]>
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 { | |
2039a272 | 249 | if (xlog_is_shutdown(log)) |
9f9c19ec CH |
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); |
2039a272 | 263 | if (xlog_is_shutdown(log)) |
9f9c19ec | 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 | ||
e1d06e5f | 301 | ASSERT(!xlog_in_recovery(log)); |
42ceedb3 CH |
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; | |
2039a272 | 368 | if (xlog_is_shutdown(mp->m_log)) |
50d25484 BF |
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 | ||
2039a272 | 385 | if (xlog_is_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 | ||
2039a272 | 453 | if (xlog_is_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 | ||
aad7272a DC |
489 | /* |
490 | * Run all the pending iclog callbacks and wake log force waiters and iclog | |
491 | * space waiters so they can process the newly set shutdown state. We really | |
492 | * don't care what order we process callbacks here because the log is shut down | |
493 | * and so state cannot change on disk anymore. | |
494 | */ | |
495 | static void | |
496 | xlog_state_shutdown_callbacks( | |
497 | struct xlog *log) | |
498 | { | |
499 | struct xlog_in_core *iclog; | |
500 | LIST_HEAD(cb_list); | |
501 | ||
502 | spin_lock(&log->l_icloglock); | |
503 | iclog = log->l_iclog; | |
504 | do { | |
505 | list_splice_init(&iclog->ic_callbacks, &cb_list); | |
506 | wake_up_all(&iclog->ic_force_wait); | |
507 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
508 | ||
509 | wake_up_all(&log->l_flush_wait); | |
510 | spin_unlock(&log->l_icloglock); | |
511 | ||
512 | xlog_cil_process_committed(&cb_list); | |
513 | } | |
514 | ||
df732b29 CH |
515 | /* |
516 | * Flush iclog to disk if this is the last reference to the given iclog and the | |
9d110014 DC |
517 | * it is in the WANT_SYNC state. |
518 | * | |
519 | * If the caller passes in a non-zero @old_tail_lsn and the current log tail | |
520 | * does not match, there may be metadata on disk that must be persisted before | |
521 | * this iclog is written. To satisfy that requirement, set the | |
522 | * XLOG_ICL_NEED_FLUSH flag as a condition for writing this iclog with the new | |
523 | * log tail value. | |
524 | * | |
525 | * If XLOG_ICL_NEED_FUA is already set on the iclog, we need to ensure that the | |
526 | * log tail is updated correctly. NEED_FUA indicates that the iclog will be | |
527 | * written to stable storage, and implies that a commit record is contained | |
528 | * within the iclog. We need to ensure that the log tail does not move beyond | |
529 | * the tail that the first commit record in the iclog ordered against, otherwise | |
530 | * correct recovery of that checkpoint becomes dependent on future operations | |
531 | * performed on this iclog. | |
532 | * | |
533 | * Hence if NEED_FUA is set and the current iclog tail lsn is empty, write the | |
534 | * current tail into iclog. Once the iclog tail is set, future operations must | |
535 | * not modify it, otherwise they potentially violate ordering constraints for | |
536 | * the checkpoint commit that wrote the initial tail lsn value. The tail lsn in | |
537 | * the iclog will get zeroed on activation of the iclog after sync, so we | |
538 | * always capture the tail lsn on the iclog on the first NEED_FUA release | |
539 | * regardless of the number of active reference counts on this iclog. | |
df732b29 | 540 | */ |
9d110014 | 541 | |
eef983ff | 542 | int |
df732b29 CH |
543 | xlog_state_release_iclog( |
544 | struct xlog *log, | |
0dc8f7f1 DC |
545 | struct xlog_in_core *iclog, |
546 | xfs_lsn_t old_tail_lsn) | |
df732b29 | 547 | { |
9d392064 | 548 | xfs_lsn_t tail_lsn; |
df732b29 CH |
549 | lockdep_assert_held(&log->l_icloglock); |
550 | ||
956f6daa | 551 | trace_xlog_iclog_release(iclog, _RET_IP_); |
5112e206 | 552 | if (xlog_is_shutdown(log)) |
df732b29 CH |
553 | return -EIO; |
554 | ||
0dc8f7f1 DC |
555 | /* |
556 | * Grabbing the current log tail needs to be atomic w.r.t. the writing | |
557 | * of the tail LSN into the iclog so we guarantee that the log tail does | |
558 | * not move between deciding if a cache flush is required and writing | |
559 | * the LSN into the iclog below. | |
560 | */ | |
561 | if (old_tail_lsn || iclog->ic_state == XLOG_STATE_WANT_SYNC) { | |
562 | tail_lsn = xlog_assign_tail_lsn(log->l_mp); | |
563 | ||
564 | if (old_tail_lsn && tail_lsn != old_tail_lsn) | |
565 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH; | |
9d110014 DC |
566 | |
567 | if ((iclog->ic_flags & XLOG_ICL_NEED_FUA) && | |
568 | !iclog->ic_header.h_tail_lsn) | |
569 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); | |
0dc8f7f1 DC |
570 | } |
571 | ||
9d392064 DC |
572 | if (!atomic_dec_and_test(&iclog->ic_refcnt)) |
573 | return 0; | |
574 | ||
575 | if (iclog->ic_state != XLOG_STATE_WANT_SYNC) { | |
576 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
577 | return 0; | |
df732b29 CH |
578 | } |
579 | ||
9d392064 | 580 | iclog->ic_state = XLOG_STATE_SYNCING; |
9d110014 DC |
581 | if (!iclog->ic_header.h_tail_lsn) |
582 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); | |
583 | xlog_verify_tail_lsn(log, iclog); | |
9d392064 DC |
584 | trace_xlog_iclog_syncing(iclog, _RET_IP_); |
585 | ||
586 | spin_unlock(&log->l_icloglock); | |
587 | xlog_sync(log, iclog); | |
588 | spin_lock(&log->l_icloglock); | |
df732b29 CH |
589 | return 0; |
590 | } | |
591 | ||
1da177e4 LT |
592 | /* |
593 | * Mount a log filesystem | |
594 | * | |
595 | * mp - ubiquitous xfs mount point structure | |
596 | * log_target - buftarg of on-disk log device | |
597 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
598 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
599 | * | |
600 | * Return error or zero. | |
601 | */ | |
602 | int | |
249a8c11 DC |
603 | xfs_log_mount( |
604 | xfs_mount_t *mp, | |
605 | xfs_buftarg_t *log_target, | |
606 | xfs_daddr_t blk_offset, | |
607 | int num_bblks) | |
1da177e4 | 608 | { |
e1d06e5f | 609 | struct xlog *log; |
9c92ee20 | 610 | bool fatal = xfs_sb_version_hascrc(&mp->m_sb); |
3e7b91cf JL |
611 | int error = 0; |
612 | int min_logfsbs; | |
249a8c11 | 613 | |
c99d609a DC |
614 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
615 | xfs_notice(mp, "Mounting V%d Filesystem", | |
616 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
617 | } else { | |
a0fa2b67 | 618 | xfs_notice(mp, |
c99d609a DC |
619 | "Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.", |
620 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
bd186aa9 | 621 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
622 | } |
623 | ||
e1d06e5f DC |
624 | log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); |
625 | if (IS_ERR(log)) { | |
626 | error = PTR_ERR(log); | |
644c3567 DC |
627 | goto out; |
628 | } | |
e1d06e5f | 629 | mp->m_log = log; |
1da177e4 | 630 | |
3e7b91cf JL |
631 | /* |
632 | * Validate the given log space and drop a critical message via syslog | |
633 | * if the log size is too small that would lead to some unexpected | |
634 | * situations in transaction log space reservation stage. | |
635 | * | |
636 | * Note: we can't just reject the mount if the validation fails. This | |
637 | * would mean that people would have to downgrade their kernel just to | |
638 | * remedy the situation as there is no way to grow the log (short of | |
639 | * black magic surgery with xfs_db). | |
640 | * | |
641 | * We can, however, reject mounts for CRC format filesystems, as the | |
642 | * mkfs binary being used to make the filesystem should never create a | |
643 | * filesystem with a log that is too small. | |
644 | */ | |
645 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
646 | ||
647 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
648 | xfs_warn(mp, | |
649 | "Log size %d blocks too small, minimum size is %d blocks", | |
650 | mp->m_sb.sb_logblocks, min_logfsbs); | |
2451337d | 651 | error = -EINVAL; |
3e7b91cf JL |
652 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { |
653 | xfs_warn(mp, | |
654 | "Log size %d blocks too large, maximum size is %lld blocks", | |
655 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
2451337d | 656 | error = -EINVAL; |
3e7b91cf JL |
657 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { |
658 | xfs_warn(mp, | |
659 | "log size %lld bytes too large, maximum size is %lld bytes", | |
660 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
661 | XFS_MAX_LOG_BYTES); | |
2451337d | 662 | error = -EINVAL; |
9c92ee20 DW |
663 | } else if (mp->m_sb.sb_logsunit > 1 && |
664 | mp->m_sb.sb_logsunit % mp->m_sb.sb_blocksize) { | |
665 | xfs_warn(mp, | |
666 | "log stripe unit %u bytes must be a multiple of block size", | |
667 | mp->m_sb.sb_logsunit); | |
668 | error = -EINVAL; | |
669 | fatal = true; | |
3e7b91cf JL |
670 | } |
671 | if (error) { | |
9c92ee20 DW |
672 | /* |
673 | * Log check errors are always fatal on v5; or whenever bad | |
674 | * metadata leads to a crash. | |
675 | */ | |
676 | if (fatal) { | |
3e7b91cf JL |
677 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); |
678 | ASSERT(0); | |
679 | goto out_free_log; | |
680 | } | |
f41febd2 | 681 | xfs_crit(mp, "Log size out of supported range."); |
3e7b91cf | 682 | xfs_crit(mp, |
f41febd2 | 683 | "Continuing onwards, but if log hangs are experienced then please report this message in the bug report."); |
3e7b91cf JL |
684 | } |
685 | ||
249a8c11 DC |
686 | /* |
687 | * Initialize the AIL now we have a log. | |
688 | */ | |
249a8c11 DC |
689 | error = xfs_trans_ail_init(mp); |
690 | if (error) { | |
a0fa2b67 | 691 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 692 | goto out_free_log; |
249a8c11 | 693 | } |
e1d06e5f | 694 | log->l_ailp = mp->m_ail; |
249a8c11 | 695 | |
1da177e4 LT |
696 | /* |
697 | * skip log recovery on a norecovery mount. pretend it all | |
698 | * just worked. | |
699 | */ | |
700 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 701 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
702 | |
703 | if (readonly) | |
bd186aa9 | 704 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 705 | |
e1d06e5f | 706 | error = xlog_recover(log); |
1da177e4 LT |
707 | |
708 | if (readonly) | |
bd186aa9 | 709 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 710 | if (error) { |
a0fa2b67 DC |
711 | xfs_warn(mp, "log mount/recovery failed: error %d", |
712 | error); | |
e1d06e5f | 713 | xlog_recover_cancel(log); |
26430752 | 714 | goto out_destroy_ail; |
1da177e4 LT |
715 | } |
716 | } | |
717 | ||
e1d06e5f | 718 | error = xfs_sysfs_init(&log->l_kobj, &xfs_log_ktype, &mp->m_kobj, |
baff4e44 BF |
719 | "log"); |
720 | if (error) | |
721 | goto out_destroy_ail; | |
722 | ||
1da177e4 | 723 | /* Normal transactions can now occur */ |
e1d06e5f | 724 | clear_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate); |
1da177e4 | 725 | |
71e330b5 DC |
726 | /* |
727 | * Now the log has been fully initialised and we know were our | |
728 | * space grant counters are, we can initialise the permanent ticket | |
729 | * needed for delayed logging to work. | |
730 | */ | |
e1d06e5f | 731 | xlog_cil_init_post_recovery(log); |
71e330b5 | 732 | |
1da177e4 | 733 | return 0; |
26430752 CH |
734 | |
735 | out_destroy_ail: | |
736 | xfs_trans_ail_destroy(mp); | |
737 | out_free_log: | |
e1d06e5f | 738 | xlog_dealloc_log(log); |
644c3567 | 739 | out: |
249a8c11 | 740 | return error; |
26430752 | 741 | } |
1da177e4 LT |
742 | |
743 | /* | |
f661f1e0 DC |
744 | * Finish the recovery of the file system. This is separate from the |
745 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
746 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
747 | * here. | |
1da177e4 | 748 | * |
f661f1e0 DC |
749 | * If we finish recovery successfully, start the background log work. If we are |
750 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
751 | * it. | |
1da177e4 LT |
752 | */ |
753 | int | |
f0b2efad BF |
754 | xfs_log_mount_finish( |
755 | struct xfs_mount *mp) | |
1da177e4 | 756 | { |
fd67d8a0 DC |
757 | struct xlog *log = mp->m_log; |
758 | bool readonly = (mp->m_flags & XFS_MOUNT_RDONLY); | |
759 | int error = 0; | |
1da177e4 | 760 | |
f0b2efad | 761 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) { |
bd186aa9 | 762 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
f0b2efad | 763 | return 0; |
6f4a1eef ES |
764 | } else if (readonly) { |
765 | /* Allow unlinked processing to proceed */ | |
766 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
1da177e4 LT |
767 | } |
768 | ||
8204f8dd DW |
769 | /* |
770 | * During the second phase of log recovery, we need iget and | |
771 | * iput to behave like they do for an active filesystem. | |
772 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
773 | * of inodes before we're done replaying log items on those | |
774 | * inodes. Turn it off immediately after recovery finishes | |
775 | * so that we don't leak the quota inodes if subsequent mount | |
776 | * activities fail. | |
799ea9e9 DW |
777 | * |
778 | * We let all inodes involved in redo item processing end up on | |
779 | * the LRU instead of being evicted immediately so that if we do | |
780 | * something to an unlinked inode, the irele won't cause | |
781 | * premature truncation and freeing of the inode, which results | |
782 | * in log recovery failure. We have to evict the unreferenced | |
1751e8a6 | 783 | * lru inodes after clearing SB_ACTIVE because we don't |
799ea9e9 DW |
784 | * otherwise clean up the lru if there's a subsequent failure in |
785 | * xfs_mountfs, which leads to us leaking the inodes if nothing | |
786 | * else (e.g. quotacheck) references the inodes before the | |
787 | * mount failure occurs. | |
8204f8dd | 788 | */ |
1751e8a6 | 789 | mp->m_super->s_flags |= SB_ACTIVE; |
e1d06e5f | 790 | if (xlog_recovery_needed(log)) |
fd67d8a0 | 791 | error = xlog_recover_finish(log); |
f0b2efad BF |
792 | if (!error) |
793 | xfs_log_work_queue(mp); | |
1751e8a6 | 794 | mp->m_super->s_flags &= ~SB_ACTIVE; |
799ea9e9 | 795 | evict_inodes(mp->m_super); |
f0b2efad | 796 | |
f1b92bbc BF |
797 | /* |
798 | * Drain the buffer LRU after log recovery. This is required for v4 | |
799 | * filesystems to avoid leaving around buffers with NULL verifier ops, | |
800 | * but we do it unconditionally to make sure we're always in a clean | |
801 | * cache state after mount. | |
802 | * | |
803 | * Don't push in the error case because the AIL may have pending intents | |
804 | * that aren't removed until recovery is cancelled. | |
805 | */ | |
e1d06e5f | 806 | if (xlog_recovery_needed(log)) { |
fd67d8a0 DC |
807 | if (!error) { |
808 | xfs_log_force(mp, XFS_LOG_SYNC); | |
809 | xfs_ail_push_all_sync(mp->m_ail); | |
810 | } | |
811 | xfs_notice(mp, "Ending recovery (logdev: %s)", | |
812 | mp->m_logname ? mp->m_logname : "internal"); | |
813 | } else { | |
814 | xfs_info(mp, "Ending clean mount"); | |
f1b92bbc | 815 | } |
10fb9ac1 | 816 | xfs_buftarg_drain(mp->m_ddev_targp); |
f1b92bbc | 817 | |
e1d06e5f | 818 | clear_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate); |
6f4a1eef ES |
819 | if (readonly) |
820 | mp->m_flags |= XFS_MOUNT_RDONLY; | |
821 | ||
4e6b8270 | 822 | /* Make sure the log is dead if we're returning failure. */ |
fd67d8a0 | 823 | ASSERT(!error || xlog_is_shutdown(log)); |
4e6b8270 | 824 | |
f0b2efad BF |
825 | return error; |
826 | } | |
827 | ||
828 | /* | |
829 | * The mount has failed. Cancel the recovery if it hasn't completed and destroy | |
830 | * the log. | |
831 | */ | |
a7a9250e | 832 | void |
f0b2efad BF |
833 | xfs_log_mount_cancel( |
834 | struct xfs_mount *mp) | |
835 | { | |
a7a9250e | 836 | xlog_recover_cancel(mp->m_log); |
f0b2efad | 837 | xfs_log_unmount(mp); |
1da177e4 LT |
838 | } |
839 | ||
45eddb41 DC |
840 | /* |
841 | * Flush out the iclog to disk ensuring that device caches are flushed and | |
842 | * the iclog hits stable storage before any completion waiters are woken. | |
843 | */ | |
844 | static inline int | |
845 | xlog_force_iclog( | |
846 | struct xlog_in_core *iclog) | |
847 | { | |
848 | atomic_inc(&iclog->ic_refcnt); | |
2bf1ec0f | 849 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA; |
45eddb41 DC |
850 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
851 | xlog_state_switch_iclogs(iclog->ic_log, iclog, 0); | |
852 | return xlog_state_release_iclog(iclog->ic_log, iclog, 0); | |
853 | } | |
854 | ||
81e5b50a | 855 | /* |
a79b28c2 DC |
856 | * Wait for the iclog and all prior iclogs to be written disk as required by the |
857 | * log force state machine. Waiting on ic_force_wait ensures iclog completions | |
858 | * have been ordered and callbacks run before we are woken here, hence | |
859 | * guaranteeing that all the iclogs up to this one are on stable storage. | |
81e5b50a | 860 | */ |
a79b28c2 | 861 | int |
81e5b50a CH |
862 | xlog_wait_on_iclog( |
863 | struct xlog_in_core *iclog) | |
864 | __releases(iclog->ic_log->l_icloglock) | |
865 | { | |
866 | struct xlog *log = iclog->ic_log; | |
867 | ||
956f6daa | 868 | trace_xlog_iclog_wait_on(iclog, _RET_IP_); |
2039a272 | 869 | if (!xlog_is_shutdown(log) && |
81e5b50a CH |
870 | iclog->ic_state != XLOG_STATE_ACTIVE && |
871 | iclog->ic_state != XLOG_STATE_DIRTY) { | |
872 | XFS_STATS_INC(log->l_mp, xs_log_force_sleep); | |
873 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
874 | } else { | |
875 | spin_unlock(&log->l_icloglock); | |
876 | } | |
877 | ||
2039a272 | 878 | if (xlog_is_shutdown(log)) |
81e5b50a CH |
879 | return -EIO; |
880 | return 0; | |
881 | } | |
882 | ||
1da177e4 | 883 | /* |
3c702f95 DC |
884 | * Write out an unmount record using the ticket provided. We have to account for |
885 | * the data space used in the unmount ticket as this write is not done from a | |
886 | * transaction context that has already done the accounting for us. | |
1da177e4 | 887 | */ |
3c702f95 DC |
888 | static int |
889 | xlog_write_unmount_record( | |
890 | struct xlog *log, | |
3468bb1c | 891 | struct xlog_ticket *ticket) |
53235f22 | 892 | { |
3c702f95 | 893 | struct xfs_unmount_log_format ulf = { |
53235f22 DW |
894 | .magic = XLOG_UNMOUNT_TYPE, |
895 | }; | |
896 | struct xfs_log_iovec reg = { | |
3c702f95 DC |
897 | .i_addr = &ulf, |
