]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
a4fbe6ab | 9 | #include "xfs_format.h" |
239880ef DC |
10 | #include "xfs_log_format.h" |
11 | #include "xfs_trans_resv.h" | |
1da177e4 | 12 | #include "xfs_mount.h" |
e9e899a2 | 13 | #include "xfs_errortag.h" |
1da177e4 | 14 | #include "xfs_error.h" |
239880ef DC |
15 | #include "xfs_trans.h" |
16 | #include "xfs_trans_priv.h" | |
17 | #include "xfs_log.h" | |
1da177e4 | 18 | #include "xfs_log_priv.h" |
0b1b213f | 19 | #include "xfs_trace.h" |
baff4e44 | 20 | #include "xfs_sysfs.h" |
61e63ecb | 21 | #include "xfs_sb.h" |
39353ff6 | 22 | #include "xfs_health.h" |
1da177e4 | 23 | |
eb01c9cd | 24 | kmem_zone_t *xfs_log_ticket_zone; |
1da177e4 | 25 | |
1da177e4 | 26 | /* Local miscellaneous function prototypes */ |
9a8d2fdb MT |
27 | STATIC struct xlog * |
28 | xlog_alloc_log( | |
29 | struct xfs_mount *mp, | |
30 | struct xfs_buftarg *log_target, | |
31 | xfs_daddr_t blk_offset, | |
32 | int num_bblks); | |
ad223e60 MT |
33 | STATIC int |
34 | xlog_space_left( | |
35 | struct xlog *log, | |
36 | atomic64_t *head); | |
9a8d2fdb MT |
37 | STATIC void |
38 | xlog_dealloc_log( | |
39 | struct xlog *log); | |
1da177e4 LT |
40 | |
41 | /* local state machine functions */ | |
d15cbf2f | 42 | STATIC void xlog_state_done_syncing( |
12e6a0f4 | 43 | struct xlog_in_core *iclog); |
9a8d2fdb MT |
44 | STATIC int |
45 | xlog_state_get_iclog_space( | |
46 | struct xlog *log, | |
47 | int len, | |
48 | struct xlog_in_core **iclog, | |
49 | struct xlog_ticket *ticket, | |
50 | int *continued_write, | |
51 | int *logoffsetp); | |
9a8d2fdb MT |
52 | STATIC void |
53 | xlog_state_switch_iclogs( | |
54 | struct xlog *log, | |
55 | struct xlog_in_core *iclog, | |
56 | int eventual_size); | |
57 | STATIC void | |
ad223e60 | 58 | xlog_grant_push_ail( |
9a8d2fdb MT |
59 | struct xlog *log, |
60 | int need_bytes); | |
61 | STATIC void | |
df732b29 CH |
62 | xlog_sync( |
63 | struct xlog *log, | |
64 | struct xlog_in_core *iclog); | |
cfcbbbd0 | 65 | #if defined(DEBUG) |
9a8d2fdb MT |
66 | STATIC void |
67 | xlog_verify_dest_ptr( | |
68 | struct xlog *log, | |
5809d5e0 | 69 | void *ptr); |
ad223e60 MT |
70 | STATIC void |
71 | xlog_verify_grant_tail( | |
9a8d2fdb MT |
72 | struct xlog *log); |
73 | STATIC void | |
74 | xlog_verify_iclog( | |
75 | struct xlog *log, | |
76 | struct xlog_in_core *iclog, | |
abca1f33 | 77 | int count); |
9a8d2fdb MT |
78 | STATIC void |
79 | xlog_verify_tail_lsn( | |
80 | struct xlog *log, | |
81 | struct xlog_in_core *iclog, | |
82 | xfs_lsn_t tail_lsn); | |
1da177e4 LT |
83 | #else |
84 | #define xlog_verify_dest_ptr(a,b) | |
3f336c6f | 85 | #define xlog_verify_grant_tail(a) |
abca1f33 | 86 | #define xlog_verify_iclog(a,b,c) |
1da177e4 LT |
87 | #define xlog_verify_tail_lsn(a,b,c) |
88 | #endif | |
89 | ||
9a8d2fdb MT |
90 | STATIC int |
91 | xlog_iclogs_empty( | |
92 | struct xlog *log); | |
1da177e4 | 93 | |
303591a0 BF |
94 | static int |
95 | xfs_log_cover(struct xfs_mount *); | |
96 | ||
dd954c69 | 97 | static void |
663e496a | 98 | xlog_grant_sub_space( |
ad223e60 MT |
99 | struct xlog *log, |
100 | atomic64_t *head, | |
101 | int bytes) | |
dd954c69 | 102 | { |
d0eb2f38 DC |
103 | int64_t head_val = atomic64_read(head); |
104 | int64_t new, old; | |
a69ed03c | 105 | |
d0eb2f38 DC |
106 | do { |
107 | int cycle, space; | |
a69ed03c | 108 | |
d0eb2f38 | 109 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 110 | |
d0eb2f38 DC |
111 | space -= bytes; |
112 | if (space < 0) { | |
113 | space += log->l_logsize; | |
114 | cycle--; | |
115 | } | |
116 | ||
117 | old = head_val; | |
118 | new = xlog_assign_grant_head_val(cycle, space); | |
119 | head_val = atomic64_cmpxchg(head, old, new); | |
120 | } while (head_val != old); | |
dd954c69 CH |
121 | } |
122 | ||
123 | static void | |
663e496a | 124 | xlog_grant_add_space( |
ad223e60 MT |
125 | struct xlog *log, |
126 | atomic64_t *head, | |
127 | int bytes) | |
dd954c69 | 128 | { |
d0eb2f38 DC |
129 | int64_t head_val = atomic64_read(head); |
130 | int64_t new, old; | |
a69ed03c | 131 | |
d0eb2f38 DC |
132 | do { |
133 | int tmp; | |
134 | int cycle, space; | |
a69ed03c | 135 | |
d0eb2f38 | 136 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 137 | |
d0eb2f38 DC |
138 | tmp = log->l_logsize - space; |
139 | if (tmp > bytes) | |
140 | space += bytes; | |
141 | else { | |
142 | space = bytes - tmp; | |
143 | cycle++; | |
144 | } | |
145 | ||
146 | old = head_val; | |
147 | new = xlog_assign_grant_head_val(cycle, space); | |
148 | head_val = atomic64_cmpxchg(head, old, new); | |
149 | } while (head_val != old); | |
dd954c69 | 150 | } |
a69ed03c | 151 | |
c303c5b8 CH |
152 | STATIC void |
153 | xlog_grant_head_init( | |
154 | struct xlog_grant_head *head) | |
155 | { | |
156 | xlog_assign_grant_head(&head->grant, 1, 0); | |
157 | INIT_LIST_HEAD(&head->waiters); | |
158 | spin_lock_init(&head->lock); | |
159 | } | |
160 | ||
a79bf2d7 CH |
161 | STATIC void |
162 | xlog_grant_head_wake_all( | |
163 | struct xlog_grant_head *head) | |
164 | { | |
165 | struct xlog_ticket *tic; | |
166 | ||
167 | spin_lock(&head->lock); | |
168 | list_for_each_entry(tic, &head->waiters, t_queue) | |
169 | wake_up_process(tic->t_task); | |
170 | spin_unlock(&head->lock); | |
171 | } | |
172 | ||
e179840d CH |
173 | static inline int |
174 | xlog_ticket_reservation( | |
ad223e60 | 175 | struct xlog *log, |
e179840d CH |
176 | struct xlog_grant_head *head, |
177 | struct xlog_ticket *tic) | |
9f9c19ec | 178 | { |
e179840d CH |
179 | if (head == &log->l_write_head) { |
180 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
181 | return tic->t_unit_res; | |
182 | } else { | |
9f9c19ec | 183 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) |
e179840d | 184 | return tic->t_unit_res * tic->t_cnt; |
9f9c19ec | 185 | else |
e179840d | 186 | return tic->t_unit_res; |
9f9c19ec | 187 | } |
9f9c19ec CH |
188 | } |
189 | ||
190 | STATIC bool | |
e179840d | 191 | xlog_grant_head_wake( |
ad223e60 | 192 | struct xlog *log, |
e179840d | 193 | struct xlog_grant_head *head, |
9f9c19ec CH |
194 | int *free_bytes) |
195 | { | |
196 | struct xlog_ticket *tic; | |
197 | int need_bytes; | |
7c107afb | 198 | bool woken_task = false; |
9f9c19ec | 199 | |
e179840d | 200 | list_for_each_entry(tic, &head->waiters, t_queue) { |
7c107afb DC |
201 | |
202 | /* | |
203 | * There is a chance that the size of the CIL checkpoints in | |
204 | * progress at the last AIL push target calculation resulted in | |
205 | * limiting the target to the log head (l_last_sync_lsn) at the | |
206 | * time. This may not reflect where the log head is now as the | |
207 | * CIL checkpoints may have completed. | |
208 | * | |
209 | * Hence when we are woken here, it may be that the head of the | |
210 | * log that has moved rather than the tail. As the tail didn't | |
211 | * move, there still won't be space available for the | |
212 | * reservation we require. However, if the AIL has already | |
213 | * pushed to the target defined by the old log head location, we | |
214 | * will hang here waiting for something else to update the AIL | |
215 | * push target. | |
216 | * | |
217 | * Therefore, if there isn't space to wake the first waiter on | |
218 | * the grant head, we need to push the AIL again to ensure the | |
219 | * target reflects both the current log tail and log head | |
220 | * position before we wait for the tail to move again. | |
221 | */ | |
222 | ||
e179840d | 223 | need_bytes = xlog_ticket_reservation(log, head, tic); |
7c107afb DC |
224 | if (*free_bytes < need_bytes) { |
225 | if (!woken_task) | |
226 | xlog_grant_push_ail(log, need_bytes); | |
9f9c19ec | 227 | return false; |
7c107afb | 228 | } |
9f9c19ec | 229 | |
e179840d CH |
230 | *free_bytes -= need_bytes; |
231 | trace_xfs_log_grant_wake_up(log, tic); | |
14a7235f | 232 | wake_up_process(tic->t_task); |
7c107afb | 233 | woken_task = true; |
9f9c19ec CH |
234 | } |
235 | ||
236 | return true; | |
237 | } | |
238 | ||
239 | STATIC int | |
23ee3df3 | 240 | xlog_grant_head_wait( |
ad223e60 | 241 | struct xlog *log, |
23ee3df3 | 242 | struct xlog_grant_head *head, |
9f9c19ec | 243 | struct xlog_ticket *tic, |
a30b0367 DC |
244 | int need_bytes) __releases(&head->lock) |
245 | __acquires(&head->lock) | |
9f9c19ec | 246 | { |
23ee3df3 | 247 | list_add_tail(&tic->t_queue, &head->waiters); |
9f9c19ec CH |
248 | |
249 | do { | |
250 | if (XLOG_FORCED_SHUTDOWN(log)) | |
251 | goto shutdown; | |
252 | xlog_grant_push_ail(log, need_bytes); | |
253 | ||
14a7235f | 254 | __set_current_state(TASK_UNINTERRUPTIBLE); |
23ee3df3 | 255 | spin_unlock(&head->lock); |
14a7235f | 256 | |
ff6d6af2 | 257 | XFS_STATS_INC(log->l_mp, xs_sleep_logspace); |
9f9c19ec | 258 | |
14a7235f CH |
259 | trace_xfs_log_grant_sleep(log, tic); |
260 | schedule(); | |
9f9c19ec CH |
261 | trace_xfs_log_grant_wake(log, tic); |
262 | ||
23ee3df3 | 263 | spin_lock(&head->lock); |
9f9c19ec CH |
264 | if (XLOG_FORCED_SHUTDOWN(log)) |
265 | goto shutdown; | |
23ee3df3 | 266 | } while (xlog_space_left(log, &head->grant) < need_bytes); |
9f9c19ec CH |
267 | |
268 | list_del_init(&tic->t_queue); | |
269 | return 0; | |
270 | shutdown: | |
271 | list_del_init(&tic->t_queue); | |
2451337d | 272 | return -EIO; |
9f9c19ec CH |
273 | } |
274 | ||
42ceedb3 CH |
275 | /* |
276 | * Atomically get the log space required for a log ticket. | |
277 | * | |
278 | * Once a ticket gets put onto head->waiters, it will only return after the | |
279 | * needed reservation is satisfied. | |
280 | * | |
281 | * This function is structured so that it has a lock free fast path. This is | |
282 | * necessary because every new transaction reservation will come through this | |
283 | * path. Hence any lock will be globally hot if we take it unconditionally on | |
284 | * every pass. | |
285 | * | |
286 | * As tickets are only ever moved on and off head->waiters under head->lock, we | |
287 | * only need to take that lock if we are going to add the ticket to the queue | |
288 | * and sleep. We can avoid taking the lock if the ticket was never added to | |
289 | * head->waiters because the t_queue list head will be empty and we hold the | |
290 | * only reference to it so it can safely be checked unlocked. | |
291 | */ | |
292 | STATIC int | |
293 | xlog_grant_head_check( | |
ad223e60 | 294 | struct xlog *log, |
42ceedb3 CH |
295 | struct xlog_grant_head *head, |
296 | struct xlog_ticket *tic, | |
297 | int *need_bytes) | |
298 | { | |
299 | int free_bytes; | |
300 | int error = 0; | |
301 | ||
302 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); | |
303 | ||
304 | /* | |
305 | * If there are other waiters on the queue then give them a chance at | |
306 | * logspace before us. Wake up the first waiters, if we do not wake | |
307 | * up all the waiters then go to sleep waiting for more free space, | |
308 | * otherwise try to get some space for this transaction. | |
309 | */ | |
310 | *need_bytes = xlog_ticket_reservation(log, head, tic); | |
311 | free_bytes = xlog_space_left(log, &head->grant); | |
312 | if (!list_empty_careful(&head->waiters)) { | |
313 | spin_lock(&head->lock); | |
314 | if (!xlog_grant_head_wake(log, head, &free_bytes) || | |
315 | free_bytes < *need_bytes) { | |
316 | error = xlog_grant_head_wait(log, head, tic, | |
317 | *need_bytes); | |
318 | } | |
319 | spin_unlock(&head->lock); | |
320 | } else if (free_bytes < *need_bytes) { | |
321 | spin_lock(&head->lock); | |
322 | error = xlog_grant_head_wait(log, head, tic, *need_bytes); | |
323 | spin_unlock(&head->lock); | |
324 | } | |
325 | ||
326 | return error; | |
327 | } | |
328 | ||
0adba536 CH |
329 | static void |
330 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
331 | { | |
332 | tic->t_res_num = 0; | |
333 | tic->t_res_arr_sum = 0; | |
334 | tic->t_res_num_ophdrs = 0; | |
335 | } | |
336 | ||
337 | static void | |
338 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
339 | { | |
340 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
341 | /* add to overflow and start again */ | |
342 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
343 | tic->t_res_num = 0; | |
344 | tic->t_res_arr_sum = 0; | |
345 | } | |
346 | ||
347 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
348 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
349 | tic->t_res_arr_sum += len; | |
350 | tic->t_res_num++; | |
351 | } | |
dd954c69 | 352 | |
50d25484 BF |
353 | bool |
354 | xfs_log_writable( | |
355 | struct xfs_mount *mp) | |
356 | { | |
357 | /* | |
8e9800f9 DW |
358 | * Do not write to the log on norecovery mounts, if the data or log |
359 | * devices are read-only, or if the filesystem is shutdown. Read-only | |
360 | * mounts allow internal writes for log recovery and unmount purposes, | |
361 | * so don't restrict that case. | |
50d25484 BF |
362 | */ |
363 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
364 | return false; | |
8e9800f9 DW |
365 | if (xfs_readonly_buftarg(mp->m_ddev_targp)) |
366 | return false; | |
50d25484 BF |
367 | if (xfs_readonly_buftarg(mp->m_log->l_targ)) |
368 | return false; | |
369 | if (XFS_FORCED_SHUTDOWN(mp)) | |
370 | return false; | |
371 | return true; | |
372 | } | |
373 | ||
9006fb91 CH |
374 | /* |
375 | * Replenish the byte reservation required by moving the grant write head. | |
376 | */ | |
377 | int | |
378 | xfs_log_regrant( | |
379 | struct xfs_mount *mp, | |
380 | struct xlog_ticket *tic) | |
381 | { | |
ad223e60 | 382 | struct xlog *log = mp->m_log; |
9006fb91 CH |
383 | int need_bytes; |
384 | int error = 0; | |
385 | ||
386 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 387 | return -EIO; |
9006fb91 | 388 | |
ff6d6af2 | 389 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
390 | |
391 | /* | |
392 | * This is a new transaction on the ticket, so we need to change the | |
393 | * transaction ID so that the next transaction has a different TID in | |
394 | * the log. Just add one to the existing tid so that we can see chains | |
395 | * of rolling transactions in the log easily. | |
396 | */ | |
397 | tic->t_tid++; | |
398 | ||
399 | xlog_grant_push_ail(log, tic->t_unit_res); | |
400 | ||
401 | tic->t_curr_res = tic->t_unit_res; | |
402 | xlog_tic_reset_res(tic); | |
403 | ||
404 | if (tic->t_cnt > 0) | |
405 | return 0; | |
406 | ||
407 | trace_xfs_log_regrant(log, tic); | |
408 | ||
409 | error = xlog_grant_head_check(log, &log->l_write_head, tic, | |
410 | &need_bytes); | |
411 | if (error) | |
412 | goto out_error; | |
413 | ||
414 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
415 | trace_xfs_log_regrant_exit(log, tic); | |
416 | xlog_verify_grant_tail(log); | |
417 | return 0; | |
418 | ||
419 | out_error: | |
420 | /* | |
421 | * If we are failing, make sure the ticket doesn't have any current | |
422 | * reservations. We don't want to add this back when the ticket/ | |
423 | * transaction gets cancelled. | |
424 | */ | |
425 | tic->t_curr_res = 0; | |
426 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
427 | return error; | |
428 | } | |
429 | ||
430 | /* | |
a0e336ba | 431 | * Reserve log space and return a ticket corresponding to the reservation. |
9006fb91 CH |
432 | * |
433 | * Each reservation is going to reserve extra space for a log record header. | |
434 | * When writes happen to the on-disk log, we don't subtract the length of the | |
435 | * log record header from any reservation. By wasting space in each | |
436 | * reservation, we prevent over allocation problems. | |
437 | */ | |
438 | int | |
439 | xfs_log_reserve( | |
440 | struct xfs_mount *mp, | |
441 | int unit_bytes, | |
442 | int cnt, | |
443 | struct xlog_ticket **ticp, | |
c8ce540d | 444 | uint8_t client, |
710b1e2c | 445 | bool permanent) |
9006fb91 | 446 | { |
ad223e60 | 447 | struct xlog *log = mp->m_log; |
9006fb91 CH |
448 | struct xlog_ticket *tic; |
449 | int need_bytes; | |
450 | int error = 0; | |
451 | ||
452 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); | |
453 | ||
454 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 455 | return -EIO; |
9006fb91 | 456 | |
ff6d6af2 | 457 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
458 | |
459 | ASSERT(*ticp == NULL); | |
ca4f2589 | 460 | tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent); |
9006fb91 CH |
461 | *ticp = tic; |
462 | ||
437a255a DC |
463 | xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt |
464 | : tic->t_unit_res); | |
9006fb91 CH |
465 | |
466 | trace_xfs_log_reserve(log, tic); | |
467 | ||
468 | error = xlog_grant_head_check(log, &log->l_reserve_head, tic, | |
469 | &need_bytes); | |
470 | if (error) | |
471 | goto out_error; | |
472 | ||
473 | xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); | |
474 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
475 | trace_xfs_log_reserve_exit(log, tic); | |
476 | xlog_verify_grant_tail(log); | |
477 | return 0; | |
478 | ||
479 | out_error: | |
480 | /* | |
481 | * If we are failing, make sure the ticket doesn't have any current | |
482 | * reservations. We don't want to add this back when the ticket/ | |
483 | * transaction gets cancelled. | |
484 | */ | |
485 | tic->t_curr_res = 0; | |
486 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
487 | return error; | |
488 | } | |
489 | ||
df732b29 CH |
490 | static bool |
491 | __xlog_state_release_iclog( | |
492 | struct xlog *log, | |
493 | struct xlog_in_core *iclog) | |
494 | { | |
495 | lockdep_assert_held(&log->l_icloglock); | |
496 | ||
497 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { | |
498 | /* update tail before writing to iclog */ | |
499 | xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp); | |
500 | ||
501 | iclog->ic_state = XLOG_STATE_SYNCING; | |
502 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); | |
503 | xlog_verify_tail_lsn(log, iclog, tail_lsn); | |
504 | /* cycle incremented when incrementing curr_block */ | |
505 | return true; | |
506 | } | |
507 | ||
508 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
509 | return false; | |
510 | } | |
511 | ||
512 | /* | |
513 | * Flush iclog to disk if this is the last reference to the given iclog and the | |
514 | * it is in the WANT_SYNC state. | |
515 | */ | |
eef983ff | 516 | int |
df732b29 CH |
517 | xlog_state_release_iclog( |
518 | struct xlog *log, | |
519 | struct xlog_in_core *iclog) | |
520 | { | |
521 | lockdep_assert_held(&log->l_icloglock); | |
522 | ||
1858bb0b | 523 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
df732b29 CH |
524 | return -EIO; |
525 | ||
526 | if (atomic_dec_and_test(&iclog->ic_refcnt) && | |
527 | __xlog_state_release_iclog(log, iclog)) { | |
528 | spin_unlock(&log->l_icloglock); | |
529 | xlog_sync(log, iclog); | |
530 | spin_lock(&log->l_icloglock); | |
531 | } | |
532 | ||
533 | return 0; | |
534 | } | |
535 | ||
1da177e4 LT |
536 | /* |
537 | * Mount a log filesystem | |
538 | * | |
539 | * mp - ubiquitous xfs mount point structure | |
540 | * log_target - buftarg of on-disk log device | |
541 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
542 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
543 | * | |
544 | * Return error or zero. | |
545 | */ | |
546 | int | |
249a8c11 DC |
547 | xfs_log_mount( |
548 | xfs_mount_t *mp, | |
549 | xfs_buftarg_t *log_target, | |
550 | xfs_daddr_t blk_offset, | |
551 | int num_bblks) | |
1da177e4 | 552 | { |
9c92ee20 | 553 | bool fatal = xfs_sb_version_hascrc(&mp->m_sb); |
3e7b91cf JL |
554 | int error = 0; |
555 | int min_logfsbs; | |
249a8c11 | 556 | |
c99d609a DC |
557 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
558 | xfs_notice(mp, "Mounting V%d Filesystem", | |
