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 */ | |
956f6daa | 505 | trace_xlog_iclog_syncing(iclog, _RET_IP_); |
df732b29 CH |
506 | return true; |
507 | } | |
508 | ||
509 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
510 | return false; | |
511 | } | |
512 | ||
513 | /* | |
514 | * Flush iclog to disk if this is the last reference to the given iclog and the | |
515 | * it is in the WANT_SYNC state. | |
516 | */ | |
eef983ff | 517 | int |
df732b29 CH |
518 | xlog_state_release_iclog( |
519 | struct xlog *log, | |
520 | struct xlog_in_core *iclog) | |
521 | { | |
522 | lockdep_assert_held(&log->l_icloglock); | |
523 | ||
956f6daa | 524 | trace_xlog_iclog_release(iclog, _RET_IP_); |
1858bb0b | 525 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
df732b29 CH |
526 | return -EIO; |
527 | ||
528 | if (atomic_dec_and_test(&iclog->ic_refcnt) && | |
529 | __xlog_state_release_iclog(log, iclog)) { | |
530 | spin_unlock(&log->l_icloglock); | |
531 | xlog_sync(log, iclog); | |
532 | spin_lock(&log->l_icloglock); | |
533 | } | |
534 | ||
535 | return 0; | |
536 | } | |
537 | ||
1da177e4 LT |
538 | /* |
539 | * Mount a log filesystem | |
540 | * | |
541 | * mp - ubiquitous xfs mount point structure | |
542 | * log_target - buftarg of on-disk log device | |
543 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
544 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
545 | * | |
546 | * Return error or zero. | |
547 | */ | |
548 | int | |
249a8c11 DC |
549 | xfs_log_mount( |
550 | xfs_mount_t *mp, | |
551 | xfs_buftarg_t *log_target, | |
552 | xfs_daddr_t blk_offset, | |
553 | int num_bblks) | |
1da177e4 | 554 | { |
9c92ee20 | 555 | bool fatal = xfs_sb_version_hascrc(&mp->m_sb); |
3e7b91cf JL |
556 | int error = 0; |
557 | int min_logfsbs; | |
249a8c11 | 558 | |
c99d609a DC |
559 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
560 | xfs_notice(mp, "Mounting V%d Filesystem", | |
561 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
562 | } else { | |
a0fa2b67 | 563 | xfs_notice(mp, |
c99d609a DC |
564 | "Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.", |
565 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
bd186aa9 | 566 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
567 | } |
568 | ||
569 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e | 570 | if (IS_ERR(mp->m_log)) { |
2451337d | 571 | error = PTR_ERR(mp->m_log); |
644c3567 DC |
572 | goto out; |
573 | } | |
1da177e4 | 574 | |
3e7b91cf JL |
575 | /* |
576 | * Validate the given log space and drop a critical message via syslog | |
577 | * if the log size is too small that would lead to some unexpected | |
578 | * situations in transaction log space reservation stage. | |
579 | * | |
580 | * Note: we can't just reject the mount if the validation fails. This | |
581 | * would mean that people would have to downgrade their kernel just to | |
582 | * remedy the situation as there is no way to grow the log (short of | |
583 | * black magic surgery with xfs_db). | |
584 | * | |
585 | * We can, however, reject mounts for CRC format filesystems, as the | |
586 | * mkfs binary being used to make the filesystem should never create a | |
587 | * filesystem with a log that is too small. | |
588 | */ | |
589 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
590 | ||
591 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
592 | xfs_warn(mp, | |
593 | "Log size %d blocks too small, minimum size is %d blocks", | |
594 | mp->m_sb.sb_logblocks, min_logfsbs); | |
2451337d | 595 | error = -EINVAL; |
3e7b91cf JL |
596 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { |
597 | xfs_warn(mp, | |
598 | "Log size %d blocks too large, maximum size is %lld blocks", | |
599 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
2451337d | 600 | error = -EINVAL; |
3e7b91cf JL |
601 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { |
602 | xfs_warn(mp, | |
603 | "log size %lld bytes too large, maximum size is %lld bytes", | |
604 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
605 | XFS_MAX_LOG_BYTES); | |
2451337d | 606 | error = -EINVAL; |
9c92ee20 DW |
607 | } else if (mp->m_sb.sb_logsunit > 1 && |
608 | mp->m_sb.sb_logsunit % mp->m_sb.sb_blocksize) { | |
609 | xfs_warn(mp, | |
610 | "log stripe unit %u bytes must be a multiple of block size", | |
611 | mp->m_sb.sb_logsunit); | |
612 | error = -EINVAL; | |
613 | fatal = true; | |
3e7b91cf JL |
614 | } |
615 | if (error) { | |
9c92ee20 DW |
616 | /* |
617 | * Log check errors are always fatal on v5; or whenever bad | |
618 | * metadata leads to a crash. | |
619 | */ | |
620 | if (fatal) { | |
3e7b91cf JL |
621 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); |
622 | ASSERT(0); | |
623 | goto out_free_log; | |
624 | } | |
f41febd2 | 625 | xfs_crit(mp, "Log size out of supported range."); |
3e7b91cf | 626 | xfs_crit(mp, |
f41febd2 | 627 | "Continuing onwards, but if log hangs are experienced then please report this message in the bug report."); |
3e7b91cf JL |
628 | } |
629 | ||
249a8c11 DC |
630 | /* |
631 | * Initialize the AIL now we have a log. | |
632 | */ | |
249a8c11 DC |
633 | error = xfs_trans_ail_init(mp); |
634 | if (error) { | |
a0fa2b67 | 635 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 636 | goto out_free_log; |
249a8c11 | 637 | } |
a9c21c1b | 638 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 639 | |
1da177e4 LT |
640 | /* |
641 | * skip log recovery on a norecovery mount. pretend it all | |
642 | * just worked. | |
643 | */ | |
644 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 645 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
646 | |
647 | if (readonly) | |
bd186aa9 | 648 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 649 | |
65be6054 | 650 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
651 | |
652 | if (readonly) | |
bd186aa9 | 653 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 654 | if (error) { |
a0fa2b67 DC |
655 | xfs_warn(mp, "log mount/recovery failed: error %d", |
656 | error); | |
f0b2efad | 657 | xlog_recover_cancel(mp->m_log); |
26430752 | 658 | goto out_destroy_ail; |
1da177e4 LT |
659 | } |
660 | } | |
661 | ||
baff4e44 BF |
662 | error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj, |
663 | "log"); | |
664 | if (error) | |
665 | goto out_destroy_ail; | |
666 | ||
1da177e4 LT |
667 | /* Normal transactions can now occur */ |
668 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
669 | ||
71e330b5 DC |
670 | /* |
671 | * Now the log has been fully initialised and we know were our | |
672 | * space grant counters are, we can initialise the permanent ticket | |
673 | * needed for delayed logging to work. | |
674 | */ | |
675 | xlog_cil_init_post_recovery(mp->m_log); | |
676 | ||
1da177e4 | 677 | return 0; |
26430752 CH |
678 | |
679 | out_destroy_ail: | |
680 | xfs_trans_ail_destroy(mp); | |
681 | out_free_log: | |
682 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 683 | out: |
249a8c11 | 684 | return error; |
26430752 | 685 | } |
1da177e4 LT |
686 | |
687 | /* | |
f661f1e0 DC |
688 | * Finish the recovery of the file system. This is separate from the |
689 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
690 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
691 | * here. | |
1da177e4 | 692 | * |
f661f1e0 DC |
693 | * If we finish recovery successfully, start the background log work. If we are |
694 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
695 | * it. | |
1da177e4 LT |
696 | */ |
697 | int | |
f0b2efad BF |
698 | xfs_log_mount_finish( |
699 | struct xfs_mount *mp) | |
1da177e4 | 700 | { |
f661f1e0 | 701 | int error = 0; |
6f4a1eef | 702 | bool readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
f1b92bbc | 703 | bool recovered = mp->m_log->l_flags & XLOG_RECOVERY_NEEDED; |
1da177e4 | 704 | |
f0b2efad | 705 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) { |
bd186aa9 | 706 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
f0b2efad | 707 | return 0; |
6f4a1eef ES |
708 | } else if (readonly) { |
709 | /* Allow unlinked processing to proceed */ | |
710 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
1da177e4 LT |
711 | } |
712 | ||
8204f8dd DW |
713 | /* |
714 | * During the second phase of log recovery, we need iget and | |
715 | * iput to behave like they do for an active filesystem. | |
716 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
717 | * of inodes before we're done replaying log items on those | |
718 | * inodes. Turn it off immediately after recovery finishes | |
719 | * so that we don't leak the quota inodes if subsequent mount | |
720 | * activities fail. | |
799ea9e9 DW |
721 | * |
722 | * We let all inodes involved in redo item processing end up on | |
723 | * the LRU instead of being evicted immediately so that if we do | |
724 | * something to an unlinked inode, the irele won't cause | |
725 | * premature truncation and freeing of the inode, which results | |
726 | * in log recovery failure. We have to evict the unreferenced | |
1751e8a6 | 727 | * lru inodes after clearing SB_ACTIVE because we don't |
799ea9e9 DW |
728 | * otherwise clean up the lru if there's a subsequent failure in |
729 | * xfs_mountfs, which leads to us leaking the inodes if nothing | |
730 | * else (e.g. quotacheck) references the inodes before the | |
731 | * mount failure occurs. | |
8204f8dd | 732 | */ |
1751e8a6 | 733 | mp->m_super->s_flags |= SB_ACTIVE; |
f0b2efad BF |
734 | error = xlog_recover_finish(mp->m_log); |
735 | if (!error) | |
736 | xfs_log_work_queue(mp); | |
1751e8a6 | 737 | mp->m_super->s_flags &= ~SB_ACTIVE; |
799ea9e9 | 738 | evict_inodes(mp->m_super); |
f0b2efad | 739 | |
f1b92bbc BF |
740 | /* |
741 | * Drain the buffer LRU after log recovery. This is required for v4 | |
742 | * filesystems to avoid leaving around buffers with NULL verifier ops, | |
743 | * but we do it unconditionally to make sure we're always in a clean | |
744 | * cache state after mount. | |
745 | * | |
746 | * Don't push in the error case because the AIL may have pending intents | |
747 | * that aren't removed until recovery is cancelled. | |
748 | */ | |
749 | if (!error && recovered) { | |
750 | xfs_log_force(mp, XFS_LOG_SYNC); | |
751 | xfs_ail_push_all_sync(mp->m_ail); | |
752 | } | |
10fb9ac1 | 753 | xfs_buftarg_drain(mp->m_ddev_targp); |
f1b92bbc | 754 | |
6f4a1eef ES |
755 | if (readonly) |
756 | mp->m_flags |= XFS_MOUNT_RDONLY; | |
757 | ||
4e6b8270 DW |
758 | /* Make sure the log is dead if we're returning failure. */ |
759 | ASSERT(!error || (mp->m_log->l_flags & XLOG_IO_ERROR)); | |
760 | ||
f0b2efad BF |
761 | return error; |
762 | } | |
763 | ||
764 | /* | |
765 | * The mount has failed. Cancel the recovery if it hasn't completed and destroy | |
766 | * the log. | |
767 | */ | |
a7a9250e | 768 | void |
f0b2efad BF |
769 | xfs_log_mount_cancel( |
770 | struct xfs_mount *mp) | |
771 | { | |
a7a9250e | 772 | xlog_recover_cancel(mp->m_log); |
f0b2efad | 773 | xfs_log_unmount(mp); |
1da177e4 LT |
774 | } |
775 | ||
81e5b50a | 776 | /* |
a79b28c2 DC |
777 | * Wait for the iclog and all prior iclogs to be written disk as required by the |
778 | * log force state machine. Waiting on ic_force_wait ensures iclog completions | |
779 | * have been ordered and callbacks run before we are woken here, hence | |
780 | * guaranteeing that all the iclogs up to this one are on stable storage. | |
81e5b50a | 781 | */ |
a79b28c2 | 782 | int |
81e5b50a CH |
783 | xlog_wait_on_iclog( |
784 | struct xlog_in_core *iclog) | |
785 | __releases(iclog->ic_log->l_icloglock) | |
786 | { | |
787 | struct xlog *log = iclog->ic_log; | |
788 | ||
956f6daa | 789 | trace_xlog_iclog_wait_on(iclog, _RET_IP_); |
81e5b50a CH |
790 | if (!XLOG_FORCED_SHUTDOWN(log) && |
791 | iclog->ic_state != XLOG_STATE_ACTIVE && | |
792 | iclog->ic_state != XLOG_STATE_DIRTY) { | |
793 | XFS_STATS_INC(log->l_mp, xs_log_force_sleep); | |
794 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
795 | } else { | |
796 | spin_unlock(&log->l_icloglock); | |
797 | } | |
798 | ||
799 | if (XLOG_FORCED_SHUTDOWN(log)) | |
800 | return -EIO; | |
801 | return 0; | |
802 | } | |
803 | ||
1da177e4 | 804 | /* |
3c702f95 DC |
805 | * Write out an unmount record using the ticket provided. We have to account for |
806 | * the data space used in the unmount ticket as this write is not done from a | |
807 | * transaction context that has already done the accounting for us. | |
1da177e4 | 808 | */ |
3c702f95 DC |
809 | static int |
810 | xlog_write_unmount_record( | |
811 | struct xlog *log, | |
3468bb1c | 812 | struct xlog_ticket *ticket) |
53235f22 | 813 | { |
3c702f95 | 814 | struct xfs_unmount_log_format ulf = { |
53235f22 DW |
815 | .magic = XLOG_UNMOUNT_TYPE, |
816 | }; | |
817 | struct xfs_log_iovec reg = { | |
3c702f95 DC |
818 | .i_addr = &ulf, |
819 | .i_len = sizeof(ulf), | |
53235f22 DW |
820 | .i_type = XLOG_REG_TYPE_UNMOUNT, |
821 | }; | |
822 | struct xfs_log_vec vec = { | |
823 | .lv_niovecs = 1, | |
824 | .lv_iovecp = ®, | |
825 | }; | |
3c702f95 DC |
826 | |
827 | /* account for space used by record data */ | |
828 | ticket->t_curr_res -= sizeof(ulf); | |
eef983ff | 829 | |
3468bb1c | 830 | return xlog_write(log, &vec, ticket, NULL, NULL, XLOG_UNMOUNT_TRANS); |
3c702f95 DC |
831 | } |
832 | ||
833 | /* | |
834 | * Mark the filesystem clean by writing an unmount record to the head of the | |
835 | * log. | |
836 | */ | |
837 | static void | |
838 | xlog_unmount_write( | |
839 | struct xlog *log) | |
840 | { | |
841 | struct xfs_mount *mp = log->l_mp; | |
53235f22 DW |
842 | struct xlog_in_core *iclog; |
843 | struct xlog_ticket *tic = NULL; | |
53235f22 DW |
844 | int error; |
845 | ||
846 | error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0); | |
847 | if (error) | |
848 | goto out_err; | |
849 | ||
3468bb1c | 850 | error = xlog_write_unmount_record(log, tic); |
53235f22 DW |
851 | /* |
852 | * At this point, we're umounting anyway, so there's no point in | |
853 | * transitioning log state to IOERROR. Just continue... | |
854 | */ | |
855 | out_err: | |
856 | if (error) | |
857 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
858 | ||
859 | spin_lock(&log->l_icloglock); | |
860 | iclog = log->l_iclog; | |
861 | atomic_inc(&iclog->ic_refcnt); | |
69363999 CH |
862 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
863 | xlog_state_switch_iclogs(log, iclog, 0); | |
864 | else | |
865 | ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
866 | iclog->ic_state == XLOG_STATE_IOERROR); | |
eef983ff DC |
867 | /* |
868 | * Ensure the journal is fully flushed and on stable storage once the | |
869 | * iclog containing the unmount record is written. | |
870 | */ | |
871 | iclog->ic_flags |= (XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA); | |
53235f22 | 872 | error = xlog_state_release_iclog(log, iclog); |
81e5b50a | 873 | xlog_wait_on_iclog(iclog); |
53235f22 DW |
874 | |
875 | if (tic) { | |
876 | trace_xfs_log_umount_write(log, tic); | |
8b41e3f9 | 877 | xfs_log_ticket_ungrant(log, tic); |
53235f22 DW |
878 | } |
879 | } | |
880 | ||
13859c98 CH |
881 | static void |
882 | xfs_log_unmount_verify_iclog( | |
883 | struct xlog *log) | |
884 | { | |
885 | struct xlog_in_core *iclog = log->l_iclog; | |
886 | ||
887 | do { | |
888 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
889 | ASSERT(iclog->ic_offset == 0); | |
890 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
891 | } | |
892 | ||
1da177e4 LT |
893 | /* |
894 | * Unmount record used to have a string "Unmount filesystem--" in the | |
895 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
896 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 897 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
898 | * As far as I know, there weren't any dependencies on the old behaviour. |
899 | */ | |
550319e9 | 900 | static void |
13859c98 CH |
901 | xfs_log_unmount_write( |
902 | struct xfs_mount *mp) | |
1da177e4 | 903 | { |
13859c98 | 904 | struct xlog *log = mp->m_log; |
1da177e4 | 905 | |
50d25484 | 906 | if (!xfs_log_writable(mp)) |
550319e9 | 907 | return; |
1da177e4 | 908 | |
550319e9 | 909 | xfs_log_force(mp, XFS_LOG_SYNC); |
1da177e4 | 910 | |
6178d104 CH |
911 | if (XLOG_FORCED_SHUTDOWN(log)) |
912 | return; | |
5cc3c006 DW |
913 | |
914 | /* | |
915 | * If we think the summary counters are bad, avoid writing the unmount | |
916 | * record to force log recovery at next mount, after which the summary | |
917 | * counters will be recalculated. Refer to xlog_check_unmount_rec for | |
918 | * more details. | |
919 | */ | |
920 | if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp, | |
921 | XFS_ERRTAG_FORCE_SUMMARY_RECALC)) { | |
922 | xfs_alert(mp, "%s: will fix summary counters at next mount", | |
923 | __func__); | |
924 | return; | |
925 | } | |
926 | ||
13859c98 | 927 | xfs_log_unmount_verify_iclog(log); |
3c702f95 | 928 | xlog_unmount_write(log); |
550319e9 | 929 | } |
1da177e4 LT |
930 | |
931 | /* | |
c75921a7 | 932 | * Empty the log for unmount/freeze. |
cf2931db DC |
933 | * |
934 | * To do this, we first need to shut down the background log work so it is not | |
935 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
936 | * the log so we can then flush them out. Once they have completed their IO and | |
303591a0 | 937 | * run the callbacks removing themselves from the AIL, we can cover the log. |