898 | .i_len = sizeof(ulf), | |
53235f22 DW |
899 | .i_type = XLOG_REG_TYPE_UNMOUNT, |
900 | }; | |
901 | struct xfs_log_vec vec = { | |
902 | .lv_niovecs = 1, | |
903 | .lv_iovecp = ®, | |
904 | }; | |
3c702f95 DC |
905 | |
906 | /* account for space used by record data */ | |
907 | ticket->t_curr_res -= sizeof(ulf); | |
eef983ff | 908 | |
3468bb1c | 909 | return xlog_write(log, &vec, ticket, NULL, NULL, XLOG_UNMOUNT_TRANS); |
3c702f95 DC |
910 | } |
911 | ||
912 | /* | |
913 | * Mark the filesystem clean by writing an unmount record to the head of the | |
914 | * log. | |
915 | */ | |
916 | static void | |
917 | xlog_unmount_write( | |
918 | struct xlog *log) | |
919 | { | |
920 | struct xfs_mount *mp = log->l_mp; | |
53235f22 DW |
921 | struct xlog_in_core *iclog; |
922 | struct xlog_ticket *tic = NULL; | |
53235f22 DW |
923 | int error; |
924 | ||
925 | error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0); | |
926 | if (error) | |
927 | goto out_err; | |
928 | ||
3468bb1c | 929 | error = xlog_write_unmount_record(log, tic); |
53235f22 DW |
930 | /* |
931 | * At this point, we're umounting anyway, so there's no point in | |
5112e206 | 932 | * transitioning log state to shutdown. Just continue... |
53235f22 DW |
933 | */ |
934 | out_err: | |
935 | if (error) | |
936 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
937 | ||
938 | spin_lock(&log->l_icloglock); | |
939 | iclog = log->l_iclog; | |
45eddb41 | 940 | error = xlog_force_iclog(iclog); |
81e5b50a | 941 | xlog_wait_on_iclog(iclog); |
53235f22 DW |
942 | |
943 | if (tic) { | |
944 | trace_xfs_log_umount_write(log, tic); | |
8b41e3f9 | 945 | xfs_log_ticket_ungrant(log, tic); |
53235f22 DW |
946 | } |
947 | } | |
948 | ||
13859c98 CH |
949 | static void |
950 | xfs_log_unmount_verify_iclog( | |
951 | struct xlog *log) | |
952 | { | |
953 | struct xlog_in_core *iclog = log->l_iclog; | |
954 | ||
955 | do { | |
956 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
957 | ASSERT(iclog->ic_offset == 0); | |
958 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
959 | } | |
960 | ||
1da177e4 LT |
961 | /* |
962 | * Unmount record used to have a string "Unmount filesystem--" in the | |
963 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
964 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 965 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
966 | * As far as I know, there weren't any dependencies on the old behaviour. |
967 | */ | |
550319e9 | 968 | static void |
13859c98 CH |
969 | xfs_log_unmount_write( |
970 | struct xfs_mount *mp) | |
1da177e4 | 971 | { |
13859c98 | 972 | struct xlog *log = mp->m_log; |
1da177e4 | 973 | |
50d25484 | 974 | if (!xfs_log_writable(mp)) |
550319e9 | 975 | return; |
1da177e4 | 976 | |
550319e9 | 977 | xfs_log_force(mp, XFS_LOG_SYNC); |
1da177e4 | 978 | |
2039a272 | 979 | if (xlog_is_shutdown(log)) |
6178d104 | 980 | return; |
5cc3c006 DW |
981 | |
982 | /* | |
983 | * If we think the summary counters are bad, avoid writing the unmount | |
984 | * record to force log recovery at next mount, after which the summary | |
985 | * counters will be recalculated. Refer to xlog_check_unmount_rec for | |
986 | * more details. | |
987 | */ | |
988 | if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp, | |
989 | XFS_ERRTAG_FORCE_SUMMARY_RECALC)) { | |
990 | xfs_alert(mp, "%s: will fix summary counters at next mount", | |
991 | __func__); | |
992 | return; | |
993 | } | |
994 | ||
13859c98 | 995 | xfs_log_unmount_verify_iclog(log); |
3c702f95 | 996 | xlog_unmount_write(log); |
550319e9 | 997 | } |
1da177e4 LT |
998 | |
999 | /* | |
c75921a7 | 1000 | * Empty the log for unmount/freeze. |
cf2931db DC |
1001 | * |
1002 | * To do this, we first need to shut down the background log work so it is not | |
1003 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
1004 | * the log so we can then flush them out. Once they have completed their IO and | |
303591a0 | 1005 | * run the callbacks removing themselves from the AIL, we can cover the log. |
1da177e4 | 1006 | */ |
303591a0 | 1007 | int |
c75921a7 DC |
1008 | xfs_log_quiesce( |
1009 | struct xfs_mount *mp) | |
1da177e4 | 1010 | { |
908ce71e DW |
1011 | /* |
1012 | * Clear log incompat features since we're quiescing the log. Report | |
1013 | * failures, though it's not fatal to have a higher log feature | |
1014 | * protection level than the log contents actually require. | |
1015 | */ | |
1016 | if (xfs_clear_incompat_log_features(mp)) { | |
1017 | int error; | |
1018 | ||
1019 | error = xfs_sync_sb(mp, false); | |
1020 | if (error) | |
1021 | xfs_warn(mp, | |
1022 | "Failed to clear log incompat features on quiesce"); | |
1023 | } | |
1024 | ||
f661f1e0 | 1025 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
1026 | xfs_log_force(mp, XFS_LOG_SYNC); |
1027 | ||
1028 | /* | |
1029 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
8321ddb2 | 1030 | * will push it, xfs_buftarg_wait() will not wait for it. Further, |
cf2931db DC |
1031 | * xfs_buf_iowait() cannot be used because it was pushed with the |
1032 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
1033 | * the IO to complete. | |
1034 | */ | |
1035 | xfs_ail_push_all_sync(mp->m_ail); | |
8321ddb2 | 1036 | xfs_buftarg_wait(mp->m_ddev_targp); |
cf2931db DC |
1037 | xfs_buf_lock(mp->m_sb_bp); |
1038 | xfs_buf_unlock(mp->m_sb_bp); | |
303591a0 BF |
1039 | |
1040 | return xfs_log_cover(mp); | |
9e54ee0f | 1041 | } |
cf2931db | 1042 | |
9e54ee0f BF |
1043 | void |
1044 | xfs_log_clean( | |
1045 | struct xfs_mount *mp) | |
1046 | { | |
1047 | xfs_log_quiesce(mp); | |
cf2931db | 1048 | xfs_log_unmount_write(mp); |
c75921a7 DC |
1049 | } |
1050 | ||
1051 | /* | |
1052 | * Shut down and release the AIL and Log. | |
1053 | * | |
1054 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
1055 | * from the AIL so that the log is empty before we write the unmount record to | |
1056 | * the log. Once this is done, we can tear down the AIL and the log. | |
1057 | */ | |
1058 | void | |
1059 | xfs_log_unmount( | |
1060 | struct xfs_mount *mp) | |
1061 | { | |
9e54ee0f | 1062 | xfs_log_clean(mp); |
cf2931db | 1063 | |
8321ddb2 BF |
1064 | xfs_buftarg_drain(mp->m_ddev_targp); |
1065 | ||
249a8c11 | 1066 | xfs_trans_ail_destroy(mp); |
baff4e44 BF |
1067 | |
1068 | xfs_sysfs_del(&mp->m_log->l_kobj); | |
1069 | ||
c41564b5 | 1070 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
1071 | } |
1072 | ||
43f5efc5 DC |
1073 | void |
1074 | xfs_log_item_init( | |
1075 | struct xfs_mount *mp, | |
1076 | struct xfs_log_item *item, | |
1077 | int type, | |
272e42b2 | 1078 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
1079 | { |
1080 | item->li_mountp = mp; | |
1081 | item->li_ailp = mp->m_ail; | |
1082 | item->li_type = type; | |
1083 | item->li_ops = ops; | |
71e330b5 DC |
1084 | item->li_lv = NULL; |
1085 | ||
1086 | INIT_LIST_HEAD(&item->li_ail); | |
1087 | INIT_LIST_HEAD(&item->li_cil); | |
643c8c05 | 1088 | INIT_LIST_HEAD(&item->li_bio_list); |
e6631f85 | 1089 | INIT_LIST_HEAD(&item->li_trans); |
43f5efc5 DC |
1090 | } |
1091 | ||
09a423a3 CH |
1092 | /* |
1093 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 1094 | */ |
1da177e4 | 1095 | void |
09a423a3 | 1096 | xfs_log_space_wake( |
cfb7cdca | 1097 | struct xfs_mount *mp) |
1da177e4 | 1098 | { |
ad223e60 | 1099 | struct xlog *log = mp->m_log; |
cfb7cdca | 1100 | int free_bytes; |
1da177e4 | 1101 | |
2039a272 | 1102 | if (xlog_is_shutdown(log)) |
1da177e4 | 1103 | return; |
1da177e4 | 1104 | |
28496968 | 1105 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
e1d06e5f | 1106 | ASSERT(!xlog_in_recovery(log)); |
09a423a3 | 1107 | |
28496968 CH |
1108 | spin_lock(&log->l_write_head.lock); |
1109 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 1110 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 1111 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 1112 | } |
10547941 | 1113 | |
28496968 | 1114 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
e1d06e5f | 1115 | ASSERT(!xlog_in_recovery(log)); |
09a423a3 | 1116 | |
28496968 CH |
1117 | spin_lock(&log->l_reserve_head.lock); |
1118 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 1119 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 1120 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 1121 | } |
3f16b985 | 1122 | } |
1da177e4 LT |
1123 | |
1124 | /* | |
2c6e24ce DC |
1125 | * Determine if we have a transaction that has gone to disk that needs to be |
1126 | * covered. To begin the transition to the idle state firstly the log needs to | |
1127 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1128 | * we start attempting to cover the log. | |
b6f8dd49 | 1129 | * |
2c6e24ce DC |
1130 | * Only if we are then in a state where covering is needed, the caller is |
1131 | * informed that dummy transactions are required to move the log into the idle | |
1132 | * state. | |
1133 | * | |
1134 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1135 | * cover the log as we may be in a situation where there isn't log space | |
1136 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1137 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1138 | * there's no point in running a dummy transaction at this point because we | |
1139 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 | 1140 | */ |
37444fc4 BF |
1141 | static bool |
1142 | xfs_log_need_covered( | |
1143 | struct xfs_mount *mp) | |
1da177e4 | 1144 | { |
37444fc4 BF |
1145 | struct xlog *log = mp->m_log; |
1146 | bool needed = false; | |
1da177e4 | 1147 | |
2c6e24ce | 1148 | if (!xlog_cil_empty(log)) |
8646b982 | 1149 | return false; |
2c6e24ce | 1150 | |
b22cd72c | 1151 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1152 | switch (log->l_covered_state) { |
1153 | case XLOG_STATE_COVER_DONE: | |
1154 | case XLOG_STATE_COVER_DONE2: | |
1155 | case XLOG_STATE_COVER_IDLE: | |
1156 | break; | |
1157 | case XLOG_STATE_COVER_NEED: | |
1158 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1159 | if (xfs_ail_min_lsn(log->l_ailp)) |
1160 | break; | |
1161 | if (!xlog_iclogs_empty(log)) | |
1162 | break; | |
1163 | ||
37444fc4 | 1164 | needed = true; |
2c6e24ce DC |
1165 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) |
1166 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1167 | else | |
1168 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1169 | break; | |
b6f8dd49 | 1170 | default: |
37444fc4 | 1171 | needed = true; |
b6f8dd49 | 1172 | break; |
1da177e4 | 1173 | } |
b22cd72c | 1174 | spin_unlock(&log->l_icloglock); |
014c2544 | 1175 | return needed; |
1da177e4 LT |
1176 | } |
1177 | ||
303591a0 BF |
1178 | /* |
1179 | * Explicitly cover the log. This is similar to background log covering but | |
1180 | * intended for usage in quiesce codepaths. The caller is responsible to ensure | |
1181 | * the log is idle and suitable for covering. The CIL, iclog buffers and AIL | |
1182 | * must all be empty. | |
1183 | */ | |
1184 | static int | |
1185 | xfs_log_cover( | |
1186 | struct xfs_mount *mp) | |
1187 | { | |
303591a0 | 1188 | int error = 0; |
f46e5a17 | 1189 | bool need_covered; |
303591a0 | 1190 | |
4533fc63 BF |
1191 | ASSERT((xlog_cil_empty(mp->m_log) && xlog_iclogs_empty(mp->m_log) && |
1192 | !xfs_ail_min_lsn(mp->m_log->l_ailp)) || | |
2039a272 | 1193 | xlog_is_shutdown(mp->m_log)); |
303591a0 BF |
1194 | |
1195 | if (!xfs_log_writable(mp)) | |
1196 | return 0; | |
1197 | ||
f46e5a17 BF |
1198 | /* |
1199 | * xfs_log_need_covered() is not idempotent because it progresses the | |
1200 | * state machine if the log requires covering. Therefore, we must call | |
1201 | * this function once and use the result until we've issued an sb sync. | |
1202 | * Do so first to make that abundantly clear. | |
1203 | * | |
1204 | * Fall into the covering sequence if the log needs covering or the | |
1205 | * mount has lazy superblock accounting to sync to disk. The sb sync | |
1206 | * used for covering accumulates the in-core counters, so covering | |
1207 | * handles this for us. | |
1208 | */ | |
1209 | need_covered = xfs_log_need_covered(mp); | |
1210 | if (!need_covered && !xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1211 | return 0; | |
1212 | ||
303591a0 BF |
1213 | /* |
1214 | * To cover the log, commit the superblock twice (at most) in | |
1215 | * independent checkpoints. The first serves as a reference for the | |
1216 | * tail pointer. The sync transaction and AIL push empties the AIL and | |
1217 | * updates the in-core tail to the LSN of the first checkpoint. The | |
1218 | * second commit updates the on-disk tail with the in-core LSN, | |
1219 | * covering the log. Push the AIL one more time to leave it empty, as | |
1220 | * we found it. | |
1221 | */ | |
f46e5a17 | 1222 | do { |
303591a0 BF |
1223 | error = xfs_sync_sb(mp, true); |
1224 | if (error) | |
1225 | break; | |
1226 | xfs_ail_push_all_sync(mp->m_ail); | |
f46e5a17 | 1227 | } while (xfs_log_need_covered(mp)); |
303591a0 BF |
1228 | |
1229 | return error; | |
1230 | } | |
1231 | ||
09a423a3 | 1232 | /* |
1da177e4 LT |
1233 | * We may be holding the log iclog lock upon entering this routine. |
1234 | */ | |
1235 | xfs_lsn_t | |
1c304625 | 1236 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1237 | struct xfs_mount *mp) |
1da177e4 | 1238 | { |
ad223e60 | 1239 | struct xlog *log = mp->m_log; |
1c304625 CH |
1240 | struct xfs_log_item *lip; |
1241 | xfs_lsn_t tail_lsn; | |
1242 | ||
57e80956 | 1243 | assert_spin_locked(&mp->m_ail->ail_lock); |
1da177e4 | 1244 | |
09a423a3 CH |
1245 | /* |
1246 | * To make sure we always have a valid LSN for the log tail we keep | |
1247 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1248 | * and use that when the AIL was empty. |
09a423a3 | 1249 | */ |
1c304625 CH |
1250 | lip = xfs_ail_min(mp->m_ail); |
1251 | if (lip) | |
1252 | tail_lsn = lip->li_lsn; | |
1253 | else | |
84f3c683 | 1254 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
750b9c90 | 1255 | trace_xfs_log_assign_tail_lsn(log, tail_lsn); |
1c3cb9ec | 1256 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1257 | return tail_lsn; |
1c3cb9ec | 1258 | } |
1da177e4 | 1259 | |
1c304625 CH |
1260 | xfs_lsn_t |
1261 | xlog_assign_tail_lsn( | |
1262 | struct xfs_mount *mp) | |
1263 | { | |
1264 | xfs_lsn_t tail_lsn; | |
1265 | ||
57e80956 | 1266 | spin_lock(&mp->m_ail->ail_lock); |
1c304625 | 1267 | tail_lsn = xlog_assign_tail_lsn_locked(mp); |
57e80956 | 1268 | spin_unlock(&mp->m_ail->ail_lock); |
1c304625 CH |
1269 | |
1270 | return tail_lsn; | |
1271 | } | |
1272 | ||
1da177e4 LT |
1273 | /* |
1274 | * Return the space in the log between the tail and the head. The head | |
1275 | * is passed in the cycle/bytes formal parms. In the special case where | |
1276 | * the reserve head has wrapped passed the tail, this calculation is no | |
1277 | * longer valid. In this case, just return 0 which means there is no space | |
1278 | * in the log. This works for all places where this function is called | |
1279 | * with the reserve head. Of course, if the write head were to ever | |
1280 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1281 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1282 | * | |
1283 | * This code also handles the case where the reservation head is behind | |
1284 | * the tail. The details of this case are described below, but the end | |
1285 | * result is that we return the size of the log as the amount of space left. | |
1286 | */ | |
a8272ce0 | 1287 | STATIC int |
a69ed03c | 1288 | xlog_space_left( |
ad223e60 | 1289 | struct xlog *log, |
c8a09ff8 | 1290 | atomic64_t *head) |
1da177e4 | 1291 | { |
a69ed03c DC |
1292 | int free_bytes; |
1293 | int tail_bytes; | |
1294 | int tail_cycle; | |
1295 | int head_cycle; | |
1296 | int head_bytes; | |
1da177e4 | 1297 | |
a69ed03c | 1298 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1299 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1300 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1301 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1302 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1303 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1304 | return 0; |
a69ed03c DC |
1305 | else if (tail_cycle < head_cycle) { |
1306 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1307 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1308 | } else { |
1309 | /* | |
1310 | * The reservation head is behind the tail. | |
1311 | * In this case we just want to return the size of the | |
1312 | * log as the amount of space left. | |
1313 | */ | |
f41febd2 | 1314 | xfs_alert(log->l_mp, "xlog_space_left: head behind tail"); |
a0fa2b67 | 1315 | xfs_alert(log->l_mp, |
f41febd2 JP |
1316 | " tail_cycle = %d, tail_bytes = %d", |
1317 | tail_cycle, tail_bytes); | |
1318 | xfs_alert(log->l_mp, | |
1319 | " GH cycle = %d, GH bytes = %d", | |
1320 | head_cycle, head_bytes); | |
1da177e4 LT |
1321 | ASSERT(0); |
1322 | free_bytes = log->l_logsize; | |
1323 | } | |
1324 | return free_bytes; | |
a69ed03c | 1325 | } |
1da177e4 LT |
1326 | |
1327 | ||
0d5a75e9 | 1328 | static void |
79b54d9b CH |
1329 | xlog_ioend_work( |
1330 | struct work_struct *work) | |
1da177e4 | 1331 | { |
79b54d9b CH |
1332 | struct xlog_in_core *iclog = |
1333 | container_of(work, struct xlog_in_core, ic_end_io_work); | |
1334 | struct xlog *log = iclog->ic_log; | |
79b54d9b | 1335 | int error; |
1da177e4 | 1336 | |
79b54d9b | 1337 | error = blk_status_to_errno(iclog->ic_bio.bi_status); |
366fc4b8 CH |
1338 | #ifdef DEBUG |
1339 | /* treat writes with injected CRC errors as failed */ | |
1340 | if (iclog->ic_fail_crc) | |
79b54d9b | 1341 | error = -EIO; |
366fc4b8 CH |
1342 | #endif |
1343 | ||
1da177e4 | 1344 | /* |
366fc4b8 | 1345 | * Race to shutdown the filesystem if we see an error. |
1da177e4 | 1346 | */ |
79b54d9b CH |
1347 | if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) { |
1348 | xfs_alert(log->l_mp, "log I/O error %d", error); | |
1349 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1da177e4 | 1350 | } |
3db296f3 | 1351 | |
12e6a0f4 | 1352 | xlog_state_done_syncing(iclog); |
79b54d9b | 1353 | bio_uninit(&iclog->ic_bio); |
9c23eccc | 1354 | |
3db296f3 | 1355 | /* |
79b54d9b CH |
1356 | * Drop the lock to signal that we are done. Nothing references the |
