
2443 | /* | |
2444 | * We have two dirty iclogs so start over | |
2445 | * This could also be num of ops indicates | |
2446 | * this is not the dummy going out. | |
2447 | */ | |
2448 | changed = 2; | |
2449 | } | |
2450 | iclog->ic_header.h_num_logops = 0; | |
2451 | memset(iclog->ic_header.h_cycle_data, 0, | |
2452 | sizeof(iclog->ic_header.h_cycle_data)); | |
2453 | iclog->ic_header.h_lsn = 0; | |
2454 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) | |
2455 | /* do nothing */; | |
2456 | else | |
2457 | break; /* stop cleaning */ | |
2458 | iclog = iclog->ic_next; | |
2459 | } while (iclog != log->l_iclog); | |
2460 | ||
2461 | /* log is locked when we are called */ | |
2462 | /* | |
2463 | * Change state for the dummy log recording. | |
2464 | * We usually go to NEED. But we go to NEED2 if the changed indicates | |
2465 | * we are done writing the dummy record. | |
2466 | * If we are done with the second dummy recored (DONE2), then | |
2467 | * we go to IDLE. | |
2468 | */ | |
2469 | if (changed) { | |
2470 | switch (log->l_covered_state) { | |
2471 | case XLOG_STATE_COVER_IDLE: | |
2472 | case XLOG_STATE_COVER_NEED: | |
2473 | case XLOG_STATE_COVER_NEED2: | |
2474 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2475 | break; | |
2476 | ||
2477 | case XLOG_STATE_COVER_DONE: | |
2478 | if (changed == 1) | |
2479 | log->l_covered_state = XLOG_STATE_COVER_NEED2; | |
2480 | else | |
2481 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2482 | break; | |
2483 | ||
2484 | case XLOG_STATE_COVER_DONE2: | |
2485 | if (changed == 1) | |
2486 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
2487 | else | |
2488 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2489 | break; | |
2490 | ||
2491 | default: | |
2492 | ASSERT(0); | |
2493 | } | |
2494 | } | |
2495 | } /* xlog_state_clean_log */ | |
2496 | ||
2497 | STATIC xfs_lsn_t | |
2498 | xlog_get_lowest_lsn( | |
9a8d2fdb | 2499 | struct xlog *log) |
1da177e4 LT |
2500 | { |
2501 | xlog_in_core_t *lsn_log; | |
2502 | xfs_lsn_t lowest_lsn, lsn; | |
2503 | ||
2504 | lsn_log = log->l_iclog; | |
2505 | lowest_lsn = 0; | |
2506 | do { | |
2507 | if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) { | |
b53e675d | 2508 | lsn = be64_to_cpu(lsn_log->ic_header.h_lsn); |
1da177e4 LT |
2509 | if ((lsn && !lowest_lsn) || |
2510 | (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) { | |
2511 | lowest_lsn = lsn; | |
2512 | } | |
2513 | } | |
2514 | lsn_log = lsn_log->ic_next; | |
2515 | } while (lsn_log != log->l_iclog); | |
014c2544 | 2516 | return lowest_lsn; |
1da177e4 LT |
2517 | } |
2518 | ||
2519 | ||
2520 | STATIC void | |
2521 | xlog_state_do_callback( | |
9a8d2fdb MT |
2522 | struct xlog *log, |
2523 | int aborted, | |
2524 | struct xlog_in_core *ciclog) | |
1da177e4 LT |
2525 | { |
2526 | xlog_in_core_t *iclog; | |
2527 | xlog_in_core_t *first_iclog; /* used to know when we've | |
2528 | * processed all iclogs once */ | |
2529 | xfs_log_callback_t *cb, *cb_next; | |
2530 | int flushcnt = 0; | |
2531 | xfs_lsn_t lowest_lsn; | |
2532 | int ioerrors; /* counter: iclogs with errors */ | |
2533 | int loopdidcallbacks; /* flag: inner loop did callbacks*/ | |
2534 | int funcdidcallbacks; /* flag: function did callbacks */ | |
2535 | int repeats; /* for issuing console warnings if | |
2536 | * looping too many times */ | |
d748c623 | 2537 | int wake = 0; |
1da177e4 | 2538 | |
b22cd72c | 2539 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2540 | first_iclog = iclog = log->l_iclog; |
2541 | ioerrors = 0; | |
2542 | funcdidcallbacks = 0; | |
2543 | repeats = 0; | |
2544 | ||
2545 | do { | |
2546 | /* | |
2547 | * Scan all iclogs starting with the one pointed to by the | |
2548 | * log. Reset this starting point each time the log is | |
2549 | * unlocked (during callbacks). | |
2550 | * | |
2551 | * Keep looping through iclogs until one full pass is made | |
2552 | * without running any callbacks. | |
2553 | */ | |
2554 | first_iclog = log->l_iclog; | |
2555 | iclog = log->l_iclog; | |
2556 | loopdidcallbacks = 0; | |
2557 | repeats++; | |
2558 | ||
2559 | do { | |
2560 | ||
2561 | /* skip all iclogs in the ACTIVE & DIRTY states */ | |
2562 | if (iclog->ic_state & | |
2563 | (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) { | |
2564 | iclog = iclog->ic_next; | |
2565 | continue; | |
2566 | } | |
2567 | ||
2568 | /* | |
2569 | * Between marking a filesystem SHUTDOWN and stopping | |
2570 | * the log, we do flush all iclogs to disk (if there | |
2571 | * wasn't a log I/O error). So, we do want things to | |
2572 | * go smoothly in case of just a SHUTDOWN w/o a | |
2573 | * LOG_IO_ERROR. | |
2574 | */ | |
2575 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
2576 | /* | |
2577 | * Can only perform callbacks in order. Since | |
2578 | * this iclog is not in the DONE_SYNC/ | |
2579 | * DO_CALLBACK state, we skip the rest and | |
2580 | * just try to clean up. If we set our iclog | |
2581 | * to DO_CALLBACK, we will not process it when | |
2582 | * we retry since a previous iclog is in the | |
2583 | * CALLBACK and the state cannot change since | |
b22cd72c | 2584 | * we are holding the l_icloglock. |
1da177e4 LT |
2585 | */ |
2586 | if (!(iclog->ic_state & | |
2587 | (XLOG_STATE_DONE_SYNC | | |
2588 | XLOG_STATE_DO_CALLBACK))) { | |
2589 | if (ciclog && (ciclog->ic_state == | |
2590 | XLOG_STATE_DONE_SYNC)) { | |
2591 | ciclog->ic_state = XLOG_STATE_DO_CALLBACK; | |
2592 | } | |
2593 | break; | |
2594 | } | |
2595 | /* | |
2596 | * We now have an iclog that is in either the | |
2597 | * DO_CALLBACK or DONE_SYNC states. The other | |
2598 | * states (WANT_SYNC, SYNCING, or CALLBACK were | |
2599 | * caught by the above if and are going to | |
2600 | * clean (i.e. we aren't doing their callbacks) | |
2601 | * see the above if. | |
2602 | */ | |
2603 | ||
2604 | /* | |
2605 | * We will do one more check here to see if we | |
2606 | * have chased our tail around. | |
2607 | */ | |
2608 | ||
2609 | lowest_lsn = xlog_get_lowest_lsn(log); | |
b53e675d CH |
2610 | if (lowest_lsn && |
2611 | XFS_LSN_CMP(lowest_lsn, | |
84f3c683 | 2612 | be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { |
1da177e4 LT |
2613 | iclog = iclog->ic_next; |
2614 | continue; /* Leave this iclog for | |
2615 | * another thread */ | |
2616 | } | |
2617 | ||
2618 | iclog->ic_state = XLOG_STATE_CALLBACK; | |
2619 | ||
1da177e4 | 2620 | |
84f3c683 | 2621 | /* |
d35e88fa DC |
2622 | * Completion of a iclog IO does not imply that |
2623 | * a transaction has completed, as transactions | |
2624 | * can be large enough to span many iclogs. We | |
2625 | * cannot change the tail of the log half way | |
2626 | * through a transaction as this may be the only | |
2627 | * transaction in the log and moving th etail to | |
2628 | * point to the middle of it will prevent | |
2629 | * recovery from finding the start of the | |
2630 | * transaction. Hence we should only update the | |
2631 | * last_sync_lsn if this iclog contains | |
2632 | * transaction completion callbacks on it. | |
2633 | * | |
2634 | * We have to do this before we drop the | |
84f3c683 DC |
2635 | * icloglock to ensure we are the only one that |
2636 | * can update it. | |
1da177e4 | 2637 | */ |
84f3c683 DC |
2638 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), |
2639 | be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); | |
d35e88fa DC |
2640 | if (iclog->ic_callback) |
2641 | atomic64_set(&log->l_last_sync_lsn, | |
2642 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1da177e4 | 2643 | |
84f3c683 | 2644 | } else |
1da177e4 | 2645 | ioerrors++; |
84f3c683 DC |
2646 | |
2647 | spin_unlock(&log->l_icloglock); | |