1da177e4 | 938 | */ |
303591a0 | 939 | int |
c75921a7 DC |
940 | xfs_log_quiesce( |
941 | struct xfs_mount *mp) | |
1da177e4 | 942 | { |
f661f1e0 | 943 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
944 | xfs_log_force(mp, XFS_LOG_SYNC); |
945 | ||
946 | /* | |
947 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
8321ddb2 | 948 | * will push it, xfs_buftarg_wait() will not wait for it. Further, |
cf2931db DC |
949 | * xfs_buf_iowait() cannot be used because it was pushed with the |
950 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
951 | * the IO to complete. | |
952 | */ | |
953 | xfs_ail_push_all_sync(mp->m_ail); | |
8321ddb2 | 954 | xfs_buftarg_wait(mp->m_ddev_targp); |
cf2931db DC |
955 | xfs_buf_lock(mp->m_sb_bp); |
956 | xfs_buf_unlock(mp->m_sb_bp); | |
303591a0 BF |
957 | |
958 | return xfs_log_cover(mp); | |
9e54ee0f | 959 | } |
cf2931db | 960 | |
9e54ee0f BF |
961 | void |
962 | xfs_log_clean( | |
963 | struct xfs_mount *mp) | |
964 | { | |
965 | xfs_log_quiesce(mp); | |
cf2931db | 966 | xfs_log_unmount_write(mp); |
c75921a7 DC |
967 | } |
968 | ||
969 | /* | |
970 | * Shut down and release the AIL and Log. | |
971 | * | |
972 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
973 | * from the AIL so that the log is empty before we write the unmount record to | |
974 | * the log. Once this is done, we can tear down the AIL and the log. | |
975 | */ | |
976 | void | |
977 | xfs_log_unmount( | |
978 | struct xfs_mount *mp) | |
979 | { | |
9e54ee0f | 980 | xfs_log_clean(mp); |
cf2931db | 981 | |
8321ddb2 BF |
982 | xfs_buftarg_drain(mp->m_ddev_targp); |
983 | ||
249a8c11 | 984 | xfs_trans_ail_destroy(mp); |
baff4e44 BF |
985 | |
986 | xfs_sysfs_del(&mp->m_log->l_kobj); | |
987 | ||
c41564b5 | 988 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
989 | } |
990 | ||
43f5efc5 DC |
991 | void |
992 | xfs_log_item_init( | |
993 | struct xfs_mount *mp, | |
994 | struct xfs_log_item *item, | |
995 | int type, | |
272e42b2 | 996 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
997 | { |
998 | item->li_mountp = mp; | |
999 | item->li_ailp = mp->m_ail; | |
1000 | item->li_type = type; | |
1001 | item->li_ops = ops; | |
71e330b5 DC |
1002 | item->li_lv = NULL; |
1003 | ||
1004 | INIT_LIST_HEAD(&item->li_ail); | |
1005 | INIT_LIST_HEAD(&item->li_cil); | |
643c8c05 | 1006 | INIT_LIST_HEAD(&item->li_bio_list); |
e6631f85 | 1007 | INIT_LIST_HEAD(&item->li_trans); |
43f5efc5 DC |
1008 | } |
1009 | ||
09a423a3 CH |
1010 | /* |
1011 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 1012 | */ |
1da177e4 | 1013 | void |
09a423a3 | 1014 | xfs_log_space_wake( |
cfb7cdca | 1015 | struct xfs_mount *mp) |
1da177e4 | 1016 | { |
ad223e60 | 1017 | struct xlog *log = mp->m_log; |
cfb7cdca | 1018 | int free_bytes; |
1da177e4 | 1019 | |
1da177e4 LT |
1020 | if (XLOG_FORCED_SHUTDOWN(log)) |
1021 | return; | |
1da177e4 | 1022 | |
28496968 | 1023 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
1024 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1025 | ||
28496968 CH |
1026 | spin_lock(&log->l_write_head.lock); |
1027 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 1028 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 1029 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 1030 | } |
10547941 | 1031 | |
28496968 | 1032 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
1033 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1034 | ||
28496968 CH |
1035 | spin_lock(&log->l_reserve_head.lock); |
1036 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 1037 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 1038 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 1039 | } |
3f16b985 | 1040 | } |
1da177e4 LT |
1041 | |
1042 | /* | |
2c6e24ce DC |
1043 | * Determine if we have a transaction that has gone to disk that needs to be |
1044 | * covered. To begin the transition to the idle state firstly the log needs to | |
1045 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1046 | * we start attempting to cover the log. | |
b6f8dd49 | 1047 | * |
2c6e24ce DC |
1048 | * Only if we are then in a state where covering is needed, the caller is |
1049 | * informed that dummy transactions are required to move the log into the idle | |
1050 | * state. | |
1051 | * | |
1052 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1053 | * cover the log as we may be in a situation where there isn't log space | |
1054 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1055 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1056 | * there's no point in running a dummy transaction at this point because we | |
1057 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 | 1058 | */ |
37444fc4 BF |
1059 | static bool |
1060 | xfs_log_need_covered( | |
1061 | struct xfs_mount *mp) | |
1da177e4 | 1062 | { |
37444fc4 BF |
1063 | struct xlog *log = mp->m_log; |
1064 | bool needed = false; | |
1da177e4 | 1065 | |
2c6e24ce | 1066 | if (!xlog_cil_empty(log)) |
8646b982 | 1067 | return false; |
2c6e24ce | 1068 | |
b22cd72c | 1069 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1070 | switch (log->l_covered_state) { |
1071 | case XLOG_STATE_COVER_DONE: | |
1072 | case XLOG_STATE_COVER_DONE2: | |
1073 | case XLOG_STATE_COVER_IDLE: | |
1074 | break; | |
1075 | case XLOG_STATE_COVER_NEED: | |
1076 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1077 | if (xfs_ail_min_lsn(log->l_ailp)) |
1078 | break; | |
1079 | if (!xlog_iclogs_empty(log)) | |
1080 | break; | |
1081 | ||
37444fc4 | 1082 | needed = true; |
2c6e24ce DC |
1083 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) |
1084 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1085 | else | |
1086 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1087 | break; | |
b6f8dd49 | 1088 | default: |
37444fc4 | 1089 | needed = true; |
b6f8dd49 | 1090 | break; |
1da177e4 | 1091 | } |
b22cd72c | 1092 | spin_unlock(&log->l_icloglock); |
014c2544 | 1093 | return needed; |
1da177e4 LT |
1094 | } |
1095 | ||
303591a0 BF |
1096 | /* |
1097 | * Explicitly cover the log. This is similar to background log covering but | |
1098 | * intended for usage in quiesce codepaths. The caller is responsible to ensure | |
1099 | * the log is idle and suitable for covering. The CIL, iclog buffers and AIL | |
1100 | * must all be empty. | |
1101 | */ | |
1102 | static int | |
1103 | xfs_log_cover( | |
1104 | struct xfs_mount *mp) | |
1105 | { | |
303591a0 | 1106 | int error = 0; |
f46e5a17 | 1107 | bool need_covered; |
303591a0 | 1108 | |
4533fc63 BF |
1109 | ASSERT((xlog_cil_empty(mp->m_log) && xlog_iclogs_empty(mp->m_log) && |
1110 | !xfs_ail_min_lsn(mp->m_log->l_ailp)) || | |
303591a0 BF |
1111 | XFS_FORCED_SHUTDOWN(mp)); |
1112 | ||
1113 | if (!xfs_log_writable(mp)) | |
1114 | return 0; | |
1115 | ||
f46e5a17 BF |
1116 | /* |
1117 | * xfs_log_need_covered() is not idempotent because it progresses the | |
1118 | * state machine if the log requires covering. Therefore, we must call | |
1119 | * this function once and use the result until we've issued an sb sync. | |
1120 | * Do so first to make that abundantly clear. | |
1121 | * | |
1122 | * Fall into the covering sequence if the log needs covering or the | |
1123 | * mount has lazy superblock accounting to sync to disk. The sb sync | |
1124 | * used for covering accumulates the in-core counters, so covering | |
1125 | * handles this for us. | |
1126 | */ | |
1127 | need_covered = xfs_log_need_covered(mp); | |
1128 | if (!need_covered && !xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1129 | return 0; | |
1130 | ||
303591a0 BF |
1131 | /* |
1132 | * To cover the log, commit the superblock twice (at most) in | |
1133 | * independent checkpoints. The first serves as a reference for the | |
1134 | * tail pointer. The sync transaction and AIL push empties the AIL and | |
1135 | * updates the in-core tail to the LSN of the first checkpoint. The | |
1136 | * second commit updates the on-disk tail with the in-core LSN, | |
1137 | * covering the log. Push the AIL one more time to leave it empty, as | |
1138 | * we found it. | |
1139 | */ | |
f46e5a17 | 1140 | do { |
303591a0 BF |
1141 | error = xfs_sync_sb(mp, true); |
1142 | if (error) | |
1143 | break; | |
1144 | xfs_ail_push_all_sync(mp->m_ail); | |
f46e5a17 | 1145 | } while (xfs_log_need_covered(mp)); |
303591a0 BF |
1146 | |
1147 | return error; | |
1148 | } | |
1149 | ||
09a423a3 | 1150 | /* |
1da177e4 LT |
1151 | * We may be holding the log iclog lock upon entering this routine. |
1152 | */ | |
1153 | xfs_lsn_t | |
1c304625 | 1154 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1155 | struct xfs_mount *mp) |
1da177e4 | 1156 | { |
ad223e60 | 1157 | struct xlog *log = mp->m_log; |
1c304625 CH |
1158 | struct xfs_log_item *lip; |
1159 | xfs_lsn_t tail_lsn; | |
1160 | ||
57e80956 | 1161 | assert_spin_locked(&mp->m_ail->ail_lock); |
1da177e4 | 1162 | |
09a423a3 CH |
1163 | /* |
1164 | * To make sure we always have a valid LSN for the log tail we keep | |
1165 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1166 | * and use that when the AIL was empty. |
09a423a3 | 1167 | */ |
1c304625 CH |
1168 | lip = xfs_ail_min(mp->m_ail); |
1169 | if (lip) | |
1170 | tail_lsn = lip->li_lsn; | |
1171 | else | |
84f3c683 | 1172 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
750b9c90 | 1173 | trace_xfs_log_assign_tail_lsn(log, tail_lsn); |
1c3cb9ec | 1174 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1175 | return tail_lsn; |
1c3cb9ec | 1176 | } |
1da177e4 | 1177 | |
1c304625 CH |
1178 | xfs_lsn_t |
1179 | xlog_assign_tail_lsn( | |
1180 | struct xfs_mount *mp) | |
1181 | { | |
1182 | xfs_lsn_t tail_lsn; | |
1183 | ||
57e80956 | 1184 | spin_lock(&mp->m_ail->ail_lock); |
1c304625 | 1185 | tail_lsn = xlog_assign_tail_lsn_locked(mp); |
57e80956 | 1186 | spin_unlock(&mp->m_ail->ail_lock); |
1c304625 CH |
1187 | |
1188 | return tail_lsn; | |
1189 | } | |
1190 | ||
1da177e4 LT |
1191 | /* |
1192 | * Return the space in the log between the tail and the head. The head | |
1193 | * is passed in the cycle/bytes formal parms. In the special case where | |
1194 | * the reserve head has wrapped passed the tail, this calculation is no | |
1195 | * longer valid. In this case, just return 0 which means there is no space | |
1196 | * in the log. This works for all places where this function is called | |
1197 | * with the reserve head. Of course, if the write head were to ever | |
1198 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1199 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1200 | * | |
1201 | * This code also handles the case where the reservation head is behind | |
1202 | * the tail. The details of this case are described below, but the end | |
1203 | * result is that we return the size of the log as the amount of space left. | |
1204 | */ | |
a8272ce0 | 1205 | STATIC int |
a69ed03c | 1206 | xlog_space_left( |
ad223e60 | 1207 | struct xlog *log, |
c8a09ff8 | 1208 | atomic64_t *head) |
1da177e4 | 1209 | { |
a69ed03c DC |
1210 | int free_bytes; |
1211 | int tail_bytes; | |
1212 | int tail_cycle; | |
1213 | int head_cycle; | |
1214 | int head_bytes; | |
1da177e4 | 1215 | |
a69ed03c | 1216 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1217 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1218 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1219 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1220 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1221 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1222 | return 0; |
a69ed03c DC |
1223 | else if (tail_cycle < head_cycle) { |
1224 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1225 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1226 | } else { |
1227 | /* | |
1228 | * The reservation head is behind the tail. | |
1229 | * In this case we just want to return the size of the | |
1230 | * log as the amount of space left. | |
1231 | */ | |
f41febd2 | 1232 | xfs_alert(log->l_mp, "xlog_space_left: head behind tail"); |
a0fa2b67 | 1233 | xfs_alert(log->l_mp, |
f41febd2 JP |
1234 | " tail_cycle = %d, tail_bytes = %d", |
1235 | tail_cycle, tail_bytes); | |
1236 | xfs_alert(log->l_mp, | |
1237 | " GH cycle = %d, GH bytes = %d", | |
1238 | head_cycle, head_bytes); | |
1da177e4 LT |
1239 | ASSERT(0); |
1240 | free_bytes = log->l_logsize; | |
1241 | } | |
1242 | return free_bytes; | |
a69ed03c | 1243 | } |
1da177e4 LT |
1244 | |
1245 | ||
0d5a75e9 | 1246 | static void |
79b54d9b CH |
1247 | xlog_ioend_work( |
1248 | struct work_struct *work) | |
1da177e4 | 1249 | { |
79b54d9b CH |
1250 | struct xlog_in_core *iclog = |
1251 | container_of(work, struct xlog_in_core, ic_end_io_work); | |
1252 | struct xlog *log = iclog->ic_log; | |
79b54d9b | 1253 | int error; |
1da177e4 | 1254 | |
79b54d9b | 1255 | error = blk_status_to_errno(iclog->ic_bio.bi_status); |
366fc4b8 CH |
1256 | #ifdef DEBUG |
1257 | /* treat writes with injected CRC errors as failed */ | |
1258 | if (iclog->ic_fail_crc) | |
79b54d9b | 1259 | error = -EIO; |
366fc4b8 CH |
1260 | #endif |
1261 | ||
1da177e4 | 1262 | /* |
366fc4b8 | 1263 | * Race to shutdown the filesystem if we see an error. |
1da177e4 | 1264 | */ |
79b54d9b CH |
1265 | if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) { |
1266 | xfs_alert(log->l_mp, "log I/O error %d", error); | |
1267 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1da177e4 | 1268 | } |
3db296f3 | 1269 | |
12e6a0f4 | 1270 | xlog_state_done_syncing(iclog); |
79b54d9b | 1271 | bio_uninit(&iclog->ic_bio); |
9c23eccc | 1272 | |
3db296f3 | 1273 | /* |
79b54d9b CH |
1274 | * Drop the lock to signal that we are done. Nothing references the |
1275 | * iclog after this, so an unmount waiting on this lock can now tear it | |
1276 | * down safely. As such, it is unsafe to reference the iclog after the | |
1277 | * unlock as we could race with it being freed. | |
3db296f3 | 1278 | */ |
79b54d9b | 1279 | up(&iclog->ic_sema); |
c3f8fc73 | 1280 | } |
1da177e4 | 1281 | |
1da177e4 LT |
1282 | /* |
1283 | * Return size of each in-core log record buffer. | |
1284 | * | |
9da096fd | 1285 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1286 | * |
1287 | * If the filesystem blocksize is too large, we may need to choose a | |
1288 | * larger size since the directory code currently logs entire blocks. | |
1289 | */ | |
1da177e4 | 1290 | STATIC void |
9a8d2fdb MT |
1291 | xlog_get_iclog_buffer_size( |
1292 | struct xfs_mount *mp, | |
1293 | struct xlog *log) | |
1da177e4 | 1294 | { |
1cb51258 | 1295 | if (mp->m_logbufs <= 0) |
4f62282a CH |
1296 | mp->m_logbufs = XLOG_MAX_ICLOGS; |
1297 | if (mp->m_logbsize <= 0) | |
1298 | mp->m_logbsize = XLOG_BIG_RECORD_BSIZE; | |
1299 | ||
1300 | log->l_iclog_bufs = mp->m_logbufs; | |
1301 | log->l_iclog_size = mp->m_logbsize; | |
1da177e4 LT |
1302 | |
1303 | /* | |
4f62282a | 1304 | * # headers = size / 32k - one header holds cycles from 32k of data. |
1da177e4 | 1305 | */ |
4f62282a CH |
1306 | log->l_iclog_heads = |
1307 | DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE); | |
1308 | log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT; | |
1309 | } | |
1da177e4 | 1310 | |
f661f1e0 DC |
1311 | void |
1312 | xfs_log_work_queue( | |
1313 | struct xfs_mount *mp) | |
1314 | { | |
696a5620 | 1315 | queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1316 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * Every sync period we need to unpin all items in the AIL and push them to | |
1321 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1322 | * indicate that the filesystem is idle. | |
1323 | */ | |
0d5a75e9 | 1324 | static void |
f661f1e0 DC |
1325 | xfs_log_worker( |
1326 | struct work_struct *work) | |
1327 | { | |
1328 | struct xlog *log = container_of(to_delayed_work(work), | |
1329 | struct xlog, l_work); | |
1330 | struct xfs_mount *mp = log->l_mp; | |
1331 | ||
1332 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
37444fc4 | 1333 | if (xfs_fs_writable(mp, SB_FREEZE_WRITE) && xfs_log_need_covered(mp)) { |
61e63ecb DC |
1334 | /* |
1335 | * Dump a transaction into the log that contains no real change. | |
1336 | * This is needed to stamp the current tail LSN into the log | |
1337 | * during the covering operation. | |
1338 | * | |
1339 | * We cannot use an inode here for this - that will push dirty | |
1340 | * state back up into the VFS and then periodic inode flushing | |
1341 | * will prevent log covering from making progress. Hence we | |
1342 | * synchronously log the superblock instead to ensure the | |
1343 | * superblock is immediately unpinned and can be written back. | |
1344 | */ | |
1345 | xfs_sync_sb(mp, true); | |
1346 | } else | |
f661f1e0 DC |
1347 | xfs_log_force(mp, 0); |
1348 | ||
1349 | /* start pushing all the metadata that is currently dirty */ | |
1350 | xfs_ail_push_all(mp->m_ail); | |
1351 | ||
1352 | /* queue us up again */ | |
1353 | xfs_log_work_queue(mp); | |
1354 | } | |
1355 | ||
1da177e4 LT |
1356 | /* |
1357 | * This routine initializes some of the log structure for a given mount point. | |
1358 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1359 | * some other stuff may be filled in too. | |
1360 | */ | |
9a8d2fdb MT |
1361 | STATIC struct xlog * |
1362 | xlog_alloc_log( | |
1363 | struct xfs_mount *mp, | |
1364 | struct xfs_buftarg *log_target, | |
1365 | xfs_daddr_t blk_offset, | |
1366 | int num_bblks) | |