1357 | * iclog after this, so an unmount waiting on this lock can now tear it | |
1358 | * down safely. As such, it is unsafe to reference the iclog after the | |
1359 | * unlock as we could race with it being freed. | |
3db296f3 | 1360 | */ |
79b54d9b | 1361 | up(&iclog->ic_sema); |
c3f8fc73 | 1362 | } |
1da177e4 | 1363 | |
1da177e4 LT |
1364 | /* |
1365 | * Return size of each in-core log record buffer. | |
1366 | * | |
9da096fd | 1367 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1368 | * |
1369 | * If the filesystem blocksize is too large, we may need to choose a | |
1370 | * larger size since the directory code currently logs entire blocks. | |
1371 | */ | |
1da177e4 | 1372 | STATIC void |
9a8d2fdb MT |
1373 | xlog_get_iclog_buffer_size( |
1374 | struct xfs_mount *mp, | |
1375 | struct xlog *log) | |
1da177e4 | 1376 | { |
1cb51258 | 1377 | if (mp->m_logbufs <= 0) |
4f62282a CH |
1378 | mp->m_logbufs = XLOG_MAX_ICLOGS; |
1379 | if (mp->m_logbsize <= 0) | |
1380 | mp->m_logbsize = XLOG_BIG_RECORD_BSIZE; | |
1381 | ||
1382 | log->l_iclog_bufs = mp->m_logbufs; | |
1383 | log->l_iclog_size = mp->m_logbsize; | |
1da177e4 LT |
1384 | |
1385 | /* | |
4f62282a | 1386 | * # headers = size / 32k - one header holds cycles from 32k of data. |
1da177e4 | 1387 | */ |
4f62282a CH |
1388 | log->l_iclog_heads = |
1389 | DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE); | |
1390 | log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT; | |
1391 | } | |
1da177e4 | 1392 | |
f661f1e0 DC |
1393 | void |
1394 | xfs_log_work_queue( | |
1395 | struct xfs_mount *mp) | |
1396 | { | |
696a5620 | 1397 | queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1398 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1399 | } | |
1400 | ||
2b73a2c8 DW |
1401 | /* |
1402 | * Clear the log incompat flags if we have the opportunity. | |
1403 | * | |
1404 | * This only happens if we're about to log the second dummy transaction as part | |
1405 | * of covering the log and we can get the log incompat feature usage lock. | |
1406 | */ | |
1407 | static inline void | |
1408 | xlog_clear_incompat( | |
1409 | struct xlog *log) | |
1410 | { | |
1411 | struct xfs_mount *mp = log->l_mp; | |
1412 | ||
1413 | if (!xfs_sb_has_incompat_log_feature(&mp->m_sb, | |
1414 | XFS_SB_FEAT_INCOMPAT_LOG_ALL)) | |
1415 | return; | |
1416 | ||
1417 | if (log->l_covered_state != XLOG_STATE_COVER_DONE2) | |
1418 | return; | |
1419 | ||
1420 | if (!down_write_trylock(&log->l_incompat_users)) | |
1421 | return; | |
1422 | ||
1423 | xfs_clear_incompat_log_features(mp); | |
1424 | up_write(&log->l_incompat_users); | |
1425 | } | |
1426 | ||
f661f1e0 DC |
1427 | /* |
1428 | * Every sync period we need to unpin all items in the AIL and push them to | |
1429 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1430 | * indicate that the filesystem is idle. | |
1431 | */ | |
0d5a75e9 | 1432 | static void |
f661f1e0 DC |
1433 | xfs_log_worker( |
1434 | struct work_struct *work) | |
1435 | { | |
1436 | struct xlog *log = container_of(to_delayed_work(work), | |
1437 | struct xlog, l_work); | |
1438 | struct xfs_mount *mp = log->l_mp; | |
1439 | ||
1440 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
37444fc4 | 1441 | if (xfs_fs_writable(mp, SB_FREEZE_WRITE) && xfs_log_need_covered(mp)) { |
61e63ecb DC |
1442 | /* |
1443 | * Dump a transaction into the log that contains no real change. | |
1444 | * This is needed to stamp the current tail LSN into the log | |
1445 | * during the covering operation. | |
1446 | * | |
1447 | * We cannot use an inode here for this - that will push dirty | |
1448 | * state back up into the VFS and then periodic inode flushing | |
1449 | * will prevent log covering from making progress. Hence we | |
1450 | * synchronously log the superblock instead to ensure the | |
1451 | * superblock is immediately unpinned and can be written back. | |
1452 | */ | |
2b73a2c8 | 1453 | xlog_clear_incompat(log); |
61e63ecb DC |
1454 | xfs_sync_sb(mp, true); |
1455 | } else | |
f661f1e0 DC |
1456 | xfs_log_force(mp, 0); |
1457 | ||
1458 | /* start pushing all the metadata that is currently dirty */ | |
1459 | xfs_ail_push_all(mp->m_ail); | |
1460 | ||
1461 | /* queue us up again */ | |
1462 | xfs_log_work_queue(mp); | |
1463 | } | |
1464 | ||
1da177e4 LT |
1465 | /* |
1466 | * This routine initializes some of the log structure for a given mount point. | |
1467 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1468 | * some other stuff may be filled in too. | |
1469 | */ | |
9a8d2fdb MT |
1470 | STATIC struct xlog * |
1471 | xlog_alloc_log( | |
1472 | struct xfs_mount *mp, | |
1473 | struct xfs_buftarg *log_target, | |
1474 | xfs_daddr_t blk_offset, | |
1475 | int num_bblks) | |
1da177e4 | 1476 | { |
9a8d2fdb | 1477 | struct xlog *log; |
1da177e4 LT |
1478 | xlog_rec_header_t *head; |
1479 | xlog_in_core_t **iclogp; | |
1480 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1da177e4 | 1481 | int i; |
2451337d | 1482 | int error = -ENOMEM; |
69ce58f0 | 1483 | uint log2_size = 0; |
1da177e4 | 1484 | |
9a8d2fdb | 1485 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1486 | if (!log) { |
a0fa2b67 | 1487 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1488 | goto out; |
1489 | } | |
1da177e4 LT |
1490 | |
1491 | log->l_mp = mp; | |
1492 | log->l_targ = log_target; | |
1493 | log->l_logsize = BBTOB(num_bblks); | |
1494 | log->l_logBBstart = blk_offset; | |
1495 | log->l_logBBsize = num_bblks; | |
1496 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
e1d06e5f | 1497 | set_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate); |
f661f1e0 | 1498 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1499 | |
1500 | log->l_prev_block = -1; | |
1da177e4 | 1501 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1502 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1503 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1504 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 | 1505 | |
a6a65fef DC |
1506 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) |
1507 | log->l_iclog_roundoff = mp->m_sb.sb_logsunit; | |
1508 | else | |
1509 | log->l_iclog_roundoff = BBSIZE; | |
1510 | ||
c303c5b8 CH |
1511 | xlog_grant_head_init(&log->l_reserve_head); |
1512 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1513 | |
2451337d | 1514 | error = -EFSCORRUPTED; |
62118709 | 1515 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1516 | log2_size = mp->m_sb.sb_logsectlog; |
1517 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1518 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1519 | log2_size, BBSHIFT); | |
a6cb767e DC |
1520 | goto out_free_log; |
1521 | } | |
1522 | ||
69ce58f0 AE |
1523 | log2_size -= BBSHIFT; |
1524 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1525 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1526 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1527 | goto out_free_log; |
1528 | } | |
69ce58f0 AE |
1529 | |
1530 | /* for larger sector sizes, must have v2 or external log */ | |
1531 | if (log2_size && log->l_logBBstart > 0 && | |
1532 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1533 | xfs_warn(mp, |
1534 | "log sector size (0x%x) invalid for configuration.", | |
1535 | log2_size); | |
a6cb767e DC |
1536 | goto out_free_log; |
1537 | } | |
1da177e4 | 1538 | } |
69ce58f0 | 1539 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 | 1540 | |
2b73a2c8 DW |
1541 | init_rwsem(&log->l_incompat_users); |
1542 | ||
1da177e4 LT |
1543 | xlog_get_iclog_buffer_size(mp, log); |
1544 | ||
007c61c6 | 1545 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1546 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1547 | |
1da177e4 LT |
1548 | iclogp = &log->l_iclog; |
1549 | /* | |
1550 | * The amount of memory to allocate for the iclog structure is | |
1551 | * rather funky due to the way the structure is defined. It is | |
1552 | * done this way so that we can use different sizes for machines | |
1553 | * with different amounts of memory. See the definition of | |
1554 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1555 | */ | |
1da177e4 | 1556 | ASSERT(log->l_iclog_size >= 4096); |
79b54d9b | 1557 | for (i = 0; i < log->l_iclog_bufs; i++) { |
89b171ac CH |
1558 | size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE) * |
1559 | sizeof(struct bio_vec); | |
79b54d9b CH |
1560 | |
1561 | iclog = kmem_zalloc(sizeof(*iclog) + bvec_size, KM_MAYFAIL); | |
1562 | if (!iclog) | |
644c3567 DC |
1563 | goto out_free_iclog; |
1564 | ||
79b54d9b | 1565 | *iclogp = iclog; |
1da177e4 LT |
1566 | iclog->ic_prev = prev_iclog; |
1567 | prev_iclog = iclog; | |
1fa40b01 | 1568 | |
d634525d DC |
1569 | iclog->ic_data = kvzalloc(log->l_iclog_size, |
1570 | GFP_KERNEL | __GFP_RETRY_MAYFAIL); | |
79b54d9b | 1571 | if (!iclog->ic_data) |
644c3567 | 1572 | goto out_free_iclog; |
4679b2d3 | 1573 | #ifdef DEBUG |
5809d5e0 | 1574 | log->l_iclog_bak[i] = &iclog->ic_header; |
4679b2d3 | 1575 | #endif |
1da177e4 LT |
1576 | head = &iclog->ic_header; |
1577 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1578 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1579 | head->h_version = cpu_to_be32( | |
62118709 | 1580 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1581 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1582 | /* new fields */ |
b53e675d | 1583 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1584 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1585 | ||
79b54d9b | 1586 | iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize; |
1da177e4 LT |
1587 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1588 | iclog->ic_log = log; | |
114d23aa | 1589 | atomic_set(&iclog->ic_refcnt, 0); |
89ae379d | 1590 | INIT_LIST_HEAD(&iclog->ic_callbacks); |
b28708d6 | 1591 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1592 | |
eb40a875 DC |
1593 | init_waitqueue_head(&iclog->ic_force_wait); |
1594 | init_waitqueue_head(&iclog->ic_write_wait); | |
79b54d9b CH |
1595 | INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work); |
1596 | sema_init(&iclog->ic_sema, 1); | |
1da177e4 LT |
1597 | |
1598 | iclogp = &iclog->ic_next; | |
1599 | } | |
1600 | *iclogp = log->l_iclog; /* complete ring */ | |
1601 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1602 | ||
1058d0f5 | 1603 | log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s", |
05a302a1 DW |
1604 | XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | |
1605 | WQ_HIGHPRI), | |
1606 | 0, mp->m_super->s_id); | |
1058d0f5 CH |
1607 | if (!log->l_ioend_workqueue) |
1608 | goto out_free_iclog; | |
1609 | ||
71e330b5 DC |
1610 | error = xlog_cil_init(log); |
1611 | if (error) | |
1058d0f5 | 1612 | goto out_destroy_workqueue; |
1da177e4 | 1613 | return log; |
644c3567 | 1614 | |
1058d0f5 CH |
1615 | out_destroy_workqueue: |
1616 | destroy_workqueue(log->l_ioend_workqueue); | |
644c3567 DC |
1617 | out_free_iclog: |
1618 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1619 | prev_iclog = iclog->ic_next; | |
79b54d9b | 1620 | kmem_free(iclog->ic_data); |
644c3567 | 1621 | kmem_free(iclog); |
798a9cad BF |
1622 | if (prev_iclog == log->l_iclog) |
1623 | break; | |
644c3567 | 1624 | } |
644c3567 DC |
1625 | out_free_log: |
1626 | kmem_free(log); | |
a6cb767e | 1627 | out: |
2451337d | 1628 | return ERR_PTR(error); |
1da177e4 LT |
1629 | } /* xlog_alloc_log */ |
1630 | ||
1da177e4 LT |
1631 | /* |
1632 | * Write out the commit record of a transaction associated with the given | |
f10e925d | 1633 | * ticket to close off a running log write. Return the lsn of the commit record. |
1da177e4 | 1634 | */ |
f10e925d | 1635 | int |
55b66332 | 1636 | xlog_commit_record( |
ad223e60 | 1637 | struct xlog *log, |
55b66332 DC |
1638 | struct xlog_ticket *ticket, |
1639 | struct xlog_in_core **iclog, | |
f10e925d | 1640 | xfs_lsn_t *lsn) |
1da177e4 | 1641 | { |
55b66332 DC |
1642 | struct xfs_log_iovec reg = { |
1643 | .i_addr = NULL, | |
1644 | .i_len = 0, | |
1645 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1646 | }; | |
1647 | struct xfs_log_vec vec = { | |
1648 | .lv_niovecs = 1, | |
1649 | .lv_iovecp = ®, | |
1650 | }; | |
f10e925d | 1651 | int error; |
1da177e4 | 1652 | |
2039a272 | 1653 | if (xlog_is_shutdown(log)) |
f10e925d DC |
1654 | return -EIO; |
1655 | ||
3468bb1c | 1656 | error = xlog_write(log, &vec, ticket, lsn, iclog, XLOG_COMMIT_TRANS); |
55b66332 | 1657 | if (error) |
f10e925d | 1658 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1659 | return error; |
55b66332 | 1660 | } |
1da177e4 LT |
1661 | |
1662 | /* | |
ed1575da DW |
1663 | * Compute the LSN that we'd need to push the log tail towards in order to have |
1664 | * (a) enough on-disk log space to log the number of bytes specified, (b) at | |
1665 | * least 25% of the log space free, and (c) at least 256 blocks free. If the | |
1666 | * log free space already meets all three thresholds, this function returns | |
1667 | * NULLCOMMITLSN. | |
1da177e4 | 1668 | */ |
ed1575da DW |
1669 | xfs_lsn_t |
1670 | xlog_grant_push_threshold( | |
ad223e60 | 1671 | struct xlog *log, |
2ced19cb | 1672 | int need_bytes) |
1da177e4 | 1673 | { |
2ced19cb | 1674 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1675 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1676 | int free_blocks; |
1677 | int free_bytes; | |
1678 | int threshold_block; | |
1679 | int threshold_cycle; | |
1680 | int free_threshold; | |
1681 | ||
1682 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1683 | ||
28496968 | 1684 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1685 | free_blocks = BTOBBT(free_bytes); |
1686 | ||
1687 | /* | |
1688 | * Set the threshold for the minimum number of free blocks in the | |
1689 | * log to the maximum of what the caller needs, one quarter of the | |
1690 | * log, and 256 blocks. | |
1691 | */ | |
1692 | free_threshold = BTOBB(need_bytes); | |
9bb54cb5 DC |
1693 | free_threshold = max(free_threshold, (log->l_logBBsize >> 2)); |
1694 | free_threshold = max(free_threshold, 256); | |
2ced19cb | 1695 | if (free_blocks >= free_threshold) |
ed1575da | 1696 | return NULLCOMMITLSN; |
2ced19cb | 1697 | |
1c3cb9ec DC |
1698 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1699 | &threshold_block); | |
1700 | threshold_block += free_threshold; | |
1da177e4 | 1701 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1702 | threshold_block -= log->l_logBBsize; |
1703 | threshold_cycle += 1; | |
1da177e4 | 1704 | } |
2ced19cb DC |
1705 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1706 | threshold_block); | |
1707 | /* | |
1708 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1709 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1710 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1711 | */ |
84f3c683 DC |
1712 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1713 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1714 | threshold_lsn = last_sync_lsn; | |
2ced19cb | 1715 | |
ed1575da DW |
1716 | return threshold_lsn; |
1717 | } | |
1718 | ||
1719 | /* | |
1720 | * Push the tail of the log if we need to do so to maintain the free log space | |
1721 | * thresholds set out by xlog_grant_push_threshold. We may need to adopt a | |
1722 | * policy which pushes on an lsn which is further along in the log once we | |
1723 | * reach the high water mark. In this manner, we would be creating a low water | |
1724 | * mark. | |
1725 | */ | |
1726 | STATIC void | |
1727 | xlog_grant_push_ail( | |
1728 | struct xlog *log, | |
1729 | int need_bytes) | |
1730 | { | |
1731 | xfs_lsn_t threshold_lsn; | |
1732 | ||
1733 | threshold_lsn = xlog_grant_push_threshold(log, need_bytes); | |
2039a272 | 1734 | if (threshold_lsn == NULLCOMMITLSN || xlog_is_shutdown(log)) |
ed1575da DW |
1735 | return; |
1736 | ||
2ced19cb DC |
1737 | /* |
1738 | * Get the transaction layer to kick the dirty buffers out to | |
1739 | * disk asynchronously. No point in trying to do this if | |
1740 | * the filesystem is shutting down. | |
1741 | */ | |
ed1575da | 1742 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1743 | } |
1da177e4 | 1744 | |
0e446be4 CH |
1745 | /* |
1746 | * Stamp cycle number in every block | |
1747 | */ | |
1748 | STATIC void | |
1749 | xlog_pack_data( | |
1750 | struct xlog *log, | |
1751 | struct xlog_in_core *iclog, | |
1752 | int roundoff) | |
1753 | { | |
1754 | int i, j, k; | |
1755 | int size = iclog->ic_offset + roundoff; | |
1756 | __be32 cycle_lsn; | |
b2a922cd | 1757 | char *dp; |
0e446be4 CH |
1758 | |
1759 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1760 | ||
1761 | dp = iclog->ic_datap; | |
1762 | for (i = 0; i < BTOBB(size); i++) { | |
1763 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1764 | break; | |
1765 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1766 | *(__be32 *)dp = cycle_lsn; | |
1767 | dp += BBSIZE; | |
1768 | } | |
1769 | ||
1770 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1771 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1772 | ||
1773 | for ( ; i < BTOBB(size); i++) { | |
1774 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1775 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1776 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1777 | *(__be32 *)dp = cycle_lsn; | |
1778 | dp += BBSIZE; | |
1779 | } | |
1780 | ||
1781 | for (i = 1; i < log->l_iclog_heads; i++) | |
1782 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1783 | } | |
1784 | } | |
1785 | ||
1786 | /* | |
1787 | * Calculate the checksum for a log buffer. | |
1788 | * | |
1789 | * This is a little more complicated than it should be because the various | |
1790 | * headers and the actual data are non-contiguous. | |
1791 | */ | |
f9668a09 | 1792 | __le32 |
0e446be4 CH |
1793 | xlog_cksum( |
1794 | struct xlog *log, | |
1795 | struct xlog_rec_header *rhead, | |
1796 | char *dp, | |
1797 | int size) | |
1798 | { | |
c8ce540d | 1799 | uint32_t crc; |
0e446be4 CH |
1800 | |
1801 | /* first generate the crc for the record header ... */ | |
cae028df | 1802 | crc = xfs_start_cksum_update((char *)rhead, |
0e446be4 CH |
1803 | sizeof(struct xlog_rec_header), |
1804 | offsetof(struct xlog_rec_header, h_crc)); | |
1805 | ||