1da177e4 | 2648 | |
114d23aa DC |
2649 | /* |
2650 | * Keep processing entries in the callback list until | |
2651 | * we come around and it is empty. We need to | |
2652 | * atomically see that the list is empty and change the | |
2653 | * state to DIRTY so that we don't miss any more | |
2654 | * callbacks being added. | |
2655 | */ | |
2656 | spin_lock(&iclog->ic_callback_lock); | |
2657 | cb = iclog->ic_callback; | |
4b80916b | 2658 | while (cb) { |
1da177e4 LT |
2659 | iclog->ic_callback_tail = &(iclog->ic_callback); |
2660 | iclog->ic_callback = NULL; | |
114d23aa | 2661 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2662 | |
2663 | /* perform callbacks in the order given */ | |
4b80916b | 2664 | for (; cb; cb = cb_next) { |
1da177e4 LT |
2665 | cb_next = cb->cb_next; |
2666 | cb->cb_func(cb->cb_arg, aborted); | |
2667 | } | |
114d23aa | 2668 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
2669 | cb = iclog->ic_callback; |
2670 | } | |
2671 | ||
2672 | loopdidcallbacks++; | |
2673 | funcdidcallbacks++; | |
2674 | ||
114d23aa | 2675 | spin_lock(&log->l_icloglock); |
4b80916b | 2676 | ASSERT(iclog->ic_callback == NULL); |
114d23aa | 2677 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2678 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) |
2679 | iclog->ic_state = XLOG_STATE_DIRTY; | |
2680 | ||
2681 | /* | |
2682 | * Transition from DIRTY to ACTIVE if applicable. | |
2683 | * NOP if STATE_IOERROR. | |
2684 | */ | |
2685 | xlog_state_clean_log(log); | |
2686 | ||
2687 | /* wake up threads waiting in xfs_log_force() */ | |
eb40a875 | 2688 | wake_up_all(&iclog->ic_force_wait); |
1da177e4 LT |
2689 | |
2690 | iclog = iclog->ic_next; | |
2691 | } while (first_iclog != iclog); | |
a3c6685e NS |
2692 | |
2693 | if (repeats > 5000) { | |
2694 | flushcnt += repeats; | |
2695 | repeats = 0; | |
a0fa2b67 | 2696 | xfs_warn(log->l_mp, |
a3c6685e | 2697 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2698 | __func__, flushcnt); |
1da177e4 LT |
2699 | } |
2700 | } while (!ioerrors && loopdidcallbacks); | |
2701 | ||
2702 | /* | |
2703 | * make one last gasp attempt to see if iclogs are being left in | |
2704 | * limbo.. | |
2705 | */ | |
2706 | #ifdef DEBUG | |
2707 | if (funcdidcallbacks) { | |
2708 | first_iclog = iclog = log->l_iclog; | |
2709 | do { | |
2710 | ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK); | |
2711 | /* | |
2712 | * Terminate the loop if iclogs are found in states | |
2713 | * which will cause other threads to clean up iclogs. | |
2714 | * | |
2715 | * SYNCING - i/o completion will go through logs | |
2716 | * DONE_SYNC - interrupt thread should be waiting for | |
b22cd72c | 2717 | * l_icloglock |
1da177e4 LT |
2718 | * IOERROR - give up hope all ye who enter here |
2719 | */ | |
2720 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
2721 | iclog->ic_state == XLOG_STATE_SYNCING || | |
2722 | iclog->ic_state == XLOG_STATE_DONE_SYNC || | |
2723 | iclog->ic_state == XLOG_STATE_IOERROR ) | |
2724 | break; | |
2725 | iclog = iclog->ic_next; | |
2726 | } while (first_iclog != iclog); | |
2727 | } | |
2728 | #endif | |
2729 | ||
d748c623 MW |
2730 | if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) |
2731 | wake = 1; | |
b22cd72c | 2732 | spin_unlock(&log->l_icloglock); |
d748c623 MW |
2733 | |
2734 | if (wake) | |
eb40a875 | 2735 | wake_up_all(&log->l_flush_wait); |
d748c623 | 2736 | } |
1da177e4 LT |
2737 | |
2738 | ||
2739 | /* | |
2740 | * Finish transitioning this iclog to the dirty state. | |
2741 | * | |
2742 | * Make sure that we completely execute this routine only when this is | |
2743 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2744 | * when we reach the end of the physical log, get turned into 2 separate | |
2745 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2746 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2747 | * the second completion goes through. | |
2748 | * | |
2749 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2750 | * global state machine log lock. |
1da177e4 | 2751 | */ |
a8272ce0 | 2752 | STATIC void |
1da177e4 LT |
2753 | xlog_state_done_syncing( |
2754 | xlog_in_core_t *iclog, | |
2755 | int aborted) | |
2756 | { | |
9a8d2fdb | 2757 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2758 | |
b22cd72c | 2759 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2760 | |
2761 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING || | |
2762 | iclog->ic_state == XLOG_STATE_IOERROR); | |
155cc6b7 | 2763 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
2764 | ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2); |
2765 | ||
2766 | ||
2767 | /* | |
2768 | * If we got an error, either on the first buffer, or in the case of | |
2769 | * split log writes, on the second, we mark ALL iclogs STATE_IOERROR, | |
2770 | * and none should ever be attempted to be written to disk | |
2771 | * again. | |
2772 | */ | |
2773 | if (iclog->ic_state != XLOG_STATE_IOERROR) { | |
2774 | if (--iclog->ic_bwritecnt == 1) { | |
b22cd72c | 2775 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2776 | return; |
2777 | } | |
2778 | iclog->ic_state = XLOG_STATE_DONE_SYNC; | |
2779 | } | |
2780 | ||
2781 | /* | |
2782 | * Someone could be sleeping prior to writing out the next | |
2783 | * iclog buffer, we wake them all, one will get to do the | |
2784 | * I/O, the others get to wait for the result. | |
2785 | */ | |
eb40a875 | 2786 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2787 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2788 | xlog_state_do_callback(log, aborted, iclog); /* also cleans log */ |
2789 | } /* xlog_state_done_syncing */ | |
2790 | ||
2791 | ||
2792 | /* | |
2793 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2794 | * sleep. We wait on the flush queue on the head iclog as that should be |
2795 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2796 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2797 | * |
2798 | * The in-core logs are used in a circular fashion. They are not used | |
2799 | * out-of-order even when an iclog past the head is free. | |
2800 | * | |
2801 | * return: | |
2802 | * * log_offset where xlog_write() can start writing into the in-core | |
2803 | * log's data space. | |
2804 | * * in-core log pointer to which xlog_write() should write. | |
2805 | * * boolean indicating this is a continued write to an in-core log. | |
2806 | * If this is the last write, then the in-core log's offset field | |
2807 | * needs to be incremented, depending on the amount of data which | |
2808 | * is copied. | |
2809 | */ | |
a8272ce0 | 2810 | STATIC int |
9a8d2fdb MT |
2811 | xlog_state_get_iclog_space( |
2812 | struct xlog *log, | |
2813 | int len, | |
2814 | struct xlog_in_core **iclogp, | |
2815 | struct xlog_ticket *ticket, | |
2816 | int *continued_write, | |
2817 | int *logoffsetp) | |
1da177e4 | 2818 | { |
1da177e4 LT |
2819 | int log_offset; |
2820 | xlog_rec_header_t *head; | |
2821 | xlog_in_core_t *iclog; | |
2822 | int error; | |
2823 | ||
2824 | restart: | |
b22cd72c | 2825 | spin_lock(&log->l_icloglock); |
1da177e4 | 2826 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2827 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2828 | return XFS_ERROR(EIO); |
2829 | } | |
2830 | ||
2831 | iclog = log->l_iclog; | |
d748c623 | 2832 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
1da177e4 | 2833 | XFS_STATS_INC(xs_log_noiclogs); |
d748c623 MW |
2834 | |
2835 | /* Wait for log writes to have flushed */ | |
eb40a875 | 2836 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
2837 | goto restart; |
2838 | } | |
d748c623 | 2839 | |
1da177e4 LT |
2840 | head = &iclog->ic_header; |
2841 | ||