1da177e4 | 1367 | { |
9a8d2fdb | 1368 | struct xlog *log; |
1da177e4 LT |
1369 | xlog_rec_header_t *head; |
1370 | xlog_in_core_t **iclogp; | |
1371 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1da177e4 | 1372 | int i; |
2451337d | 1373 | int error = -ENOMEM; |
69ce58f0 | 1374 | uint log2_size = 0; |
1da177e4 | 1375 | |
9a8d2fdb | 1376 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1377 | if (!log) { |
a0fa2b67 | 1378 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1379 | goto out; |
1380 | } | |
1da177e4 LT |
1381 | |
1382 | log->l_mp = mp; | |
1383 | log->l_targ = log_target; | |
1384 | log->l_logsize = BBTOB(num_bblks); | |
1385 | log->l_logBBstart = blk_offset; | |
1386 | log->l_logBBsize = num_bblks; | |
1387 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1388 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1389 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1390 | |
1391 | log->l_prev_block = -1; | |
1da177e4 | 1392 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1393 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1394 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1395 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 | 1396 | |
a6a65fef DC |
1397 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) |
1398 | log->l_iclog_roundoff = mp->m_sb.sb_logsunit; | |
1399 | else | |
1400 | log->l_iclog_roundoff = BBSIZE; | |
1401 | ||
c303c5b8 CH |
1402 | xlog_grant_head_init(&log->l_reserve_head); |
1403 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1404 | |
2451337d | 1405 | error = -EFSCORRUPTED; |
62118709 | 1406 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1407 | log2_size = mp->m_sb.sb_logsectlog; |
1408 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1409 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1410 | log2_size, BBSHIFT); | |
a6cb767e DC |
1411 | goto out_free_log; |
1412 | } | |
1413 | ||
69ce58f0 AE |
1414 | log2_size -= BBSHIFT; |
1415 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1416 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1417 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1418 | goto out_free_log; |
1419 | } | |
69ce58f0 AE |
1420 | |
1421 | /* for larger sector sizes, must have v2 or external log */ | |
1422 | if (log2_size && log->l_logBBstart > 0 && | |
1423 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1424 | xfs_warn(mp, |
1425 | "log sector size (0x%x) invalid for configuration.", | |
1426 | log2_size); | |
a6cb767e DC |
1427 | goto out_free_log; |
1428 | } | |
1da177e4 | 1429 | } |
69ce58f0 | 1430 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1431 | |
1432 | xlog_get_iclog_buffer_size(mp, log); | |
1433 | ||
007c61c6 | 1434 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1435 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1436 | |
1da177e4 LT |
1437 | iclogp = &log->l_iclog; |
1438 | /* | |
1439 | * The amount of memory to allocate for the iclog structure is | |
1440 | * rather funky due to the way the structure is defined. It is | |
1441 | * done this way so that we can use different sizes for machines | |
1442 | * with different amounts of memory. See the definition of | |
1443 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1444 | */ | |
1da177e4 | 1445 | ASSERT(log->l_iclog_size >= 4096); |
79b54d9b | 1446 | for (i = 0; i < log->l_iclog_bufs; i++) { |
f8f9ee47 | 1447 | int align_mask = xfs_buftarg_dma_alignment(mp->m_logdev_targp); |
89b171ac CH |
1448 | size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE) * |
1449 | sizeof(struct bio_vec); | |
79b54d9b CH |
1450 | |
1451 | iclog = kmem_zalloc(sizeof(*iclog) + bvec_size, KM_MAYFAIL); | |
1452 | if (!iclog) | |
644c3567 DC |
1453 | goto out_free_iclog; |
1454 | ||
79b54d9b | 1455 | *iclogp = iclog; |
1da177e4 LT |
1456 | iclog->ic_prev = prev_iclog; |
1457 | prev_iclog = iclog; | |
1fa40b01 | 1458 | |
f8f9ee47 | 1459 | iclog->ic_data = kmem_alloc_io(log->l_iclog_size, align_mask, |
3219e8cf | 1460 | KM_MAYFAIL | KM_ZERO); |
79b54d9b | 1461 | if (!iclog->ic_data) |
644c3567 | 1462 | goto out_free_iclog; |
4679b2d3 | 1463 | #ifdef DEBUG |
5809d5e0 | 1464 | log->l_iclog_bak[i] = &iclog->ic_header; |
4679b2d3 | 1465 | #endif |
1da177e4 LT |
1466 | head = &iclog->ic_header; |
1467 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1468 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1469 | head->h_version = cpu_to_be32( | |
62118709 | 1470 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1471 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1472 | /* new fields */ |
b53e675d | 1473 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1474 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1475 | ||
79b54d9b | 1476 | iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize; |
1da177e4 LT |
1477 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1478 | iclog->ic_log = log; | |
114d23aa | 1479 | atomic_set(&iclog->ic_refcnt, 0); |
89ae379d | 1480 | INIT_LIST_HEAD(&iclog->ic_callbacks); |
b28708d6 | 1481 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1482 | |
eb40a875 DC |
1483 | init_waitqueue_head(&iclog->ic_force_wait); |
1484 | init_waitqueue_head(&iclog->ic_write_wait); | |
79b54d9b CH |
1485 | INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work); |
1486 | sema_init(&iclog->ic_sema, 1); | |
1da177e4 LT |
1487 | |
1488 | iclogp = &iclog->ic_next; | |
1489 | } | |
1490 | *iclogp = log->l_iclog; /* complete ring */ | |
1491 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1492 | ||
1058d0f5 | 1493 | log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s", |
05a302a1 DW |
1494 | XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | |
1495 | WQ_HIGHPRI), | |
1496 | 0, mp->m_super->s_id); | |
1058d0f5 CH |
1497 | if (!log->l_ioend_workqueue) |
1498 | goto out_free_iclog; | |
1499 | ||
71e330b5 DC |
1500 | error = xlog_cil_init(log); |
1501 | if (error) | |
1058d0f5 | 1502 | goto out_destroy_workqueue; |
1da177e4 | 1503 | return log; |
644c3567 | 1504 | |
1058d0f5 CH |
1505 | out_destroy_workqueue: |
1506 | destroy_workqueue(log->l_ioend_workqueue); | |
644c3567 DC |
1507 | out_free_iclog: |
1508 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1509 | prev_iclog = iclog->ic_next; | |
79b54d9b | 1510 | kmem_free(iclog->ic_data); |
644c3567 | 1511 | kmem_free(iclog); |
798a9cad BF |
1512 | if (prev_iclog == log->l_iclog) |
1513 | break; | |
644c3567 | 1514 | } |
644c3567 DC |
1515 | out_free_log: |
1516 | kmem_free(log); | |
a6cb767e | 1517 | out: |
2451337d | 1518 | return ERR_PTR(error); |
1da177e4 LT |
1519 | } /* xlog_alloc_log */ |
1520 | ||
1da177e4 LT |
1521 | /* |
1522 | * Write out the commit record of a transaction associated with the given | |
f10e925d | 1523 | * ticket to close off a running log write. Return the lsn of the commit record. |
1da177e4 | 1524 | */ |
f10e925d | 1525 | int |
55b66332 | 1526 | xlog_commit_record( |
ad223e60 | 1527 | struct xlog *log, |
55b66332 DC |
1528 | struct xlog_ticket *ticket, |
1529 | struct xlog_in_core **iclog, | |
f10e925d | 1530 | xfs_lsn_t *lsn) |
1da177e4 | 1531 | { |
55b66332 DC |
1532 | struct xfs_log_iovec reg = { |
1533 | .i_addr = NULL, | |
1534 | .i_len = 0, | |
1535 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1536 | }; | |
1537 | struct xfs_log_vec vec = { | |
1538 | .lv_niovecs = 1, | |
1539 | .lv_iovecp = ®, | |
1540 | }; | |
f10e925d | 1541 | int error; |
1da177e4 | 1542 | |
f10e925d DC |
1543 | if (XLOG_FORCED_SHUTDOWN(log)) |
1544 | return -EIO; | |
1545 | ||
3468bb1c | 1546 | error = xlog_write(log, &vec, ticket, lsn, iclog, XLOG_COMMIT_TRANS); |
55b66332 | 1547 | if (error) |
f10e925d | 1548 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1549 | return error; |
55b66332 | 1550 | } |
1da177e4 LT |
1551 | |
1552 | /* | |
ed1575da DW |
1553 | * Compute the LSN that we'd need to push the log tail towards in order to have |
1554 | * (a) enough on-disk log space to log the number of bytes specified, (b) at | |
1555 | * least 25% of the log space free, and (c) at least 256 blocks free. If the | |
1556 | * log free space already meets all three thresholds, this function returns | |
1557 | * NULLCOMMITLSN. | |
1da177e4 | 1558 | */ |
ed1575da DW |
1559 | xfs_lsn_t |
1560 | xlog_grant_push_threshold( | |
ad223e60 | 1561 | struct xlog *log, |
2ced19cb | 1562 | int need_bytes) |
1da177e4 | 1563 | { |
2ced19cb | 1564 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1565 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1566 | int free_blocks; |
1567 | int free_bytes; | |
1568 | int threshold_block; | |
1569 | int threshold_cycle; | |
1570 | int free_threshold; | |
1571 | ||
1572 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1573 | ||
28496968 | 1574 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1575 | free_blocks = BTOBBT(free_bytes); |
1576 | ||
1577 | /* | |
1578 | * Set the threshold for the minimum number of free blocks in the | |
1579 | * log to the maximum of what the caller needs, one quarter of the | |
1580 | * log, and 256 blocks. | |
1581 | */ | |
1582 | free_threshold = BTOBB(need_bytes); | |
9bb54cb5 DC |
1583 | free_threshold = max(free_threshold, (log->l_logBBsize >> 2)); |
1584 | free_threshold = max(free_threshold, 256); | |
2ced19cb | 1585 | if (free_blocks >= free_threshold) |
ed1575da | 1586 | return NULLCOMMITLSN; |
2ced19cb | 1587 | |
1c3cb9ec DC |
1588 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1589 | &threshold_block); | |
1590 | threshold_block += free_threshold; | |
1da177e4 | 1591 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1592 | threshold_block -= log->l_logBBsize; |
1593 | threshold_cycle += 1; | |
1da177e4 | 1594 | } |
2ced19cb DC |
1595 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1596 | threshold_block); | |
1597 | /* | |
1598 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1599 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1600 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1601 | */ |
84f3c683 DC |
1602 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1603 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1604 | threshold_lsn = last_sync_lsn; | |
2ced19cb | 1605 | |
ed1575da DW |
1606 | return threshold_lsn; |
1607 | } | |
1608 | ||
1609 | /* | |
1610 | * Push the tail of the log if we need to do so to maintain the free log space | |
1611 | * thresholds set out by xlog_grant_push_threshold. We may need to adopt a | |
1612 | * policy which pushes on an lsn which is further along in the log once we | |
1613 | * reach the high water mark. In this manner, we would be creating a low water | |
1614 | * mark. | |
1615 | */ | |
1616 | STATIC void | |
1617 | xlog_grant_push_ail( | |
1618 | struct xlog *log, | |
1619 | int need_bytes) | |
1620 | { | |
1621 | xfs_lsn_t threshold_lsn; | |
1622 | ||
1623 | threshold_lsn = xlog_grant_push_threshold(log, need_bytes); | |
1624 | if (threshold_lsn == NULLCOMMITLSN || XLOG_FORCED_SHUTDOWN(log)) | |
1625 | return; | |
1626 | ||
2ced19cb DC |
1627 | /* |
1628 | * Get the transaction layer to kick the dirty buffers out to | |
1629 | * disk asynchronously. No point in trying to do this if | |
1630 | * the filesystem is shutting down. | |
1631 | */ | |
ed1575da | 1632 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1633 | } |
1da177e4 | 1634 | |
0e446be4 CH |
1635 | /* |
1636 | * Stamp cycle number in every block | |
1637 | */ | |
1638 | STATIC void | |
1639 | xlog_pack_data( | |
1640 | struct xlog *log, | |
1641 | struct xlog_in_core *iclog, | |
1642 | int roundoff) | |
1643 | { | |
1644 | int i, j, k; | |
1645 | int size = iclog->ic_offset + roundoff; | |
1646 | __be32 cycle_lsn; | |
b2a922cd | 1647 | char *dp; |
0e446be4 CH |
1648 | |
1649 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1650 | ||
1651 | dp = iclog->ic_datap; | |
1652 | for (i = 0; i < BTOBB(size); i++) { | |
1653 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1654 | break; | |
1655 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1656 | *(__be32 *)dp = cycle_lsn; | |
1657 | dp += BBSIZE; | |
1658 | } | |
1659 | ||
1660 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1661 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1662 | ||
1663 | for ( ; i < BTOBB(size); i++) { | |
1664 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1665 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1666 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1667 | *(__be32 *)dp = cycle_lsn; | |
1668 | dp += BBSIZE; | |
1669 | } | |
1670 | ||
1671 | for (i = 1; i < log->l_iclog_heads; i++) | |
1672 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1673 | } | |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * Calculate the checksum for a log buffer. | |
1678 | * | |
1679 | * This is a little more complicated than it should be because the various | |
1680 | * headers and the actual data are non-contiguous. | |
1681 | */ | |
f9668a09 | 1682 | __le32 |
0e446be4 CH |
1683 | xlog_cksum( |
1684 | struct xlog *log, | |
1685 | struct xlog_rec_header *rhead, | |
1686 | char *dp, | |
1687 | int size) | |
1688 | { | |
c8ce540d | 1689 | uint32_t crc; |
0e446be4 CH |
1690 | |
1691 | /* first generate the crc for the record header ... */ | |
cae028df | 1692 | crc = xfs_start_cksum_update((char *)rhead, |
0e446be4 CH |
1693 | sizeof(struct xlog_rec_header), |
1694 | offsetof(struct xlog_rec_header, h_crc)); | |
1695 | ||
1696 | /* ... then for additional cycle data for v2 logs ... */ | |
1697 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1698 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1699 | int i; | |
a3f20014 | 1700 | int xheads; |
0e446be4 | 1701 | |
0c771b99 | 1702 | xheads = DIV_ROUND_UP(size, XLOG_HEADER_CYCLE_SIZE); |
0e446be4 | 1703 | |
a3f20014 | 1704 | for (i = 1; i < xheads; i++) { |
0e446be4 CH |
1705 | crc = crc32c(crc, &xhdr[i].hic_xheader, |
1706 | sizeof(struct xlog_rec_ext_header)); | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | /* ... and finally for the payload */ | |
1711 | crc = crc32c(crc, dp, size); | |
1712 | ||
1713 | return xfs_end_cksum(crc); | |
1714 | } | |
1715 | ||
79b54d9b CH |
1716 | static void |
1717 | xlog_bio_end_io( | |
1718 | struct bio *bio) | |
1719 | { | |
1720 | struct xlog_in_core *iclog = bio->bi_private; | |
1721 | ||
1058d0f5 | 1722 | queue_work(iclog->ic_log->l_ioend_workqueue, |
79b54d9b CH |
1723 | &iclog->ic_end_io_work); |
1724 | } | |
1725 | ||
842a42d1 | 1726 | static int |
79b54d9b CH |
1727 | xlog_map_iclog_data( |
1728 | struct bio *bio, | |
1729 | void *data, | |
1730 | size_t count) | |
1731 | { | |
1732 | do { | |
1733 | struct page *page = kmem_to_page(data); | |
1734 | unsigned int off = offset_in_page(data); | |
1735 | size_t len = min_t(size_t, count, PAGE_SIZE - off); | |
1736 | ||
842a42d1 BF |
1737 | if (bio_add_page(bio, page, len, off) != len) |
1738 | return -EIO; | |
79b54d9b CH |
1739 | |
1740 | data += len; | |
1741 | count -= len; | |
1742 | } while (count); | |
842a42d1 BF |
1743 | |
1744 | return 0; | |
79b54d9b CH |
1745 | } |
1746 | ||
94860a30 CH |
1747 | STATIC void |
1748 | xlog_write_iclog( | |
1749 | struct xlog *log, | |
1750 | struct xlog_in_core *iclog, | |
94860a30 | 1751 | uint64_t bno, |
eef983ff | 1752 | unsigned int count) |
873ff550 | 1753 | { |
94860a30 | 1754 | ASSERT(bno < log->l_logBBsize); |
956f6daa | 1755 | trace_xlog_iclog_write(iclog, _RET_IP_); |
94860a30 CH |
1756 | |
1757 | /* | |
1758 | * We lock the iclogbufs here so that we can serialise against I/O | |
1759 | * completion during unmount. We might be processing a shutdown | |
1760 | * triggered during unmount, and that can occur asynchronously to the | |
1761 | * unmount thread, and hence we need to ensure that completes before | |
1762 | * tearing down the iclogbufs. Hence we need to hold the buffer lock | |
1763 | * across the log IO to archieve that. | |
1764 | */ | |
79b54d9b | 1765 | down(&iclog->ic_sema); |
1858bb0b | 1766 | if (unlikely(iclog->ic_state == XLOG_STATE_IOERROR)) { |
873ff550 CH |
1767 | /* |
1768 | * It would seem logical to return EIO here, but we rely on | |
1769 | * the log state machine to propagate I/O errors instead of | |
79b54d9b CH |
1770 | * doing it here. We kick of the state machine and unlock |
1771 | * the buffer manually, the code needs to be kept in sync | |
1772 | * with the I/O completion path. | |
873ff550 | 1773 | */ |
12e6a0f4 | 1774 | xlog_state_done_syncing(iclog); |
79b54d9b | 1775 | up(&iclog->ic_sema); |
94860a30 | 1776 | return; |
873ff550 CH |
1777 | } |
1778 | ||
79b54d9b CH |
1779 | bio_init(&iclog->ic_bio, iclog->ic_bvec, howmany(count, PAGE_SIZE)); |
1780 | bio_set_dev(&iclog->ic_bio, log->l_targ->bt_bdev); | |
1781 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno; | |
1782 | iclog->ic_bio.bi_end_io = xlog_bio_end_io; | |
1783 | iclog->ic_bio.bi_private = iclog; | |
2def2845 DC |
1784 | |
1785 | /* | |
1786 | * We use REQ_SYNC | REQ_IDLE here to tell the block layer the are more | |
1787 | * IOs coming immediately after this one. This prevents the block layer | |
1788 | * writeback throttle from throttling log writes behind background | |
1789 | * metadata writeback and causing priority inversions. | |
1790 | */ | |
eef983ff | 1791 | iclog->ic_bio.bi_opf = REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_IDLE; |
b5d721ea | 1792 | if (iclog->ic_flags & XLOG_ICL_NEED_FLUSH) { |