1806 | /* ... then for additional cycle data for v2 logs ... */ | |
1807 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1808 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1809 | int i; | |
a3f20014 | 1810 | int xheads; |
0e446be4 | 1811 | |
0c771b99 | 1812 | xheads = DIV_ROUND_UP(size, XLOG_HEADER_CYCLE_SIZE); |
0e446be4 | 1813 | |
a3f20014 | 1814 | for (i = 1; i < xheads; i++) { |
0e446be4 CH |
1815 | crc = crc32c(crc, &xhdr[i].hic_xheader, |
1816 | sizeof(struct xlog_rec_ext_header)); | |
1817 | } | |
1818 | } | |
1819 | ||
1820 | /* ... and finally for the payload */ | |
1821 | crc = crc32c(crc, dp, size); | |
1822 | ||
1823 | return xfs_end_cksum(crc); | |
1824 | } | |
1825 | ||
79b54d9b CH |
1826 | static void |
1827 | xlog_bio_end_io( | |
1828 | struct bio *bio) | |
1829 | { | |
1830 | struct xlog_in_core *iclog = bio->bi_private; | |
1831 | ||
1058d0f5 | 1832 | queue_work(iclog->ic_log->l_ioend_workqueue, |
79b54d9b CH |
1833 | &iclog->ic_end_io_work); |
1834 | } | |
1835 | ||
842a42d1 | 1836 | static int |
79b54d9b CH |
1837 | xlog_map_iclog_data( |
1838 | struct bio *bio, | |
1839 | void *data, | |
1840 | size_t count) | |
1841 | { | |
1842 | do { | |
1843 | struct page *page = kmem_to_page(data); | |
1844 | unsigned int off = offset_in_page(data); | |
1845 | size_t len = min_t(size_t, count, PAGE_SIZE - off); | |
1846 | ||
842a42d1 BF |
1847 | if (bio_add_page(bio, page, len, off) != len) |
1848 | return -EIO; | |
79b54d9b CH |
1849 | |
1850 | data += len; | |
1851 | count -= len; | |
1852 | } while (count); | |
842a42d1 BF |
1853 | |
1854 | return 0; | |
79b54d9b CH |
1855 | } |
1856 | ||
94860a30 CH |
1857 | STATIC void |
1858 | xlog_write_iclog( | |
1859 | struct xlog *log, | |
1860 | struct xlog_in_core *iclog, | |
94860a30 | 1861 | uint64_t bno, |
eef983ff | 1862 | unsigned int count) |
873ff550 | 1863 | { |
94860a30 | 1864 | ASSERT(bno < log->l_logBBsize); |
956f6daa | 1865 | trace_xlog_iclog_write(iclog, _RET_IP_); |
94860a30 CH |
1866 | |
1867 | /* | |
1868 | * We lock the iclogbufs here so that we can serialise against I/O | |
1869 | * completion during unmount. We might be processing a shutdown | |
1870 | * triggered during unmount, and that can occur asynchronously to the | |
1871 | * unmount thread, and hence we need to ensure that completes before | |
1872 | * tearing down the iclogbufs. Hence we need to hold the buffer lock | |
1873 | * across the log IO to archieve that. | |
1874 | */ | |
79b54d9b | 1875 | down(&iclog->ic_sema); |
5112e206 | 1876 | if (xlog_is_shutdown(log)) { |
873ff550 CH |
1877 | /* |
1878 | * It would seem logical to return EIO here, but we rely on | |
1879 | * the log state machine to propagate I/O errors instead of | |
79b54d9b CH |
1880 | * doing it here. We kick of the state machine and unlock |
1881 | * the buffer manually, the code needs to be kept in sync | |
1882 | * with the I/O completion path. | |
873ff550 | 1883 | */ |
12e6a0f4 | 1884 | xlog_state_done_syncing(iclog); |
79b54d9b | 1885 | up(&iclog->ic_sema); |
94860a30 | 1886 | return; |
873ff550 CH |
1887 | } |
1888 | ||
79b54d9b CH |
1889 | bio_init(&iclog->ic_bio, iclog->ic_bvec, howmany(count, PAGE_SIZE)); |
1890 | bio_set_dev(&iclog->ic_bio, log->l_targ->bt_bdev); | |
1891 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno; | |
1892 | iclog->ic_bio.bi_end_io = xlog_bio_end_io; | |
1893 | iclog->ic_bio.bi_private = iclog; | |
2def2845 DC |
1894 | |
1895 | /* | |
1896 | * We use REQ_SYNC | REQ_IDLE here to tell the block layer the are more | |
1897 | * IOs coming immediately after this one. This prevents the block layer | |
1898 | * writeback throttle from throttling log writes behind background | |
1899 | * metadata writeback and causing priority inversions. | |
1900 | */ | |
eef983ff | 1901 | iclog->ic_bio.bi_opf = REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_IDLE; |
b5d721ea | 1902 | if (iclog->ic_flags & XLOG_ICL_NEED_FLUSH) { |
79b54d9b | 1903 | iclog->ic_bio.bi_opf |= REQ_PREFLUSH; |
b5d721ea DC |
1904 | /* |
1905 | * For external log devices, we also need to flush the data | |
1906 | * device cache first to ensure all metadata writeback covered | |
1907 | * by the LSN in this iclog is on stable storage. This is slow, | |
1908 | * but it *must* complete before we issue the external log IO. | |
1909 | */ | |
1910 | if (log->l_targ != log->l_mp->m_ddev_targp) | |
1911 | blkdev_issue_flush(log->l_mp->m_ddev_targp->bt_bdev); | |
1912 | } | |
eef983ff DC |
1913 | if (iclog->ic_flags & XLOG_ICL_NEED_FUA) |
1914 | iclog->ic_bio.bi_opf |= REQ_FUA; | |
b5d721ea | 1915 | |
eef983ff | 1916 | iclog->ic_flags &= ~(XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA); |
79b54d9b | 1917 | |
842a42d1 BF |
1918 | if (xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, count)) { |
1919 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1920 | return; | |
1921 | } | |
79b54d9b | 1922 | if (is_vmalloc_addr(iclog->ic_data)) |
2c68a1df | 1923 | flush_kernel_vmap_range(iclog->ic_data, count); |
79b54d9b CH |
1924 | |
1925 | /* | |
1926 | * If this log buffer would straddle the end of the log we will have | |
1927 | * to split it up into two bios, so that we can continue at the start. | |
1928 | */ | |
1929 | if (bno + BTOBB(count) > log->l_logBBsize) { | |
1930 | struct bio *split; | |
1931 | ||
1932 | split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno, | |
1933 | GFP_NOIO, &fs_bio_set); | |
1934 | bio_chain(split, &iclog->ic_bio); | |
1935 | submit_bio(split); | |
1936 | ||
1937 | /* restart at logical offset zero for the remainder */ | |
1938 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart; | |
1939 | } | |
1940 | ||
1941 | submit_bio(&iclog->ic_bio); | |
873ff550 | 1942 | } |
1da177e4 | 1943 | |
56933848 CH |
1944 | /* |
1945 | * We need to bump cycle number for the part of the iclog that is | |
1946 | * written to the start of the log. Watch out for the header magic | |
1947 | * number case, though. | |
1948 | */ | |
79b54d9b | 1949 | static void |
56933848 CH |
1950 | xlog_split_iclog( |
1951 | struct xlog *log, | |
1952 | void *data, | |
1953 | uint64_t bno, | |
1954 | unsigned int count) | |
1955 | { | |
1956 | unsigned int split_offset = BBTOB(log->l_logBBsize - bno); | |
1957 | unsigned int i; | |
1958 | ||
1959 | for (i = split_offset; i < count; i += BBSIZE) { | |
1960 | uint32_t cycle = get_unaligned_be32(data + i); | |
1961 | ||
1962 | if (++cycle == XLOG_HEADER_MAGIC_NUM) | |
1963 | cycle++; | |
1964 | put_unaligned_be32(cycle, data + i); | |
1965 | } | |
56933848 CH |
1966 | } |
1967 | ||
db0a6faf CH |
1968 | static int |
1969 | xlog_calc_iclog_size( | |
1970 | struct xlog *log, | |
1971 | struct xlog_in_core *iclog, | |
1972 | uint32_t *roundoff) | |
1973 | { | |
1974 | uint32_t count_init, count; | |
db0a6faf CH |
1975 | |
1976 | /* Add for LR header */ | |
1977 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
a6a65fef | 1978 | count = roundup(count_init, log->l_iclog_roundoff); |
db0a6faf | 1979 | |
db0a6faf CH |
1980 | *roundoff = count - count_init; |
1981 | ||
a6a65fef DC |
1982 | ASSERT(count >= count_init); |
1983 | ASSERT(*roundoff < log->l_iclog_roundoff); | |
db0a6faf CH |
1984 | return count; |
1985 | } | |
1986 | ||
1da177e4 LT |
1987 | /* |
1988 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1989 | * fashion. Previously, we should have moved the current iclog | |
1990 | * ptr in the log to point to the next available iclog. This allows further | |
1991 | * write to continue while this code syncs out an iclog ready to go. | |
1992 | * Before an in-core log can be written out, the data section must be scanned | |
1993 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1994 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1995 | * cycle count because there in an implicit assumption that drives will | |
1996 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1997 | * we can't have part of a 512 byte block written and part not written. By | |
1998 | * tagging each block, we will know which blocks are valid when recovering | |
1999 | * after an unclean shutdown. | |
2000 | * | |
2001 | * This routine is single threaded on the iclog. No other thread can be in | |
2002 | * this routine with the same iclog. Changing contents of iclog can there- | |
2003 | * fore be done without grabbing the state machine lock. Updating the global | |
2004 | * log will require grabbing the lock though. | |
2005 | * | |
2006 | * The entire log manager uses a logical block numbering scheme. Only | |
94860a30 CH |
2007 | * xlog_write_iclog knows about the fact that the log may not start with |
2008 | * block zero on a given device. | |
1da177e4 | 2009 | */ |
94860a30 | 2010 | STATIC void |
9a8d2fdb MT |
2011 | xlog_sync( |
2012 | struct xlog *log, | |
2013 | struct xlog_in_core *iclog) | |
1da177e4 | 2014 | { |
db0a6faf CH |
2015 | unsigned int count; /* byte count of bwrite */ |
2016 | unsigned int roundoff; /* roundoff to BB or stripe */ | |
2017 | uint64_t bno; | |
db0a6faf | 2018 | unsigned int size; |
1da177e4 | 2019 | |
155cc6b7 | 2020 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
956f6daa | 2021 | trace_xlog_iclog_sync(iclog, _RET_IP_); |
1da177e4 | 2022 | |
db0a6faf | 2023 | count = xlog_calc_iclog_size(log, iclog, &roundoff); |
1da177e4 LT |
2024 | |
2025 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
2026 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
2027 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
2028 | |
2029 | /* put cycle number in every block */ | |
2030 | xlog_pack_data(log, iclog, roundoff); | |
2031 | ||
2032 | /* real byte length */ | |
0e446be4 | 2033 | size = iclog->ic_offset; |
db0a6faf | 2034 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) |
0e446be4 CH |
2035 | size += roundoff; |
2036 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 2037 | |
9b0489c1 | 2038 | XFS_STATS_INC(log->l_mp, xs_log_writes); |
ff6d6af2 | 2039 | XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count)); |
1da177e4 | 2040 | |
94860a30 CH |
2041 | bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)); |
2042 | ||
1da177e4 | 2043 | /* Do we need to split this write into 2 parts? */ |
eef983ff | 2044 | if (bno + BTOBB(count) > log->l_logBBsize) |
79b54d9b | 2045 | xlog_split_iclog(log, &iclog->ic_header, bno, count); |
0e446be4 CH |
2046 | |
2047 | /* calculcate the checksum */ | |
2048 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
2049 | iclog->ic_datap, size); | |
609adfc2 BF |
2050 | /* |
2051 | * Intentionally corrupt the log record CRC based on the error injection | |
2052 | * frequency, if defined. This facilitates testing log recovery in the | |
2053 | * event of torn writes. Hence, set the IOABORT state to abort the log | |
2054 | * write on I/O completion and shutdown the fs. The subsequent mount | |
2055 | * detects the bad CRC and attempts to recover. | |
2056 | */ | |
366fc4b8 | 2057 | #ifdef DEBUG |
3e88a007 | 2058 | if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) { |
e2a64192 | 2059 | iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA); |
366fc4b8 | 2060 | iclog->ic_fail_crc = true; |
609adfc2 BF |
2061 | xfs_warn(log->l_mp, |
2062 | "Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.", | |
2063 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
2064 | } | |
366fc4b8 | 2065 | #endif |
abca1f33 | 2066 | xlog_verify_iclog(log, iclog, count); |
eef983ff | 2067 | xlog_write_iclog(log, iclog, bno, count); |
94860a30 | 2068 | } |
1da177e4 | 2069 | |
1da177e4 | 2070 | /* |
c41564b5 | 2071 | * Deallocate a log structure |
1da177e4 | 2072 | */ |
a8272ce0 | 2073 | STATIC void |
9a8d2fdb MT |
2074 | xlog_dealloc_log( |
2075 | struct xlog *log) | |
1da177e4 LT |
2076 | { |
2077 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
2078 | int i; |
2079 | ||
71e330b5 DC |
2080 | xlog_cil_destroy(log); |
2081 | ||
44396476 | 2082 | /* |
9c23eccc DC |
2083 | * Cycle all the iclogbuf locks to make sure all log IO completion |
2084 | * is done before we tear down these buffers. | |
2085 | */ | |
2086 | iclog = log->l_iclog; | |
2087 | for (i = 0; i < log->l_iclog_bufs; i++) { | |
79b54d9b CH |
2088 | down(&iclog->ic_sema); |
2089 | up(&iclog->ic_sema); | |
9c23eccc DC |
2090 | iclog = iclog->ic_next; |
2091 | } | |
2092 | ||
1da177e4 | 2093 | iclog = log->l_iclog; |
9c23eccc | 2094 | for (i = 0; i < log->l_iclog_bufs; i++) { |
1da177e4 | 2095 | next_iclog = iclog->ic_next; |
79b54d9b | 2096 | kmem_free(iclog->ic_data); |
f0e2d93c | 2097 | kmem_free(iclog); |
1da177e4 LT |
2098 | iclog = next_iclog; |
2099 | } | |
1da177e4 | 2100 | |
1da177e4 | 2101 | log->l_mp->m_log = NULL; |
1058d0f5 | 2102 | destroy_workqueue(log->l_ioend_workqueue); |
f0e2d93c | 2103 | kmem_free(log); |
b843299b | 2104 | } |
1da177e4 LT |
2105 | |
2106 | /* | |
2107 | * Update counters atomically now that memcpy is done. | |
2108 | */ | |
1da177e4 | 2109 | static inline void |
9a8d2fdb MT |
2110 | xlog_state_finish_copy( |
2111 | struct xlog *log, | |
2112 | struct xlog_in_core *iclog, | |
2113 | int record_cnt, | |
2114 | int copy_bytes) | |
1da177e4 | 2115 | { |
390aab0a | 2116 | lockdep_assert_held(&log->l_icloglock); |
1da177e4 | 2117 | |
413d57c9 | 2118 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 | 2119 | iclog->ic_offset += copy_bytes; |
390aab0a | 2120 | } |
1da177e4 | 2121 | |
7e9c6396 TS |
2122 | /* |
2123 | * print out info relating to regions written which consume | |
2124 | * the reservation | |
2125 | */ | |
71e330b5 DC |
2126 | void |
2127 | xlog_print_tic_res( | |
2128 | struct xfs_mount *mp, | |
2129 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
2130 | { |
2131 | uint i; | |
2132 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
2133 | ||
2134 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
5110cd82 | 2135 | #define REG_TYPE_STR(type, str) [XLOG_REG_TYPE_##type] = str |
d31d7185 | 2136 | static char *res_type_str[] = { |
5110cd82 DW |
2137 | REG_TYPE_STR(BFORMAT, "bformat"), |
2138 | REG_TYPE_STR(BCHUNK, "bchunk"), | |
2139 | REG_TYPE_STR(EFI_FORMAT, "efi_format"), | |
2140 | REG_TYPE_STR(EFD_FORMAT, "efd_format"), | |
2141 | REG_TYPE_STR(IFORMAT, "iformat"), | |
2142 | REG_TYPE_STR(ICORE, "icore"), | |
2143 | REG_TYPE_STR(IEXT, "iext"), | |
2144 | REG_TYPE_STR(IBROOT, "ibroot"), | |
2145 | REG_TYPE_STR(ILOCAL, "ilocal"), | |
2146 | REG_TYPE_STR(IATTR_EXT, "iattr_ext"), | |
2147 | REG_TYPE_STR(IATTR_BROOT, "iattr_broot"), | |
2148 | REG_TYPE_STR(IATTR_LOCAL, "iattr_local"), | |
2149 | REG_TYPE_STR(QFORMAT, "qformat"), | |
2150 | REG_TYPE_STR(DQUOT, "dquot"), | |
2151 | REG_TYPE_STR(QUOTAOFF, "quotaoff"), | |
2152 | REG_TYPE_STR(LRHEADER, "LR header"), | |
2153 | REG_TYPE_STR(UNMOUNT, "unmount"), | |
2154 | REG_TYPE_STR(COMMIT, "commit"), | |
2155 | REG_TYPE_STR(TRANSHDR, "trans header"), | |
d31d7185 DW |
2156 | REG_TYPE_STR(ICREATE, "inode create"), |
2157 | REG_TYPE_STR(RUI_FORMAT, "rui_format"), | |
2158 | REG_TYPE_STR(RUD_FORMAT, "rud_format"), | |
2159 | REG_TYPE_STR(CUI_FORMAT, "cui_format"), | |
2160 | REG_TYPE_STR(CUD_FORMAT, "cud_format"), | |
2161 | REG_TYPE_STR(BUI_FORMAT, "bui_format"), | |
2162 | REG_TYPE_STR(BUD_FORMAT, "bud_format"), | |
7e9c6396 | 2163 | }; |
d31d7185 | 2164 | BUILD_BUG_ON(ARRAY_SIZE(res_type_str) != XLOG_REG_TYPE_MAX + 1); |
5110cd82 | 2165 | #undef REG_TYPE_STR |
7e9c6396 | 2166 | |
7d2d5653 | 2167 | xfs_warn(mp, "ticket reservation summary:"); |
f41febd2 JP |
2168 | xfs_warn(mp, " unit res = %d bytes", |
2169 | ticket->t_unit_res); | |
2170 | xfs_warn(mp, " current res = %d bytes", | |
2171 | ticket->t_curr_res); | |
2172 | xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)", | |
2173 | ticket->t_res_arr_sum, ticket->t_res_o_flow); | |
2174 | xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)", | |
2175 | ticket->t_res_num_ophdrs, ophdr_spc); | |
2176 | xfs_warn(mp, " ophdr + reg = %u bytes", | |
2177 | ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc); | |
2178 | xfs_warn(mp, " num regions = %u", | |
2179 | ticket->t_res_num); | |
7e9c6396 TS |
2180 | |
2181 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 2182 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 2183 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 | 2184 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
5110cd82 | 2185 | "bad-rtype" : res_type_str[r_type]), |
7e9c6396 TS |
2186 | ticket->t_res_arr[i].r_len); |
2187 | } | |
2188 | } | |
7e9c6396 | 2189 | |
d4ca1d55 BF |
2190 | /* |
2191 | * Print a summary of the transaction. | |
2192 | */ | |
2193 | void | |
2194 | xlog_print_trans( | |
e6631f85 | 2195 | struct xfs_trans *tp) |
d4ca1d55 | 2196 | { |
e6631f85 DC |
2197 | struct xfs_mount *mp = tp->t_mountp; |
2198 | struct xfs_log_item *lip; | |
d4ca1d55 BF |
2199 | |
2200 | /* dump core transaction and ticket info */ | |
2201 | xfs_warn(mp, "transaction summary:"); | |
2c8f6265 BF |
2202 | xfs_warn(mp, " log res = %d", tp->t_log_res); |
2203 | xfs_warn(mp, " log count = %d", tp->t_log_count); | |
2204 | xfs_warn(mp, " flags = 0x%x", tp->t_flags); | |
d4ca1d55 BF |
2205 | |
2206 | xlog_print_tic_res(mp, tp->t_ticket); | |
2207 | ||
2208 | /* dump each log item */ | |
e6631f85 | 2209 | list_for_each_entry(lip, &tp->t_items, li_trans) { |
d4ca1d55 BF |
2210 | struct xfs_log_vec *lv = lip->li_lv; |
2211 | struct xfs_log_iovec *vec; | |
2212 | int i; | |
2213 | ||
2214 | xfs_warn(mp, "log item: "); | |
2215 | xfs_warn(mp, " type = 0x%x", lip->li_type); | |
22525c17 | 2216 | xfs_warn(mp, " flags = 0x%lx", lip->li_flags); |
d4ca1d55 BF |
2217 | if (!lv) |
2218 | continue; | |
2219 | xfs_warn(mp, " niovecs = %d", lv->lv_niovecs); | |
2220 | xfs_warn(mp, " size = %d", lv->lv_size); | |
2221 | xfs_warn(mp, " bytes = %d", lv->lv_bytes); | |
2222 | xfs_warn(mp, " buf len = %d", lv->lv_buf_len); | |
2223 | ||
2224 | /* dump each iovec for the log item */ | |
2225 | vec = lv->lv_iovecp; | |
2226 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2227 | int dumplen = min(vec->i_len, 32); | |
2228 | ||
2229 | xfs_warn(mp, " iovec[%d]", i); | |
2230 | xfs_warn(mp, " type = 0x%x", vec->i_type); | |
2231 | xfs_warn(mp, " len = %d", vec->i_len); | |
2232 | xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i); | |