155cc6b7 | 2842 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
2843 | log_offset = iclog->ic_offset; |
2844 | ||
2845 | /* On the 1st write to an iclog, figure out lsn. This works | |
2846 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
2847 | * committing to. If the offset is set, that's how many blocks | |
2848 | * must be written. | |
2849 | */ | |
2850 | if (log_offset == 0) { | |
2851 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 2852 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
2853 | log->l_iclog_hsize, |
2854 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
2855 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
2856 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 2857 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
2858 | ASSERT(log->l_curr_block >= 0); |
2859 | } | |
2860 | ||
2861 | /* If there is enough room to write everything, then do it. Otherwise, | |
2862 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
2863 | * bit is on, so this will get flushed out. Don't update ic_offset | |
2864 | * until you know exactly how many bytes get copied. Therefore, wait | |
2865 | * until later to update ic_offset. | |
2866 | * | |
2867 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
2868 | * can fit into remaining data section. | |
2869 | */ | |
2870 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
2871 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2872 | ||
49641f1a DC |
2873 | /* |
2874 | * If I'm the only one writing to this iclog, sync it to disk. | |
2875 | * We need to do an atomic compare and decrement here to avoid | |
2876 | * racing with concurrent atomic_dec_and_lock() calls in | |
2877 | * xlog_state_release_iclog() when there is more than one | |
2878 | * reference to the iclog. | |
2879 | */ | |
2880 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) { | |
2881 | /* we are the only one */ | |
b22cd72c | 2882 | spin_unlock(&log->l_icloglock); |
49641f1a DC |
2883 | error = xlog_state_release_iclog(log, iclog); |
2884 | if (error) | |
014c2544 | 2885 | return error; |
1da177e4 | 2886 | } else { |
b22cd72c | 2887 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2888 | } |
2889 | goto restart; | |
2890 | } | |
2891 | ||
2892 | /* Do we have enough room to write the full amount in the remainder | |
2893 | * of this iclog? Or must we continue a write on the next iclog and | |
2894 | * mark this iclog as completely taken? In the case where we switch | |
2895 | * iclogs (to mark it taken), this particular iclog will release/sync | |
2896 | * to disk in xlog_write(). | |
2897 | */ | |
2898 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
2899 | *continued_write = 0; | |
2900 | iclog->ic_offset += len; | |
2901 | } else { | |
2902 | *continued_write = 1; | |
2903 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2904 | } | |
2905 | *iclogp = iclog; | |
2906 | ||
2907 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 2908 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2909 | |
2910 | *logoffsetp = log_offset; | |
2911 | return 0; | |
2912 | } /* xlog_state_get_iclog_space */ | |
2913 | ||
1da177e4 LT |
2914 | /* The first cnt-1 times through here we don't need to |
2915 | * move the grant write head because the permanent | |
2916 | * reservation has reserved cnt times the unit amount. | |
2917 | * Release part of current permanent unit reservation and | |
2918 | * reset current reservation to be one units worth. Also | |
2919 | * move grant reservation head forward. | |
2920 | */ | |
2921 | STATIC void | |
9a8d2fdb MT |
2922 | xlog_regrant_reserve_log_space( |
2923 | struct xlog *log, | |
2924 | struct xlog_ticket *ticket) | |
1da177e4 | 2925 | { |
0b1b213f CH |
2926 | trace_xfs_log_regrant_reserve_enter(log, ticket); |
2927 | ||
1da177e4 LT |
2928 | if (ticket->t_cnt > 0) |
2929 | ticket->t_cnt--; | |
2930 | ||
28496968 | 2931 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 2932 | ticket->t_curr_res); |
28496968 | 2933 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 2934 | ticket->t_curr_res); |
1da177e4 | 2935 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2936 | xlog_tic_reset_res(ticket); |
0b1b213f CH |
2937 | |
2938 | trace_xfs_log_regrant_reserve_sub(log, ticket); | |
2939 | ||
1da177e4 | 2940 | /* just return if we still have some of the pre-reserved space */ |
d0eb2f38 | 2941 | if (ticket->t_cnt > 0) |
1da177e4 | 2942 | return; |
1da177e4 | 2943 | |
28496968 | 2944 | xlog_grant_add_space(log, &log->l_reserve_head.grant, |
a69ed03c | 2945 | ticket->t_unit_res); |
0b1b213f CH |
2946 | |
2947 | trace_xfs_log_regrant_reserve_exit(log, ticket); | |
2948 | ||
1da177e4 | 2949 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2950 | xlog_tic_reset_res(ticket); |
1da177e4 LT |
2951 | } /* xlog_regrant_reserve_log_space */ |
2952 | ||
2953 | ||
2954 | /* | |
2955 | * Give back the space left from a reservation. | |
2956 | * | |
2957 | * All the information we need to make a correct determination of space left | |
2958 | * is present. For non-permanent reservations, things are quite easy. The | |
2959 | * count should have been decremented to zero. We only need to deal with the | |
2960 | * space remaining in the current reservation part of the ticket. If the | |
2961 | * ticket contains a permanent reservation, there may be left over space which | |
2962 | * needs to be released. A count of N means that N-1 refills of the current | |
2963 | * reservation can be done before we need to ask for more space. The first | |
2964 | * one goes to fill up the first current reservation. Once we run out of | |
2965 | * space, the count will stay at zero and the only space remaining will be | |
2966 | * in the current reservation field. | |
2967 | */ | |
2968 | STATIC void | |
9a8d2fdb MT |
2969 | xlog_ungrant_log_space( |
2970 | struct xlog *log, | |
2971 | struct xlog_ticket *ticket) | |
1da177e4 | 2972 | { |
663e496a DC |
2973 | int bytes; |
2974 | ||
1da177e4 LT |
2975 | if (ticket->t_cnt > 0) |
2976 | ticket->t_cnt--; | |
2977 | ||
0b1b213f | 2978 | trace_xfs_log_ungrant_enter(log, ticket); |
0b1b213f | 2979 | trace_xfs_log_ungrant_sub(log, ticket); |
1da177e4 | 2980 | |
663e496a DC |
2981 | /* |
2982 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
2983 | * up more space based on the remaining count. |
2984 | */ | |
663e496a | 2985 | bytes = ticket->t_curr_res; |
1da177e4 LT |
2986 | if (ticket->t_cnt > 0) { |
2987 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 2988 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
2989 | } |
2990 | ||
28496968 CH |
2991 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
2992 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 2993 | |
0b1b213f CH |
2994 | trace_xfs_log_ungrant_exit(log, ticket); |
2995 | ||
cfb7cdca | 2996 | xfs_log_space_wake(log->l_mp); |
09a423a3 | 2997 | } |
1da177e4 | 2998 | |
1da177e4 LT |
2999 | /* |
3000 | * Flush iclog to disk if this is the last reference to the given iclog and | |
3001 | * the WANT_SYNC bit is set. | |
3002 | * | |
3003 | * When this function is entered, the iclog is not necessarily in the | |
3004 | * WANT_SYNC state. It may be sitting around waiting to get filled. | |
3005 | * | |
3006 | * | |
3007 | */ | |
a8272ce0 | 3008 | STATIC int |
b589334c | 3009 | xlog_state_release_iclog( |
9a8d2fdb MT |
3010 | struct xlog *log, |
3011 | struct xlog_in_core *iclog) | |
1da177e4 | 3012 | { |
1da177e4 LT |
3013 | int sync = 0; /* do we sync? */ |
3014 | ||
155cc6b7 DC |
3015 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3016 | return XFS_ERROR(EIO); | |
3017 | ||
3018 | ASSERT(atomic_read(&iclog->ic_refcnt) > 0); | |