79b54d9b | 1793 | iclog->ic_bio.bi_opf |= REQ_PREFLUSH; |
b5d721ea DC |
1794 | /* |
1795 | * For external log devices, we also need to flush the data | |
1796 | * device cache first to ensure all metadata writeback covered | |
1797 | * by the LSN in this iclog is on stable storage. This is slow, | |
1798 | * but it *must* complete before we issue the external log IO. | |
1799 | */ | |
1800 | if (log->l_targ != log->l_mp->m_ddev_targp) | |
1801 | blkdev_issue_flush(log->l_mp->m_ddev_targp->bt_bdev); | |
1802 | } | |
eef983ff DC |
1803 | if (iclog->ic_flags & XLOG_ICL_NEED_FUA) |
1804 | iclog->ic_bio.bi_opf |= REQ_FUA; | |
b5d721ea | 1805 | |
eef983ff | 1806 | iclog->ic_flags &= ~(XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA); |
79b54d9b | 1807 | |
842a42d1 BF |
1808 | if (xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, count)) { |
1809 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1810 | return; | |
1811 | } | |
79b54d9b | 1812 | if (is_vmalloc_addr(iclog->ic_data)) |
2c68a1df | 1813 | flush_kernel_vmap_range(iclog->ic_data, count); |
79b54d9b CH |
1814 | |
1815 | /* | |
1816 | * If this log buffer would straddle the end of the log we will have | |
1817 | * to split it up into two bios, so that we can continue at the start. | |
1818 | */ | |
1819 | if (bno + BTOBB(count) > log->l_logBBsize) { | |
1820 | struct bio *split; | |
1821 | ||
1822 | split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno, | |
1823 | GFP_NOIO, &fs_bio_set); | |
1824 | bio_chain(split, &iclog->ic_bio); | |
1825 | submit_bio(split); | |
1826 | ||
1827 | /* restart at logical offset zero for the remainder */ | |
1828 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart; | |
1829 | } | |
1830 | ||
1831 | submit_bio(&iclog->ic_bio); | |
873ff550 | 1832 | } |
1da177e4 | 1833 | |
56933848 CH |
1834 | /* |
1835 | * We need to bump cycle number for the part of the iclog that is | |
1836 | * written to the start of the log. Watch out for the header magic | |
1837 | * number case, though. | |
1838 | */ | |
79b54d9b | 1839 | static void |
56933848 CH |
1840 | xlog_split_iclog( |
1841 | struct xlog *log, | |
1842 | void *data, | |
1843 | uint64_t bno, | |
1844 | unsigned int count) | |
1845 | { | |
1846 | unsigned int split_offset = BBTOB(log->l_logBBsize - bno); | |
1847 | unsigned int i; | |
1848 | ||
1849 | for (i = split_offset; i < count; i += BBSIZE) { | |
1850 | uint32_t cycle = get_unaligned_be32(data + i); | |
1851 | ||
1852 | if (++cycle == XLOG_HEADER_MAGIC_NUM) | |
1853 | cycle++; | |
1854 | put_unaligned_be32(cycle, data + i); | |
1855 | } | |
56933848 CH |
1856 | } |
1857 | ||
db0a6faf CH |
1858 | static int |
1859 | xlog_calc_iclog_size( | |
1860 | struct xlog *log, | |
1861 | struct xlog_in_core *iclog, | |
1862 | uint32_t *roundoff) | |
1863 | { | |
1864 | uint32_t count_init, count; | |
db0a6faf CH |
1865 | |
1866 | /* Add for LR header */ | |
1867 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
a6a65fef | 1868 | count = roundup(count_init, log->l_iclog_roundoff); |
db0a6faf | 1869 | |
db0a6faf CH |
1870 | *roundoff = count - count_init; |
1871 | ||
a6a65fef DC |
1872 | ASSERT(count >= count_init); |
1873 | ASSERT(*roundoff < log->l_iclog_roundoff); | |
db0a6faf CH |
1874 | return count; |
1875 | } | |
1876 | ||
1da177e4 LT |
1877 | /* |
1878 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1879 | * fashion. Previously, we should have moved the current iclog | |
1880 | * ptr in the log to point to the next available iclog. This allows further | |
1881 | * write to continue while this code syncs out an iclog ready to go. | |
1882 | * Before an in-core log can be written out, the data section must be scanned | |
1883 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1884 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1885 | * cycle count because there in an implicit assumption that drives will | |
1886 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1887 | * we can't have part of a 512 byte block written and part not written. By | |
1888 | * tagging each block, we will know which blocks are valid when recovering | |
1889 | * after an unclean shutdown. | |
1890 | * | |
1891 | * This routine is single threaded on the iclog. No other thread can be in | |
1892 | * this routine with the same iclog. Changing contents of iclog can there- | |
1893 | * fore be done without grabbing the state machine lock. Updating the global | |
1894 | * log will require grabbing the lock though. | |
1895 | * | |
1896 | * The entire log manager uses a logical block numbering scheme. Only | |
94860a30 CH |
1897 | * xlog_write_iclog knows about the fact that the log may not start with |
1898 | * block zero on a given device. | |
1da177e4 | 1899 | */ |
94860a30 | 1900 | STATIC void |
9a8d2fdb MT |
1901 | xlog_sync( |
1902 | struct xlog *log, | |
1903 | struct xlog_in_core *iclog) | |
1da177e4 | 1904 | { |
db0a6faf CH |
1905 | unsigned int count; /* byte count of bwrite */ |
1906 | unsigned int roundoff; /* roundoff to BB or stripe */ | |
1907 | uint64_t bno; | |
db0a6faf | 1908 | unsigned int size; |
1da177e4 | 1909 | |
155cc6b7 | 1910 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
956f6daa | 1911 | trace_xlog_iclog_sync(iclog, _RET_IP_); |
1da177e4 | 1912 | |
db0a6faf | 1913 | count = xlog_calc_iclog_size(log, iclog, &roundoff); |
1da177e4 LT |
1914 | |
1915 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1916 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1917 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1918 | |
1919 | /* put cycle number in every block */ | |
1920 | xlog_pack_data(log, iclog, roundoff); | |
1921 | ||
1922 | /* real byte length */ | |
0e446be4 | 1923 | size = iclog->ic_offset; |
db0a6faf | 1924 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) |
0e446be4 CH |
1925 | size += roundoff; |
1926 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 1927 | |
9b0489c1 | 1928 | XFS_STATS_INC(log->l_mp, xs_log_writes); |
ff6d6af2 | 1929 | XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count)); |
1da177e4 | 1930 | |
94860a30 CH |
1931 | bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)); |
1932 | ||
1da177e4 | 1933 | /* Do we need to split this write into 2 parts? */ |
eef983ff | 1934 | if (bno + BTOBB(count) > log->l_logBBsize) |
79b54d9b | 1935 | xlog_split_iclog(log, &iclog->ic_header, bno, count); |
0e446be4 CH |
1936 | |
1937 | /* calculcate the checksum */ | |
1938 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
1939 | iclog->ic_datap, size); | |
609adfc2 BF |
1940 | /* |
1941 | * Intentionally corrupt the log record CRC based on the error injection | |
1942 | * frequency, if defined. This facilitates testing log recovery in the | |
1943 | * event of torn writes. Hence, set the IOABORT state to abort the log | |
1944 | * write on I/O completion and shutdown the fs. The subsequent mount | |
1945 | * detects the bad CRC and attempts to recover. | |
1946 | */ | |
366fc4b8 | 1947 | #ifdef DEBUG |
3e88a007 | 1948 | if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) { |
e2a64192 | 1949 | iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA); |
366fc4b8 | 1950 | iclog->ic_fail_crc = true; |
609adfc2 BF |
1951 | xfs_warn(log->l_mp, |
1952 | "Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.", | |
1953 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1954 | } | |
366fc4b8 | 1955 | #endif |
abca1f33 | 1956 | xlog_verify_iclog(log, iclog, count); |
eef983ff | 1957 | xlog_write_iclog(log, iclog, bno, count); |
94860a30 | 1958 | } |
1da177e4 | 1959 | |
1da177e4 | 1960 | /* |
c41564b5 | 1961 | * Deallocate a log structure |
1da177e4 | 1962 | */ |
a8272ce0 | 1963 | STATIC void |
9a8d2fdb MT |
1964 | xlog_dealloc_log( |
1965 | struct xlog *log) | |
1da177e4 LT |
1966 | { |
1967 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1968 | int i; |
1969 | ||
71e330b5 DC |
1970 | xlog_cil_destroy(log); |
1971 | ||
44396476 | 1972 | /* |
9c23eccc DC |
1973 | * Cycle all the iclogbuf locks to make sure all log IO completion |
1974 | * is done before we tear down these buffers. | |
1975 | */ | |
1976 | iclog = log->l_iclog; | |
1977 | for (i = 0; i < log->l_iclog_bufs; i++) { | |
79b54d9b CH |
1978 | down(&iclog->ic_sema); |
1979 | up(&iclog->ic_sema); | |
9c23eccc DC |
1980 | iclog = iclog->ic_next; |
1981 | } | |
1982 | ||
1da177e4 | 1983 | iclog = log->l_iclog; |
9c23eccc | 1984 | for (i = 0; i < log->l_iclog_bufs; i++) { |
1da177e4 | 1985 | next_iclog = iclog->ic_next; |
79b54d9b | 1986 | kmem_free(iclog->ic_data); |
f0e2d93c | 1987 | kmem_free(iclog); |
1da177e4 LT |
1988 | iclog = next_iclog; |
1989 | } | |
1da177e4 | 1990 | |
1da177e4 | 1991 | log->l_mp->m_log = NULL; |
1058d0f5 | 1992 | destroy_workqueue(log->l_ioend_workqueue); |
f0e2d93c | 1993 | kmem_free(log); |
b843299b | 1994 | } |
1da177e4 LT |
1995 | |
1996 | /* | |
1997 | * Update counters atomically now that memcpy is done. | |
1998 | */ | |
1da177e4 | 1999 | static inline void |
9a8d2fdb MT |
2000 | xlog_state_finish_copy( |
2001 | struct xlog *log, | |
2002 | struct xlog_in_core *iclog, | |
2003 | int record_cnt, | |
2004 | int copy_bytes) | |
1da177e4 | 2005 | { |
390aab0a | 2006 | lockdep_assert_held(&log->l_icloglock); |
1da177e4 | 2007 | |
413d57c9 | 2008 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 | 2009 | iclog->ic_offset += copy_bytes; |
390aab0a | 2010 | } |
1da177e4 | 2011 | |
7e9c6396 TS |
2012 | /* |
2013 | * print out info relating to regions written which consume | |
2014 | * the reservation | |
2015 | */ | |
71e330b5 DC |
2016 | void |
2017 | xlog_print_tic_res( | |
2018 | struct xfs_mount *mp, | |
2019 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
2020 | { |
2021 | uint i; | |
2022 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
2023 | ||
2024 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
5110cd82 | 2025 | #define REG_TYPE_STR(type, str) [XLOG_REG_TYPE_##type] = str |
d31d7185 | 2026 | static char *res_type_str[] = { |
5110cd82 DW |
2027 | REG_TYPE_STR(BFORMAT, "bformat"), |
2028 | REG_TYPE_STR(BCHUNK, "bchunk"), | |
2029 | REG_TYPE_STR(EFI_FORMAT, "efi_format"), | |
2030 | REG_TYPE_STR(EFD_FORMAT, "efd_format"), | |
2031 | REG_TYPE_STR(IFORMAT, "iformat"), | |
2032 | REG_TYPE_STR(ICORE, "icore"), | |
2033 | REG_TYPE_STR(IEXT, "iext"), | |
2034 | REG_TYPE_STR(IBROOT, "ibroot"), | |
2035 | REG_TYPE_STR(ILOCAL, "ilocal"), | |
2036 | REG_TYPE_STR(IATTR_EXT, "iattr_ext"), | |
2037 | REG_TYPE_STR(IATTR_BROOT, "iattr_broot"), | |
2038 | REG_TYPE_STR(IATTR_LOCAL, "iattr_local"), | |
2039 | REG_TYPE_STR(QFORMAT, "qformat"), | |
2040 | REG_TYPE_STR(DQUOT, "dquot"), | |
2041 | REG_TYPE_STR(QUOTAOFF, "quotaoff"), | |
2042 | REG_TYPE_STR(LRHEADER, "LR header"), | |
2043 | REG_TYPE_STR(UNMOUNT, "unmount"), | |
2044 | REG_TYPE_STR(COMMIT, "commit"), | |
2045 | REG_TYPE_STR(TRANSHDR, "trans header"), | |
d31d7185 DW |
2046 | REG_TYPE_STR(ICREATE, "inode create"), |
2047 | REG_TYPE_STR(RUI_FORMAT, "rui_format"), | |
2048 | REG_TYPE_STR(RUD_FORMAT, "rud_format"), | |
2049 | REG_TYPE_STR(CUI_FORMAT, "cui_format"), | |
2050 | REG_TYPE_STR(CUD_FORMAT, "cud_format"), | |
2051 | REG_TYPE_STR(BUI_FORMAT, "bui_format"), | |
2052 | REG_TYPE_STR(BUD_FORMAT, "bud_format"), | |
7e9c6396 | 2053 | }; |
d31d7185 | 2054 | BUILD_BUG_ON(ARRAY_SIZE(res_type_str) != XLOG_REG_TYPE_MAX + 1); |
5110cd82 | 2055 | #undef REG_TYPE_STR |
7e9c6396 | 2056 | |
7d2d5653 | 2057 | xfs_warn(mp, "ticket reservation summary:"); |
f41febd2 JP |
2058 | xfs_warn(mp, " unit res = %d bytes", |
2059 | ticket->t_unit_res); | |
2060 | xfs_warn(mp, " current res = %d bytes", | |
2061 | ticket->t_curr_res); | |
2062 | xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)", | |
2063 | ticket->t_res_arr_sum, ticket->t_res_o_flow); | |
2064 | xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)", | |
2065 | ticket->t_res_num_ophdrs, ophdr_spc); | |
2066 | xfs_warn(mp, " ophdr + reg = %u bytes", | |
2067 | ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc); | |
2068 | xfs_warn(mp, " num regions = %u", | |
2069 | ticket->t_res_num); | |
7e9c6396 TS |
2070 | |
2071 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 2072 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 2073 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 | 2074 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
5110cd82 | 2075 | "bad-rtype" : res_type_str[r_type]), |
7e9c6396 TS |
2076 | ticket->t_res_arr[i].r_len); |
2077 | } | |
2078 | } | |
7e9c6396 | 2079 | |
d4ca1d55 BF |
2080 | /* |
2081 | * Print a summary of the transaction. | |
2082 | */ | |
2083 | void | |
2084 | xlog_print_trans( | |
e6631f85 | 2085 | struct xfs_trans *tp) |
d4ca1d55 | 2086 | { |
e6631f85 DC |
2087 | struct xfs_mount *mp = tp->t_mountp; |
2088 | struct xfs_log_item *lip; | |
d4ca1d55 BF |
2089 | |
2090 | /* dump core transaction and ticket info */ | |
2091 | xfs_warn(mp, "transaction summary:"); | |
2c8f6265 BF |
2092 | xfs_warn(mp, " log res = %d", tp->t_log_res); |
2093 | xfs_warn(mp, " log count = %d", tp->t_log_count); | |
2094 | xfs_warn(mp, " flags = 0x%x", tp->t_flags); | |
d4ca1d55 BF |
2095 | |
2096 | xlog_print_tic_res(mp, tp->t_ticket); | |
2097 | ||
2098 | /* dump each log item */ | |
e6631f85 | 2099 | list_for_each_entry(lip, &tp->t_items, li_trans) { |
d4ca1d55 BF |
2100 | struct xfs_log_vec *lv = lip->li_lv; |
2101 | struct xfs_log_iovec *vec; | |
2102 | int i; | |
2103 | ||
2104 | xfs_warn(mp, "log item: "); | |
2105 | xfs_warn(mp, " type = 0x%x", lip->li_type); | |
22525c17 | 2106 | xfs_warn(mp, " flags = 0x%lx", lip->li_flags); |
d4ca1d55 BF |
2107 | if (!lv) |
2108 | continue; | |
2109 | xfs_warn(mp, " niovecs = %d", lv->lv_niovecs); | |
2110 | xfs_warn(mp, " size = %d", lv->lv_size); | |
2111 | xfs_warn(mp, " bytes = %d", lv->lv_bytes); | |
2112 | xfs_warn(mp, " buf len = %d", lv->lv_buf_len); | |
2113 | ||
2114 | /* dump each iovec for the log item */ | |
2115 | vec = lv->lv_iovecp; | |
2116 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2117 | int dumplen = min(vec->i_len, 32); | |
2118 | ||
2119 | xfs_warn(mp, " iovec[%d]", i); | |
2120 | xfs_warn(mp, " type = 0x%x", vec->i_type); | |
2121 | xfs_warn(mp, " len = %d", vec->i_len); | |
2122 | xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i); | |
244e3dea | 2123 | xfs_hex_dump(vec->i_addr, dumplen); |
d4ca1d55 BF |
2124 | |
2125 | vec++; | |
2126 | } | |
2127 | } | |
2128 | } | |
2129 | ||
b5203cd0 | 2130 | /* |
7ec94921 DC |
2131 | * Calculate the potential space needed by the log vector. We may need a start |
2132 | * record, and each region gets its own struct xlog_op_header and may need to be | |
2133 | * double word aligned. | |
b5203cd0 DC |
2134 | */ |
2135 | static int | |
2136 | xlog_write_calc_vec_length( | |
2137 | struct xlog_ticket *ticket, | |
7ec94921 | 2138 | struct xfs_log_vec *log_vector, |
3468bb1c | 2139 | uint optype) |
b5203cd0 | 2140 | { |
55b66332 | 2141 | struct xfs_log_vec *lv; |
3468bb1c | 2142 | int headers = 0; |
b5203cd0 DC |
2143 | int len = 0; |
2144 | int i; | |
2145 | ||
3468bb1c DC |
2146 | if (optype & XLOG_START_TRANS) |
2147 | headers++; | |
2148 | ||
55b66332 | 2149 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2150 | /* we don't write ordered log vectors */ |
2151 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2152 | continue; | |
2153 | ||
55b66332 DC |
2154 | headers += lv->lv_niovecs; |
2155 | ||
2156 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2157 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2158 | |
55b66332 DC |
2159 | len += vecp->i_len; |
2160 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2161 | } | |
b5203cd0 DC |
2162 | } |
2163 | ||
2164 | ticket->t_res_num_ophdrs += headers; | |
2165 | len += headers * sizeof(struct xlog_op_header); | |
2166 | ||
2167 | return len; | |
2168 | } | |
2169 | ||
7ec94921 | 2170 | static void |
b5203cd0 | 2171 | xlog_write_start_rec( |
e6b1f273 | 2172 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2173 | struct xlog_ticket *ticket) |
2174 | { | |
b5203cd0 DC |
2175 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2176 | ophdr->oh_clientid = ticket->t_clientid; | |
2177 | ophdr->oh_len = 0; | |
2178 | ophdr->oh_flags = XLOG_START_TRANS; | |
2179 | ophdr->oh_res2 = 0; | |
b5203cd0 DC |
2180 | } |
2181 | ||
2182 | static xlog_op_header_t * | |
2183 | xlog_write_setup_ophdr( | |
ad223e60 | 2184 | struct xlog *log, |
e6b1f273 | 2185 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2186 | struct xlog_ticket *ticket, |
2187 | uint flags) | |
2188 | { | |
b5203cd0 DC |
2189 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2190 | ophdr->oh_clientid = ticket->t_clientid; | |
2191 | ophdr->oh_res2 = 0; | |
2192 | ||
2193 | /* are we copying a commit or unmount record? */ | |
2194 | ophdr->oh_flags = flags; | |
2195 | ||
2196 | /* | |
2197 | * We've seen logs corrupted with bad transaction client ids. This | |
2198 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2199 | * and shut down the filesystem. | |
2200 | */ | |
2201 | switch (ophdr->oh_clientid) { | |
2202 | case XFS_TRANSACTION: | |
2203 | case XFS_VOLUME: | |
2204 | case XFS_LOG: | |
2205 | break; | |
2206 | default: | |
a0fa2b67 | 2207 | xfs_warn(log->l_mp, |
c9690043 | 2208 | "Bad XFS transaction clientid 0x%x in ticket "PTR_FMT, |
b5203cd0 DC |