559 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
560 | } else { | |
a0fa2b67 | 561 | xfs_notice(mp, |
c99d609a DC |
562 | "Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.", |
563 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
bd186aa9 | 564 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
565 | } |
566 | ||
567 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e | 568 | if (IS_ERR(mp->m_log)) { |
2451337d | 569 | error = PTR_ERR(mp->m_log); |
644c3567 DC |
570 | goto out; |
571 | } | |
1da177e4 | 572 | |
3e7b91cf JL |
573 | /* |
574 | * Validate the given log space and drop a critical message via syslog | |
575 | * if the log size is too small that would lead to some unexpected | |
576 | * situations in transaction log space reservation stage. | |
577 | * | |
578 | * Note: we can't just reject the mount if the validation fails. This | |
579 | * would mean that people would have to downgrade their kernel just to | |
580 | * remedy the situation as there is no way to grow the log (short of | |
581 | * black magic surgery with xfs_db). | |
582 | * | |
583 | * We can, however, reject mounts for CRC format filesystems, as the | |
584 | * mkfs binary being used to make the filesystem should never create a | |
585 | * filesystem with a log that is too small. | |
586 | */ | |
587 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
588 | ||
589 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
590 | xfs_warn(mp, | |
591 | "Log size %d blocks too small, minimum size is %d blocks", | |
592 | mp->m_sb.sb_logblocks, min_logfsbs); | |
2451337d | 593 | error = -EINVAL; |
3e7b91cf JL |
594 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { |
595 | xfs_warn(mp, | |
596 | "Log size %d blocks too large, maximum size is %lld blocks", | |
597 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
2451337d | 598 | error = -EINVAL; |
3e7b91cf JL |
599 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { |
600 | xfs_warn(mp, | |
601 | "log size %lld bytes too large, maximum size is %lld bytes", | |
602 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
603 | XFS_MAX_LOG_BYTES); | |
2451337d | 604 | error = -EINVAL; |
9c92ee20 DW |
605 | } else if (mp->m_sb.sb_logsunit > 1 && |
606 | mp->m_sb.sb_logsunit % mp->m_sb.sb_blocksize) { | |
607 | xfs_warn(mp, | |
608 | "log stripe unit %u bytes must be a multiple of block size", | |
609 | mp->m_sb.sb_logsunit); | |
610 | error = -EINVAL; | |
611 | fatal = true; | |
3e7b91cf JL |
612 | } |
613 | if (error) { | |
9c92ee20 DW |
614 | /* |
615 | * Log check errors are always fatal on v5; or whenever bad | |
616 | * metadata leads to a crash. | |
617 | */ | |
618 | if (fatal) { | |
3e7b91cf JL |
619 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); |
620 | ASSERT(0); | |
621 | goto out_free_log; | |
622 | } | |
f41febd2 | 623 | xfs_crit(mp, "Log size out of supported range."); |
3e7b91cf | 624 | xfs_crit(mp, |
f41febd2 | 625 | "Continuing onwards, but if log hangs are experienced then please report this message in the bug report."); |
3e7b91cf JL |
626 | } |
627 | ||
249a8c11 DC |
628 | /* |
629 | * Initialize the AIL now we have a log. | |
630 | */ | |
249a8c11 DC |
631 | error = xfs_trans_ail_init(mp); |
632 | if (error) { | |
a0fa2b67 | 633 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 634 | goto out_free_log; |
249a8c11 | 635 | } |
a9c21c1b | 636 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 637 | |
1da177e4 LT |
638 | /* |
639 | * skip log recovery on a norecovery mount. pretend it all | |
640 | * just worked. | |
641 | */ | |
642 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 643 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
644 | |
645 | if (readonly) | |
bd186aa9 | 646 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 647 | |
65be6054 | 648 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
649 | |
650 | if (readonly) | |
bd186aa9 | 651 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 652 | if (error) { |
a0fa2b67 DC |
653 | xfs_warn(mp, "log mount/recovery failed: error %d", |
654 | error); | |
f0b2efad | 655 | xlog_recover_cancel(mp->m_log); |
26430752 | 656 | goto out_destroy_ail; |
1da177e4 LT |
657 | } |
658 | } | |
659 | ||
baff4e44 BF |
660 | error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj, |
661 | "log"); | |
662 | if (error) | |
663 | goto out_destroy_ail; | |
664 | ||
1da177e4 LT |
665 | /* Normal transactions can now occur */ |
666 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
667 | ||
71e330b5 DC |
668 | /* |
669 | * Now the log has been fully initialised and we know were our | |
670 | * space grant counters are, we can initialise the permanent ticket | |
671 | * needed for delayed logging to work. | |
672 | */ | |
673 | xlog_cil_init_post_recovery(mp->m_log); | |
674 | ||
1da177e4 | 675 | return 0; |
26430752 CH |
676 | |
677 | out_destroy_ail: | |
678 | xfs_trans_ail_destroy(mp); | |
679 | out_free_log: | |
680 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 681 | out: |
249a8c11 | 682 | return error; |
26430752 | 683 | } |
1da177e4 LT |
684 | |
685 | /* | |
f661f1e0 DC |
686 | * Finish the recovery of the file system. This is separate from the |
687 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
688 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
689 | * here. | |
1da177e4 | 690 | * |
f661f1e0 DC |
691 | * If we finish recovery successfully, start the background log work. If we are |
692 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
693 | * it. | |
1da177e4 LT |
694 | */ |
695 | int | |
f0b2efad BF |
696 | xfs_log_mount_finish( |
697 | struct xfs_mount *mp) | |
1da177e4 | 698 | { |
f661f1e0 | 699 | int error = 0; |
6f4a1eef | 700 | bool readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
f1b92bbc | 701 | bool recovered = mp->m_log->l_flags & XLOG_RECOVERY_NEEDED; |
1da177e4 | 702 | |
f0b2efad | 703 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) { |
bd186aa9 | 704 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
f0b2efad | 705 | return 0; |
6f4a1eef ES |
706 | } else if (readonly) { |
707 | /* Allow unlinked processing to proceed */ | |
708 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
1da177e4 LT |
709 | } |
710 | ||
8204f8dd DW |
711 | /* |
712 | * During the second phase of log recovery, we need iget and | |
713 | * iput to behave like they do for an active filesystem. | |
714 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
715 | * of inodes before we're done replaying log items on those | |
716 | * inodes. Turn it off immediately after recovery finishes | |
717 | * so that we don't leak the quota inodes if subsequent mount | |
718 | * activities fail. | |
799ea9e9 DW |
719 | * |
720 | * We let all inodes involved in redo item processing end up on | |
721 | * the LRU instead of being evicted immediately so that if we do | |
722 | * something to an unlinked inode, the irele won't cause | |
723 | * premature truncation and freeing of the inode, which results | |
724 | * in log recovery failure. We have to evict the unreferenced | |
1751e8a6 | 725 | * lru inodes after clearing SB_ACTIVE because we don't |
799ea9e9 DW |
726 | * otherwise clean up the lru if there's a subsequent failure in |
727 | * xfs_mountfs, which leads to us leaking the inodes if nothing | |
728 | * else (e.g. quotacheck) references the inodes before the | |
729 | * mount failure occurs. | |
8204f8dd | 730 | */ |
1751e8a6 | 731 | mp->m_super->s_flags |= SB_ACTIVE; |
f0b2efad BF |
732 | error = xlog_recover_finish(mp->m_log); |
733 | if (!error) | |
734 | xfs_log_work_queue(mp); | |
1751e8a6 | 735 | mp->m_super->s_flags &= ~SB_ACTIVE; |
799ea9e9 | 736 | evict_inodes(mp->m_super); |
f0b2efad | 737 | |
f1b92bbc BF |
738 | /* |
739 | * Drain the buffer LRU after log recovery. This is required for v4 | |
740 | * filesystems to avoid leaving around buffers with NULL verifier ops, | |
741 | * but we do it unconditionally to make sure we're always in a clean | |
742 | * cache state after mount. | |
743 | * | |
744 | * Don't push in the error case because the AIL may have pending intents | |
745 | * that aren't removed until recovery is cancelled. | |
746 | */ | |
747 | if (!error && recovered) { | |
748 | xfs_log_force(mp, XFS_LOG_SYNC); | |
749 | xfs_ail_push_all_sync(mp->m_ail); | |
750 | } | |
10fb9ac1 | 751 | xfs_buftarg_drain(mp->m_ddev_targp); |
f1b92bbc | 752 | |
6f4a1eef ES |
753 | if (readonly) |
754 | mp->m_flags |= XFS_MOUNT_RDONLY; | |
755 | ||
f0b2efad BF |
756 | return error; |
757 | } | |
758 | ||
759 | /* | |
760 | * The mount has failed. Cancel the recovery if it hasn't completed and destroy | |
761 | * the log. | |
762 | */ | |
a7a9250e | 763 | void |
f0b2efad BF |
764 | xfs_log_mount_cancel( |
765 | struct xfs_mount *mp) | |
766 | { | |
a7a9250e | 767 | xlog_recover_cancel(mp->m_log); |
f0b2efad | 768 | xfs_log_unmount(mp); |
1da177e4 LT |
769 | } |
770 | ||
81e5b50a | 771 | /* |
a79b28c2 DC |
772 | * Wait for the iclog and all prior iclogs to be written disk as required by the |
773 | * log force state machine. Waiting on ic_force_wait ensures iclog completions | |
774 | * have been ordered and callbacks run before we are woken here, hence | |
775 | * guaranteeing that all the iclogs up to this one are on stable storage. | |
81e5b50a | 776 | */ |
a79b28c2 | 777 | int |
81e5b50a CH |
778 | xlog_wait_on_iclog( |
779 | struct xlog_in_core *iclog) | |
780 | __releases(iclog->ic_log->l_icloglock) | |
781 | { | |
782 | struct xlog *log = iclog->ic_log; | |
783 | ||
784 | if (!XLOG_FORCED_SHUTDOWN(log) && | |
785 | iclog->ic_state != XLOG_STATE_ACTIVE && | |
786 | iclog->ic_state != XLOG_STATE_DIRTY) { | |
787 | XFS_STATS_INC(log->l_mp, xs_log_force_sleep); | |
788 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
789 | } else { | |
790 | spin_unlock(&log->l_icloglock); | |
791 | } | |
792 | ||
793 | if (XLOG_FORCED_SHUTDOWN(log)) | |
794 | return -EIO; | |
795 | return 0; | |
796 | } | |
797 | ||
1da177e4 | 798 | /* |
3c702f95 DC |
799 | * Write out an unmount record using the ticket provided. We have to account for |
800 | * the data space used in the unmount ticket as this write is not done from a | |
801 | * transaction context that has already done the accounting for us. | |
1da177e4 | 802 | */ |
3c702f95 DC |
803 | static int |
804 | xlog_write_unmount_record( | |
805 | struct xlog *log, | |
3468bb1c | 806 | struct xlog_ticket *ticket) |
53235f22 | 807 | { |
3c702f95 | 808 | struct xfs_unmount_log_format ulf = { |
53235f22 DW |
809 | .magic = XLOG_UNMOUNT_TYPE, |
810 | }; | |
811 | struct xfs_log_iovec reg = { | |
3c702f95 DC |
812 | .i_addr = &ulf, |
813 | .i_len = sizeof(ulf), | |
53235f22 DW |
814 | .i_type = XLOG_REG_TYPE_UNMOUNT, |
815 | }; | |
816 | struct xfs_log_vec vec = { | |
817 | .lv_niovecs = 1, | |
818 | .lv_iovecp = ®, | |
819 | }; | |
3c702f95 DC |
820 | |
821 | /* account for space used by record data */ | |
822 | ticket->t_curr_res -= sizeof(ulf); | |
eef983ff DC |
823 | |
824 | /* | |
825 | * For external log devices, we need to flush the data device cache | |
826 | * first to ensure all metadata writeback is on stable storage before we | |
827 | * stamp the tail LSN into the unmount record. | |
828 | */ | |
829 | if (log->l_targ != log->l_mp->m_ddev_targp) | |
830 | blkdev_issue_flush(log->l_targ->bt_bdev); | |
3468bb1c | 831 | return xlog_write(log, &vec, ticket, NULL, NULL, XLOG_UNMOUNT_TRANS); |
3c702f95 DC |
832 | } |
833 | ||
834 | /* | |
835 | * Mark the filesystem clean by writing an unmount record to the head of the | |
836 | * log. | |
837 | */ | |
838 | static void | |
839 | xlog_unmount_write( | |
840 | struct xlog *log) | |
841 | { | |
842 | struct xfs_mount *mp = log->l_mp; | |
53235f22 DW |
843 | struct xlog_in_core *iclog; |
844 | struct xlog_ticket *tic = NULL; | |
53235f22 DW |
845 | int error; |
846 | ||
847 | error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0); | |
848 | if (error) | |
849 | goto out_err; | |
850 | ||
3468bb1c | 851 | error = xlog_write_unmount_record(log, tic); |
53235f22 DW |
852 | /* |
853 | * At this point, we're umounting anyway, so there's no point in | |
854 | * transitioning log state to IOERROR. Just continue... | |
855 | */ | |
856 | out_err: | |
857 | if (error) | |
858 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
859 | ||
860 | spin_lock(&log->l_icloglock); | |
861 | iclog = log->l_iclog; | |
862 | atomic_inc(&iclog->ic_refcnt); | |
69363999 CH |
863 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
864 | xlog_state_switch_iclogs(log, iclog, 0); | |
865 | else | |
866 | ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
867 | iclog->ic_state == XLOG_STATE_IOERROR); | |
eef983ff DC |
868 | /* |
869 | * Ensure the journal is fully flushed and on stable storage once the | |
870 | * iclog containing the unmount record is written. | |
871 | */ | |
872 | iclog->ic_flags |= (XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA); | |
53235f22 | 873 | error = xlog_state_release_iclog(log, iclog); |
81e5b50a | 874 | xlog_wait_on_iclog(iclog); |
53235f22 DW |
875 | |
876 | if (tic) { | |
877 | trace_xfs_log_umount_write(log, tic); | |
8b41e3f9 | 878 | xfs_log_ticket_ungrant(log, tic); |
53235f22 DW |
879 | } |
880 | } | |
881 | ||
13859c98 CH |
882 | static void |
883 | xfs_log_unmount_verify_iclog( | |
884 | struct xlog *log) | |
885 | { | |
886 | struct xlog_in_core *iclog = log->l_iclog; | |
887 | ||
888 | do { | |
889 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
890 | ASSERT(iclog->ic_offset == 0); | |
891 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
892 | } | |
893 | ||
1da177e4 LT |
894 | /* |
895 | * Unmount record used to have a string "Unmount filesystem--" in the | |
896 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
897 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 898 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
899 | * As far as I know, there weren't any dependencies on the old behaviour. |
900 | */ | |
550319e9 | 901 | static void |
13859c98 CH |
902 | xfs_log_unmount_write( |
903 | struct xfs_mount *mp) | |
1da177e4 | 904 | { |
13859c98 | 905 | struct xlog *log = mp->m_log; |
1da177e4 | 906 | |
50d25484 | 907 | if (!xfs_log_writable(mp)) |
550319e9 | 908 | return; |
1da177e4 | 909 | |
550319e9 | 910 | xfs_log_force(mp, XFS_LOG_SYNC); |
1da177e4 | 911 | |
6178d104 CH |
912 | if (XLOG_FORCED_SHUTDOWN(log)) |
913 | return; | |
5cc3c006 DW |
914 | |
915 | /* | |
916 | * If we think the summary counters are bad, avoid writing the unmount | |
917 | * record to force log recovery at next mount, after which the summary | |
918 | * counters will be recalculated. Refer to xlog_check_unmount_rec for | |
919 | * more details. | |
920 | */ | |
921 | if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp, | |
922 | XFS_ERRTAG_FORCE_SUMMARY_RECALC)) { | |
923 | xfs_alert(mp, "%s: will fix summary counters at next mount", | |
924 | __func__); | |
925 | return; | |
926 | } | |
927 | ||
13859c98 | 928 | xfs_log_unmount_verify_iclog(log); |
3c702f95 | 929 | xlog_unmount_write(log); |
550319e9 | 930 | } |
1da177e4 LT |
931 | |
932 | /* | |
c75921a7 | 933 | * Empty the log for unmount/freeze. |
cf2931db DC |
934 | * |
935 | * To do this, we first need to shut down the background log work so it is not | |
936 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
937 | * the log so we can then flush them out. Once they have completed their IO and | |
303591a0 | 938 | * run the callbacks removing themselves from the AIL, we can cover the log. |
1da177e4 | 939 | */ |
303591a0 | 940 | int |
c75921a7 DC |
941 | xfs_log_quiesce( |
942 | struct xfs_mount *mp) | |
1da177e4 | 943 | { |
f661f1e0 | 944 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
945 | xfs_log_force(mp, XFS_LOG_SYNC); |
946 | ||
947 | /* | |
948 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
8321ddb2 | 949 | * will push it, xfs_buftarg_wait() will not wait for it. Further, |
cf2931db DC |
950 | * xfs_buf_iowait() cannot be used because it was pushed with the |
951 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
952 | * the IO to complete. | |
953 | */ | |
954 | xfs_ail_push_all_sync(mp->m_ail); | |
8321ddb2 | 955 | xfs_buftarg_wait(mp->m_ddev_targp); |
cf2931db DC |
956 | xfs_buf_lock(mp->m_sb_bp); |
957 | xfs_buf_unlock(mp->m_sb_bp); | |
303591a0 BF |
958 | |
959 | return xfs_log_cover(mp); | |
9e54ee0f | 960 | } |
cf2931db | 961 | |
9e54ee0f BF |
962 | void |
963 | xfs_log_clean( | |
964 | struct xfs_mount *mp) | |
965 | { | |
966 | xfs_log_quiesce(mp); | |
cf2931db | 967 | xfs_log_unmount_write(mp); |
c75921a7 DC |
968 | } |
969 | ||
970 | /* | |
971 | * Shut down and release the AIL and Log. | |
972 | * | |
973 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
974 | * from the AIL so that the log is empty before we write the unmount record to | |
975 | * the log. Once this is done, we can tear down the AIL and the log. | |
976 | */ | |
977 | void | |
978 | xfs_log_unmount( | |
979 | struct xfs_mount *mp) | |
980 | { | |
9e54ee0f | 981 | xfs_log_clean(mp); |
cf2931db | 982 | |
8321ddb2 BF |
983 | xfs_buftarg_drain(mp->m_ddev_targp); |
984 | ||
249a8c11 | 985 | xfs_trans_ail_destroy(mp); |
baff4e44 BF |
986 | |
987 | xfs_sysfs_del(&mp->m_log->l_kobj); | |
988 | ||
c41564b5 | 989 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
990 | } |
991 | ||
43f5efc5 DC |
992 | void |
993 | xfs_log_item_init( | |
994 | struct xfs_mount *mp, | |
995 | struct xfs_log_item *item, | |
996 | int type, | |
272e42b2 | 997 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
998 | { |
999 | item->li_mountp = mp; | |
1000 | item->li_ailp = mp->m_ail; | |
1001 | item->li_type = type; | |
1002 | item->li_ops = ops; | |
71e330b5 DC |
1003 | item->li_lv = NULL; |
1004 | ||
1005 | INIT_LIST_HEAD(&item->li_ail); | |
1006 | INIT_LIST_HEAD(&item->li_cil); | |
643c8c05 | 1007 | INIT_LIST_HEAD(&item->li_bio_list); |
e6631f85 | 1008 | INIT_LIST_HEAD(&item->li_trans); |
43f5efc5 DC |
1009 | } |
1010 | ||
09a423a3 CH |
1011 | /* |
1012 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 1013 | */ |
1da177e4 | 1014 | void |
09a423a3 | 1015 | xfs_log_space_wake( |
cfb7cdca | 1016 | struct xfs_mount *mp) |
1da177e4 | 1017 | { |
ad223e60 | 1018 | struct xlog *log = mp->m_log; |
cfb7cdca | 1019 | int free_bytes; |
1da177e4 | 1020 | |
1da177e4 LT |
1021 | if (XLOG_FORCED_SHUTDOWN(log)) |
1022 | return; | |
1da177e4 | 1023 | |
28496968 | 1024 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
1025 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1026 | ||
28496968 CH |
1027 | spin_lock(&log->l_write_head.lock); |
1028 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 1029 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 1030 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 1031 | } |
10547941 | 1032 | |
28496968 | 1033 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
1034 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1035 | ||
28496968 CH |
1036 | spin_lock(&log->l_reserve_head.lock); |
1037 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 1038 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 1039 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 1040 | } |
3f16b985 | 1041 | } |
1da177e4 LT |
1042 | |
1043 | /* | |
2c6e24ce DC |
1044 | * Determine if we have a transaction that has gone to disk that needs to be |
1045 | * covered. To begin the transition to the idle state firstly the log needs to | |
1046 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1047 | * we start attempting to cover the log. | |
b6f8dd49 | 1048 | * |
2c6e24ce DC |