3019 | if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) | |
3020 | return 0; | |
3021 | ||
1da177e4 | 3022 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
b22cd72c | 3023 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3024 | return XFS_ERROR(EIO); |
3025 | } | |
1da177e4 LT |
3026 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE || |
3027 | iclog->ic_state == XLOG_STATE_WANT_SYNC); | |
3028 | ||
155cc6b7 | 3029 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { |
b589334c | 3030 | /* update tail before writing to iclog */ |
1c3cb9ec | 3031 | xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp); |
1da177e4 LT |
3032 | sync++; |
3033 | iclog->ic_state = XLOG_STATE_SYNCING; | |
1c3cb9ec DC |
3034 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); |
3035 | xlog_verify_tail_lsn(log, iclog, tail_lsn); | |
1da177e4 LT |
3036 | /* cycle incremented when incrementing curr_block */ |
3037 | } | |
b22cd72c | 3038 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3039 | |
3040 | /* | |
3041 | * We let the log lock go, so it's possible that we hit a log I/O | |
c41564b5 | 3042 | * error or some other SHUTDOWN condition that marks the iclog |
1da177e4 LT |
3043 | * as XLOG_STATE_IOERROR before the bwrite. However, we know that |
3044 | * this iclog has consistent data, so we ignore IOERROR | |
3045 | * flags after this point. | |
3046 | */ | |
b589334c | 3047 | if (sync) |
1da177e4 | 3048 | return xlog_sync(log, iclog); |
014c2544 | 3049 | return 0; |
1da177e4 LT |
3050 | } /* xlog_state_release_iclog */ |
3051 | ||
3052 | ||
3053 | /* | |
3054 | * This routine will mark the current iclog in the ring as WANT_SYNC | |
3055 | * and move the current iclog pointer to the next iclog in the ring. | |
3056 | * When this routine is called from xlog_state_get_iclog_space(), the | |
3057 | * exact size of the iclog has not yet been determined. All we know is | |
3058 | * that every data block. We have run out of space in this log record. | |
3059 | */ | |
3060 | STATIC void | |
9a8d2fdb MT |
3061 | xlog_state_switch_iclogs( |
3062 | struct xlog *log, | |
3063 | struct xlog_in_core *iclog, | |
3064 | int eventual_size) | |
1da177e4 LT |
3065 | { |
3066 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
3067 | if (!eventual_size) | |
3068 | eventual_size = iclog->ic_offset; | |
3069 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3070 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3071 | log->l_prev_block = log->l_curr_block; |
3072 | log->l_prev_cycle = log->l_curr_cycle; | |
3073 | ||
3074 | /* roll log?: ic_offset changed later */ | |
3075 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3076 | ||
3077 | /* Round up to next log-sunit */ | |
62118709 | 3078 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
1da177e4 LT |
3079 | log->l_mp->m_sb.sb_logsunit > 1) { |
3080 | __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit); | |
3081 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); | |
3082 | } | |
3083 | ||
3084 | if (log->l_curr_block >= log->l_logBBsize) { | |
3085 | log->l_curr_cycle++; | |
3086 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3087 | log->l_curr_cycle++; | |
3088 | log->l_curr_block -= log->l_logBBsize; | |
3089 | ASSERT(log->l_curr_block >= 0); | |
3090 | } | |
3091 | ASSERT(iclog == log->l_iclog); | |
3092 | log->l_iclog = iclog->ic_next; | |
3093 | } /* xlog_state_switch_iclogs */ | |
3094 | ||
1da177e4 LT |
3095 | /* |
3096 | * Write out all data in the in-core log as of this exact moment in time. | |
3097 | * | |
3098 | * Data may be written to the in-core log during this call. However, | |
3099 | * we don't guarantee this data will be written out. A change from past | |
3100 | * implementation means this routine will *not* write out zero length LRs. | |
3101 | * | |
3102 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3103 | * If we determine there is no flushable data, we just return. There is no | |
3104 | * flushable data if: | |
3105 | * | |
3106 | * 1. the current iclog is active and has no data; the previous iclog | |
3107 | * is in the active or dirty state. | |
3108 | * 2. the current iclog is drity, and the previous iclog is in the | |
3109 | * active or dirty state. | |
3110 | * | |
12017faf | 3111 | * We may sleep if: |
1da177e4 LT |
3112 | * |
3113 | * 1. the current iclog is not in the active nor dirty state. | |
3114 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3115 | * active nor dirty state. | |
3116 | * 3. the current iclog is active, and there is another thread writing | |
3117 | * to this particular iclog. | |
3118 | * 4. a) the current iclog is active and has no other writers | |
3119 | * b) when we return from flushing out this iclog, it is still | |
3120 | * not in the active nor dirty state. | |
3121 | */ | |
a14a348b CH |
3122 | int |
3123 | _xfs_log_force( | |
3124 | struct xfs_mount *mp, | |
3125 | uint flags, | |
3126 | int *log_flushed) | |
1da177e4 | 3127 | { |
ad223e60 | 3128 | struct xlog *log = mp->m_log; |
a14a348b CH |
3129 | struct xlog_in_core *iclog; |
3130 | xfs_lsn_t lsn; | |
3131 | ||
3132 | XFS_STATS_INC(xs_log_force); | |
1da177e4 | 3133 | |
93b8a585 | 3134 | xlog_cil_force(log); |
71e330b5 | 3135 | |
b22cd72c | 3136 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3137 | |
3138 | iclog = log->l_iclog; | |
3139 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3140 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3141 | return XFS_ERROR(EIO); |
3142 | } | |
3143 | ||
3144 | /* If the head iclog is not active nor dirty, we just attach | |
3145 | * ourselves to the head and go to sleep. | |
3146 | */ | |
3147 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3148 | iclog->ic_state == XLOG_STATE_DIRTY) { | |
3149 | /* | |
3150 | * If the head is dirty or (active and empty), then | |
3151 | * we need to look at the previous iclog. If the previous | |
3152 | * iclog is active or dirty we are done. There is nothing | |
3153 | * to sync out. Otherwise, we attach ourselves to the | |
3154 | * previous iclog and go to sleep. | |
3155 | */ | |
3156 | if (iclog->ic_state == XLOG_STATE_DIRTY || | |
155cc6b7 DC |
3157 | (atomic_read(&iclog->ic_refcnt) == 0 |
3158 | && iclog->ic_offset == 0)) { | |
1da177e4 LT |
3159 | iclog = iclog->ic_prev; |
3160 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3161 | iclog->ic_state == XLOG_STATE_DIRTY) | |
3162 | goto no_sleep; | |
3163 | else | |
3164 | goto maybe_sleep; | |
3165 | } else { | |
155cc6b7 | 3166 | if (atomic_read(&iclog->ic_refcnt) == 0) { |
1da177e4 LT |
3167 | /* We are the only one with access to this |
3168 | * iclog. Flush it out now. There should | |
3169 | * be a roundoff of zero to show that someone | |
3170 | * has already taken care of the roundoff from | |
3171 | * the previous sync. | |
3172 | */ | |
155cc6b7 | 3173 | atomic_inc(&iclog->ic_refcnt); |
b53e675d | 3174 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
1da177e4 | 3175 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 3176 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3177 | |
3178 | if (xlog_state_release_iclog(log, iclog)) | |
3179 | return XFS_ERROR(EIO); | |
a14a348b CH |
3180 | |
3181 | if (log_flushed) | |
3182 | *log_flushed = 1; | |
b22cd72c | 3183 | spin_lock(&log->l_icloglock); |
b53e675d | 3184 | if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn && |
1da177e4 LT |
3185 | iclog->ic_state != XLOG_STATE_DIRTY) |
3186 | goto maybe_sleep; | |
3187 | else | |
3188 | goto no_sleep; | |
3189 | } else { | |
3190 | /* Someone else is writing to this iclog. | |
3191 | * Use its call to flush out the data. However, | |
3192 | * the other thread may not force out this LR, | |