2209 | ophdr->oh_clientid, ticket); |
2210 | return NULL; | |
2211 | } | |
2212 | ||
2213 | return ophdr; | |
2214 | } | |
2215 | ||
2216 | /* | |
2217 | * Set up the parameters of the region copy into the log. This has | |
2218 | * to handle region write split across multiple log buffers - this | |
2219 | * state is kept external to this function so that this code can | |
ac0e300f | 2220 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2221 | */ |
2222 | static int | |
2223 | xlog_write_setup_copy( | |
2224 | struct xlog_ticket *ticket, | |
2225 | struct xlog_op_header *ophdr, | |
2226 | int space_available, | |
2227 | int space_required, | |
2228 | int *copy_off, | |
2229 | int *copy_len, | |
2230 | int *last_was_partial_copy, | |
2231 | int *bytes_consumed) | |
2232 | { | |
2233 | int still_to_copy; | |
2234 | ||
2235 | still_to_copy = space_required - *bytes_consumed; | |
2236 | *copy_off = *bytes_consumed; | |
2237 | ||
2238 | if (still_to_copy <= space_available) { | |
2239 | /* write of region completes here */ | |
2240 | *copy_len = still_to_copy; | |
2241 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2242 | if (*last_was_partial_copy) | |
2243 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2244 | *last_was_partial_copy = 0; | |
2245 | *bytes_consumed = 0; | |
2246 | return 0; | |
2247 | } | |
2248 | ||
2249 | /* partial write of region, needs extra log op header reservation */ | |
2250 | *copy_len = space_available; | |
2251 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2252 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2253 | if (*last_was_partial_copy) | |
2254 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2255 | *bytes_consumed += *copy_len; | |
2256 | (*last_was_partial_copy)++; | |
2257 | ||
2258 | /* account for new log op header */ | |
2259 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2260 | ticket->t_res_num_ophdrs++; | |
2261 | ||
2262 | return sizeof(struct xlog_op_header); | |
2263 | } | |
2264 | ||
2265 | static int | |
2266 | xlog_write_copy_finish( | |
ad223e60 | 2267 | struct xlog *log, |
b5203cd0 DC |
2268 | struct xlog_in_core *iclog, |
2269 | uint flags, | |
2270 | int *record_cnt, | |
2271 | int *data_cnt, | |
2272 | int *partial_copy, | |
2273 | int *partial_copy_len, | |
2274 | int log_offset, | |
2275 | struct xlog_in_core **commit_iclog) | |
2276 | { | |
df732b29 CH |
2277 | int error; |
2278 | ||
b5203cd0 DC |
2279 | if (*partial_copy) { |
2280 | /* | |
2281 | * This iclog has already been marked WANT_SYNC by | |
2282 | * xlog_state_get_iclog_space. | |
2283 | */ | |
390aab0a | 2284 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2285 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2286 | *record_cnt = 0; | |
2287 | *data_cnt = 0; | |
df732b29 | 2288 | goto release_iclog; |
b5203cd0 DC |
2289 | } |
2290 | ||
2291 | *partial_copy = 0; | |
2292 | *partial_copy_len = 0; | |
2293 | ||
2294 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2295 | /* no more space in this iclog - push it. */ | |
390aab0a | 2296 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2297 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2298 | *record_cnt = 0; | |
2299 | *data_cnt = 0; | |
2300 | ||
69363999 CH |
2301 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
2302 | xlog_state_switch_iclogs(log, iclog, 0); | |
2303 | else | |
2304 | ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
2305 | iclog->ic_state == XLOG_STATE_IOERROR); | |
b5203cd0 | 2306 | if (!commit_iclog) |
df732b29 CH |
2307 | goto release_iclog; |
2308 | spin_unlock(&log->l_icloglock); | |
b5203cd0 DC |
2309 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2310 | *commit_iclog = iclog; | |
2311 | } | |
2312 | ||
2313 | return 0; | |
df732b29 CH |
2314 | |
2315 | release_iclog: | |
2316 | error = xlog_state_release_iclog(log, iclog); | |
2317 | spin_unlock(&log->l_icloglock); | |
2318 | return error; | |
b5203cd0 DC |
2319 | } |
2320 | ||
1da177e4 LT |
2321 | /* |
2322 | * Write some region out to in-core log | |
2323 | * | |
2324 | * This will be called when writing externally provided regions or when | |
2325 | * writing out a commit record for a given transaction. | |
2326 | * | |
2327 | * General algorithm: | |
2328 | * 1. Find total length of this write. This may include adding to the | |
2329 | * lengths passed in. | |
2330 | * 2. Check whether we violate the tickets reservation. | |
2331 | * 3. While writing to this iclog | |
2332 | * A. Reserve as much space in this iclog as can get | |
2333 | * B. If this is first write, save away start lsn | |
2334 | * C. While writing this region: | |
2335 | * 1. If first write of transaction, write start record | |
2336 | * 2. Write log operation header (header per region) | |
2337 | * 3. Find out if we can fit entire region into this iclog | |
2338 | * 4. Potentially, verify destination memcpy ptr | |
2339 | * 5. Memcpy (partial) region | |
2340 | * 6. If partial copy, release iclog; otherwise, continue | |
2341 | * copying more regions into current iclog | |
2342 | * 4. Mark want sync bit (in simulation mode) | |
2343 | * 5. Release iclog for potential flush to on-disk log. | |
2344 | * | |
2345 | * ERRORS: | |
2346 | * 1. Panic if reservation is overrun. This should never happen since | |
2347 | * reservation amounts are generated internal to the filesystem. | |
2348 | * NOTES: | |
2349 | * 1. Tickets are single threaded data structures. | |
2350 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2351 | * syncing routine. When a single log_write region needs to span | |
2352 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2353 | * on all log operation writes which don't contain the end of the | |
2354 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2355 | * operation which contains the end of the continued log_write region. | |
2356 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2357 | * we don't really know exactly how much space will be used. As a result, | |
2358 | * we don't update ic_offset until the end when we know exactly how many | |
2359 | * bytes have been written out. | |
2360 | */ | |
71e330b5 | 2361 | int |
35a8a72f | 2362 | xlog_write( |
ad223e60 | 2363 | struct xlog *log, |
55b66332 | 2364 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2365 | struct xlog_ticket *ticket, |
2366 | xfs_lsn_t *start_lsn, | |
2367 | struct xlog_in_core **commit_iclog, | |
3468bb1c | 2368 | uint optype) |
1da177e4 | 2369 | { |
99428ad0 | 2370 | struct xlog_in_core *iclog = NULL; |
9590e9c6 DC |
2371 | struct xfs_log_vec *lv = log_vector; |
2372 | struct xfs_log_iovec *vecp = lv->lv_iovecp; | |
2373 | int index = 0; | |
99428ad0 | 2374 | int len; |
99428ad0 CH |
2375 | int partial_copy = 0; |
2376 | int partial_copy_len = 0; | |
2377 | int contwr = 0; | |
2378 | int record_cnt = 0; | |
2379 | int data_cnt = 0; | |
df732b29 | 2380 | int error = 0; |
99428ad0 | 2381 | |
93b8a585 | 2382 | /* |
9590e9c6 DC |
2383 | * If this is a commit or unmount transaction, we don't need a start |
2384 | * record to be written. We do, however, have to account for the | |
2385 | * commit or unmount header that gets written. Hence we always have | |
2386 | * to account for an extra xlog_op_header here. | |
93b8a585 | 2387 | */ |
9590e9c6 | 2388 | ticket->t_curr_res -= sizeof(struct xlog_op_header); |
7d2d5653 BF |
2389 | if (ticket->t_curr_res < 0) { |
2390 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
2391 | "ctx ticket reservation ran out. Need to up reservation"); | |
55b66332 | 2392 | xlog_print_tic_res(log->l_mp, ticket); |
7d2d5653 BF |
2393 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
2394 | } | |
1da177e4 | 2395 | |
3468bb1c DC |
2396 | len = xlog_write_calc_vec_length(ticket, log_vector, optype); |
2397 | if (start_lsn) | |
2398 | *start_lsn = 0; | |
fd63875c | 2399 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2400 | void *ptr; |
99428ad0 | 2401 | int log_offset; |
1da177e4 | 2402 | |
99428ad0 CH |
2403 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2404 | &contwr, &log_offset); | |
2405 | if (error) | |
2406 | return error; | |
1da177e4 | 2407 | |
99428ad0 | 2408 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2409 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2410 | |
eef983ff | 2411 | /* Start_lsn is the first lsn written to. */ |
3468bb1c | 2412 | if (start_lsn && !*start_lsn) |
99428ad0 | 2413 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
b5203cd0 | 2414 | |
99428ad0 CH |
2415 | /* |
2416 | * This loop writes out as many regions as can fit in the amount | |
2417 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2418 | */ | |
fd63875c DC |
2419 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2420 | struct xfs_log_iovec *reg; | |
99428ad0 | 2421 | struct xlog_op_header *ophdr; |
99428ad0 CH |
2422 | int copy_len; |
2423 | int copy_off; | |
fd63875c | 2424 | bool ordered = false; |
3468bb1c | 2425 | bool wrote_start_rec = false; |
fd63875c DC |
2426 | |
2427 | /* ordered log vectors have no regions to write */ | |
2428 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2429 | ASSERT(lv->lv_niovecs == 0); | |
2430 | ordered = true; | |
2431 | goto next_lv; | |
2432 | } | |
99428ad0 | 2433 | |
fd63875c | 2434 | reg = &vecp[index]; |
c8ce540d DW |
2435 | ASSERT(reg->i_len % sizeof(int32_t) == 0); |
2436 | ASSERT((unsigned long)ptr % sizeof(int32_t) == 0); | |
99428ad0 | 2437 | |
7ec94921 DC |
2438 | /* |
2439 | * Before we start formatting log vectors, we need to | |
2440 | * write a start record. Only do this for the first | |
2441 | * iclog we write to. | |
2442 | */ | |
3468bb1c | 2443 | if (optype & XLOG_START_TRANS) { |
7ec94921 | 2444 | xlog_write_start_rec(ptr, ticket); |
e6b1f273 | 2445 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
7ec94921 | 2446 | sizeof(struct xlog_op_header)); |
3468bb1c DC |
2447 | optype &= ~XLOG_START_TRANS; |
2448 | wrote_start_rec = true; | |
99428ad0 | 2449 | } |
b5203cd0 | 2450 | |
3468bb1c | 2451 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, optype); |
99428ad0 | 2452 | if (!ophdr) |
2451337d | 2453 | return -EIO; |
99428ad0 | 2454 | |
e6b1f273 | 2455 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2456 | sizeof(struct xlog_op_header)); |
2457 | ||
2458 | len += xlog_write_setup_copy(ticket, ophdr, | |
2459 | iclog->ic_size-log_offset, | |
55b66332 | 2460 | reg->i_len, |
99428ad0 CH |
2461 | ©_off, ©_len, |
2462 | &partial_copy, | |
2463 | &partial_copy_len); | |
2464 | xlog_verify_dest_ptr(log, ptr); | |
2465 | ||
91f9f5fe ES |
2466 | /* |
2467 | * Copy region. | |
2468 | * | |
2469 | * Unmount records just log an opheader, so can have | |
2470 | * empty payloads with no data region to copy. Hence we | |
2471 | * only copy the payload if the vector says it has data | |
2472 | * to copy. | |
2473 | */ | |
99428ad0 | 2474 | ASSERT(copy_len >= 0); |
91f9f5fe ES |
2475 | if (copy_len > 0) { |
2476 | memcpy(ptr, reg->i_addr + copy_off, copy_len); | |
2477 | xlog_write_adv_cnt(&ptr, &len, &log_offset, | |
2478 | copy_len); | |
2479 | } | |
7ec94921 | 2480 | copy_len += sizeof(struct xlog_op_header); |
99428ad0 | 2481 | record_cnt++; |
3468bb1c | 2482 | if (wrote_start_rec) { |
7ec94921 DC |
2483 | copy_len += sizeof(struct xlog_op_header); |
2484 | record_cnt++; | |
7ec94921 | 2485 | } |
99428ad0 CH |
2486 | data_cnt += contwr ? copy_len : 0; |
2487 | ||
3468bb1c | 2488 | error = xlog_write_copy_finish(log, iclog, optype, |
99428ad0 CH |
2489 | &record_cnt, &data_cnt, |
2490 | &partial_copy, | |
2491 | &partial_copy_len, | |
2492 | log_offset, | |
2493 | commit_iclog); | |
2494 | if (error) | |
2495 | return error; | |
2496 | ||
2497 | /* | |
2498 | * if we had a partial copy, we need to get more iclog | |
2499 | * space but we don't want to increment the region | |
2500 | * index because there is still more is this region to | |
2501 | * write. | |
2502 | * | |
2503 | * If we completed writing this region, and we flushed | |
2504 | * the iclog (indicated by resetting of the record | |
2505 | * count), then we also need to get more log space. If | |
2506 | * this was the last record, though, we are done and | |
2507 | * can just return. | |
2508 | */ | |
2509 | if (partial_copy) | |
2510 | break; | |
2511 | ||
55b66332 | 2512 | if (++index == lv->lv_niovecs) { |
fd63875c | 2513 | next_lv: |
55b66332 DC |
2514 | lv = lv->lv_next; |
2515 | index = 0; | |
2516 | if (lv) | |
2517 | vecp = lv->lv_iovecp; | |
2518 | } | |
749f24f3 | 2519 | if (record_cnt == 0 && !ordered) { |
55b66332 | 2520 | if (!lv) |
99428ad0 CH |
2521 | return 0; |
2522 | break; | |
2523 | } | |
2524 | } | |
2525 | } | |
2526 | ||
2527 | ASSERT(len == 0); | |
2528 | ||
390aab0a | 2529 | spin_lock(&log->l_icloglock); |
99428ad0 | 2530 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); |
df732b29 | 2531 | if (commit_iclog) { |
3468bb1c | 2532 | ASSERT(optype & XLOG_COMMIT_TRANS); |
df732b29 CH |
2533 | *commit_iclog = iclog; |
2534 | } else { | |
2535 | error = xlog_state_release_iclog(log, iclog); | |
2536 | } | |
390aab0a | 2537 | spin_unlock(&log->l_icloglock); |
1da177e4 | 2538 | |
df732b29 | 2539 | return error; |
99428ad0 | 2540 | } |
1da177e4 | 2541 | |
c814b4f2 CH |
2542 | static void |
2543 | xlog_state_activate_iclog( | |
2544 | struct xlog_in_core *iclog, | |
2545 | int *iclogs_changed) | |
2546 | { | |
2547 | ASSERT(list_empty_careful(&iclog->ic_callbacks)); | |
956f6daa | 2548 | trace_xlog_iclog_activate(iclog, _RET_IP_); |
c814b4f2 CH |
2549 | |
2550 | /* | |
2551 | * If the number of ops in this iclog indicate it just contains the | |
2552 | * dummy transaction, we can change state into IDLE (the second time | |
2553 | * around). Otherwise we should change the state into NEED a dummy. | |
2554 | * We don't need to cover the dummy. | |
2555 | */ | |
2556 | if (*iclogs_changed == 0 && | |
2557 | iclog->ic_header.h_num_logops == cpu_to_be32(XLOG_COVER_OPS)) { | |
2558 | *iclogs_changed = 1; | |
2559 | } else { | |
2560 | /* | |
2561 | * We have two dirty iclogs so start over. This could also be | |
2562 | * num of ops indicating this is not the dummy going out. | |
2563 | */ | |
2564 | *iclogs_changed = 2; | |
2565 | } | |
2566 | ||
2567 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2568 | iclog->ic_offset = 0; | |
2569 | iclog->ic_header.h_num_logops = 0; | |
2570 | memset(iclog->ic_header.h_cycle_data, 0, | |
2571 | sizeof(iclog->ic_header.h_cycle_data)); | |
2572 | iclog->ic_header.h_lsn = 0; | |
2573 | } | |
2574 | ||
0383f543 | 2575 | /* |
c814b4f2 CH |
2576 | * Loop through all iclogs and mark all iclogs currently marked DIRTY as |
2577 | * ACTIVE after iclog I/O has completed. | |
1da177e4 | 2578 | */ |
c814b4f2 CH |
2579 | static void |
2580 | xlog_state_activate_iclogs( | |
0383f543 | 2581 | struct xlog *log, |
c814b4f2 | 2582 | int *iclogs_changed) |
1da177e4 | 2583 | { |
c814b4f2 | 2584 | struct xlog_in_core *iclog = log->l_iclog; |
1da177e4 | 2585 | |
1da177e4 | 2586 | do { |
c814b4f2 CH |
2587 | if (iclog->ic_state == XLOG_STATE_DIRTY) |
2588 | xlog_state_activate_iclog(iclog, iclogs_changed); | |
2589 | /* | |
2590 | * The ordering of marking iclogs ACTIVE must be maintained, so | |
2591 | * an iclog doesn't become ACTIVE beyond one that is SYNCING. | |
2592 | */ | |
2593 | else if (iclog->ic_state != XLOG_STATE_ACTIVE) | |
2594 | break; | |
2595 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
2596 | } | |
0383f543 | 2597 | |
c814b4f2 CH |
2598 | static int |
2599 | xlog_covered_state( | |
2600 | int prev_state, | |
2601 | int iclogs_changed) | |
2602 | { | |
0383f543 | 2603 | /* |
b0eb9e11 BF |
2604 | * We go to NEED for any non-covering writes. We go to NEED2 if we just |
2605 | * wrote the first covering record (DONE). We go to IDLE if we just | |
2606 | * wrote the second covering record (DONE2) and remain in IDLE until a | |
2607 | * non-covering write occurs. | |
0383f543 | 2608 | */ |
c814b4f2 CH |
2609 | switch (prev_state) { |
2610 | case XLOG_STATE_COVER_IDLE: | |
b0eb9e11 BF |
2611 | if (iclogs_changed == 1) |
2612 | return XLOG_STATE_COVER_IDLE; | |
53004ee7 | 2613 | fallthrough; |
c814b4f2 CH |
2614 | case XLOG_STATE_COVER_NEED: |
2615 | case XLOG_STATE_COVER_NEED2: | |
2616 | break; | |
2617 | case XLOG_STATE_COVER_DONE: | |
2618 | if (iclogs_changed == 1) | |
2619 | return XLOG_STATE_COVER_NEED2; | |
2620 | break; | |
2621 | case XLOG_STATE_COVER_DONE2: | |
2622 | if (iclogs_changed == 1) | |
2623 | return XLOG_STATE_COVER_IDLE; | |
2624 | break; | |
2625 | default: | |
2626 | ASSERT(0); | |
2627 | } | |
0383f543 | 2628 | |
c814b4f2 CH |
2629 | return XLOG_STATE_COVER_NEED; |
2630 | } | |
1da177e4 | 2631 | |
c814b4f2 CH |
2632 | STATIC void |
2633 | xlog_state_clean_iclog( | |
2634 | struct xlog *log, | |
2635 | struct xlog_in_core *dirty_iclog) | |
2636 | { | |
2637 | int iclogs_changed = 0; | |
1da177e4 | 2638 | |
956f6daa DC |
2639 | trace_xlog_iclog_clean(dirty_iclog, _RET_IP_); |
2640 | ||
5781464b | 2641 | dirty_iclog->ic_state = XLOG_STATE_DIRTY; |
1da177e4 | 2642 | |
c814b4f2 CH |
2643 | xlog_state_activate_iclogs(log, &iclogs_changed); |
2644 | wake_up_all(&dirty_iclog->ic_force_wait); | |
2645 | ||
2646 | if (iclogs_changed) { | |
2647 | log->l_covered_state = xlog_covered_state(log->l_covered_state, | |
2648 | iclogs_changed); | |
1da177e4 | 2649 | } |
0383f543 | 2650 | } |
1da177e4 LT |