1049 | * Only if we are then in a state where covering is needed, the caller is |
1050 | * informed that dummy transactions are required to move the log into the idle | |
1051 | * state. | |
1052 | * | |
1053 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1054 | * cover the log as we may be in a situation where there isn't log space | |
1055 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1056 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1057 | * there's no point in running a dummy transaction at this point because we | |
1058 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 | 1059 | */ |
37444fc4 BF |
1060 | static bool |
1061 | xfs_log_need_covered( | |
1062 | struct xfs_mount *mp) | |
1da177e4 | 1063 | { |
37444fc4 BF |
1064 | struct xlog *log = mp->m_log; |
1065 | bool needed = false; | |
1da177e4 | 1066 | |
2c6e24ce | 1067 | if (!xlog_cil_empty(log)) |
8646b982 | 1068 | return false; |
2c6e24ce | 1069 | |
b22cd72c | 1070 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1071 | switch (log->l_covered_state) { |
1072 | case XLOG_STATE_COVER_DONE: | |
1073 | case XLOG_STATE_COVER_DONE2: | |
1074 | case XLOG_STATE_COVER_IDLE: | |
1075 | break; | |
1076 | case XLOG_STATE_COVER_NEED: | |
1077 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1078 | if (xfs_ail_min_lsn(log->l_ailp)) |
1079 | break; | |
1080 | if (!xlog_iclogs_empty(log)) | |
1081 | break; | |
1082 | ||
37444fc4 | 1083 | needed = true; |
2c6e24ce DC |
1084 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) |
1085 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1086 | else | |
1087 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1088 | break; | |
b6f8dd49 | 1089 | default: |
37444fc4 | 1090 | needed = true; |
b6f8dd49 | 1091 | break; |
1da177e4 | 1092 | } |
b22cd72c | 1093 | spin_unlock(&log->l_icloglock); |
014c2544 | 1094 | return needed; |
1da177e4 LT |
1095 | } |
1096 | ||
303591a0 BF |
1097 | /* |
1098 | * Explicitly cover the log. This is similar to background log covering but | |
1099 | * intended for usage in quiesce codepaths. The caller is responsible to ensure | |
1100 | * the log is idle and suitable for covering. The CIL, iclog buffers and AIL | |
1101 | * must all be empty. | |
1102 | */ | |
1103 | static int | |
1104 | xfs_log_cover( | |
1105 | struct xfs_mount *mp) | |
1106 | { | |
303591a0 | 1107 | int error = 0; |
f46e5a17 | 1108 | bool need_covered; |
303591a0 | 1109 | |
4533fc63 BF |
1110 | ASSERT((xlog_cil_empty(mp->m_log) && xlog_iclogs_empty(mp->m_log) && |
1111 | !xfs_ail_min_lsn(mp->m_log->l_ailp)) || | |
303591a0 BF |
1112 | XFS_FORCED_SHUTDOWN(mp)); |
1113 | ||
1114 | if (!xfs_log_writable(mp)) | |
1115 | return 0; | |
1116 | ||
f46e5a17 BF |
1117 | /* |
1118 | * xfs_log_need_covered() is not idempotent because it progresses the | |
1119 | * state machine if the log requires covering. Therefore, we must call | |
1120 | * this function once and use the result until we've issued an sb sync. | |
1121 | * Do so first to make that abundantly clear. | |
1122 | * | |
1123 | * Fall into the covering sequence if the log needs covering or the | |
1124 | * mount has lazy superblock accounting to sync to disk. The sb sync | |
1125 | * used for covering accumulates the in-core counters, so covering | |
1126 | * handles this for us. | |
1127 | */ | |
1128 | need_covered = xfs_log_need_covered(mp); | |
1129 | if (!need_covered && !xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1130 | return 0; | |
1131 | ||
303591a0 BF |
1132 | /* |
1133 | * To cover the log, commit the superblock twice (at most) in | |
1134 | * independent checkpoints. The first serves as a reference for the | |
1135 | * tail pointer. The sync transaction and AIL push empties the AIL and | |
1136 | * updates the in-core tail to the LSN of the first checkpoint. The | |
1137 | * second commit updates the on-disk tail with the in-core LSN, | |
1138 | * covering the log. Push the AIL one more time to leave it empty, as | |
1139 | * we found it. | |
1140 | */ | |
f46e5a17 | 1141 | do { |
303591a0 BF |
1142 | error = xfs_sync_sb(mp, true); |
1143 | if (error) | |
1144 | break; | |
1145 | xfs_ail_push_all_sync(mp->m_ail); | |
f46e5a17 | 1146 | } while (xfs_log_need_covered(mp)); |
303591a0 BF |
1147 | |
1148 | return error; | |
1149 | } | |
1150 | ||
09a423a3 | 1151 | /* |
1da177e4 LT |
1152 | * We may be holding the log iclog lock upon entering this routine. |
1153 | */ | |
1154 | xfs_lsn_t | |
1c304625 | 1155 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1156 | struct xfs_mount *mp) |
1da177e4 | 1157 | { |
ad223e60 | 1158 | struct xlog *log = mp->m_log; |
1c304625 CH |
1159 | struct xfs_log_item *lip; |
1160 | xfs_lsn_t tail_lsn; | |
1161 | ||
57e80956 | 1162 | assert_spin_locked(&mp->m_ail->ail_lock); |
1da177e4 | 1163 | |
09a423a3 CH |
1164 | /* |
1165 | * To make sure we always have a valid LSN for the log tail we keep | |
1166 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1167 | * and use that when the AIL was empty. |
09a423a3 | 1168 | */ |
1c304625 CH |
1169 | lip = xfs_ail_min(mp->m_ail); |
1170 | if (lip) | |
1171 | tail_lsn = lip->li_lsn; | |
1172 | else | |
84f3c683 | 1173 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
750b9c90 | 1174 | trace_xfs_log_assign_tail_lsn(log, tail_lsn); |
1c3cb9ec | 1175 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1176 | return tail_lsn; |
1c3cb9ec | 1177 | } |
1da177e4 | 1178 | |
1c304625 CH |
1179 | xfs_lsn_t |
1180 | xlog_assign_tail_lsn( | |
1181 | struct xfs_mount *mp) | |
1182 | { | |
1183 | xfs_lsn_t tail_lsn; | |
1184 | ||
57e80956 | 1185 | spin_lock(&mp->m_ail->ail_lock); |
1c304625 | 1186 | tail_lsn = xlog_assign_tail_lsn_locked(mp); |
57e80956 | 1187 | spin_unlock(&mp->m_ail->ail_lock); |
1c304625 CH |
1188 | |
1189 | return tail_lsn; | |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* |
1193 | * Return the space in the log between the tail and the head. The head | |
1194 | * is passed in the cycle/bytes formal parms. In the special case where | |
1195 | * the reserve head has wrapped passed the tail, this calculation is no | |
1196 | * longer valid. In this case, just return 0 which means there is no space | |
1197 | * in the log. This works for all places where this function is called | |
1198 | * with the reserve head. Of course, if the write head were to ever | |
1199 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1200 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1201 | * | |
1202 | * This code also handles the case where the reservation head is behind | |
1203 | * the tail. The details of this case are described below, but the end | |
1204 | * result is that we return the size of the log as the amount of space left. | |
1205 | */ | |
a8272ce0 | 1206 | STATIC int |
a69ed03c | 1207 | xlog_space_left( |
ad223e60 | 1208 | struct xlog *log, |
c8a09ff8 | 1209 | atomic64_t *head) |
1da177e4 | 1210 | { |
a69ed03c DC |
1211 | int free_bytes; |
1212 | int tail_bytes; | |
1213 | int tail_cycle; | |
1214 | int head_cycle; | |
1215 | int head_bytes; | |
1da177e4 | 1216 | |
a69ed03c | 1217 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1218 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1219 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1220 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1221 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1222 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1223 | return 0; |
a69ed03c DC |
1224 | else if (tail_cycle < head_cycle) { |
1225 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1226 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1227 | } else { |
1228 | /* | |
1229 | * The reservation head is behind the tail. | |
1230 | * In this case we just want to return the size of the | |
1231 | * log as the amount of space left. | |
1232 | */ | |
f41febd2 | 1233 | xfs_alert(log->l_mp, "xlog_space_left: head behind tail"); |
a0fa2b67 | 1234 | xfs_alert(log->l_mp, |
f41febd2 JP |
1235 | " tail_cycle = %d, tail_bytes = %d", |
1236 | tail_cycle, tail_bytes); | |
1237 | xfs_alert(log->l_mp, | |
1238 | " GH cycle = %d, GH bytes = %d", | |
1239 | head_cycle, head_bytes); | |
1da177e4 LT |
1240 | ASSERT(0); |
1241 | free_bytes = log->l_logsize; | |
1242 | } | |
1243 | return free_bytes; | |
a69ed03c | 1244 | } |
1da177e4 LT |
1245 | |
1246 | ||
0d5a75e9 | 1247 | static void |
79b54d9b CH |
1248 | xlog_ioend_work( |
1249 | struct work_struct *work) | |
1da177e4 | 1250 | { |
79b54d9b CH |
1251 | struct xlog_in_core *iclog = |
1252 | container_of(work, struct xlog_in_core, ic_end_io_work); | |
1253 | struct xlog *log = iclog->ic_log; | |
79b54d9b | 1254 | int error; |
1da177e4 | 1255 | |
79b54d9b | 1256 | error = blk_status_to_errno(iclog->ic_bio.bi_status); |
366fc4b8 CH |
1257 | #ifdef DEBUG |
1258 | /* treat writes with injected CRC errors as failed */ | |
1259 | if (iclog->ic_fail_crc) | |
79b54d9b | 1260 | error = -EIO; |
366fc4b8 CH |
1261 | #endif |
1262 | ||
1da177e4 | 1263 | /* |
366fc4b8 | 1264 | * Race to shutdown the filesystem if we see an error. |
1da177e4 | 1265 | */ |
79b54d9b CH |
1266 | if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) { |
1267 | xfs_alert(log->l_mp, "log I/O error %d", error); | |
1268 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1da177e4 | 1269 | } |
3db296f3 | 1270 | |
12e6a0f4 | 1271 | xlog_state_done_syncing(iclog); |
79b54d9b | 1272 | bio_uninit(&iclog->ic_bio); |
9c23eccc | 1273 | |
3db296f3 | 1274 | /* |
79b54d9b CH |
1275 | * Drop the lock to signal that we are done. Nothing references the |
1276 | * iclog after this, so an unmount waiting on this lock can now tear it | |
1277 | * down safely. As such, it is unsafe to reference the iclog after the | |
1278 | * unlock as we could race with it being freed. | |
3db296f3 | 1279 | */ |
79b54d9b | 1280 | up(&iclog->ic_sema); |
c3f8fc73 | 1281 | } |
1da177e4 | 1282 | |
1da177e4 LT |
1283 | /* |
1284 | * Return size of each in-core log record buffer. | |
1285 | * | |
9da096fd | 1286 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1287 | * |
1288 | * If the filesystem blocksize is too large, we may need to choose a | |
1289 | * larger size since the directory code currently logs entire blocks. | |
1290 | */ | |
1da177e4 | 1291 | STATIC void |
9a8d2fdb MT |
1292 | xlog_get_iclog_buffer_size( |
1293 | struct xfs_mount *mp, | |
1294 | struct xlog *log) | |
1da177e4 | 1295 | { |
1cb51258 | 1296 | if (mp->m_logbufs <= 0) |
4f62282a CH |
1297 | mp->m_logbufs = XLOG_MAX_ICLOGS; |
1298 | if (mp->m_logbsize <= 0) | |
1299 | mp->m_logbsize = XLOG_BIG_RECORD_BSIZE; | |
1300 | ||
1301 | log->l_iclog_bufs = mp->m_logbufs; | |
1302 | log->l_iclog_size = mp->m_logbsize; | |
1da177e4 LT |
1303 | |
1304 | /* | |
4f62282a | 1305 | * # headers = size / 32k - one header holds cycles from 32k of data. |
1da177e4 | 1306 | */ |
4f62282a CH |
1307 | log->l_iclog_heads = |
1308 | DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE); | |
1309 | log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT; | |
1310 | } | |
1da177e4 | 1311 | |
f661f1e0 DC |
1312 | void |
1313 | xfs_log_work_queue( | |
1314 | struct xfs_mount *mp) | |
1315 | { | |
696a5620 | 1316 | queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1317 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Every sync period we need to unpin all items in the AIL and push them to | |
1322 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1323 | * indicate that the filesystem is idle. | |
1324 | */ | |
0d5a75e9 | 1325 | static void |
f661f1e0 DC |
1326 | xfs_log_worker( |
1327 | struct work_struct *work) | |
1328 | { | |
1329 | struct xlog *log = container_of(to_delayed_work(work), | |
1330 | struct xlog, l_work); | |
1331 | struct xfs_mount *mp = log->l_mp; | |
1332 | ||
1333 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
37444fc4 | 1334 | if (xfs_fs_writable(mp, SB_FREEZE_WRITE) && xfs_log_need_covered(mp)) { |
61e63ecb DC |
1335 | /* |
1336 | * Dump a transaction into the log that contains no real change. | |
1337 | * This is needed to stamp the current tail LSN into the log | |
1338 | * during the covering operation. | |
1339 | * | |
1340 | * We cannot use an inode here for this - that will push dirty | |
1341 | * state back up into the VFS and then periodic inode flushing | |
1342 | * will prevent log covering from making progress. Hence we | |
1343 | * synchronously log the superblock instead to ensure the | |
1344 | * superblock is immediately unpinned and can be written back. | |
1345 | */ | |
1346 | xfs_sync_sb(mp, true); | |
1347 | } else | |
f661f1e0 DC |
1348 | xfs_log_force(mp, 0); |
1349 | ||
1350 | /* start pushing all the metadata that is currently dirty */ | |
1351 | xfs_ail_push_all(mp->m_ail); | |
1352 | ||
1353 | /* queue us up again */ | |
1354 | xfs_log_work_queue(mp); | |
1355 | } | |
1356 | ||
1da177e4 LT |
1357 | /* |
1358 | * This routine initializes some of the log structure for a given mount point. | |
1359 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1360 | * some other stuff may be filled in too. | |
1361 | */ | |
9a8d2fdb MT |
1362 | STATIC struct xlog * |
1363 | xlog_alloc_log( | |
1364 | struct xfs_mount *mp, | |
1365 | struct xfs_buftarg *log_target, | |
1366 | xfs_daddr_t blk_offset, | |
1367 | int num_bblks) | |
1da177e4 | 1368 | { |
9a8d2fdb | 1369 | struct xlog *log; |
1da177e4 LT |
1370 | xlog_rec_header_t *head; |
1371 | xlog_in_core_t **iclogp; | |
1372 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1da177e4 | 1373 | int i; |
2451337d | 1374 | int error = -ENOMEM; |
69ce58f0 | 1375 | uint log2_size = 0; |
1da177e4 | 1376 | |
9a8d2fdb | 1377 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1378 | if (!log) { |
a0fa2b67 | 1379 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1380 | goto out; |
1381 | } | |
1da177e4 LT |
1382 | |
1383 | log->l_mp = mp; | |
1384 | log->l_targ = log_target; | |
1385 | log->l_logsize = BBTOB(num_bblks); | |
1386 | log->l_logBBstart = blk_offset; | |
1387 | log->l_logBBsize = num_bblks; | |
1388 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1389 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1390 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1391 | |
1392 | log->l_prev_block = -1; | |
1da177e4 | 1393 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1394 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1395 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1396 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 | 1397 | |
a6a65fef DC |
1398 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) |
1399 | log->l_iclog_roundoff = mp->m_sb.sb_logsunit; | |
1400 | else | |
1401 | log->l_iclog_roundoff = BBSIZE; | |
1402 | ||
c303c5b8 CH |
1403 | xlog_grant_head_init(&log->l_reserve_head); |
1404 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1405 | |
2451337d | 1406 | error = -EFSCORRUPTED; |
62118709 | 1407 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1408 | log2_size = mp->m_sb.sb_logsectlog; |
1409 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1410 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1411 | log2_size, BBSHIFT); | |
a6cb767e DC |
1412 | goto out_free_log; |
1413 | } | |
1414 | ||
69ce58f0 AE |
1415 | log2_size -= BBSHIFT; |
1416 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1417 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1418 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1419 | goto out_free_log; |
1420 | } | |
69ce58f0 AE |
1421 | |
1422 | /* for larger sector sizes, must have v2 or external log */ | |
1423 | if (log2_size && log->l_logBBstart > 0 && | |
1424 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1425 | xfs_warn(mp, |
1426 | "log sector size (0x%x) invalid for configuration.", | |
1427 | log2_size); | |
a6cb767e DC |
1428 | goto out_free_log; |
1429 | } | |
1da177e4 | 1430 | } |
69ce58f0 | 1431 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1432 | |
1433 | xlog_get_iclog_buffer_size(mp, log); | |
1434 | ||
007c61c6 | 1435 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1436 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1437 | |
1da177e4 LT |
1438 | iclogp = &log->l_iclog; |
1439 | /* | |
1440 | * The amount of memory to allocate for the iclog structure is | |
1441 | * rather funky due to the way the structure is defined. It is | |
1442 | * done this way so that we can use different sizes for machines | |
1443 | * with different amounts of memory. See the definition of | |
1444 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1445 | */ | |
1da177e4 | 1446 | ASSERT(log->l_iclog_size >= 4096); |
79b54d9b | 1447 | for (i = 0; i < log->l_iclog_bufs; i++) { |
f8f9ee47 | 1448 | int align_mask = xfs_buftarg_dma_alignment(mp->m_logdev_targp); |
89b171ac CH |
1449 | size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE) * |
1450 | sizeof(struct bio_vec); | |
79b54d9b CH |
1451 | |
1452 | iclog = kmem_zalloc(sizeof(*iclog) + bvec_size, KM_MAYFAIL); | |
1453 | if (!iclog) | |
644c3567 DC |
1454 | goto out_free_iclog; |
1455 | ||
79b54d9b | 1456 | *iclogp = iclog; |
1da177e4 LT |
1457 | iclog->ic_prev = prev_iclog; |
1458 | prev_iclog = iclog; | |
1fa40b01 | 1459 | |
f8f9ee47 | 1460 | iclog->ic_data = kmem_alloc_io(log->l_iclog_size, align_mask, |
3219e8cf | 1461 | KM_MAYFAIL | KM_ZERO); |
79b54d9b | 1462 | if (!iclog->ic_data) |
644c3567 | 1463 | goto out_free_iclog; |
4679b2d3 | 1464 | #ifdef DEBUG |
5809d5e0 | 1465 | log->l_iclog_bak[i] = &iclog->ic_header; |
4679b2d3 | 1466 | #endif |
1da177e4 LT |
1467 | head = &iclog->ic_header; |
1468 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1469 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1470 | head->h_version = cpu_to_be32( | |
62118709 | 1471 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1472 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1473 | /* new fields */ |
b53e675d | 1474 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1475 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1476 | ||
79b54d9b | 1477 | iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize; |
1da177e4 LT |
1478 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1479 | iclog->ic_log = log; | |
114d23aa DC |
1480 | atomic_set(&iclog->ic_refcnt, 0); |
1481 | spin_lock_init(&iclog->ic_callback_lock); | |
89ae379d | 1482 | INIT_LIST_HEAD(&iclog->ic_callbacks); |
b28708d6 | 1483 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1484 | |
eb40a875 DC |
1485 | init_waitqueue_head(&iclog->ic_force_wait); |
1486 | init_waitqueue_head(&iclog->ic_write_wait); | |
79b54d9b CH |
1487 | INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work); |
1488 | sema_init(&iclog->ic_sema, 1); | |
1da177e4 LT |
1489 | |
1490 | iclogp = &iclog->ic_next; | |
1491 | } | |
1492 | *iclogp = log->l_iclog; /* complete ring */ | |
1493 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1494 | ||
1058d0f5 | 1495 | log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s", |
05a302a1 DW |
1496 | XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | |
1497 | WQ_HIGHPRI), | |
1498 | 0, mp->m_super->s_id); | |
1058d0f5 CH |
1499 | if (!log->l_ioend_workqueue) |
1500 | goto out_free_iclog; | |
1501 | ||
71e330b5 DC |
1502 | error = xlog_cil_init(log); |
1503 | if (error) | |
1058d0f5 | 1504 | goto out_destroy_workqueue; |
1da177e4 | 1505 | return log; |
644c3567 | 1506 | |
1058d0f5 CH |
1507 | out_destroy_workqueue: |
1508 | destroy_workqueue(log->l_ioend_workqueue); | |
644c3567 DC |
1509 | out_free_iclog: |
1510 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1511 | prev_iclog = iclog->ic_next; | |