3193 | * so we mark it WANT_SYNC. | |
3194 | */ | |
3195 | xlog_state_switch_iclogs(log, iclog, 0); | |
3196 | goto maybe_sleep; | |
3197 | } | |
3198 | } | |
3199 | } | |
3200 | ||
3201 | /* By the time we come around again, the iclog could've been filled | |
3202 | * which would give it another lsn. If we have a new lsn, just | |
3203 | * return because the relevant data has been flushed. | |
3204 | */ | |
3205 | maybe_sleep: | |
3206 | if (flags & XFS_LOG_SYNC) { | |
3207 | /* | |
3208 | * We must check if we're shutting down here, before | |
b22cd72c | 3209 | * we wait, while we're holding the l_icloglock. |
1da177e4 LT |
3210 | * Then we check again after waking up, in case our |
3211 | * sleep was disturbed by a bad news. | |
3212 | */ | |
3213 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3214 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3215 | return XFS_ERROR(EIO); |
3216 | } | |
3217 | XFS_STATS_INC(xs_log_force_sleep); | |
eb40a875 | 3218 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); |
1da177e4 LT |
3219 | /* |
3220 | * No need to grab the log lock here since we're | |
3221 | * only deciding whether or not to return EIO | |
3222 | * and the memory read should be atomic. | |
3223 | */ | |
3224 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3225 | return XFS_ERROR(EIO); | |
a14a348b CH |
3226 | if (log_flushed) |
3227 | *log_flushed = 1; | |
1da177e4 LT |
3228 | } else { |
3229 | ||
3230 | no_sleep: | |
b22cd72c | 3231 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3232 | } |
3233 | return 0; | |
a14a348b | 3234 | } |
1da177e4 | 3235 | |
a14a348b CH |
3236 | /* |
3237 | * Wrapper for _xfs_log_force(), to be used when caller doesn't care | |
3238 | * about errors or whether the log was flushed or not. This is the normal | |
3239 | * interface to use when trying to unpin items or move the log forward. | |
3240 | */ | |
3241 | void | |
3242 | xfs_log_force( | |
3243 | xfs_mount_t *mp, | |
3244 | uint flags) | |
3245 | { | |
3246 | int error; | |
3247 | ||
14c26c6a | 3248 | trace_xfs_log_force(mp, 0); |
a14a348b | 3249 | error = _xfs_log_force(mp, flags, NULL); |
a0fa2b67 DC |
3250 | if (error) |
3251 | xfs_warn(mp, "%s: error %d returned.", __func__, error); | |
a14a348b | 3252 | } |
1da177e4 LT |
3253 | |
3254 | /* | |
a14a348b | 3255 | * Force the in-core log to disk for a specific LSN. |
1da177e4 LT |
3256 | * |
3257 | * Find in-core log with lsn. | |
3258 | * If it is in the DIRTY state, just return. | |
3259 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3260 | * state and go to sleep or return. | |
3261 | * If it is in any other state, go to sleep or return. | |
3262 | * | |
a14a348b CH |
3263 | * Synchronous forces are implemented with a signal variable. All callers |
3264 | * to force a given lsn to disk will wait on a the sv attached to the | |
3265 | * specific in-core log. When given in-core log finally completes its | |
3266 | * write to disk, that thread will wake up all threads waiting on the | |
3267 | * sv. | |
1da177e4 | 3268 | */ |
a14a348b CH |
3269 | int |
3270 | _xfs_log_force_lsn( | |
3271 | struct xfs_mount *mp, | |
3272 | xfs_lsn_t lsn, | |
3273 | uint flags, | |
3274 | int *log_flushed) | |
1da177e4 | 3275 | { |
ad223e60 | 3276 | struct xlog *log = mp->m_log; |
a14a348b CH |
3277 | struct xlog_in_core *iclog; |
3278 | int already_slept = 0; | |
1da177e4 | 3279 | |
a14a348b | 3280 | ASSERT(lsn != 0); |
1da177e4 | 3281 | |
a14a348b | 3282 | XFS_STATS_INC(xs_log_force); |
1da177e4 | 3283 | |
93b8a585 CH |
3284 | lsn = xlog_cil_force_lsn(log, lsn); |
3285 | if (lsn == NULLCOMMITLSN) | |
3286 | return 0; | |
71e330b5 | 3287 | |
a14a348b CH |
3288 | try_again: |
3289 | spin_lock(&log->l_icloglock); | |
3290 | iclog = log->l_iclog; | |
3291 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3292 | spin_unlock(&log->l_icloglock); |
a14a348b | 3293 | return XFS_ERROR(EIO); |
1da177e4 LT |
3294 | } |
3295 | ||
a14a348b CH |
3296 | do { |
3297 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { | |
3298 | iclog = iclog->ic_next; | |
3299 | continue; | |
3300 | } | |
3301 | ||
3302 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
3303 | spin_unlock(&log->l_icloglock); | |
3304 | return 0; | |
3305 | } | |
3306 | ||
3307 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3308 | /* | |
3309 | * We sleep here if we haven't already slept (e.g. | |
3310 | * this is the first time we've looked at the correct | |
3311 | * iclog buf) and the buffer before us is going to | |
3312 | * be sync'ed. The reason for this is that if we | |
3313 | * are doing sync transactions here, by waiting for | |
3314 | * the previous I/O to complete, we can allow a few | |
3315 | * more transactions into this iclog before we close | |
3316 | * it down. | |
3317 | * | |
3318 | * Otherwise, we mark the buffer WANT_SYNC, and bump | |
3319 | * up the refcnt so we can release the log (which | |
3320 | * drops the ref count). The state switch keeps new | |
3321 | * transaction commits from using this buffer. When | |
3322 | * the current commits finish writing into the buffer, | |
3323 | * the refcount will drop to zero and the buffer will | |
3324 | * go out then. | |
3325 | */ | |
3326 | if (!already_slept && | |
3327 | (iclog->ic_prev->ic_state & | |
3328 | (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) { | |
3329 | ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); | |
3330 | ||
3331 | XFS_STATS_INC(xs_log_force_sleep); | |
3332 | ||
eb40a875 DC |
3333 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3334 | &log->l_icloglock); | |
a14a348b CH |
3335 | if (log_flushed) |
3336 | *log_flushed = 1; | |
3337 | already_slept = 1; | |
3338 | goto try_again; | |
3339 | } | |
155cc6b7 | 3340 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 3341 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 3342 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3343 | if (xlog_state_release_iclog(log, iclog)) |
3344 | return XFS_ERROR(EIO); | |
a14a348b CH |
3345 | if (log_flushed) |
3346 | *log_flushed = 1; | |
b22cd72c | 3347 | spin_lock(&log->l_icloglock); |
1da177e4 | 3348 | } |
1da177e4 | 3349 | |
a14a348b CH |
3350 | if ((flags & XFS_LOG_SYNC) && /* sleep */ |
3351 | !(iclog->ic_state & | |
3352 | (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) { | |
3353 | /* | |
3354 | * Don't wait on completion if we know that we've | |
3355 | * gotten a log write error. | |
3356 | */ | |
3357 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
3358 | spin_unlock(&log->l_icloglock); | |
3359 | return XFS_ERROR(EIO); | |
3360 | } | |
3361 | XFS_STATS_INC(xs_log_force_sleep); | |
eb40a875 | 3362 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); |
a14a348b CH |
3363 | /* |
3364 | * No need to grab the log lock here since we're | |
3365 | * only deciding whether or not to return EIO | |
3366 | * and the memory read should be atomic. | |
3367 | */ | |
3368 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3369 | return XFS_ERROR(EIO); | |
1da177e4 | 3370 | |
a14a348b CH |
3371 | if (log_flushed) |
3372 | *log_flushed = 1; | |
3373 | } else { /* just return */ | |
b22cd72c | 3374 | spin_unlock(&log->l_icloglock); |
1da177e4 | 3375 | } |
1da177e4 | 3376 | |
a14a348b CH |
3377 | return 0; |
3378 | } while (iclog != log->l_iclog); | |
1da177e4 | 3379 | |
a14a348b CH |
3380 | spin_unlock(&log->l_icloglock); |
3381 | return 0; | |
3382 | } | |
3383 | ||
3384 | /* | |
3385 | * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care | |
3386 | * about errors or whether the log was flushed or not. This is the normal | |
3387 | * interface to use when trying to unpin items or move the log forward. | |
