]>
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, | |
9d110014 | 81 | struct xlog_in_core *iclog); |
1da177e4 LT |
82 | #else |
83 | #define xlog_verify_dest_ptr(a,b) | |
3f336c6f | 84 | #define xlog_verify_grant_tail(a) |
abca1f33 | 85 | #define xlog_verify_iclog(a,b,c) |
9d110014 | 86 | #define xlog_verify_tail_lsn(a,b) |
1da177e4 LT |
87 | #endif |
88 | ||
9a8d2fdb MT |
89 | STATIC int |
90 | xlog_iclogs_empty( | |
91 | struct xlog *log); | |
1da177e4 | 92 | |
303591a0 BF |
93 | static int |
94 | xfs_log_cover(struct xfs_mount *); | |
95 | ||
dd954c69 | 96 | static void |
663e496a | 97 | xlog_grant_sub_space( |
ad223e60 MT |
98 | struct xlog *log, |
99 | atomic64_t *head, | |
100 | int bytes) | |
dd954c69 | 101 | { |
d0eb2f38 DC |
102 | int64_t head_val = atomic64_read(head); |
103 | int64_t new, old; | |
a69ed03c | 104 | |
d0eb2f38 DC |
105 | do { |
106 | int cycle, space; | |
a69ed03c | 107 | |
d0eb2f38 | 108 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 109 | |
d0eb2f38 DC |
110 | space -= bytes; |
111 | if (space < 0) { | |
112 | space += log->l_logsize; | |
113 | cycle--; | |
114 | } | |
115 | ||
116 | old = head_val; | |
117 | new = xlog_assign_grant_head_val(cycle, space); | |
118 | head_val = atomic64_cmpxchg(head, old, new); | |
119 | } while (head_val != old); | |
dd954c69 CH |
120 | } |
121 | ||
122 | static void | |
663e496a | 123 | xlog_grant_add_space( |
ad223e60 MT |
124 | struct xlog *log, |
125 | atomic64_t *head, | |
126 | int bytes) | |
dd954c69 | 127 | { |
d0eb2f38 DC |
128 | int64_t head_val = atomic64_read(head); |
129 | int64_t new, old; | |
a69ed03c | 130 | |
d0eb2f38 DC |
131 | do { |
132 | int tmp; | |
133 | int cycle, space; | |
a69ed03c | 134 | |
d0eb2f38 | 135 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 136 | |
d0eb2f38 DC |
137 | tmp = log->l_logsize - space; |
138 | if (tmp > bytes) | |
139 | space += bytes; | |
140 | else { | |
141 | space = bytes - tmp; | |
142 | cycle++; | |
143 | } | |
144 | ||
145 | old = head_val; | |
146 | new = xlog_assign_grant_head_val(cycle, space); | |
147 | head_val = atomic64_cmpxchg(head, old, new); | |
148 | } while (head_val != old); | |
dd954c69 | 149 | } |
a69ed03c | 150 | |
c303c5b8 CH |
151 | STATIC void |
152 | xlog_grant_head_init( | |
153 | struct xlog_grant_head *head) | |
154 | { | |
155 | xlog_assign_grant_head(&head->grant, 1, 0); | |
156 | INIT_LIST_HEAD(&head->waiters); | |
157 | spin_lock_init(&head->lock); | |
158 | } | |
159 | ||
a79bf2d7 CH |
160 | STATIC void |
161 | xlog_grant_head_wake_all( | |
162 | struct xlog_grant_head *head) | |
163 | { | |
164 | struct xlog_ticket *tic; | |
165 | ||
166 | spin_lock(&head->lock); | |
167 | list_for_each_entry(tic, &head->waiters, t_queue) | |
168 | wake_up_process(tic->t_task); | |
169 | spin_unlock(&head->lock); | |
170 | } | |
171 | ||
e179840d CH |
172 | static inline int |
173 | xlog_ticket_reservation( | |
ad223e60 | 174 | struct xlog *log, |
e179840d CH |
175 | struct xlog_grant_head *head, |
176 | struct xlog_ticket *tic) | |
9f9c19ec | 177 | { |
e179840d CH |
178 | if (head == &log->l_write_head) { |
179 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
180 | return tic->t_unit_res; | |
181 | } else { | |
9f9c19ec | 182 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) |
e179840d | 183 | return tic->t_unit_res * tic->t_cnt; |
9f9c19ec | 184 | else |
e179840d | 185 | return tic->t_unit_res; |
9f9c19ec | 186 | } |
9f9c19ec CH |
187 | } |
188 | ||
189 | STATIC bool | |
e179840d | 190 | xlog_grant_head_wake( |
ad223e60 | 191 | struct xlog *log, |
e179840d | 192 | struct xlog_grant_head *head, |
9f9c19ec CH |
193 | int *free_bytes) |
194 | { | |
195 | struct xlog_ticket *tic; | |
196 | int need_bytes; | |
7c107afb | 197 | bool woken_task = false; |
9f9c19ec | 198 | |
e179840d | 199 | list_for_each_entry(tic, &head->waiters, t_queue) { |
7c107afb DC |
200 | |
201 | /* | |
202 | * There is a chance that the size of the CIL checkpoints in | |
203 | * progress at the last AIL push target calculation resulted in | |
204 | * limiting the target to the log head (l_last_sync_lsn) at the | |
205 | * time. This may not reflect where the log head is now as the | |
206 | * CIL checkpoints may have completed. | |
207 | * | |
208 | * Hence when we are woken here, it may be that the head of the | |
209 | * log that has moved rather than the tail. As the tail didn't | |
210 | * move, there still won't be space available for the | |
211 | * reservation we require. However, if the AIL has already | |
212 | * pushed to the target defined by the old log head location, we | |
213 | * will hang here waiting for something else to update the AIL | |
214 | * push target. | |
215 | * | |
216 | * Therefore, if there isn't space to wake the first waiter on | |
217 | * the grant head, we need to push the AIL again to ensure the | |
218 | * target reflects both the current log tail and log head | |
219 | * position before we wait for the tail to move again. | |
220 | */ | |
221 | ||
e179840d | 222 | need_bytes = xlog_ticket_reservation(log, head, tic); |
7c107afb DC |
223 | if (*free_bytes < need_bytes) { |
224 | if (!woken_task) | |
225 | xlog_grant_push_ail(log, need_bytes); | |
9f9c19ec | 226 | return false; |
7c107afb | 227 | } |
9f9c19ec | 228 | |
e179840d CH |
229 | *free_bytes -= need_bytes; |
230 | trace_xfs_log_grant_wake_up(log, tic); | |
14a7235f | 231 | wake_up_process(tic->t_task); |
7c107afb | 232 | woken_task = true; |
9f9c19ec CH |
233 | } |
234 | ||
235 | return true; | |
236 | } | |
237 | ||
238 | STATIC int | |
23ee3df3 | 239 | xlog_grant_head_wait( |
ad223e60 | 240 | struct xlog *log, |
23ee3df3 | 241 | struct xlog_grant_head *head, |
9f9c19ec | 242 | struct xlog_ticket *tic, |
a30b0367 DC |
243 | int need_bytes) __releases(&head->lock) |
244 | __acquires(&head->lock) | |
9f9c19ec | 245 | { |
23ee3df3 | 246 | list_add_tail(&tic->t_queue, &head->waiters); |
9f9c19ec CH |
247 | |
248 | do { | |
249 | if (XLOG_FORCED_SHUTDOWN(log)) | |
250 | goto shutdown; | |
251 | xlog_grant_push_ail(log, need_bytes); | |
252 | ||
14a7235f | 253 | __set_current_state(TASK_UNINTERRUPTIBLE); |
23ee3df3 | 254 | spin_unlock(&head->lock); |
14a7235f | 255 | |
ff6d6af2 | 256 | XFS_STATS_INC(log->l_mp, xs_sleep_logspace); |
9f9c19ec | 257 | |
14a7235f CH |
258 | trace_xfs_log_grant_sleep(log, tic); |
259 | schedule(); | |
9f9c19ec CH |
260 | trace_xfs_log_grant_wake(log, tic); |
261 | ||
23ee3df3 | 262 | spin_lock(&head->lock); |
9f9c19ec CH |
263 | if (XLOG_FORCED_SHUTDOWN(log)) |
264 | goto shutdown; | |
23ee3df3 | 265 | } while (xlog_space_left(log, &head->grant) < need_bytes); |
9f9c19ec CH |
266 | |
267 | list_del_init(&tic->t_queue); | |
268 | return 0; | |
269 | shutdown: | |
270 | list_del_init(&tic->t_queue); | |
2451337d | 271 | return -EIO; |
9f9c19ec CH |
272 | } |
273 | ||
42ceedb3 CH |
274 | /* |
275 | * Atomically get the log space required for a log ticket. | |
276 | * | |
277 | * Once a ticket gets put onto head->waiters, it will only return after the | |
278 | * needed reservation is satisfied. | |
279 | * | |
280 | * This function is structured so that it has a lock free fast path. This is | |
281 | * necessary because every new transaction reservation will come through this | |
282 | * path. Hence any lock will be globally hot if we take it unconditionally on | |
283 | * every pass. | |
284 | * | |
285 | * As tickets are only ever moved on and off head->waiters under head->lock, we | |
286 | * only need to take that lock if we are going to add the ticket to the queue | |
287 | * and sleep. We can avoid taking the lock if the ticket was never added to | |
288 | * head->waiters because the t_queue list head will be empty and we hold the | |
289 | * only reference to it so it can safely be checked unlocked. | |
290 | */ | |
291 | STATIC int | |
292 | xlog_grant_head_check( | |
ad223e60 | 293 | struct xlog *log, |
42ceedb3 CH |
294 | struct xlog_grant_head *head, |
295 | struct xlog_ticket *tic, | |
296 | int *need_bytes) | |
297 | { | |
298 | int free_bytes; | |
299 | int error = 0; | |
300 | ||
301 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); | |
302 | ||
303 | /* | |
304 | * If there are other waiters on the queue then give them a chance at | |
305 | * logspace before us. Wake up the first waiters, if we do not wake | |
306 | * up all the waiters then go to sleep waiting for more free space, | |
307 | * otherwise try to get some space for this transaction. | |
308 | */ | |
309 | *need_bytes = xlog_ticket_reservation(log, head, tic); | |
310 | free_bytes = xlog_space_left(log, &head->grant); | |
311 | if (!list_empty_careful(&head->waiters)) { | |
312 | spin_lock(&head->lock); | |
313 | if (!xlog_grant_head_wake(log, head, &free_bytes) || | |
314 | free_bytes < *need_bytes) { | |
315 | error = xlog_grant_head_wait(log, head, tic, | |
316 | *need_bytes); | |
317 | } | |
318 | spin_unlock(&head->lock); | |
319 | } else if (free_bytes < *need_bytes) { | |
320 | spin_lock(&head->lock); | |
321 | error = xlog_grant_head_wait(log, head, tic, *need_bytes); | |
322 | spin_unlock(&head->lock); | |
323 | } | |
324 | ||
325 | return error; | |
326 | } | |
327 | ||
0adba536 CH |
328 | static void |
329 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
330 | { | |
331 | tic->t_res_num = 0; | |
332 | tic->t_res_arr_sum = 0; | |
333 | tic->t_res_num_ophdrs = 0; | |
334 | } | |
335 | ||
336 | static void | |
337 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
338 | { | |
339 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
340 | /* add to overflow and start again */ | |
341 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
342 | tic->t_res_num = 0; | |
343 | tic->t_res_arr_sum = 0; | |
344 | } | |
345 | ||
346 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
347 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
348 | tic->t_res_arr_sum += len; | |
349 | tic->t_res_num++; | |
350 | } | |
dd954c69 | 351 | |
50d25484 BF |
352 | bool |
353 | xfs_log_writable( | |
354 | struct xfs_mount *mp) | |
355 | { | |
356 | /* | |
8e9800f9 DW |
357 | * Do not write to the log on norecovery mounts, if the data or log |
358 | * devices are read-only, or if the filesystem is shutdown. Read-only | |
359 | * mounts allow internal writes for log recovery and unmount purposes, | |
360 | * so don't restrict that case. | |
50d25484 BF |
361 | */ |
362 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
363 | return false; | |
8e9800f9 DW |
364 | if (xfs_readonly_buftarg(mp->m_ddev_targp)) |
365 | return false; | |
50d25484 BF |
366 | if (xfs_readonly_buftarg(mp->m_log->l_targ)) |
367 | return false; | |
368 | if (XFS_FORCED_SHUTDOWN(mp)) | |
369 | return false; | |
370 | return true; | |
371 | } | |
372 | ||
9006fb91 CH |
373 | /* |
374 | * Replenish the byte reservation required by moving the grant write head. | |
375 | */ | |
376 | int | |
377 | xfs_log_regrant( | |
378 | struct xfs_mount *mp, | |
379 | struct xlog_ticket *tic) | |
380 | { | |
ad223e60 | 381 | struct xlog *log = mp->m_log; |
9006fb91 CH |
382 | int need_bytes; |
383 | int error = 0; | |
384 | ||
385 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 386 | return -EIO; |
9006fb91 | 387 | |
ff6d6af2 | 388 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
389 | |
390 | /* | |
391 | * This is a new transaction on the ticket, so we need to change the | |
392 | * transaction ID so that the next transaction has a different TID in | |
393 | * the log. Just add one to the existing tid so that we can see chains | |
394 | * of rolling transactions in the log easily. | |
395 | */ | |
396 | tic->t_tid++; | |
397 | ||
398 | xlog_grant_push_ail(log, tic->t_unit_res); | |
399 | ||
400 | tic->t_curr_res = tic->t_unit_res; | |
401 | xlog_tic_reset_res(tic); | |
402 | ||
403 | if (tic->t_cnt > 0) | |
404 | return 0; | |
405 | ||
406 | trace_xfs_log_regrant(log, tic); | |
407 | ||
408 | error = xlog_grant_head_check(log, &log->l_write_head, tic, | |
409 | &need_bytes); | |
410 | if (error) | |
411 | goto out_error; | |
412 | ||
413 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
414 | trace_xfs_log_regrant_exit(log, tic); | |
415 | xlog_verify_grant_tail(log); | |
416 | return 0; | |
417 | ||
418 | out_error: | |
419 | /* | |
420 | * If we are failing, make sure the ticket doesn't have any current | |
421 | * reservations. We don't want to add this back when the ticket/ | |
422 | * transaction gets cancelled. | |
423 | */ | |
424 | tic->t_curr_res = 0; | |
425 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
426 | return error; | |
427 | } | |
428 | ||
429 | /* | |
a0e336ba | 430 | * Reserve log space and return a ticket corresponding to the reservation. |
9006fb91 CH |
431 | * |
432 | * Each reservation is going to reserve extra space for a log record header. | |
433 | * When writes happen to the on-disk log, we don't subtract the length of the | |
434 | * log record header from any reservation. By wasting space in each | |
435 | * reservation, we prevent over allocation problems. | |
436 | */ | |
437 | int | |
438 | xfs_log_reserve( | |
439 | struct xfs_mount *mp, | |
440 | int unit_bytes, | |
441 | int cnt, | |
442 | struct xlog_ticket **ticp, | |
c8ce540d | 443 | uint8_t client, |
710b1e2c | 444 | bool permanent) |
9006fb91 | 445 | { |
ad223e60 | 446 | struct xlog *log = mp->m_log; |
9006fb91 CH |
447 | struct xlog_ticket *tic; |
448 | int need_bytes; | |
449 | int error = 0; | |
450 | ||
451 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); | |
452 | ||
453 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 454 | return -EIO; |
9006fb91 | 455 | |
ff6d6af2 | 456 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
457 | |
458 | ASSERT(*ticp == NULL); | |
ca4f2589 | 459 | tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent); |
9006fb91 CH |
460 | *ticp = tic; |
461 | ||
437a255a DC |
462 | xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt |
463 | : tic->t_unit_res); | |
9006fb91 CH |
464 | |
465 | trace_xfs_log_reserve(log, tic); | |
466 | ||
467 | error = xlog_grant_head_check(log, &log->l_reserve_head, tic, | |
468 | &need_bytes); | |
469 | if (error) | |
470 | goto out_error; | |
471 | ||
472 | xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); | |
473 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
474 | trace_xfs_log_reserve_exit(log, tic); | |
475 | xlog_verify_grant_tail(log); | |
476 | return 0; | |
477 | ||
478 | out_error: | |
479 | /* | |
480 | * If we are failing, make sure the ticket doesn't have any current | |
481 | * reservations. We don't want to add this back when the ticket/ | |
482 | * transaction gets cancelled. | |
483 | */ | |
484 | tic->t_curr_res = 0; | |
485 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
486 | return error; | |
487 | } | |
488 | ||
df732b29 CH |
489 | /* |
490 | * Flush iclog to disk if this is the last reference to the given iclog and the | |
9d110014 DC |
491 | * it is in the WANT_SYNC state. |
492 | * | |
493 | * If the caller passes in a non-zero @old_tail_lsn and the current log tail | |
494 | * does not match, there may be metadata on disk that must be persisted before | |
495 | * this iclog is written. To satisfy that requirement, set the | |
496 | * XLOG_ICL_NEED_FLUSH flag as a condition for writing this iclog with the new | |
497 | * log tail value. | |
498 | * | |
499 | * If XLOG_ICL_NEED_FUA is already set on the iclog, we need to ensure that the | |
500 | * log tail is updated correctly. NEED_FUA indicates that the iclog will be | |
501 | * written to stable storage, and implies that a commit record is contained | |
502 | * within the iclog. We need to ensure that the log tail does not move beyond | |
503 | * the tail that the first commit record in the iclog ordered against, otherwise | |
504 | * correct recovery of that checkpoint becomes dependent on future operations | |
505 | * performed on this iclog. | |
506 | * | |
507 | * Hence if NEED_FUA is set and the current iclog tail lsn is empty, write the | |
508 | * current tail into iclog. Once the iclog tail is set, future operations must | |
509 | * not modify it, otherwise they potentially violate ordering constraints for | |
510 | * the checkpoint commit that wrote the initial tail lsn value. The tail lsn in | |
511 | * the iclog will get zeroed on activation of the iclog after sync, so we | |
512 | * always capture the tail lsn on the iclog on the first NEED_FUA release | |
513 | * regardless of the number of active reference counts on this iclog. | |
df732b29 | 514 | */ |
9d110014 | 515 | |
eef983ff | 516 | int |
df732b29 CH |
517 | xlog_state_release_iclog( |
518 | struct xlog *log, | |
0dc8f7f1 DC |
519 | struct xlog_in_core *iclog, |
520 | xfs_lsn_t old_tail_lsn) | |
df732b29 | 521 | { |
9d392064 | 522 | xfs_lsn_t tail_lsn; |
df732b29 CH |
523 | lockdep_assert_held(&log->l_icloglock); |
524 | ||
956f6daa | 525 | trace_xlog_iclog_release(iclog, _RET_IP_); |
1858bb0b | 526 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
df732b29 CH |
527 | return -EIO; |
528 | ||
0dc8f7f1 DC |
529 | /* |
530 | * Grabbing the current log tail needs to be atomic w.r.t. the writing | |
531 | * of the tail LSN into the iclog so we guarantee that the log tail does | |
532 | * not move between deciding if a cache flush is required and writing | |
533 | * the LSN into the iclog below. | |
534 | */ | |
535 | if (old_tail_lsn || iclog->ic_state == XLOG_STATE_WANT_SYNC) { | |
536 | tail_lsn = xlog_assign_tail_lsn(log->l_mp); | |
537 | ||
538 | if (old_tail_lsn && tail_lsn != old_tail_lsn) | |
539 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH; | |
9d110014 DC |
540 | |
541 | if ((iclog->ic_flags & XLOG_ICL_NEED_FUA) && | |
542 | !iclog->ic_header.h_tail_lsn) | |
543 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); | |
0dc8f7f1 DC |
544 | } |
545 | ||
9d392064 DC |
546 | if (!atomic_dec_and_test(&iclog->ic_refcnt)) |
547 | return 0; | |
548 | ||
549 | if (iclog->ic_state != XLOG_STATE_WANT_SYNC) { | |
550 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
551 | return 0; | |
df732b29 CH |
552 | } |
553 | ||
9d392064 | 554 | iclog->ic_state = XLOG_STATE_SYNCING; |
9d110014 DC |
555 | if (!iclog->ic_header.h_tail_lsn) |
556 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); | |
557 | xlog_verify_tail_lsn(log, iclog); | |
9d392064 DC |
558 | trace_xlog_iclog_syncing(iclog, _RET_IP_); |
559 | ||
560 | spin_unlock(&log->l_icloglock); | |
561 | xlog_sync(log, iclog); | |
562 | spin_lock(&log->l_icloglock); | |
df732b29 CH |
563 | return 0; |
564 | } | |
565 | ||
1da177e4 LT |
566 | /* |
567 | * Mount a log filesystem | |
568 | * | |
569 | * mp - ubiquitous xfs mount point structure | |
570 | * log_target - buftarg of on-disk log device | |
571 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
572 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
573 | * | |
574 | * Return error or zero. | |
575 | */ | |
576 | int | |
249a8c11 DC |
577 | xfs_log_mount( |
578 | xfs_mount_t *mp, | |
579 | xfs_buftarg_t *log_target, | |
580 | xfs_daddr_t blk_offset, | |
581 | int num_bblks) | |
1da177e4 | 582 | { |
9c92ee20 | 583 | bool fatal = xfs_sb_version_hascrc(&mp->m_sb); |
3e7b91cf JL |
584 | int error = 0; |
585 | int min_logfsbs; | |
249a8c11 | 586 | |
c99d609a DC |
587 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
588 | xfs_notice(mp, "Mounting V%d Filesystem", | |
589 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
590 | } else { | |
a0fa2b67 | 591 | xfs_notice(mp, |
c99d609a DC |
592 | "Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.", |
593 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
bd186aa9 | 594 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
595 | } |
596 | ||
597 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e | 598 | if (IS_ERR(mp->m_log)) { |
2451337d | 599 | error = PTR_ERR(mp->m_log); |
644c3567 DC |
600 | goto out; |
601 | } | |
1da177e4 | 602 | |
3e7b91cf JL |
603 | /* |
604 | * Validate the given log space and drop a critical message via syslog | |