244e3dea | 2233 | xfs_hex_dump(vec->i_addr, dumplen); |
d4ca1d55 BF |
2234 | |
2235 | vec++; | |
2236 | } | |
2237 | } | |
2238 | } | |
2239 | ||
b5203cd0 | 2240 | /* |
7ec94921 DC |
2241 | * Calculate the potential space needed by the log vector. We may need a start |
2242 | * record, and each region gets its own struct xlog_op_header and may need to be | |
2243 | * double word aligned. | |
b5203cd0 DC |
2244 | */ |
2245 | static int | |
2246 | xlog_write_calc_vec_length( | |
2247 | struct xlog_ticket *ticket, | |
7ec94921 | 2248 | struct xfs_log_vec *log_vector, |
3468bb1c | 2249 | uint optype) |
b5203cd0 | 2250 | { |
55b66332 | 2251 | struct xfs_log_vec *lv; |
3468bb1c | 2252 | int headers = 0; |
b5203cd0 DC |
2253 | int len = 0; |
2254 | int i; | |
2255 | ||
3468bb1c DC |
2256 | if (optype & XLOG_START_TRANS) |
2257 | headers++; | |
2258 | ||
55b66332 | 2259 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2260 | /* we don't write ordered log vectors */ |
2261 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2262 | continue; | |
2263 | ||
55b66332 DC |
2264 | headers += lv->lv_niovecs; |
2265 | ||
2266 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2267 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2268 | |
55b66332 DC |
2269 | len += vecp->i_len; |
2270 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2271 | } | |
b5203cd0 DC |
2272 | } |
2273 | ||
2274 | ticket->t_res_num_ophdrs += headers; | |
2275 | len += headers * sizeof(struct xlog_op_header); | |
2276 | ||
2277 | return len; | |
2278 | } | |
2279 | ||
7ec94921 | 2280 | static void |
b5203cd0 | 2281 | xlog_write_start_rec( |
e6b1f273 | 2282 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2283 | struct xlog_ticket *ticket) |
2284 | { | |
b5203cd0 DC |
2285 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2286 | ophdr->oh_clientid = ticket->t_clientid; | |
2287 | ophdr->oh_len = 0; | |
2288 | ophdr->oh_flags = XLOG_START_TRANS; | |
2289 | ophdr->oh_res2 = 0; | |
b5203cd0 DC |
2290 | } |
2291 | ||
2292 | static xlog_op_header_t * | |
2293 | xlog_write_setup_ophdr( | |
ad223e60 | 2294 | struct xlog *log, |
e6b1f273 | 2295 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2296 | struct xlog_ticket *ticket, |
2297 | uint flags) | |
2298 | { | |
b5203cd0 DC |
2299 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2300 | ophdr->oh_clientid = ticket->t_clientid; | |
2301 | ophdr->oh_res2 = 0; | |
2302 | ||
2303 | /* are we copying a commit or unmount record? */ | |
2304 | ophdr->oh_flags = flags; | |
2305 | ||
2306 | /* | |
2307 | * We've seen logs corrupted with bad transaction client ids. This | |
2308 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2309 | * and shut down the filesystem. | |
2310 | */ | |
2311 | switch (ophdr->oh_clientid) { | |
2312 | case XFS_TRANSACTION: | |
2313 | case XFS_VOLUME: | |
2314 | case XFS_LOG: | |
2315 | break; | |
2316 | default: | |
a0fa2b67 | 2317 | xfs_warn(log->l_mp, |
c9690043 | 2318 | "Bad XFS transaction clientid 0x%x in ticket "PTR_FMT, |
b5203cd0 DC |
2319 | ophdr->oh_clientid, ticket); |
2320 | return NULL; | |
2321 | } | |
2322 | ||
2323 | return ophdr; | |
2324 | } | |
2325 | ||
2326 | /* | |
2327 | * Set up the parameters of the region copy into the log. This has | |
2328 | * to handle region write split across multiple log buffers - this | |
2329 | * state is kept external to this function so that this code can | |
ac0e300f | 2330 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2331 | */ |
2332 | static int | |
2333 | xlog_write_setup_copy( | |
2334 | struct xlog_ticket *ticket, | |
2335 | struct xlog_op_header *ophdr, | |
2336 | int space_available, | |
2337 | int space_required, | |
2338 | int *copy_off, | |
2339 | int *copy_len, | |
2340 | int *last_was_partial_copy, | |
2341 | int *bytes_consumed) | |
2342 | { | |
2343 | int still_to_copy; | |
2344 | ||
2345 | still_to_copy = space_required - *bytes_consumed; | |
2346 | *copy_off = *bytes_consumed; | |
2347 | ||
2348 | if (still_to_copy <= space_available) { | |
2349 | /* write of region completes here */ | |
2350 | *copy_len = still_to_copy; | |
2351 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2352 | if (*last_was_partial_copy) | |
2353 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2354 | *last_was_partial_copy = 0; | |
2355 | *bytes_consumed = 0; | |
2356 | return 0; | |
2357 | } | |
2358 | ||
2359 | /* partial write of region, needs extra log op header reservation */ | |
2360 | *copy_len = space_available; | |
2361 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2362 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2363 | if (*last_was_partial_copy) | |
2364 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2365 | *bytes_consumed += *copy_len; | |
2366 | (*last_was_partial_copy)++; | |
2367 | ||
2368 | /* account for new log op header */ | |
2369 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2370 | ticket->t_res_num_ophdrs++; | |
2371 | ||
2372 | return sizeof(struct xlog_op_header); | |
2373 | } | |
2374 | ||
2375 | static int | |
2376 | xlog_write_copy_finish( | |
ad223e60 | 2377 | struct xlog *log, |
b5203cd0 DC |
2378 | struct xlog_in_core *iclog, |
2379 | uint flags, | |
2380 | int *record_cnt, | |
2381 | int *data_cnt, | |
2382 | int *partial_copy, | |
2383 | int *partial_copy_len, | |
2384 | int log_offset, | |
2385 | struct xlog_in_core **commit_iclog) | |
2386 | { | |
df732b29 CH |
2387 | int error; |
2388 | ||
b5203cd0 DC |
2389 | if (*partial_copy) { |
2390 | /* | |
2391 | * This iclog has already been marked WANT_SYNC by | |
2392 | * xlog_state_get_iclog_space. | |
2393 | */ | |
390aab0a | 2394 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2395 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2396 | *record_cnt = 0; | |
2397 | *data_cnt = 0; | |
df732b29 | 2398 | goto release_iclog; |
b5203cd0 DC |
2399 | } |
2400 | ||
2401 | *partial_copy = 0; | |
2402 | *partial_copy_len = 0; | |
2403 | ||
2404 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2405 | /* no more space in this iclog - push it. */ | |
390aab0a | 2406 | spin_lock(&log->l_icloglock); |
b5203cd0 DC |
2407 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); |
2408 | *record_cnt = 0; | |
2409 | *data_cnt = 0; | |
2410 | ||
69363999 CH |
2411 | if (iclog->ic_state == XLOG_STATE_ACTIVE) |
2412 | xlog_state_switch_iclogs(log, iclog, 0); | |
2413 | else | |
2414 | ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
5112e206 | 2415 | xlog_is_shutdown(log)); |
b5203cd0 | 2416 | if (!commit_iclog) |
df732b29 CH |
2417 | goto release_iclog; |
2418 | spin_unlock(&log->l_icloglock); | |
b5203cd0 DC |
2419 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2420 | *commit_iclog = iclog; | |
2421 | } | |
2422 | ||
2423 | return 0; | |
df732b29 CH |
2424 | |
2425 | release_iclog: | |
0dc8f7f1 | 2426 | error = xlog_state_release_iclog(log, iclog, 0); |
df732b29 CH |
2427 | spin_unlock(&log->l_icloglock); |
2428 | return error; | |
b5203cd0 DC |
2429 | } |
2430 | ||
1da177e4 LT |
2431 | /* |
2432 | * Write some region out to in-core log | |
2433 | * | |
2434 | * This will be called when writing externally provided regions or when | |
2435 | * writing out a commit record for a given transaction. | |
2436 | * | |
2437 | * General algorithm: | |
2438 | * 1. Find total length of this write. This may include adding to the | |
2439 | * lengths passed in. | |
2440 | * 2. Check whether we violate the tickets reservation. | |
2441 | * 3. While writing to this iclog | |
2442 | * A. Reserve as much space in this iclog as can get | |
2443 | * B. If this is first write, save away start lsn | |
2444 | * C. While writing this region: | |
2445 | * 1. If first write of transaction, write start record | |
2446 | * 2. Write log operation header (header per region) | |
2447 | * 3. Find out if we can fit entire region into this iclog | |
2448 | * 4. Potentially, verify destination memcpy ptr | |
2449 | * 5. Memcpy (partial) region | |
2450 | * 6. If partial copy, release iclog; otherwise, continue | |
2451 | * copying more regions into current iclog | |
2452 | * 4. Mark want sync bit (in simulation mode) | |
2453 | * 5. Release iclog for potential flush to on-disk log. | |
2454 | * | |
2455 | * ERRORS: | |
2456 | * 1. Panic if reservation is overrun. This should never happen since | |
2457 | * reservation amounts are generated internal to the filesystem. | |
2458 | * NOTES: | |
2459 | * 1. Tickets are single threaded data structures. | |
2460 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2461 | * syncing routine. When a single log_write region needs to span | |
2462 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2463 | * on all log operation writes which don't contain the end of the | |
2464 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2465 | * operation which contains the end of the continued log_write region. | |
2466 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2467 | * we don't really know exactly how much space will be used. As a result, | |
2468 | * we don't update ic_offset until the end when we know exactly how many | |
2469 | * bytes have been written out. | |
2470 | */ | |
71e330b5 | 2471 | int |
35a8a72f | 2472 | xlog_write( |
ad223e60 | 2473 | struct xlog *log, |
55b66332 | 2474 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2475 | struct xlog_ticket *ticket, |
2476 | xfs_lsn_t *start_lsn, | |
2477 | struct xlog_in_core **commit_iclog, | |
3468bb1c | 2478 | uint optype) |
1da177e4 | 2479 | { |
99428ad0 | 2480 | struct xlog_in_core *iclog = NULL; |
9590e9c6 DC |
2481 | struct xfs_log_vec *lv = log_vector; |
2482 | struct xfs_log_iovec *vecp = lv->lv_iovecp; | |
2483 | int index = 0; | |
99428ad0 | 2484 | int len; |
99428ad0 CH |
2485 | int partial_copy = 0; |
2486 | int partial_copy_len = 0; | |
2487 | int contwr = 0; | |
2488 | int record_cnt = 0; | |
2489 | int data_cnt = 0; | |
df732b29 | 2490 | int error = 0; |
99428ad0 | 2491 | |
93b8a585 | 2492 | /* |
9590e9c6 DC |
2493 | * If this is a commit or unmount transaction, we don't need a start |
2494 | * record to be written. We do, however, have to account for the | |
2495 | * commit or unmount header that gets written. Hence we always have | |
2496 | * to account for an extra xlog_op_header here. | |
93b8a585 | 2497 | */ |
9590e9c6 | 2498 | ticket->t_curr_res -= sizeof(struct xlog_op_header); |
7d2d5653 BF |
2499 | if (ticket->t_curr_res < 0) { |
2500 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
2501 | "ctx ticket reservation ran out. Need to up reservation"); | |
55b66332 | 2502 | xlog_print_tic_res(log->l_mp, ticket); |
7d2d5653 BF |
2503 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
2504 | } | |
1da177e4 | 2505 | |
3468bb1c DC |
2506 | len = xlog_write_calc_vec_length(ticket, log_vector, optype); |
2507 | if (start_lsn) | |
2508 | *start_lsn = 0; | |
fd63875c | 2509 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2510 | void *ptr; |
99428ad0 | 2511 | int log_offset; |
1da177e4 | 2512 | |
99428ad0 CH |
2513 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2514 | &contwr, &log_offset); | |
2515 | if (error) | |
2516 | return error; | |
1da177e4 | 2517 | |
99428ad0 | 2518 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2519 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2520 | |
eef983ff | 2521 | /* Start_lsn is the first lsn written to. */ |
3468bb1c | 2522 | if (start_lsn && !*start_lsn) |
99428ad0 | 2523 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
b5203cd0 | 2524 | |
99428ad0 CH |
2525 | /* |
2526 | * This loop writes out as many regions as can fit in the amount | |
2527 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2528 | */ | |
fd63875c DC |
2529 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2530 | struct xfs_log_iovec *reg; | |
99428ad0 | 2531 | struct xlog_op_header *ophdr; |
99428ad0 CH |
2532 | int copy_len; |
2533 | int copy_off; | |
fd63875c | 2534 | bool ordered = false; |
3468bb1c | 2535 | bool wrote_start_rec = false; |
fd63875c DC |
2536 | |
2537 | /* ordered log vectors have no regions to write */ | |
2538 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2539 | ASSERT(lv->lv_niovecs == 0); | |
2540 | ordered = true; | |
2541 | goto next_lv; | |
2542 | } | |
99428ad0 | 2543 | |
fd63875c | 2544 | reg = &vecp[index]; |
c8ce540d DW |
2545 | ASSERT(reg->i_len % sizeof(int32_t) == 0); |
2546 | ASSERT((unsigned long)ptr % sizeof(int32_t) == 0); | |
99428ad0 | 2547 | |
7ec94921 DC |
2548 | /* |
2549 | * Before we start formatting log vectors, we need to | |
2550 | * write a start record. Only do this for the first | |
2551 | * iclog we write to. | |
2552 | */ | |
3468bb1c | 2553 | if (optype & XLOG_START_TRANS) { |
7ec94921 | 2554 | xlog_write_start_rec(ptr, ticket); |
e6b1f273 | 2555 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
7ec94921 | 2556 | sizeof(struct xlog_op_header)); |
3468bb1c DC |
2557 | optype &= ~XLOG_START_TRANS; |
2558 | wrote_start_rec = true; | |
99428ad0 | 2559 | } |
b5203cd0 | 2560 | |
3468bb1c | 2561 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, optype); |
99428ad0 | 2562 | if (!ophdr) |
2451337d | 2563 | return -EIO; |
99428ad0 | 2564 | |
e6b1f273 | 2565 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2566 | sizeof(struct xlog_op_header)); |
2567 | ||
2568 | len += xlog_write_setup_copy(ticket, ophdr, | |
2569 | iclog->ic_size-log_offset, | |
55b66332 | 2570 | reg->i_len, |
99428ad0 CH |
2571 | ©_off, ©_len, |
2572 | &partial_copy, | |
2573 | &partial_copy_len); | |
2574 | xlog_verify_dest_ptr(log, ptr); | |
2575 | ||
91f9f5fe ES |
2576 | /* |
2577 | * Copy region. | |
2578 | * | |
2579 | * Unmount records just log an opheader, so can have | |
2580 | * empty payloads with no data region to copy. Hence we | |
2581 | * only copy the payload if the vector says it has data | |
2582 | * to copy. | |
2583 | */ | |
99428ad0 | 2584 | ASSERT(copy_len >= 0); |
91f9f5fe ES |
2585 | if (copy_len > 0) { |
2586 | memcpy(ptr, reg->i_addr + copy_off, copy_len); | |
2587 | xlog_write_adv_cnt(&ptr, &len, &log_offset, | |
2588 | copy_len); | |
2589 | } | |
7ec94921 | 2590 | copy_len += sizeof(struct xlog_op_header); |
99428ad0 | 2591 | record_cnt++; |
3468bb1c | 2592 | if (wrote_start_rec) { |
7ec94921 DC |
2593 | copy_len += sizeof(struct xlog_op_header); |
2594 | record_cnt++; | |
7ec94921 | 2595 | } |
99428ad0 CH |
2596 | data_cnt += contwr ? copy_len : 0; |
2597 | ||
3468bb1c | 2598 | error = xlog_write_copy_finish(log, iclog, optype, |
99428ad0 CH |
2599 | &record_cnt, &data_cnt, |
2600 | &partial_copy, | |
2601 | &partial_copy_len, | |
2602 | log_offset, | |
2603 | commit_iclog); | |
2604 | if (error) | |
2605 | return error; | |
2606 | ||
2607 | /* | |
2608 | * if we had a partial copy, we need to get more iclog | |
2609 | * space but we don't want to increment the region | |
2610 | * index because there is still more is this region to | |
2611 | * write. | |
2612 | * | |
2613 | * If we completed writing this region, and we flushed | |
2614 | * the iclog (indicated by resetting of the record | |
2615 | * count), then we also need to get more log space. If | |
2616 | * this was the last record, though, we are done and | |
2617 | * can just return. | |
2618 | */ | |
2619 | if (partial_copy) | |
2620 | break; | |
2621 | ||
55b66332 | 2622 | if (++index == lv->lv_niovecs) { |
fd63875c | 2623 | next_lv: |
55b66332 DC |
2624 | lv = lv->lv_next; |
2625 | index = 0; | |
2626 | if (lv) | |
2627 | vecp = lv->lv_iovecp; | |
2628 | } | |
749f24f3 | 2629 | if (record_cnt == 0 && !ordered) { |
55b66332 | 2630 | if (!lv) |
99428ad0 CH |
2631 | return 0; |
2632 | break; | |
2633 | } | |
2634 | } | |
2635 | } | |
2636 | ||
2637 | ASSERT(len == 0); | |
2638 | ||
390aab0a | 2639 | spin_lock(&log->l_icloglock); |
99428ad0 | 2640 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); |
df732b29 | 2641 | if (commit_iclog) { |
3468bb1c | 2642 | ASSERT(optype & XLOG_COMMIT_TRANS); |
df732b29 CH |
2643 | *commit_iclog = iclog; |
2644 | } else { | |
0dc8f7f1 | 2645 | error = xlog_state_release_iclog(log, iclog, 0); |
df732b29 | 2646 | } |
390aab0a | 2647 | spin_unlock(&log->l_icloglock); |
1da177e4 | 2648 | |
df732b29 | 2649 | return error; |
99428ad0 | 2650 | } |
1da177e4 | 2651 | |
c814b4f2 CH |
2652 | static void |
2653 | xlog_state_activate_iclog( | |
2654 | struct xlog_in_core *iclog, | |
2655 | int *iclogs_changed) | |
2656 | { | |
2657 | ASSERT(list_empty_careful(&iclog->ic_callbacks)); | |
956f6daa | 2658 | trace_xlog_iclog_activate(iclog, _RET_IP_); |
c814b4f2 CH |
2659 | |
2660 | /* | |
2661 | * If the number of ops in this iclog indicate it just contains the | |
2662 | * dummy transaction, we can change state into IDLE (the second time | |
2663 | * around). Otherwise we should change the state into NEED a dummy. | |
2664 | * We don't need to cover the dummy. | |
2665 | */ | |
2666 | if (*iclogs_changed == 0 && | |
2667 | iclog->ic_header.h_num_logops == cpu_to_be32(XLOG_COVER_OPS)) { | |
2668 | *iclogs_changed = 1; | |
2669 | } else { | |
2670 | /* | |
2671 | * We have two dirty iclogs so start over. This could also be | |
2672 | * num of ops indicating this is not the dummy going out. | |
2673 | */ | |
2674 | *iclogs_changed = 2; | |
2675 | } | |
2676 | ||
2677 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2678 | iclog->ic_offset = 0; | |
2679 | iclog->ic_header.h_num_logops = 0; | |
2680 | memset(iclog->ic_header.h_cycle_data, 0, | |
2681 | sizeof(iclog->ic_header.h_cycle_data)); | |
2682 | iclog->ic_header.h_lsn = 0; | |
9d110014 | 2683 | iclog->ic_header.h_tail_lsn = 0; |
c814b4f2 CH |
2684 | } |
2685 | ||
0383f543 | 2686 | /* |
c814b4f2 CH |
2687 | * Loop through all iclogs and mark all iclogs currently marked DIRTY as |
2688 | * ACTIVE after iclog I/O has completed. | |
1da177e4 | 2689 | */ |
c814b4f2 CH |
2690 | static void |
2691 | xlog_state_activate_iclogs( | |
0383f543 | 2692 | struct xlog *log, |
c814b4f2 | 2693 | int *iclogs_changed) |
1da177e4 | 2694 | { |
c814b4f2 | 2695 | struct xlog_in_core *iclog = log->l_iclog; |
1da177e4 | 2696 | |
1da177e4 | 2697 | do { |
c814b4f2 CH |
2698 | if (iclog->ic_state == XLOG_STATE_DIRTY) |
2699 | xlog_state_activate_iclog(iclog, iclogs_changed); | |