2651 | |
2652 | STATIC xfs_lsn_t | |
2653 | xlog_get_lowest_lsn( | |
9bff3132 | 2654 | struct xlog *log) |
1da177e4 | 2655 | { |
9bff3132 CH |
2656 | struct xlog_in_core *iclog = log->l_iclog; |
2657 | xfs_lsn_t lowest_lsn = 0, lsn; | |
1da177e4 | 2658 | |
1da177e4 | 2659 | do { |
1858bb0b CH |
2660 | if (iclog->ic_state == XLOG_STATE_ACTIVE || |
2661 | iclog->ic_state == XLOG_STATE_DIRTY) | |
9bff3132 CH |
2662 | continue; |
2663 | ||
2664 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2665 | if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0) | |
1da177e4 | 2666 | lowest_lsn = lsn; |
9bff3132 CH |
2667 | } while ((iclog = iclog->ic_next) != log->l_iclog); |
2668 | ||
014c2544 | 2669 | return lowest_lsn; |
1da177e4 LT |
2670 | } |
2671 | ||
14e15f1b DC |
2672 | /* |
2673 | * Completion of a iclog IO does not imply that a transaction has completed, as | |
2674 | * transactions can be large enough to span many iclogs. We cannot change the | |
2675 | * tail of the log half way through a transaction as this may be the only | |
2676 | * transaction in the log and moving the tail to point to the middle of it | |
2677 | * will prevent recovery from finding the start of the transaction. Hence we | |
2678 | * should only update the last_sync_lsn if this iclog contains transaction | |
2679 | * completion callbacks on it. | |
2680 | * | |
2681 | * We have to do this before we drop the icloglock to ensure we are the only one | |
2682 | * that can update it. | |
2683 | * | |
2684 | * If we are moving the last_sync_lsn forwards, we also need to ensure we kick | |
2685 | * the reservation grant head pushing. This is due to the fact that the push | |
2686 | * target is bound by the current last_sync_lsn value. Hence if we have a large | |
2687 | * amount of log space bound up in this committing transaction then the | |
2688 | * last_sync_lsn value may be the limiting factor preventing tail pushing from | |
2689 | * freeing space in the log. Hence once we've updated the last_sync_lsn we | |
2690 | * should push the AIL to ensure the push target (and hence the grant head) is | |
2691 | * no longer bound by the old log head location and can move forwards and make | |
2692 | * progress again. | |
2693 | */ | |
2694 | static void | |
2695 | xlog_state_set_callback( | |
2696 | struct xlog *log, | |
2697 | struct xlog_in_core *iclog, | |
2698 | xfs_lsn_t header_lsn) | |
2699 | { | |
956f6daa | 2700 | trace_xlog_iclog_callback(iclog, _RET_IP_); |
14e15f1b DC |
2701 | iclog->ic_state = XLOG_STATE_CALLBACK; |
2702 | ||
2703 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), | |
2704 | header_lsn) <= 0); | |
2705 | ||
2706 | if (list_empty_careful(&iclog->ic_callbacks)) | |
2707 | return; | |
2708 | ||
2709 | atomic64_set(&log->l_last_sync_lsn, header_lsn); | |
2710 | xlog_grant_push_ail(log, 0); | |
2711 | } | |
2712 | ||
5e96fa8d DC |
2713 | /* |
2714 | * Return true if we need to stop processing, false to continue to the next | |
2715 | * iclog. The caller will need to run callbacks if the iclog is returned in the | |
2716 | * XLOG_STATE_CALLBACK state. | |
2717 | */ | |
2718 | static bool | |
2719 | xlog_state_iodone_process_iclog( | |
2720 | struct xlog *log, | |
2721 | struct xlog_in_core *iclog, | |
5e96fa8d DC |
2722 | bool *ioerror) |
2723 | { | |
2724 | xfs_lsn_t lowest_lsn; | |
14e15f1b | 2725 | xfs_lsn_t header_lsn; |
5e96fa8d | 2726 | |
1858bb0b CH |
2727 | switch (iclog->ic_state) { |
2728 | case XLOG_STATE_ACTIVE: | |
2729 | case XLOG_STATE_DIRTY: | |
2730 | /* | |
2731 | * Skip all iclogs in the ACTIVE & DIRTY states: | |
2732 | */ | |
5e96fa8d | 2733 | return false; |
1858bb0b CH |
2734 | case XLOG_STATE_IOERROR: |
2735 | /* | |
2736 | * Between marking a filesystem SHUTDOWN and stopping the log, | |
2737 | * we do flush all iclogs to disk (if there wasn't a log I/O | |
2738 | * error). So, we do want things to go smoothly in case of just | |
4b29ab04 | 2739 | * a SHUTDOWN w/o a LOG_IO_ERROR. |
1858bb0b | 2740 | */ |
5e96fa8d DC |
2741 | *ioerror = true; |
2742 | return false; | |
1858bb0b | 2743 | case XLOG_STATE_DONE_SYNC: |
1858bb0b | 2744 | /* |
4b29ab04 CH |
2745 | * Now that we have an iclog that is in the DONE_SYNC state, do |
2746 | * one more check here to see if we have chased our tail around. | |
2747 | * If this is not the lowest lsn iclog, then we will leave it | |
2748 | * for another completion to process. | |
1858bb0b CH |
2749 | */ |
2750 | header_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2751 | lowest_lsn = xlog_get_lowest_lsn(log); | |
2752 | if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, header_lsn) < 0) | |
2753 | return false; | |
2754 | xlog_state_set_callback(log, iclog, header_lsn); | |
2755 | return false; | |
2756 | default: | |
2757 | /* | |
2758 | * Can only perform callbacks in order. Since this iclog is not | |
4b29ab04 CH |
2759 | * in the DONE_SYNC state, we skip the rest and just try to |
2760 | * clean up. | |
1858bb0b | 2761 | */ |
5e96fa8d DC |
2762 | return true; |
2763 | } | |
5e96fa8d DC |
2764 | } |
2765 | ||
1da177e4 LT |
2766 | STATIC void |
2767 | xlog_state_do_callback( | |
12e6a0f4 | 2768 | struct xlog *log) |
1da177e4 | 2769 | { |
5e96fa8d DC |
2770 | struct xlog_in_core *iclog; |
2771 | struct xlog_in_core *first_iclog; | |
5e96fa8d DC |
2772 | bool cycled_icloglock; |
2773 | bool ioerror; | |
2774 | int flushcnt = 0; | |
2775 | int repeats = 0; | |
1da177e4 | 2776 | |
b22cd72c | 2777 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2778 | do { |
2779 | /* | |
2780 | * Scan all iclogs starting with the one pointed to by the | |
2781 | * log. Reset this starting point each time the log is | |
2782 | * unlocked (during callbacks). | |
2783 | * | |
2784 | * Keep looping through iclogs until one full pass is made | |
2785 | * without running any callbacks. | |
2786 | */ | |
2787 | first_iclog = log->l_iclog; | |
2788 | iclog = log->l_iclog; | |
6546818c | 2789 | cycled_icloglock = false; |
5e96fa8d | 2790 | ioerror = false; |
1da177e4 LT |
2791 | repeats++; |
2792 | ||
2793 | do { | |
a1bb8505 DC |
2794 | LIST_HEAD(cb_list); |
2795 | ||
5e96fa8d | 2796 | if (xlog_state_iodone_process_iclog(log, iclog, |
4b29ab04 | 2797 | &ioerror)) |
5e96fa8d | 2798 | break; |
1da177e4 | 2799 | |
1858bb0b CH |
2800 | if (iclog->ic_state != XLOG_STATE_CALLBACK && |
2801 | iclog->ic_state != XLOG_STATE_IOERROR) { | |
1da177e4 LT |
2802 | iclog = iclog->ic_next; |
2803 | continue; | |
2804 | } | |
a1bb8505 | 2805 | list_splice_init(&iclog->ic_callbacks, &cb_list); |
6be00102 | 2806 | spin_unlock(&log->l_icloglock); |
1da177e4 | 2807 | |
a1bb8505 DC |
2808 | trace_xlog_iclog_callbacks_start(iclog, _RET_IP_); |
2809 | xlog_cil_process_committed(&cb_list); | |
2810 | trace_xlog_iclog_callbacks_done(iclog, _RET_IP_); | |
6546818c | 2811 | cycled_icloglock = true; |
6be00102 DC |
2812 | |
2813 | spin_lock(&log->l_icloglock); | |
5781464b CH |
2814 | if (XLOG_FORCED_SHUTDOWN(log)) |
2815 | wake_up_all(&iclog->ic_force_wait); | |
2816 | else | |
2817 | xlog_state_clean_iclog(log, iclog); | |
1da177e4 LT |
2818 | iclog = iclog->ic_next; |
2819 | } while (first_iclog != iclog); | |
a3c6685e NS |
2820 | |
2821 | if (repeats > 5000) { | |
2822 | flushcnt += repeats; | |
2823 | repeats = 0; | |
a0fa2b67 | 2824 | xfs_warn(log->l_mp, |
a3c6685e | 2825 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2826 | __func__, flushcnt); |
1da177e4 | 2827 | } |
5e96fa8d | 2828 | } while (!ioerror && cycled_icloglock); |
1da177e4 | 2829 | |
1858bb0b CH |
2830 | if (log->l_iclog->ic_state == XLOG_STATE_ACTIVE || |
2831 | log->l_iclog->ic_state == XLOG_STATE_IOERROR) | |
eb40a875 | 2832 | wake_up_all(&log->l_flush_wait); |
cdea5459 RR |
2833 | |
2834 | spin_unlock(&log->l_icloglock); | |
d748c623 | 2835 | } |
1da177e4 LT |
2836 | |
2837 | ||
2838 | /* | |
2839 | * Finish transitioning this iclog to the dirty state. | |
2840 | * | |
2841 | * Make sure that we completely execute this routine only when this is | |
2842 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2843 | * when we reach the end of the physical log, get turned into 2 separate | |
2844 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2845 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2846 | * the second completion goes through. | |
2847 | * | |
2848 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2849 | * global state machine log lock. |
1da177e4 | 2850 | */ |
a8272ce0 | 2851 | STATIC void |
1da177e4 | 2852 | xlog_state_done_syncing( |
12e6a0f4 | 2853 | struct xlog_in_core *iclog) |
1da177e4 | 2854 | { |
d15cbf2f | 2855 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2856 | |
b22cd72c | 2857 | spin_lock(&log->l_icloglock); |
155cc6b7 | 2858 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
956f6daa | 2859 | trace_xlog_iclog_sync_done(iclog, _RET_IP_); |
1da177e4 LT |
2860 | |
2861 | /* | |
2862 | * If we got an error, either on the first buffer, or in the case of | |
12e6a0f4 CH |
2863 | * split log writes, on the second, we shut down the file system and |
2864 | * no iclogs should ever be attempted to be written to disk again. | |
1da177e4 | 2865 | */ |
12e6a0f4 CH |
2866 | if (!XLOG_FORCED_SHUTDOWN(log)) { |
2867 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING); | |
1da177e4 | 2868 | iclog->ic_state = XLOG_STATE_DONE_SYNC; |
12e6a0f4 | 2869 | } |
1da177e4 LT |
2870 | |
2871 | /* | |
2872 | * Someone could be sleeping prior to writing out the next | |
2873 | * iclog buffer, we wake them all, one will get to do the | |
2874 | * I/O, the others get to wait for the result. | |
2875 | */ | |
eb40a875 | 2876 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2877 | spin_unlock(&log->l_icloglock); |
b843299b | 2878 | xlog_state_do_callback(log); |
12e6a0f4 | 2879 | } |
1da177e4 LT |
2880 | |
2881 | /* | |
2882 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2883 | * sleep. We wait on the flush queue on the head iclog as that should be |
2884 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2885 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2886 | * |
2887 | * The in-core logs are used in a circular fashion. They are not used | |
2888 | * out-of-order even when an iclog past the head is free. | |
2889 | * | |
2890 | * return: | |
2891 | * * log_offset where xlog_write() can start writing into the in-core | |
2892 | * log's data space. | |
2893 | * * in-core log pointer to which xlog_write() should write. | |
2894 | * * boolean indicating this is a continued write to an in-core log. | |
2895 | * If this is the last write, then the in-core log's offset field | |
2896 | * needs to be incremented, depending on the amount of data which | |
2897 | * is copied. | |
2898 | */ | |
a8272ce0 | 2899 | STATIC int |
9a8d2fdb MT |
2900 | xlog_state_get_iclog_space( |
2901 | struct xlog *log, | |
2902 | int len, | |
2903 | struct xlog_in_core **iclogp, | |
2904 | struct xlog_ticket *ticket, | |
2905 | int *continued_write, | |
2906 | int *logoffsetp) | |
1da177e4 | 2907 | { |
1da177e4 LT |
2908 | int log_offset; |
2909 | xlog_rec_header_t *head; | |
2910 | xlog_in_core_t *iclog; | |
1da177e4 LT |
2911 | |
2912 | restart: | |
b22cd72c | 2913 | spin_lock(&log->l_icloglock); |
1da177e4 | 2914 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2915 | spin_unlock(&log->l_icloglock); |
2451337d | 2916 | return -EIO; |
1da177e4 LT |
2917 | } |
2918 | ||
2919 | iclog = log->l_iclog; | |
d748c623 | 2920 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
ff6d6af2 | 2921 | XFS_STATS_INC(log->l_mp, xs_log_noiclogs); |
d748c623 MW |
2922 | |
2923 | /* Wait for log writes to have flushed */ | |
eb40a875 | 2924 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
2925 | goto restart; |
2926 | } | |
d748c623 | 2927 | |
1da177e4 LT |
2928 | head = &iclog->ic_header; |
2929 | ||
155cc6b7 | 2930 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
2931 | log_offset = iclog->ic_offset; |
2932 | ||
956f6daa DC |
2933 | trace_xlog_iclog_get_space(iclog, _RET_IP_); |
2934 | ||
1da177e4 LT |
2935 | /* On the 1st write to an iclog, figure out lsn. This works |
2936 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
2937 | * committing to. If the offset is set, that's how many blocks | |
2938 | * must be written. | |
2939 | */ | |
2940 | if (log_offset == 0) { | |
2941 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 2942 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
2943 | log->l_iclog_hsize, |
2944 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
2945 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
2946 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 2947 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
2948 | ASSERT(log->l_curr_block >= 0); |
2949 | } | |
2950 | ||
2951 | /* If there is enough room to write everything, then do it. Otherwise, | |
2952 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
2953 | * bit is on, so this will get flushed out. Don't update ic_offset | |
2954 | * until you know exactly how many bytes get copied. Therefore, wait | |
2955 | * until later to update ic_offset. | |
2956 | * | |
2957 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
2958 | * can fit into remaining data section. | |
2959 | */ | |
2960 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
df732b29 CH |
2961 | int error = 0; |
2962 | ||
1da177e4 LT |
2963 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); |
2964 | ||
49641f1a | 2965 | /* |
df732b29 CH |
2966 | * If we are the only one writing to this iclog, sync it to |
2967 | * disk. We need to do an atomic compare and decrement here to | |
2968 | * avoid racing with concurrent atomic_dec_and_lock() calls in | |
49641f1a DC |
2969 | * xlog_state_release_iclog() when there is more than one |
2970 | * reference to the iclog. | |
2971 | */ | |
df732b29 | 2972 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) |
49641f1a | 2973 | error = xlog_state_release_iclog(log, iclog); |
df732b29 CH |
2974 | spin_unlock(&log->l_icloglock); |
2975 | if (error) | |
2976 | return error; | |
1da177e4 LT |
2977 | goto restart; |
2978 | } | |
2979 | ||
2980 | /* Do we have enough room to write the full amount in the remainder | |
2981 | * of this iclog? Or must we continue a write on the next iclog and | |
2982 | * mark this iclog as completely taken? In the case where we switch | |
2983 | * iclogs (to mark it taken), this particular iclog will release/sync | |
2984 | * to disk in xlog_write(). | |
2985 | */ | |
2986 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
2987 | *continued_write = 0; | |
2988 | iclog->ic_offset += len; | |
2989 | } else { | |
2990 | *continued_write = 1; | |
2991 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2992 | } | |
2993 | *iclogp = iclog; | |
2994 | ||
2995 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 2996 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2997 | |
2998 | *logoffsetp = log_offset; | |
2999 | return 0; | |
b843299b | 3000 | } |
1da177e4 | 3001 | |
8b41e3f9 | 3002 | /* |
b843299b DC |
3003 | * The first cnt-1 times a ticket goes through here we don't need to move the |
3004 | * grant write head because the permanent reservation has reserved cnt times the | |
3005 | * unit amount. Release part of current permanent unit reservation and reset | |
3006 | * current reservation to be one units worth. Also move grant reservation head | |
3007 | * forward. | |
1da177e4 | 3008 | */ |
8b41e3f9 CH |
3009 | void |
3010 | xfs_log_ticket_regrant( | |
9a8d2fdb MT |
3011 | struct xlog *log, |
3012 | struct xlog_ticket *ticket) | |
1da177e4 | 3013 | { |
8b41e3f9 | 3014 | trace_xfs_log_ticket_regrant(log, ticket); |
0b1b213f | 3015 | |
1da177e4 LT |
3016 | if (ticket->t_cnt > 0) |
3017 | ticket->t_cnt--; | |
3018 | ||
28496968 | 3019 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3020 | ticket->t_curr_res); |
28496968 | 3021 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 3022 | ticket->t_curr_res); |
1da177e4 | 3023 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3024 | xlog_tic_reset_res(ticket); |
0b1b213f | 3025 | |
8b41e3f9 | 3026 | trace_xfs_log_ticket_regrant_sub(log, ticket); |
0b1b213f | 3027 | |
1da177e4 | 3028 | /* just return if we still have some of the pre-reserved space */ |
8b41e3f9 CH |
3029 | if (!ticket->t_cnt) { |
3030 | xlog_grant_add_space(log, &log->l_reserve_head.grant, | |
3031 | ticket->t_unit_res); | |
3032 | trace_xfs_log_ticket_regrant_exit(log, ticket); | |
1da177e4 | 3033 | |
8b41e3f9 CH |
3034 | ticket->t_curr_res = ticket->t_unit_res; |
3035 | xlog_tic_reset_res(ticket); | |
3036 | } | |
1da177e4 | 3037 | |
8b41e3f9 CH |
3038 | xfs_log_ticket_put(ticket); |
3039 | } | |
1da177e4 LT |
3040 | |
3041 | /* | |
3042 | * Give back the space left from a reservation. | |
3043 | * | |
3044 | * All the information we need to make a correct determination of space left | |
3045 | * is present. For non-permanent reservations, things are quite easy. The | |
3046 | * count should have been decremented to zero. We only need to deal with the | |
3047 | * space remaining in the current reservation part of the ticket. If the | |
3048 | * ticket contains a permanent reservation, there may be left over space which | |
3049 | * needs to be released. A count of N means that N-1 refills of the current | |
3050 | * reservation can be done before we need to ask for more space. The first | |