605 | * if the log size is too small that would lead to some unexpected | |
606 | * situations in transaction log space reservation stage. | |
607 | * | |
608 | * Note: we can't just reject the mount if the validation fails. This | |
609 | * would mean that people would have to downgrade their kernel just to | |
610 | * remedy the situation as there is no way to grow the log (short of | |
611 | * black magic surgery with xfs_db). | |
612 | * | |
613 | * We can, however, reject mounts for CRC format filesystems, as the | |
614 | * mkfs binary being used to make the filesystem should never create a | |
615 | * filesystem with a log that is too small. | |
616 | */ | |
617 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
618 | ||
619 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
620 | xfs_warn(mp, | |
621 | "Log size %d blocks too small, minimum size is %d blocks", | |
622 | mp->m_sb.sb_logblocks, min_logfsbs); | |
2451337d | 623 | error = -EINVAL; |
3e7b91cf JL |
624 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { |
625 | xfs_warn(mp, | |
626 | "Log size %d blocks too large, maximum size is %lld blocks", | |
627 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
2451337d | 628 | error = -EINVAL; |
3e7b91cf JL |
629 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { |
630 | xfs_warn(mp, | |
631 | "log size %lld bytes too large, maximum size is %lld bytes", | |
632 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
633 | XFS_MAX_LOG_BYTES); | |
2451337d | 634 | error = -EINVAL; |
9c92ee20 DW |
635 | } else if (mp->m_sb.sb_logsunit > 1 && |
636 | mp->m_sb.sb_logsunit % mp->m_sb.sb_blocksize) { | |
637 | xfs_warn(mp, | |
638 | "log stripe unit %u bytes must be a multiple of block size", | |
639 | mp->m_sb.sb_logsunit); | |
640 | error = -EINVAL; | |
641 | fatal = true; | |
3e7b91cf JL |
642 | } |
643 | if (error) { | |
9c92ee20 DW |
644 | /* |
645 | * Log check errors are always fatal on v5; or whenever bad | |
646 | * metadata leads to a crash. | |
647 | */ | |
648 | if (fatal) { | |
3e7b91cf JL |
649 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); |
650 | ASSERT(0); | |
651 | goto out_free_log; | |
652 | } | |
f41febd2 | 653 | xfs_crit(mp, "Log size out of supported range."); |
3e7b91cf | 654 | xfs_crit(mp, |
f41febd2 | 655 | "Continuing onwards, but if log hangs are experienced then please report this message in the bug report."); |
3e7b91cf JL |
656 | } |
657 | ||
249a8c11 DC |
658 | /* |
659 | * Initialize the AIL now we have a log. | |
660 | */ | |
249a8c11 DC |
661 | error = xfs_trans_ail_init(mp); |
662 | if (error) { | |
a0fa2b67 | 663 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 664 | goto out_free_log; |
249a8c11 | 665 | } |
a9c21c1b | 666 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 667 | |
1da177e4 LT |
668 | /* |
669 | * skip log recovery on a norecovery mount. pretend it all | |
670 | * just worked. | |
671 | */ | |
672 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 673 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
674 | |
675 | if (readonly) | |
bd186aa9 | 676 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 677 | |
65be6054 | 678 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
679 | |
680 | if (readonly) | |
bd186aa9 | 681 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 682 | if (error) { |
a0fa2b67 DC |
683 | xfs_warn(mp, "log mount/recovery failed: error %d", |
684 | error); | |
f0b2efad | 685 | xlog_recover_cancel(mp->m_log); |
26430752 | 686 | goto out_destroy_ail; |
1da177e4 LT |
687 | } |
688 | } | |
689 | ||
baff4e44 BF |
690 | error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj, |
691 | "log"); | |
692 | if (error) | |
693 | goto out_destroy_ail; | |
694 | ||
1da177e4 LT |
695 | /* Normal transactions can now occur */ |
696 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
697 | ||
71e330b5 DC |
698 | /* |
699 | * Now the log has been fully initialised and we know were our | |
700 | * space grant counters are, we can initialise the permanent ticket | |
701 | * needed for delayed logging to work. | |
702 | */ | |
703 | xlog_cil_init_post_recovery(mp->m_log); | |
704 | ||
1da177e4 | 705 | return 0; |
26430752 CH |
706 | |
707 | out_destroy_ail: | |
708 | xfs_trans_ail_destroy(mp); | |
709 | out_free_log: | |
710 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 711 | out: |
249a8c11 | 712 | return error; |
26430752 | 713 | } |
1da177e4 LT |
714 | |
715 | /* | |
f661f1e0 DC |
716 | * Finish the recovery of the file system. This is separate from the |
717 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
718 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
719 | * here. | |
1da177e4 | 720 | * |
f661f1e0 DC |
721 | * If we finish recovery successfully, start the background log work. If we are |
722 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
723 | * it. | |
1da177e4 LT |
724 | */ |
725 | int | |
f0b2efad BF |
726 | xfs_log_mount_finish( |
727 | struct xfs_mount *mp) | |
1da177e4 | 728 | { |
f661f1e0 | 729 | int error = 0; |
6f4a1eef | 730 | bool readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
f1b92bbc | 731 | bool recovered = mp->m_log->l_flags & XLOG_RECOVERY_NEEDED; |
1da177e4 | 732 | |
f0b2efad | 733 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) { |
bd186aa9 | 734 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
f0b2efad | 735 | return 0; |
6f4a1eef ES |
736 | } else if (readonly) { |
737 | /* Allow unlinked processing to proceed */ | |
738 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
1da177e4 LT |
739 | } |
740 | ||
8204f8dd DW |
741 | /* |
742 | * During the second phase of log recovery, we need iget and | |
743 | * iput to behave like they do for an active filesystem. | |
744 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
745 | * of inodes before we're done replaying log items on those | |
746 | * inodes. Turn it off immediately after recovery finishes | |
747 | * so that we don't leak the quota inodes if subsequent mount | |
748 | * activities fail. | |
799ea9e9 DW |
749 | * |
750 | * We let all inodes involved in redo item processing end up on | |
751 | * the LRU instead of being evicted immediately so that if we do | |
752 | * something to an unlinked inode, the irele won't cause | |
753 | * premature truncation and freeing of the inode, which results | |
754 | * in log recovery failure. We have to evict the unreferenced | |
1751e8a6 | 755 | * lru inodes after clearing SB_ACTIVE because we don't |
799ea9e9 DW |
756 | * otherwise clean up the lru if there's a subsequent failure in |
757 | * xfs_mountfs, which leads to us leaking the inodes if nothing | |
758 | * else (e.g. quotacheck) references the inodes before the | |
759 | * mount failure occurs. | |
8204f8dd | 760 | */ |
1751e8a6 | 761 | mp->m_super->s_flags |= SB_ACTIVE; |
f0b2efad BF |
762 | error = xlog_recover_finish(mp->m_log); |
763 | if (!error) | |
764 | xfs_log_work_queue(mp); | |
1751e8a6 | 765 | mp->m_super->s_flags &= ~SB_ACTIVE; |
799ea9e9 | 766 | evict_inodes(mp->m_super); |
f0b2efad | 767 | |
f1b92bbc BF |
768 | /* |
769 | * Drain the buffer LRU after log recovery. This is required for v4 | |
770 | * filesystems to avoid leaving around buffers with NULL verifier ops, | |
771 | * but we do it unconditionally to make sure we're always in a clean | |
772 | * cache state after mount. | |
773 | * | |
774 | * Don't push in the error case because the AIL may have pending intents | |
775 | * that aren't removed until recovery is cancelled. | |
776 | */ | |
777 | if (!error && recovered) { | |
778 | xfs_log_force(mp, XFS_LOG_SYNC); | |
779 | xfs_ail_push_all_sync(mp->m_ail); | |
780 | } | |
10fb9ac1 | 781 | xfs_buftarg_drain(mp->m_ddev_targp); |
f1b92bbc | 782 | |
6f4a1eef ES |
783 | if (readonly) |
784 | mp->m_flags |= XFS_MOUNT_RDONLY; | |
785 | ||
4e6b8270 DW |
786 | /* Make sure the log is dead if we're returning failure. */ |
787 | ASSERT(!error || (mp->m_log->l_flags & XLOG_IO_ERROR)); | |
788 | ||
f0b2efad BF |
789 | return error; |
790 | } | |
791 | ||
792 | /* | |
793 | * The mount has failed. Cancel the recovery if it hasn't completed and destroy | |
794 | * the log. | |
795 | */ | |
a7a9250e | 796 | void |
f0b2efad BF |
797 | xfs_log_mount_cancel( |
798 | struct xfs_mount *mp) | |
799 | { | |
a7a9250e | 800 | xlog_recover_cancel(mp->m_log); |
f0b2efad | 801 | xfs_log_unmount(mp); |
1da177e4 LT |
802 | } |
803 | ||
45eddb41 DC |
804 | /* |
805 | * Flush out the iclog to disk ensuring that device caches are flushed and | |
806 | * the iclog hits stable storage before any completion waiters are woken. | |
807 | */ | |
808 | static inline int | |
809 | xlog_force_iclog( | |
810 | struct xlog_in_core *iclog) | |
811 | { | |
812 | atomic_inc(&iclog->ic_refcnt); | |
2bf1ec0f | 813 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA; |
45eddb41 DC |
814 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
815 | xlog_state_switch_iclogs(iclog->ic_log, iclog, 0); | |
816 | return xlog_state_release_iclog(iclog->ic_log, iclog, 0); | |
817 | } | |
818 | ||
81e5b50a | 819 | /* |
a79b28c2 DC |
820 | * Wait for the iclog and all prior iclogs to be written disk as required by the |
821 | * log force state machine. Waiting on ic_force_wait ensures iclog completions | |
822 | * have been ordered and callbacks run before we are woken here, hence | |
823 | * guaranteeing that all the iclogs up to this one are on stable storage. | |
81e5b50a | 824 | */ |
a79b28c2 | 825 | int |
81e5b50a CH |
826 | xlog_wait_on_iclog( |
827 | struct xlog_in_core *iclog) | |
828 | __releases(iclog->ic_log->l_icloglock) | |
829 | { | |
830 | struct xlog *log = iclog->ic_log; | |
831 | ||
956f6daa | 832 | trace_xlog_iclog_wait_on(iclog, _RET_IP_); |
81e5b50a CH |
833 | if (!XLOG_FORCED_SHUTDOWN(log) && |
834 | iclog->ic_state != XLOG_STATE_ACTIVE && | |
835 | iclog->ic_state != XLOG_STATE_DIRTY) { | |
836 | XFS_STATS_INC(log->l_mp, xs_log_force_sleep); | |
837 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
838 | } else { | |
839 | spin_unlock(&log->l_icloglock); | |
840 | } | |
841 | ||
842 | if (XLOG_FORCED_SHUTDOWN(log)) | |
843 | return -EIO; | |
844 | return 0; | |
845 | } | |
846 | ||
1da177e4 | 847 | /* |
3c702f95 DC |
848 | * Write out an unmount record using the ticket provided. We have to account for |
849 | * the data space used in the unmount ticket as this write is not done from a | |
850 | * transaction context that has already done the accounting for us. | |
1da177e4 | 851 | */ |
3c702f95 DC |
852 | static int |
853 | xlog_write_unmount_record( | |
854 | struct xlog *log, | |
3468bb1c | 855 | struct xlog_ticket *ticket) |
53235f22 | 856 | { |
3c702f95 | 857 | struct xfs_unmount_log_format ulf = { |
53235f22 DW |
858 | .magic = XLOG_UNMOUNT_TYPE, |
859 | }; | |
860 | struct xfs_log_iovec reg = { | |
3c702f95 DC |
861 | .i_addr = &ulf, |
862 | .i_len = sizeof(ulf), | |
53235f22 DW |
863 | .i_type = XLOG_REG_TYPE_UNMOUNT, |
864 | }; | |
865 | struct xfs_log_vec vec = { | |
866 | .lv_niovecs = 1, | |
867 | .lv_iovecp = ®, | |
868 | }; | |
3c702f95 DC |
869 | |
870 | /* account for space used by record data */ | |
871 | ticket->t_curr_res -= sizeof(ulf); | |
eef983ff | 872 | |
3468bb1c | 873 | return xlog_write(log, &vec, ticket, NULL, NULL, XLOG_UNMOUNT_TRANS); |
3c702f95 DC |
874 | } |
875 | ||
876 | /* | |
877 | * Mark the filesystem clean by writing an unmount record to the head of the | |
878 | * log. | |
879 | */ | |
880 | static void | |
881 | xlog_unmount_write( | |
882 | struct xlog *log) | |
883 | { | |
884 | struct xfs_mount *mp = log->l_mp; | |
53235f22 DW |
885 | struct xlog_in_core *iclog; |
886 | struct xlog_ticket *tic = NULL; | |
53235f22 DW |
887 | int error; |
888 | ||
889 | error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0); | |
890 | if (error) | |
891 | goto out_err; | |
892 | ||
3468bb1c | 893 | error = xlog_write_unmount_record(log, tic); |
53235f22 DW |
894 | /* |
895 | * At this point, we're umounting anyway, so there's no point in | |
896 | * transitioning log state to IOERROR. Just continue... | |
897 | */ | |
898 | out_err: | |
899 | if (error) | |
900 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
901 | ||
902 | spin_lock(&log->l_icloglock); | |
903 | iclog = log->l_iclog; | |
45eddb41 | 904 | error = xlog_force_iclog(iclog); |
81e5b50a | 905 | xlog_wait_on_iclog(iclog); |
53235f22 DW |
906 | |
907 | if (tic) { | |
908 | trace_xfs_log_umount_write(log, tic); | |
8b41e3f9 | 909 | xfs_log_ticket_ungrant(log, tic); |
53235f22 DW |
910 | } |
911 | } | |
912 | ||
13859c98 CH |
913 | static void |
914 | xfs_log_unmount_verify_iclog( | |
915 | struct xlog *log) | |
916 | { | |
917 | struct xlog_in_core *iclog = log->l_iclog; | |
918 | ||
919 | do { | |
920 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
921 | ASSERT(iclog->ic_offset == 0); | |
922 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
923 | } | |
924 | ||
1da177e4 LT |
925 | /* |
926 | * Unmount record used to have a string "Unmount filesystem--" in the | |
927 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
928 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 929 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
930 | * As far as I know, there weren't any dependencies on the old behaviour. |
931 | */ | |
550319e9 | 932 | static void |
13859c98 CH |
933 | xfs_log_unmount_write( |
934 | struct xfs_mount *mp) | |
1da177e4 | 935 | { |
13859c98 | 936 | struct xlog *log = mp->m_log; |
1da177e4 | 937 | |
50d25484 | 938 | if (!xfs_log_writable(mp)) |
550319e9 | 939 | return; |
1da177e4 | 940 | |
550319e9 | 941 | xfs_log_force(mp, XFS_LOG_SYNC); |
1da177e4 | 942 | |
6178d104 CH |
943 | if (XLOG_FORCED_SHUTDOWN(log)) |
944 | return; | |
5cc3c006 DW |
945 | |
946 | /* | |
947 | * If we think the summary counters are bad, avoid writing the unmount | |
948 | * record to force log recovery at next mount, after which the summary | |
949 | * counters will be recalculated. Refer to xlog_check_unmount_rec for | |
950 | * more details. | |
951 | */ | |
952 | if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp, | |
953 | XFS_ERRTAG_FORCE_SUMMARY_RECALC)) { | |
954 | xfs_alert(mp, "%s: will fix summary counters at next mount", | |
955 | __func__); | |
956 | return; | |
957 | } | |
958 | ||
13859c98 | 959 | xfs_log_unmount_verify_iclog(log); |
3c702f95 | 960 | xlog_unmount_write(log); |
550319e9 | 961 | } |
1da177e4 LT |
962 | |
963 | /* | |
c75921a7 | 964 | * Empty the log for unmount/freeze. |
cf2931db DC |
965 | * |
966 | * To do this, we first need to shut down the background log work so it is not | |
967 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
968 | * the log so we can then flush them out. Once they have completed their IO and | |
303591a0 | 969 | * run the callbacks removing themselves from the AIL, we can cover the log. |
1da177e4 | 970 | */ |
303591a0 | 971 | int |
c75921a7 DC |
972 | xfs_log_quiesce( |
973 | struct xfs_mount *mp) | |
1da177e4 | 974 | { |
908ce71e DW |
975 | /* |
976 | * Clear log incompat features since we're quiescing the log. Report | |
977 | * failures, though it's not fatal to have a higher log feature | |
978 | * protection level than the log contents actually require. | |
979 | */ | |
980 | if (xfs_clear_incompat_log_features(mp)) { | |
981 | int error; | |
982 | ||
983 | error = xfs_sync_sb(mp, false); | |
984 | if (error) | |
985 | xfs_warn(mp, | |
986 | "Failed to clear log incompat features on quiesce"); | |
987 | } | |
988 | ||
f661f1e0 | 989 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
990 | xfs_log_force(mp, XFS_LOG_SYNC); |
991 | ||
992 | /* | |
993 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
8321ddb2 | 994 | * will push it, xfs_buftarg_wait() will not wait for it. Further, |
cf2931db DC |
995 | * xfs_buf_iowait() cannot be used because it was pushed with the |
996 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
997 | * the IO to complete. | |
998 | */ | |
999 | xfs_ail_push_all_sync(mp->m_ail); | |
8321ddb2 | 1000 | xfs_buftarg_wait(mp->m_ddev_targp); |
cf2931db DC |
1001 | xfs_buf_lock(mp->m_sb_bp); |
1002 | xfs_buf_unlock(mp->m_sb_bp); | |
303591a0 BF |
1003 | |
1004 | return xfs_log_cover(mp); | |
9e54ee0f | 1005 | } |
cf2931db | 1006 | |
9e54ee0f BF |
1007 | void |
1008 | xfs_log_clean( | |
1009 | struct xfs_mount *mp) | |
1010 | { | |
1011 | xfs_log_quiesce(mp); | |
cf2931db | 1012 | xfs_log_unmount_write(mp); |
c75921a7 DC |
1013 | } |
1014 | ||
1015 | /* | |
1016 | * Shut down and release the AIL and Log. | |
1017 | * | |
1018 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
1019 | * from the AIL so that the log is empty before we write the unmount record to | |
1020 | * the log. Once this is done, we can tear down the AIL and the log. | |
1021 | */ | |
1022 | void | |
1023 | xfs_log_unmount( | |
1024 | struct xfs_mount *mp) | |
1025 | { | |
9e54ee0f | 1026 | xfs_log_clean(mp); |
cf2931db | 1027 | |
8321ddb2 BF |
1028 | xfs_buftarg_drain(mp->m_ddev_targp); |
1029 | ||
249a8c11 | 1030 | xfs_trans_ail_destroy(mp); |
baff4e44 BF |
1031 | |
1032 | xfs_sysfs_del(&mp->m_log->l_kobj); | |
1033 | ||
c41564b5 | 1034 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
1035 | } |
1036 | ||
43f5efc5 DC |
1037 | void |
1038 | xfs_log_item_init( | |
1039 | struct xfs_mount *mp, | |
1040 | struct xfs_log_item *item, | |
1041 | int type, | |
272e42b2 | 1042 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
1043 | { |
1044 | item->li_mountp = mp; | |
1045 | item->li_ailp = mp->m_ail; | |
1046 | item->li_type = type; | |
1047 | item->li_ops = ops; | |
71e330b5 DC |
1048 | item->li_lv = NULL; |
1049 | ||
1050 | INIT_LIST_HEAD(&item->li_ail); | |
1051 | INIT_LIST_HEAD(&item->li_cil); | |
643c8c05 | 1052 | INIT_LIST_HEAD(&item->li_bio_list); |
e6631f85 | 1053 | INIT_LIST_HEAD(&item->li_trans); |
43f5efc5 DC |
1054 | } |
1055 | ||
09a423a3 CH |
1056 | /* |
1057 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 1058 | */ |
1da177e4 | 1059 | void |
09a423a3 | 1060 | xfs_log_space_wake( |
cfb7cdca | 1061 | struct xfs_mount *mp) |
1da177e4 | 1062 | { |
ad223e60 | 1063 | struct xlog *log = mp->m_log; |
cfb7cdca | 1064 | int free_bytes; |
1da177e4 | 1065 | |
1da177e4 LT |
1066 | if (XLOG_FORCED_SHUTDOWN(log)) |
1067 | return; | |
1da177e4 | 1068 | |
28496968 | 1069 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
1070 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1071 | ||
28496968 CH |
1072 | spin_lock(&log->l_write_head.lock); |
1073 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 1074 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 1075 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 1076 | } |
10547941 | 1077 | |
28496968 | 1078 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
1079 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1080 | ||
28496968 CH |
1081 | spin_lock(&log->l_reserve_head.lock); |
1082 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 1083 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 1084 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 1085 | } |
3f16b985 | 1086 | } |
1da177e4 LT |
1087 | |
1088 | /* | |
2c6e24ce DC |
1089 | * Determine if we have a transaction that has gone to disk that needs to be |
1090 | * covered. To begin the transition to the idle state firstly the log needs to | |
1091 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1092 | * we start attempting to cover the log. | |
b6f8dd49 | 1093 | * |
2c6e24ce DC |
1094 | * Only if we are then in a state where covering is needed, the caller is |
1095 | * informed that dummy transactions are required to move the log into the idle | |
1096 | * state. | |
1097 | * | |
1098 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1099 | * cover the log as we may be in a situation where there isn't log space | |
1100 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1101 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1102 | * there's no point in running a dummy transaction at this point because we | |
1103 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 | 1104 | */ |