79b54d9b | 1512 | kmem_free(iclog->ic_data); |
644c3567 | 1513 | kmem_free(iclog); |
798a9cad BF |
1514 | if (prev_iclog == log->l_iclog) |
1515 | break; | |
644c3567 | 1516 | } |
644c3567 DC |
1517 | out_free_log: |
1518 | kmem_free(log); | |
a6cb767e | 1519 | out: |
2451337d | 1520 | return ERR_PTR(error); |
1da177e4 LT |
1521 | } /* xlog_alloc_log */ |
1522 | ||
1da177e4 LT |
1523 | /* |
1524 | * Write out the commit record of a transaction associated with the given | |
f10e925d | 1525 | * ticket to close off a running log write. Return the lsn of the commit record. |
1da177e4 | 1526 | */ |
f10e925d | 1527 | int |
55b66332 | 1528 | xlog_commit_record( |
ad223e60 | 1529 | struct xlog *log, |
55b66332 DC |
1530 | struct xlog_ticket *ticket, |
1531 | struct xlog_in_core **iclog, | |
f10e925d | 1532 | xfs_lsn_t *lsn) |
1da177e4 | 1533 | { |
55b66332 DC |
1534 | struct xfs_log_iovec reg = { |
1535 | .i_addr = NULL, | |
1536 | .i_len = 0, | |
1537 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1538 | }; | |
1539 | struct xfs_log_vec vec = { | |
1540 | .lv_niovecs = 1, | |
1541 | .lv_iovecp = ®, | |
1542 | }; | |
f10e925d | 1543 | int error; |
1da177e4 | 1544 | |
f10e925d DC |
1545 | if (XLOG_FORCED_SHUTDOWN(log)) |
1546 | return -EIO; | |
1547 | ||
3468bb1c | 1548 | error = xlog_write(log, &vec, ticket, lsn, iclog, XLOG_COMMIT_TRANS); |
55b66332 | 1549 | if (error) |
f10e925d | 1550 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1551 | return error; |
55b66332 | 1552 | } |
1da177e4 LT |
1553 | |
1554 | /* | |
ed1575da DW |
1555 | * Compute the LSN that we'd need to push the log tail towards in order to have |
1556 | * (a) enough on-disk log space to log the number of bytes specified, (b) at | |
1557 | * least 25% of the log space free, and (c) at least 256 blocks free. If the | |
1558 | * log free space already meets all three thresholds, this function returns | |
1559 | * NULLCOMMITLSN. | |
1da177e4 | 1560 | */ |
ed1575da DW |
1561 | xfs_lsn_t |
1562 | xlog_grant_push_threshold( | |
ad223e60 | 1563 | struct xlog *log, |
2ced19cb | 1564 | int need_bytes) |
1da177e4 | 1565 | { |
2ced19cb | 1566 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1567 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1568 | int free_blocks; |
1569 | int free_bytes; | |
1570 | int threshold_block; | |
1571 | int threshold_cycle; | |
1572 | int free_threshold; | |
1573 | ||
1574 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1575 | ||
28496968 | 1576 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1577 | free_blocks = BTOBBT(free_bytes); |
1578 | ||
1579 | /* | |
1580 | * Set the threshold for the minimum number of free blocks in the | |
1581 | * log to the maximum of what the caller needs, one quarter of the | |
1582 | * log, and 256 blocks. | |
1583 | */ | |
1584 | free_threshold = BTOBB(need_bytes); | |
9bb54cb5 DC |
1585 | free_threshold = max(free_threshold, (log->l_logBBsize >> 2)); |
1586 | free_threshold = max(free_threshold, 256); | |
2ced19cb | 1587 | if (free_blocks >= free_threshold) |
ed1575da | 1588 | return NULLCOMMITLSN; |
2ced19cb | 1589 | |
1c3cb9ec DC |
1590 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1591 | &threshold_block); | |
1592 | threshold_block += free_threshold; | |
1da177e4 | 1593 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1594 | threshold_block -= log->l_logBBsize; |
1595 | threshold_cycle += 1; | |
1da177e4 | 1596 | } |
2ced19cb DC |
1597 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1598 | threshold_block); | |
1599 | /* | |
1600 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1601 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1602 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1603 | */ |
84f3c683 DC |
1604 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1605 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1606 | threshold_lsn = last_sync_lsn; | |
2ced19cb | 1607 | |
ed1575da DW |
1608 | return threshold_lsn; |
1609 | } | |
1610 | ||
1611 | /* | |
1612 | * Push the tail of the log if we need to do so to maintain the free log space | |
1613 | * thresholds set out by xlog_grant_push_threshold. We may need to adopt a | |
1614 | * policy which pushes on an lsn which is further along in the log once we | |
1615 | * reach the high water mark. In this manner, we would be creating a low water | |
1616 | * mark. | |
1617 | */ | |
1618 | STATIC void | |
1619 | xlog_grant_push_ail( | |
1620 | struct xlog *log, | |
1621 | int need_bytes) | |
1622 | { | |
1623 | xfs_lsn_t threshold_lsn; | |
1624 | ||
1625 | threshold_lsn = xlog_grant_push_threshold(log, need_bytes); | |
1626 | if (threshold_lsn == NULLCOMMITLSN || XLOG_FORCED_SHUTDOWN(log)) | |
1627 | return; | |
1628 | ||
2ced19cb DC |
1629 | /* |
1630 | * Get the transaction layer to kick the dirty buffers out to | |
1631 | * disk asynchronously. No point in trying to do this if | |
1632 | * the filesystem is shutting down. | |
1633 | */ | |
ed1575da | 1634 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1635 | } |
1da177e4 | 1636 | |
0e446be4 CH |
1637 | /* |
1638 | * Stamp cycle number in every block | |
1639 | */ | |
1640 | STATIC void | |
1641 | xlog_pack_data( | |
1642 | struct xlog *log, | |
1643 | struct xlog_in_core *iclog, | |
1644 | int roundoff) | |
1645 | { | |
1646 | int i, j, k; | |
1647 | int size = iclog->ic_offset + roundoff; | |
1648 | __be32 cycle_lsn; | |
b2a922cd | 1649 | char *dp; |
0e446be4 CH |
1650 | |
1651 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1652 | ||
1653 | dp = iclog->ic_datap; | |
1654 | for (i = 0; i < BTOBB(size); i++) { | |
1655 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1656 | break; | |
1657 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1658 | *(__be32 *)dp = cycle_lsn; | |
1659 | dp += BBSIZE; | |
1660 | } | |
1661 | ||
1662 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1663 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1664 | ||
1665 | for ( ; i < BTOBB(size); i++) { | |
1666 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1667 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1668 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1669 | *(__be32 *)dp = cycle_lsn; | |
1670 | dp += BBSIZE; | |
1671 | } | |
1672 | ||
1673 | for (i = 1; i < log->l_iclog_heads; i++) | |
1674 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1675 | } | |
1676 | } | |
1677 | ||
1678 | /* | |
1679 | * Calculate the checksum for a log buffer. | |
1680 | * | |
1681 | * This is a little more complicated than it should be because the various | |
1682 | * headers and the actual data are non-contiguous. | |
1683 | */ | |
f9668a09 | 1684 | __le32 |
0e446be4 CH |
1685 | xlog_cksum( |
1686 | struct xlog *log, | |
1687 | struct xlog_rec_header *rhead, | |
1688 | char *dp, | |
1689 | int size) | |
1690 | { | |
c8ce540d | 1691 | uint32_t crc; |
0e446be4 CH |
1692 | |
1693 | /* first generate the crc for the record header ... */ | |
cae028df | 1694 | crc = xfs_start_cksum_update((char *)rhead, |
0e446be4 CH |
1695 | sizeof(struct xlog_rec_header), |
1696 | offsetof(struct xlog_rec_header, h_crc)); | |
1697 | ||
1698 | /* ... then for additional cycle data for v2 logs ... */ | |
1699 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1700 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1701 | int i; | |
a3f20014 | 1702 | int xheads; |
0e446be4 | 1703 | |
0c771b99 | 1704 | xheads = DIV_ROUND_UP(size, XLOG_HEADER_CYCLE_SIZE); |
0e446be4 | 1705 | |
a3f20014 | 1706 | for (i = 1; i < xheads; i++) { |
0e446be4 CH |
1707 | crc = crc32c(crc, &xhdr[i].hic_xheader, |
1708 | sizeof(struct xlog_rec_ext_header)); | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | /* ... and finally for the payload */ | |
1713 | crc = crc32c(crc, dp, size); | |
1714 | ||
1715 | return xfs_end_cksum(crc); | |
1716 | } | |
1717 | ||
79b54d9b CH |
1718 | static void |
1719 | xlog_bio_end_io( | |
1720 | struct bio *bio) | |
1721 | { | |
1722 | struct xlog_in_core *iclog = bio->bi_private; | |
1723 | ||
1058d0f5 | 1724 | queue_work(iclog->ic_log->l_ioend_workqueue, |
79b54d9b CH |
1725 | &iclog->ic_end_io_work); |
1726 | } | |
1727 | ||
842a42d1 | 1728 | static int |
79b54d9b CH |
1729 | xlog_map_iclog_data( |
1730 | struct bio *bio, | |
1731 | void *data, | |
1732 | size_t count) | |
1733 | { | |
1734 | do { | |
1735 | struct page *page = kmem_to_page(data); | |
1736 | unsigned int off = offset_in_page(data); | |
1737 | size_t len = min_t(size_t, count, PAGE_SIZE - off); | |
1738 | ||
842a42d1 BF |
1739 | if (bio_add_page(bio, page, len, off) != len) |
1740 | return -EIO; | |
79b54d9b CH |
1741 | |
1742 | data += len; | |
1743 | count -= len; | |
1744 | } while (count); | |
842a42d1 BF |
1745 | |
1746 | return 0; | |
79b54d9b CH |
1747 | } |
1748 | ||
94860a30 CH |
1749 | STATIC void |
1750 | xlog_write_iclog( | |
1751 | struct xlog *log, | |
1752 | struct xlog_in_core *iclog, | |
94860a30 | 1753 | uint64_t bno, |
eef983ff | 1754 | unsigned int count) |
873ff550 | 1755 | { |
94860a30 | 1756 | ASSERT(bno < log->l_logBBsize); |
94860a30 CH |
1757 | |
1758 | /* | |
1759 | * We lock the iclogbufs here so that we can serialise against I/O | |
1760 | * completion during unmount. We might be processing a shutdown | |
1761 | * triggered during unmount, and that can occur asynchronously to the | |
1762 | * unmount thread, and hence we need to ensure that completes before | |
1763 | * tearing down the iclogbufs. Hence we need to hold the buffer lock | |
1764 | * across the log IO to archieve that. | |
1765 | */ | |
79b54d9b | 1766 | down(&iclog->ic_sema); |
1858bb0b | 1767 | if (unlikely(iclog->ic_state == XLOG_STATE_IOERROR)) { |
873ff550 CH |
1768 | /* |
1769 | * It would seem logical to return EIO here, but we rely on | |
1770 | * the log state machine to propagate I/O errors instead of | |
79b54d9b CH |
1771 | * doing it here. We kick of the state machine and unlock |
1772 | * the buffer manually, the code needs to be kept in sync | |
1773 | * with the I/O completion path. | |
873ff550 | 1774 | */ |
12e6a0f4 | 1775 | xlog_state_done_syncing(iclog); |
79b54d9b | 1776 | up(&iclog->ic_sema); |
94860a30 | 1777 | return; |
873ff550 CH |
1778 | } |
1779 | ||
79b54d9b CH |
1780 | bio_init(&iclog->ic_bio, iclog->ic_bvec, howmany(count, PAGE_SIZE)); |
1781 | bio_set_dev(&iclog->ic_bio, log->l_targ->bt_bdev); | |
1782 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno; | |
1783 | iclog->ic_bio.bi_end_io = xlog_bio_end_io; | |
1784 | iclog->ic_bio.bi_private = iclog; | |
2def2845 DC |
1785 | |
1786 | /* | |
1787 | * We use REQ_SYNC | REQ_IDLE here to tell the block layer the are more | |
1788 | * IOs coming immediately after this one. This prevents the block layer | |
1789 | * writeback throttle from throttling log writes behind background | |
1790 | * metadata writeback and causing priority inversions. | |
1791 | */ | |
eef983ff DC |
1792 | iclog->ic_bio.bi_opf = REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_IDLE; |
1793 | if (iclog->ic_flags & XLOG_ICL_NEED_FLUSH) | |
79b54d9b | 1794 | iclog->ic_bio.bi_opf |= REQ_PREFLUSH; |
eef983ff DC |
1795 | if (iclog->ic_flags & XLOG_ICL_NEED_FUA) |
1796 | iclog->ic_bio.bi_opf |= REQ_FUA; | |
1797 | iclog->ic_flags &= ~(XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA); | |
79b54d9b | 1798 | |
842a42d1 BF |
1799 | if (xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, count)) { |
1800 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1801 | return; | |
1802 | } | |
79b54d9b | 1803 | if (is_vmalloc_addr(iclog->ic_data)) |
2c68a1df | 1804 | flush_kernel_vmap_range(iclog->ic_data, count); |
79b54d9b CH |
1805 | |
1806 | /* | |
1807 | * If this log buffer would straddle the end of the log we will have | |
1808 | * to split it up into two bios, so that we can continue at the start. | |
1809 | */ | |
1810 | if (bno + BTOBB(count) > log->l_logBBsize) { | |
1811 | struct bio *split; | |
1812 | ||
1813 | split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno, | |
1814 | GFP_NOIO, &fs_bio_set); | |
1815 | bio_chain(split, &iclog->ic_bio); | |
1816 | submit_bio(split); | |
1817 | ||
1818 | /* restart at logical offset zero for the remainder */ | |
1819 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart; | |
1820 | } | |
1821 | ||
1822 | submit_bio(&iclog->ic_bio); | |
873ff550 | 1823 | } |
1da177e4 | 1824 | |
56933848 CH |
1825 | /* |
1826 | * We need to bump cycle number for the part of the iclog that is | |
1827 | * written to the start of the log. Watch out for the header magic | |
1828 | * number case, though. | |
1829 | */ | |
79b54d9b | 1830 | static void |
56933848 CH |
1831 | xlog_split_iclog( |
1832 | struct xlog *log, | |
1833 | void *data, | |
1834 | uint64_t bno, | |
1835 | unsigned int count) | |
1836 | { | |
1837 | unsigned int split_offset = BBTOB(log->l_logBBsize - bno); | |
1838 | unsigned int i; | |
1839 | ||
1840 | for (i = split_offset; i < count; i += BBSIZE) { | |
1841 | uint32_t cycle = get_unaligned_be32(data + i); | |
1842 | ||
1843 | if (++cycle == XLOG_HEADER_MAGIC_NUM) | |
1844 | cycle++; | |
1845 | put_unaligned_be32(cycle, data + i); | |
1846 | } | |
56933848 CH |
1847 | } |
1848 | ||
db0a6faf CH |
1849 | static int |
1850 | xlog_calc_iclog_size( | |
1851 | struct xlog *log, | |
1852 | struct xlog_in_core *iclog, | |
1853 | uint32_t *roundoff) | |
1854 | { | |
1855 | uint32_t count_init, count; | |
db0a6faf CH |
1856 | |
1857 | /* Add for LR header */ | |
1858 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
a6a65fef | 1859 | count = roundup(count_init, log->l_iclog_roundoff); |
db0a6faf | 1860 | |
db0a6faf CH |
1861 | *roundoff = count - count_init; |
1862 | ||
a6a65fef DC |
1863 | ASSERT(count >= count_init); |
1864 | ASSERT(*roundoff < log->l_iclog_roundoff); | |
db0a6faf CH |
1865 | return count; |
1866 | } | |
1867 | ||
1da177e4 LT |
1868 | /* |
1869 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1870 | * fashion. Previously, we should have moved the current iclog | |
1871 | * ptr in the log to point to the next available iclog. This allows further | |
1872 | * write to continue while this code syncs out an iclog ready to go. | |
1873 | * Before an in-core log can be written out, the data section must be scanned | |
1874 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1875 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1876 | * cycle count because there in an implicit assumption that drives will | |
1877 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1878 | * we can't have part of a 512 byte block written and part not written. By | |
1879 | * tagging each block, we will know which blocks are valid when recovering | |
1880 | * after an unclean shutdown. | |
1881 | * | |
1882 | * This routine is single threaded on the iclog. No other thread can be in | |
1883 | * this routine with the same iclog. Changing contents of iclog can there- | |
1884 | * fore be done without grabbing the state machine lock. Updating the global | |
1885 | * log will require grabbing the lock though. | |
1886 | * | |
1887 | * The entire log manager uses a logical block numbering scheme. Only | |
94860a30 CH |
1888 | * xlog_write_iclog knows about the fact that the log may not start with |
1889 | * block zero on a given device. | |
1da177e4 | 1890 | */ |
94860a30 | 1891 | STATIC void |
9a8d2fdb MT |
1892 | xlog_sync( |
1893 | struct xlog *log, | |
1894 | struct xlog_in_core *iclog) | |
1da177e4 | 1895 | { |
db0a6faf CH |
1896 | unsigned int count; /* byte count of bwrite */ |
1897 | unsigned int roundoff; /* roundoff to BB or stripe */ | |
1898 | uint64_t bno; | |
db0a6faf | 1899 | unsigned int size; |
1da177e4 | 1900 | |
155cc6b7 | 1901 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 | 1902 | |
db0a6faf | 1903 | count = xlog_calc_iclog_size(log, iclog, &roundoff); |
1da177e4 LT |
1904 | |
1905 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1906 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1907 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1908 | |
1909 | /* put cycle number in every block */ | |
1910 | xlog_pack_data(log, iclog, roundoff); | |
1911 | ||
1912 | /* real byte length */ | |
0e446be4 | 1913 | size = iclog->ic_offset; |
db0a6faf | 1914 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) |
0e446be4 CH |
1915 | size += roundoff; |
1916 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 1917 | |
9b0489c1 | 1918 | XFS_STATS_INC(log->l_mp, xs_log_writes); |
ff6d6af2 | 1919 | XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count)); |
1da177e4 | 1920 | |
94860a30 CH |
1921 | bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)); |
1922 | ||
1da177e4 | 1923 | /* Do we need to split this write into 2 parts? */ |
eef983ff | 1924 | if (bno + BTOBB(count) > log->l_logBBsize) |
79b54d9b | 1925 | xlog_split_iclog(log, &iclog->ic_header, bno, count); |
0e446be4 CH |
1926 | |
1927 | /* calculcate the checksum */ | |
1928 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
1929 | iclog->ic_datap, size); | |
609adfc2 BF |
1930 | /* |
1931 | * Intentionally corrupt the log record CRC based on the error injection | |
1932 | * frequency, if defined. This facilitates testing log recovery in the | |
1933 | * event of torn writes. Hence, set the IOABORT state to abort the log | |
1934 | * write on I/O completion and shutdown the fs. The subsequent mount | |
1935 | * detects the bad CRC and attempts to recover. | |
1936 | */ | |
366fc4b8 | 1937 | #ifdef DEBUG |
3e88a007 | 1938 | if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) { |
e2a64192 | 1939 | iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA); |
366fc4b8 | 1940 | iclog->ic_fail_crc = true; |
609adfc2 BF |
1941 | xfs_warn(log->l_mp, |
1942 | "Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.", | |
1943 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1944 | } | |
366fc4b8 | 1945 | #endif |
abca1f33 | 1946 | xlog_verify_iclog(log, iclog, count); |
eef983ff | 1947 | xlog_write_iclog(log, iclog, bno, count); |
94860a30 | 1948 | } |
1da177e4 | 1949 | |
1da177e4 | 1950 | /* |
c41564b5 | 1951 | * Deallocate a log structure |
1da177e4 | 1952 | */ |
a8272ce0 | 1953 | STATIC void |
9a8d2fdb MT |
1954 | xlog_dealloc_log( |
1955 | struct xlog *log) | |
1da177e4 LT |
1956 | { |
1957 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1958 | int i; |
1959 | ||
71e330b5 DC |
1960 | xlog_cil_destroy(log); |
1961 | ||
44396476 | 1962 | /* |
9c23eccc DC |
1963 | * Cycle all the iclogbuf locks to make sure all log IO completion |
1964 | * is done before we tear down these buffers. | |
1965 | */ | |
1966 | iclog = log->l_iclog; | |
1967 | for (i = 0; i < log->l_iclog_bufs; i++) { | |
79b54d9b CH |
1968 | down(&iclog->ic_sema); |
1969 | up(&iclog->ic_sema); | |
9c23eccc DC |
1970 | iclog = iclog->ic_next; |
1971 | } | |
1972 | ||
1da177e4 | 1973 | iclog = log->l_iclog; |
9c23eccc | 1974 | for (i = 0; i < log->l_iclog_bufs; i++) { |
1da177e4 | 1975 | next_iclog = iclog->ic_next; |
79b54d9b | 1976 | kmem_free(iclog->ic_data); |
f0e2d93c | 1977 | kmem_free(iclog); |
1da177e4 LT |
1978 | iclog = next_iclog; |
1979 | } | |
1da177e4 | 1980 | |
1da177e4 | 1981 | log->l_mp->m_log = NULL; |
1058d0f5 | 1982 | destroy_workqueue(log->l_ioend_workqueue); |
f0e2d93c | 1983 | kmem_free(log); |
b843299b | 1984 | } |
1da177e4 LT |
1985 | |
1986 | /* | |
1987 | * Update counters atomically now that memcpy is done. | |
1988 | */ | |
1da177e4 | 1989 | static inline void |
9a8d2fdb MT |
1990 | xlog_state_finish_copy( |
1991 | struct xlog *log, | |
1992 | struct xlog_in_core *iclog, | |
1993 | int record_cnt, | |
