]>
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" |
1da177e4 | 19 | #include "xfs_log_recover.h" |
a844f451 | 20 | #include "xfs_inode.h" |
0b1b213f | 21 | #include "xfs_trace.h" |
f661f1e0 | 22 | #include "xfs_fsops.h" |
0e446be4 | 23 | #include "xfs_cksum.h" |
baff4e44 | 24 | #include "xfs_sysfs.h" |
61e63ecb | 25 | #include "xfs_sb.h" |
39353ff6 | 26 | #include "xfs_health.h" |
1da177e4 | 27 | |
eb01c9cd | 28 | kmem_zone_t *xfs_log_ticket_zone; |
1da177e4 | 29 | |
1da177e4 | 30 | /* Local miscellaneous function prototypes */ |
ad223e60 MT |
31 | STATIC int |
32 | xlog_commit_record( | |
33 | struct xlog *log, | |
34 | struct xlog_ticket *ticket, | |
35 | struct xlog_in_core **iclog, | |
36 | xfs_lsn_t *commitlsnp); | |
37 | ||
9a8d2fdb MT |
38 | STATIC struct xlog * |
39 | xlog_alloc_log( | |
40 | struct xfs_mount *mp, | |
41 | struct xfs_buftarg *log_target, | |
42 | xfs_daddr_t blk_offset, | |
43 | int num_bblks); | |
ad223e60 MT |
44 | STATIC int |
45 | xlog_space_left( | |
46 | struct xlog *log, | |
47 | atomic64_t *head); | |
9a8d2fdb MT |
48 | STATIC void |
49 | xlog_dealloc_log( | |
50 | struct xlog *log); | |
1da177e4 LT |
51 | |
52 | /* local state machine functions */ | |
d15cbf2f CH |
53 | STATIC void xlog_state_done_syncing( |
54 | struct xlog_in_core *iclog, | |
55 | bool aborted); | |
9a8d2fdb MT |
56 | STATIC int |
57 | xlog_state_get_iclog_space( | |
58 | struct xlog *log, | |
59 | int len, | |
60 | struct xlog_in_core **iclog, | |
61 | struct xlog_ticket *ticket, | |
62 | int *continued_write, | |
63 | int *logoffsetp); | |
64 | STATIC int | |
65 | xlog_state_release_iclog( | |
66 | struct xlog *log, | |
67 | struct xlog_in_core *iclog); | |
68 | STATIC void | |
69 | xlog_state_switch_iclogs( | |
70 | struct xlog *log, | |
71 | struct xlog_in_core *iclog, | |
72 | int eventual_size); | |
73 | STATIC void | |
74 | xlog_state_want_sync( | |
75 | struct xlog *log, | |
76 | struct xlog_in_core *iclog); | |
1da177e4 | 77 | |
ad223e60 MT |
78 | STATIC void |
79 | xlog_grant_push_ail( | |
9a8d2fdb MT |
80 | struct xlog *log, |
81 | int need_bytes); | |
82 | STATIC void | |
83 | xlog_regrant_reserve_log_space( | |
84 | struct xlog *log, | |
85 | struct xlog_ticket *ticket); | |
86 | STATIC void | |
87 | xlog_ungrant_log_space( | |
88 | struct xlog *log, | |
89 | struct xlog_ticket *ticket); | |
1da177e4 | 90 | |
cfcbbbd0 | 91 | #if defined(DEBUG) |
9a8d2fdb MT |
92 | STATIC void |
93 | xlog_verify_dest_ptr( | |
94 | struct xlog *log, | |
5809d5e0 | 95 | void *ptr); |
ad223e60 MT |
96 | STATIC void |
97 | xlog_verify_grant_tail( | |
9a8d2fdb MT |
98 | struct xlog *log); |
99 | STATIC void | |
100 | xlog_verify_iclog( | |
101 | struct xlog *log, | |
102 | struct xlog_in_core *iclog, | |
abca1f33 | 103 | int count); |
9a8d2fdb MT |
104 | STATIC void |
105 | xlog_verify_tail_lsn( | |
106 | struct xlog *log, | |
107 | struct xlog_in_core *iclog, | |
108 | xfs_lsn_t tail_lsn); | |
1da177e4 LT |
109 | #else |
110 | #define xlog_verify_dest_ptr(a,b) | |
3f336c6f | 111 | #define xlog_verify_grant_tail(a) |
abca1f33 | 112 | #define xlog_verify_iclog(a,b,c) |
1da177e4 LT |
113 | #define xlog_verify_tail_lsn(a,b,c) |
114 | #endif | |
115 | ||
9a8d2fdb MT |
116 | STATIC int |
117 | xlog_iclogs_empty( | |
118 | struct xlog *log); | |
1da177e4 | 119 | |
dd954c69 | 120 | static void |
663e496a | 121 | xlog_grant_sub_space( |
ad223e60 MT |
122 | struct xlog *log, |
123 | atomic64_t *head, | |
124 | int bytes) | |
dd954c69 | 125 | { |
d0eb2f38 DC |
126 | int64_t head_val = atomic64_read(head); |
127 | int64_t new, old; | |