37444fc4 BF |
1105 | static bool |
1106 | xfs_log_need_covered( | |
1107 | struct xfs_mount *mp) | |
1da177e4 | 1108 | { |
37444fc4 BF |
1109 | struct xlog *log = mp->m_log; |
1110 | bool needed = false; | |
1da177e4 | 1111 | |
2c6e24ce | 1112 | if (!xlog_cil_empty(log)) |
8646b982 | 1113 | return false; |
2c6e24ce | 1114 | |
b22cd72c | 1115 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1116 | switch (log->l_covered_state) { |
1117 | case XLOG_STATE_COVER_DONE: | |
1118 | case XLOG_STATE_COVER_DONE2: | |
1119 | case XLOG_STATE_COVER_IDLE: | |
1120 | break; | |
1121 | case XLOG_STATE_COVER_NEED: | |
1122 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1123 | if (xfs_ail_min_lsn(log->l_ailp)) |
1124 | break; | |
1125 | if (!xlog_iclogs_empty(log)) | |
1126 | break; | |
1127 | ||
37444fc4 | 1128 | needed = true; |
2c6e24ce DC |
1129 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) |
1130 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1131 | else | |
1132 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1133 | break; | |
b6f8dd49 | 1134 | default: |
37444fc4 | 1135 | needed = true; |
b6f8dd49 | 1136 | break; |
1da177e4 | 1137 | } |
b22cd72c | 1138 | spin_unlock(&log->l_icloglock); |
014c2544 | 1139 | return needed; |
1da177e4 LT |
1140 | } |
1141 | ||
303591a0 BF |
1142 | /* |
1143 | * Explicitly cover the log. This is similar to background log covering but | |
1144 | * intended for usage in quiesce codepaths. The caller is responsible to ensure | |
1145 | * the log is idle and suitable for covering. The CIL, iclog buffers and AIL | |
1146 | * must all be empty. | |
1147 | */ | |
1148 | static int | |
1149 | xfs_log_cover( | |
1150 | struct xfs_mount *mp) | |
1151 | { | |
303591a0 | 1152 | int error = 0; |
f46e5a17 | 1153 | bool need_covered; |
303591a0 | 1154 | |
4533fc63 BF |
1155 | ASSERT((xlog_cil_empty(mp->m_log) && xlog_iclogs_empty(mp->m_log) && |
1156 | !xfs_ail_min_lsn(mp->m_log->l_ailp)) || | |
303591a0 BF |
1157 | XFS_FORCED_SHUTDOWN(mp)); |
1158 | ||
1159 | if (!xfs_log_writable(mp)) | |
1160 | return 0; | |
1161 | ||
f46e5a17 BF |
1162 | /* |
1163 | * xfs_log_need_covered() is not idempotent because it progresses the | |
1164 | * state machine if the log requires covering. Therefore, we must call | |
1165 | * this function once and use the result until we've issued an sb sync. | |
1166 | * Do so first to make that abundantly clear. | |
1167 | * | |
1168 | * Fall into the covering sequence if the log needs covering or the | |
1169 | * mount has lazy superblock accounting to sync to disk. The sb sync | |
1170 | * used for covering accumulates the in-core counters, so covering | |
1171 | * handles this for us. | |
1172 | */ | |
1173 | need_covered = xfs_log_need_covered(mp); | |
1174 | if (!need_covered && !xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1175 | return 0; | |
1176 | ||
303591a0 BF |
1177 | /* |
1178 | * To cover the log, commit the superblock twice (at most) in | |
1179 | * independent checkpoints. The first serves as a reference for the | |
1180 | * tail pointer. The sync transaction and AIL push empties the AIL and | |
1181 | * updates the in-core tail to the LSN of the first checkpoint. The | |
1182 | * second commit updates the on-disk tail with the in-core LSN, | |
1183 | * covering the log. Push the AIL one more time to leave it empty, as | |
1184 | * we found it. | |
1185 | */ | |
f46e5a17 | 1186 | do { |
303591a0 BF |
1187 | error = xfs_sync_sb(mp, true); |
1188 | if (error) | |
1189 | break; | |
1190 | xfs_ail_push_all_sync(mp->m_ail); | |
f46e5a17 | 1191 | } while (xfs_log_need_covered(mp)); |
303591a0 BF |
1192 | |
1193 | return error; | |
1194 | } | |
1195 | ||
09a423a3 | 1196 | /* |
1da177e4 LT |
1197 | * We may be holding the log iclog lock upon entering this routine. |
1198 | */ | |
1199 | xfs_lsn_t | |
1c304625 | 1200 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1201 | struct xfs_mount *mp) |
1da177e4 | 1202 | { |
ad223e60 | 1203 | struct xlog *log = mp->m_log; |
1c304625 CH |
1204 | struct xfs_log_item *lip; |
1205 | xfs_lsn_t tail_lsn; | |
1206 | ||
57e80956 | 1207 | assert_spin_locked(&mp->m_ail->ail_lock); |
1da177e4 | 1208 | |
09a423a3 CH |
1209 | /* |
1210 | * To make sure we always have a valid LSN for the log tail we keep | |
1211 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1212 | * and use that when the AIL was empty. |
09a423a3 | 1213 | */ |
1c304625 CH |
1214 | lip = xfs_ail_min(mp->m_ail); |
1215 | if (lip) | |
1216 | tail_lsn = lip->li_lsn; | |
1217 | else | |
84f3c683 | 1218 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
750b9c90 | 1219 | trace_xfs_log_assign_tail_lsn(log, tail_lsn); |
1c3cb9ec | 1220 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1221 | return tail_lsn; |
1c3cb9ec | 1222 | } |
1da177e4 | 1223 | |
1c304625 CH |
1224 | xfs_lsn_t |
1225 | xlog_assign_tail_lsn( | |
1226 | struct xfs_mount *mp) | |
1227 | { | |
1228 | xfs_lsn_t tail_lsn; | |
1229 | ||
57e80956 | 1230 | spin_lock(&mp->m_ail->ail_lock); |
1c304625 | 1231 | tail_lsn = xlog_assign_tail_lsn_locked(mp); |
57e80956 | 1232 | spin_unlock(&mp->m_ail->ail_lock); |
1c304625 CH |
1233 | |
1234 | return tail_lsn; | |
1235 | } | |
1236 | ||
1da177e4 LT |
1237 | /* |
1238 | * Return the space in the log between the tail and the head. The head | |
1239 | * is passed in the cycle/bytes formal parms. In the special case where | |
1240 | * the reserve head has wrapped passed the tail, this calculation is no | |
1241 | * longer valid. In this case, just return 0 which means there is no space | |
1242 | * in the log. This works for all places where this function is called | |
1243 | * with the reserve head. Of course, if the write head were to ever | |
1244 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1245 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1246 | * | |
1247 | * This code also handles the case where the reservation head is behind | |
1248 | * the tail. The details of this case are described below, but the end | |
1249 | * result is that we return the size of the log as the amount of space left. | |
1250 | */ | |
a8272ce0 | 1251 | STATIC int |
a69ed03c | 1252 | xlog_space_left( |
ad223e60 | 1253 | struct xlog *log, |
c8a09ff8 | 1254 | atomic64_t *head) |
1da177e4 | 1255 | { |
a69ed03c DC |
1256 | int free_bytes; |
1257 | int tail_bytes; | |
1258 | int tail_cycle; | |
1259 | int head_cycle; | |
1260 | int head_bytes; | |
1da177e4 | 1261 | |
a69ed03c | 1262 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1263 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1264 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1265 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1266 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1267 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1268 | return 0; |
a69ed03c DC |
1269 | else if (tail_cycle < head_cycle) { |
1270 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1271 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1272 | } else { |
1273 | /* | |
1274 | * The reservation head is behind the tail. | |
1275 | * In this case we just want to return the size of the | |
1276 | * log as the amount of space left. | |
1277 | */ | |
f41febd2 | 1278 | xfs_alert(log->l_mp, "xlog_space_left: head behind tail"); |
a0fa2b67 | 1279 | xfs_alert(log->l_mp, |
f41febd2 JP |
1280 | " tail_cycle = %d, tail_bytes = %d", |
1281 | tail_cycle, tail_bytes); | |
1282 | xfs_alert(log->l_mp, | |
1283 | " GH cycle = %d, GH bytes = %d", | |
1284 | head_cycle, head_bytes); | |
1da177e4 LT |
1285 | ASSERT(0); |
1286 | free_bytes = log->l_logsize; | |
1287 | } | |
1288 | return free_bytes; | |
a69ed03c | 1289 | } |
1da177e4 LT |
1290 | |
1291 | ||
0d5a75e9 | 1292 | static void |
79b54d9b CH |
1293 | xlog_ioend_work( |
1294 | struct work_struct *work) | |
1da177e4 | 1295 | { |
79b54d9b CH |
1296 | struct xlog_in_core *iclog = |
1297 | container_of(work, struct xlog_in_core, ic_end_io_work); | |
1298 | struct xlog *log = iclog->ic_log; | |
79b54d9b | 1299 | int error; |
1da177e4 | 1300 | |
79b54d9b | 1301 | error = blk_status_to_errno(iclog->ic_bio.bi_status); |
366fc4b8 CH |
1302 | #ifdef DEBUG |
1303 | /* treat writes with injected CRC errors as failed */ | |
1304 | if (iclog->ic_fail_crc) | |
79b54d9b | 1305 | error = -EIO; |
366fc4b8 CH |
1306 | #endif |
1307 | ||
1da177e4 | 1308 | /* |
366fc4b8 | 1309 | * Race to shutdown the filesystem if we see an error. |
1da177e4 | 1310 | */ |
79b54d9b CH |
1311 | if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) { |
1312 | xfs_alert(log->l_mp, "log I/O error %d", error); | |
1313 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1da177e4 | 1314 | } |
3db296f3 | 1315 | |
12e6a0f4 | 1316 | xlog_state_done_syncing(iclog); |
79b54d9b | 1317 | bio_uninit(&iclog->ic_bio); |
9c23eccc | 1318 | |
3db296f3 | 1319 | /* |
79b54d9b CH |
1320 | * Drop the lock to signal that we are done. Nothing references the |
1321 | * iclog after this, so an unmount waiting on this lock can now tear it | |
1322 | * down safely. As such, it is unsafe to reference the iclog after the | |
1323 | * unlock as we could race with it being freed. | |
3db296f3 | 1324 | */ |
79b54d9b | 1325 | up(&iclog->ic_sema); |
c3f8fc73 | 1326 | } |
1da177e4 | 1327 | |
1da177e4 LT |
1328 | /* |
1329 | * Return size of each in-core log record buffer. | |
1330 | * | |
9da096fd | 1331 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1332 | * |
1333 | * If the filesystem blocksize is too large, we may need to choose a | |
1334 | * larger size since the directory code currently logs entire blocks. | |
1335 | */ | |
1da177e4 | 1336 | STATIC void |
9a8d2fdb MT |
1337 | xlog_get_iclog_buffer_size( |
1338 | struct xfs_mount *mp, | |
1339 | struct xlog *log) | |
1da177e4 | 1340 | { |
1cb51258 | 1341 | if (mp->m_logbufs <= 0) |
4f62282a CH |
1342 | mp->m_logbufs = XLOG_MAX_ICLOGS; |
1343 | if (mp->m_logbsize <= 0) | |
1344 | mp->m_logbsize = XLOG_BIG_RECORD_BSIZE; | |
1345 | ||
1346 | log->l_iclog_bufs = mp->m_logbufs; | |
1347 | log->l_iclog_size = mp->m_logbsize; | |
1da177e4 LT |
1348 | |
1349 | /* | |
4f62282a | 1350 | * # headers = size / 32k - one header holds cycles from 32k of data. |
1da177e4 | 1351 | */ |
4f62282a CH |
1352 | log->l_iclog_heads = |
1353 | DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE); | |
1354 | log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT; | |
1355 | } | |
1da177e4 | 1356 | |
f661f1e0 DC |
1357 | void |
1358 | xfs_log_work_queue( | |
1359 | struct xfs_mount *mp) | |
1360 | { | |
696a5620 | 1361 | queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1362 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1363 | } | |
1364 | ||
2b73a2c8 DW |
1365 | /* |
1366 | * Clear the log incompat flags if we have the opportunity. | |
1367 | * | |
1368 | * This only happens if we're about to log the second dummy transaction as part | |
1369 | * of covering the log and we can get the log incompat feature usage lock. | |
1370 | */ | |
1371 | static inline void | |
1372 | xlog_clear_incompat( | |
1373 | struct xlog *log) | |
1374 | { | |
1375 | struct xfs_mount *mp = log->l_mp; | |
1376 | ||
1377 | if (!xfs_sb_has_incompat_log_feature(&mp->m_sb, | |
1378 | XFS_SB_FEAT_INCOMPAT_LOG_ALL)) | |
1379 | return; | |
1380 | ||
1381 | if (log->l_covered_state != XLOG_STATE_COVER_DONE2) | |
1382 | return; | |
1383 | ||
1384 | if (!down_write_trylock(&log->l_incompat_users)) | |
1385 | return; | |
1386 | ||
1387 | xfs_clear_incompat_log_features(mp); | |
1388 | up_write(&log->l_incompat_users); | |
1389 | } | |
1390 | ||
f661f1e0 DC |
1391 | /* |
1392 | * Every sync period we need to unpin all items in the AIL and push them to | |
1393 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1394 | * indicate that the filesystem is idle. | |
1395 | */ | |
0d5a75e9 | 1396 | static void |
f661f1e0 DC |
1397 | xfs_log_worker( |
1398 | struct work_struct *work) | |
1399 | { | |
1400 | struct xlog *log = container_of(to_delayed_work(work), | |
1401 | struct xlog, l_work); | |
1402 | struct xfs_mount *mp = log->l_mp; | |
1403 | ||
1404 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
37444fc4 | 1405 | if (xfs_fs_writable(mp, SB_FREEZE_WRITE) && xfs_log_need_covered(mp)) { |
61e63ecb DC |
1406 | /* |
1407 | * Dump a transaction into the log that contains no real change. | |
1408 | * This is needed to stamp the current tail LSN into the log | |
1409 | * during the covering operation. | |
1410 | * | |
1411 | * We cannot use an inode here for this - that will push dirty | |
1412 | * state back up into the VFS and then periodic inode flushing | |
1413 | * will prevent log covering from making progress. Hence we | |
1414 | * synchronously log the superblock instead to ensure the | |
1415 | * superblock is immediately unpinned and can be written back. | |
1416 | */ | |
2b73a2c8 | 1417 | xlog_clear_incompat(log); |
61e63ecb DC |
1418 | xfs_sync_sb(mp, true); |
1419 | } else | |
f661f1e0 DC |
1420 | xfs_log_force(mp, 0); |
1421 | ||
1422 | /* start pushing all the metadata that is currently dirty */ | |
1423 | xfs_ail_push_all(mp->m_ail); | |
1424 | ||
1425 | /* queue us up again */ | |
1426 | xfs_log_work_queue(mp); | |
1427 | } | |
1428 | ||
1da177e4 LT |
1429 | /* |
1430 | * This routine initializes some of the log structure for a given mount point. | |
1431 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1432 | * some other stuff may be filled in too. | |
1433 | */ | |
9a8d2fdb MT |
1434 | STATIC struct xlog * |
1435 | xlog_alloc_log( | |
1436 | struct xfs_mount *mp, | |
1437 | struct xfs_buftarg *log_target, | |
1438 | xfs_daddr_t blk_offset, | |
1439 | int num_bblks) | |
1da177e4 | 1440 | { |
9a8d2fdb | 1441 | struct xlog *log; |
1da177e4 LT |
1442 | xlog_rec_header_t *head; |
1443 | xlog_in_core_t **iclogp; | |
1444 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1da177e4 | 1445 | int i; |
2451337d | 1446 | int error = -ENOMEM; |
69ce58f0 | 1447 | uint log2_size = 0; |
1da177e4 | 1448 | |
9a8d2fdb | 1449 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1450 | if (!log) { |
a0fa2b67 | 1451 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1452 | goto out; |
1453 | } | |
1da177e4 LT |
1454 | |
1455 | log->l_mp = mp; | |
1456 | log->l_targ = log_target; | |
1457 | log->l_logsize = BBTOB(num_bblks); | |
1458 | log->l_logBBstart = blk_offset; | |
1459 | log->l_logBBsize = num_bblks; | |
1460 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1461 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1462 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1463 | |
1464 | log->l_prev_block = -1; | |
1da177e4 | 1465 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1466 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1467 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1468 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 | 1469 | |
a6a65fef DC |
1470 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) |
1471 | log->l_iclog_roundoff = mp->m_sb.sb_logsunit; | |
1472 | else | |
1473 | log->l_iclog_roundoff = BBSIZE; | |
1474 | ||
c303c5b8 CH |
1475 | xlog_grant_head_init(&log->l_reserve_head); |
1476 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1477 | |
2451337d | 1478 | error = -EFSCORRUPTED; |
62118709 | 1479 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1480 | log2_size = mp->m_sb.sb_logsectlog; |
1481 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1482 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1483 | log2_size, BBSHIFT); | |
a6cb767e DC |
1484 | goto out_free_log; |
1485 | } | |
1486 | ||
69ce58f0 AE |
1487 | log2_size -= BBSHIFT; |
1488 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1489 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1490 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1491 | goto out_free_log; |
1492 | } | |
69ce58f0 AE |
1493 | |
1494 | /* for larger sector sizes, must have v2 or external log */ | |
1495 | if (log2_size && log->l_logBBstart > 0 && | |
1496 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1497 | xfs_warn(mp, |
1498 | "log sector size (0x%x) invalid for configuration.", | |
1499 | log2_size); | |
a6cb767e DC |
1500 | goto out_free_log; |
1501 | } | |
1da177e4 | 1502 | } |
69ce58f0 | 1503 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 | 1504 | |
2b73a2c8 DW |
1505 | init_rwsem(&log->l_incompat_users); |
1506 | ||
1da177e4 LT |
1507 | xlog_get_iclog_buffer_size(mp, log); |
1508 | ||
007c61c6 | 1509 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1510 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1511 | |
1da177e4 LT |
1512 | iclogp = &log->l_iclog; |
1513 | /* | |
1514 | * The amount of memory to allocate for the iclog structure is | |
1515 | * rather funky due to the way the structure is defined. It is | |
1516 | * done this way so that we can use different sizes for machines | |
1517 | * with different amounts of memory. See the definition of | |
1518 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1519 | */ | |
1da177e4 | 1520 | ASSERT(log->l_iclog_size >= 4096); |
79b54d9b | 1521 | for (i = 0; i < log->l_iclog_bufs; i++) { |
89b171ac CH |
1522 | size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE) * |
1523 | sizeof(struct bio_vec); | |
79b54d9b CH |
1524 | |
1525 | iclog = kmem_zalloc(sizeof(*iclog) + bvec_size, KM_MAYFAIL); | |
1526 | if (!iclog) | |
644c3567 DC |
1527 | goto out_free_iclog; |
1528 | ||
79b54d9b | 1529 | *iclogp = iclog; |
1da177e4 LT |
1530 | iclog->ic_prev = prev_iclog; |
1531 | prev_iclog = iclog; | |
1fa40b01 | 1532 | |
d634525d DC |
1533 | iclog->ic_data = kvzalloc(log->l_iclog_size, |
1534 | GFP_KERNEL | __GFP_RETRY_MAYFAIL); | |
79b54d9b | 1535 | if (!iclog->ic_data) |
644c3567 | 1536 | goto out_free_iclog; |
4679b2d3 | 1537 | #ifdef DEBUG |
5809d5e0 | 1538 | log->l_iclog_bak[i] = &iclog->ic_header; |
4679b2d3 | 1539 | #endif |
1da177e4 LT |
1540 | head = &iclog->ic_header; |
1541 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1542 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1543 | head->h_version = cpu_to_be32( | |
62118709 | 1544 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1545 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1546 | /* new fields */ |
b53e675d | 1547 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1548 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1549 | ||
79b54d9b | 1550 | iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize; |
1da177e4 LT |
1551 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1552 | iclog->ic_log = log; | |
114d23aa | 1553 | atomic_set(&iclog->ic_refcnt, 0); |
89ae379d | 1554 | INIT_LIST_HEAD(&iclog->ic_callbacks); |
b28708d6 | 1555 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1556 | |
eb40a875 DC |
1557 | init_waitqueue_head(&iclog->ic_force_wait); |
1558 | init_waitqueue_head(&iclog->ic_write_wait); | |
79b54d9b CH |
1559 | INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work); |
1560 | sema_init(&iclog->ic_sema, 1); | |
1da177e4 LT |
1561 | |
1562 | iclogp = &iclog->ic_next; | |
1563 | } | |
1564 | *iclogp = log->l_iclog; /* complete ring */ | |
1565 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1566 | ||
1058d0f5 | 1567 | log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s", |
05a302a1 DW |
1568 | XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | |
1569 | WQ_HIGHPRI), | |
1570 | 0, mp->m_super->s_id); | |
1058d0f5 CH |
1571 | if (!log->l_ioend_workqueue) |
1572 | goto out_free_iclog; | |
1573 | ||
71e330b5 DC |
1574 | error = xlog_cil_init(log); |
1575 | if (error) | |
1058d0f5 | 1576 | goto out_destroy_workqueue; |
1da177e4 | 1577 | return log; |
644c3567 | 1578 | |
1058d0f5 CH |
1579 | out_destroy_workqueue: |
1580 | destroy_workqueue(log->l_ioend_workqueue); | |
644c3567 DC |
1581 | out_free_iclog: |
1582 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1583 | prev_iclog = iclog->ic_next; | |
79b54d9b | 1584 | kmem_free(iclog->ic_data); |
644c3567 | 1585 | kmem_free(iclog); |
798a9cad BF |
1586 | if (prev_iclog == log->l_iclog) |
1587 | break; | |