1994 | int copy_bytes) | |
1da177e4 | 1995 | { |
390aab0a | 1996 | lockdep_assert_held(&log->l_icloglock); |
1da177e4 | 1997 | |
413d57c9 | 1998 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 | 1999 | iclog->ic_offset += copy_bytes; |
390aab0a | 2000 | } |
1da177e4 | 2001 | |
7e9c6396 TS |
2002 | /* |
2003 | * print out info relating to regions written which consume | |
2004 | * the reservation | |
2005 | */ | |
71e330b5 DC |
2006 | void |
2007 | xlog_print_tic_res( | |
2008 | struct xfs_mount *mp, | |
2009 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
2010 | { |
2011 | uint i; | |
2012 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
2013 | ||
2014 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
5110cd82 | 2015 | #define REG_TYPE_STR(type, str) [XLOG_REG_TYPE_##type] = str |
d31d7185 | 2016 | static char *res_type_str[] = { |
5110cd82 DW |
2017 | REG_TYPE_STR(BFORMAT, "bformat"), |
2018 | REG_TYPE_STR(BCHUNK, "bchunk"), | |
2019 | REG_TYPE_STR(EFI_FORMAT, "efi_format"), | |
2020 | REG_TYPE_STR(EFD_FORMAT, "efd_format"), | |
2021 | REG_TYPE_STR(IFORMAT, "iformat"), | |
2022 | REG_TYPE_STR(ICORE, "icore"), | |
2023 | REG_TYPE_STR(IEXT, "iext"), | |
2024 | REG_TYPE_STR(IBROOT, "ibroot"), | |
2025 | REG_TYPE_STR(ILOCAL, "ilocal"), | |
2026 | REG_TYPE_STR(IATTR_EXT, "iattr_ext"), | |
2027 | REG_TYPE_STR(IATTR_BROOT, "iattr_broot"), | |
2028 | REG_TYPE_STR(IATTR_LOCAL, "iattr_local"), | |
2029 | REG_TYPE_STR(QFORMAT, "qformat"), | |
2030 | REG_TYPE_STR(DQUOT, "dquot"), | |
2031 | REG_TYPE_STR(QUOTAOFF, "quotaoff"), | |
2032 | REG_TYPE_STR(LRHEADER, "LR header"), | |
2033 | REG_TYPE_STR(UNMOUNT, "unmount"), | |
2034 | REG_TYPE_STR(COMMIT, "commit"), | |
2035 | REG_TYPE_STR(TRANSHDR, "trans header"), | |
d31d7185 DW |
2036 | REG_TYPE_STR(ICREATE, "inode create"), |
2037 | REG_TYPE_STR(RUI_FORMAT, "rui_format"), | |
2038 | REG_TYPE_STR(RUD_FORMAT, "rud_format"), | |
2039 | REG_TYPE_STR(CUI_FORMAT, "cui_format"), | |
2040 | REG_TYPE_STR(CUD_FORMAT, "cud_format"), | |
2041 | REG_TYPE_STR(BUI_FORMAT, "bui_format"), | |
2042 | REG_TYPE_STR(BUD_FORMAT, "bud_format"), | |
7e9c6396 | 2043 | }; |
d31d7185 | 2044 | BUILD_BUG_ON(ARRAY_SIZE(res_type_str) != XLOG_REG_TYPE_MAX + 1); |
5110cd82 | 2045 | #undef REG_TYPE_STR |
7e9c6396 | 2046 | |
7d2d5653 | 2047 | xfs_warn(mp, "ticket reservation summary:"); |
f41febd2 JP |
2048 | xfs_warn(mp, " unit res = %d bytes", |
2049 | ticket->t_unit_res); | |
2050 | xfs_warn(mp, " current res = %d bytes", | |
2051 | ticket->t_curr_res); | |
2052 | xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)", | |
2053 | ticket->t_res_arr_sum, ticket->t_res_o_flow); | |
2054 | xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)", | |
2055 | ticket->t_res_num_ophdrs, ophdr_spc); | |
2056 | xfs_warn(mp, " ophdr + reg = %u bytes", | |
2057 | ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc); | |
2058 | xfs_warn(mp, " num regions = %u", | |
2059 | ticket->t_res_num); | |
7e9c6396 TS |
2060 | |
2061 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 2062 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 2063 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 | 2064 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
5110cd82 | 2065 | "bad-rtype" : res_type_str[r_type]), |
7e9c6396 TS |
2066 | ticket->t_res_arr[i].r_len); |
2067 | } | |
2068 | } | |
7e9c6396 | 2069 | |
d4ca1d55 BF |
2070 | /* |
2071 | * Print a summary of the transaction. | |
2072 | */ | |
2073 | void | |
2074 | xlog_print_trans( | |
e6631f85 | 2075 | struct xfs_trans *tp) |
d4ca1d55 | 2076 | { |
e6631f85 DC |
2077 | struct xfs_mount *mp = tp->t_mountp; |
2078 | struct xfs_log_item *lip; | |
d4ca1d55 BF |
2079 | |
2080 | /* dump core transaction and ticket info */ | |
2081 | xfs_warn(mp, "transaction summary:"); | |
2c8f6265 BF |
2082 | xfs_warn(mp, " log res = %d", tp->t_log_res); |
2083 | xfs_warn(mp, " log count = %d", tp->t_log_count); | |
2084 | xfs_warn(mp, " flags = 0x%x", tp->t_flags); | |
d4ca1d55 BF |
2085 | |
2086 | xlog_print_tic_res(mp, tp->t_ticket); | |
2087 | ||
2088 | /* dump each log item */ | |
e6631f85 | 2089 | list_for_each_entry(lip, &tp->t_items, li_trans) { |
d4ca1d55 BF |
2090 | struct xfs_log_vec *lv = lip->li_lv; |
2091 | struct xfs_log_iovec *vec; | |
2092 | int i; | |
2093 | ||
2094 | xfs_warn(mp, "log item: "); | |
2095 | xfs_warn(mp, " type = 0x%x", lip->li_type); | |
22525c17 | 2096 | xfs_warn(mp, " flags = 0x%lx", lip->li_flags); |
d4ca1d55 BF |
2097 | if (!lv) |
2098 | continue; | |
2099 | xfs_warn(mp, " niovecs = %d", lv->lv_niovecs); | |
2100 | xfs_warn(mp, " size = %d", lv->lv_size); | |
2101 | xfs_warn(mp, " bytes = %d", lv->lv_bytes); | |
2102 | xfs_warn(mp, " buf len = %d", lv->lv_buf_len); | |
2103 | ||
2104 | /* dump each iovec for the log item */ | |
2105 | vec = lv->lv_iovecp; | |
2106 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2107 | int dumplen = min(vec->i_len, 32); | |
2108 | ||
2109 | xfs_warn(mp, " iovec[%d]", i); | |
2110 | xfs_warn(mp, " type = 0x%x", vec->i_type); | |
2111 | xfs_warn(mp, " len = %d", vec->i_len); | |
2112 | xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i); | |
244e3dea | 2113 | xfs_hex_dump(vec->i_addr, dumplen); |
d4ca1d55 BF |
2114 | |
2115 | vec++; | |
2116 | } | |
2117 | } | |
2118 | } | |
2119 | ||
b5203cd0 | 2120 | /* |
7ec94921 DC |
2121 | * Calculate the potential space needed by the log vector. We may need a start |
2122 | * record, and each region gets its own struct xlog_op_header and may need to be | |
2123 | * double word aligned. | |
b5203cd0 DC |
2124 | */ |
2125 | static int | |
2126 | xlog_write_calc_vec_length( | |
2127 | struct xlog_ticket *ticket, | |
7ec94921 | 2128 | struct xfs_log_vec *log_vector, |
3468bb1c | 2129 | uint optype) |
b5203cd0 | 2130 | { |
55b66332 | 2131 | struct xfs_log_vec *lv; |
3468bb1c | 2132 | int headers = 0; |
b5203cd0 DC |
2133 | int len = 0; |
2134 | int i; | |
2135 | ||
3468bb1c DC |
2136 | if (optype & XLOG_START_TRANS) |
2137 | headers++; | |
2138 | ||
55b66332 | 2139 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2140 | /* we don't write ordered log vectors */ |
2141 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2142 | continue; | |
2143 | ||
55b66332 DC |
2144 | headers += lv->lv_niovecs; |
2145 | ||
2146 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2147 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2148 | |
55b66332 DC |
2149 | len += vecp->i_len; |
2150 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2151 | } | |
b5203cd0 DC |
2152 | } |
2153 | ||
2154 | ticket->t_res_num_ophdrs += headers; | |
2155 | len += headers * sizeof(struct xlog_op_header); | |
2156 | ||
2157 | return len; | |
2158 | } | |
2159 | ||
7ec94921 | 2160 | static void |
b5203cd0 | 2161 | xlog_write_start_rec( |
e6b1f273 | 2162 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2163 | struct xlog_ticket *ticket) |
2164 | { | |
b5203cd0 DC |
2165 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2166 | ophdr->oh_clientid = ticket->t_clientid; | |
2167 | ophdr->oh_len = 0; | |
2168 | ophdr->oh_flags = XLOG_START_TRANS; | |
2169 | ophdr->oh_res2 = 0; | |
b5203cd0 DC |
2170 | } |
2171 | ||
2172 | static xlog_op_header_t * | |
2173 | xlog_write_setup_ophdr( | |
ad223e60 | 2174 | struct xlog *log, |
e6b1f273 | 2175 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2176 | struct xlog_ticket *ticket, |
2177 | uint flags) | |
2178 | { | |
b5203cd0 DC |
2179 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2180 | ophdr->oh_clientid = ticket->t_clientid; | |
2181 | ophdr->oh_res2 = 0; | |
2182 | ||
2183 | /* are we copying a commit or unmount record? */ | |
2184 | ophdr->oh_flags = flags; | |
2185 | ||
2186 | /* | |
2187 | * We've seen logs corrupted with bad transaction client ids. This | |
2188 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2189 | * and shut down the filesystem. | |
2190 | */ | |
2191 | switch (ophdr->oh_clientid) { | |
2192 | case XFS_TRANSACTION: | |
2193 | case XFS_VOLUME: | |
2194 | case XFS_LOG: | |
2195 | break; | |
2196 | default: | |
a0fa2b67 | 2197 | xfs_warn(log->l_mp, |
c9690043 | 2198 | "Bad XFS transaction clientid 0x%x in ticket "PTR_FMT, |
b5203cd0 DC |
2199 | ophdr->oh_clientid, ticket); |
2200 | return NULL; | |
2201 | } | |
2202 | ||
2203 | return ophdr; | |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * Set up the parameters of the region copy into the log. This has | |
2208 | * to handle region write split across multiple log buffers - this | |
2209 | * state is kept external to this function so that this code can | |
ac0e300f | 2210 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2211 | */ |
2212 | static int | |
2213 | xlog_write_setup_copy( | |
2214 | struct xlog_ticket *ticket, | |
2215 | struct xlog_op_header *ophdr, | |
2216 | int space_available, | |
2217 | int space_required, | |
2218 | int *copy_off, | |
2219 | int *copy_len, | |
2220 | int *last_was_partial_copy, | |
2221 | int *bytes_consumed) | |
2222 | { | |
2223 | int still_to_copy; | |
2224 | ||
2225 | still_to_copy = space_required - *bytes_consumed; | |
2226 | *copy_off = *bytes_consumed; | |
2227 | ||
2228 | if (still_to_copy <= space_available) { | |
2229 | /* write of region completes here */ | |
2230 | *copy_len = still_to_copy; | |
2231 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2232 | if (*last_was_partial_copy) | |
2233 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2234 | *last_was_partial_copy = 0; | |
2235 | *bytes_consumed = 0; | |
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | /* partial write of region, needs extra log op header reservation */ | |
2240 | *copy_len = space_available; | |
2241 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2242 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2243 | if (*last_was_partial_copy) | |
2244 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2245 | *bytes_consumed += *copy_len; | |
2246 | (*last_was_partial_copy)++; | |
2247 | ||
2248 | /* account for new log op header */ | |
2249 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2250 | ticket->t_res_num_ophdrs++; | |
2251 | ||
2252 | return sizeof(struct xlog_op_header); | |
2253 | } | |
2254 | ||
2255 | static int | |
2256 | xlog_write_copy_finish( | |
ad223e60 | 2257 | struct xlog *log, |
b5203cd0 DC |
2258 | struct xlog_in_core *iclog, |
2259 | uint flags, | |
2260 | int *record_cnt, | |
2261 | int *data_cnt, | |
2262 | int *partial_copy, | |
2263 | int *partial_copy_len, | |
2264 | int log_offset, | |
2265 | struct xlog_in_core **commit_iclog) | |
2266 | { | |
df732b29 CH |
2267 | int error; |
2268 | ||
b5203cd0 DC |
2269 | if (*partial_copy) { |
2270 | /* | |
2271 | * This iclog has already been marked WANT_SYNC by | |
2272 | * xlog_state_get_iclog_space. | |
2273 | */ | |
390aab0a | 2274 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2275 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2276 | *record_cnt = 0; | |
2277 | *data_cnt = 0; | |
df732b29 | 2278 | goto release_iclog; |
b5203cd0 DC |
2279 | } |
2280 | ||
2281 | *partial_copy = 0; | |
2282 | *partial_copy_len = 0; | |
2283 | ||
2284 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2285 | /* no more space in this iclog - push it. */ | |
390aab0a | 2286 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2287 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2288 | *record_cnt = 0; | |
2289 | *data_cnt = 0; | |
2290 | ||
69363999 CH |
2291 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
2292 | xlog_state_switch_iclogs(log, iclog, 0); | |
2293 | else | |
2294 | ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
2295 | iclog->ic_state == XLOG_STATE_IOERROR); | |
b5203cd0 | 2296 | if (!commit_iclog) |
df732b29 CH |
2297 | goto release_iclog; |
2298 | spin_unlock(&log->l_icloglock); | |
b5203cd0 DC |
2299 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2300 | *commit_iclog = iclog; | |
2301 | } | |
2302 | ||
2303 | return 0; | |
df732b29 CH |
2304 | |
2305 | release_iclog: | |
2306 | error = xlog_state_release_iclog(log, iclog); | |
2307 | spin_unlock(&log->l_icloglock); | |
2308 | return error; | |
b5203cd0 DC |
2309 | } |
2310 | ||
1da177e4 LT |
2311 | /* |
2312 | * Write some region out to in-core log | |
2313 | * | |
2314 | * This will be called when writing externally provided regions or when | |
2315 | * writing out a commit record for a given transaction. | |
2316 | * | |
2317 | * General algorithm: | |
2318 | * 1. Find total length of this write. This may include adding to the | |
2319 | * lengths passed in. | |
2320 | * 2. Check whether we violate the tickets reservation. | |
2321 | * 3. While writing to this iclog | |
2322 | * A. Reserve as much space in this iclog as can get | |
2323 | * B. If this is first write, save away start lsn | |
2324 | * C. While writing this region: | |
2325 | * 1. If first write of transaction, write start record | |
2326 | * 2. Write log operation header (header per region) | |
2327 | * 3. Find out if we can fit entire region into this iclog | |
2328 | * 4. Potentially, verify destination memcpy ptr | |
2329 | * 5. Memcpy (partial) region | |
2330 | * 6. If partial copy, release iclog; otherwise, continue | |
2331 | * copying more regions into current iclog | |
2332 | * 4. Mark want sync bit (in simulation mode) | |
2333 | * 5. Release iclog for potential flush to on-disk log. | |
2334 | * | |
2335 | * ERRORS: | |
2336 | * 1. Panic if reservation is overrun. This should never happen since | |
2337 | * reservation amounts are generated internal to the filesystem. | |
2338 | * NOTES: | |
2339 | * 1. Tickets are single threaded data structures. | |
2340 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2341 | * syncing routine. When a single log_write region needs to span | |
2342 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2343 | * on all log operation writes which don't contain the end of the | |
2344 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2345 | * operation which contains the end of the continued log_write region. | |
2346 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2347 | * we don't really know exactly how much space will be used. As a result, | |
2348 | * we don't update ic_offset until the end when we know exactly how many | |
2349 | * bytes have been written out. | |
2350 | */ | |
71e330b5 | 2351 | int |
35a8a72f | 2352 | xlog_write( |
ad223e60 | 2353 | struct xlog *log, |
55b66332 | 2354 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2355 | struct xlog_ticket *ticket, |
2356 | xfs_lsn_t *start_lsn, | |
2357 | struct xlog_in_core **commit_iclog, | |
3468bb1c | 2358 | uint optype) |
1da177e4 | 2359 | { |
99428ad0 | 2360 | struct xlog_in_core *iclog = NULL; |
9590e9c6 DC |
2361 | struct xfs_log_vec *lv = log_vector; |
2362 | struct xfs_log_iovec *vecp = lv->lv_iovecp; | |
2363 | int index = 0; | |
99428ad0 | 2364 | int len; |
99428ad0 CH |
2365 | int partial_copy = 0; |
2366 | int partial_copy_len = 0; | |
2367 | int contwr = 0; | |
2368 | int record_cnt = 0; | |
2369 | int data_cnt = 0; | |
df732b29 | 2370 | int error = 0; |
99428ad0 | 2371 | |
93b8a585 | 2372 | /* |
9590e9c6 DC |
2373 | * If this is a commit or unmount transaction, we don't need a start |
2374 | * record to be written. We do, however, have to account for the | |
2375 | * commit or unmount header that gets written. Hence we always have | |
2376 | * to account for an extra xlog_op_header here. | |
93b8a585 | 2377 | */ |
9590e9c6 | 2378 | ticket->t_curr_res -= sizeof(struct xlog_op_header); |
7d2d5653 BF |
2379 | if (ticket->t_curr_res < 0) { |
2380 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
2381 | "ctx ticket reservation ran out. Need to up reservation"); | |
55b66332 | 2382 | xlog_print_tic_res(log->l_mp, ticket); |
7d2d5653 BF |
2383 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
2384 | } | |
1da177e4 | 2385 | |
3468bb1c DC |
2386 | len = xlog_write_calc_vec_length(ticket, log_vector, optype); |
2387 | if (start_lsn) | |
2388 | *start_lsn = 0; | |
fd63875c | 2389 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2390 | void *ptr; |
99428ad0 | 2391 | int log_offset; |
1da177e4 | 2392 | |
99428ad0 CH |
2393 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2394 | &contwr, &log_offset); | |
2395 | if (error) | |
2396 | return error; | |
1da177e4 | 2397 | |
99428ad0 | 2398 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2399 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2400 | |
eef983ff | 2401 | /* Start_lsn is the first lsn written to. */ |
3468bb1c | 2402 | if (start_lsn && !*start_lsn) |
99428ad0 | 2403 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
b5203cd0 | 2404 | |
99428ad0 CH |
2405 | /* |
2406 | * This loop writes out as many regions as can fit in the amount | |
2407 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2408 | */ | |
fd63875c DC |
2409 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2410 | struct xfs_log_iovec *reg; | |
99428ad0 | 2411 | struct xlog_op_header *ophdr; |
99428ad0 CH |
2412 | int copy_len; |
2413 | int copy_off; | |
fd63875c | 2414 | bool ordered = false; |
3468bb1c | 2415 | bool wrote_start_rec = false; |
fd63875c DC |
2416 | |
2417 | /* ordered log vectors have no regions to write */ | |
2418 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2419 | ASSERT(lv->lv_niovecs == 0); | |
2420 | ordered = true; | |
2421 | goto next_lv; | |
2422 | } | |
99428ad0 | 2423 | |
fd63875c | 2424 | reg = &vecp[index]; |
c8ce540d DW |
2425 | ASSERT(reg->i_len % sizeof(int32_t) == 0); |
2426 | ASSERT((unsigned long)ptr % sizeof(int32_t) == 0); | |
99428ad0 | 2427 | |
7ec94921 DC |
2428 | /* |
2429 | * Before we start formatting log vectors, we need to | |
2430 | * write a start record. Only do this for the first | |
2431 | * iclog we write to. | |
2432 | */ | |
3468bb1c | 2433 | if (optype & XLOG_START_TRANS) { |
7ec94921 | 2434 | xlog_write_start_rec(ptr, ticket); |
e6b1f273 | 2435 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
7ec94921 | 2436 | sizeof(struct xlog_op_header)); |
3468bb1c DC |
2437 | optype &= ~XLOG_START_TRANS; |
2438 | wrote_start_rec = true; | |
99428ad0 | 2439 | } |
b5203cd0 | 2440 | |
3468bb1c | 2441 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, optype); |
99428ad0 | 2442 | if (!ophdr) |
2451337d | 2443 | return -EIO; |
99428ad0 | 2444 | |
e6b1f273 | 2445 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2446 | sizeof(struct xlog_op_header)); |
2447 | ||
2448 | len += xlog_write_setup_copy(ticket, ophdr, | |
2449 | iclog->ic_size-log_offset, | |
55b66332 | 2450 | reg->i_len, |
99428ad0 CH |
2451 | ©_off, ©_len, |
2452 | &partial_copy, | |
2453 | &partial_copy_len); | |
2454 | xlog_verify_dest_ptr(log, ptr); | |
2455 | ||
91f9f5fe ES |
2456 | /* |
2457 | * Copy region. | |
2458 | * | |
2459 | * Unmount records just log an opheader, so can have | |
2460 | * empty payloads with no data region to copy. Hence we | |
2461 | * only copy the payload if the vector says it has data | |
2462 | * to copy. | |
2463 | */ | |
99428ad0 | 2464 | ASSERT(copy_len >= 0); |
91f9f5fe ES |
2465 | if (copy_len > 0) { |
2466 | memcpy(ptr, reg->i_addr + copy_off, copy_len); | |
2467 | xlog_write_adv_cnt(&ptr, &len, &log_offset, | |
2468 | copy_len); | |
2469 | } | |
7ec94921 | 2470 | copy_len += sizeof(struct xlog_op_header); |
99428ad0 | 2471 | record_cnt++; |
3468bb1c | 2472 | if (wrote_start_rec) { |
7ec94921 DC |
2473 | copy_len += sizeof(struct xlog_op_header); |
2474 | record_cnt++; | |
7ec94921 | 2475 | } |
99428ad0 CH |
2476 | data_cnt += contwr ? copy_len : 0; |
2477 | ||