a69ed03c | 128 | |
d0eb2f38 DC |
129 | do { |
130 | int cycle, space; | |
a69ed03c | 131 | |
d0eb2f38 | 132 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 133 | |
d0eb2f38 DC |
134 | space -= bytes; |
135 | if (space < 0) { | |
136 | space += log->l_logsize; | |
137 | cycle--; | |
138 | } | |
139 | ||
140 | old = head_val; | |
141 | new = xlog_assign_grant_head_val(cycle, space); | |
142 | head_val = atomic64_cmpxchg(head, old, new); | |
143 | } while (head_val != old); | |
dd954c69 CH |
144 | } |
145 | ||
146 | static void | |
663e496a | 147 | xlog_grant_add_space( |
ad223e60 MT |
148 | struct xlog *log, |
149 | atomic64_t *head, | |
150 | int bytes) | |
dd954c69 | 151 | { |
d0eb2f38 DC |
152 | int64_t head_val = atomic64_read(head); |
153 | int64_t new, old; | |
a69ed03c | 154 | |
d0eb2f38 DC |
155 | do { |
156 | int tmp; | |
157 | int cycle, space; | |
a69ed03c | 158 | |
d0eb2f38 | 159 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 160 | |
d0eb2f38 DC |
161 | tmp = log->l_logsize - space; |
162 | if (tmp > bytes) | |
163 | space += bytes; | |
164 | else { | |
165 | space = bytes - tmp; | |
166 | cycle++; | |
167 | } | |
168 | ||
169 | old = head_val; | |
170 | new = xlog_assign_grant_head_val(cycle, space); | |
171 | head_val = atomic64_cmpxchg(head, old, new); | |
172 | } while (head_val != old); | |
dd954c69 | 173 | } |
a69ed03c | 174 | |
c303c5b8 CH |
175 | STATIC void |
176 | xlog_grant_head_init( | |
177 | struct xlog_grant_head *head) | |
178 | { | |
179 | xlog_assign_grant_head(&head->grant, 1, 0); | |
180 | INIT_LIST_HEAD(&head->waiters); | |
181 | spin_lock_init(&head->lock); | |
182 | } | |
183 | ||
a79bf2d7 CH |
184 | STATIC void |
185 | xlog_grant_head_wake_all( | |
186 | struct xlog_grant_head *head) | |
187 | { | |
188 | struct xlog_ticket *tic; | |
189 | ||
190 | spin_lock(&head->lock); | |
191 | list_for_each_entry(tic, &head->waiters, t_queue) | |
192 | wake_up_process(tic->t_task); | |
193 | spin_unlock(&head->lock); | |
194 | } | |
195 | ||
e179840d CH |
196 | static inline int |
197 | xlog_ticket_reservation( | |
ad223e60 | 198 | struct xlog *log, |
e179840d CH |
199 | struct xlog_grant_head *head, |
200 | struct xlog_ticket *tic) | |
9f9c19ec | 201 | { |
e179840d CH |
202 | if (head == &log->l_write_head) { |
203 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
204 | return tic->t_unit_res; | |
205 | } else { | |
9f9c19ec | 206 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) |
e179840d | 207 | return tic->t_unit_res * tic->t_cnt; |
9f9c19ec | 208 | else |
e179840d | 209 | return tic->t_unit_res; |
9f9c19ec | 210 | } |
9f9c19ec CH |
211 | } |
212 | ||
213 | STATIC bool | |
e179840d | 214 | xlog_grant_head_wake( |
ad223e60 | 215 | struct xlog *log, |
e179840d | 216 | struct xlog_grant_head *head, |
9f9c19ec CH |
217 | int *free_bytes) |
218 | { | |
219 | struct xlog_ticket *tic; | |
220 | int need_bytes; | |
221 | ||
e179840d CH |
222 | list_for_each_entry(tic, &head->waiters, t_queue) { |
223 | need_bytes = xlog_ticket_reservation(log, head, tic); | |
9f9c19ec CH |
224 | if (*free_bytes < need_bytes) |
225 | return false; | |
9f9c19ec | 226 | |
e179840d CH |
227 | *free_bytes -= need_bytes; |
228 | trace_xfs_log_grant_wake_up(log, tic); | |
14a7235f | 229 | wake_up_process(tic->t_task); |
9f9c19ec CH |
230 | } |
231 | ||
232 | return true; | |
233 | } | |
234 | ||
235 | STATIC int | |
23ee3df3 | 236 | xlog_grant_head_wait( |
ad223e60 | 237 | struct xlog *log, |
23ee3df3 | 238 | struct xlog_grant_head *head, |
9f9c19ec | 239 | struct xlog_ticket *tic, |
a30b0367 DC |