2700 | /* | |
2701 | * The ordering of marking iclogs ACTIVE must be maintained, so | |
2702 | * an iclog doesn't become ACTIVE beyond one that is SYNCING. | |
2703 | */ | |
2704 | else if (iclog->ic_state != XLOG_STATE_ACTIVE) | |
2705 | break; | |
2706 | } while ((iclog = iclog->ic_next) != log->l_iclog); | |
2707 | } | |
0383f543 | 2708 | |
c814b4f2 CH |
2709 | static int |
2710 | xlog_covered_state( | |
2711 | int prev_state, | |
2712 | int iclogs_changed) | |
2713 | { | |
0383f543 | 2714 | /* |
b0eb9e11 BF |
2715 | * We go to NEED for any non-covering writes. We go to NEED2 if we just |
2716 | * wrote the first covering record (DONE). We go to IDLE if we just | |
2717 | * wrote the second covering record (DONE2) and remain in IDLE until a | |
2718 | * non-covering write occurs. | |
0383f543 | 2719 | */ |
c814b4f2 CH |
2720 | switch (prev_state) { |
2721 | case XLOG_STATE_COVER_IDLE: | |
b0eb9e11 BF |
2722 | if (iclogs_changed == 1) |
2723 | return XLOG_STATE_COVER_IDLE; | |
53004ee7 | 2724 | fallthrough; |
c814b4f2 CH |
2725 | case XLOG_STATE_COVER_NEED: |
2726 | case XLOG_STATE_COVER_NEED2: | |
2727 | break; | |
2728 | case XLOG_STATE_COVER_DONE: | |
2729 | if (iclogs_changed == 1) | |
2730 | return XLOG_STATE_COVER_NEED2; | |
2731 | break; | |
2732 | case XLOG_STATE_COVER_DONE2: | |
2733 | if (iclogs_changed == 1) | |
2734 | return XLOG_STATE_COVER_IDLE; | |
2735 | break; | |
2736 | default: | |
2737 | ASSERT(0); | |
2738 | } | |
0383f543 | 2739 | |
c814b4f2 CH |
2740 | return XLOG_STATE_COVER_NEED; |
2741 | } | |
1da177e4 | 2742 | |
c814b4f2 CH |
2743 | STATIC void |
2744 | xlog_state_clean_iclog( | |
2745 | struct xlog *log, | |
2746 | struct xlog_in_core *dirty_iclog) | |
2747 | { | |
2748 | int iclogs_changed = 0; | |
1da177e4 | 2749 | |
956f6daa DC |
2750 | trace_xlog_iclog_clean(dirty_iclog, _RET_IP_); |
2751 | ||
5781464b | 2752 | dirty_iclog->ic_state = XLOG_STATE_DIRTY; |
1da177e4 | 2753 | |
c814b4f2 CH |
2754 | xlog_state_activate_iclogs(log, &iclogs_changed); |
2755 | wake_up_all(&dirty_iclog->ic_force_wait); | |
2756 | ||
2757 | if (iclogs_changed) { | |
2758 | log->l_covered_state = xlog_covered_state(log->l_covered_state, | |
2759 | iclogs_changed); | |
1da177e4 | 2760 | } |
0383f543 | 2761 | } |
1da177e4 LT |
2762 | |
2763 | STATIC xfs_lsn_t | |
2764 | xlog_get_lowest_lsn( | |
9bff3132 | 2765 | struct xlog *log) |
1da177e4 | 2766 | { |
9bff3132 CH |
2767 | struct xlog_in_core *iclog = log->l_iclog; |
2768 | xfs_lsn_t lowest_lsn = 0, lsn; | |
1da177e4 | 2769 | |
1da177e4 | 2770 | do { |
1858bb0b CH |
2771 | if (iclog->ic_state == XLOG_STATE_ACTIVE || |
2772 | iclog->ic_state == XLOG_STATE_DIRTY) | |
9bff3132 CH |
2773 | continue; |
2774 | ||
2775 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2776 | if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0) | |
1da177e4 | 2777 | lowest_lsn = lsn; |
9bff3132 CH |
2778 | } while ((iclog = iclog->ic_next) != log->l_iclog); |
2779 | ||
014c2544 | 2780 | return lowest_lsn; |
1da177e4 LT |
2781 | } |
2782 | ||
14e15f1b DC |
2783 | /* |
2784 | * Completion of a iclog IO does not imply that a transaction has completed, as | |
2785 | * transactions can be large enough to span many iclogs. We cannot change the | |
2786 | * tail of the log half way through a transaction as this may be the only | |
2787 | * transaction in the log and moving the tail to point to the middle of it | |
2788 | * will prevent recovery from finding the start of the transaction. Hence we | |
2789 | * should only update the last_sync_lsn if this iclog contains transaction | |
2790 | * completion callbacks on it. | |
2791 | * | |
2792 | * We have to do this before we drop the icloglock to ensure we are the only one | |
2793 | * that can update it. | |
2794 | * | |
2795 | * If we are moving the last_sync_lsn forwards, we also need to ensure we kick | |
2796 | * the reservation grant head pushing. This is due to the fact that the push | |
2797 | * target is bound by the current last_sync_lsn value. Hence if we have a large | |
2798 | * amount of log space bound up in this committing transaction then the | |
2799 | * last_sync_lsn value may be the limiting factor preventing tail pushing from | |
2800 | * freeing space in the log. Hence once we've updated the last_sync_lsn we | |
2801 | * should push the AIL to ensure the push target (and hence the grant head) is | |
2802 | * no longer bound by the old log head location and can move forwards and make | |
2803 | * progress again. | |
2804 | */ | |
2805 | static void | |
2806 | xlog_state_set_callback( | |
2807 | struct xlog *log, | |
2808 | struct xlog_in_core *iclog, | |
2809 | xfs_lsn_t header_lsn) | |
2810 | { | |
956f6daa | 2811 | trace_xlog_iclog_callback(iclog, _RET_IP_); |
14e15f1b DC |
2812 | iclog->ic_state = XLOG_STATE_CALLBACK; |
2813 | ||
2814 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), | |
2815 | header_lsn) <= 0); | |
2816 | ||
2817 | if (list_empty_careful(&iclog->ic_callbacks)) | |
2818 | return; | |
2819 | ||
2820 | atomic64_set(&log->l_last_sync_lsn, header_lsn); | |
2821 | xlog_grant_push_ail(log, 0); | |
2822 | } | |
2823 | ||
5e96fa8d DC |
2824 | /* |
2825 | * Return true if we need to stop processing, false to continue to the next | |
2826 | * iclog. The caller will need to run callbacks if the iclog is returned in the | |
2827 | * XLOG_STATE_CALLBACK state. | |
2828 | */ | |
2829 | static bool | |
2830 | xlog_state_iodone_process_iclog( | |
2831 | struct xlog *log, | |
5112e206 | 2832 | struct xlog_in_core *iclog) |
5e96fa8d DC |
2833 | { |
2834 | xfs_lsn_t lowest_lsn; | |
14e15f1b | 2835 | xfs_lsn_t header_lsn; |
5e96fa8d | 2836 | |
1858bb0b CH |
2837 | switch (iclog->ic_state) { |
2838 | case XLOG_STATE_ACTIVE: | |
2839 | case XLOG_STATE_DIRTY: | |
2840 | /* | |
2841 | * Skip all iclogs in the ACTIVE & DIRTY states: | |
2842 | */ | |
5e96fa8d | 2843 | return false; |
1858bb0b | 2844 | case XLOG_STATE_DONE_SYNC: |
1858bb0b | 2845 | /* |
4b29ab04 CH |
2846 | * Now that we have an iclog that is in the DONE_SYNC state, do |
2847 | * one more check here to see if we have chased our tail around. | |
2848 | * If this is not the lowest lsn iclog, then we will leave it | |
2849 | * for another completion to process. | |
1858bb0b CH |
2850 | */ |
2851 | header_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2852 | lowest_lsn = xlog_get_lowest_lsn(log); | |
2853 | if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, header_lsn) < 0) | |
2854 | return false; | |
2855 | xlog_state_set_callback(log, iclog, header_lsn); | |
2856 | return false; | |
2857 | default: | |
2858 | /* | |
2859 | * Can only perform callbacks in order. Since this iclog is not | |
4b29ab04 CH |
2860 | * in the DONE_SYNC state, we skip the rest and just try to |
2861 | * clean up. | |
1858bb0b | 2862 | */ |
5e96fa8d DC |
2863 | return true; |
2864 | } | |
5e96fa8d DC |
2865 | } |
2866 | ||
8bb92005 DC |
2867 | /* |
2868 | * Loop over all the iclogs, running attached callbacks on them. Return true if | |
aad7272a DC |
2869 | * we ran any callbacks, indicating that we dropped the icloglock. We don't need |
2870 | * to handle transient shutdown state here at all because | |
2871 | * xlog_state_shutdown_callbacks() will be run to do the necessary shutdown | |
2872 | * cleanup of the callbacks. | |
8bb92005 DC |
2873 | */ |
2874 | static bool | |
2875 | xlog_state_do_iclog_callbacks( | |
2876 | struct xlog *log) | |
2877 | __releases(&log->l_icloglock) | |
2878 | __acquires(&log->l_icloglock) | |
2879 | { | |
2880 | struct xlog_in_core *first_iclog = log->l_iclog; | |
2881 | struct xlog_in_core *iclog = first_iclog; | |
2882 | bool ran_callback = false; | |
2883 | ||
2884 | do { | |
2885 | LIST_HEAD(cb_list); | |
2886 | ||
aad7272a DC |
2887 | if (xlog_state_iodone_process_iclog(log, iclog)) |
2888 | break; | |
2889 | if (iclog->ic_state != XLOG_STATE_CALLBACK) { | |
2890 | iclog = iclog->ic_next; | |
2891 | continue; | |
8bb92005 DC |
2892 | } |
2893 | list_splice_init(&iclog->ic_callbacks, &cb_list); | |
2894 | spin_unlock(&log->l_icloglock); | |
2895 | ||
2896 | trace_xlog_iclog_callbacks_start(iclog, _RET_IP_); | |
2897 | xlog_cil_process_committed(&cb_list); | |
2898 | trace_xlog_iclog_callbacks_done(iclog, _RET_IP_); | |
2899 | ran_callback = true; | |
2900 | ||
2901 | spin_lock(&log->l_icloglock); | |
aad7272a | 2902 | xlog_state_clean_iclog(log, iclog); |
8bb92005 DC |
2903 | iclog = iclog->ic_next; |
2904 | } while (iclog != first_iclog); | |
2905 | ||
2906 | return ran_callback; | |
2907 | } | |
2908 | ||
2909 | ||
2910 | /* | |
2911 | * Loop running iclog completion callbacks until there are no more iclogs in a | |
2912 | * state that can run callbacks. | |
2913 | */ | |
1da177e4 LT |
2914 | STATIC void |
2915 | xlog_state_do_callback( | |
12e6a0f4 | 2916 | struct xlog *log) |
1da177e4 | 2917 | { |
5e96fa8d DC |
2918 | int flushcnt = 0; |
2919 | int repeats = 0; | |
1da177e4 | 2920 | |
b22cd72c | 2921 | spin_lock(&log->l_icloglock); |
8bb92005 DC |
2922 | while (xlog_state_do_iclog_callbacks(log)) { |
2923 | if (xlog_is_shutdown(log)) | |
2924 | break; | |
a3c6685e | 2925 | |
5112e206 | 2926 | if (++repeats > 5000) { |
a3c6685e NS |
2927 | flushcnt += repeats; |
2928 | repeats = 0; | |
a0fa2b67 | 2929 | xfs_warn(log->l_mp, |
a3c6685e | 2930 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2931 | __func__, flushcnt); |
1da177e4 | 2932 | } |
8bb92005 | 2933 | } |
1da177e4 | 2934 | |
aad7272a | 2935 | if (log->l_iclog->ic_state == XLOG_STATE_ACTIVE) |
eb40a875 | 2936 | wake_up_all(&log->l_flush_wait); |
cdea5459 RR |
2937 | |
2938 | spin_unlock(&log->l_icloglock); | |
d748c623 | 2939 | } |
1da177e4 LT |
2940 | |
2941 | ||
2942 | /* | |
2943 | * Finish transitioning this iclog to the dirty state. | |
2944 | * | |
1da177e4 | 2945 | * Callbacks could take time, so they are done outside the scope of the |
12017faf | 2946 | * global state machine log lock. |
1da177e4 | 2947 | */ |
a8272ce0 | 2948 | STATIC void |
1da177e4 | 2949 | xlog_state_done_syncing( |
12e6a0f4 | 2950 | struct xlog_in_core *iclog) |
1da177e4 | 2951 | { |
d15cbf2f | 2952 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2953 | |
b22cd72c | 2954 | spin_lock(&log->l_icloglock); |
155cc6b7 | 2955 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
956f6daa | 2956 | trace_xlog_iclog_sync_done(iclog, _RET_IP_); |
1da177e4 LT |
2957 | |
2958 | /* | |
2959 | * If we got an error, either on the first buffer, or in the case of | |
12e6a0f4 CH |
2960 | * split log writes, on the second, we shut down the file system and |
2961 | * no iclogs should ever be attempted to be written to disk again. | |
1da177e4 | 2962 | */ |
2039a272 | 2963 | if (!xlog_is_shutdown(log)) { |
12e6a0f4 | 2964 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING); |
1da177e4 | 2965 | iclog->ic_state = XLOG_STATE_DONE_SYNC; |
12e6a0f4 | 2966 | } |
1da177e4 LT |
2967 | |
2968 | /* | |
2969 | * Someone could be sleeping prior to writing out the next | |
2970 | * iclog buffer, we wake them all, one will get to do the | |
2971 | * I/O, the others get to wait for the result. | |
2972 | */ | |
eb40a875 | 2973 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2974 | spin_unlock(&log->l_icloglock); |
b843299b | 2975 | xlog_state_do_callback(log); |
12e6a0f4 | 2976 | } |
1da177e4 LT |
2977 | |
2978 | /* | |
2979 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2980 | * sleep. We wait on the flush queue on the head iclog as that should be |
2981 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2982 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2983 | * |
2984 | * The in-core logs are used in a circular fashion. They are not used | |
2985 | * out-of-order even when an iclog past the head is free. | |
2986 | * | |
2987 | * return: | |
2988 | * * log_offset where xlog_write() can start writing into the in-core | |
2989 | * log's data space. | |
2990 | * * in-core log pointer to which xlog_write() should write. | |
2991 | * * boolean indicating this is a continued write to an in-core log. | |
2992 | * If this is the last write, then the in-core log's offset field | |
2993 | * needs to be incremented, depending on the amount of data which | |
2994 | * is copied. | |
2995 | */ | |
a8272ce0 | 2996 | STATIC int |
9a8d2fdb MT |
2997 | xlog_state_get_iclog_space( |
2998 | struct xlog *log, | |
2999 | int len, | |
3000 | struct xlog_in_core **iclogp, | |
3001 | struct xlog_ticket *ticket, | |
3002 | int *continued_write, | |
3003 | int *logoffsetp) | |
1da177e4 | 3004 | { |
1da177e4 LT |
3005 | int log_offset; |
3006 | xlog_rec_header_t *head; | |
3007 | xlog_in_core_t *iclog; | |
1da177e4 LT |
3008 | |
3009 | restart: | |
b22cd72c | 3010 | spin_lock(&log->l_icloglock); |
2039a272 | 3011 | if (xlog_is_shutdown(log)) { |
b22cd72c | 3012 | spin_unlock(&log->l_icloglock); |
2451337d | 3013 | return -EIO; |
1da177e4 LT |
3014 | } |
3015 | ||
3016 | iclog = log->l_iclog; | |
d748c623 | 3017 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
ff6d6af2 | 3018 | XFS_STATS_INC(log->l_mp, xs_log_noiclogs); |
d748c623 MW |
3019 | |
3020 | /* Wait for log writes to have flushed */ | |
eb40a875 | 3021 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
3022 | goto restart; |
3023 | } | |
d748c623 | 3024 | |
1da177e4 LT |
3025 | head = &iclog->ic_header; |
3026 | ||
155cc6b7 | 3027 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
3028 | log_offset = iclog->ic_offset; |
3029 | ||
956f6daa DC |
3030 | trace_xlog_iclog_get_space(iclog, _RET_IP_); |
3031 | ||
1da177e4 LT |
3032 | /* On the 1st write to an iclog, figure out lsn. This works |
3033 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
3034 | * committing to. If the offset is set, that's how many blocks | |
3035 | * must be written. | |
3036 | */ | |
3037 | if (log_offset == 0) { | |
3038 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 3039 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
3040 | log->l_iclog_hsize, |
3041 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
3042 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
3043 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 3044 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
3045 | ASSERT(log->l_curr_block >= 0); |
3046 | } | |
3047 | ||
3048 | /* If there is enough room to write everything, then do it. Otherwise, | |
3049 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
3050 | * bit is on, so this will get flushed out. Don't update ic_offset | |
3051 | * until you know exactly how many bytes get copied. Therefore, wait | |
3052 | * until later to update ic_offset. | |
3053 | * | |
3054 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
3055 | * can fit into remaining data section. | |
3056 | */ | |
3057 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
df732b29 CH |
3058 | int error = 0; |
3059 | ||
1da177e4 LT |
3060 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); |
3061 | ||
49641f1a | 3062 | /* |
df732b29 CH |
3063 | * If we are the only one writing to this iclog, sync it to |
3064 | * disk. We need to do an atomic compare and decrement here to | |
3065 | * avoid racing with concurrent atomic_dec_and_lock() calls in | |
49641f1a DC |
3066 | * xlog_state_release_iclog() when there is more than one |
3067 | * reference to the iclog. | |
3068 | */ | |
df732b29 | 3069 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) |
0dc8f7f1 | 3070 | error = xlog_state_release_iclog(log, iclog, 0); |
df732b29 CH |
3071 | spin_unlock(&log->l_icloglock); |
3072 | if (error) | |
3073 | return error; | |
1da177e4 LT |
3074 | goto restart; |
3075 | } | |
3076 | ||
3077 | /* Do we have enough room to write the full amount in the remainder | |
3078 | * of this iclog? Or must we continue a write on the next iclog and | |
3079 | * mark this iclog as completely taken? In the case where we switch | |
3080 | * iclogs (to mark it taken), this particular iclog will release/sync | |
3081 | * to disk in xlog_write(). | |
3082 | */ | |
3083 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
3084 | *continued_write = 0; | |
3085 | iclog->ic_offset += len; | |
3086 | } else { | |
3087 | *continued_write = 1; | |
3088 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
3089 | } | |
3090 | *iclogp = iclog; | |
3091 | ||
3092 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 3093 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3094 | |
3095 | *logoffsetp = log_offset; | |
3096 | return 0; | |
b843299b | 3097 | } |
1da177e4 | 3098 | |
8b41e3f9 | 3099 | /* |
b843299b DC |
3100 | * The first cnt-1 times a ticket goes through here we don't need to move the |
3101 | * grant write head because the permanent reservation has reserved cnt times the | |
3102 | * unit amount. Release part of current permanent unit reservation and reset | |
3103 | * current reservation to be one units worth. Also move grant reservation head | |
3104 | * forward. | |
1da177e4 | 3105 | */ |
8b41e3f9 CH |
3106 | void |
3107 | xfs_log_ticket_regrant( | |
9a8d2fdb MT |
3108 | struct xlog *log, |
3109 | struct xlog_ticket *ticket) | |
1da177e4 | 3110 | { |
8b41e3f9 | 3111 | trace_xfs_log_ticket_regrant(log, ticket); |
0b1b213f | 3112 | |
1da177e4 LT |
3113 | if (ticket->t_cnt > 0) |
3114 | ticket->t_cnt--; | |
3115 | ||
28496968 | 3116 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3117 | ticket->t_curr_res); |
28496968 | 3118 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 3119 | ticket->t_curr_res); |
1da177e4 | 3120 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3121 | xlog_tic_reset_res(ticket); |
0b1b213f | 3122 | |
8b41e3f9 | 3123 | trace_xfs_log_ticket_regrant_sub(log, ticket); |
0b1b213f | 3124 | |
1da177e4 | 3125 | /* just return if we still have some of the pre-reserved space */ |
8b41e3f9 CH |
3126 | if (!ticket->t_cnt) { |
3127 | xlog_grant_add_space(log, &log->l_reserve_head.grant, | |
3128 | ticket->t_unit_res); | |
3129 | trace_xfs_log_ticket_regrant_exit(log, ticket); | |
1da177e4 | 3130 | |
8b41e3f9 CH |
3131 | ticket->t_curr_res = ticket->t_unit_res; |
3132 | xlog_tic_reset_res(ticket); | |
3133 | } | |
1da177e4 | 3134 | |
8b41e3f9 CH |
3135 | xfs_log_ticket_put(ticket); |
3136 | } | |
1da177e4 LT |
3137 | |