3468bb1c | 2478 | error = xlog_write_copy_finish(log, iclog, optype, |
99428ad0 CH |
2479 | &record_cnt, &data_cnt, |
2480 | &partial_copy, | |
2481 | &partial_copy_len, | |
2482 | log_offset, | |
2483 | commit_iclog); | |
2484 | if (error) | |
2485 | return error; | |
2486 | ||
2487 | /* | |
2488 | * if we had a partial copy, we need to get more iclog | |
2489 | * space but we don't want to increment the region | |
2490 | * index because there is still more is this region to | |
2491 | * write. | |
2492 | * | |
2493 | * If we completed writing this region, and we flushed | |
2494 | * the iclog (indicated by resetting of the record | |
2495 | * count), then we also need to get more log space. If | |
2496 | * this was the last record, though, we are done and | |
2497 | * can just return. | |
2498 | */ | |
2499 | if (partial_copy) | |
2500 | break; | |
2501 | ||
55b66332 | 2502 | if (++index == lv->lv_niovecs) { |
fd63875c | 2503 | next_lv: |
55b66332 DC |
2504 | lv = lv->lv_next; |
2505 | index = 0; | |
2506 | if (lv) | |
2507 | vecp = lv->lv_iovecp; | |
2508 | } | |
749f24f3 | 2509 | if (record_cnt == 0 && !ordered) { |
55b66332 | 2510 | if (!lv) |
99428ad0 CH |
2511 | return 0; |
2512 | break; | |
2513 | } | |
2514 | } | |
2515 | } | |
2516 | ||
2517 | ASSERT(len == 0); | |
2518 | ||
390aab0a | 2519 | spin_lock(&log->l_icloglock); |
99428ad0 | 2520 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); |
df732b29 | 2521 | if (commit_iclog) { |
3468bb1c | 2522 | ASSERT(optype & XLOG_COMMIT_TRANS); |
df732b29 CH |
2523 | *commit_iclog = iclog; |
2524 | } else { | |
2525 | error = xlog_state_release_iclog(log, iclog); | |
2526 | } | |
390aab0a | 2527 | spin_unlock(&log->l_icloglock); |
1da177e4 | 2528 | |
df732b29 | 2529 | return error; |
99428ad0 | 2530 | } |
1da177e4 | 2531 | |
c814b4f2 CH |
2532 | static void |
2533 | xlog_state_activate_iclog( | |
2534 | struct xlog_in_core *iclog, | |
2535 | int *iclogs_changed) | |
2536 | { | |
2537 | ASSERT(list_empty_careful(&iclog->ic_callbacks)); | |
2538 | ||
2539 | /* | |
2540 | * If the number of ops in this iclog indicate it just contains the | |
2541 | * dummy transaction, we can change state into IDLE (the second time | |
2542 | * around). Otherwise we should change the state into NEED a dummy. | |
2543 | * We don't need to cover the dummy. | |
2544 | */ | |
2545 | if (*iclogs_changed == 0 && | |
2546 | iclog->ic_header.h_num_logops == cpu_to_be32(XLOG_COVER_OPS)) { | |
2547 | *iclogs_changed = 1; | |
2548 | } else { | |
2549 | /* | |
2550 | * We have two dirty iclogs so start over. This could also be | |
2551 | * num of ops indicating this is not the dummy going out. | |
2552 | */ | |
2553 | *iclogs_changed = 2; | |
2554 | } | |
2555 | ||
2556 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2557 | iclog->ic_offset = 0; | |
2558 | iclog->ic_header.h_num_logops = 0; | |
2559 | memset(iclog->ic_header.h_cycle_data, 0, | |
2560 | sizeof(iclog->ic_header.h_cycle_data)); | |
2561 | iclog->ic_header.h_lsn = 0; | |
2562 | } | |
2563 | ||
0383f543 | 2564 | /* |
c814b4f2 CH |
2565 | * Loop through all iclogs and mark all iclogs currently marked DIRTY as |
2566 | * ACTIVE after iclog I/O has completed. | |
1da177e4 | 2567 | */ |
c814b4f2 CH |
2568 | static void |
2569 | xlog_state_activate_iclogs( | |
0383f543 | 2570 | struct xlog *log, |
c814b4f2 | 2571 | int *iclogs_changed) |
1da177e4 | 2572 | { |
c814b4f2 | 2573 | struct xlog_in_core *iclog = log->l_iclog; |
1da177e4 | 2574 | |
1da177e4 | 2575 | do { |
c814b4f2 CH |
2576 | if (iclog->ic_state == XLOG_STATE_DIRTY) |
2577 | xlog_state_activate_iclog(iclog, iclogs_changed); | |
2578 | /* | |
2579 | * The ordering of marking iclogs ACTIVE must be maintained, so | |
2580 | * an iclog doesn't become ACTIVE beyond one that is SYNCING. | |
2581 | */ | |
2582 | else if (iclog->ic_state != XLOG_STATE_ACTIVE) | |
2583 | break; | |
2584 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
2585 | } | |
0383f543 | 2586 | |
c814b4f2 CH |
2587 | static int |
2588 | xlog_covered_state( | |
2589 | int prev_state, | |
2590 | int iclogs_changed) | |
2591 | { | |
0383f543 | 2592 | /* |
b0eb9e11 BF |
2593 | * We go to NEED for any non-covering writes. We go to NEED2 if we just |
2594 | * wrote the first covering record (DONE). We go to IDLE if we just | |
2595 | * wrote the second covering record (DONE2) and remain in IDLE until a | |
2596 | * non-covering write occurs. | |
0383f543 | 2597 | */ |
c814b4f2 CH |
2598 | switch (prev_state) { |
2599 | case XLOG_STATE_COVER_IDLE: | |
b0eb9e11 BF |
2600 | if (iclogs_changed == 1) |
2601 | return XLOG_STATE_COVER_IDLE; | |
c814b4f2 CH |
2602 | case XLOG_STATE_COVER_NEED: |
2603 | case XLOG_STATE_COVER_NEED2: | |
2604 | break; | |
2605 | case XLOG_STATE_COVER_DONE: | |
2606 | if (iclogs_changed == 1) | |
2607 | return XLOG_STATE_COVER_NEED2; | |
2608 | break; | |
2609 | case XLOG_STATE_COVER_DONE2: | |
2610 | if (iclogs_changed == 1) | |
2611 | return XLOG_STATE_COVER_IDLE; | |
2612 | break; | |
2613 | default: | |
2614 | ASSERT(0); | |
2615 | } | |
0383f543 | 2616 | |
c814b4f2 CH |
2617 | return XLOG_STATE_COVER_NEED; |
2618 | } | |
1da177e4 | 2619 | |
c814b4f2 CH |
2620 | STATIC void |
2621 | xlog_state_clean_iclog( | |
2622 | struct xlog *log, | |
2623 | struct xlog_in_core *dirty_iclog) | |
2624 | { | |
2625 | int iclogs_changed = 0; | |
1da177e4 | 2626 | |
5781464b | 2627 | dirty_iclog->ic_state = XLOG_STATE_DIRTY; |
1da177e4 | 2628 | |
c814b4f2 CH |
2629 | xlog_state_activate_iclogs(log, &iclogs_changed); |
2630 | wake_up_all(&dirty_iclog->ic_force_wait); | |
2631 | ||
2632 | if (iclogs_changed) { | |
2633 | log->l_covered_state = xlog_covered_state(log->l_covered_state, | |
2634 | iclogs_changed); | |
1da177e4 | 2635 | } |
0383f543 | 2636 | } |
1da177e4 LT |
2637 | |
2638 | STATIC xfs_lsn_t | |
2639 | xlog_get_lowest_lsn( | |
9bff3132 | 2640 | struct xlog *log) |
1da177e4 | 2641 | { |
9bff3132 CH |
2642 | struct xlog_in_core *iclog = log->l_iclog; |
2643 | xfs_lsn_t lowest_lsn = 0, lsn; | |
1da177e4 | 2644 | |
1da177e4 | 2645 | do { |
1858bb0b CH |
2646 | if (iclog->ic_state == XLOG_STATE_ACTIVE || |
2647 | iclog->ic_state == XLOG_STATE_DIRTY) | |
9bff3132 CH |
2648 | continue; |
2649 | ||
2650 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2651 | if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0) | |
1da177e4 | 2652 | lowest_lsn = lsn; |
9bff3132 CH |
2653 | } while ((iclog = iclog->ic_next) != log->l_iclog); |
2654 | ||
014c2544 | 2655 | return lowest_lsn; |
1da177e4 LT |
2656 | } |
2657 | ||
14e15f1b DC |
2658 | /* |
2659 | * Completion of a iclog IO does not imply that a transaction has completed, as | |
2660 | * transactions can be large enough to span many iclogs. We cannot change the | |
2661 | * tail of the log half way through a transaction as this may be the only | |
2662 | * transaction in the log and moving the tail to point to the middle of it | |
2663 | * will prevent recovery from finding the start of the transaction. Hence we | |
2664 | * should only update the last_sync_lsn if this iclog contains transaction | |
2665 | * completion callbacks on it. | |
2666 | * | |
2667 | * We have to do this before we drop the icloglock to ensure we are the only one | |
2668 | * that can update it. | |
2669 | * | |
2670 | * If we are moving the last_sync_lsn forwards, we also need to ensure we kick | |
2671 | * the reservation grant head pushing. This is due to the fact that the push | |
2672 | * target is bound by the current last_sync_lsn value. Hence if we have a large | |
2673 | * amount of log space bound up in this committing transaction then the | |
2674 | * last_sync_lsn value may be the limiting factor preventing tail pushing from | |
2675 | * freeing space in the log. Hence once we've updated the last_sync_lsn we | |
2676 | * should push the AIL to ensure the push target (and hence the grant head) is | |
2677 | * no longer bound by the old log head location and can move forwards and make | |
2678 | * progress again. | |
2679 | */ | |
2680 | static void | |
2681 | xlog_state_set_callback( | |
2682 | struct xlog *log, | |
2683 | struct xlog_in_core *iclog, | |
2684 | xfs_lsn_t header_lsn) | |
2685 | { | |
2686 | iclog->ic_state = XLOG_STATE_CALLBACK; | |
2687 | ||
2688 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), | |
2689 | header_lsn) <= 0); | |
2690 | ||
2691 | if (list_empty_careful(&iclog->ic_callbacks)) | |
2692 | return; | |
2693 | ||
2694 | atomic64_set(&log->l_last_sync_lsn, header_lsn); | |
2695 | xlog_grant_push_ail(log, 0); | |
2696 | } | |
2697 | ||
5e96fa8d DC |
2698 | /* |
2699 | * Return true if we need to stop processing, false to continue to the next | |
2700 | * iclog. The caller will need to run callbacks if the iclog is returned in the | |
2701 | * XLOG_STATE_CALLBACK state. | |
2702 | */ | |
2703 | static bool | |
2704 | xlog_state_iodone_process_iclog( | |
2705 | struct xlog *log, | |
2706 | struct xlog_in_core *iclog, | |
5e96fa8d DC |
2707 | bool *ioerror) |
2708 | { | |
2709 | xfs_lsn_t lowest_lsn; | |
14e15f1b | 2710 | xfs_lsn_t header_lsn; |
5e96fa8d | 2711 | |
1858bb0b CH |
2712 | switch (iclog->ic_state) { |
2713 | case XLOG_STATE_ACTIVE: | |
2714 | case XLOG_STATE_DIRTY: | |
2715 | /* | |
2716 | * Skip all iclogs in the ACTIVE & DIRTY states: | |
2717 | */ | |
5e96fa8d | 2718 | return false; |
1858bb0b CH |
2719 | case XLOG_STATE_IOERROR: |
2720 | /* | |
2721 | * Between marking a filesystem SHUTDOWN and stopping the log, | |
2722 | * we do flush all iclogs to disk (if there wasn't a log I/O | |
2723 | * error). So, we do want things to go smoothly in case of just | |
4b29ab04 | 2724 | * a SHUTDOWN w/o a LOG_IO_ERROR. |
1858bb0b | 2725 | */ |
5e96fa8d DC |
2726 | *ioerror = true; |
2727 | return false; | |
1858bb0b | 2728 | case XLOG_STATE_DONE_SYNC: |
1858bb0b | 2729 | /* |
4b29ab04 CH |
2730 | * Now that we have an iclog that is in the DONE_SYNC state, do |
2731 | * one more check here to see if we have chased our tail around. | |
2732 | * If this is not the lowest lsn iclog, then we will leave it | |
2733 | * for another completion to process. | |
1858bb0b CH |
2734 | */ |
2735 | header_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2736 | lowest_lsn = xlog_get_lowest_lsn(log); | |
2737 | if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, header_lsn) < 0) | |
2738 | return false; | |
2739 | xlog_state_set_callback(log, iclog, header_lsn); | |
2740 | return false; | |
2741 | default: | |
2742 | /* | |
2743 | * Can only perform callbacks in order. Since this iclog is not | |
4b29ab04 CH |
2744 | * in the DONE_SYNC state, we skip the rest and just try to |
2745 | * clean up. | |
1858bb0b | 2746 | */ |
5e96fa8d DC |
2747 | return true; |
2748 | } | |
5e96fa8d DC |
2749 | } |
2750 | ||
6546818c DC |
2751 | /* |
2752 | * Keep processing entries in the iclog callback list until we come around and | |
2753 | * it is empty. We need to atomically see that the list is empty and change the | |
2754 | * state to DIRTY so that we don't miss any more callbacks being added. | |
2755 | * | |
2756 | * This function is called with the icloglock held and returns with it held. We | |
2757 | * drop it while running callbacks, however, as holding it over thousands of | |
2758 | * callbacks is unnecessary and causes excessive contention if we do. | |
2759 | */ | |
2760 | static void | |
2761 | xlog_state_do_iclog_callbacks( | |
2762 | struct xlog *log, | |
12e6a0f4 | 2763 | struct xlog_in_core *iclog) |
f7559793 DW |
2764 | __releases(&log->l_icloglock) |
2765 | __acquires(&log->l_icloglock) | |
6546818c DC |
2766 | { |
2767 | spin_unlock(&log->l_icloglock); | |
2768 | spin_lock(&iclog->ic_callback_lock); | |
2769 | while (!list_empty(&iclog->ic_callbacks)) { | |
2770 | LIST_HEAD(tmp); | |
2771 | ||
2772 | list_splice_init(&iclog->ic_callbacks, &tmp); | |
2773 | ||
2774 | spin_unlock(&iclog->ic_callback_lock); | |
12e6a0f4 | 2775 | xlog_cil_process_committed(&tmp); |
6546818c DC |
2776 | spin_lock(&iclog->ic_callback_lock); |
2777 | } | |
2778 | ||
2779 | /* | |
2780 | * Pick up the icloglock while still holding the callback lock so we | |
2781 | * serialise against anyone trying to add more callbacks to this iclog | |
2782 | * now we've finished processing. | |
2783 | */ | |
2784 | spin_lock(&log->l_icloglock); | |
2785 | spin_unlock(&iclog->ic_callback_lock); | |
2786 | } | |
2787 | ||
1da177e4 LT |
2788 | STATIC void |
2789 | xlog_state_do_callback( | |
12e6a0f4 | 2790 | struct xlog *log) |
1da177e4 | 2791 | { |
5e96fa8d DC |
2792 | struct xlog_in_core *iclog; |
2793 | struct xlog_in_core *first_iclog; | |
5e96fa8d DC |
2794 | bool cycled_icloglock; |
2795 | bool ioerror; | |
2796 | int flushcnt = 0; | |
2797 | int repeats = 0; | |
1da177e4 | 2798 | |
b22cd72c | 2799 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2800 | do { |
2801 | /* | |
2802 | * Scan all iclogs starting with the one pointed to by the | |
2803 | * log. Reset this starting point each time the log is | |
2804 | * unlocked (during callbacks). | |
2805 | * | |
2806 | * Keep looping through iclogs until one full pass is made | |
2807 | * without running any callbacks. | |
2808 | */ | |
2809 | first_iclog = log->l_iclog; | |
2810 | iclog = log->l_iclog; | |
6546818c | 2811 | cycled_icloglock = false; |
5e96fa8d | 2812 | ioerror = false; |
1da177e4 LT |
2813 | repeats++; |
2814 | ||
2815 | do { | |
5e96fa8d | 2816 | if (xlog_state_iodone_process_iclog(log, iclog, |
4b29ab04 | 2817 | &ioerror)) |
5e96fa8d | 2818 | break; |
1da177e4 | 2819 | |
1858bb0b CH |
2820 | if (iclog->ic_state != XLOG_STATE_CALLBACK && |
2821 | iclog->ic_state != XLOG_STATE_IOERROR) { | |
1da177e4 LT |
2822 | iclog = iclog->ic_next; |
2823 | continue; | |
2824 | } | |
2825 | ||
114d23aa | 2826 | /* |
6546818c DC |
2827 | * Running callbacks will drop the icloglock which means |
2828 | * we'll have to run at least one more complete loop. | |
114d23aa | 2829 | */ |
6546818c | 2830 | cycled_icloglock = true; |
12e6a0f4 | 2831 | xlog_state_do_iclog_callbacks(log, iclog); |
5781464b CH |
2832 | if (XLOG_FORCED_SHUTDOWN(log)) |
2833 | wake_up_all(&iclog->ic_force_wait); | |
2834 | else | |
2835 | xlog_state_clean_iclog(log, iclog); | |
1da177e4 LT |
2836 | iclog = iclog->ic_next; |
2837 | } while (first_iclog != iclog); | |
a3c6685e NS |
2838 | |
2839 | if (repeats > 5000) { | |
2840 | flushcnt += repeats; | |
2841 | repeats = 0; | |
a0fa2b67 | 2842 | xfs_warn(log->l_mp, |
a3c6685e | 2843 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2844 | __func__, flushcnt); |
1da177e4 | 2845 | } |
5e96fa8d | 2846 | } while (!ioerror && cycled_icloglock); |
1da177e4 | 2847 | |
1858bb0b CH |
2848 | if (log->l_iclog->ic_state == XLOG_STATE_ACTIVE || |
2849 | log->l_iclog->ic_state == XLOG_STATE_IOERROR) | |
eb40a875 | 2850 | wake_up_all(&log->l_flush_wait); |
cdea5459 RR |
2851 | |
2852 | spin_unlock(&log->l_icloglock); | |
d748c623 | 2853 | } |
1da177e4 LT |
2854 | |
2855 | ||
2856 | /* | |
2857 | * Finish transitioning this iclog to the dirty state. | |
2858 | * | |
2859 | * Make sure that we completely execute this routine only when this is | |
2860 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2861 | * when we reach the end of the physical log, get turned into 2 separate | |
2862 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2863 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2864 | * the second completion goes through. | |
2865 | * | |
2866 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2867 | * global state machine log lock. |
1da177e4 | 2868 | */ |
a8272ce0 | 2869 | STATIC void |
1da177e4 | 2870 | xlog_state_done_syncing( |
12e6a0f4 | 2871 | struct xlog_in_core *iclog) |
1da177e4 | 2872 | { |
d15cbf2f | 2873 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2874 | |
b22cd72c | 2875 | spin_lock(&log->l_icloglock); |
155cc6b7 | 2876 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
2877 | |
2878 | /* | |
2879 | * If we got an error, either on the first buffer, or in the case of | |
12e6a0f4 CH |
2880 | * split log writes, on the second, we shut down the file system and |
2881 | * no iclogs should ever be attempted to be written to disk again. | |
1da177e4 | 2882 | */ |
12e6a0f4 CH |
2883 | if (!XLOG_FORCED_SHUTDOWN(log)) { |
2884 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING); | |
1da177e4 | 2885 | iclog->ic_state = XLOG_STATE_DONE_SYNC; |
12e6a0f4 | 2886 | } |
1da177e4 LT |
2887 | |
2888 | /* | |
2889 | * Someone could be sleeping prior to writing out the next | |
2890 | * iclog buffer, we wake them all, one will get to do the | |
2891 | * I/O, the others get to wait for the result. | |
2892 | */ | |
eb40a875 | 2893 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2894 | spin_unlock(&log->l_icloglock); |
b843299b | 2895 | xlog_state_do_callback(log); |
12e6a0f4 | 2896 | } |
1da177e4 LT |
2897 | |
2898 | /* | |
2899 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2900 | * sleep. We wait on the flush queue on the head iclog as that should be |
2901 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2902 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2903 | * |
2904 | * The in-core logs are used in a circular fashion. They are not used | |
2905 | * out-of-order even when an iclog past the head is free. | |
2906 | * | |
2907 | * return: | |
2908 | * * log_offset where xlog_write() can start writing into the in-core | |
2909 | * log's data space. | |
2910 | * * in-core log pointer to which xlog_write() should write. | |
2911 | * * boolean indicating this is a continued write to an in-core log. | |
2912 | * If this is the last write, then the in-core log's offset field | |
2913 | * needs to be incremented, depending on the amount of data which | |
2914 | * is copied. | |
2915 | */ | |
a8272ce0 | 2916 | STATIC int |
9a8d2fdb MT |
2917 | xlog_state_get_iclog_space( |
2918 | struct xlog *log, | |
2919 | int len, | |
2920 | struct xlog_in_core **iclogp, | |
2921 | struct xlog_ticket *ticket, | |
2922 | int *continued_write, | |
2923 | int *logoffsetp) | |
1da177e4 | 2924 | { |
1da177e4 LT |
2925 | int log_offset; |
2926 | xlog_rec_header_t *head; | |
2927 | xlog_in_core_t *iclog; | |
1da177e4 LT |
2928 | |
2929 | restart: | |
b22cd72c | 2930 | spin_lock(&log->l_icloglock); |
1da177e4 | 2931 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2932 | spin_unlock(&log->l_icloglock); |
2451337d | 2933 | return -EIO; |
1da177e4 LT |
2934 | } |
2935 | ||
2936 | iclog = log->l_iclog; | |
d748c623 | 2937 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
ff6d6af2 | 2938 | XFS_STATS_INC(log->l_mp, xs_log_noiclogs); |
d748c623 MW |
2939 | |
2940 | /* Wait for log writes to have flushed */ | |
eb40a875 | 2941 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
2942 | goto restart; |
2943 | } | |
d748c623 | 2944 | |
1da177e4 LT |
2945 | head = &iclog->ic_header; |
2946 | ||
155cc6b7 | 2947 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
2948 | log_offset = iclog->ic_offset; |
2949 | ||
2950 | /* On the 1st write to an iclog, figure out lsn. This works | |
2951 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
2952 | * committing to. If the offset is set, that's how many blocks | |
2953 | * must be written. | |
2954 | */ | |
2955 | if (log_offset == 0) { | |
2956 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 2957 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
2958 | log->l_iclog_hsize, |
2959 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
2960 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
2961 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 2962 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