644c3567 | 1588 | } |
644c3567 DC |
1589 | out_free_log: |
1590 | kmem_free(log); | |
a6cb767e | 1591 | out: |
2451337d | 1592 | return ERR_PTR(error); |
1da177e4 LT |
1593 | } /* xlog_alloc_log */ |
1594 | ||
1da177e4 LT |
1595 | /* |
1596 | * Write out the commit record of a transaction associated with the given | |
f10e925d | 1597 | * ticket to close off a running log write. Return the lsn of the commit record. |
1da177e4 | 1598 | */ |
f10e925d | 1599 | int |
55b66332 | 1600 | xlog_commit_record( |
ad223e60 | 1601 | struct xlog *log, |
55b66332 DC |
1602 | struct xlog_ticket *ticket, |
1603 | struct xlog_in_core **iclog, | |
f10e925d | 1604 | xfs_lsn_t *lsn) |
1da177e4 | 1605 | { |
55b66332 DC |
1606 | struct xfs_log_iovec reg = { |
1607 | .i_addr = NULL, | |
1608 | .i_len = 0, | |
1609 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1610 | }; | |
1611 | struct xfs_log_vec vec = { | |
1612 | .lv_niovecs = 1, | |
1613 | .lv_iovecp = ®, | |
1614 | }; | |
f10e925d | 1615 | int error; |
1da177e4 | 1616 | |
f10e925d DC |
1617 | if (XLOG_FORCED_SHUTDOWN(log)) |
1618 | return -EIO; | |
1619 | ||
3468bb1c | 1620 | error = xlog_write(log, &vec, ticket, lsn, iclog, XLOG_COMMIT_TRANS); |
55b66332 | 1621 | if (error) |
f10e925d | 1622 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1623 | return error; |
55b66332 | 1624 | } |
1da177e4 LT |
1625 | |
1626 | /* | |
ed1575da DW |
1627 | * Compute the LSN that we'd need to push the log tail towards in order to have |
1628 | * (a) enough on-disk log space to log the number of bytes specified, (b) at | |
1629 | * least 25% of the log space free, and (c) at least 256 blocks free. If the | |
1630 | * log free space already meets all three thresholds, this function returns | |
1631 | * NULLCOMMITLSN. | |
1da177e4 | 1632 | */ |
ed1575da DW |
1633 | xfs_lsn_t |
1634 | xlog_grant_push_threshold( | |
ad223e60 | 1635 | struct xlog *log, |
2ced19cb | 1636 | int need_bytes) |
1da177e4 | 1637 | { |
2ced19cb | 1638 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1639 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1640 | int free_blocks; |
1641 | int free_bytes; | |
1642 | int threshold_block; | |
1643 | int threshold_cycle; | |
1644 | int free_threshold; | |
1645 | ||
1646 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1647 | ||
28496968 | 1648 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1649 | free_blocks = BTOBBT(free_bytes); |
1650 | ||
1651 | /* | |
1652 | * Set the threshold for the minimum number of free blocks in the | |
1653 | * log to the maximum of what the caller needs, one quarter of the | |
1654 | * log, and 256 blocks. | |
1655 | */ | |
1656 | free_threshold = BTOBB(need_bytes); | |
9bb54cb5 DC |
1657 | free_threshold = max(free_threshold, (log->l_logBBsize >> 2)); |
1658 | free_threshold = max(free_threshold, 256); | |
2ced19cb | 1659 | if (free_blocks >= free_threshold) |
ed1575da | 1660 | return NULLCOMMITLSN; |
2ced19cb | 1661 | |
1c3cb9ec DC |
1662 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1663 | &threshold_block); | |
1664 | threshold_block += free_threshold; | |
1da177e4 | 1665 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1666 | threshold_block -= log->l_logBBsize; |
1667 | threshold_cycle += 1; | |
1da177e4 | 1668 | } |
2ced19cb DC |
1669 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1670 | threshold_block); | |
1671 | /* | |
1672 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1673 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1674 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1675 | */ |
84f3c683 DC |
1676 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1677 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1678 | threshold_lsn = last_sync_lsn; | |
2ced19cb | 1679 | |
ed1575da DW |
1680 | return threshold_lsn; |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * Push the tail of the log if we need to do so to maintain the free log space | |
1685 | * thresholds set out by xlog_grant_push_threshold. We may need to adopt a | |
1686 | * policy which pushes on an lsn which is further along in the log once we | |
1687 | * reach the high water mark. In this manner, we would be creating a low water | |
1688 | * mark. | |
1689 | */ | |
1690 | STATIC void | |
1691 | xlog_grant_push_ail( | |
1692 | struct xlog *log, | |
1693 | int need_bytes) | |
1694 | { | |
1695 | xfs_lsn_t threshold_lsn; | |
1696 | ||
1697 | threshold_lsn = xlog_grant_push_threshold(log, need_bytes); | |
1698 | if (threshold_lsn == NULLCOMMITLSN || XLOG_FORCED_SHUTDOWN(log)) | |
1699 | return; | |
1700 | ||
2ced19cb DC |
1701 | /* |
1702 | * Get the transaction layer to kick the dirty buffers out to | |
1703 | * disk asynchronously. No point in trying to do this if | |
1704 | * the filesystem is shutting down. | |
1705 | */ | |
ed1575da | 1706 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1707 | } |
1da177e4 | 1708 | |
0e446be4 CH |
1709 | /* |
1710 | * Stamp cycle number in every block | |
1711 | */ | |
1712 | STATIC void | |
1713 | xlog_pack_data( | |
1714 | struct xlog *log, | |
1715 | struct xlog_in_core *iclog, | |
1716 | int roundoff) | |
1717 | { | |
1718 | int i, j, k; | |
1719 | int size = iclog->ic_offset + roundoff; | |
1720 | __be32 cycle_lsn; | |
b2a922cd | 1721 | char *dp; |
0e446be4 CH |
1722 | |
1723 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1724 | ||
1725 | dp = iclog->ic_datap; | |
1726 | for (i = 0; i < BTOBB(size); i++) { | |
1727 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1728 | break; | |
1729 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1730 | *(__be32 *)dp = cycle_lsn; | |
1731 | dp += BBSIZE; | |
1732 | } | |
1733 | ||
1734 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1735 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1736 | ||
1737 | for ( ; i < BTOBB(size); i++) { | |
1738 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1739 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1740 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1741 | *(__be32 *)dp = cycle_lsn; | |
1742 | dp += BBSIZE; | |
1743 | } | |
1744 | ||
1745 | for (i = 1; i < log->l_iclog_heads; i++) | |
1746 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1747 | } | |
1748 | } | |
1749 | ||
1750 | /* | |
1751 | * Calculate the checksum for a log buffer. | |
1752 | * | |
1753 | * This is a little more complicated than it should be because the various | |
1754 | * headers and the actual data are non-contiguous. | |
1755 | */ | |
f9668a09 | 1756 | __le32 |
0e446be4 CH |
1757 | xlog_cksum( |
1758 | struct xlog *log, | |
1759 | struct xlog_rec_header *rhead, | |
1760 | char *dp, | |
1761 | int size) | |
1762 | { | |
c8ce540d | 1763 | uint32_t crc; |
0e446be4 CH |
1764 | |
1765 | /* first generate the crc for the record header ... */ | |
cae028df | 1766 | crc = xfs_start_cksum_update((char *)rhead, |
0e446be4 CH |
1767 | sizeof(struct xlog_rec_header), |
1768 | offsetof(struct xlog_rec_header, h_crc)); | |
1769 | ||
1770 | /* ... then for additional cycle data for v2 logs ... */ | |
1771 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1772 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1773 | int i; | |
a3f20014 | 1774 | int xheads; |
0e446be4 | 1775 | |
0c771b99 | 1776 | xheads = DIV_ROUND_UP(size, XLOG_HEADER_CYCLE_SIZE); |
0e446be4 | 1777 | |
a3f20014 | 1778 | for (i = 1; i < xheads; i++) { |
0e446be4 CH |
1779 | crc = crc32c(crc, &xhdr[i].hic_xheader, |
1780 | sizeof(struct xlog_rec_ext_header)); | |
1781 | } | |
1782 | } | |
1783 | ||
1784 | /* ... and finally for the payload */ | |
1785 | crc = crc32c(crc, dp, size); | |
1786 | ||
1787 | return xfs_end_cksum(crc); | |
1788 | } | |
1789 | ||
79b54d9b CH |
1790 | static void |
1791 | xlog_bio_end_io( | |
1792 | struct bio *bio) | |
1793 | { | |
1794 | struct xlog_in_core *iclog = bio->bi_private; | |
1795 | ||
1058d0f5 | 1796 | queue_work(iclog->ic_log->l_ioend_workqueue, |
79b54d9b CH |
1797 | &iclog->ic_end_io_work); |
1798 | } | |
1799 | ||
842a42d1 | 1800 | static int |
79b54d9b CH |
1801 | xlog_map_iclog_data( |
1802 | struct bio *bio, | |
1803 | void *data, | |
1804 | size_t count) | |
1805 | { | |
1806 | do { | |
1807 | struct page *page = kmem_to_page(data); | |
1808 | unsigned int off = offset_in_page(data); | |
1809 | size_t len = min_t(size_t, count, PAGE_SIZE - off); | |
1810 | ||
842a42d1 BF |
1811 | if (bio_add_page(bio, page, len, off) != len) |
1812 | return -EIO; | |
79b54d9b CH |
1813 | |
1814 | data += len; | |
1815 | count -= len; | |
1816 | } while (count); | |
842a42d1 BF |
1817 | |
1818 | return 0; | |
79b54d9b CH |
1819 | } |
1820 | ||
94860a30 CH |
1821 | STATIC void |
1822 | xlog_write_iclog( | |
1823 | struct xlog *log, | |
1824 | struct xlog_in_core *iclog, | |
94860a30 | 1825 | uint64_t bno, |
eef983ff | 1826 | unsigned int count) |
873ff550 | 1827 | { |
94860a30 | 1828 | ASSERT(bno < log->l_logBBsize); |
956f6daa | 1829 | trace_xlog_iclog_write(iclog, _RET_IP_); |
94860a30 CH |
1830 | |
1831 | /* | |
1832 | * We lock the iclogbufs here so that we can serialise against I/O | |
1833 | * completion during unmount. We might be processing a shutdown | |
1834 | * triggered during unmount, and that can occur asynchronously to the | |
1835 | * unmount thread, and hence we need to ensure that completes before | |
1836 | * tearing down the iclogbufs. Hence we need to hold the buffer lock | |
1837 | * across the log IO to archieve that. | |
1838 | */ | |
79b54d9b | 1839 | down(&iclog->ic_sema); |
1858bb0b | 1840 | if (unlikely(iclog->ic_state == XLOG_STATE_IOERROR)) { |
873ff550 CH |
1841 | /* |
1842 | * It would seem logical to return EIO here, but we rely on | |
1843 | * the log state machine to propagate I/O errors instead of | |
79b54d9b CH |
1844 | * doing it here. We kick of the state machine and unlock |
1845 | * the buffer manually, the code needs to be kept in sync | |
1846 | * with the I/O completion path. | |
873ff550 | 1847 | */ |
12e6a0f4 | 1848 | xlog_state_done_syncing(iclog); |
79b54d9b | 1849 | up(&iclog->ic_sema); |
94860a30 | 1850 | return; |
873ff550 CH |
1851 | } |
1852 | ||
79b54d9b CH |
1853 | bio_init(&iclog->ic_bio, iclog->ic_bvec, howmany(count, PAGE_SIZE)); |
1854 | bio_set_dev(&iclog->ic_bio, log->l_targ->bt_bdev); | |
1855 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno; | |
1856 | iclog->ic_bio.bi_end_io = xlog_bio_end_io; | |
1857 | iclog->ic_bio.bi_private = iclog; | |
2def2845 DC |
1858 | |
1859 | /* | |
1860 | * We use REQ_SYNC | REQ_IDLE here to tell the block layer the are more | |
1861 | * IOs coming immediately after this one. This prevents the block layer | |
1862 | * writeback throttle from throttling log writes behind background | |
1863 | * metadata writeback and causing priority inversions. | |
1864 | */ | |
eef983ff | 1865 | iclog->ic_bio.bi_opf = REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_IDLE; |
b5d721ea | 1866 | if (iclog->ic_flags & XLOG_ICL_NEED_FLUSH) { |
79b54d9b | 1867 | iclog->ic_bio.bi_opf |= REQ_PREFLUSH; |
b5d721ea DC |
1868 | /* |
1869 | * For external log devices, we also need to flush the data | |
1870 | * device cache first to ensure all metadata writeback covered | |
1871 | * by the LSN in this iclog is on stable storage. This is slow, | |
1872 | * but it *must* complete before we issue the external log IO. | |
1873 | */ | |
1874 | if (log->l_targ != log->l_mp->m_ddev_targp) | |
1875 | blkdev_issue_flush(log->l_mp->m_ddev_targp->bt_bdev); | |
1876 | } | |
eef983ff DC |
1877 | if (iclog->ic_flags & XLOG_ICL_NEED_FUA) |
1878 | iclog->ic_bio.bi_opf |= REQ_FUA; | |
b5d721ea | 1879 | |
eef983ff | 1880 | iclog->ic_flags &= ~(XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA); |
79b54d9b | 1881 | |
842a42d1 BF |
1882 | if (xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, count)) { |
1883 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1884 | return; | |
1885 | } | |
79b54d9b | 1886 | if (is_vmalloc_addr(iclog->ic_data)) |
2c68a1df | 1887 | flush_kernel_vmap_range(iclog->ic_data, count); |
79b54d9b CH |
1888 | |
1889 | /* | |
1890 | * If this log buffer would straddle the end of the log we will have | |
1891 | * to split it up into two bios, so that we can continue at the start. | |
1892 | */ | |
1893 | if (bno + BTOBB(count) > log->l_logBBsize) { | |
1894 | struct bio *split; | |
1895 | ||
1896 | split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno, | |
1897 | GFP_NOIO, &fs_bio_set); | |
1898 | bio_chain(split, &iclog->ic_bio); | |
1899 | submit_bio(split); | |
1900 | ||
1901 | /* restart at logical offset zero for the remainder */ | |
1902 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart; | |
1903 | } | |
1904 | ||
1905 | submit_bio(&iclog->ic_bio); | |
873ff550 | 1906 | } |
1da177e4 | 1907 | |
56933848 CH |
1908 | /* |
1909 | * We need to bump cycle number for the part of the iclog that is | |
1910 | * written to the start of the log. Watch out for the header magic | |
1911 | * number case, though. | |
1912 | */ | |
79b54d9b | 1913 | static void |
56933848 CH |
1914 | xlog_split_iclog( |
1915 | struct xlog *log, | |
1916 | void *data, | |
1917 | uint64_t bno, | |
1918 | unsigned int count) | |
1919 | { | |
1920 | unsigned int split_offset = BBTOB(log->l_logBBsize - bno); | |
1921 | unsigned int i; | |
1922 | ||
1923 | for (i = split_offset; i < count; i += BBSIZE) { | |
1924 | uint32_t cycle = get_unaligned_be32(data + i); | |
1925 | ||
1926 | if (++cycle == XLOG_HEADER_MAGIC_NUM) | |
1927 | cycle++; | |
1928 | put_unaligned_be32(cycle, data + i); | |
1929 | } | |
56933848 CH |
1930 | } |
1931 | ||
db0a6faf CH |
1932 | static int |
1933 | xlog_calc_iclog_size( | |
1934 | struct xlog *log, | |
1935 | struct xlog_in_core *iclog, | |
1936 | uint32_t *roundoff) | |
1937 | { | |
1938 | uint32_t count_init, count; | |
db0a6faf CH |
1939 | |
1940 | /* Add for LR header */ | |
1941 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
a6a65fef | 1942 | count = roundup(count_init, log->l_iclog_roundoff); |
db0a6faf | 1943 | |
db0a6faf CH |
1944 | *roundoff = count - count_init; |
1945 | ||
a6a65fef DC |
1946 | ASSERT(count >= count_init); |
1947 | ASSERT(*roundoff < log->l_iclog_roundoff); | |
db0a6faf CH |
1948 | return count; |
1949 | } | |
1950 | ||
1da177e4 LT |
1951 | /* |
1952 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1953 | * fashion. Previously, we should have moved the current iclog | |
1954 | * ptr in the log to point to the next available iclog. This allows further | |
1955 | * write to continue while this code syncs out an iclog ready to go. | |
1956 | * Before an in-core log can be written out, the data section must be scanned | |
1957 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1958 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1959 | * cycle count because there in an implicit assumption that drives will | |
1960 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1961 | * we can't have part of a 512 byte block written and part not written. By | |
1962 | * tagging each block, we will know which blocks are valid when recovering | |
1963 | * after an unclean shutdown. | |
1964 | * | |
1965 | * This routine is single threaded on the iclog. No other thread can be in | |
1966 | * this routine with the same iclog. Changing contents of iclog can there- | |
1967 | * fore be done without grabbing the state machine lock. Updating the global | |
1968 | * log will require grabbing the lock though. | |
1969 | * | |
1970 | * The entire log manager uses a logical block numbering scheme. Only | |
94860a30 CH |
1971 | * xlog_write_iclog knows about the fact that the log may not start with |
1972 | * block zero on a given device. | |
1da177e4 | 1973 | */ |
94860a30 | 1974 | STATIC void |
9a8d2fdb MT |
1975 | xlog_sync( |
1976 | struct xlog *log, | |
1977 | struct xlog_in_core *iclog) | |
1da177e4 | 1978 | { |
db0a6faf CH |
1979 | unsigned int count; /* byte count of bwrite */ |
1980 | unsigned int roundoff; /* roundoff to BB or stripe */ | |
1981 | uint64_t bno; | |
db0a6faf | 1982 | unsigned int size; |
1da177e4 | 1983 | |
155cc6b7 | 1984 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
956f6daa | 1985 | trace_xlog_iclog_sync(iclog, _RET_IP_); |
1da177e4 | 1986 | |
db0a6faf | 1987 | count = xlog_calc_iclog_size(log, iclog, &roundoff); |
1da177e4 LT |
1988 | |
1989 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1990 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1991 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1992 | |
1993 | /* put cycle number in every block */ | |
1994 | xlog_pack_data(log, iclog, roundoff); | |
1995 | ||
1996 | /* real byte length */ | |
0e446be4 | 1997 | size = iclog->ic_offset; |
db0a6faf | 1998 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) |
0e446be4 CH |
1999 | size += roundoff; |
2000 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 2001 | |
9b0489c1 | 2002 | XFS_STATS_INC(log->l_mp, xs_log_writes); |
ff6d6af2 | 2003 | XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count)); |
1da177e4 | 2004 | |
94860a30 CH |
2005 | bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)); |
2006 | ||
1da177e4 | 2007 | /* Do we need to split this write into 2 parts? */ |
eef983ff | 2008 | if (bno + BTOBB(count) > log->l_logBBsize) |
79b54d9b | 2009 | xlog_split_iclog(log, &iclog->ic_header, bno, count); |
0e446be4 CH |
2010 | |
2011 | /* calculcate the checksum */ | |
2012 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
2013 | iclog->ic_datap, size); | |
609adfc2 BF |
2014 | /* |
2015 | * Intentionally corrupt the log record CRC based on the error injection | |
2016 | * frequency, if defined. This facilitates testing log recovery in the | |
2017 | * event of torn writes. Hence, set the IOABORT state to abort the log | |
2018 | * write on I/O completion and shutdown the fs. The subsequent mount | |
2019 | * detects the bad CRC and attempts to recover. | |
2020 | */ | |
366fc4b8 | 2021 | #ifdef DEBUG |
3e88a007 | 2022 | if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) { |
e2a64192 | 2023 | iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA); |
366fc4b8 | 2024 | iclog->ic_fail_crc = true; |
609adfc2 BF |
2025 | xfs_warn(log->l_mp, |
2026 | "Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.", | |
2027 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
2028 | } | |
366fc4b8 | 2029 | #endif |
abca1f33 | 2030 | xlog_verify_iclog(log, iclog, count); |
eef983ff | 2031 | xlog_write_iclog(log, iclog, bno, count); |
94860a30 | 2032 | } |
1da177e4 | 2033 | |
1da177e4 | 2034 | /* |
c41564b5 | 2035 | * Deallocate a log structure |
1da177e4 | 2036 | */ |
a8272ce0 | 2037 | STATIC void |
9a8d2fdb MT |
2038 | xlog_dealloc_log( |
2039 | struct xlog *log) | |
1da177e4 LT |
2040 | { |
2041 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
2042 | int i; |
2043 | ||
71e330b5 DC |
2044 | xlog_cil_destroy(log); |
2045 | ||
44396476 | 2046 | /* |
9c23eccc DC |
2047 | * Cycle all the iclogbuf locks to make sure all log IO completion |
2048 | * is done before we tear down these buffers. | |
2049 | */ | |
2050 | iclog = log->l_iclog; | |
2051 | for (i = 0; i < log->l_iclog_bufs; i++) { | |
79b54d9b CH |
2052 | down(&iclog->ic_sema); |
2053 | up(&iclog->ic_sema); | |
9c23eccc DC |
2054 | iclog = iclog->ic_next; |
2055 | } | |
2056 | ||
1da177e4 | 2057 | iclog = log->l_iclog; |
9c23eccc | 2058 | for (i = 0; i < log->l_iclog_bufs; i++) { |
1da177e4 | 2059 | next_iclog = iclog->ic_next; |
79b54d9b | 2060 | kmem_free(iclog->ic_data); |
f0e2d93c | 2061 | kmem_free(iclog); |
1da177e4 LT |
2062 | iclog = next_iclog; |
2063 | } | |
1da177e4 | 2064 | |
1da177e4 | 2065 | log->l_mp->m_log = NULL; |
1058d0f5 | 2066 | destroy_workqueue(log->l_ioend_workqueue); |
f0e2d93c | 2067 | kmem_free(log); |
b843299b | 2068 | } |
1da177e4 LT |
2069 | |
2070 | /* | |
2071 | * Update counters atomically now that memcpy is done. | |
2072 | */ | |
1da177e4 | 2073 | static inline void |
9a8d2fdb MT |
2074 | xlog_state_finish_copy( |
2075 | struct xlog *log, | |
2076 | struct xlog_in_core *iclog, | |
2077 | int record_cnt, | |