3388 | */ | |
3389 | void | |
3390 | xfs_log_force_lsn( | |
3391 | xfs_mount_t *mp, | |
3392 | xfs_lsn_t lsn, | |
3393 | uint flags) | |
3394 | { | |
3395 | int error; | |
1da177e4 | 3396 | |
14c26c6a | 3397 | trace_xfs_log_force(mp, lsn); |
a14a348b | 3398 | error = _xfs_log_force_lsn(mp, lsn, flags, NULL); |
a0fa2b67 DC |
3399 | if (error) |
3400 | xfs_warn(mp, "%s: error %d returned.", __func__, error); | |
a14a348b | 3401 | } |
1da177e4 LT |
3402 | |
3403 | /* | |
3404 | * Called when we want to mark the current iclog as being ready to sync to | |
3405 | * disk. | |
3406 | */ | |
a8272ce0 | 3407 | STATIC void |
9a8d2fdb MT |
3408 | xlog_state_want_sync( |
3409 | struct xlog *log, | |
3410 | struct xlog_in_core *iclog) | |
1da177e4 | 3411 | { |
a8914f3a | 3412 | assert_spin_locked(&log->l_icloglock); |
1da177e4 LT |
3413 | |
3414 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3415 | xlog_state_switch_iclogs(log, iclog, 0); | |
3416 | } else { | |
3417 | ASSERT(iclog->ic_state & | |
3418 | (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR)); | |
3419 | } | |
39e2defe | 3420 | } |
1da177e4 LT |
3421 | |
3422 | ||
3423 | /***************************************************************************** | |
3424 | * | |
3425 | * TICKET functions | |
3426 | * | |
3427 | ***************************************************************************** | |
3428 | */ | |
3429 | ||
3430 | /* | |
9da096fd | 3431 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3432 | */ |
cc09c0dc DC |
3433 | void |
3434 | xfs_log_ticket_put( | |
3435 | xlog_ticket_t *ticket) | |
1da177e4 | 3436 | { |
cc09c0dc | 3437 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3438 | if (atomic_dec_and_test(&ticket->t_ref)) |
cc09c0dc | 3439 | kmem_zone_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3440 | } |
1da177e4 | 3441 | |
cc09c0dc DC |
3442 | xlog_ticket_t * |
3443 | xfs_log_ticket_get( | |
3444 | xlog_ticket_t *ticket) | |
3445 | { | |
3446 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3447 | atomic_inc(&ticket->t_ref); | |
3448 | return ticket; | |
3449 | } | |
1da177e4 LT |
3450 | |
3451 | /* | |
e773fc93 JL |
3452 | * Figure out the total log space unit (in bytes) that would be |
3453 | * required for a log ticket. | |
1da177e4 | 3454 | */ |
e773fc93 JL |
3455 | int |
3456 | xfs_log_calc_unit_res( | |
3457 | struct xfs_mount *mp, | |
3458 | int unit_bytes) | |
1da177e4 | 3459 | { |
e773fc93 JL |
3460 | struct xlog *log = mp->m_log; |
3461 | int iclog_space; | |
3462 | uint num_headers; | |
1da177e4 LT |
3463 | |
3464 | /* | |
3465 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3466 | * in the log. A unit in this case is the amount of space for one | |
3467 | * of these log operations. Normal reservations have a cnt of 1 | |
3468 | * and their unit amount is the total amount of space required. | |
3469 | * | |
3470 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3471 | * which occupy space in the on-disk log. |
3472 | * | |
3473 | * Normal form of a transaction is: | |
3474 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3475 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3476 | * | |
3477 | * We need to account for all the leadup data and trailer data | |
3478 | * around the transaction data. | |
3479 | * And then we need to account for the worst case in terms of using | |
3480 | * more space. | |
3481 | * The worst case will happen if: | |
3482 | * - the placement of the transaction happens to be such that the | |
3483 | * roundoff is at its maximum | |
3484 | * - the transaction data is synced before the commit record is synced | |
3485 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3486 | * Therefore the commit record is in its own Log Record. | |
3487 | * This can happen as the commit record is called with its | |
3488 | * own region to xlog_write(). | |
3489 | * This then means that in the worst case, roundoff can happen for | |
3490 | * the commit-rec as well. | |
3491 | * The commit-rec is smaller than padding in this scenario and so it is | |
3492 | * not added separately. | |
1da177e4 LT |
3493 | */ |
3494 | ||
32fb9b57 TS |
3495 | /* for trans header */ |
3496 | unit_bytes += sizeof(xlog_op_header_t); | |
3497 | unit_bytes += sizeof(xfs_trans_header_t); | |
3498 | ||
1da177e4 | 3499 | /* for start-rec */ |
32fb9b57 TS |
3500 | unit_bytes += sizeof(xlog_op_header_t); |
3501 | ||
9b9fc2b7 DC |
3502 | /* |
3503 | * for LR headers - the space for data in an iclog is the size minus | |
3504 | * the space used for the headers. If we use the iclog size, then we | |
3505 | * undercalculate the number of headers required. | |
3506 | * | |
3507 | * Furthermore - the addition of op headers for split-recs might | |
3508 | * increase the space required enough to require more log and op | |
3509 | * headers, so take that into account too. | |
3510 | * | |
3511 | * IMPORTANT: This reservation makes the assumption that if this | |
3512 | * transaction is the first in an iclog and hence has the LR headers | |
3513 | * accounted to it, then the remaining space in the iclog is | |
3514 | * exclusively for this transaction. i.e. if the transaction is larger | |
3515 | * than the iclog, it will be the only thing in that iclog. | |
3516 | * Fundamentally, this means we must pass the entire log vector to | |
3517 | * xlog_write to guarantee this. | |
3518 | */ | |
3519 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3520 | num_headers = howmany(unit_bytes, iclog_space); | |
3521 | ||
3522 | /* for split-recs - ophdrs added when data split over LRs */ | |
3523 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3524 | ||
3525 | /* add extra header reservations if we overrun */ | |
3526 | while (!num_headers || | |
3527 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3528 | unit_bytes += sizeof(xlog_op_header_t); | |
3529 | num_headers++; | |
3530 | } | |
32fb9b57 | 3531 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3532 | |
32fb9b57 TS |
3533 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3534 | unit_bytes += log->l_iclog_hsize; | |
3535 | ||
32fb9b57 | 3536 | /* for roundoff padding for transaction data and one for commit record */ |
e773fc93 | 3537 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) { |
1da177e4 | 3538 | /* log su roundoff */ |
e773fc93 | 3539 | unit_bytes += 2 * mp->m_sb.sb_logsunit; |
1da177e4 LT |
3540 | } else { |
3541 | /* BB roundoff */ | |
e773fc93 | 3542 | unit_bytes += 2 * BBSIZE; |
1da177e4 LT |
3543 | } |
3544 | ||
e773fc93 JL |
3545 | return unit_bytes; |
3546 | } | |
3547 | ||
3548 | /* | |
3549 | * Allocate and initialise a new log ticket. | |
3550 | */ | |
3551 | struct xlog_ticket * | |
3552 | xlog_ticket_alloc( | |
3553 | struct xlog *log, | |
3554 | int unit_bytes, | |
3555 | int cnt, | |
3556 | char client, | |
3557 | bool permanent, | |
3558 | xfs_km_flags_t alloc_flags) | |
3559 | { | |
3560 | struct xlog_ticket *tic; | |
3561 | int unit_res; | |
3562 | ||
3563 | tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags); | |
3564 | if (!tic) | |
3565 | return NULL; | |
3566 | ||
3567 | unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes); | |
3568 | ||
cc09c0dc | 3569 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3570 | tic->t_task = current; |
10547941 | 3571 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3572 | tic->t_unit_res = unit_res; |
3573 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3574 | tic->t_cnt = cnt; |
3575 | tic->t_ocnt = cnt; | |
ecb3403d | 3576 | tic->t_tid = prandom_u32(); |
1da177e4 LT |
3577 | tic->t_clientid = client; |
3578 | tic->t_flags = XLOG_TIC_INITED; | |