3138 | /* | |
3139 | * Give back the space left from a reservation. | |
3140 | * | |
3141 | * All the information we need to make a correct determination of space left | |
3142 | * is present. For non-permanent reservations, things are quite easy. The | |
3143 | * count should have been decremented to zero. We only need to deal with the | |
3144 | * space remaining in the current reservation part of the ticket. If the | |
3145 | * ticket contains a permanent reservation, there may be left over space which | |
3146 | * needs to be released. A count of N means that N-1 refills of the current | |
3147 | * reservation can be done before we need to ask for more space. The first | |
3148 | * one goes to fill up the first current reservation. Once we run out of | |
3149 | * space, the count will stay at zero and the only space remaining will be | |
3150 | * in the current reservation field. | |
3151 | */ | |
8b41e3f9 CH |
3152 | void |
3153 | xfs_log_ticket_ungrant( | |
9a8d2fdb MT |
3154 | struct xlog *log, |
3155 | struct xlog_ticket *ticket) | |
1da177e4 | 3156 | { |
8b41e3f9 CH |
3157 | int bytes; |
3158 | ||
3159 | trace_xfs_log_ticket_ungrant(log, ticket); | |
663e496a | 3160 | |
1da177e4 LT |
3161 | if (ticket->t_cnt > 0) |
3162 | ticket->t_cnt--; | |
3163 | ||
8b41e3f9 | 3164 | trace_xfs_log_ticket_ungrant_sub(log, ticket); |
1da177e4 | 3165 | |
663e496a DC |
3166 | /* |
3167 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
3168 | * up more space based on the remaining count. |
3169 | */ | |
663e496a | 3170 | bytes = ticket->t_curr_res; |
1da177e4 LT |
3171 | if (ticket->t_cnt > 0) { |
3172 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 3173 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
3174 | } |
3175 | ||
28496968 CH |
3176 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
3177 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 3178 | |
8b41e3f9 | 3179 | trace_xfs_log_ticket_ungrant_exit(log, ticket); |
0b1b213f | 3180 | |
cfb7cdca | 3181 | xfs_log_space_wake(log->l_mp); |
8b41e3f9 | 3182 | xfs_log_ticket_put(ticket); |
09a423a3 | 3183 | } |
1da177e4 | 3184 | |
1da177e4 | 3185 | /* |
b843299b DC |
3186 | * This routine will mark the current iclog in the ring as WANT_SYNC and move |
3187 | * the current iclog pointer to the next iclog in the ring. | |
1da177e4 LT |
3188 | */ |
3189 | STATIC void | |
9a8d2fdb MT |
3190 | xlog_state_switch_iclogs( |
3191 | struct xlog *log, | |
3192 | struct xlog_in_core *iclog, | |
3193 | int eventual_size) | |
1da177e4 LT |
3194 | { |
3195 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
69363999 | 3196 | assert_spin_locked(&log->l_icloglock); |
956f6daa | 3197 | trace_xlog_iclog_switch(iclog, _RET_IP_); |
69363999 | 3198 | |
1da177e4 LT |
3199 | if (!eventual_size) |
3200 | eventual_size = iclog->ic_offset; | |
3201 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3202 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3203 | log->l_prev_block = log->l_curr_block; |
3204 | log->l_prev_cycle = log->l_curr_cycle; | |
3205 | ||
3206 | /* roll log?: ic_offset changed later */ | |
3207 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3208 | ||
3209 | /* Round up to next log-sunit */ | |
a6a65fef | 3210 | if (log->l_iclog_roundoff > BBSIZE) { |
18842e0a | 3211 | uint32_t sunit_bb = BTOBB(log->l_iclog_roundoff); |
1da177e4 LT |
3212 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); |
3213 | } | |
3214 | ||
3215 | if (log->l_curr_block >= log->l_logBBsize) { | |
a45086e2 BF |
3216 | /* |
3217 | * Rewind the current block before the cycle is bumped to make | |
3218 | * sure that the combined LSN never transiently moves forward | |
3219 | * when the log wraps to the next cycle. This is to support the | |
3220 | * unlocked sample of these fields from xlog_valid_lsn(). Most | |
3221 | * other cases should acquire l_icloglock. | |
3222 | */ | |
3223 | log->l_curr_block -= log->l_logBBsize; | |
3224 | ASSERT(log->l_curr_block >= 0); | |
3225 | smp_wmb(); | |
1da177e4 LT |
3226 | log->l_curr_cycle++; |
3227 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3228 | log->l_curr_cycle++; | |
1da177e4 LT |
3229 | } |
3230 | ASSERT(iclog == log->l_iclog); | |
3231 | log->l_iclog = iclog->ic_next; | |
b843299b | 3232 | } |
1da177e4 | 3233 | |
8191d822 DC |
3234 | /* |
3235 | * Force the iclog to disk and check if the iclog has been completed before | |
3236 | * xlog_force_iclog() returns. This can happen on synchronous (e.g. | |
3237 | * pmem) or fast async storage because we drop the icloglock to issue the IO. | |
3238 | * If completion has already occurred, tell the caller so that it can avoid an | |
3239 | * unnecessary wait on the iclog. | |
3240 | */ | |
3241 | static int | |
3242 | xlog_force_and_check_iclog( | |
3243 | struct xlog_in_core *iclog, | |
3244 | bool *completed) | |
3245 | { | |
3246 | xfs_lsn_t lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
3247 | int error; | |
3248 | ||
3249 | *completed = false; | |
3250 | error = xlog_force_iclog(iclog); | |
3251 | if (error) | |
3252 | return error; | |
3253 | ||
3254 | /* | |
3255 | * If the iclog has already been completed and reused the header LSN | |
3256 | * will have been rewritten by completion | |
3257 | */ | |
3258 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) | |
3259 | *completed = true; | |
3260 | return 0; | |
3261 | } | |
3262 | ||
1da177e4 LT |
3263 | /* |
3264 | * Write out all data in the in-core log as of this exact moment in time. | |
3265 | * | |
3266 | * Data may be written to the in-core log during this call. However, | |
3267 | * we don't guarantee this data will be written out. A change from past | |
3268 | * implementation means this routine will *not* write out zero length LRs. | |
3269 | * | |
3270 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3271 | * If we determine there is no flushable data, we just return. There is no | |
3272 | * flushable data if: | |
3273 | * | |
3274 | * 1. the current iclog is active and has no data; the previous iclog | |
3275 | * is in the active or dirty state. | |
3276 | * 2. the current iclog is drity, and the previous iclog is in the | |
3277 | * active or dirty state. | |
3278 | * | |
12017faf | 3279 | * We may sleep if: |
1da177e4 LT |
3280 | * |
3281 | * 1. the current iclog is not in the active nor dirty state. | |
3282 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3283 | * active nor dirty state. | |
3284 | * 3. the current iclog is active, and there is another thread writing | |
3285 | * to this particular iclog. | |
3286 | * 4. a) the current iclog is active and has no other writers | |
3287 | * b) when we return from flushing out this iclog, it is still | |
3288 | * not in the active nor dirty state. | |
3289 | */ | |
a14a348b | 3290 | int |
60e5bb78 | 3291 | xfs_log_force( |
a14a348b | 3292 | struct xfs_mount *mp, |
60e5bb78 | 3293 | uint flags) |
1da177e4 | 3294 | { |
ad223e60 | 3295 | struct xlog *log = mp->m_log; |
a14a348b | 3296 | struct xlog_in_core *iclog; |
a14a348b | 3297 | |
ff6d6af2 | 3298 | XFS_STATS_INC(mp, xs_log_force); |
60e5bb78 | 3299 | trace_xfs_log_force(mp, 0, _RET_IP_); |
1da177e4 | 3300 | |
93b8a585 | 3301 | xlog_cil_force(log); |
71e330b5 | 3302 | |
b22cd72c | 3303 | spin_lock(&log->l_icloglock); |
5112e206 | 3304 | if (xlog_is_shutdown(log)) |
e6b96570 | 3305 | goto out_error; |
1da177e4 | 3306 | |
5112e206 | 3307 | iclog = log->l_iclog; |
956f6daa DC |
3308 | trace_xlog_iclog_force(iclog, _RET_IP_); |
3309 | ||
e6b96570 CH |
3310 | if (iclog->ic_state == XLOG_STATE_DIRTY || |
3311 | (iclog->ic_state == XLOG_STATE_ACTIVE && | |
3312 | atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) { | |
1da177e4 | 3313 | /* |
e6b96570 CH |
3314 | * If the head is dirty or (active and empty), then we need to |
3315 | * look at the previous iclog. | |
3316 | * | |
3317 | * If the previous iclog is active or dirty we are done. There | |
3318 | * is nothing to sync out. Otherwise, we attach ourselves to the | |
1da177e4 LT |
3319 | * previous iclog and go to sleep. |
3320 | */ | |
e6b96570 | 3321 | iclog = iclog->ic_prev; |
e6b96570 CH |
3322 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3323 | if (atomic_read(&iclog->ic_refcnt) == 0) { | |
45eddb41 | 3324 | /* We have exclusive access to this iclog. */ |
8191d822 DC |
3325 | bool completed; |
3326 | ||
3327 | if (xlog_force_and_check_iclog(iclog, &completed)) | |
df732b29 | 3328 | goto out_error; |
1da177e4 | 3329 | |
8191d822 | 3330 | if (completed) |
e6b96570 CH |
3331 | goto out_unlock; |
3332 | } else { | |
3333 | /* | |
2bf1ec0f DC |
3334 | * Someone else is still writing to this iclog, so we |
3335 | * need to ensure that when they release the iclog it | |
3336 | * gets synced immediately as we may be waiting on it. | |
e6b96570 CH |
3337 | */ |
3338 | xlog_state_switch_iclogs(log, iclog, 0); | |
1da177e4 | 3339 | } |
1da177e4 | 3340 | } |
e6b96570 | 3341 | |
2bf1ec0f DC |
3342 | /* |
3343 | * The iclog we are about to wait on may contain the checkpoint pushed | |
3344 | * by the above xlog_cil_force() call, but it may not have been pushed | |
3345 | * to disk yet. Like the ACTIVE case above, we need to make sure caches | |
3346 | * are flushed when this iclog is written. | |
3347 | */ | |
3348 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) | |
3349 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA; | |
3350 | ||
81e5b50a CH |
3351 | if (flags & XFS_LOG_SYNC) |
3352 | return xlog_wait_on_iclog(iclog); | |
e6b96570 CH |
3353 | out_unlock: |
3354 | spin_unlock(&log->l_icloglock); | |
3355 | return 0; | |
3356 | out_error: | |
3357 | spin_unlock(&log->l_icloglock); | |
3358 | return -EIO; | |
a14a348b | 3359 | } |
1da177e4 | 3360 | |
3e4da466 | 3361 | static int |
5f9b4b0d DC |
3362 | xlog_force_lsn( |
3363 | struct xlog *log, | |
a14a348b CH |
3364 | xfs_lsn_t lsn, |
3365 | uint flags, | |
3e4da466 CH |
3366 | int *log_flushed, |
3367 | bool already_slept) | |
1da177e4 | 3368 | { |
a14a348b | 3369 | struct xlog_in_core *iclog; |
8191d822 | 3370 | bool completed; |
71e330b5 | 3371 | |
a14a348b | 3372 | spin_lock(&log->l_icloglock); |
5112e206 | 3373 | if (xlog_is_shutdown(log)) |
93806299 | 3374 | goto out_error; |
1da177e4 | 3375 | |
5112e206 | 3376 | iclog = log->l_iclog; |
93806299 | 3377 | while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { |
956f6daa | 3378 | trace_xlog_iclog_force_lsn(iclog, _RET_IP_); |
93806299 CH |
3379 | iclog = iclog->ic_next; |
3380 | if (iclog == log->l_iclog) | |
3381 | goto out_unlock; | |
3382 | } | |
a14a348b | 3383 | |
2bf1ec0f DC |
3384 | switch (iclog->ic_state) { |
3385 | case XLOG_STATE_ACTIVE: | |
93806299 CH |
3386 | /* |
3387 | * We sleep here if we haven't already slept (e.g. this is the | |
3388 | * first time we've looked at the correct iclog buf) and the | |
3389 | * buffer before us is going to be sync'ed. The reason for this | |
3390 | * is that if we are doing sync transactions here, by waiting | |
3391 | * for the previous I/O to complete, we can allow a few more | |
3392 | * transactions into this iclog before we close it down. | |
3393 | * | |
3394 | * Otherwise, we mark the buffer WANT_SYNC, and bump up the | |
3395 | * refcnt so we can release the log (which drops the ref count). | |
3396 | * The state switch keeps new transaction commits from using | |
3397 | * this buffer. When the current commits finish writing into | |
3398 | * the buffer, the refcount will drop to zero and the buffer | |
3399 | * will go out then. | |
3400 | */ | |
3401 | if (!already_slept && | |
1858bb0b CH |
3402 | (iclog->ic_prev->ic_state == XLOG_STATE_WANT_SYNC || |
3403 | iclog->ic_prev->ic_state == XLOG_STATE_SYNCING)) { | |
93806299 CH |
3404 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3405 | &log->l_icloglock); | |
3e4da466 | 3406 | return -EAGAIN; |
1da177e4 | 3407 | } |
8191d822 | 3408 | if (xlog_force_and_check_iclog(iclog, &completed)) |
df732b29 | 3409 | goto out_error; |
93806299 CH |
3410 | if (log_flushed) |
3411 | *log_flushed = 1; | |
8191d822 DC |
3412 | if (completed) |
3413 | goto out_unlock; | |
2bf1ec0f DC |
3414 | break; |
3415 | case XLOG_STATE_WANT_SYNC: | |
3416 | /* | |
3417 | * This iclog may contain the checkpoint pushed by the | |
3418 | * xlog_cil_force_seq() call, but there are other writers still | |
3419 | * accessing it so it hasn't been pushed to disk yet. Like the | |
3420 | * ACTIVE case above, we need to make sure caches are flushed | |
3421 | * when this iclog is written. | |
3422 | */ | |
3423 | iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA; | |
3424 | break; | |
3425 | default: | |
3426 | /* | |
3427 | * The entire checkpoint was written by the CIL force and is on | |
3428 | * its way to disk already. It will be stable when it | |
3429 | * completes, so we don't need to manipulate caches here at all. | |
3430 | * We just need to wait for completion if necessary. | |
3431 | */ | |
3432 | break; | |
93806299 | 3433 | } |
1da177e4 | 3434 | |
81e5b50a CH |
3435 | if (flags & XFS_LOG_SYNC) |
3436 | return xlog_wait_on_iclog(iclog); | |
93806299 | 3437 | out_unlock: |
a14a348b CH |
3438 | spin_unlock(&log->l_icloglock); |
3439 | return 0; | |
93806299 CH |
3440 | out_error: |
3441 | spin_unlock(&log->l_icloglock); | |
3442 | return -EIO; | |
a14a348b CH |
3443 | } |
3444 | ||
3e4da466 CH |
3445 | /* |
3446 | * Force the in-core log to disk for a specific LSN. | |
3447 | * | |
3448 | * Find in-core log with lsn. | |
3449 | * If it is in the DIRTY state, just return. | |
3450 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3451 | * state and go to sleep or return. | |
3452 | * If it is in any other state, go to sleep or return. | |
3453 | * | |
3454 | * Synchronous forces are implemented with a wait queue. All callers trying | |
3455 | * to force a given lsn to disk must wait on the queue attached to the | |
3456 | * specific in-core log. When given in-core log finally completes its write | |
3457 | * to disk, that thread will wake up all threads waiting on the queue. | |
3458 | */ | |
3459 | int | |
5f9b4b0d | 3460 | xfs_log_force_seq( |
3e4da466 | 3461 | struct xfs_mount *mp, |
5f9b4b0d | 3462 | xfs_csn_t seq, |
3e4da466 CH |
3463 | uint flags, |
3464 | int *log_flushed) | |
3465 | { | |
5f9b4b0d DC |
3466 | struct xlog *log = mp->m_log; |
3467 | xfs_lsn_t lsn; | |
3e4da466 | 3468 | int ret; |
5f9b4b0d | 3469 | ASSERT(seq != 0); |
3e4da466 CH |
3470 | |
3471 | XFS_STATS_INC(mp, xs_log_force); | |
5f9b4b0d | 3472 | trace_xfs_log_force(mp, seq, _RET_IP_); |
3e4da466 | 3473 | |
5f9b4b0d | 3474 | lsn = xlog_cil_force_seq(log, seq); |
3e4da466 CH |
3475 | if (lsn == NULLCOMMITLSN) |
3476 | return 0; | |
3477 | ||
5f9b4b0d DC |
3478 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, false); |
3479 | if (ret == -EAGAIN) { | |
3480 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3481 | ret = xlog_force_lsn(log, lsn, flags, log_flushed, true); | |
3482 | } | |
3e4da466 CH |
3483 | return ret; |
3484 | } | |
3485 | ||
1da177e4 | 3486 | /* |
9da096fd | 3487 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3488 | */ |
cc09c0dc DC |
3489 | void |
3490 | xfs_log_ticket_put( | |
3491 | xlog_ticket_t *ticket) | |
1da177e4 | 3492 | { |
cc09c0dc | 3493 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3494 | if (atomic_dec_and_test(&ticket->t_ref)) |
377bcd5f | 3495 | kmem_cache_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3496 | } |
1da177e4 | 3497 | |
cc09c0dc DC |
3498 | xlog_ticket_t * |
3499 | xfs_log_ticket_get( | |
3500 | xlog_ticket_t *ticket) | |
3501 | { | |
3502 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3503 | atomic_inc(&ticket->t_ref); | |
3504 | return ticket; | |
3505 | } | |
1da177e4 LT |
3506 | |
3507 | /* | |
e773fc93 JL |
3508 | * Figure out the total log space unit (in bytes) that would be |
3509 | * required for a log ticket. | |
1da177e4 | 3510 | */ |
a6a65fef DC |
3511 | static int |
3512 | xlog_calc_unit_res( | |
3513 | struct xlog *log, | |
e773fc93 | 3514 | int unit_bytes) |
1da177e4 | 3515 | { |
e773fc93 JL |
3516 | int iclog_space; |
3517 | uint num_headers; | |
1da177e4 LT |
3518 | |
3519 | /* | |
3520 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3521 | * in the log. A unit in this case is the amount of space for one | |
3522 | * of these log operations. Normal reservations have a cnt of 1 | |
3523 | * and their unit amount is the total amount of space required. | |
3524 | * | |
3525 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3526 | * which occupy space in the on-disk log. |
3527 | * | |
3528 | * Normal form of a transaction is: | |
3529 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3530 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3531 | * | |
3532 | * We need to account for all the leadup data and trailer data | |
3533 | * around the transaction data. | |
3534 | * And then we need to account for the worst case in terms of using | |
3535 | * more space. | |
3536 | * The worst case will happen if: | |
3537 | * - the placement of the transaction happens to be such that the | |
3538 | * roundoff is at its maximum | |
3539 | * - the transaction data is synced before the commit record is synced | |
3540 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3541 | * Therefore the commit record is in its own Log Record. | |
3542 | * This can happen as the commit record is called with its | |
3543 | * own region to xlog_write(). | |
3544 | * This then means that in the worst case, roundoff can happen for | |
3545 | * the commit-rec as well. | |
3546 | * The commit-rec is smaller than padding in this scenario and so it is | |
3547 | * not added separately. | |
1da177e4 LT |
3548 | */ |
3549 | ||
32fb9b57 TS |
3550 | /* for trans header */ |
3551 | unit_bytes += sizeof(xlog_op_header_t); | |
3552 | unit_bytes += sizeof(xfs_trans_header_t); | |
3553 | ||
1da177e4 | 3554 | /* for start-rec */ |
32fb9b57 TS |
3555 | unit_bytes += sizeof(xlog_op_header_t); |
3556 | ||
9b9fc2b7 DC |
3557 | /* |
3558 | * for LR headers - the space for data in an iclog is the size minus | |
3559 | * the space used for the headers. If we use the iclog size, then we | |
3560 | * undercalculate the number of headers required. | |
3561 | * | |