2078 | int copy_bytes) | |
1da177e4 | 2079 | { |
390aab0a | 2080 | lockdep_assert_held(&log->l_icloglock); |
1da177e4 | 2081 | |
413d57c9 | 2082 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 | 2083 | iclog->ic_offset += copy_bytes; |
390aab0a | 2084 | } |
1da177e4 | 2085 | |
7e9c6396 TS |
2086 | /* |
2087 | * print out info relating to regions written which consume | |
2088 | * the reservation | |
2089 | */ | |
71e330b5 DC |
2090 | void |
2091 | xlog_print_tic_res( | |
2092 | struct xfs_mount *mp, | |
2093 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
2094 | { |
2095 | uint i; | |
2096 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
2097 | ||
2098 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
5110cd82 | 2099 | #define REG_TYPE_STR(type, str) [XLOG_REG_TYPE_##type] = str |
d31d7185 | 2100 | static char *res_type_str[] = { |
5110cd82 DW |
2101 | REG_TYPE_STR(BFORMAT, "bformat"), |
2102 | REG_TYPE_STR(BCHUNK, "bchunk"), | |
2103 | REG_TYPE_STR(EFI_FORMAT, "efi_format"), | |
2104 | REG_TYPE_STR(EFD_FORMAT, "efd_format"), | |
2105 | REG_TYPE_STR(IFORMAT, "iformat"), | |
2106 | REG_TYPE_STR(ICORE, "icore"), | |
2107 | REG_TYPE_STR(IEXT, "iext"), | |
2108 | REG_TYPE_STR(IBROOT, "ibroot"), | |
2109 | REG_TYPE_STR(ILOCAL, "ilocal"), | |
2110 | REG_TYPE_STR(IATTR_EXT, "iattr_ext"), | |
2111 | REG_TYPE_STR(IATTR_BROOT, "iattr_broot"), | |
2112 | REG_TYPE_STR(IATTR_LOCAL, "iattr_local"), | |
2113 | REG_TYPE_STR(QFORMAT, "qformat"), | |
2114 | REG_TYPE_STR(DQUOT, "dquot"), | |
2115 | REG_TYPE_STR(QUOTAOFF, "quotaoff"), | |
2116 | REG_TYPE_STR(LRHEADER, "LR header"), | |
2117 | REG_TYPE_STR(UNMOUNT, "unmount"), | |
2118 | REG_TYPE_STR(COMMIT, "commit"), | |
2119 | REG_TYPE_STR(TRANSHDR, "trans header"), | |
d31d7185 DW |
2120 | REG_TYPE_STR(ICREATE, "inode create"), |
2121 | REG_TYPE_STR(RUI_FORMAT, "rui_format"), | |
2122 | REG_TYPE_STR(RUD_FORMAT, "rud_format"), | |
2123 | REG_TYPE_STR(CUI_FORMAT, "cui_format"), | |
2124 | REG_TYPE_STR(CUD_FORMAT, "cud_format"), | |
2125 | REG_TYPE_STR(BUI_FORMAT, "bui_format"), | |
2126 | REG_TYPE_STR(BUD_FORMAT, "bud_format"), | |
7e9c6396 | 2127 | }; |
d31d7185 | 2128 | BUILD_BUG_ON(ARRAY_SIZE(res_type_str) != XLOG_REG_TYPE_MAX + 1); |
5110cd82 | 2129 | #undef REG_TYPE_STR |
7e9c6396 | 2130 | |
7d2d5653 | 2131 | xfs_warn(mp, "ticket reservation summary:"); |
f41febd2 JP |
2132 | xfs_warn(mp, " unit res = %d bytes", |
2133 | ticket->t_unit_res); | |
2134 | xfs_warn(mp, " current res = %d bytes", | |
2135 | ticket->t_curr_res); | |
2136 | xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)", | |
2137 | ticket->t_res_arr_sum, ticket->t_res_o_flow); | |
2138 | xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)", | |
2139 | ticket->t_res_num_ophdrs, ophdr_spc); | |
2140 | xfs_warn(mp, " ophdr + reg = %u bytes", | |
2141 | ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc); | |
2142 | xfs_warn(mp, " num regions = %u", | |
2143 | ticket->t_res_num); | |
7e9c6396 TS |
2144 | |
2145 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 2146 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 2147 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 | 2148 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
5110cd82 | 2149 | "bad-rtype" : res_type_str[r_type]), |
7e9c6396 TS |
2150 | ticket->t_res_arr[i].r_len); |
2151 | } | |
2152 | } | |
7e9c6396 | 2153 | |
d4ca1d55 BF |
2154 | /* |
2155 | * Print a summary of the transaction. | |
2156 | */ | |
2157 | void | |
2158 | xlog_print_trans( | |
e6631f85 | 2159 | struct xfs_trans *tp) |
d4ca1d55 | 2160 | { |
e6631f85 DC |
2161 | struct xfs_mount *mp = tp->t_mountp; |
2162 | struct xfs_log_item *lip; | |
d4ca1d55 BF |
2163 | |
2164 | /* dump core transaction and ticket info */ | |
2165 | xfs_warn(mp, "transaction summary:"); | |
2c8f6265 BF |
2166 | xfs_warn(mp, " log res = %d", tp->t_log_res); |
2167 | xfs_warn(mp, " log count = %d", tp->t_log_count); | |
2168 | xfs_warn(mp, " flags = 0x%x", tp->t_flags); | |
d4ca1d55 BF |
2169 | |
2170 | xlog_print_tic_res(mp, tp->t_ticket); | |
2171 | ||
2172 | /* dump each log item */ | |
e6631f85 | 2173 | list_for_each_entry(lip, &tp->t_items, li_trans) { |
d4ca1d55 BF |
2174 | struct xfs_log_vec *lv = lip->li_lv; |
2175 | struct xfs_log_iovec *vec; | |
2176 | int i; | |
2177 | ||
2178 | xfs_warn(mp, "log item: "); | |
2179 | xfs_warn(mp, " type = 0x%x", lip->li_type); | |
22525c17 | 2180 | xfs_warn(mp, " flags = 0x%lx", lip->li_flags); |
d4ca1d55 BF |
2181 | if (!lv) |
2182 | continue; | |
2183 | xfs_warn(mp, " niovecs = %d", lv->lv_niovecs); | |
2184 | xfs_warn(mp, " size = %d", lv->lv_size); | |
2185 | xfs_warn(mp, " bytes = %d", lv->lv_bytes); | |
2186 | xfs_warn(mp, " buf len = %d", lv->lv_buf_len); | |
2187 | ||
2188 | /* dump each iovec for the log item */ | |
2189 | vec = lv->lv_iovecp; | |
2190 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2191 | int dumplen = min(vec->i_len, 32); | |
2192 | ||
2193 | xfs_warn(mp, " iovec[%d]", i); | |
2194 | xfs_warn(mp, " type = 0x%x", vec->i_type); | |
2195 | xfs_warn(mp, " len = %d", vec->i_len); | |
2196 | xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i); | |
244e3dea | 2197 | xfs_hex_dump(vec->i_addr, dumplen); |
d4ca1d55 BF |
2198 | |
2199 | vec++; | |
2200 | } | |
2201 | } | |
2202 | } | |
2203 | ||
b5203cd0 | 2204 | /* |
7ec94921 DC |
2205 | * Calculate the potential space needed by the log vector. We may need a start |
2206 | * record, and each region gets its own struct xlog_op_header and may need to be | |
2207 | * double word aligned. | |
b5203cd0 DC |
2208 | */ |
2209 | static int | |
2210 | xlog_write_calc_vec_length( | |
2211 | struct xlog_ticket *ticket, | |
7ec94921 | 2212 | struct xfs_log_vec *log_vector, |
3468bb1c | 2213 | uint optype) |
b5203cd0 | 2214 | { |
55b66332 | 2215 | struct xfs_log_vec *lv; |
3468bb1c | 2216 | int headers = 0; |
b5203cd0 DC |
2217 | int len = 0; |
2218 | int i; | |
2219 | ||
3468bb1c DC |
2220 | if (optype & XLOG_START_TRANS) |
2221 | headers++; | |
2222 | ||
55b66332 | 2223 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2224 | /* we don't write ordered log vectors */ |
2225 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2226 | continue; | |
2227 | ||
55b66332 DC |
2228 | headers += lv->lv_niovecs; |
2229 | ||
2230 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2231 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2232 | |
55b66332 DC |
2233 | len += vecp->i_len; |
2234 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2235 | } | |
b5203cd0 DC |
2236 | } |
2237 | ||
2238 | ticket->t_res_num_ophdrs += headers; | |
2239 | len += headers * sizeof(struct xlog_op_header); | |
2240 | ||
2241 | return len; | |
2242 | } | |
2243 | ||
7ec94921 | 2244 | static void |
b5203cd0 | 2245 | xlog_write_start_rec( |
e6b1f273 | 2246 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2247 | struct xlog_ticket *ticket) |
2248 | { | |
b5203cd0 DC |
2249 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2250 | ophdr->oh_clientid = ticket->t_clientid; | |
2251 | ophdr->oh_len = 0; | |
2252 | ophdr->oh_flags = XLOG_START_TRANS; | |
2253 | ophdr->oh_res2 = 0; | |
b5203cd0 DC |
2254 | } |
2255 | ||
2256 | static xlog_op_header_t * | |
2257 | xlog_write_setup_ophdr( | |
ad223e60 | 2258 | struct xlog *log, |
e6b1f273 | 2259 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2260 | struct xlog_ticket *ticket, |
2261 | uint flags) | |
2262 | { | |
b5203cd0 DC |
2263 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2264 | ophdr->oh_clientid = ticket->t_clientid; | |
2265 | ophdr->oh_res2 = 0; | |
2266 | ||
2267 | /* are we copying a commit or unmount record? */ | |
2268 | ophdr->oh_flags = flags; | |
2269 | ||
2270 | /* | |
2271 | * We've seen logs corrupted with bad transaction client ids. This | |
2272 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2273 | * and shut down the filesystem. | |
2274 | */ | |
2275 | switch (ophdr->oh_clientid) { | |
2276 | case XFS_TRANSACTION: | |
2277 | case XFS_VOLUME: | |
2278 | case XFS_LOG: | |
2279 | break; | |
2280 | default: | |
a0fa2b67 | 2281 | xfs_warn(log->l_mp, |
c9690043 | 2282 | "Bad XFS transaction clientid 0x%x in ticket "PTR_FMT, |
b5203cd0 DC |
2283 | ophdr->oh_clientid, ticket); |
2284 | return NULL; | |
2285 | } | |
2286 | ||
2287 | return ophdr; | |
2288 | } | |
2289 | ||
2290 | /* | |
2291 | * Set up the parameters of the region copy into the log. This has | |
2292 | * to handle region write split across multiple log buffers - this | |
2293 | * state is kept external to this function so that this code can | |
ac0e300f | 2294 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2295 | */ |
2296 | static int | |
2297 | xlog_write_setup_copy( | |
2298 | struct xlog_ticket *ticket, | |
2299 | struct xlog_op_header *ophdr, | |
2300 | int space_available, | |
2301 | int space_required, | |
2302 | int *copy_off, | |
2303 | int *copy_len, | |
2304 | int *last_was_partial_copy, | |
2305 | int *bytes_consumed) | |
2306 | { | |
2307 | int still_to_copy; | |
2308 | ||
2309 | still_to_copy = space_required - *bytes_consumed; | |
2310 | *copy_off = *bytes_consumed; | |
2311 | ||
2312 | if (still_to_copy <= space_available) { | |
2313 | /* write of region completes here */ | |
2314 | *copy_len = still_to_copy; | |
2315 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2316 | if (*last_was_partial_copy) | |
2317 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2318 | *last_was_partial_copy = 0; | |
2319 | *bytes_consumed = 0; | |
2320 | return 0; | |
2321 | } | |
2322 | ||
2323 | /* partial write of region, needs extra log op header reservation */ | |
2324 | *copy_len = space_available; | |
2325 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2326 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2327 | if (*last_was_partial_copy) | |
2328 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2329 | *bytes_consumed += *copy_len; | |
2330 | (*last_was_partial_copy)++; | |
2331 | ||
2332 | /* account for new log op header */ | |
2333 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2334 | ticket->t_res_num_ophdrs++; | |
2335 | ||
2336 | return sizeof(struct xlog_op_header); | |
2337 | } | |
2338 | ||
2339 | static int | |
2340 | xlog_write_copy_finish( | |
ad223e60 | 2341 | struct xlog *log, |
b5203cd0 DC |
2342 | struct xlog_in_core *iclog, |
2343 | uint flags, | |
2344 | int *record_cnt, | |
2345 | int *data_cnt, | |
2346 | int *partial_copy, | |
2347 | int *partial_copy_len, | |
2348 | int log_offset, | |
2349 | struct xlog_in_core **commit_iclog) | |
2350 | { | |
df732b29 CH |
2351 | int error; |
2352 | ||
b5203cd0 DC |
2353 | if (*partial_copy) { |
2354 | /* | |
2355 | * This iclog has already been marked WANT_SYNC by | |
2356 | * xlog_state_get_iclog_space. | |
2357 | */ | |
390aab0a | 2358 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2359 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2360 | *record_cnt = 0; | |
2361 | *data_cnt = 0; | |
df732b29 | 2362 | goto release_iclog; |
b5203cd0 DC |
2363 | } |
2364 | ||
2365 | *partial_copy = 0; | |
2366 | *partial_copy_len = 0; | |
2367 | ||
2368 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2369 | /* no more space in this iclog - push it. */ | |
390aab0a | 2370 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2371 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2372 | *record_cnt = 0; | |
2373 | *data_cnt = 0; | |
2374 | ||
69363999 CH |
2375 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
2376 | xlog_state_switch_iclogs(log, iclog, 0); | |
2377 | else | |
2378 | ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
2379 | iclog->ic_state == XLOG_STATE_IOERROR); | |
b5203cd0 | 2380 | if (!commit_iclog) |
df732b29 CH |
2381 | goto release_iclog; |
2382 | spin_unlock(&log->l_icloglock); | |
b5203cd0 DC |
2383 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2384 | *commit_iclog = iclog; | |
2385 | } | |
2386 | ||
2387 | return 0; | |
df732b29 CH |
2388 | |
2389 | release_iclog: | |
0dc8f7f1 | 2390 | error = xlog_state_release_iclog(log, iclog, 0); |
df732b29 CH |
2391 | spin_unlock(&log->l_icloglock); |
2392 | return error; | |
b5203cd0 DC |
2393 | } |
2394 | ||
1da177e4 LT |
2395 | /* |
2396 | * Write some region out to in-core log | |
2397 | * | |
2398 | * This will be called when writing externally provided regions or when | |
2399 | * writing out a commit record for a given transaction. | |
2400 | * | |
2401 | * General algorithm: | |
2402 | * 1. Find total length of this write. This may include adding to the | |
2403 | * lengths passed in. | |
2404 | * 2. Check whether we violate the tickets reservation. | |
2405 | * 3. While writing to this iclog | |
2406 | * A. Reserve as much space in this iclog as can get | |
2407 | * B. If this is first write, save away start lsn | |
2408 | * C. While writing this region: | |
2409 | * 1. If first write of transaction, write start record | |
2410 | * 2. Write log operation header (header per region) | |
2411 | * 3. Find out if we can fit entire region into this iclog | |
2412 | * 4. Potentially, verify destination memcpy ptr | |
2413 | * 5. Memcpy (partial) region | |
2414 | * 6. If partial copy, release iclog; otherwise, continue | |
2415 | * copying more regions into current iclog | |
2416 | * 4. Mark want sync bit (in simulation mode) | |
2417 | * 5. Release iclog for potential flush to on-disk log. | |
2418 | * | |
2419 | * ERRORS: | |
2420 | * 1. Panic if reservation is overrun. This should never happen since | |
2421 | * reservation amounts are generated internal to the filesystem. | |
2422 | * NOTES: | |
2423 | * 1. Tickets are single threaded data structures. | |
2424 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2425 | * syncing routine. When a single log_write region needs to span | |
2426 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2427 | * on all log operation writes which don't contain the end of the | |
2428 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2429 | * operation which contains the end of the continued log_write region. | |
2430 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2431 | * we don't really know exactly how much space will be used. As a result, | |
2432 | * we don't update ic_offset until the end when we know exactly how many | |
2433 | * bytes have been written out. | |
2434 | */ | |
71e330b5 | 2435 | int |
35a8a72f | 2436 | xlog_write( |
ad223e60 | 2437 | struct xlog *log, |
55b66332 | 2438 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2439 | struct xlog_ticket *ticket, |
2440 | xfs_lsn_t *start_lsn, | |
2441 | struct xlog_in_core **commit_iclog, | |
3468bb1c | 2442 | uint optype) |
1da177e4 | 2443 | { |
99428ad0 | 2444 | struct xlog_in_core *iclog = NULL; |
9590e9c6 DC |
2445 | struct xfs_log_vec *lv = log_vector; |
2446 | struct xfs_log_iovec *vecp = lv->lv_iovecp; | |
2447 | int index = 0; | |
99428ad0 | 2448 | int len; |
99428ad0 CH |
2449 | int partial_copy = 0; |
2450 | int partial_copy_len = 0; | |
2451 | int contwr = 0; | |
2452 | int record_cnt = 0; | |
2453 | int data_cnt = 0; | |
df732b29 | 2454 | int error = 0; |
99428ad0 | 2455 | |
93b8a585 | 2456 | /* |
9590e9c6 DC |
2457 | * If this is a commit or unmount transaction, we don't need a start |
2458 | * record to be written. We do, however, have to account for the | |
2459 | * commit or unmount header that gets written. Hence we always have | |
2460 | * to account for an extra xlog_op_header here. | |
93b8a585 | 2461 | */ |
9590e9c6 | 2462 | ticket->t_curr_res -= sizeof(struct xlog_op_header); |
7d2d5653 BF |
2463 | if (ticket->t_curr_res < 0) { |
2464 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
2465 | "ctx ticket reservation ran out. Need to up reservation"); | |
55b66332 | 2466 | xlog_print_tic_res(log->l_mp, ticket); |
7d2d5653 BF |
2467 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
2468 | } | |
1da177e4 | 2469 | |
3468bb1c DC |
2470 | len = xlog_write_calc_vec_length(ticket, log_vector, optype); |
2471 | if (start_lsn) | |
2472 | *start_lsn = 0; | |
fd63875c | 2473 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2474 | void *ptr; |
99428ad0 | 2475 | int log_offset; |
1da177e4 | 2476 | |
99428ad0 CH |
2477 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2478 | &contwr, &log_offset); | |
2479 | if (error) | |
2480 | return error; | |
1da177e4 | 2481 | |
99428ad0 | 2482 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2483 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2484 | |
eef983ff | 2485 | /* Start_lsn is the first lsn written to. */ |
3468bb1c | 2486 | if (start_lsn && !*start_lsn) |
99428ad0 | 2487 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
b5203cd0 | 2488 | |
99428ad0 CH |
2489 | /* |
2490 | * This loop writes out as many regions as can fit in the amount | |
2491 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2492 | */ | |
fd63875c DC |
2493 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2494 | struct xfs_log_iovec *reg; | |
99428ad0 | 2495 | struct xlog_op_header *ophdr; |
99428ad0 CH |
2496 | int copy_len; |
2497 | int copy_off; | |
fd63875c | 2498 | bool ordered = false; |
3468bb1c | 2499 | bool wrote_start_rec = false; |
fd63875c DC |
2500 | |
2501 | /* ordered log vectors have no regions to write */ | |
2502 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2503 | ASSERT(lv->lv_niovecs == 0); | |
2504 | ordered = true; | |
2505 | goto next_lv; | |
2506 | } | |
99428ad0 | 2507 | |
fd63875c | 2508 | reg = &vecp[index]; |
c8ce540d DW |
2509 | ASSERT(reg->i_len % sizeof(int32_t) == 0); |
2510 | ASSERT((unsigned long)ptr % sizeof(int32_t) == 0); | |
99428ad0 | 2511 | |
7ec94921 DC |
2512 | /* |
2513 | * Before we start formatting log vectors, we need to | |
2514 | * write a start record. Only do this for the first | |
2515 | * iclog we write to. | |
2516 | */ | |
3468bb1c | 2517 | if (optype & XLOG_START_TRANS) { |
7ec94921 | 2518 | xlog_write_start_rec(ptr, ticket); |
e6b1f273 | 2519 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
7ec94921 | 2520 | sizeof(struct xlog_op_header)); |
3468bb1c DC |
2521 | optype &= ~XLOG_START_TRANS; |
2522 | wrote_start_rec = true; | |
99428ad0 | 2523 | } |
b5203cd0 | 2524 | |
3468bb1c | 2525 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, optype); |
99428ad0 | 2526 | if (!ophdr) |
2451337d | 2527 | return -EIO; |
99428ad0 | 2528 | |
e6b1f273 | 2529 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2530 | sizeof(struct xlog_op_header)); |
2531 | ||
2532 | len += xlog_write_setup_copy(ticket, ophdr, | |
2533 | iclog->ic_size-log_offset, | |
55b66332 | 2534 | reg->i_len, |
99428ad0 CH |
2535 | ©_off, ©_len, |
2536 | &partial_copy, | |
2537 | &partial_copy_len); | |
2538 | xlog_verify_dest_ptr(log, ptr); | |
2539 | ||
91f9f5fe ES |
2540 | /* |
2541 | * Copy region. | |
2542 | * | |
2543 | * Unmount records just log an opheader, so can have | |
2544 | * empty payloads with no data region to copy. Hence we | |
2545 | * only copy the payload if the vector says it has data | |
2546 | * to copy. | |
2547 | */ | |
99428ad0 | 2548 | ASSERT(copy_len >= 0); |
91f9f5fe ES |
2549 | if (copy_len > 0) { |
2550 | memcpy(ptr, reg->i_addr + copy_off, copy_len); | |
2551 | xlog_write_adv_cnt(&ptr, &len, &log_offset, | |
2552 | copy_len); | |
2553 | } | |
7ec94921 | 2554 | copy_len += sizeof(struct xlog_op_header); |
99428ad0 | 2555 | record_cnt++; |
3468bb1c | 2556 | if (wrote_start_rec) { |
7ec94921 DC |
2557 | copy_len += sizeof(struct xlog_op_header); |
2558 | record_cnt++; | |
7ec94921 | 2559 | } |
99428ad0 CH |
2560 | data_cnt += contwr ? copy_len : 0; |
2561 | ||
3468bb1c | 2562 | error = xlog_write_copy_finish(log, iclog, optype, |
99428ad0 CH |
2563 | &record_cnt, &data_cnt, |
2564 | &partial_copy, | |
2565 | &partial_copy_len, | |
2566 | log_offset, | |
2567 | commit_iclog); | |
2568 | if (error) | |
2569 | return error; | |
2570 | ||
2571 | /* | |
2572 | * if we had a partial copy, we need to get more iclog | |
2573 | * space but we don't want to increment the region | |
2574 | * index because there is still more is this region to | |