7e9c6396 | 3579 | tic->t_trans_type = 0; |
9006fb91 | 3580 | if (permanent) |
1da177e4 | 3581 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3582 | |
0adba536 | 3583 | xlog_tic_reset_res(tic); |
7e9c6396 | 3584 | |
1da177e4 | 3585 | return tic; |
cc09c0dc | 3586 | } |
1da177e4 LT |
3587 | |
3588 | ||
3589 | /****************************************************************************** | |
3590 | * | |
3591 | * Log debug routines | |
3592 | * | |
3593 | ****************************************************************************** | |
3594 | */ | |
cfcbbbd0 | 3595 | #if defined(DEBUG) |
1da177e4 LT |
3596 | /* |
3597 | * Make sure that the destination ptr is within the valid data region of | |
3598 | * one of the iclogs. This uses backup pointers stored in a different | |
3599 | * part of the log in case we trash the log structure. | |
3600 | */ | |
3601 | void | |
e6b1f273 | 3602 | xlog_verify_dest_ptr( |
ad223e60 | 3603 | struct xlog *log, |
e6b1f273 | 3604 | char *ptr) |
1da177e4 LT |
3605 | { |
3606 | int i; | |
3607 | int good_ptr = 0; | |
3608 | ||
e6b1f273 CH |
3609 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3610 | if (ptr >= log->l_iclog_bak[i] && | |
3611 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3612 | good_ptr++; |
3613 | } | |
e6b1f273 CH |
3614 | |
3615 | if (!good_ptr) | |
a0fa2b67 | 3616 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3617 | } |
1da177e4 | 3618 | |
da8a1a4a DC |
3619 | /* |
3620 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3621 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3622 | * the cycles differ by exactly one and check the byte count. | |
3623 | * | |
3624 | * This check is run unlocked, so can give false positives. Rather than assert | |
3625 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3626 | * determine if they want to panic the machine when such an error occurs. For | |
3627 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3628 | * an assert, it can be turned off at runtime. | |
3629 | */ | |
3f336c6f DC |
3630 | STATIC void |
3631 | xlog_verify_grant_tail( | |
ad223e60 | 3632 | struct xlog *log) |
3f336c6f | 3633 | { |
1c3cb9ec | 3634 | int tail_cycle, tail_blocks; |
a69ed03c | 3635 | int cycle, space; |
3f336c6f | 3636 | |
28496968 | 3637 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3638 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3639 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3640 | if (cycle - 1 != tail_cycle && |
3641 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3642 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3643 | "%s: cycle - 1 != tail_cycle", __func__); | |
3644 | log->l_flags |= XLOG_TAIL_WARN; | |
3645 | } | |
3646 | ||
3647 | if (space > BBTOB(tail_blocks) && | |
3648 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3649 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3650 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3651 | log->l_flags |= XLOG_TAIL_WARN; | |
3652 | } | |
3f336c6f DC |
3653 | } |
3654 | } | |
3655 | ||
1da177e4 LT |
3656 | /* check if it will fit */ |
3657 | STATIC void | |
9a8d2fdb MT |
3658 | xlog_verify_tail_lsn( |
3659 | struct xlog *log, | |
3660 | struct xlog_in_core *iclog, | |
3661 | xfs_lsn_t tail_lsn) | |
1da177e4 LT |
3662 | { |
3663 | int blocks; | |
3664 | ||
3665 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3666 | blocks = | |
3667 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3668 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3669 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3670 | } else { |
3671 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3672 | ||
3673 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3674 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3675 | |
3676 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3677 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3678 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3679 | } |
3680 | } /* xlog_verify_tail_lsn */ | |
3681 | ||
3682 | /* | |
3683 | * Perform a number of checks on the iclog before writing to disk. | |
3684 | * | |
3685 | * 1. Make sure the iclogs are still circular | |
3686 | * 2. Make sure we have a good magic number | |
3687 | * 3. Make sure we don't have magic numbers in the data | |
3688 | * 4. Check fields of each log operation header for: | |
3689 | * A. Valid client identifier | |
3690 | * B. tid ptr value falls in valid ptr space (user space code) | |
3691 | * C. Length in log record header is correct according to the | |
3692 | * individual operation headers within record. | |
3693 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3694 | * log, check the preceding blocks of the physical log to make sure all | |
3695 | * the cycle numbers agree with the current cycle number. | |
3696 | */ | |
3697 | STATIC void | |
9a8d2fdb MT |
3698 | xlog_verify_iclog( |
3699 | struct xlog *log, | |
3700 | struct xlog_in_core *iclog, | |
3701 | int count, | |
667a9291 | 3702 | bool syncing) |
1da177e4 LT |
3703 | { |
3704 | xlog_op_header_t *ophead; | |
3705 | xlog_in_core_t *icptr; | |
3706 | xlog_in_core_2_t *xhdr; | |
3707 | xfs_caddr_t ptr; | |
3708 | xfs_caddr_t base_ptr; | |
3709 | __psint_t field_offset; | |
3710 | __uint8_t clientid; | |
3711 | int len, i, j, k, op_len; | |
3712 | int idx; | |
1da177e4 LT |
3713 | |
3714 | /* check validity of iclog pointers */ | |
b22cd72c | 3715 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3716 | icptr = log->l_iclog; |
3717 | for (i=0; i < log->l_iclog_bufs; i++) { | |
4b80916b | 3718 | if (icptr == NULL) |
a0fa2b67 | 3719 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
1da177e4 LT |
3720 | icptr = icptr->ic_next; |
3721 | } | |
3722 | if (icptr != log->l_iclog) | |
a0fa2b67 | 3723 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3724 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3725 | |
3726 | /* check log magic numbers */ | |
69ef921b | 3727 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3728 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3729 | |
b53e675d CH |
3730 | ptr = (xfs_caddr_t) &iclog->ic_header; |
3731 | for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count; | |
1da177e4 | 3732 | ptr += BBSIZE) { |
69ef921b | 3733 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3734 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3735 | __func__); | |
1da177e4 LT |
3736 | } |
3737 | ||
3738 | /* check fields */ | |
b53e675d | 3739 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
1da177e4 LT |
3740 | ptr = iclog->ic_datap; |
3741 | base_ptr = ptr; | |
3742 | ophead = (xlog_op_header_t *)ptr; | |
b28708d6 | 3743 | xhdr = iclog->ic_data; |
1da177e4 LT |
3744 | for (i = 0; i < len; i++) { |
3745 | ophead = (xlog_op_header_t *)ptr; | |
3746 | ||
3747 | /* clientid is only 1 byte */ | |
3748 | field_offset = (__psint_t) | |
3749 | ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr); | |
667a9291 | 3750 | if (!syncing || (field_offset & 0x1ff)) { |
1da177e4 LT |
3751 | clientid = ophead->oh_clientid; |
3752 | } else { | |
3753 | idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap); | |
3754 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3755 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3756 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3757 | clientid = xlog_get_client_id( |
3758 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3759 | } else { |
03bea6fe CH |
3760 | clientid = xlog_get_client_id( |
3761 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3762 | } |
3763 | } | |
3764 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 DC |
3765 | xfs_warn(log->l_mp, |
3766 | "%s: invalid clientid %d op 0x%p offset 0x%lx", | |