3562 | * Furthermore - the addition of op headers for split-recs might | |
3563 | * increase the space required enough to require more log and op | |
3564 | * headers, so take that into account too. | |
3565 | * | |
3566 | * IMPORTANT: This reservation makes the assumption that if this | |
3567 | * transaction is the first in an iclog and hence has the LR headers | |
3568 | * accounted to it, then the remaining space in the iclog is | |
3569 | * exclusively for this transaction. i.e. if the transaction is larger | |
3570 | * than the iclog, it will be the only thing in that iclog. | |
3571 | * Fundamentally, this means we must pass the entire log vector to | |
3572 | * xlog_write to guarantee this. | |
3573 | */ | |
3574 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3575 | num_headers = howmany(unit_bytes, iclog_space); | |
3576 | ||
3577 | /* for split-recs - ophdrs added when data split over LRs */ | |
3578 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3579 | ||
3580 | /* add extra header reservations if we overrun */ | |
3581 | while (!num_headers || | |
3582 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3583 | unit_bytes += sizeof(xlog_op_header_t); | |
3584 | num_headers++; | |
3585 | } | |
32fb9b57 | 3586 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3587 | |
32fb9b57 TS |
3588 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3589 | unit_bytes += log->l_iclog_hsize; | |
3590 | ||
a6a65fef DC |
3591 | /* roundoff padding for transaction data and one for commit record */ |
3592 | unit_bytes += 2 * log->l_iclog_roundoff; | |
1da177e4 | 3593 | |
e773fc93 JL |
3594 | return unit_bytes; |
3595 | } | |
3596 | ||
a6a65fef DC |
3597 | int |
3598 | xfs_log_calc_unit_res( | |
3599 | struct xfs_mount *mp, | |
3600 | int unit_bytes) | |
3601 | { | |
3602 | return xlog_calc_unit_res(mp->m_log, unit_bytes); | |
3603 | } | |
3604 | ||
e773fc93 JL |
3605 | /* |
3606 | * Allocate and initialise a new log ticket. | |
3607 | */ | |
3608 | struct xlog_ticket * | |
3609 | xlog_ticket_alloc( | |
3610 | struct xlog *log, | |
3611 | int unit_bytes, | |
3612 | int cnt, | |
3613 | char client, | |
ca4f2589 | 3614 | bool permanent) |
e773fc93 JL |
3615 | { |
3616 | struct xlog_ticket *tic; | |
3617 | int unit_res; | |
3618 | ||
ca4f2589 | 3619 | tic = kmem_cache_zalloc(xfs_log_ticket_zone, GFP_NOFS | __GFP_NOFAIL); |
e773fc93 | 3620 | |
a6a65fef | 3621 | unit_res = xlog_calc_unit_res(log, unit_bytes); |
e773fc93 | 3622 | |
cc09c0dc | 3623 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3624 | tic->t_task = current; |
10547941 | 3625 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3626 | tic->t_unit_res = unit_res; |
3627 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3628 | tic->t_cnt = cnt; |
3629 | tic->t_ocnt = cnt; | |
ecb3403d | 3630 | tic->t_tid = prandom_u32(); |
1da177e4 | 3631 | tic->t_clientid = client; |
9006fb91 | 3632 | if (permanent) |
1da177e4 | 3633 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3634 | |
0adba536 | 3635 | xlog_tic_reset_res(tic); |
7e9c6396 | 3636 | |
1da177e4 | 3637 | return tic; |
cc09c0dc | 3638 | } |
1da177e4 | 3639 | |
cfcbbbd0 | 3640 | #if defined(DEBUG) |
1da177e4 LT |
3641 | /* |
3642 | * Make sure that the destination ptr is within the valid data region of | |
3643 | * one of the iclogs. This uses backup pointers stored in a different | |
3644 | * part of the log in case we trash the log structure. | |
3645 | */ | |
181fdfe6 | 3646 | STATIC void |
e6b1f273 | 3647 | xlog_verify_dest_ptr( |
ad223e60 | 3648 | struct xlog *log, |
5809d5e0 | 3649 | void *ptr) |
1da177e4 LT |
3650 | { |
3651 | int i; | |
3652 | int good_ptr = 0; | |
3653 | ||
e6b1f273 CH |
3654 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3655 | if (ptr >= log->l_iclog_bak[i] && | |
3656 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3657 | good_ptr++; |
3658 | } | |
e6b1f273 CH |
3659 | |
3660 | if (!good_ptr) | |
a0fa2b67 | 3661 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3662 | } |
1da177e4 | 3663 | |
da8a1a4a DC |
3664 | /* |
3665 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3666 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3667 | * the cycles differ by exactly one and check the byte count. | |
3668 | * | |
3669 | * This check is run unlocked, so can give false positives. Rather than assert | |
3670 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3671 | * determine if they want to panic the machine when such an error occurs. For | |
3672 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3673 | * an assert, it can be turned off at runtime. | |
3674 | */ | |
3f336c6f DC |
3675 | STATIC void |
3676 | xlog_verify_grant_tail( | |
ad223e60 | 3677 | struct xlog *log) |
3f336c6f | 3678 | { |
1c3cb9ec | 3679 | int tail_cycle, tail_blocks; |
a69ed03c | 3680 | int cycle, space; |
3f336c6f | 3681 | |
28496968 | 3682 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3683 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3684 | if (tail_cycle != cycle) { | |
da8a1a4a | 3685 | if (cycle - 1 != tail_cycle && |
e1d06e5f | 3686 | !test_and_set_bit(XLOG_TAIL_WARN, &log->l_opstate)) { |
da8a1a4a DC |
3687 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, |
3688 | "%s: cycle - 1 != tail_cycle", __func__); | |
da8a1a4a DC |
3689 | } |
3690 | ||
3691 | if (space > BBTOB(tail_blocks) && | |
e1d06e5f | 3692 | !test_and_set_bit(XLOG_TAIL_WARN, &log->l_opstate)) { |
da8a1a4a DC |
3693 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, |
3694 | "%s: space > BBTOB(tail_blocks)", __func__); | |
da8a1a4a | 3695 | } |
3f336c6f DC |
3696 | } |
3697 | } | |
3698 | ||
1da177e4 LT |
3699 | /* check if it will fit */ |
3700 | STATIC void | |
9a8d2fdb MT |
3701 | xlog_verify_tail_lsn( |
3702 | struct xlog *log, | |
9d110014 | 3703 | struct xlog_in_core *iclog) |
1da177e4 | 3704 | { |
9d110014 DC |
3705 | xfs_lsn_t tail_lsn = be64_to_cpu(iclog->ic_header.h_tail_lsn); |
3706 | int blocks; | |
1da177e4 LT |
3707 | |
3708 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3709 | blocks = | |
3710 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3711 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3712 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3713 | } else { |
3714 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3715 | ||
3716 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3717 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3718 | |
3719 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3720 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3721 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 | 3722 | } |
b843299b | 3723 | } |
1da177e4 LT |
3724 | |
3725 | /* | |
3726 | * Perform a number of checks on the iclog before writing to disk. | |
3727 | * | |
3728 | * 1. Make sure the iclogs are still circular | |
3729 | * 2. Make sure we have a good magic number | |
3730 | * 3. Make sure we don't have magic numbers in the data | |
3731 | * 4. Check fields of each log operation header for: | |
3732 | * A. Valid client identifier | |
3733 | * B. tid ptr value falls in valid ptr space (user space code) | |
3734 | * C. Length in log record header is correct according to the | |
3735 | * individual operation headers within record. | |
3736 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3737 | * log, check the preceding blocks of the physical log to make sure all | |
3738 | * the cycle numbers agree with the current cycle number. | |
3739 | */ | |
3740 | STATIC void | |
9a8d2fdb MT |
3741 | xlog_verify_iclog( |
3742 | struct xlog *log, | |
3743 | struct xlog_in_core *iclog, | |
abca1f33 | 3744 | int count) |
1da177e4 LT |
3745 | { |
3746 | xlog_op_header_t *ophead; | |
3747 | xlog_in_core_t *icptr; | |
3748 | xlog_in_core_2_t *xhdr; | |
5809d5e0 | 3749 | void *base_ptr, *ptr, *p; |
db9d67d6 | 3750 | ptrdiff_t field_offset; |
c8ce540d | 3751 | uint8_t clientid; |
1da177e4 LT |
3752 | int len, i, j, k, op_len; |
3753 | int idx; | |
1da177e4 LT |
3754 | |
3755 | /* check validity of iclog pointers */ | |
b22cd72c | 3756 | spin_lock(&log->l_icloglock); |
1da177e4 | 3757 | icptr = log->l_iclog; |
643f7c4e GB |
3758 | for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next) |
3759 | ASSERT(icptr); | |
3760 | ||
1da177e4 | 3761 | if (icptr != log->l_iclog) |
a0fa2b67 | 3762 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3763 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3764 | |
3765 | /* check log magic numbers */ | |
69ef921b | 3766 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3767 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3768 | |
5809d5e0 CH |
3769 | base_ptr = ptr = &iclog->ic_header; |
3770 | p = &iclog->ic_header; | |
3771 | for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) { | |
69ef921b | 3772 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3773 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3774 | __func__); | |
1da177e4 LT |
3775 | } |
3776 | ||
3777 | /* check fields */ | |
b53e675d | 3778 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
5809d5e0 CH |
3779 | base_ptr = ptr = iclog->ic_datap; |
3780 | ophead = ptr; | |
b28708d6 | 3781 | xhdr = iclog->ic_data; |
1da177e4 | 3782 | for (i = 0; i < len; i++) { |
5809d5e0 | 3783 | ophead = ptr; |
1da177e4 LT |
3784 | |
3785 | /* clientid is only 1 byte */ | |
5809d5e0 CH |
3786 | p = &ophead->oh_clientid; |
3787 | field_offset = p - base_ptr; | |
abca1f33 | 3788 | if (field_offset & 0x1ff) { |
1da177e4 LT |
3789 | clientid = ophead->oh_clientid; |
3790 | } else { | |
b2a922cd | 3791 | idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap); |
1da177e4 LT |
3792 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3793 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3794 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3795 | clientid = xlog_get_client_id( |
3796 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3797 | } else { |
03bea6fe CH |
3798 | clientid = xlog_get_client_id( |
3799 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3800 | } |
3801 | } | |
3802 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 | 3803 | xfs_warn(log->l_mp, |
c9690043 | 3804 | "%s: invalid clientid %d op "PTR_FMT" offset 0x%lx", |
a0fa2b67 DC |
3805 | __func__, clientid, ophead, |
3806 | (unsigned long)field_offset); | |
1da177e4 LT |
3807 | |
3808 | /* check length */ | |
5809d5e0 CH |
3809 | p = &ophead->oh_len; |
3810 | field_offset = p - base_ptr; | |
abca1f33 | 3811 | if (field_offset & 0x1ff) { |
67fcb7bf | 3812 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 | 3813 | } else { |
db9d67d6 CH |
3814 | idx = BTOBBT((uintptr_t)&ophead->oh_len - |
3815 | (uintptr_t)iclog->ic_datap); | |
1da177e4 LT |
3816 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3817 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3818 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3819 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3820 | } else { |
b53e675d | 3821 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3822 | } |
3823 | } | |
3824 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3825 | } | |
b843299b | 3826 | } |
cfcbbbd0 | 3827 | #endif |
1da177e4 | 3828 | |
1da177e4 | 3829 | /* |
b36d4651 DC |
3830 | * Perform a forced shutdown on the log. This should be called once and once |
3831 | * only by the high level filesystem shutdown code to shut the log subsystem | |
3832 | * down cleanly. | |
9da1ab18 | 3833 | * |
b36d4651 DC |
3834 | * Our main objectives here are to make sure that: |
3835 | * a. if the shutdown was not due to a log IO error, flush the logs to | |
3836 | * disk. Anything modified after this is ignored. | |
3837 | * b. the log gets atomically marked 'XLOG_IO_ERROR' for all interested | |
3838 | * parties to find out. Nothing new gets queued after this is done. | |
3839 | * c. Tasks sleeping on log reservations, pinned objects and | |
3840 | * other resources get woken up. | |
5112e206 | 3841 | * |
b36d4651 DC |
3842 | * Return true if the shutdown cause was a log IO error and we actually shut the |
3843 | * log down. | |
1da177e4 | 3844 | */ |
b36d4651 DC |
3845 | bool |
3846 | xlog_force_shutdown( | |
3847 | struct xlog *log, | |
3848 | int shutdown_flags) | |
1da177e4 | 3849 | { |
b36d4651 | 3850 | bool log_error = (shutdown_flags & SHUTDOWN_LOG_IO_ERROR); |
1da177e4 LT |
3851 | |
3852 | /* | |
b36d4651 DC |
3853 | * If this happens during log recovery then we aren't using the runtime |
3854 | * log mechanisms yet so there's nothing to shut down. | |
1da177e4 | 3855 | */ |
b36d4651 DC |
3856 | if (!log || xlog_in_recovery(log)) |
3857 | return false; | |
1da177e4 | 3858 | |
b36d4651 | 3859 | ASSERT(!xlog_is_shutdown(log)); |
9da1ab18 DC |
3860 | |
3861 | /* | |
a870fe6d | 3862 | * Flush all the completed transactions to disk before marking the log |
b36d4651 DC |
3863 | * being shut down. We need to do this first as shutting down the log |
3864 | * before the force will prevent the log force from flushing the iclogs | |
3865 | * to disk. | |
3866 | * | |
3867 | * Re-entry due to a log IO error shutdown during the log force is | |
3868 | * prevented by the atomicity of higher level shutdown code. | |
9da1ab18 | 3869 | */ |
b36d4651 DC |
3870 | if (!log_error) |
3871 | xfs_log_force(log->l_mp, XFS_LOG_SYNC); | |
9da1ab18 | 3872 | |
1da177e4 | 3873 | /* |
b36d4651 DC |
3874 | * Atomically set the shutdown state. If the shutdown state is already |
3875 | * set, there someone else is performing the shutdown and so we are done | |
3876 | * here. This should never happen because we should only ever get called | |
3877 | * once by the first shutdown caller. | |
3878 | * | |
3879 | * Much of the log state machine transitions assume that shutdown state | |
3880 | * cannot change once they hold the log->l_icloglock. Hence we need to | |
3881 | * hold that lock here, even though we use the atomic test_and_set_bit() | |
3882 | * operation to set the shutdown state. | |
1da177e4 | 3883 | */ |
b22cd72c | 3884 | spin_lock(&log->l_icloglock); |
b36d4651 DC |
3885 | if (test_and_set_bit(XLOG_IO_ERROR, &log->l_opstate)) { |
3886 | spin_unlock(&log->l_icloglock); | |
3887 | ASSERT(0); | |
3888 | return false; | |
3889 | } | |
b22cd72c | 3890 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3891 | |
3892 | /* | |
10547941 DC |
3893 | * We don't want anybody waiting for log reservations after this. That |
3894 | * means we have to wake up everybody queued up on reserveq as well as | |
3895 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3896 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3897 | * action is protected by the grant locks. |
1da177e4 | 3898 | */ |
a79bf2d7 CH |
3899 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3900 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3901 | |
1da177e4 | 3902 | /* |
ac983517 DC |
3903 | * Wake up everybody waiting on xfs_log_force. Wake the CIL push first |
3904 | * as if the log writes were completed. The abort handling in the log | |
3905 | * item committed callback functions will do this again under lock to | |
3906 | * avoid races. | |
1da177e4 | 3907 | */ |
cdea5459 | 3908 | spin_lock(&log->l_cilp->xc_push_lock); |
ac983517 | 3909 | wake_up_all(&log->l_cilp->xc_commit_wait); |
cdea5459 | 3910 | spin_unlock(&log->l_cilp->xc_push_lock); |
aad7272a | 3911 | xlog_state_shutdown_callbacks(log); |
1da177e4 | 3912 | |
b36d4651 | 3913 | return log_error; |
1da177e4 LT |
3914 | } |
3915 | ||
ba0f32d4 | 3916 | STATIC int |
9a8d2fdb MT |
3917 | xlog_iclogs_empty( |
3918 | struct xlog *log) | |
1da177e4 LT |
3919 | { |
3920 | xlog_in_core_t *iclog; | |
3921 | ||
3922 | iclog = log->l_iclog; | |
3923 | do { | |
3924 | /* endianness does not matter here, zero is zero in | |
3925 | * any language. | |
3926 | */ | |
3927 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3928 | return 0; |
1da177e4 LT |
3929 | iclog = iclog->ic_next; |
3930 | } while (iclog != log->l_iclog); | |
014c2544 | 3931 | return 1; |
1da177e4 | 3932 | } |
f661f1e0 | 3933 | |
a45086e2 BF |
3934 | /* |
3935 | * Verify that an LSN stamped into a piece of metadata is valid. This is | |
3936 | * intended for use in read verifiers on v5 superblocks. | |
3937 | */ | |
3938 | bool | |
3939 | xfs_log_check_lsn( | |
3940 | struct xfs_mount *mp, | |
3941 | xfs_lsn_t lsn) | |
3942 | { | |
3943 | struct xlog *log = mp->m_log; | |
3944 | bool valid; | |
3945 | ||
3946 | /* | |
3947 | * norecovery mode skips mount-time log processing and unconditionally | |
3948 | * resets the in-core LSN. We can't validate in this mode, but | |
3949 | * modifications are not allowed anyways so just return true. | |
3950 | */ | |
3951 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
3952 | return true; | |
3953 | ||
3954 | /* | |
3955 | * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is | |
3956 | * handled by recovery and thus safe to ignore here. | |
3957 | */ | |
3958 | if (lsn == NULLCOMMITLSN) | |
3959 | return true; | |
3960 | ||
3961 | valid = xlog_valid_lsn(mp->m_log, lsn); | |
3962 | ||
3963 | /* warn the user about what's gone wrong before verifier failure */ | |
3964 | if (!valid) { | |
3965 | spin_lock(&log->l_icloglock); | |
3966 | xfs_warn(mp, | |
3967 | "Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). " | |
3968 | "Please unmount and run xfs_repair (>= v4.3) to resolve.", | |
3969 | CYCLE_LSN(lsn), BLOCK_LSN(lsn), | |
3970 | log->l_curr_cycle, log->l_curr_block); | |
3971 | spin_unlock(&log->l_icloglock); | |
3972 | } | |
3973 | ||
3974 | return valid; | |
3975 | } | |
0c60d3aa | 3976 | |
2b73a2c8 DW |
3977 | /* |
3978 | * Notify the log that we're about to start using a feature that is protected | |
3979 | * by a log incompat feature flag. This will prevent log covering from | |
3980 | * clearing those flags. | |
3981 | */ | |
3982 | void | |
3983 | xlog_use_incompat_feat( | |
3984 | struct xlog *log) | |
3985 | { | |
3986 | down_read(&log->l_incompat_users); | |
3987 | } | |
3988 | ||
3989 | /* Notify the log that we've finished using log incompat features. */ | |
3990 | void | |
3991 | xlog_drop_incompat_feat( | |
3992 | struct xlog *log) | |
3993 | { | |
3994 | up_read(&log->l_incompat_users); | |
3995 | } |