3051 | * one goes to fill up the first current reservation. Once we run out of | |
3052 | * space, the count will stay at zero and the only space remaining will be | |
3053 | * in the current reservation field. | |
3054 | */ | |
8b41e3f9 CH |
3055 | void |
3056 | xfs_log_ticket_ungrant( | |
9a8d2fdb MT |
3057 | struct xlog *log, |
3058 | struct xlog_ticket *ticket) | |
1da177e4 | 3059 | { |
8b41e3f9 CH |
3060 | int bytes; |
3061 | ||
3062 | trace_xfs_log_ticket_ungrant(log, ticket); | |
663e496a | 3063 | |
1da177e4 LT |
3064 | if (ticket->t_cnt > 0) |
3065 | ticket->t_cnt--; | |
3066 | ||
8b41e3f9 | 3067 | trace_xfs_log_ticket_ungrant_sub(log, ticket); |
1da177e4 | 3068 | |
663e496a DC |
3069 | /* |
3070 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
3071 | * up more space based on the remaining count. |
3072 | */ | |
663e496a | 3073 | bytes = ticket->t_curr_res; |
1da177e4 LT |
3074 | if (ticket->t_cnt > 0) { |
3075 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 3076 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
3077 | } |
3078 | ||
28496968 CH |
3079 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
3080 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 3081 | |
8b41e3f9 | 3082 | trace_xfs_log_ticket_ungrant_exit(log, ticket); |
0b1b213f | 3083 | |
cfb7cdca | 3084 | xfs_log_space_wake(log->l_mp); |
8b41e3f9 | 3085 | xfs_log_ticket_put(ticket); |
09a423a3 | 3086 | } |
1da177e4 | 3087 | |
1da177e4 | 3088 | /* |
b843299b DC |
3089 | * This routine will mark the current iclog in the ring as WANT_SYNC and move |
3090 | * the current iclog pointer to the next iclog in the ring. | |
1da177e4 LT |
3091 | */ |
3092 | STATIC void | |
9a8d2fdb MT |
3093 | xlog_state_switch_iclogs( |
3094 | struct xlog *log, | |
3095 | struct xlog_in_core *iclog, | |
3096 | int eventual_size) | |
1da177e4 LT |
3097 | { |
3098 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
69363999 | 3099 | assert_spin_locked(&log->l_icloglock); |
956f6daa | 3100 | trace_xlog_iclog_switch(iclog, _RET_IP_); |
69363999 | 3101 | |
1da177e4 LT |
3102 | if (!eventual_size) |
3103 | eventual_size = iclog->ic_offset; | |
3104 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3105 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3106 | log->l_prev_block = log->l_curr_block; |
3107 | log->l_prev_cycle = log->l_curr_cycle; | |
3108 | ||
3109 | /* roll log?: ic_offset changed later */ | |
3110 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3111 | ||
3112 | /* Round up to next log-sunit */ | |
a6a65fef | 3113 | if (log->l_iclog_roundoff > BBSIZE) { |
18842e0a | 3114 | uint32_t sunit_bb = BTOBB(log->l_iclog_roundoff); |
1da177e4 LT |
3115 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); |
3116 | } | |
3117 | ||
3118 | if (log->l_curr_block >= log->l_logBBsize) { | |
a45086e2 BF |
3119 | /* |
3120 | * Rewind the current block before the cycle is bumped to make | |
3121 | * sure that the combined LSN never transiently moves forward | |
3122 | * when the log wraps to the next cycle. This is to support the | |
3123 | * unlocked sample of these fields from xlog_valid_lsn(). Most | |
3124 | * other cases should acquire l_icloglock. | |
3125 | */ | |
3126 | log->l_curr_block -= log->l_logBBsize; | |
3127 | ASSERT(log->l_curr_block >= 0); | |
3128 | smp_wmb(); | |
1da177e4 LT |
3129 | log->l_curr_cycle++; |
3130 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3131 | log->l_curr_cycle++; | |
1da177e4 LT |
3132 | } |
3133 | ASSERT(iclog == log->l_iclog); | |
3134 | log->l_iclog = iclog->ic_next; | |
b843299b | 3135 | } |
1da177e4 | 3136 | |
1da177e4 LT |
3137 | /* |
3138 | * Write out all data in the in-core log as of this exact moment in time. | |
3139 | * | |
3140 | * Data may be written to the in-core log during this call. However, | |
3141 | * we don't guarantee this data will be written out. A change from past | |
3142 | * implementation means this routine will *not* write out zero length LRs. | |
3143 | * | |
3144 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3145 | * If we determine there is no flushable data, we just return. There is no | |
3146 | * flushable data if: | |
3147 | * | |
3148 | * 1. the current iclog is active and has no data; the previous iclog | |
3149 | * is in the active or dirty state. | |
3150 | * 2. the current iclog is drity, and the previous iclog is in the | |
3151 | * active or dirty state. | |
3152 | * | |
12017faf | 3153 | * We may sleep if: |
1da177e4 LT |
3154 | * |
3155 | * 1. the current iclog is not in the active nor dirty state. | |
3156 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3157 | * active nor dirty state. | |
3158 | * 3. the current iclog is active, and there is another thread writing | |
3159 | * to this particular iclog. | |
3160 | * 4. a) the current iclog is active and has no other writers | |
3161 | * b) when we return from flushing out this iclog, it is still | |
3162 | * not in the active nor dirty state. | |
3163 | */ | |
a14a348b | 3164 | int |
60e5bb78 | 3165 | xfs_log_force( |
a14a348b | 3166 | struct xfs_mount *mp, |
60e5bb78 | 3167 | uint flags) |
1da177e4 | 3168 | { |
ad223e60 | 3169 | struct xlog *log = mp->m_log; |
a14a348b CH |
3170 | struct xlog_in_core *iclog; |
3171 | xfs_lsn_t lsn; | |
3172 | ||
ff6d6af2 | 3173 | XFS_STATS_INC(mp, xs_log_force); |
60e5bb78 | 3174 | trace_xfs_log_force(mp, 0, _RET_IP_); |
1da177e4 | 3175 | |
93b8a585 | 3176 | xlog_cil_force(log); |
71e330b5 | 3177 | |
b22cd72c | 3178 | spin_lock(&log->l_icloglock); |
1da177e4 | 3179 | iclog = log->l_iclog; |
1858bb0b | 3180 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
e6b96570 | 3181 | goto out_error; |
1da177e4 | 3182 | |
956f6daa DC |
3183 | trace_xlog_iclog_force(iclog, _RET_IP_); |
3184 | ||
e6b96570 CH |
3185 | if (iclog->ic_state == XLOG_STATE_DIRTY || |
3186 | (iclog->ic_state == XLOG_STATE_ACTIVE && | |
3187 | atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) { | |
1da177e4 | 3188 | /* |
e6b96570 CH |
3189 | * If the head is dirty or (active and empty), then we need to |
3190 | * look at the previous iclog. | |
3191 | * | |
3192 | * If the previous iclog is active or dirty we are done. There | |
3193 | * is nothing to sync out. Otherwise, we attach ourselves to the | |
1da177e4 LT |
3194 | * previous iclog and go to sleep. |
3195 | */ | |
e6b96570 | 3196 | iclog = iclog->ic_prev; |
e6b96570 CH |
3197 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3198 | if (atomic_read(&iclog->ic_refcnt) == 0) { | |
3199 | /* | |
3200 | * We are the only one with access to this iclog. | |
3201 | * | |
3202 | * Flush it out now. There should be a roundoff of zero | |
3203 | * to show that someone has already taken care of the | |
3204 | * roundoff from the previous sync. | |
3205 | */ | |
3206 | atomic_inc(&iclog->ic_refcnt); | |
3207 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
3208 | xlog_state_switch_iclogs(log, iclog, 0); | |
e6b96570 | 3209 | if (xlog_state_release_iclog(log, iclog)) |
df732b29 | 3210 | goto out_error; |
1da177e4 | 3211 | |
81e5b50a | 3212 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) |
e6b96570 CH |
3213 | goto out_unlock; |
3214 | } else { | |
3215 | /* | |
3216 | * Someone else is writing to this iclog. | |
3217 | * | |
3218 | * Use its call to flush out the data. However, the | |
3219 | * other thread may not force out this LR, so we mark | |
3220 | * it WANT_SYNC. | |
3221 | */ | |
3222 | xlog_state_switch_iclogs(log, iclog, 0); | |
1da177e4 | 3223 | } |
e6b96570 | 3224 | } else { |
1da177e4 | 3225 | /* |
e6b96570 CH |
3226 | * If the head iclog is not active nor dirty, we just attach |
3227 | * ourselves to the head and go to sleep if necessary. | |
1da177e4 | 3228 | */ |
e6b96570 | 3229 | ; |
1da177e4 | 3230 | } |
e6b96570 | 3231 | |
81e5b50a CH |
3232 | if (flags & XFS_LOG_SYNC) |
3233 | return xlog_wait_on_iclog(iclog); | |
e6b96570 CH |
3234 | out_unlock: |
3235 | spin_unlock(&log->l_icloglock); | |
3236 | return 0; | |
3237 | out_error: | |
3238 | spin_unlock(&log->l_icloglock); | |
3239 | return -EIO; | |
a14a348b | 3240 | } |
1da177e4 | 3241 | |
3e4da466 | 3242 | static int |
5f9b4b0d DC |
3243 | xlog_force_lsn( |
3244 | struct xlog *log, | |
a14a348b CH |
3245 | xfs_lsn_t lsn, |
3246 | uint flags, | |
3e4da466 CH |
3247 | int *log_flushed, |
3248 | bool already_slept) | |
1da177e4 | 3249 | { |
a14a348b | 3250 | struct xlog_in_core *iclog; |
71e330b5 | 3251 | |
a14a348b CH |
3252 | spin_lock(&log->l_icloglock); |
3253 | iclog = log->l_iclog; | |
1858bb0b | 3254 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
93806299 | 3255 | goto out_error; |
1da177e4 | 3256 | |
93806299 | 3257 | while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { |
956f6daa | 3258 | trace_xlog_iclog_force_lsn(iclog, _RET_IP_); |
93806299 CH |
3259 | iclog = iclog->ic_next; |
3260 | if (iclog == log->l_iclog) | |
3261 | goto out_unlock; | |
3262 | } | |
a14a348b | 3263 | |
93806299 CH |
3264 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3265 | /* | |
3266 | * We sleep here if we haven't already slept (e.g. this is the | |
3267 | * first time we've looked at the correct iclog buf) and the | |
3268 | * buffer before us is going to be sync'ed. The reason for this | |
3269 | * is that if we are doing sync transactions here, by waiting | |
3270 | * for the previous I/O to complete, we can allow a few more | |
3271 | * transactions into this iclog before we close it down. | |
3272 | * | |
3273 | * Otherwise, we mark the buffer WANT_SYNC, and bump up the | |
3274 | * refcnt so we can release the log (which drops the ref count). | |
3275 | * The state switch keeps new transaction commits from using | |
3276 | * this buffer. When the current commits finish writing into | |
3277 | * the buffer, the refcount will drop to zero and the buffer | |
3278 | * will go out then. | |
3279 | */ | |
3280 | if (!already_slept && | |
1858bb0b CH |
3281 | (iclog->ic_prev->ic_state == XLOG_STATE_WANT_SYNC || |
3282 | iclog->ic_prev->ic_state == XLOG_STATE_SYNCING)) { | |
93806299 CH |
3283 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3284 | &log->l_icloglock); | |
3e4da466 | 3285 | return -EAGAIN; |
1da177e4 | 3286 | } |
93806299 CH |
3287 | atomic_inc(&iclog->ic_refcnt); |
3288 | xlog_state_switch_iclogs(log, iclog, 0); | |
93806299 | 3289 | if (xlog_state_release_iclog(log, iclog)) |
df732b29 | 3290 | goto out_error; |
93806299 CH |
3291 | if (log_flushed) |
3292 | *log_flushed = 1; | |
93806299 | 3293 | } |
1da177e4 | 3294 | |
81e5b50a CH |
3295 | if (flags & XFS_LOG_SYNC) |
3296 | return xlog_wait_on_iclog(iclog); | |
93806299 | 3297 | out_unlock: |
a14a348b CH |
3298 | spin_unlock(&log->l_icloglock); |
3299 | return 0; | |
93806299 CH |
3300 | out_error: |
3301 | spin_unlock(&log->l_icloglock); | |
3302 | return -EIO; | |
a14a348b CH |
3303 | } |
3304 | ||
3e4da466 CH |
3305 | /* |
3306 | * Force the in-core log to disk for a specific LSN. | |
3307 | * | |
3308 | * Find in-core log with lsn. | |
3309 | * If it is in the DIRTY state, just return. | |
3310 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3311 | * state and go to sleep or return. | |
3312 | * If it is in any other state, go to sleep or return. | |
3313 | * | |
3314 | * Synchronous forces are implemented with a wait queue. All callers trying | |
3315 | * to force a given lsn to disk must wait on the queue attached to the | |
3316 | * specific in-core log. When given in-core log finally completes its write | |
3317 | * to disk, that thread will wake up all threads waiting on the queue. | |
3318 | */ | |
3319 | int | |
5f9b4b0d | 3320 | xfs_log_force_seq( |
3e4da466 | 3321 | struct xfs_mount *mp, |
5f9b4b0d | 3322 | xfs_csn_t seq, |
3e4da466 CH |
3323 | uint flags, |
3324 | int *log_flushed) | |
3325 | { | |
5f9b4b0d DC |
3326 | struct xlog *log = mp->m_log; |
3327 | xfs_lsn_t lsn; | |
3e4da466 | 3328 | int ret; |
5f9b4b0d | 3329 | ASSERT(seq != 0); |
3e4da466 CH |
3330 | |
3331 | XFS_STATS_INC(mp, xs_log_force); | |
5f9b4b0d | 3332 | trace_xfs_log_force(mp, seq, _RET_IP_); |
3e4da466 | 3333 | |
5f9b4b0d | 3334 | lsn = xlog_cil_force_seq(log, seq); |
3e4da466 CH |
3335 | if (lsn == NULLCOMMITLSN) |
3336 | return 0; | |
3337 | ||
5f9b4b0d DC |
3338 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, false); |
3339 | if (ret == -EAGAIN) { | |
3340 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3341 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, true); | |
3342 | } | |
3e4da466 CH |
3343 | return ret; |
3344 | } | |
3345 | ||
1da177e4 | 3346 | /* |
9da096fd | 3347 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3348 | */ |
cc09c0dc DC |
3349 | void |
3350 | xfs_log_ticket_put( | |
3351 | xlog_ticket_t *ticket) | |
1da177e4 | 3352 | { |
cc09c0dc | 3353 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3354 | if (atomic_dec_and_test(&ticket->t_ref)) |
377bcd5f | 3355 | kmem_cache_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3356 | } |
1da177e4 | 3357 | |
cc09c0dc DC |
3358 | xlog_ticket_t * |
3359 | xfs_log_ticket_get( | |
3360 | xlog_ticket_t *ticket) | |
3361 | { | |
3362 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3363 | atomic_inc(&ticket->t_ref); | |
3364 | return ticket; | |
3365 | } | |
1da177e4 LT |
3366 | |
3367 | /* | |
e773fc93 JL |
3368 | * Figure out the total log space unit (in bytes) that would be |
3369 | * required for a log ticket. | |
1da177e4 | 3370 | */ |
a6a65fef DC |
3371 | static int |
3372 | xlog_calc_unit_res( | |
3373 | struct xlog *log, | |
e773fc93 | 3374 | int unit_bytes) |
1da177e4 | 3375 | { |
e773fc93 JL |
3376 | int iclog_space; |
3377 | uint num_headers; | |
1da177e4 LT |
3378 | |
3379 | /* | |
3380 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3381 | * in the log. A unit in this case is the amount of space for one | |
3382 | * of these log operations. Normal reservations have a cnt of 1 | |
3383 | * and their unit amount is the total amount of space required. | |
3384 | * | |
3385 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3386 | * which occupy space in the on-disk log. |
3387 | * | |
3388 | * Normal form of a transaction is: | |
3389 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3390 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3391 | * | |
3392 | * We need to account for all the leadup data and trailer data | |
3393 | * around the transaction data. | |
3394 | * And then we need to account for the worst case in terms of using | |
3395 | * more space. | |
3396 | * The worst case will happen if: | |
3397 | * - the placement of the transaction happens to be such that the | |
3398 | * roundoff is at its maximum | |
3399 | * - the transaction data is synced before the commit record is synced | |
3400 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3401 | * Therefore the commit record is in its own Log Record. | |
3402 | * This can happen as the commit record is called with its | |
3403 | * own region to xlog_write(). | |
3404 | * This then means that in the worst case, roundoff can happen for | |
3405 | * the commit-rec as well. | |
3406 | * The commit-rec is smaller than padding in this scenario and so it is | |
3407 | * not added separately. | |
1da177e4 LT |
3408 | */ |
3409 | ||
32fb9b57 TS |
3410 | /* for trans header */ |
3411 | unit_bytes += sizeof(xlog_op_header_t); | |
3412 | unit_bytes += sizeof(xfs_trans_header_t); | |
3413 | ||
1da177e4 | 3414 | /* for start-rec */ |
32fb9b57 TS |
3415 | unit_bytes += sizeof(xlog_op_header_t); |
3416 | ||
9b9fc2b7 DC |
3417 | /* |
3418 | * for LR headers - the space for data in an iclog is the size minus | |
3419 | * the space used for the headers. If we use the iclog size, then we | |
3420 | * undercalculate the number of headers required. | |
3421 | * | |
3422 | * Furthermore - the addition of op headers for split-recs might | |
3423 | * increase the space required enough to require more log and op | |
3424 | * headers, so take that into account too. | |
3425 | * | |
3426 | * IMPORTANT: This reservation makes the assumption that if this | |
3427 | * transaction is the first in an iclog and hence has the LR headers | |
3428 | * accounted to it, then the remaining space in the iclog is | |
3429 | * exclusively for this transaction. i.e. if the transaction is larger | |
3430 | * than the iclog, it will be the only thing in that iclog. | |
3431 | * Fundamentally, this means we must pass the entire log vector to | |
3432 | * xlog_write to guarantee this. | |
3433 | */ | |
3434 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3435 | num_headers = howmany(unit_bytes, iclog_space); | |
3436 | ||
3437 | /* for split-recs - ophdrs added when data split over LRs */ | |
3438 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3439 | ||
3440 | /* add extra header reservations if we overrun */ | |
3441 | while (!num_headers || | |
3442 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3443 | unit_bytes += sizeof(xlog_op_header_t); | |
3444 | num_headers++; | |
3445 | } | |
32fb9b57 | 3446 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3447 | |
32fb9b57 TS |
3448 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3449 | unit_bytes += log->l_iclog_hsize; | |
3450 | ||
a6a65fef DC |
3451 | /* roundoff padding for transaction data and one for commit record */ |
3452 | unit_bytes += 2 * log->l_iclog_roundoff; | |