2575 | * write. | |
2576 | * | |
2577 | * If we completed writing this region, and we flushed | |
2578 | * the iclog (indicated by resetting of the record | |
2579 | * count), then we also need to get more log space. If | |
2580 | * this was the last record, though, we are done and | |
2581 | * can just return. | |
2582 | */ | |
2583 | if (partial_copy) | |
2584 | break; | |
2585 | ||
55b66332 | 2586 | if (++index == lv->lv_niovecs) { |
fd63875c | 2587 | next_lv: |
55b66332 DC |
2588 | lv = lv->lv_next; |
2589 | index = 0; | |
2590 | if (lv) | |
2591 | vecp = lv->lv_iovecp; | |
2592 | } | |
749f24f3 | 2593 | if (record_cnt == 0 && !ordered) { |
55b66332 | 2594 | if (!lv) |
99428ad0 CH |
2595 | return 0; |
2596 | break; | |
2597 | } | |
2598 | } | |
2599 | } | |
2600 | ||
2601 | ASSERT(len == 0); | |
2602 | ||
390aab0a | 2603 | spin_lock(&log->l_icloglock); |
99428ad0 | 2604 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); |
df732b29 | 2605 | if (commit_iclog) { |
3468bb1c | 2606 | ASSERT(optype & XLOG_COMMIT_TRANS); |
df732b29 CH |
2607 | *commit_iclog = iclog; |
2608 | } else { | |
0dc8f7f1 | 2609 | error = xlog_state_release_iclog(log, iclog, 0); |
df732b29 | 2610 | } |
390aab0a | 2611 | spin_unlock(&log->l_icloglock); |
1da177e4 | 2612 | |
df732b29 | 2613 | return error; |
99428ad0 | 2614 | } |
1da177e4 | 2615 | |
c814b4f2 CH |
2616 | static void |
2617 | xlog_state_activate_iclog( | |
2618 | struct xlog_in_core *iclog, | |
2619 | int *iclogs_changed) | |
2620 | { | |
2621 | ASSERT(list_empty_careful(&iclog->ic_callbacks)); | |
956f6daa | 2622 | trace_xlog_iclog_activate(iclog, _RET_IP_); |
c814b4f2 CH |
2623 | |
2624 | /* | |
2625 | * If the number of ops in this iclog indicate it just contains the | |
2626 | * dummy transaction, we can change state into IDLE (the second time | |
2627 | * around). Otherwise we should change the state into NEED a dummy. | |
2628 | * We don't need to cover the dummy. | |
2629 | */ | |
2630 | if (*iclogs_changed == 0 && | |
2631 | iclog->ic_header.h_num_logops == cpu_to_be32(XLOG_COVER_OPS)) { | |
2632 | *iclogs_changed = 1; | |
2633 | } else { | |
2634 | /* | |
2635 | * We have two dirty iclogs so start over. This could also be | |
2636 | * num of ops indicating this is not the dummy going out. | |
2637 | */ | |
2638 | *iclogs_changed = 2; | |
2639 | } | |
2640 | ||
2641 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2642 | iclog->ic_offset = 0; | |
2643 | iclog->ic_header.h_num_logops = 0; | |
2644 | memset(iclog->ic_header.h_cycle_data, 0, | |
2645 | sizeof(iclog->ic_header.h_cycle_data)); | |
2646 | iclog->ic_header.h_lsn = 0; | |
9d110014 | 2647 | iclog->ic_header.h_tail_lsn = 0; |
c814b4f2 CH |
2648 | } |
2649 | ||
0383f543 | 2650 | /* |
c814b4f2 CH |
2651 | * Loop through all iclogs and mark all iclogs currently marked DIRTY as |
2652 | * ACTIVE after iclog I/O has completed. | |
1da177e4 | 2653 | */ |
c814b4f2 CH |
2654 | static void |
2655 | xlog_state_activate_iclogs( | |
0383f543 | 2656 | struct xlog *log, |
c814b4f2 | 2657 | int *iclogs_changed) |
1da177e4 | 2658 | { |
c814b4f2 | 2659 | struct xlog_in_core *iclog = log->l_iclog; |
1da177e4 | 2660 | |
1da177e4 | 2661 | do { |
c814b4f2 CH |
2662 | if (iclog->ic_state == XLOG_STATE_DIRTY) |
2663 | xlog_state_activate_iclog(iclog, iclogs_changed); | |
2664 | /* | |
2665 | * The ordering of marking iclogs ACTIVE must be maintained, so | |
2666 | * an iclog doesn't become ACTIVE beyond one that is SYNCING. | |
2667 | */ | |
2668 | else if (iclog->ic_state != XLOG_STATE_ACTIVE) | |
2669 | break; | |
2670 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
2671 | } | |
0383f543 | 2672 | |
c814b4f2 CH |
2673 | static int |
2674 | xlog_covered_state( | |
2675 | int prev_state, | |
2676 | int iclogs_changed) | |
2677 | { | |
0383f543 | 2678 | /* |
b0eb9e11 BF |
2679 | * We go to NEED for any non-covering writes. We go to NEED2 if we just |
2680 | * wrote the first covering record (DONE). We go to IDLE if we just | |
2681 | * wrote the second covering record (DONE2) and remain in IDLE until a | |
2682 | * non-covering write occurs. | |
0383f543 | 2683 | */ |
c814b4f2 CH |
2684 | switch (prev_state) { |
2685 | case XLOG_STATE_COVER_IDLE: | |
b0eb9e11 BF |
2686 | if (iclogs_changed == 1) |
2687 | return XLOG_STATE_COVER_IDLE; | |
53004ee7 | 2688 | fallthrough; |
c814b4f2 CH |
2689 | case XLOG_STATE_COVER_NEED: |
2690 | case XLOG_STATE_COVER_NEED2: | |
2691 | break; | |
2692 | case XLOG_STATE_COVER_DONE: | |
2693 | if (iclogs_changed == 1) | |
2694 | return XLOG_STATE_COVER_NEED2; | |
2695 | break; | |
2696 | case XLOG_STATE_COVER_DONE2: | |
2697 | if (iclogs_changed == 1) | |
2698 | return XLOG_STATE_COVER_IDLE; | |
2699 | break; | |
2700 | default: | |
2701 | ASSERT(0); | |
2702 | } | |
0383f543 | 2703 | |
c814b4f2 CH |
2704 | return XLOG_STATE_COVER_NEED; |
2705 | } | |
1da177e4 | 2706 | |
c814b4f2 CH |
2707 | STATIC void |
2708 | xlog_state_clean_iclog( | |
2709 | struct xlog *log, | |
2710 | struct xlog_in_core *dirty_iclog) | |
2711 | { | |
2712 | int iclogs_changed = 0; | |
1da177e4 | 2713 | |
956f6daa DC |
2714 | trace_xlog_iclog_clean(dirty_iclog, _RET_IP_); |
2715 | ||
5781464b | 2716 | dirty_iclog->ic_state = XLOG_STATE_DIRTY; |
1da177e4 | 2717 | |
c814b4f2 CH |
2718 | xlog_state_activate_iclogs(log, &iclogs_changed); |
2719 | wake_up_all(&dirty_iclog->ic_force_wait); | |
2720 | ||
2721 | if (iclogs_changed) { | |
2722 | log->l_covered_state = xlog_covered_state(log->l_covered_state, | |
2723 | iclogs_changed); | |
1da177e4 | 2724 | } |
0383f543 | 2725 | } |
1da177e4 LT |
2726 | |
2727 | STATIC xfs_lsn_t | |
2728 | xlog_get_lowest_lsn( | |
9bff3132 | 2729 | struct xlog *log) |
1da177e4 | 2730 | { |
9bff3132 CH |
2731 | struct xlog_in_core *iclog = log->l_iclog; |
2732 | xfs_lsn_t lowest_lsn = 0, lsn; | |
1da177e4 | 2733 | |
1da177e4 | 2734 | do { |
1858bb0b CH |
2735 | if (iclog->ic_state == XLOG_STATE_ACTIVE || |
2736 | iclog->ic_state == XLOG_STATE_DIRTY) | |
9bff3132 CH |
2737 | continue; |
2738 | ||
2739 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2740 | if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0) | |
1da177e4 | 2741 | lowest_lsn = lsn; |
9bff3132 CH |
2742 | } while ((iclog = iclog->ic_next) != log->l_iclog); |
2743 | ||
014c2544 | 2744 | return lowest_lsn; |
1da177e4 LT |
2745 | } |
2746 | ||
14e15f1b DC |
2747 | /* |
2748 | * Completion of a iclog IO does not imply that a transaction has completed, as | |
2749 | * transactions can be large enough to span many iclogs. We cannot change the | |
2750 | * tail of the log half way through a transaction as this may be the only | |
2751 | * transaction in the log and moving the tail to point to the middle of it | |
2752 | * will prevent recovery from finding the start of the transaction. Hence we | |
2753 | * should only update the last_sync_lsn if this iclog contains transaction | |
2754 | * completion callbacks on it. | |
2755 | * | |
2756 | * We have to do this before we drop the icloglock to ensure we are the only one | |
2757 | * that can update it. | |
2758 | * | |
2759 | * If we are moving the last_sync_lsn forwards, we also need to ensure we kick | |
2760 | * the reservation grant head pushing. This is due to the fact that the push | |
2761 | * target is bound by the current last_sync_lsn value. Hence if we have a large | |
2762 | * amount of log space bound up in this committing transaction then the | |
2763 | * last_sync_lsn value may be the limiting factor preventing tail pushing from | |
2764 | * freeing space in the log. Hence once we've updated the last_sync_lsn we | |
2765 | * should push the AIL to ensure the push target (and hence the grant head) is | |
2766 | * no longer bound by the old log head location and can move forwards and make | |
2767 | * progress again. | |
2768 | */ | |
2769 | static void | |
2770 | xlog_state_set_callback( | |
2771 | struct xlog *log, | |
2772 | struct xlog_in_core *iclog, | |
2773 | xfs_lsn_t header_lsn) | |
2774 | { | |
956f6daa | 2775 | trace_xlog_iclog_callback(iclog, _RET_IP_); |
14e15f1b DC |
2776 | iclog->ic_state = XLOG_STATE_CALLBACK; |
2777 | ||
2778 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), | |
2779 | header_lsn) <= 0); | |
2780 | ||
2781 | if (list_empty_careful(&iclog->ic_callbacks)) | |
2782 | return; | |
2783 | ||
2784 | atomic64_set(&log->l_last_sync_lsn, header_lsn); | |
2785 | xlog_grant_push_ail(log, 0); | |
2786 | } | |
2787 | ||
5e96fa8d DC |
2788 | /* |
2789 | * Return true if we need to stop processing, false to continue to the next | |
2790 | * iclog. The caller will need to run callbacks if the iclog is returned in the | |
2791 | * XLOG_STATE_CALLBACK state. | |
2792 | */ | |
2793 | static bool | |
2794 | xlog_state_iodone_process_iclog( | |
2795 | struct xlog *log, | |
2796 | struct xlog_in_core *iclog, | |
5e96fa8d DC |
2797 | bool *ioerror) |
2798 | { | |
2799 | xfs_lsn_t lowest_lsn; | |
14e15f1b | 2800 | xfs_lsn_t header_lsn; |
5e96fa8d | 2801 | |
1858bb0b CH |
2802 | switch (iclog->ic_state) { |
2803 | case XLOG_STATE_ACTIVE: | |
2804 | case XLOG_STATE_DIRTY: | |
2805 | /* | |
2806 | * Skip all iclogs in the ACTIVE & DIRTY states: | |
2807 | */ | |
5e96fa8d | 2808 | return false; |
1858bb0b CH |
2809 | case XLOG_STATE_IOERROR: |
2810 | /* | |
2811 | * Between marking a filesystem SHUTDOWN and stopping the log, | |
2812 | * we do flush all iclogs to disk (if there wasn't a log I/O | |
2813 | * error). So, we do want things to go smoothly in case of just | |
4b29ab04 | 2814 | * a SHUTDOWN w/o a LOG_IO_ERROR. |
1858bb0b | 2815 | */ |
5e96fa8d DC |
2816 | *ioerror = true; |
2817 | return false; | |
1858bb0b | 2818 | case XLOG_STATE_DONE_SYNC: |
1858bb0b | 2819 | /* |
4b29ab04 CH |
2820 | * Now that we have an iclog that is in the DONE_SYNC state, do |
2821 | * one more check here to see if we have chased our tail around. | |
2822 | * If this is not the lowest lsn iclog, then we will leave it | |
2823 | * for another completion to process. | |
1858bb0b CH |
2824 | */ |
2825 | header_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2826 | lowest_lsn = xlog_get_lowest_lsn(log); | |
2827 | if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, header_lsn) < 0) | |
2828 | return false; | |
2829 | xlog_state_set_callback(log, iclog, header_lsn); | |
2830 | return false; | |
2831 | default: | |
2832 | /* | |
2833 | * Can only perform callbacks in order. Since this iclog is not | |
4b29ab04 CH |
2834 | * in the DONE_SYNC state, we skip the rest and just try to |
2835 | * clean up. | |
1858bb0b | 2836 | */ |
5e96fa8d DC |
2837 | return true; |
2838 | } | |
5e96fa8d DC |
2839 | } |
2840 | ||
1da177e4 LT |
2841 | STATIC void |
2842 | xlog_state_do_callback( | |
12e6a0f4 | 2843 | struct xlog *log) |
1da177e4 | 2844 | { |
5e96fa8d DC |
2845 | struct xlog_in_core *iclog; |
2846 | struct xlog_in_core *first_iclog; | |
5e96fa8d DC |
2847 | bool cycled_icloglock; |
2848 | bool ioerror; | |
2849 | int flushcnt = 0; | |
2850 | int repeats = 0; | |
1da177e4 | 2851 | |
b22cd72c | 2852 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2853 | do { |
2854 | /* | |
2855 | * Scan all iclogs starting with the one pointed to by the | |
2856 | * log. Reset this starting point each time the log is | |
2857 | * unlocked (during callbacks). | |
2858 | * | |
2859 | * Keep looping through iclogs until one full pass is made | |
2860 | * without running any callbacks. | |
2861 | */ | |
2862 | first_iclog = log->l_iclog; | |
2863 | iclog = log->l_iclog; | |
6546818c | 2864 | cycled_icloglock = false; |
5e96fa8d | 2865 | ioerror = false; |
1da177e4 LT |
2866 | repeats++; |
2867 | ||
2868 | do { | |
a1bb8505 DC |
2869 | LIST_HEAD(cb_list); |
2870 | ||
5e96fa8d | 2871 | if (xlog_state_iodone_process_iclog(log, iclog, |
4b29ab04 | 2872 | &ioerror)) |
5e96fa8d | 2873 | break; |
1da177e4 | 2874 | |
1858bb0b CH |
2875 | if (iclog->ic_state != XLOG_STATE_CALLBACK && |
2876 | iclog->ic_state != XLOG_STATE_IOERROR) { | |
1da177e4 LT |
2877 | iclog = iclog->ic_next; |
2878 | continue; | |
2879 | } | |
a1bb8505 | 2880 | list_splice_init(&iclog->ic_callbacks, &cb_list); |
6be00102 | 2881 | spin_unlock(&log->l_icloglock); |
1da177e4 | 2882 | |
a1bb8505 DC |
2883 | trace_xlog_iclog_callbacks_start(iclog, _RET_IP_); |
2884 | xlog_cil_process_committed(&cb_list); | |
2885 | trace_xlog_iclog_callbacks_done(iclog, _RET_IP_); | |
6546818c | 2886 | cycled_icloglock = true; |
6be00102 DC |
2887 | |
2888 | spin_lock(&log->l_icloglock); | |
5781464b CH |
2889 | if (XLOG_FORCED_SHUTDOWN(log)) |
2890 | wake_up_all(&iclog->ic_force_wait); | |
2891 | else | |
2892 | xlog_state_clean_iclog(log, iclog); | |
1da177e4 LT |
2893 | iclog = iclog->ic_next; |
2894 | } while (first_iclog != iclog); | |
a3c6685e NS |
2895 | |
2896 | if (repeats > 5000) { | |
2897 | flushcnt += repeats; | |
2898 | repeats = 0; | |
a0fa2b67 | 2899 | xfs_warn(log->l_mp, |
a3c6685e | 2900 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2901 | __func__, flushcnt); |
1da177e4 | 2902 | } |
5e96fa8d | 2903 | } while (!ioerror && cycled_icloglock); |
1da177e4 | 2904 | |
1858bb0b CH |
2905 | if (log->l_iclog->ic_state == XLOG_STATE_ACTIVE || |
2906 | log->l_iclog->ic_state == XLOG_STATE_IOERROR) | |
eb40a875 | 2907 | wake_up_all(&log->l_flush_wait); |
cdea5459 RR |
2908 | |
2909 | spin_unlock(&log->l_icloglock); | |
d748c623 | 2910 | } |
1da177e4 LT |
2911 | |
2912 | ||
2913 | /* | |
2914 | * Finish transitioning this iclog to the dirty state. | |
2915 | * | |
2916 | * Make sure that we completely execute this routine only when this is | |
2917 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2918 | * when we reach the end of the physical log, get turned into 2 separate | |
2919 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2920 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2921 | * the second completion goes through. | |
2922 | * | |
2923 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2924 | * global state machine log lock. |
1da177e4 | 2925 | */ |
a8272ce0 | 2926 | STATIC void |
1da177e4 | 2927 | xlog_state_done_syncing( |
12e6a0f4 | 2928 | struct xlog_in_core *iclog) |
1da177e4 | 2929 | { |
d15cbf2f | 2930 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2931 | |
b22cd72c | 2932 | spin_lock(&log->l_icloglock); |
155cc6b7 | 2933 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
956f6daa | 2934 | trace_xlog_iclog_sync_done(iclog, _RET_IP_); |
1da177e4 LT |
2935 | |
2936 | /* | |
2937 | * If we got an error, either on the first buffer, or in the case of | |
12e6a0f4 CH |
2938 | * split log writes, on the second, we shut down the file system and |
2939 | * no iclogs should ever be attempted to be written to disk again. | |
1da177e4 | 2940 | */ |
12e6a0f4 CH |
2941 | if (!XLOG_FORCED_SHUTDOWN(log)) { |
2942 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING); | |
1da177e4 | 2943 | iclog->ic_state = XLOG_STATE_DONE_SYNC; |
12e6a0f4 | 2944 | } |
1da177e4 LT |
2945 | |
2946 | /* | |
2947 | * Someone could be sleeping prior to writing out the next | |
2948 | * iclog buffer, we wake them all, one will get to do the | |
2949 | * I/O, the others get to wait for the result. | |
2950 | */ | |
eb40a875 | 2951 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2952 | spin_unlock(&log->l_icloglock); |
b843299b | 2953 | xlog_state_do_callback(log); |
12e6a0f4 | 2954 | } |
1da177e4 LT |
2955 | |
2956 | /* | |
2957 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2958 | * sleep. We wait on the flush queue on the head iclog as that should be |
2959 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2960 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2961 | * |
2962 | * The in-core logs are used in a circular fashion. They are not used | |
2963 | * out-of-order even when an iclog past the head is free. | |
2964 | * | |
2965 | * return: | |
2966 | * * log_offset where xlog_write() can start writing into the in-core | |
2967 | * log's data space. | |
2968 | * * in-core log pointer to which xlog_write() should write. | |
2969 | * * boolean indicating this is a continued write to an in-core log. | |
2970 | * If this is the last write, then the in-core log's offset field | |
2971 | * needs to be incremented, depending on the amount of data which | |
2972 | * is copied. | |
2973 | */ | |
a8272ce0 | 2974 | STATIC int |
9a8d2fdb MT |
2975 | xlog_state_get_iclog_space( |
2976 | struct xlog *log, | |
2977 | int len, | |
2978 | struct xlog_in_core **iclogp, | |
2979 | struct xlog_ticket *ticket, | |
2980 | int *continued_write, | |
2981 | int *logoffsetp) | |
1da177e4 | 2982 | { |
1da177e4 LT |
2983 | int log_offset; |
2984 | xlog_rec_header_t *head; | |
2985 | xlog_in_core_t *iclog; | |
1da177e4 LT |
2986 | |
2987 | restart: | |
b22cd72c | 2988 | spin_lock(&log->l_icloglock); |
1da177e4 | 2989 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2990 | spin_unlock(&log->l_icloglock); |
2451337d | 2991 | return -EIO; |
1da177e4 LT |
2992 | } |
2993 | ||
2994 | iclog = log->l_iclog; | |
d748c623 | 2995 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
ff6d6af2 | 2996 | XFS_STATS_INC(log->l_mp, xs_log_noiclogs); |
d748c623 MW |
2997 | |
2998 | /* Wait for log writes to have flushed */ | |
eb40a875 | 2999 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
3000 | goto restart; |
3001 | } | |
d748c623 | 3002 | |
1da177e4 LT |
3003 | head = &iclog->ic_header; |
3004 | ||
155cc6b7 | 3005 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
3006 | log_offset = iclog->ic_offset; |
3007 | ||
956f6daa DC |
3008 | trace_xlog_iclog_get_space(iclog, _RET_IP_); |
3009 | ||
1da177e4 LT |
3010 | /* On the 1st write to an iclog, figure out lsn. This works |
3011 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
3012 | * committing to. If the offset is set, that's how many blocks | |
3013 | * must be written. | |
3014 | */ | |
3015 | if (log_offset == 0) { | |
3016 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 3017 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
3018 | log->l_iclog_hsize, |
3019 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
3020 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
3021 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 3022 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
3023 | ASSERT(log->l_curr_block >= 0); |
3024 | } | |
3025 | ||
3026 | /* If there is enough room to write everything, then do it. Otherwise, | |
3027 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
3028 | * bit is on, so this will get flushed out. Don't update ic_offset | |
3029 | * until you know exactly how many bytes get copied. Therefore, wait | |
3030 | * until later to update ic_offset. | |
3031 | * | |
3032 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
3033 | * can fit into remaining data section. | |
3034 | */ | |
3035 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
df732b29 CH |
3036 | int error = 0; |
3037 | ||
1da177e4 LT |
3038 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); |
3039 | ||
49641f1a | 3040 | /* |
df732b29 CH |
3041 | * If we are the only one writing to this iclog, sync it to |
3042 | * disk. We need to do an atomic compare and decrement here to | |
3043 | * avoid racing with concurrent atomic_dec_and_lock() calls in | |
49641f1a DC |
3044 | * xlog_state_release_iclog() when there is more than one |
3045 | * reference to the iclog. | |
3046 | */ | |
df732b29 | 3047 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) |
0dc8f7f1 | 3048 | error = xlog_state_release_iclog(log, iclog, 0); |
df732b29 CH |
3049 | spin_unlock(&log->l_icloglock); |
3050 | if (error) | |
3051 | return error; | |
1da177e4 LT |
3052 | goto restart; |
3053 | } | |
3054 | ||
3055 | /* Do we have enough room to write the full amount in the remainder | |
3056 | * of this iclog? Or must we continue a write on the next iclog and | |
3057 | * mark this iclog as completely taken? In the case where we switch | |
3058 | * iclogs (to mark it taken), this particular iclog will release/sync | |
3059 | * to disk in xlog_write(). | |
3060 | */ | |