2963 | ASSERT(log->l_curr_block >= 0); |
2964 | } | |
2965 | ||
2966 | /* If there is enough room to write everything, then do it. Otherwise, | |
2967 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
2968 | * bit is on, so this will get flushed out. Don't update ic_offset | |
2969 | * until you know exactly how many bytes get copied. Therefore, wait | |
2970 | * until later to update ic_offset. | |
2971 | * | |
2972 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
2973 | * can fit into remaining data section. | |
2974 | */ | |
2975 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
df732b29 CH |
2976 | int error = 0; |
2977 | ||
1da177e4 LT |
2978 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); |
2979 | ||
49641f1a | 2980 | /* |
df732b29 CH |
2981 | * If we are the only one writing to this iclog, sync it to |
2982 | * disk. We need to do an atomic compare and decrement here to | |
2983 | * avoid racing with concurrent atomic_dec_and_lock() calls in | |
49641f1a DC |
2984 | * xlog_state_release_iclog() when there is more than one |
2985 | * reference to the iclog. | |
2986 | */ | |
df732b29 | 2987 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) |
49641f1a | 2988 | error = xlog_state_release_iclog(log, iclog); |
df732b29 CH |
2989 | spin_unlock(&log->l_icloglock); |
2990 | if (error) | |
2991 | return error; | |
1da177e4 LT |
2992 | goto restart; |
2993 | } | |
2994 | ||
2995 | /* Do we have enough room to write the full amount in the remainder | |
2996 | * of this iclog? Or must we continue a write on the next iclog and | |
2997 | * mark this iclog as completely taken? In the case where we switch | |
2998 | * iclogs (to mark it taken), this particular iclog will release/sync | |
2999 | * to disk in xlog_write(). | |
3000 | */ | |
3001 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
3002 | *continued_write = 0; | |
3003 | iclog->ic_offset += len; | |
3004 | } else { | |
3005 | *continued_write = 1; | |
3006 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
3007 | } | |
3008 | *iclogp = iclog; | |
3009 | ||
3010 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 3011 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3012 | |
3013 | *logoffsetp = log_offset; | |
3014 | return 0; | |
b843299b | 3015 | } |
1da177e4 | 3016 | |
8b41e3f9 | 3017 | /* |
b843299b DC |
3018 | * The first cnt-1 times a ticket goes through here we don't need to move the |
3019 | * grant write head because the permanent reservation has reserved cnt times the | |
3020 | * unit amount. Release part of current permanent unit reservation and reset | |
3021 | * current reservation to be one units worth. Also move grant reservation head | |
3022 | * forward. | |
1da177e4 | 3023 | */ |
8b41e3f9 CH |
3024 | void |
3025 | xfs_log_ticket_regrant( | |
9a8d2fdb MT |
3026 | struct xlog *log, |
3027 | struct xlog_ticket *ticket) | |
1da177e4 | 3028 | { |
8b41e3f9 | 3029 | trace_xfs_log_ticket_regrant(log, ticket); |
0b1b213f | 3030 | |
1da177e4 LT |
3031 | if (ticket->t_cnt > 0) |
3032 | ticket->t_cnt--; | |
3033 | ||
28496968 | 3034 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3035 | ticket->t_curr_res); |
28496968 | 3036 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 3037 | ticket->t_curr_res); |
1da177e4 | 3038 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3039 | xlog_tic_reset_res(ticket); |
0b1b213f | 3040 | |
8b41e3f9 | 3041 | trace_xfs_log_ticket_regrant_sub(log, ticket); |
0b1b213f | 3042 | |
1da177e4 | 3043 | /* just return if we still have some of the pre-reserved space */ |
8b41e3f9 CH |
3044 | if (!ticket->t_cnt) { |
3045 | xlog_grant_add_space(log, &log->l_reserve_head.grant, | |
3046 | ticket->t_unit_res); | |
3047 | trace_xfs_log_ticket_regrant_exit(log, ticket); | |
1da177e4 | 3048 | |
8b41e3f9 CH |
3049 | ticket->t_curr_res = ticket->t_unit_res; |
3050 | xlog_tic_reset_res(ticket); | |
3051 | } | |
1da177e4 | 3052 | |
8b41e3f9 CH |
3053 | xfs_log_ticket_put(ticket); |
3054 | } | |
1da177e4 LT |
3055 | |
3056 | /* | |
3057 | * Give back the space left from a reservation. | |
3058 | * | |
3059 | * All the information we need to make a correct determination of space left | |
3060 | * is present. For non-permanent reservations, things are quite easy. The | |
3061 | * count should have been decremented to zero. We only need to deal with the | |
3062 | * space remaining in the current reservation part of the ticket. If the | |
3063 | * ticket contains a permanent reservation, there may be left over space which | |
3064 | * needs to be released. A count of N means that N-1 refills of the current | |
3065 | * reservation can be done before we need to ask for more space. The first | |
3066 | * one goes to fill up the first current reservation. Once we run out of | |
3067 | * space, the count will stay at zero and the only space remaining will be | |
3068 | * in the current reservation field. | |
3069 | */ | |
8b41e3f9 CH |
3070 | void |
3071 | xfs_log_ticket_ungrant( | |
9a8d2fdb MT |
3072 | struct xlog *log, |
3073 | struct xlog_ticket *ticket) | |
1da177e4 | 3074 | { |
8b41e3f9 CH |
3075 | int bytes; |
3076 | ||
3077 | trace_xfs_log_ticket_ungrant(log, ticket); | |
663e496a | 3078 | |
1da177e4 LT |
3079 | if (ticket->t_cnt > 0) |
3080 | ticket->t_cnt--; | |
3081 | ||
8b41e3f9 | 3082 | trace_xfs_log_ticket_ungrant_sub(log, ticket); |
1da177e4 | 3083 | |
663e496a DC |
3084 | /* |
3085 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
3086 | * up more space based on the remaining count. |
3087 | */ | |
663e496a | 3088 | bytes = ticket->t_curr_res; |
1da177e4 LT |
3089 | if (ticket->t_cnt > 0) { |
3090 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 3091 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
3092 | } |
3093 | ||
28496968 CH |
3094 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
3095 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 3096 | |
8b41e3f9 | 3097 | trace_xfs_log_ticket_ungrant_exit(log, ticket); |
0b1b213f | 3098 | |
cfb7cdca | 3099 | xfs_log_space_wake(log->l_mp); |
8b41e3f9 | 3100 | xfs_log_ticket_put(ticket); |
09a423a3 | 3101 | } |
1da177e4 | 3102 | |
1da177e4 | 3103 | /* |
b843299b DC |
3104 | * This routine will mark the current iclog in the ring as WANT_SYNC and move |
3105 | * the current iclog pointer to the next iclog in the ring. | |
1da177e4 LT |
3106 | */ |
3107 | STATIC void | |
9a8d2fdb MT |
3108 | xlog_state_switch_iclogs( |
3109 | struct xlog *log, | |
3110 | struct xlog_in_core *iclog, | |
3111 | int eventual_size) | |
1da177e4 LT |
3112 | { |
3113 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
69363999 CH |
3114 | assert_spin_locked(&log->l_icloglock); |
3115 | ||
1da177e4 LT |
3116 | if (!eventual_size) |
3117 | eventual_size = iclog->ic_offset; | |
3118 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3119 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3120 | log->l_prev_block = log->l_curr_block; |
3121 | log->l_prev_cycle = log->l_curr_cycle; | |
3122 | ||
3123 | /* roll log?: ic_offset changed later */ | |
3124 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3125 | ||
3126 | /* Round up to next log-sunit */ | |
a6a65fef | 3127 | if (log->l_iclog_roundoff > BBSIZE) { |
18842e0a GU |
3128 | uint32_t sunit_bb = BTOBB(log->l_iclog_roundoff); |
3129 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); | |
1da177e4 LT |
3130 | } |
3131 | ||
3132 | if (log->l_curr_block >= log->l_logBBsize) { | |
a45086e2 BF |
3133 | /* |
3134 | * Rewind the current block before the cycle is bumped to make | |
3135 | * sure that the combined LSN never transiently moves forward | |
3136 | * when the log wraps to the next cycle. This is to support the | |
3137 | * unlocked sample of these fields from xlog_valid_lsn(). Most | |
3138 | * other cases should acquire l_icloglock. | |
3139 | */ | |
3140 | log->l_curr_block -= log->l_logBBsize; | |
3141 | ASSERT(log->l_curr_block >= 0); | |
3142 | smp_wmb(); | |
1da177e4 LT |
3143 | log->l_curr_cycle++; |
3144 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3145 | log->l_curr_cycle++; | |
1da177e4 LT |
3146 | } |
3147 | ASSERT(iclog == log->l_iclog); | |
3148 | log->l_iclog = iclog->ic_next; | |
b843299b | 3149 | } |
1da177e4 | 3150 | |
1da177e4 LT |
3151 | /* |
3152 | * Write out all data in the in-core log as of this exact moment in time. | |
3153 | * | |
3154 | * Data may be written to the in-core log during this call. However, | |
3155 | * we don't guarantee this data will be written out. A change from past | |
3156 | * implementation means this routine will *not* write out zero length LRs. | |
3157 | * | |
3158 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3159 | * If we determine there is no flushable data, we just return. There is no | |
3160 | * flushable data if: | |
3161 | * | |
3162 | * 1. the current iclog is active and has no data; the previous iclog | |
3163 | * is in the active or dirty state. | |
3164 | * 2. the current iclog is drity, and the previous iclog is in the | |
3165 | * active or dirty state. | |
3166 | * | |
12017faf | 3167 | * We may sleep if: |
1da177e4 LT |
3168 | * |
3169 | * 1. the current iclog is not in the active nor dirty state. | |
3170 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3171 | * active nor dirty state. | |
3172 | * 3. the current iclog is active, and there is another thread writing | |
3173 | * to this particular iclog. | |
3174 | * 4. a) the current iclog is active and has no other writers | |
3175 | * b) when we return from flushing out this iclog, it is still | |
3176 | * not in the active nor dirty state. | |
3177 | */ | |
a14a348b | 3178 | int |
60e5bb78 | 3179 | xfs_log_force( |
a14a348b | 3180 | struct xfs_mount *mp, |
60e5bb78 | 3181 | uint flags) |
1da177e4 | 3182 | { |
ad223e60 | 3183 | struct xlog *log = mp->m_log; |
a14a348b CH |
3184 | struct xlog_in_core *iclog; |
3185 | xfs_lsn_t lsn; | |
3186 | ||
ff6d6af2 | 3187 | XFS_STATS_INC(mp, xs_log_force); |
60e5bb78 | 3188 | trace_xfs_log_force(mp, 0, _RET_IP_); |
1da177e4 | 3189 | |
93b8a585 | 3190 | xlog_cil_force(log); |
71e330b5 | 3191 | |
b22cd72c | 3192 | spin_lock(&log->l_icloglock); |
1da177e4 | 3193 | iclog = log->l_iclog; |
1858bb0b | 3194 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
e6b96570 | 3195 | goto out_error; |
1da177e4 | 3196 | |
e6b96570 CH |
3197 | if (iclog->ic_state == XLOG_STATE_DIRTY || |
3198 | (iclog->ic_state == XLOG_STATE_ACTIVE && | |
3199 | atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) { | |
1da177e4 | 3200 | /* |
e6b96570 CH |
3201 | * If the head is dirty or (active and empty), then we need to |
3202 | * look at the previous iclog. | |
3203 | * | |
3204 | * If the previous iclog is active or dirty we are done. There | |
3205 | * is nothing to sync out. Otherwise, we attach ourselves to the | |
1da177e4 LT |
3206 | * previous iclog and go to sleep. |
3207 | */ | |
e6b96570 | 3208 | iclog = iclog->ic_prev; |
e6b96570 CH |
3209 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3210 | if (atomic_read(&iclog->ic_refcnt) == 0) { | |
3211 | /* | |
3212 | * We are the only one with access to this iclog. | |
3213 | * | |
3214 | * Flush it out now. There should be a roundoff of zero | |
3215 | * to show that someone has already taken care of the | |
3216 | * roundoff from the previous sync. | |
3217 | */ | |
3218 | atomic_inc(&iclog->ic_refcnt); | |
3219 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
3220 | xlog_state_switch_iclogs(log, iclog, 0); | |
e6b96570 | 3221 | if (xlog_state_release_iclog(log, iclog)) |
df732b29 | 3222 | goto out_error; |
1da177e4 | 3223 | |
81e5b50a | 3224 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) |
e6b96570 CH |
3225 | goto out_unlock; |
3226 | } else { | |
3227 | /* | |
3228 | * Someone else is writing to this iclog. | |
3229 | * | |
3230 | * Use its call to flush out the data. However, the | |
3231 | * other thread may not force out this LR, so we mark | |
3232 | * it WANT_SYNC. | |
3233 | */ | |
3234 | xlog_state_switch_iclogs(log, iclog, 0); | |
1da177e4 | 3235 | } |
e6b96570 | 3236 | } else { |
1da177e4 | 3237 | /* |
e6b96570 CH |
3238 | * If the head iclog is not active nor dirty, we just attach |
3239 | * ourselves to the head and go to sleep if necessary. | |
1da177e4 | 3240 | */ |
e6b96570 | 3241 | ; |
1da177e4 | 3242 | } |
e6b96570 | 3243 | |
81e5b50a CH |
3244 | if (flags & XFS_LOG_SYNC) |
3245 | return xlog_wait_on_iclog(iclog); | |
e6b96570 CH |
3246 | out_unlock: |
3247 | spin_unlock(&log->l_icloglock); | |
3248 | return 0; | |
3249 | out_error: | |
3250 | spin_unlock(&log->l_icloglock); | |
3251 | return -EIO; | |
a14a348b | 3252 | } |
1da177e4 | 3253 | |
3e4da466 | 3254 | static int |
5f9b4b0d DC |
3255 | xlog_force_lsn( |
3256 | struct xlog *log, | |
a14a348b CH |
3257 | xfs_lsn_t lsn, |
3258 | uint flags, | |
3e4da466 CH |
3259 | int *log_flushed, |
3260 | bool already_slept) | |
1da177e4 | 3261 | { |
a14a348b | 3262 | struct xlog_in_core *iclog; |
71e330b5 | 3263 | |
a14a348b CH |
3264 | spin_lock(&log->l_icloglock); |
3265 | iclog = log->l_iclog; | |
1858bb0b | 3266 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
93806299 | 3267 | goto out_error; |
1da177e4 | 3268 | |
93806299 CH |
3269 | while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { |
3270 | iclog = iclog->ic_next; | |
3271 | if (iclog == log->l_iclog) | |
3272 | goto out_unlock; | |
3273 | } | |
a14a348b | 3274 | |
93806299 CH |
3275 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3276 | /* | |
3277 | * We sleep here if we haven't already slept (e.g. this is the | |
3278 | * first time we've looked at the correct iclog buf) and the | |
3279 | * buffer before us is going to be sync'ed. The reason for this | |
3280 | * is that if we are doing sync transactions here, by waiting | |
3281 | * for the previous I/O to complete, we can allow a few more | |
3282 | * transactions into this iclog before we close it down. | |
3283 | * | |
3284 | * Otherwise, we mark the buffer WANT_SYNC, and bump up the | |
3285 | * refcnt so we can release the log (which drops the ref count). | |
3286 | * The state switch keeps new transaction commits from using | |
3287 | * this buffer. When the current commits finish writing into | |
3288 | * the buffer, the refcount will drop to zero and the buffer | |
3289 | * will go out then. | |
3290 | */ | |
3291 | if (!already_slept && | |
1858bb0b CH |
3292 | (iclog->ic_prev->ic_state == XLOG_STATE_WANT_SYNC || |
3293 | iclog->ic_prev->ic_state == XLOG_STATE_SYNCING)) { | |
93806299 CH |
3294 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3295 | &log->l_icloglock); | |
3e4da466 | 3296 | return -EAGAIN; |
1da177e4 | 3297 | } |
93806299 CH |
3298 | atomic_inc(&iclog->ic_refcnt); |
3299 | xlog_state_switch_iclogs(log, iclog, 0); | |
93806299 | 3300 | if (xlog_state_release_iclog(log, iclog)) |
df732b29 | 3301 | goto out_error; |
93806299 CH |
3302 | if (log_flushed) |
3303 | *log_flushed = 1; | |
93806299 | 3304 | } |
1da177e4 | 3305 | |
81e5b50a CH |
3306 | if (flags & XFS_LOG_SYNC) |
3307 | return xlog_wait_on_iclog(iclog); | |
93806299 | 3308 | out_unlock: |
a14a348b CH |
3309 | spin_unlock(&log->l_icloglock); |
3310 | return 0; | |
93806299 CH |
3311 | out_error: |
3312 | spin_unlock(&log->l_icloglock); | |
3313 | return -EIO; | |
a14a348b CH |
3314 | } |
3315 | ||
3e4da466 CH |
3316 | /* |
3317 | * Force the in-core log to disk for a specific LSN. | |
3318 | * | |
3319 | * Find in-core log with lsn. | |
3320 | * If it is in the DIRTY state, just return. | |
3321 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3322 | * state and go to sleep or return. | |
3323 | * If it is in any other state, go to sleep or return. | |
3324 | * | |
3325 | * Synchronous forces are implemented with a wait queue. All callers trying | |
3326 | * to force a given lsn to disk must wait on the queue attached to the | |
3327 | * specific in-core log. When given in-core log finally completes its write | |
3328 | * to disk, that thread will wake up all threads waiting on the queue. | |
3329 | */ | |
3330 | int | |
5f9b4b0d | 3331 | xfs_log_force_seq( |
3e4da466 | 3332 | struct xfs_mount *mp, |
5f9b4b0d | 3333 | xfs_csn_t seq, |
3e4da466 CH |
3334 | uint flags, |
3335 | int *log_flushed) | |
3336 | { | |
5f9b4b0d DC |
3337 | struct xlog *log = mp->m_log; |
3338 | xfs_lsn_t lsn; | |
3e4da466 | 3339 | int ret; |
5f9b4b0d | 3340 | ASSERT(seq != 0); |
3e4da466 CH |
3341 | |
3342 | XFS_STATS_INC(mp, xs_log_force); | |
5f9b4b0d | 3343 | trace_xfs_log_force(mp, seq, _RET_IP_); |
3e4da466 | 3344 | |
5f9b4b0d | 3345 | lsn = xlog_cil_force_seq(log, seq); |
3e4da466 CH |
3346 | if (lsn == NULLCOMMITLSN) |
3347 | return 0; | |
3348 | ||
5f9b4b0d DC |
3349 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, false); |
3350 | if (ret == -EAGAIN) { | |
3351 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3352 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, true); | |
3353 | } | |
3e4da466 CH |
3354 | return ret; |
3355 | } | |
3356 | ||
1da177e4 | 3357 | /* |
9da096fd | 3358 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3359 | */ |
cc09c0dc DC |
3360 | void |
3361 | xfs_log_ticket_put( | |
3362 | xlog_ticket_t *ticket) | |
1da177e4 | 3363 | { |
cc09c0dc | 3364 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3365 | if (atomic_dec_and_test(&ticket->t_ref)) |
377bcd5f | 3366 | kmem_cache_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3367 | } |
1da177e4 | 3368 | |
cc09c0dc DC |
3369 | xlog_ticket_t * |
3370 | xfs_log_ticket_get( | |
3371 | xlog_ticket_t *ticket) | |
3372 | { | |
3373 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3374 | atomic_inc(&ticket->t_ref); | |
3375 | return ticket; | |
3376 | } | |
1da177e4 LT |
3377 | |
3378 | /* | |
e773fc93 JL |
3379 | * Figure out the total log space unit (in bytes) that would be |
3380 | * required for a log ticket. | |
1da177e4 | 3381 | */ |
a6a65fef DC |
3382 | static int |
3383 | xlog_calc_unit_res( | |
3384 | struct xlog *log, | |
e773fc93 | 3385 | int unit_bytes) |
1da177e4 | 3386 | { |
e773fc93 JL |
3387 | int iclog_space; |
3388 | uint num_headers; | |
1da177e4 LT |
3389 | |
3390 | /* | |
3391 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3392 | * in the log. A unit in this case is the amount of space for one | |
3393 | * of these log operations. Normal reservations have a cnt of 1 | |
3394 | * and their unit amount is the total amount of space required. | |
3395 | * | |
3396 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3397 | * which occupy space in the on-disk log. |
3398 | * | |
3399 | * Normal form of a transaction is: | |
3400 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3401 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3402 | * | |
3403 | * We need to account for all the leadup data and trailer data | |
3404 | * around the transaction data. | |
3405 | * And then we need to account for the worst case in terms of using | |
3406 | * more space. | |
3407 | * The worst case will happen if: | |
3408 | * - the placement of the transaction happens to be such that the | |
3409 | * roundoff is at its maximum | |
3410 | * - the transaction data is synced before the commit record is synced | |
3411 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3412 | * Therefore the commit record is in its own Log Record. | |
3413 | * This can happen as the commit record is called with its | |
3414 | * own region to xlog_write(). | |
3415 | * This then means that in the worst case, roundoff can happen for | |
3416 | * the commit-rec as well. | |