1da177e4 | 3453 | |
e773fc93 JL |
3454 | return unit_bytes; |
3455 | } | |
3456 | ||
a6a65fef DC |
3457 | int |
3458 | xfs_log_calc_unit_res( | |
3459 | struct xfs_mount *mp, | |
3460 | int unit_bytes) | |
3461 | { | |
3462 | return xlog_calc_unit_res(mp->m_log, unit_bytes); | |
3463 | } | |
3464 | ||
e773fc93 JL |
3465 | /* |
3466 | * Allocate and initialise a new log ticket. | |
3467 | */ | |
3468 | struct xlog_ticket * | |
3469 | xlog_ticket_alloc( | |
3470 | struct xlog *log, | |
3471 | int unit_bytes, | |
3472 | int cnt, | |
3473 | char client, | |
ca4f2589 | 3474 | bool permanent) |
e773fc93 JL |
3475 | { |
3476 | struct xlog_ticket *tic; | |
3477 | int unit_res; | |
3478 | ||
ca4f2589 | 3479 | tic = kmem_cache_zalloc(xfs_log_ticket_zone, GFP_NOFS | __GFP_NOFAIL); |
e773fc93 | 3480 | |
a6a65fef | 3481 | unit_res = xlog_calc_unit_res(log, unit_bytes); |
e773fc93 | 3482 | |
cc09c0dc | 3483 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3484 | tic->t_task = current; |
10547941 | 3485 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3486 | tic->t_unit_res = unit_res; |
3487 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3488 | tic->t_cnt = cnt; |
3489 | tic->t_ocnt = cnt; | |
ecb3403d | 3490 | tic->t_tid = prandom_u32(); |
1da177e4 | 3491 | tic->t_clientid = client; |
9006fb91 | 3492 | if (permanent) |
1da177e4 | 3493 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3494 | |
0adba536 | 3495 | xlog_tic_reset_res(tic); |
7e9c6396 | 3496 | |
1da177e4 | 3497 | return tic; |
cc09c0dc | 3498 | } |
1da177e4 | 3499 | |
cfcbbbd0 | 3500 | #if defined(DEBUG) |
1da177e4 LT |
3501 | /* |
3502 | * Make sure that the destination ptr is within the valid data region of | |
3503 | * one of the iclogs. This uses backup pointers stored in a different | |
3504 | * part of the log in case we trash the log structure. | |
3505 | */ | |
181fdfe6 | 3506 | STATIC void |
e6b1f273 | 3507 | xlog_verify_dest_ptr( |
ad223e60 | 3508 | struct xlog *log, |
5809d5e0 | 3509 | void *ptr) |
1da177e4 LT |
3510 | { |
3511 | int i; | |
3512 | int good_ptr = 0; | |
3513 | ||
e6b1f273 CH |
3514 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3515 | if (ptr >= log->l_iclog_bak[i] && | |
3516 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3517 | good_ptr++; |
3518 | } | |
e6b1f273 CH |
3519 | |
3520 | if (!good_ptr) | |
a0fa2b67 | 3521 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3522 | } |
1da177e4 | 3523 | |
da8a1a4a DC |
3524 | /* |
3525 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3526 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3527 | * the cycles differ by exactly one and check the byte count. | |
3528 | * | |
3529 | * This check is run unlocked, so can give false positives. Rather than assert | |
3530 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3531 | * determine if they want to panic the machine when such an error occurs. For | |
3532 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3533 | * an assert, it can be turned off at runtime. | |
3534 | */ | |
3f336c6f DC |
3535 | STATIC void |
3536 | xlog_verify_grant_tail( | |
ad223e60 | 3537 | struct xlog *log) |
3f336c6f | 3538 | { |
1c3cb9ec | 3539 | int tail_cycle, tail_blocks; |
a69ed03c | 3540 | int cycle, space; |
3f336c6f | 3541 | |
28496968 | 3542 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3543 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3544 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3545 | if (cycle - 1 != tail_cycle && |
3546 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3547 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3548 | "%s: cycle - 1 != tail_cycle", __func__); | |
3549 | log->l_flags |= XLOG_TAIL_WARN; | |
3550 | } | |
3551 | ||
3552 | if (space > BBTOB(tail_blocks) && | |
3553 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3554 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3555 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3556 | log->l_flags |= XLOG_TAIL_WARN; | |
3557 | } | |
3f336c6f DC |
3558 | } |
3559 | } | |
3560 | ||
1da177e4 LT |
3561 | /* check if it will fit */ |
3562 | STATIC void | |
9a8d2fdb MT |
3563 | xlog_verify_tail_lsn( |
3564 | struct xlog *log, | |
3565 | struct xlog_in_core *iclog, | |
3566 | xfs_lsn_t tail_lsn) | |
1da177e4 LT |
3567 | { |
3568 | int blocks; | |
3569 | ||
3570 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3571 | blocks = | |
3572 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3573 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3574 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3575 | } else { |
3576 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3577 | ||
3578 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3579 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3580 | |
3581 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3582 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3583 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 | 3584 | } |
b843299b | 3585 | } |
1da177e4 LT |
3586 | |
3587 | /* | |
3588 | * Perform a number of checks on the iclog before writing to disk. | |
3589 | * | |
3590 | * 1. Make sure the iclogs are still circular | |
3591 | * 2. Make sure we have a good magic number | |
3592 | * 3. Make sure we don't have magic numbers in the data | |
3593 | * 4. Check fields of each log operation header for: | |
3594 | * A. Valid client identifier | |
3595 | * B. tid ptr value falls in valid ptr space (user space code) | |
3596 | * C. Length in log record header is correct according to the | |
3597 | * individual operation headers within record. | |
3598 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3599 | * log, check the preceding blocks of the physical log to make sure all | |
3600 | * the cycle numbers agree with the current cycle number. | |
3601 | */ | |
3602 | STATIC void | |
9a8d2fdb MT |
3603 | xlog_verify_iclog( |
3604 | struct xlog *log, | |
3605 | struct xlog_in_core *iclog, | |
abca1f33 | 3606 | int count) |
1da177e4 LT |
3607 | { |
3608 | xlog_op_header_t *ophead; | |
3609 | xlog_in_core_t *icptr; | |
3610 | xlog_in_core_2_t *xhdr; | |
5809d5e0 | 3611 | void *base_ptr, *ptr, *p; |
db9d67d6 | 3612 | ptrdiff_t field_offset; |
c8ce540d | 3613 | uint8_t clientid; |
1da177e4 LT |
3614 | int len, i, j, k, op_len; |
3615 | int idx; | |
1da177e4 LT |
3616 | |
3617 | /* check validity of iclog pointers */ | |
b22cd72c | 3618 | spin_lock(&log->l_icloglock); |
1da177e4 | 3619 | icptr = log->l_iclog; |
643f7c4e GB |
3620 | for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next) |
3621 | ASSERT(icptr); | |
3622 | ||
1da177e4 | 3623 | if (icptr != log->l_iclog) |
a0fa2b67 | 3624 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3625 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3626 | |
3627 | /* check log magic numbers */ | |
69ef921b | 3628 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3629 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3630 | |
5809d5e0 CH |
3631 | base_ptr = ptr = &iclog->ic_header; |
3632 | p = &iclog->ic_header; | |
3633 | for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) { | |
69ef921b | 3634 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3635 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3636 | __func__); | |
1da177e4 LT |
3637 | } |
3638 | ||
3639 | /* check fields */ | |
b53e675d | 3640 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
5809d5e0 CH |
3641 | base_ptr = ptr = iclog->ic_datap; |
3642 | ophead = ptr; | |
b28708d6 | 3643 | xhdr = iclog->ic_data; |
1da177e4 | 3644 | for (i = 0; i < len; i++) { |
5809d5e0 | 3645 | ophead = ptr; |
1da177e4 LT |
3646 | |
3647 | /* clientid is only 1 byte */ | |
5809d5e0 CH |
3648 | p = &ophead->oh_clientid; |
3649 | field_offset = p - base_ptr; | |
abca1f33 | 3650 | if (field_offset & 0x1ff) { |
1da177e4 LT |
3651 | clientid = ophead->oh_clientid; |
3652 | } else { | |
b2a922cd | 3653 | idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap); |
1da177e4 LT |
3654 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3655 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3656 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3657 | clientid = xlog_get_client_id( |
3658 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3659 | } else { |
03bea6fe CH |
3660 | clientid = xlog_get_client_id( |
3661 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3662 | } |
3663 | } | |
3664 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 | 3665 | xfs_warn(log->l_mp, |
c9690043 | 3666 | "%s: invalid clientid %d op "PTR_FMT" offset 0x%lx", |
a0fa2b67 DC |
3667 | __func__, clientid, ophead, |
3668 | (unsigned long)field_offset); | |
1da177e4 LT |
3669 | |
3670 | /* check length */ | |
5809d5e0 CH |
3671 | p = &ophead->oh_len; |
3672 | field_offset = p - base_ptr; | |
abca1f33 | 3673 | if (field_offset & 0x1ff) { |
67fcb7bf | 3674 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 | 3675 | } else { |
db9d67d6 CH |
3676 | idx = BTOBBT((uintptr_t)&ophead->oh_len - |
3677 | (uintptr_t)iclog->ic_datap); | |
1da177e4 LT |
3678 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3679 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3680 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3681 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3682 | } else { |
b53e675d | 3683 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3684 | } |
3685 | } | |
3686 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3687 | } | |
b843299b | 3688 | } |
cfcbbbd0 | 3689 | #endif |
1da177e4 LT |
3690 | |
3691 | /* | |
b22cd72c | 3692 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3693 | */ |
3694 | STATIC int | |
3695 | xlog_state_ioerror( | |
9a8d2fdb | 3696 | struct xlog *log) |
1da177e4 LT |
3697 | { |
3698 | xlog_in_core_t *iclog, *ic; | |
3699 | ||
3700 | iclog = log->l_iclog; | |
1858bb0b | 3701 | if (iclog->ic_state != XLOG_STATE_IOERROR) { |
1da177e4 LT |
3702 | /* |
3703 | * Mark all the incore logs IOERROR. | |
3704 | * From now on, no log flushes will result. | |
3705 | */ | |
3706 | ic = iclog; | |
3707 | do { | |
3708 | ic->ic_state = XLOG_STATE_IOERROR; | |
3709 | ic = ic->ic_next; | |
3710 | } while (ic != iclog); | |
014c2544 | 3711 | return 0; |
1da177e4 LT |
3712 | } |
3713 | /* | |
3714 | * Return non-zero, if state transition has already happened. | |
3715 | */ | |
014c2544 | 3716 | return 1; |
1da177e4 LT |
3717 | } |
3718 | ||
3719 | /* | |
3720 | * This is called from xfs_force_shutdown, when we're forcibly | |
3721 | * shutting down the filesystem, typically because of an IO error. | |
3722 | * Our main objectives here are to make sure that: | |
a870fe6d DC |
3723 | * a. if !logerror, flush the logs to disk. Anything modified |
3724 | * after this is ignored. | |
3725 | * b. the filesystem gets marked 'SHUTDOWN' for all interested | |
1da177e4 | 3726 | * parties to find out, 'atomically'. |
a870fe6d | 3727 | * c. those who're sleeping on log reservations, pinned objects and |
1da177e4 | 3728 | * other resources get woken up, and be told the bad news. |
a870fe6d | 3729 | * d. nothing new gets queued up after (b) and (c) are done. |
9da1ab18 | 3730 | * |
a870fe6d DC |
3731 | * Note: for the !logerror case we need to flush the regions held in memory out |
3732 | * to disk first. This needs to be done before the log is marked as shutdown, | |
3733 | * otherwise the iclog writes will fail. | |
1da177e4 LT |
3734 | */ |
3735 | int | |
3736 | xfs_log_force_umount( | |
3737 | struct xfs_mount *mp, | |
3738 | int logerror) | |
3739 | { | |
9a8d2fdb | 3740 | struct xlog *log; |
1da177e4 | 3741 | int retval; |
1da177e4 LT |
3742 | |
3743 | log = mp->m_log; | |
3744 | ||
3745 | /* | |
3746 | * If this happens during log recovery, don't worry about | |
3747 | * locking; the log isn't open for business yet. | |
3748 | */ | |
3749 | if (!log || | |
3750 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3751 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 | 3752 | if (mp->m_sb_bp) |
b0388bf1 | 3753 | mp->m_sb_bp->b_flags |= XBF_DONE; |
014c2544 | 3754 | return 0; |
1da177e4 LT |
3755 | } |
3756 | ||
3757 | /* | |
3758 | * Somebody could've already done the hard work for us. | |
3759 | * No need to get locks for this. | |
3760 | */ | |
1858bb0b | 3761 | if (logerror && log->l_iclog->ic_state == XLOG_STATE_IOERROR) { |
1da177e4 | 3762 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); |
014c2544 | 3763 | return 1; |
1da177e4 | 3764 | } |
9da1ab18 DC |
3765 | |
3766 | /* | |
a870fe6d DC |
3767 | * Flush all the completed transactions to disk before marking the log |
3768 | * being shut down. We need to do it in this order to ensure that | |
3769 | * completed operations are safely on disk before we shut down, and that | |
3770 | * we don't have to issue any buffer IO after the shutdown flags are set | |
3771 | * to guarantee this. | |
9da1ab18 | 3772 | */ |
93b8a585 | 3773 | if (!logerror) |
60e5bb78 | 3774 | xfs_log_force(mp, XFS_LOG_SYNC); |
9da1ab18 | 3775 | |
1da177e4 | 3776 | /* |
3f16b985 DC |
3777 | * mark the filesystem and the as in a shutdown state and wake |
3778 | * everybody up to tell them the bad news. | |
1da177e4 | 3779 | */ |
b22cd72c | 3780 | spin_lock(&log->l_icloglock); |
1da177e4 | 3781 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 | 3782 | if (mp->m_sb_bp) |
b0388bf1 | 3783 | mp->m_sb_bp->b_flags |= XBF_DONE; |
bac8dca9 | 3784 | |
1da177e4 | 3785 | /* |
a870fe6d DC |
3786 | * Mark the log and the iclogs with IO error flags to prevent any |
3787 | * further log IO from being issued or completed. | |
1da177e4 LT |
3788 | */ |
3789 | log->l_flags |= XLOG_IO_ERROR; | |
a870fe6d | 3790 | retval = xlog_state_ioerror(log); |
b22cd72c | 3791 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3792 | |
3793 | /* | |
10547941 DC |
3794 | * We don't want anybody waiting for log reservations after this. That |
3795 | * means we have to wake up everybody queued up on reserveq as well as | |
3796 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3797 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3798 | * action is protected by the grant locks. |
1da177e4 | 3799 | */ |
a79bf2d7 CH |
3800 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3801 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3802 | |
1da177e4 | 3803 | /* |
ac983517 DC |
3804 | * Wake up everybody waiting on xfs_log_force. Wake the CIL push first |
3805 | * as if the log writes were completed. The abort handling in the log | |
3806 | * item committed callback functions will do this again under lock to | |
3807 | * avoid races. | |
1da177e4 | 3808 | */ |
cdea5459 | 3809 | spin_lock(&log->l_cilp->xc_push_lock); |
ac983517 | 3810 | wake_up_all(&log->l_cilp->xc_commit_wait); |
cdea5459 | 3811 | spin_unlock(&log->l_cilp->xc_push_lock); |
12e6a0f4 | 3812 | xlog_state_do_callback(log); |
1da177e4 | 3813 | |
1da177e4 | 3814 | /* return non-zero if log IOERROR transition had already happened */ |
014c2544 | 3815 | return retval; |
1da177e4 LT |
3816 | } |
3817 | ||
ba0f32d4 | 3818 | STATIC int |
9a8d2fdb MT |
3819 | xlog_iclogs_empty( |
3820 | struct xlog *log) | |
1da177e4 LT |
3821 | { |
3822 | xlog_in_core_t *iclog; | |
3823 | ||
3824 | iclog = log->l_iclog; | |
3825 | do { | |
3826 | /* endianness does not matter here, zero is zero in | |
3827 | * any language. | |
3828 | */ | |
3829 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3830 | return 0; |
1da177e4 LT |
3831 | iclog = iclog->ic_next; |
3832 | } while (iclog != log->l_iclog); | |
014c2544 | 3833 | return 1; |
1da177e4 | 3834 | } |
f661f1e0 | 3835 | |
a45086e2 BF |
3836 | /* |
3837 | * Verify that an LSN stamped into a piece of metadata is valid. This is | |
3838 | * intended for use in read verifiers on v5 superblocks. | |
3839 | */ | |
3840 | bool | |
3841 | xfs_log_check_lsn( | |
3842 | struct xfs_mount *mp, | |
3843 | xfs_lsn_t lsn) | |
3844 | { | |
3845 | struct xlog *log = mp->m_log; | |
3846 | bool valid; | |
3847 | ||
3848 | /* | |
3849 | * norecovery mode skips mount-time log processing and unconditionally | |
3850 | * resets the in-core LSN. We can't validate in this mode, but | |
3851 | * modifications are not allowed anyways so just return true. | |
3852 | */ | |
3853 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
3854 | return true; | |
3855 | ||
3856 | /* | |
3857 | * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is | |
3858 | * handled by recovery and thus safe to ignore here. | |
3859 | */ | |
3860 | if (lsn == NULLCOMMITLSN) | |
3861 | return true; | |
3862 | ||
3863 | valid = xlog_valid_lsn(mp->m_log, lsn); | |
3864 | ||
3865 | /* warn the user about what's gone wrong before verifier failure */ | |
3866 | if (!valid) { | |
3867 | spin_lock(&log->l_icloglock); | |
3868 | xfs_warn(mp, | |
3869 | "Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). " | |
3870 | "Please unmount and run xfs_repair (>= v4.3) to resolve.", | |
3871 | CYCLE_LSN(lsn), BLOCK_LSN(lsn), | |
3872 | log->l_curr_cycle, log->l_curr_block); | |
3873 | spin_unlock(&log->l_icloglock); | |
3874 | } | |
3875 | ||
3876 | return valid; | |
3877 | } | |
0c60d3aa DW |
3878 | |
3879 | bool | |
3880 | xfs_log_in_recovery( | |
3881 | struct xfs_mount *mp) | |
3882 | { | |
3883 | struct xlog *log = mp->m_log; | |
3884 | ||
3885 | return log->l_flags & XLOG_ACTIVE_RECOVERY; | |
3886 | } |