3061 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
3062 | *continued_write = 0; | |
3063 | iclog->ic_offset += len; | |
3064 | } else { | |
3065 | *continued_write = 1; | |
3066 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
3067 | } | |
3068 | *iclogp = iclog; | |
3069 | ||
3070 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 3071 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3072 | |
3073 | *logoffsetp = log_offset; | |
3074 | return 0; | |
b843299b | 3075 | } |
1da177e4 | 3076 | |
8b41e3f9 | 3077 | /* |
b843299b DC |
3078 | * The first cnt-1 times a ticket goes through here we don't need to move the |
3079 | * grant write head because the permanent reservation has reserved cnt times the | |
3080 | * unit amount. Release part of current permanent unit reservation and reset | |
3081 | * current reservation to be one units worth. Also move grant reservation head | |
3082 | * forward. | |
1da177e4 | 3083 | */ |
8b41e3f9 CH |
3084 | void |
3085 | xfs_log_ticket_regrant( | |
9a8d2fdb MT |
3086 | struct xlog *log, |
3087 | struct xlog_ticket *ticket) | |
1da177e4 | 3088 | { |
8b41e3f9 | 3089 | trace_xfs_log_ticket_regrant(log, ticket); |
0b1b213f | 3090 | |
1da177e4 LT |
3091 | if (ticket->t_cnt > 0) |
3092 | ticket->t_cnt--; | |
3093 | ||
28496968 | 3094 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3095 | ticket->t_curr_res); |
28496968 | 3096 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 3097 | ticket->t_curr_res); |
1da177e4 | 3098 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3099 | xlog_tic_reset_res(ticket); |
0b1b213f | 3100 | |
8b41e3f9 | 3101 | trace_xfs_log_ticket_regrant_sub(log, ticket); |
0b1b213f | 3102 | |
1da177e4 | 3103 | /* just return if we still have some of the pre-reserved space */ |
8b41e3f9 CH |
3104 | if (!ticket->t_cnt) { |
3105 | xlog_grant_add_space(log, &log->l_reserve_head.grant, | |
3106 | ticket->t_unit_res); | |
3107 | trace_xfs_log_ticket_regrant_exit(log, ticket); | |
1da177e4 | 3108 | |
8b41e3f9 CH |
3109 | ticket->t_curr_res = ticket->t_unit_res; |
3110 | xlog_tic_reset_res(ticket); | |
3111 | } | |
1da177e4 | 3112 | |
8b41e3f9 CH |
3113 | xfs_log_ticket_put(ticket); |
3114 | } | |
1da177e4 LT |
3115 | |
3116 | /* | |
3117 | * Give back the space left from a reservation. | |
3118 | * | |
3119 | * All the information we need to make a correct determination of space left | |
3120 | * is present. For non-permanent reservations, things are quite easy. The | |
3121 | * count should have been decremented to zero. We only need to deal with the | |
3122 | * space remaining in the current reservation part of the ticket. If the | |
3123 | * ticket contains a permanent reservation, there may be left over space which | |
3124 | * needs to be released. A count of N means that N-1 refills of the current | |
3125 | * reservation can be done before we need to ask for more space. The first | |
3126 | * one goes to fill up the first current reservation. Once we run out of | |
3127 | * space, the count will stay at zero and the only space remaining will be | |
3128 | * in the current reservation field. | |
3129 | */ | |
8b41e3f9 CH |
3130 | void |
3131 | xfs_log_ticket_ungrant( | |
9a8d2fdb MT |
3132 | struct xlog *log, |
3133 | struct xlog_ticket *ticket) | |
1da177e4 | 3134 | { |
8b41e3f9 CH |
3135 | int bytes; |
3136 | ||
3137 | trace_xfs_log_ticket_ungrant(log, ticket); | |
663e496a | 3138 | |
1da177e4 LT |
3139 | if (ticket->t_cnt > 0) |
3140 | ticket->t_cnt--; | |
3141 | ||
8b41e3f9 | 3142 | trace_xfs_log_ticket_ungrant_sub(log, ticket); |
1da177e4 | 3143 | |
663e496a DC |
3144 | /* |
3145 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
3146 | * up more space based on the remaining count. |
3147 | */ | |
663e496a | 3148 | bytes = ticket->t_curr_res; |
1da177e4 LT |
3149 | if (ticket->t_cnt > 0) { |
3150 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 3151 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
3152 | } |
3153 | ||
28496968 CH |
3154 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
3155 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 3156 | |
8b41e3f9 | 3157 | trace_xfs_log_ticket_ungrant_exit(log, ticket); |
0b1b213f | 3158 | |
cfb7cdca | 3159 | xfs_log_space_wake(log->l_mp); |
8b41e3f9 | 3160 | xfs_log_ticket_put(ticket); |
09a423a3 | 3161 | } |
1da177e4 | 3162 | |
1da177e4 | 3163 | /* |
b843299b DC |
3164 | * This routine will mark the current iclog in the ring as WANT_SYNC and move |
3165 | * the current iclog pointer to the next iclog in the ring. | |
1da177e4 LT |
3166 | */ |
3167 | STATIC void | |
9a8d2fdb MT |
3168 | xlog_state_switch_iclogs( |
3169 | struct xlog *log, | |
3170 | struct xlog_in_core *iclog, | |
3171 | int eventual_size) | |
1da177e4 LT |
3172 | { |
3173 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
69363999 | 3174 | assert_spin_locked(&log->l_icloglock); |
956f6daa | 3175 | trace_xlog_iclog_switch(iclog, _RET_IP_); |
69363999 | 3176 | |
1da177e4 LT |
3177 | if (!eventual_size) |
3178 | eventual_size = iclog->ic_offset; | |
3179 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3180 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3181 | log->l_prev_block = log->l_curr_block; |
3182 | log->l_prev_cycle = log->l_curr_cycle; | |
3183 | ||
3184 | /* roll log?: ic_offset changed later */ | |
3185 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3186 | ||
3187 | /* Round up to next log-sunit */ | |
a6a65fef | 3188 | if (log->l_iclog_roundoff > BBSIZE) { |
18842e0a | 3189 | uint32_t sunit_bb = BTOBB(log->l_iclog_roundoff); |
1da177e4 LT |
3190 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); |
3191 | } | |
3192 | ||
3193 | if (log->l_curr_block >= log->l_logBBsize) { | |
a45086e2 BF |
3194 | /* |
3195 | * Rewind the current block before the cycle is bumped to make | |
3196 | * sure that the combined LSN never transiently moves forward | |
3197 | * when the log wraps to the next cycle. This is to support the | |
3198 | * unlocked sample of these fields from xlog_valid_lsn(). Most | |
3199 | * other cases should acquire l_icloglock. | |
3200 | */ | |
3201 | log->l_curr_block -= log->l_logBBsize; | |
3202 | ASSERT(log->l_curr_block >= 0); | |
3203 | smp_wmb(); | |
1da177e4 LT |
3204 | log->l_curr_cycle++; |
3205 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3206 | log->l_curr_cycle++; | |
1da177e4 LT |
3207 | } |
3208 | ASSERT(iclog == log->l_iclog); | |
3209 | log->l_iclog = iclog->ic_next; | |
b843299b | 3210 | } |
1da177e4 | 3211 | |
8191d822 DC |
3212 | /* |
3213 | * Force the iclog to disk and check if the iclog has been completed before | |
3214 | * xlog_force_iclog() returns. This can happen on synchronous (e.g. | |
3215 | * pmem) or fast async storage because we drop the icloglock to issue the IO. | |
3216 | * If completion has already occurred, tell the caller so that it can avoid an | |
3217 | * unnecessary wait on the iclog. | |
3218 | */ | |
3219 | static int | |
3220 | xlog_force_and_check_iclog( | |
3221 | struct xlog_in_core *iclog, | |
3222 | bool *completed) | |
3223 | { | |
3224 | xfs_lsn_t lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
3225 | int error; | |
3226 | ||
3227 | *completed = false; | |
3228 | error = xlog_force_iclog(iclog); | |
3229 | if (error) | |
3230 | return error; | |
3231 | ||
3232 | /* | |
3233 | * If the iclog has already been completed and reused the header LSN | |
3234 | * will have been rewritten by completion | |
3235 | */ | |
3236 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) | |
3237 | *completed = true; | |
3238 | return 0; | |
3239 | } | |
3240 | ||
1da177e4 LT |
3241 | /* |
3242 | * Write out all data in the in-core log as of this exact moment in time. | |
3243 | * | |
3244 | * Data may be written to the in-core log during this call. However, | |
3245 | * we don't guarantee this data will be written out. A change from past | |
3246 | * implementation means this routine will *not* write out zero length LRs. | |
3247 | * | |
3248 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3249 | * If we determine there is no flushable data, we just return. There is no | |
3250 | * flushable data if: | |
3251 | * | |
3252 | * 1. the current iclog is active and has no data; the previous iclog | |
3253 | * is in the active or dirty state. | |
3254 | * 2. the current iclog is drity, and the previous iclog is in the | |
3255 | * active or dirty state. | |
3256 | * | |
12017faf | 3257 | * We may sleep if: |
1da177e4 LT |
3258 | * |
3259 | * 1. the current iclog is not in the active nor dirty state. | |
3260 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3261 | * active nor dirty state. | |
3262 | * 3. the current iclog is active, and there is another thread writing | |
3263 | * to this particular iclog. | |
3264 | * 4. a) the current iclog is active and has no other writers | |
3265 | * b) when we return from flushing out this iclog, it is still | |
3266 | * not in the active nor dirty state. | |
3267 | */ | |
a14a348b | 3268 | int |
60e5bb78 | 3269 | xfs_log_force( |
a14a348b | 3270 | struct xfs_mount *mp, |
60e5bb78 | 3271 | uint flags) |
1da177e4 | 3272 | { |
ad223e60 | 3273 | struct xlog *log = mp->m_log; |
a14a348b | 3274 | struct xlog_in_core *iclog; |
a14a348b | 3275 | |
ff6d6af2 | 3276 | XFS_STATS_INC(mp, xs_log_force); |
60e5bb78 | 3277 | trace_xfs_log_force(mp, 0, _RET_IP_); |
1da177e4 | 3278 | |
93b8a585 | 3279 | xlog_cil_force(log); |
71e330b5 | 3280 | |
b22cd72c | 3281 | spin_lock(&log->l_icloglock); |
1da177e4 | 3282 | iclog = log->l_iclog; |
1858bb0b | 3283 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
e6b96570 | 3284 | goto out_error; |
1da177e4 | 3285 | |
956f6daa DC |
3286 | trace_xlog_iclog_force(iclog, _RET_IP_); |
3287 | ||
e6b96570 CH |
3288 | if (iclog->ic_state == XLOG_STATE_DIRTY || |
3289 | (iclog->ic_state == XLOG_STATE_ACTIVE && | |
3290 | atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) { | |
1da177e4 | 3291 | /* |
e6b96570 CH |
3292 | * If the head is dirty or (active and empty), then we need to |
3293 | * look at the previous iclog. | |
3294 | * | |
3295 | * If the previous iclog is active or dirty we are done. There | |
3296 | * is nothing to sync out. Otherwise, we attach ourselves to the | |
1da177e4 LT |
3297 | * previous iclog and go to sleep. |
3298 | */ | |
e6b96570 | 3299 | iclog = iclog->ic_prev; |
e6b96570 CH |
3300 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3301 | if (atomic_read(&iclog->ic_refcnt) == 0) { | |
45eddb41 | 3302 | /* We have exclusive access to this iclog. */ |
8191d822 DC |
3303 | bool completed; |
3304 | ||
3305 | if (xlog_force_and_check_iclog(iclog, &completed)) | |
df732b29 | 3306 | goto out_error; |
1da177e4 | 3307 | |
8191d822 | 3308 | if (completed) |
e6b96570 CH |
3309 | goto out_unlock; |
3310 | } else { | |
3311 | /* | |
2bf1ec0f DC |
3312 | * Someone else is still writing to this iclog, so we |
3313 | * need to ensure that when they release the iclog it | |
3314 | * gets synced immediately as we may be waiting on it. | |
e6b96570 CH |
3315 | */ |
3316 | xlog_state_switch_iclogs(log, iclog, 0); | |
1da177e4 | 3317 | } |
1da177e4 | 3318 | } |
e6b96570 | 3319 | |
2bf1ec0f DC |
3320 | /* |
3321 | * The iclog we are about to wait on may contain the checkpoint pushed | |
3322 | * by the above xlog_cil_force() call, but it may not have been pushed | |
3323 | * to disk yet. Like the ACTIVE case above, we need to make sure caches | |
3324 | * are flushed when this iclog is written. | |
3325 | */ | |
3326 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) | |
3327 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA; | |
3328 | ||
81e5b50a CH |
3329 | if (flags & XFS_LOG_SYNC) |
3330 | return xlog_wait_on_iclog(iclog); | |
e6b96570 CH |
3331 | out_unlock: |
3332 | spin_unlock(&log->l_icloglock); | |
3333 | return 0; | |
3334 | out_error: | |
3335 | spin_unlock(&log->l_icloglock); | |
3336 | return -EIO; | |
a14a348b | 3337 | } |
1da177e4 | 3338 | |
3e4da466 | 3339 | static int |
5f9b4b0d DC |
3340 | xlog_force_lsn( |
3341 | struct xlog *log, | |
a14a348b CH |
3342 | xfs_lsn_t lsn, |
3343 | uint flags, | |
3e4da466 CH |
3344 | int *log_flushed, |
3345 | bool already_slept) | |
1da177e4 | 3346 | { |
a14a348b | 3347 | struct xlog_in_core *iclog; |
8191d822 | 3348 | bool completed; |
71e330b5 | 3349 | |
a14a348b CH |
3350 | spin_lock(&log->l_icloglock); |
3351 | iclog = log->l_iclog; | |
1858bb0b | 3352 | if (iclog->ic_state == XLOG_STATE_IOERROR) |
93806299 | 3353 | goto out_error; |
1da177e4 | 3354 | |
93806299 | 3355 | while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { |
956f6daa | 3356 | trace_xlog_iclog_force_lsn(iclog, _RET_IP_); |
93806299 CH |
3357 | iclog = iclog->ic_next; |
3358 | if (iclog == log->l_iclog) | |
3359 | goto out_unlock; | |
3360 | } | |
a14a348b | 3361 | |
2bf1ec0f DC |
3362 | switch (iclog->ic_state) { |
3363 | case XLOG_STATE_ACTIVE: | |
93806299 CH |
3364 | /* |
3365 | * We sleep here if we haven't already slept (e.g. this is the | |
3366 | * first time we've looked at the correct iclog buf) and the | |
3367 | * buffer before us is going to be sync'ed. The reason for this | |
3368 | * is that if we are doing sync transactions here, by waiting | |
3369 | * for the previous I/O to complete, we can allow a few more | |
3370 | * transactions into this iclog before we close it down. | |
3371 | * | |
3372 | * Otherwise, we mark the buffer WANT_SYNC, and bump up the | |
3373 | * refcnt so we can release the log (which drops the ref count). | |
3374 | * The state switch keeps new transaction commits from using | |
3375 | * this buffer. When the current commits finish writing into | |
3376 | * the buffer, the refcount will drop to zero and the buffer | |
3377 | * will go out then. | |
3378 | */ | |
3379 | if (!already_slept && | |
1858bb0b CH |
3380 | (iclog->ic_prev->ic_state == XLOG_STATE_WANT_SYNC || |
3381 | iclog->ic_prev->ic_state == XLOG_STATE_SYNCING)) { | |
93806299 CH |
3382 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3383 | &log->l_icloglock); | |
3e4da466 | 3384 | return -EAGAIN; |
1da177e4 | 3385 | } |
8191d822 | 3386 | if (xlog_force_and_check_iclog(iclog, &completed)) |
df732b29 | 3387 | goto out_error; |
93806299 CH |
3388 | if (log_flushed) |
3389 | *log_flushed = 1; | |
8191d822 DC |
3390 | if (completed) |
3391 | goto out_unlock; | |
2bf1ec0f DC |
3392 | break; |
3393 | case XLOG_STATE_WANT_SYNC: | |
3394 | /* | |
3395 | * This iclog may contain the checkpoint pushed by the | |
3396 | * xlog_cil_force_seq() call, but there are other writers still | |
3397 | * accessing it so it hasn't been pushed to disk yet. Like the | |
3398 | * ACTIVE case above, we need to make sure caches are flushed | |
3399 | * when this iclog is written. | |
3400 | */ | |
3401 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA; | |
3402 | break; | |
3403 | default: | |
3404 | /* | |
3405 | * The entire checkpoint was written by the CIL force and is on | |
3406 | * its way to disk already. It will be stable when it | |
3407 | * completes, so we don't need to manipulate caches here at all. | |
3408 | * We just need to wait for completion if necessary. | |
3409 | */ | |
3410 | break; | |
93806299 | 3411 | } |
1da177e4 | 3412 | |
81e5b50a CH |
3413 | if (flags & XFS_LOG_SYNC) |
3414 | return xlog_wait_on_iclog(iclog); | |
93806299 | 3415 | out_unlock: |
a14a348b CH |
3416 | spin_unlock(&log->l_icloglock); |
3417 | return 0; | |
93806299 CH |
3418 | out_error: |
3419 | spin_unlock(&log->l_icloglock); | |
3420 | return -EIO; | |
a14a348b CH |
3421 | } |
3422 | ||
3e4da466 CH |
3423 | /* |
3424 | * Force the in-core log to disk for a specific LSN. | |
3425 | * | |
3426 | * Find in-core log with lsn. | |
3427 | * If it is in the DIRTY state, just return. | |
3428 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3429 | * state and go to sleep or return. | |
3430 | * If it is in any other state, go to sleep or return. | |
3431 | * | |
3432 | * Synchronous forces are implemented with a wait queue. All callers trying | |
3433 | * to force a given lsn to disk must wait on the queue attached to the | |
3434 | * specific in-core log. When given in-core log finally completes its write | |
3435 | * to disk, that thread will wake up all threads waiting on the queue. | |
3436 | */ | |
3437 | int | |
5f9b4b0d | 3438 | xfs_log_force_seq( |
3e4da466 | 3439 | struct xfs_mount *mp, |
5f9b4b0d | 3440 | xfs_csn_t seq, |
3e4da466 CH |
3441 | uint flags, |
3442 | int *log_flushed) | |
3443 | { | |
5f9b4b0d DC |
3444 | struct xlog *log = mp->m_log; |
3445 | xfs_lsn_t lsn; | |
3e4da466 | 3446 | int ret; |
5f9b4b0d | 3447 | ASSERT(seq != 0); |
3e4da466 CH |
3448 | |
3449 | XFS_STATS_INC(mp, xs_log_force); | |
5f9b4b0d | 3450 | trace_xfs_log_force(mp, seq, _RET_IP_); |
3e4da466 | 3451 | |
5f9b4b0d | 3452 | lsn = xlog_cil_force_seq(log, seq); |
3e4da466 CH |
3453 | if (lsn == NULLCOMMITLSN) |
3454 | return 0; | |
3455 | ||
5f9b4b0d DC |
3456 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, false); |
3457 | if (ret == -EAGAIN) { | |
3458 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3459 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, true); | |
3460 | } | |
3e4da466 CH |
3461 | return ret; |
3462 | } | |
3463 | ||
1da177e4 | 3464 | /* |
9da096fd | 3465 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3466 | */ |
cc09c0dc DC |
3467 | void |
3468 | xfs_log_ticket_put( | |
3469 | xlog_ticket_t *ticket) | |
1da177e4 | 3470 | { |
cc09c0dc | 3471 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3472 | if (atomic_dec_and_test(&ticket->t_ref)) |
377bcd5f | 3473 | kmem_cache_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3474 | } |
1da177e4 | 3475 | |
cc09c0dc DC |
3476 | xlog_ticket_t * |
3477 | xfs_log_ticket_get( | |
3478 | xlog_ticket_t *ticket) | |
3479 | { | |
3480 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3481 | atomic_inc(&ticket->t_ref); | |
3482 | return ticket; | |
3483 | } | |
1da177e4 LT |
3484 | |
3485 | /* | |
e773fc93 JL |
3486 | * Figure out the total log space unit (in bytes) that would be |
3487 | * required for a log ticket. | |
1da177e4 | 3488 | */ |
a6a65fef DC |
3489 | static int |
3490 | xlog_calc_unit_res( | |
3491 | struct xlog *log, | |
e773fc93 | 3492 | int unit_bytes) |
1da177e4 | 3493 | { |
e773fc93 JL |
3494 | int iclog_space; |
3495 | uint num_headers; | |
1da177e4 LT |
3496 | |
3497 | /* | |
3498 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3499 | * in the log. A unit in this case is the amount of space for one | |
3500 | * of these log operations. Normal reservations have a cnt of 1 | |
3501 | * and their unit amount is the total amount of space required. | |
3502 | * | |
3503 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3504 | * which occupy space in the on-disk log. |
3505 | * | |
3506 | * Normal form of a transaction is: | |
3507 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3508 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3509 | * | |
3510 | * We need to account for all the leadup data and trailer data | |
3511 | * around the transaction data. | |
3512 | * And then we need to account for the worst case in terms of using | |
3513 | * more space. | |
3514 | * The worst case will happen if: | |
3515 | * - the placement of the transaction happens to be such that the | |
3516 | * roundoff is at its maximum | |
3517 | * - the transaction data is synced before the commit record is synced | |
3518 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3519 | * Therefore the commit record is in its own Log Record. | |
3520 | * This can happen as the commit record is called with its | |
3521 | * own region to xlog_write(). | |
3522 | * This then means that in the worst case, roundoff can happen for | |
3523 | * the commit-rec as well. | |
3524 | * The commit-rec is smaller than padding in this scenario and so it is | |
3525 | * not added separately. | |
1da177e4 LT |
3526 | */ |
3527 | ||
32fb9b57 TS |
3528 | /* for trans header */ |
3529 | unit_bytes += sizeof(xlog_op_header_t); | |