3417 | * The commit-rec is smaller than padding in this scenario and so it is | |
3418 | * not added separately. | |
1da177e4 LT |
3419 | */ |
3420 | ||
32fb9b57 TS |
3421 | /* for trans header */ |
3422 | unit_bytes += sizeof(xlog_op_header_t); | |
3423 | unit_bytes += sizeof(xfs_trans_header_t); | |
3424 | ||
1da177e4 | 3425 | /* for start-rec */ |
32fb9b57 TS |
3426 | unit_bytes += sizeof(xlog_op_header_t); |
3427 | ||
9b9fc2b7 DC |
3428 | /* |
3429 | * for LR headers - the space for data in an iclog is the size minus | |
3430 | * the space used for the headers. If we use the iclog size, then we | |
3431 | * undercalculate the number of headers required. | |
3432 | * | |
3433 | * Furthermore - the addition of op headers for split-recs might | |
3434 | * increase the space required enough to require more log and op | |
3435 | * headers, so take that into account too. | |
3436 | * | |
3437 | * IMPORTANT: This reservation makes the assumption that if this | |
3438 | * transaction is the first in an iclog and hence has the LR headers | |
3439 | * accounted to it, then the remaining space in the iclog is | |
3440 | * exclusively for this transaction. i.e. if the transaction is larger | |
3441 | * than the iclog, it will be the only thing in that iclog. | |
3442 | * Fundamentally, this means we must pass the entire log vector to | |
3443 | * xlog_write to guarantee this. | |
3444 | */ | |
3445 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3446 | num_headers = howmany(unit_bytes, iclog_space); | |
3447 | ||
3448 | /* for split-recs - ophdrs added when data split over LRs */ | |
3449 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3450 | ||
3451 | /* add extra header reservations if we overrun */ | |
3452 | while (!num_headers || | |
3453 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3454 | unit_bytes += sizeof(xlog_op_header_t); | |
3455 | num_headers++; | |
3456 | } | |
32fb9b57 | 3457 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3458 | |
32fb9b57 TS |
3459 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3460 | unit_bytes += log->l_iclog_hsize; | |
3461 | ||
a6a65fef DC |
3462 | /* roundoff padding for transaction data and one for commit record */ |
3463 | unit_bytes += 2 * log->l_iclog_roundoff; | |
1da177e4 | 3464 | |
e773fc93 JL |
3465 | return unit_bytes; |
3466 | } | |
3467 | ||
a6a65fef DC |
3468 | int |
3469 | xfs_log_calc_unit_res( | |
3470 | struct xfs_mount *mp, | |
3471 | int unit_bytes) | |
3472 | { | |
3473 | return xlog_calc_unit_res(mp->m_log, unit_bytes); | |
3474 | } | |
3475 | ||
e773fc93 JL |
3476 | /* |
3477 | * Allocate and initialise a new log ticket. | |
3478 | */ | |
3479 | struct xlog_ticket * | |
3480 | xlog_ticket_alloc( | |
3481 | struct xlog *log, | |
3482 | int unit_bytes, | |
3483 | int cnt, | |
3484 | char client, | |
ca4f2589 | 3485 | bool permanent) |
e773fc93 JL |
3486 | { |
3487 | struct xlog_ticket *tic; | |
3488 | int unit_res; | |
3489 | ||
ca4f2589 | 3490 | tic = kmem_cache_zalloc(xfs_log_ticket_zone, GFP_NOFS | __GFP_NOFAIL); |
e773fc93 | 3491 | |
a6a65fef | 3492 | unit_res = xlog_calc_unit_res(log, unit_bytes); |
e773fc93 | 3493 | |
cc09c0dc | 3494 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3495 | tic->t_task = current; |
10547941 | 3496 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3497 | tic->t_unit_res = unit_res; |
3498 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3499 | tic->t_cnt = cnt; |
3500 | tic->t_ocnt = cnt; | |
ecb3403d | 3501 | tic->t_tid = prandom_u32(); |
1da177e4 | 3502 | tic->t_clientid = client; |
9006fb91 | 3503 | if (permanent) |
1da177e4 | 3504 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3505 | |
0adba536 | 3506 | xlog_tic_reset_res(tic); |
7e9c6396 | 3507 | |
1da177e4 | 3508 | return tic; |
cc09c0dc | 3509 | } |
1da177e4 | 3510 | |
cfcbbbd0 | 3511 | #if defined(DEBUG) |
1da177e4 LT |
3512 | /* |
3513 | * Make sure that the destination ptr is within the valid data region of | |
3514 | * one of the iclogs. This uses backup pointers stored in a different | |
3515 | * part of the log in case we trash the log structure. | |
3516 | */ | |
181fdfe6 | 3517 | STATIC void |
e6b1f273 | 3518 | xlog_verify_dest_ptr( |
ad223e60 | 3519 | struct xlog *log, |
5809d5e0 | 3520 | void *ptr) |
1da177e4 LT |
3521 | { |
3522 | int i; | |
3523 | int good_ptr = 0; | |
3524 | ||
e6b1f273 CH |
3525 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3526 | if (ptr >= log->l_iclog_bak[i] && | |
3527 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3528 | good_ptr++; |
3529 | } | |
e6b1f273 CH |
3530 | |
3531 | if (!good_ptr) | |
a0fa2b67 | 3532 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3533 | } |
1da177e4 | 3534 | |
da8a1a4a DC |
3535 | /* |
3536 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3537 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3538 | * the cycles differ by exactly one and check the byte count. | |
3539 | * | |
3540 | * This check is run unlocked, so can give false positives. Rather than assert | |
3541 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3542 | * determine if they want to panic the machine when such an error occurs. For | |
3543 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3544 | * an assert, it can be turned off at runtime. | |
3545 | */ | |
3f336c6f DC |
3546 | STATIC void |
3547 | xlog_verify_grant_tail( | |
ad223e60 | 3548 | struct xlog *log) |
3f336c6f | 3549 | { |
1c3cb9ec | 3550 | int tail_cycle, tail_blocks; |
a69ed03c | 3551 | int cycle, space; |
3f336c6f | 3552 | |
28496968 | 3553 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3554 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3555 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3556 | if (cycle - 1 != tail_cycle && |
3557 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3558 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3559 | "%s: cycle - 1 != tail_cycle", __func__); | |
3560 | log->l_flags |= XLOG_TAIL_WARN; | |
3561 | } | |
3562 | ||
3563 | if (space > BBTOB(tail_blocks) && | |
3564 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3565 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3566 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3567 | log->l_flags |= XLOG_TAIL_WARN; | |
3568 | } | |
3f336c6f DC |
3569 | } |
3570 | } | |
3571 | ||
1da177e4 LT |
3572 | /* check if it will fit */ |
3573 | STATIC void | |
9a8d2fdb MT |
3574 | xlog_verify_tail_lsn( |
3575 | struct xlog *log, | |
3576 | struct xlog_in_core *iclog, | |
3577 | xfs_lsn_t tail_lsn) | |
1da177e4 LT |
3578 | { |
3579 | int blocks; | |
3580 | ||
3581 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3582 | blocks = | |
3583 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3584 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3585 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3586 | } else { |
3587 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3588 | ||
3589 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3590 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3591 | |
3592 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3593 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3594 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 | 3595 | } |
b843299b | 3596 | } |
1da177e4 LT |
3597 | |
3598 | /* | |
3599 | * Perform a number of checks on the iclog before writing to disk. | |
3600 | * | |
3601 | * 1. Make sure the iclogs are still circular | |
3602 | * 2. Make sure we have a good magic number | |
3603 | * 3. Make sure we don't have magic numbers in the data | |
3604 | * 4. Check fields of each log operation header for: | |
3605 | * A. Valid client identifier | |
3606 | * B. tid ptr value falls in valid ptr space (user space code) | |
3607 | * C. Length in log record header is correct according to the | |
3608 | * individual operation headers within record. | |
3609 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3610 | * log, check the preceding blocks of the physical log to make sure all | |
3611 | * the cycle numbers agree with the current cycle number. | |
3612 | */ | |
3613 | STATIC void | |
9a8d2fdb MT |
3614 | xlog_verify_iclog( |
3615 | struct xlog *log, | |
3616 | struct xlog_in_core *iclog, | |
abca1f33 | 3617 | int count) |
1da177e4 LT |
3618 | { |
3619 | xlog_op_header_t *ophead; | |
3620 | xlog_in_core_t *icptr; | |
3621 | xlog_in_core_2_t *xhdr; | |
5809d5e0 | 3622 | void *base_ptr, *ptr, *p; |
db9d67d6 | 3623 | ptrdiff_t field_offset; |
c8ce540d | 3624 | uint8_t clientid; |
1da177e4 LT |
3625 | int len, i, j, k, op_len; |
3626 | int idx; | |
1da177e4 LT |
3627 | |
3628 | /* check validity of iclog pointers */ | |
b22cd72c | 3629 | spin_lock(&log->l_icloglock); |
1da177e4 | 3630 | icptr = log->l_iclog; |
643f7c4e GB |
3631 | for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next) |
3632 | ASSERT(icptr); | |
3633 | ||
1da177e4 | 3634 | if (icptr != log->l_iclog) |
a0fa2b67 | 3635 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3636 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3637 | |
3638 | /* check log magic numbers */ | |
69ef921b | 3639 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3640 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3641 | |
5809d5e0 CH |
3642 | base_ptr = ptr = &iclog->ic_header; |
3643 | p = &iclog->ic_header; | |
3644 | for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) { | |
69ef921b | 3645 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3646 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3647 | __func__); | |
1da177e4 LT |
3648 | } |
3649 | ||
3650 | /* check fields */ | |
b53e675d | 3651 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
5809d5e0 CH |
3652 | base_ptr = ptr = iclog->ic_datap; |
3653 | ophead = ptr; | |
b28708d6 | 3654 | xhdr = iclog->ic_data; |
1da177e4 | 3655 | for (i = 0; i < len; i++) { |
5809d5e0 | 3656 | ophead = ptr; |
1da177e4 LT |
3657 | |
3658 | /* clientid is only 1 byte */ | |
5809d5e0 CH |
3659 | p = &ophead->oh_clientid; |
3660 | field_offset = p - base_ptr; | |
abca1f33 | 3661 | if (field_offset & 0x1ff) { |
1da177e4 LT |
3662 | clientid = ophead->oh_clientid; |
3663 | } else { | |
b2a922cd | 3664 | idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap); |
1da177e4 LT |
3665 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3666 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3667 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3668 | clientid = xlog_get_client_id( |
3669 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3670 | } else { |
03bea6fe CH |
3671 | clientid = xlog_get_client_id( |
3672 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3673 | } |
3674 | } | |
3675 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 | 3676 | xfs_warn(log->l_mp, |
c9690043 | 3677 | "%s: invalid clientid %d op "PTR_FMT" offset 0x%lx", |
a0fa2b67 DC |
3678 | __func__, clientid, ophead, |
3679 | (unsigned long)field_offset); | |
1da177e4 LT |
3680 | |
3681 | /* check length */ | |
5809d5e0 CH |
3682 | p = &ophead->oh_len; |
3683 | field_offset = p - base_ptr; | |
abca1f33 | 3684 | if (field_offset & 0x1ff) { |
67fcb7bf | 3685 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 | 3686 | } else { |
db9d67d6 CH |
3687 | idx = BTOBBT((uintptr_t)&ophead->oh_len - |
3688 | (uintptr_t)iclog->ic_datap); | |
1da177e4 LT |
3689 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3690 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3691 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3692 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3693 | } else { |
b53e675d | 3694 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3695 | } |
3696 | } | |
3697 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3698 | } | |
b843299b | 3699 | } |
cfcbbbd0 | 3700 | #endif |
1da177e4 LT |
3701 | |
3702 | /* | |
b22cd72c | 3703 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3704 | */ |
3705 | STATIC int | |
3706 | xlog_state_ioerror( | |
9a8d2fdb | 3707 | struct xlog *log) |
1da177e4 LT |
3708 | { |
3709 | xlog_in_core_t *iclog, *ic; | |
3710 | ||
3711 | iclog = log->l_iclog; | |
1858bb0b | 3712 | if (iclog->ic_state != XLOG_STATE_IOERROR) { |
1da177e4 LT |
3713 | /* |
3714 | * Mark all the incore logs IOERROR. | |
3715 | * From now on, no log flushes will result. | |
3716 | */ | |
3717 | ic = iclog; | |
3718 | do { | |
3719 | ic->ic_state = XLOG_STATE_IOERROR; | |
3720 | ic = ic->ic_next; | |
3721 | } while (ic != iclog); | |
014c2544 | 3722 | return 0; |
1da177e4 LT |
3723 | } |
3724 | /* | |
3725 | * Return non-zero, if state transition has already happened. | |
3726 | */ | |
014c2544 | 3727 | return 1; |
1da177e4 LT |
3728 | } |
3729 | ||
3730 | /* | |
3731 | * This is called from xfs_force_shutdown, when we're forcibly | |
3732 | * shutting down the filesystem, typically because of an IO error. | |
3733 | * Our main objectives here are to make sure that: | |
a870fe6d DC |
3734 | * a. if !logerror, flush the logs to disk. Anything modified |
3735 | * after this is ignored. | |
3736 | * b. the filesystem gets marked 'SHUTDOWN' for all interested | |
1da177e4 | 3737 | * parties to find out, 'atomically'. |
a870fe6d | 3738 | * c. those who're sleeping on log reservations, pinned objects and |
1da177e4 | 3739 | * other resources get woken up, and be told the bad news. |
a870fe6d | 3740 | * d. nothing new gets queued up after (b) and (c) are done. |
9da1ab18 | 3741 | * |
a870fe6d DC |
3742 | * Note: for the !logerror case we need to flush the regions held in memory out |
3743 | * to disk first. This needs to be done before the log is marked as shutdown, | |
3744 | * otherwise the iclog writes will fail. | |
1da177e4 LT |
3745 | */ |
3746 | int | |
3747 | xfs_log_force_umount( | |
3748 | struct xfs_mount *mp, | |
3749 | int logerror) | |
3750 | { | |
9a8d2fdb | 3751 | struct xlog *log; |
1da177e4 | 3752 | int retval; |
1da177e4 LT |
3753 | |
3754 | log = mp->m_log; | |
3755 | ||
3756 | /* | |
3757 | * If this happens during log recovery, don't worry about | |
3758 | * locking; the log isn't open for business yet. | |
3759 | */ | |
3760 | if (!log || | |
3761 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3762 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 | 3763 | if (mp->m_sb_bp) |
b0388bf1 | 3764 | mp->m_sb_bp->b_flags |= XBF_DONE; |
014c2544 | 3765 | return 0; |
1da177e4 LT |
3766 | } |
3767 | ||
3768 | /* | |
3769 | * Somebody could've already done the hard work for us. | |
3770 | * No need to get locks for this. | |
3771 | */ | |
1858bb0b | 3772 | if (logerror && log->l_iclog->ic_state == XLOG_STATE_IOERROR) { |
1da177e4 | 3773 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); |
014c2544 | 3774 | return 1; |
1da177e4 | 3775 | } |
9da1ab18 DC |
3776 | |
3777 | /* | |
a870fe6d DC |
3778 | * Flush all the completed transactions to disk before marking the log |
3779 | * being shut down. We need to do it in this order to ensure that | |
3780 | * completed operations are safely on disk before we shut down, and that | |
3781 | * we don't have to issue any buffer IO after the shutdown flags are set | |
3782 | * to guarantee this. | |
9da1ab18 | 3783 | */ |
93b8a585 | 3784 | if (!logerror) |
60e5bb78 | 3785 | xfs_log_force(mp, XFS_LOG_SYNC); |
9da1ab18 | 3786 | |
1da177e4 | 3787 | /* |
3f16b985 DC |
3788 | * mark the filesystem and the as in a shutdown state and wake |
3789 | * everybody up to tell them the bad news. | |
1da177e4 | 3790 | */ |
b22cd72c | 3791 | spin_lock(&log->l_icloglock); |
1da177e4 | 3792 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 | 3793 | if (mp->m_sb_bp) |
b0388bf1 | 3794 | mp->m_sb_bp->b_flags |= XBF_DONE; |
bac8dca9 | 3795 | |
1da177e4 | 3796 | /* |
a870fe6d DC |
3797 | * Mark the log and the iclogs with IO error flags to prevent any |
3798 | * further log IO from being issued or completed. | |
1da177e4 LT |
3799 | */ |
3800 | log->l_flags |= XLOG_IO_ERROR; | |
a870fe6d | 3801 | retval = xlog_state_ioerror(log); |
b22cd72c | 3802 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3803 | |
3804 | /* | |
10547941 DC |
3805 | * We don't want anybody waiting for log reservations after this. That |
3806 | * means we have to wake up everybody queued up on reserveq as well as | |
3807 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3808 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3809 | * action is protected by the grant locks. |
1da177e4 | 3810 | */ |
a79bf2d7 CH |
3811 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3812 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3813 | |
1da177e4 | 3814 | /* |
ac983517 DC |
3815 | * Wake up everybody waiting on xfs_log_force. Wake the CIL push first |
3816 | * as if the log writes were completed. The abort handling in the log | |
3817 | * item committed callback functions will do this again under lock to | |
3818 | * avoid races. | |
1da177e4 | 3819 | */ |
cdea5459 | 3820 | spin_lock(&log->l_cilp->xc_push_lock); |
ac983517 | 3821 | wake_up_all(&log->l_cilp->xc_commit_wait); |
cdea5459 | 3822 | spin_unlock(&log->l_cilp->xc_push_lock); |
12e6a0f4 | 3823 | xlog_state_do_callback(log); |
1da177e4 | 3824 | |
1da177e4 | 3825 | /* return non-zero if log IOERROR transition had already happened */ |
014c2544 | 3826 | return retval; |
1da177e4 LT |
3827 | } |
3828 | ||
ba0f32d4 | 3829 | STATIC int |
9a8d2fdb MT |
3830 | xlog_iclogs_empty( |
3831 | struct xlog *log) | |
1da177e4 LT |
3832 | { |
3833 | xlog_in_core_t *iclog; | |
3834 | ||
3835 | iclog = log->l_iclog; | |
3836 | do { | |
3837 | /* endianness does not matter here, zero is zero in | |
3838 | * any language. | |
3839 | */ | |
3840 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3841 | return 0; |
1da177e4 LT |
3842 | iclog = iclog->ic_next; |
3843 | } while (iclog != log->l_iclog); | |
014c2544 | 3844 | return 1; |
1da177e4 | 3845 | } |
f661f1e0 | 3846 | |
a45086e2 BF |
3847 | /* |
3848 | * Verify that an LSN stamped into a piece of metadata is valid. This is | |
3849 | * intended for use in read verifiers on v5 superblocks. | |
3850 | */ | |
3851 | bool | |
3852 | xfs_log_check_lsn( | |
3853 | struct xfs_mount *mp, | |
3854 | xfs_lsn_t lsn) | |
3855 | { | |
3856 | struct xlog *log = mp->m_log; | |
3857 | bool valid; | |
3858 | ||
3859 | /* | |
3860 | * norecovery mode skips mount-time log processing and unconditionally | |
3861 | * resets the in-core LSN. We can't validate in this mode, but | |
3862 | * modifications are not allowed anyways so just return true. | |
3863 | */ | |
3864 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
3865 | return true; | |
3866 | ||
3867 | /* | |
3868 | * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is | |
3869 | * handled by recovery and thus safe to ignore here. | |
3870 | */ | |
3871 | if (lsn == NULLCOMMITLSN) | |
3872 | return true; | |
3873 | ||
3874 | valid = xlog_valid_lsn(mp->m_log, lsn); | |
3875 | ||
3876 | /* warn the user about what's gone wrong before verifier failure */ | |
3877 | if (!valid) { | |
3878 | spin_lock(&log->l_icloglock); | |
3879 | xfs_warn(mp, | |
3880 | "Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). " | |
3881 | "Please unmount and run xfs_repair (>= v4.3) to resolve.", | |
3882 | CYCLE_LSN(lsn), BLOCK_LSN(lsn), | |
3883 | log->l_curr_cycle, log->l_curr_block); | |
3884 | spin_unlock(&log->l_icloglock); | |
3885 | } | |
3886 | ||
3887 | return valid; | |
3888 | } | |
0c60d3aa DW |
3889 | |
3890 | bool | |
3891 | xfs_log_in_recovery( | |
3892 | struct xfs_mount *mp) | |
3893 | { | |
3894 | struct xlog *log = mp->m_log; | |
3895 | ||
3896 | return log->l_flags & XLOG_ACTIVE_RECOVERY; | |
3897 | } |