3767 | __func__, clientid, ophead, | |
3768 | (unsigned long)field_offset); | |
1da177e4 LT |
3769 | |
3770 | /* check length */ | |
3771 | field_offset = (__psint_t) | |
3772 | ((xfs_caddr_t)&(ophead->oh_len) - base_ptr); | |
667a9291 | 3773 | if (!syncing || (field_offset & 0x1ff)) { |
67fcb7bf | 3774 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 LT |
3775 | } else { |
3776 | idx = BTOBBT((__psint_t)&ophead->oh_len - | |
3777 | (__psint_t)iclog->ic_datap); | |
3778 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3779 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3780 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3781 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3782 | } else { |
b53e675d | 3783 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3784 | } |
3785 | } | |
3786 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3787 | } | |
3788 | } /* xlog_verify_iclog */ | |
cfcbbbd0 | 3789 | #endif |
1da177e4 LT |
3790 | |
3791 | /* | |
b22cd72c | 3792 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3793 | */ |
3794 | STATIC int | |
3795 | xlog_state_ioerror( | |
9a8d2fdb | 3796 | struct xlog *log) |
1da177e4 LT |
3797 | { |
3798 | xlog_in_core_t *iclog, *ic; | |
3799 | ||
3800 | iclog = log->l_iclog; | |
3801 | if (! (iclog->ic_state & XLOG_STATE_IOERROR)) { | |
3802 | /* | |
3803 | * Mark all the incore logs IOERROR. | |
3804 | * From now on, no log flushes will result. | |
3805 | */ | |
3806 | ic = iclog; | |
3807 | do { | |
3808 | ic->ic_state = XLOG_STATE_IOERROR; | |
3809 | ic = ic->ic_next; | |
3810 | } while (ic != iclog); | |
014c2544 | 3811 | return 0; |
1da177e4 LT |
3812 | } |
3813 | /* | |
3814 | * Return non-zero, if state transition has already happened. | |
3815 | */ | |
014c2544 | 3816 | return 1; |
1da177e4 LT |
3817 | } |
3818 | ||
3819 | /* | |
3820 | * This is called from xfs_force_shutdown, when we're forcibly | |
3821 | * shutting down the filesystem, typically because of an IO error. | |
3822 | * Our main objectives here are to make sure that: | |
3823 | * a. the filesystem gets marked 'SHUTDOWN' for all interested | |
3824 | * parties to find out, 'atomically'. | |
3825 | * b. those who're sleeping on log reservations, pinned objects and | |
3826 | * other resources get woken up, and be told the bad news. | |
3827 | * c. nothing new gets queued up after (a) and (b) are done. | |
3828 | * d. if !logerror, flush the iclogs to disk, then seal them off | |
3829 | * for business. | |
9da1ab18 DC |
3830 | * |
3831 | * Note: for delayed logging the !logerror case needs to flush the regions | |
3832 | * held in memory out to the iclogs before flushing them to disk. This needs | |
3833 | * to be done before the log is marked as shutdown, otherwise the flush to the | |
3834 | * iclogs will fail. | |
1da177e4 LT |
3835 | */ |
3836 | int | |
3837 | xfs_log_force_umount( | |
3838 | struct xfs_mount *mp, | |
3839 | int logerror) | |
3840 | { | |
9a8d2fdb | 3841 | struct xlog *log; |
1da177e4 | 3842 | int retval; |
1da177e4 LT |
3843 | |
3844 | log = mp->m_log; | |
3845 | ||
3846 | /* | |
3847 | * If this happens during log recovery, don't worry about | |
3848 | * locking; the log isn't open for business yet. | |
3849 | */ | |
3850 | if (!log || | |
3851 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3852 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 CH |
3853 | if (mp->m_sb_bp) |
3854 | XFS_BUF_DONE(mp->m_sb_bp); | |
014c2544 | 3855 | return 0; |
1da177e4 LT |
3856 | } |
3857 | ||
3858 | /* | |
3859 | * Somebody could've already done the hard work for us. | |
3860 | * No need to get locks for this. | |
3861 | */ | |
3862 | if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) { | |
3863 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); | |
014c2544 | 3864 | return 1; |
1da177e4 LT |
3865 | } |
3866 | retval = 0; | |
9da1ab18 DC |
3867 | |
3868 | /* | |
3869 | * Flush the in memory commit item list before marking the log as | |
3870 | * being shut down. We need to do it in this order to ensure all the | |
3871 | * completed transactions are flushed to disk with the xfs_log_force() | |
3872 | * call below. | |
3873 | */ | |
93b8a585 | 3874 | if (!logerror) |
a44f13ed | 3875 | xlog_cil_force(log); |
9da1ab18 | 3876 | |
1da177e4 | 3877 | /* |
3f16b985 DC |
3878 | * mark the filesystem and the as in a shutdown state and wake |
3879 | * everybody up to tell them the bad news. | |
1da177e4 | 3880 | */ |
b22cd72c | 3881 | spin_lock(&log->l_icloglock); |
1da177e4 | 3882 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 CH |
3883 | if (mp->m_sb_bp) |
3884 | XFS_BUF_DONE(mp->m_sb_bp); | |
3885 | ||
1da177e4 LT |
3886 | /* |
3887 | * This flag is sort of redundant because of the mount flag, but | |
3888 | * it's good to maintain the separation between the log and the rest | |
3889 | * of XFS. | |
3890 | */ | |
3891 | log->l_flags |= XLOG_IO_ERROR; | |
3892 | ||
3893 | /* | |
3894 | * If we hit a log error, we want to mark all the iclogs IOERROR | |
3895 | * while we're still holding the loglock. | |
3896 | */ | |
3897 | if (logerror) | |
3898 | retval = xlog_state_ioerror(log); | |
b22cd72c | 3899 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3900 | |
3901 | /* | |
10547941 DC |
3902 | * We don't want anybody waiting for log reservations after this. That |
3903 | * means we have to wake up everybody queued up on reserveq as well as | |
3904 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3905 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3906 | * action is protected by the grant locks. |
1da177e4 | 3907 | */ |
a79bf2d7 CH |
3908 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3909 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3910 | |
a14a348b | 3911 | if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) { |
1da177e4 LT |
3912 | ASSERT(!logerror); |
3913 | /* | |
3914 | * Force the incore logs to disk before shutting the | |
3915 | * log down completely. | |
3916 | */ | |
a14a348b CH |
3917 | _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
3918 | ||
b22cd72c | 3919 | spin_lock(&log->l_icloglock); |
1da177e4 | 3920 | retval = xlog_state_ioerror(log); |
b22cd72c | 3921 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3922 | } |
3923 | /* | |
3924 | * Wake up everybody waiting on xfs_log_force. | |
3925 | * Callback all log item committed functions as if the | |
3926 | * log writes were completed. | |
3927 | */ | |
3928 | xlog_state_do_callback(log, XFS_LI_ABORTED, NULL); | |
3929 | ||
3930 | #ifdef XFSERRORDEBUG | |
3931 | { | |
3932 | xlog_in_core_t *iclog; | |
3933 | ||
b22cd72c | 3934 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3935 | iclog = log->l_iclog; |
3936 | do { | |
3937 | ASSERT(iclog->ic_callback == 0); | |
3938 | iclog = iclog->ic_next; | |
3939 | } while (iclog != log->l_iclog); | |
b22cd72c | 3940 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3941 | } |
3942 | #endif | |
3943 | /* return non-zero if log IOERROR transition had already happened */ | |
014c2544 | 3944 | return retval; |
1da177e4 LT |
3945 | } |
3946 | ||
ba0f32d4 | 3947 | STATIC int |
9a8d2fdb MT |
3948 | xlog_iclogs_empty( |
3949 | struct xlog *log) | |
1da177e4 LT |
3950 | { |
3951 | xlog_in_core_t *iclog; | |
3952 | ||
3953 | iclog = log->l_iclog; | |
3954 | do { | |
3955 | /* endianness does not matter here, zero is zero in | |
3956 | * any language. | |
3957 | */ | |
3958 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3959 | return 0; |
1da177e4 LT |
3960 | iclog = iclog->ic_next; |
3961 | } while (iclog != log->l_iclog); | |
014c2544 | 3962 | return 1; |
1da177e4 | 3963 | } |
f661f1e0 | 3964 |