3530 | unit_bytes += sizeof(xfs_trans_header_t); | |
3531 | ||
1da177e4 | 3532 | /* for start-rec */ |
32fb9b57 TS |
3533 | unit_bytes += sizeof(xlog_op_header_t); |
3534 | ||
9b9fc2b7 DC |
3535 | /* |
3536 | * for LR headers - the space for data in an iclog is the size minus | |
3537 | * the space used for the headers. If we use the iclog size, then we | |
3538 | * undercalculate the number of headers required. | |
3539 | * | |
3540 | * Furthermore - the addition of op headers for split-recs might | |
3541 | * increase the space required enough to require more log and op | |
3542 | * headers, so take that into account too. | |
3543 | * | |
3544 | * IMPORTANT: This reservation makes the assumption that if this | |
3545 | * transaction is the first in an iclog and hence has the LR headers | |
3546 | * accounted to it, then the remaining space in the iclog is | |
3547 | * exclusively for this transaction. i.e. if the transaction is larger | |
3548 | * than the iclog, it will be the only thing in that iclog. | |
3549 | * Fundamentally, this means we must pass the entire log vector to | |
3550 | * xlog_write to guarantee this. | |
3551 | */ | |
3552 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3553 | num_headers = howmany(unit_bytes, iclog_space); | |
3554 | ||
3555 | /* for split-recs - ophdrs added when data split over LRs */ | |
3556 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3557 | ||
3558 | /* add extra header reservations if we overrun */ | |
3559 | while (!num_headers || | |
3560 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3561 | unit_bytes += sizeof(xlog_op_header_t); | |
3562 | num_headers++; | |
3563 | } | |
32fb9b57 | 3564 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3565 | |
32fb9b57 TS |
3566 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3567 | unit_bytes += log->l_iclog_hsize; | |
3568 | ||
a6a65fef DC |
3569 | /* roundoff padding for transaction data and one for commit record */ |
3570 | unit_bytes += 2 * log->l_iclog_roundoff; | |
1da177e4 | 3571 | |
e773fc93 JL |
3572 | return unit_bytes; |
3573 | } | |
3574 | ||
a6a65fef DC |
3575 | int |
3576 | xfs_log_calc_unit_res( | |
3577 | struct xfs_mount *mp, | |
3578 | int unit_bytes) | |
3579 | { | |
3580 | return xlog_calc_unit_res(mp->m_log, unit_bytes); | |
3581 | } | |
3582 | ||
e773fc93 JL |
3583 | /* |
3584 | * Allocate and initialise a new log ticket. | |
3585 | */ | |
3586 | struct xlog_ticket * | |
3587 | xlog_ticket_alloc( | |
3588 | struct xlog *log, | |
3589 | int unit_bytes, | |
3590 | int cnt, | |
3591 | char client, | |
ca4f2589 | 3592 | bool permanent) |
e773fc93 JL |
3593 | { |
3594 | struct xlog_ticket *tic; | |
3595 | int unit_res; | |
3596 | ||
ca4f2589 | 3597 | tic = kmem_cache_zalloc(xfs_log_ticket_zone, GFP_NOFS | __GFP_NOFAIL); |
e773fc93 | 3598 | |
a6a65fef | 3599 | unit_res = xlog_calc_unit_res(log, unit_bytes); |
e773fc93 | 3600 | |
cc09c0dc | 3601 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3602 | tic->t_task = current; |
10547941 | 3603 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3604 | tic->t_unit_res = unit_res; |
3605 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3606 | tic->t_cnt = cnt; |
3607 | tic->t_ocnt = cnt; | |
ecb3403d | 3608 | tic->t_tid = prandom_u32(); |
1da177e4 | 3609 | tic->t_clientid = client; |
9006fb91 | 3610 | if (permanent) |
1da177e4 | 3611 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3612 | |
0adba536 | 3613 | xlog_tic_reset_res(tic); |
7e9c6396 | 3614 | |
1da177e4 | 3615 | return tic; |
cc09c0dc | 3616 | } |
1da177e4 | 3617 | |
cfcbbbd0 | 3618 | #if defined(DEBUG) |
1da177e4 LT |
3619 | /* |
3620 | * Make sure that the destination ptr is within the valid data region of | |
3621 | * one of the iclogs. This uses backup pointers stored in a different | |
3622 | * part of the log in case we trash the log structure. | |
3623 | */ | |
181fdfe6 | 3624 | STATIC void |
e6b1f273 | 3625 | xlog_verify_dest_ptr( |
ad223e60 | 3626 | struct xlog *log, |
5809d5e0 | 3627 | void *ptr) |
1da177e4 LT |
3628 | { |
3629 | int i; | |
3630 | int good_ptr = 0; | |
3631 | ||
e6b1f273 CH |
3632 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3633 | if (ptr >= log->l_iclog_bak[i] && | |
3634 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3635 | good_ptr++; |
3636 | } | |
e6b1f273 CH |
3637 | |
3638 | if (!good_ptr) | |
a0fa2b67 | 3639 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3640 | } |
1da177e4 | 3641 | |
da8a1a4a DC |
3642 | /* |
3643 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3644 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3645 | * the cycles differ by exactly one and check the byte count. | |
3646 | * | |
3647 | * This check is run unlocked, so can give false positives. Rather than assert | |
3648 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3649 | * determine if they want to panic the machine when such an error occurs. For | |
3650 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3651 | * an assert, it can be turned off at runtime. | |
3652 | */ | |
3f336c6f DC |
3653 | STATIC void |
3654 | xlog_verify_grant_tail( | |
ad223e60 | 3655 | struct xlog *log) |
3f336c6f | 3656 | { |
1c3cb9ec | 3657 | int tail_cycle, tail_blocks; |
a69ed03c | 3658 | int cycle, space; |
3f336c6f | 3659 | |
28496968 | 3660 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3661 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3662 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3663 | if (cycle - 1 != tail_cycle && |
3664 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3665 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3666 | "%s: cycle - 1 != tail_cycle", __func__); | |
3667 | log->l_flags |= XLOG_TAIL_WARN; | |
3668 | } | |
3669 | ||
3670 | if (space > BBTOB(tail_blocks) && | |
3671 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3672 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3673 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3674 | log->l_flags |= XLOG_TAIL_WARN; | |
3675 | } | |
3f336c6f DC |
3676 | } |
3677 | } | |
3678 | ||
1da177e4 LT |
3679 | /* check if it will fit */ |
3680 | STATIC void | |
9a8d2fdb MT |
3681 | xlog_verify_tail_lsn( |
3682 | struct xlog *log, | |
9d110014 | 3683 | struct xlog_in_core *iclog) |
1da177e4 | 3684 | { |
9d110014 DC |
3685 | xfs_lsn_t tail_lsn = be64_to_cpu(iclog->ic_header.h_tail_lsn); |
3686 | int blocks; | |
1da177e4 LT |
3687 | |
3688 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3689 | blocks = | |
3690 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3691 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3692 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3693 | } else { |
3694 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3695 | ||
3696 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3697 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3698 | |
3699 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3700 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3701 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 | 3702 | } |
b843299b | 3703 | } |
1da177e4 LT |
3704 | |
3705 | /* | |
3706 | * Perform a number of checks on the iclog before writing to disk. | |
3707 | * | |
3708 | * 1. Make sure the iclogs are still circular | |
3709 | * 2. Make sure we have a good magic number | |
3710 | * 3. Make sure we don't have magic numbers in the data | |
3711 | * 4. Check fields of each log operation header for: | |
3712 | * A. Valid client identifier | |
3713 | * B. tid ptr value falls in valid ptr space (user space code) | |
3714 | * C. Length in log record header is correct according to the | |
3715 | * individual operation headers within record. | |
3716 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3717 | * log, check the preceding blocks of the physical log to make sure all | |
3718 | * the cycle numbers agree with the current cycle number. | |
3719 | */ | |
3720 | STATIC void | |
9a8d2fdb MT |
3721 | xlog_verify_iclog( |
3722 | struct xlog *log, | |
3723 | struct xlog_in_core *iclog, | |
abca1f33 | 3724 | int count) |
1da177e4 LT |
3725 | { |
3726 | xlog_op_header_t *ophead; | |
3727 | xlog_in_core_t *icptr; | |
3728 | xlog_in_core_2_t *xhdr; | |
5809d5e0 | 3729 | void *base_ptr, *ptr, *p; |
db9d67d6 | 3730 | ptrdiff_t field_offset; |
c8ce540d | 3731 | uint8_t clientid; |
1da177e4 LT |
3732 | int len, i, j, k, op_len; |
3733 | int idx; | |
1da177e4 LT |
3734 | |
3735 | /* check validity of iclog pointers */ | |
b22cd72c | 3736 | spin_lock(&log->l_icloglock); |
1da177e4 | 3737 | icptr = log->l_iclog; |
643f7c4e GB |
3738 | for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next) |
3739 | ASSERT(icptr); | |
3740 | ||
1da177e4 | 3741 | if (icptr != log->l_iclog) |
a0fa2b67 | 3742 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3743 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3744 | |
3745 | /* check log magic numbers */ | |
69ef921b | 3746 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3747 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3748 | |
5809d5e0 CH |
3749 | base_ptr = ptr = &iclog->ic_header; |
3750 | p = &iclog->ic_header; | |
3751 | for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) { | |
69ef921b | 3752 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3753 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3754 | __func__); | |
1da177e4 LT |
3755 | } |
3756 | ||
3757 | /* check fields */ | |
b53e675d | 3758 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
5809d5e0 CH |
3759 | base_ptr = ptr = iclog->ic_datap; |
3760 | ophead = ptr; | |
b28708d6 | 3761 | xhdr = iclog->ic_data; |
1da177e4 | 3762 | for (i = 0; i < len; i++) { |
5809d5e0 | 3763 | ophead = ptr; |
1da177e4 LT |
3764 | |
3765 | /* clientid is only 1 byte */ | |
5809d5e0 CH |
3766 | p = &ophead->oh_clientid; |
3767 | field_offset = p - base_ptr; | |
abca1f33 | 3768 | if (field_offset & 0x1ff) { |
1da177e4 LT |
3769 | clientid = ophead->oh_clientid; |
3770 | } else { | |
b2a922cd | 3771 | idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap); |
1da177e4 LT |
3772 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3773 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3774 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3775 | clientid = xlog_get_client_id( |
3776 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3777 | } else { |
03bea6fe CH |
3778 | clientid = xlog_get_client_id( |
3779 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3780 | } |
3781 | } | |
3782 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 | 3783 | xfs_warn(log->l_mp, |
c9690043 | 3784 | "%s: invalid clientid %d op "PTR_FMT" offset 0x%lx", |
a0fa2b67 DC |
3785 | __func__, clientid, ophead, |
3786 | (unsigned long)field_offset); | |
1da177e4 LT |
3787 | |
3788 | /* check length */ | |
5809d5e0 CH |
3789 | p = &ophead->oh_len; |
3790 | field_offset = p - base_ptr; | |
abca1f33 | 3791 | if (field_offset & 0x1ff) { |
67fcb7bf | 3792 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 | 3793 | } else { |
db9d67d6 CH |
3794 | idx = BTOBBT((uintptr_t)&ophead->oh_len - |
3795 | (uintptr_t)iclog->ic_datap); | |
1da177e4 LT |
3796 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3797 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3798 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3799 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3800 | } else { |
b53e675d | 3801 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3802 | } |
3803 | } | |
3804 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3805 | } | |
b843299b | 3806 | } |
cfcbbbd0 | 3807 | #endif |
1da177e4 LT |
3808 | |
3809 | /* | |
b22cd72c | 3810 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3811 | */ |
3812 | STATIC int | |
3813 | xlog_state_ioerror( | |
9a8d2fdb | 3814 | struct xlog *log) |
1da177e4 LT |
3815 | { |
3816 | xlog_in_core_t *iclog, *ic; | |
3817 | ||
3818 | iclog = log->l_iclog; | |
1858bb0b | 3819 | if (iclog->ic_state != XLOG_STATE_IOERROR) { |
1da177e4 LT |
3820 | /* |
3821 | * Mark all the incore logs IOERROR. | |
3822 | * From now on, no log flushes will result. | |
3823 | */ | |
3824 | ic = iclog; | |
3825 | do { | |
3826 | ic->ic_state = XLOG_STATE_IOERROR; | |
3827 | ic = ic->ic_next; | |
3828 | } while (ic != iclog); | |
014c2544 | 3829 | return 0; |
1da177e4 LT |
3830 | } |
3831 | /* | |
3832 | * Return non-zero, if state transition has already happened. | |
3833 | */ | |
014c2544 | 3834 | return 1; |
1da177e4 LT |
3835 | } |
3836 | ||
3837 | /* | |
3838 | * This is called from xfs_force_shutdown, when we're forcibly | |
3839 | * shutting down the filesystem, typically because of an IO error. | |
3840 | * Our main objectives here are to make sure that: | |
a870fe6d DC |
3841 | * a. if !logerror, flush the logs to disk. Anything modified |
3842 | * after this is ignored. | |
3843 | * b. the filesystem gets marked 'SHUTDOWN' for all interested | |
1da177e4 | 3844 | * parties to find out, 'atomically'. |
a870fe6d | 3845 | * c. those who're sleeping on log reservations, pinned objects and |
1da177e4 | 3846 | * other resources get woken up, and be told the bad news. |
a870fe6d | 3847 | * d. nothing new gets queued up after (b) and (c) are done. |
9da1ab18 | 3848 | * |
a870fe6d DC |
3849 | * Note: for the !logerror case we need to flush the regions held in memory out |
3850 | * to disk first. This needs to be done before the log is marked as shutdown, | |
3851 | * otherwise the iclog writes will fail. | |
1da177e4 LT |
3852 | */ |
3853 | int | |
3854 | xfs_log_force_umount( | |
3855 | struct xfs_mount *mp, | |
3856 | int logerror) | |
3857 | { | |
9a8d2fdb | 3858 | struct xlog *log; |
1da177e4 | 3859 | int retval; |
1da177e4 LT |
3860 | |
3861 | log = mp->m_log; | |
3862 | ||
3863 | /* | |
3864 | * If this happens during log recovery, don't worry about | |
3865 | * locking; the log isn't open for business yet. | |
3866 | */ | |
3867 | if (!log || | |
3868 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3869 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 | 3870 | if (mp->m_sb_bp) |
b0388bf1 | 3871 | mp->m_sb_bp->b_flags |= XBF_DONE; |
014c2544 | 3872 | return 0; |
1da177e4 LT |
3873 | } |
3874 | ||
3875 | /* | |
3876 | * Somebody could've already done the hard work for us. | |
3877 | * No need to get locks for this. | |
3878 | */ | |
1858bb0b | 3879 | if (logerror && log->l_iclog->ic_state == XLOG_STATE_IOERROR) { |
1da177e4 | 3880 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); |
014c2544 | 3881 | return 1; |
1da177e4 | 3882 | } |
9da1ab18 DC |
3883 | |
3884 | /* | |
a870fe6d DC |
3885 | * Flush all the completed transactions to disk before marking the log |
3886 | * being shut down. We need to do it in this order to ensure that | |
3887 | * completed operations are safely on disk before we shut down, and that | |
3888 | * we don't have to issue any buffer IO after the shutdown flags are set | |
3889 | * to guarantee this. | |
9da1ab18 | 3890 | */ |
93b8a585 | 3891 | if (!logerror) |
60e5bb78 | 3892 | xfs_log_force(mp, XFS_LOG_SYNC); |
9da1ab18 | 3893 | |
1da177e4 | 3894 | /* |
3f16b985 DC |
3895 | * mark the filesystem and the as in a shutdown state and wake |
3896 | * everybody up to tell them the bad news. | |
1da177e4 | 3897 | */ |
b22cd72c | 3898 | spin_lock(&log->l_icloglock); |
1da177e4 | 3899 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 | 3900 | if (mp->m_sb_bp) |
b0388bf1 | 3901 | mp->m_sb_bp->b_flags |= XBF_DONE; |
bac8dca9 | 3902 | |
1da177e4 | 3903 | /* |
a870fe6d DC |
3904 | * Mark the log and the iclogs with IO error flags to prevent any |
3905 | * further log IO from being issued or completed. | |
1da177e4 LT |
3906 | */ |
3907 | log->l_flags |= XLOG_IO_ERROR; | |
a870fe6d | 3908 | retval = xlog_state_ioerror(log); |
b22cd72c | 3909 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3910 | |
3911 | /* | |
10547941 DC |
3912 | * We don't want anybody waiting for log reservations after this. That |
3913 | * means we have to wake up everybody queued up on reserveq as well as | |
3914 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3915 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3916 | * action is protected by the grant locks. |
1da177e4 | 3917 | */ |
a79bf2d7 CH |
3918 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3919 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3920 | |
1da177e4 | 3921 | /* |
ac983517 DC |
3922 | * Wake up everybody waiting on xfs_log_force. Wake the CIL push first |
3923 | * as if the log writes were completed. The abort handling in the log | |
3924 | * item committed callback functions will do this again under lock to | |
3925 | * avoid races. | |
1da177e4 | 3926 | */ |
cdea5459 | 3927 | spin_lock(&log->l_cilp->xc_push_lock); |
ac983517 | 3928 | wake_up_all(&log->l_cilp->xc_commit_wait); |
cdea5459 | 3929 | spin_unlock(&log->l_cilp->xc_push_lock); |
12e6a0f4 | 3930 | xlog_state_do_callback(log); |
1da177e4 | 3931 | |
1da177e4 | 3932 | /* return non-zero if log IOERROR transition had already happened */ |
014c2544 | 3933 | return retval; |
1da177e4 LT |
3934 | } |
3935 | ||
ba0f32d4 | 3936 | STATIC int |
9a8d2fdb MT |
3937 | xlog_iclogs_empty( |
3938 | struct xlog *log) | |
1da177e4 LT |
3939 | { |
3940 | xlog_in_core_t *iclog; | |
3941 | ||
3942 | iclog = log->l_iclog; | |
3943 | do { | |
3944 | /* endianness does not matter here, zero is zero in | |
3945 | * any language. | |
3946 | */ | |
3947 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3948 | return 0; |
1da177e4 LT |
3949 | iclog = iclog->ic_next; |
3950 | } while (iclog != log->l_iclog); | |
014c2544 | 3951 | return 1; |
1da177e4 | 3952 | } |
f661f1e0 | 3953 | |
a45086e2 BF |
3954 | /* |
3955 | * Verify that an LSN stamped into a piece of metadata is valid. This is | |
3956 | * intended for use in read verifiers on v5 superblocks. | |
3957 | */ | |
3958 | bool | |
3959 | xfs_log_check_lsn( | |
3960 | struct xfs_mount *mp, | |
3961 | xfs_lsn_t lsn) | |
3962 | { | |
3963 | struct xlog *log = mp->m_log; | |
3964 | bool valid; | |
3965 | ||
3966 | /* | |
3967 | * norecovery mode skips mount-time log processing and unconditionally | |
3968 | * resets the in-core LSN. We can't validate in this mode, but | |
3969 | * modifications are not allowed anyways so just return true. | |
3970 | */ | |
3971 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
3972 | return true; | |
3973 | ||
3974 | /* | |
3975 | * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is | |
3976 | * handled by recovery and thus safe to ignore here. | |
3977 | */ | |
3978 | if (lsn == NULLCOMMITLSN) | |
3979 | return true; | |
3980 | ||
3981 | valid = xlog_valid_lsn(mp->m_log, lsn); | |
3982 | ||
3983 | /* warn the user about what's gone wrong before verifier failure */ | |
3984 | if (!valid) { | |
3985 | spin_lock(&log->l_icloglock); | |
3986 | xfs_warn(mp, | |
3987 | "Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). " | |
3988 | "Please unmount and run xfs_repair (>= v4.3) to resolve.", | |
3989 | CYCLE_LSN(lsn), BLOCK_LSN(lsn), | |
3990 | log->l_curr_cycle, log->l_curr_block); | |
3991 | spin_unlock(&log->l_icloglock); | |
3992 | } | |
3993 | ||
3994 | return valid; | |
3995 | } | |
0c60d3aa DW |
3996 | |
3997 | bool | |
3998 | xfs_log_in_recovery( | |
3999 | struct xfs_mount *mp) | |
4000 | { | |
4001 | struct xlog *log = mp->m_log; | |
4002 | ||
4003 | return log->l_flags & XLOG_ACTIVE_RECOVERY; | |
4004 | } | |
2b73a2c8 DW |
4005 | |
4006 | /* | |
4007 | * Notify the log that we're about to start using a feature that is protected | |
4008 | * by a log incompat feature flag. This will prevent log covering from | |
4009 | * clearing those flags. | |
4010 | */ | |
4011 | void | |
4012 | xlog_use_incompat_feat( | |
4013 | struct xlog *log) | |
4014 | { | |
4015 | down_read(&log->l_incompat_users); | |
4016 | } | |
4017 | ||
4018 | /* Notify the log that we've finished using log incompat features. */ | |
4019 | void | |
4020 | xlog_drop_incompat_feat( | |
4021 | struct xlog *log) | |
4022 | { | |
4023 | up_read(&log->l_incompat_users); | |
4024 | } |