240 | int need_bytes) __releases(&head->lock) |
241 | __acquires(&head->lock) | |
9f9c19ec | 242 | { |
23ee3df3 | 243 | list_add_tail(&tic->t_queue, &head->waiters); |
9f9c19ec CH |
244 | |
245 | do { | |
246 | if (XLOG_FORCED_SHUTDOWN(log)) | |
247 | goto shutdown; | |
248 | xlog_grant_push_ail(log, need_bytes); | |
249 | ||
14a7235f | 250 | __set_current_state(TASK_UNINTERRUPTIBLE); |
23ee3df3 | 251 | spin_unlock(&head->lock); |
14a7235f | 252 | |
ff6d6af2 | 253 | XFS_STATS_INC(log->l_mp, xs_sleep_logspace); |
9f9c19ec | 254 | |
14a7235f CH |
255 | trace_xfs_log_grant_sleep(log, tic); |
256 | schedule(); | |
9f9c19ec CH |
257 | trace_xfs_log_grant_wake(log, tic); |
258 | ||
23ee3df3 | 259 | spin_lock(&head->lock); |
9f9c19ec CH |
260 | if (XLOG_FORCED_SHUTDOWN(log)) |
261 | goto shutdown; | |
23ee3df3 | 262 | } while (xlog_space_left(log, &head->grant) < need_bytes); |
9f9c19ec CH |
263 | |
264 | list_del_init(&tic->t_queue); | |
265 | return 0; | |
266 | shutdown: | |
267 | list_del_init(&tic->t_queue); | |
2451337d | 268 | return -EIO; |
9f9c19ec CH |
269 | } |
270 | ||
42ceedb3 CH |
271 | /* |
272 | * Atomically get the log space required for a log ticket. | |
273 | * | |
274 | * Once a ticket gets put onto head->waiters, it will only return after the | |
275 | * needed reservation is satisfied. | |
276 | * | |
277 | * This function is structured so that it has a lock free fast path. This is | |
278 | * necessary because every new transaction reservation will come through this | |
279 | * path. Hence any lock will be globally hot if we take it unconditionally on | |
280 | * every pass. | |
281 | * | |
282 | * As tickets are only ever moved on and off head->waiters under head->lock, we | |
283 | * only need to take that lock if we are going to add the ticket to the queue | |
284 | * and sleep. We can avoid taking the lock if the ticket was never added to | |
285 | * head->waiters because the t_queue list head will be empty and we hold the | |
286 | * only reference to it so it can safely be checked unlocked. | |
287 | */ | |
288 | STATIC int | |
289 | xlog_grant_head_check( | |
ad223e60 | 290 | struct xlog *log, |
42ceedb3 CH |
291 | struct xlog_grant_head *head, |
292 | struct xlog_ticket *tic, | |
293 | int *need_bytes) | |
294 | { | |
295 | int free_bytes; | |
296 | int error = 0; | |
297 | ||
298 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); | |
299 | ||
300 | /* | |
301 | * If there are other waiters on the queue then give them a chance at | |
302 | * logspace before us. Wake up the first waiters, if we do not wake | |
303 | * up all the waiters then go to sleep waiting for more free space, | |
304 | * otherwise try to get some space for this transaction. | |
305 | */ | |
306 | *need_bytes = xlog_ticket_reservation(log, head, tic); | |
307 | free_bytes = xlog_space_left(log, &head->grant); | |
308 | if (!list_empty_careful(&head->waiters)) { | |
309 | spin_lock(&head->lock); | |
310 | if (!xlog_grant_head_wake(log, head, &free_bytes) || | |
311 | free_bytes < *need_bytes) { | |
312 | error = xlog_grant_head_wait(log, head, tic, | |
313 | *need_bytes); | |
314 | } | |
315 | spin_unlock(&head->lock); | |
316 | } else if (free_bytes < *need_bytes) { | |
317 | spin_lock(&head->lock); | |
318 | error = xlog_grant_head_wait(log, head, tic, *need_bytes); | |
319 | spin_unlock(&head->lock); | |
320 | } | |
321 | ||
322 | return error; | |
323 | } | |
324 | ||
0adba536 CH |
325 | static void |
326 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
327 | { | |
328 | tic->t_res_num = 0; | |
329 | tic->t_res_arr_sum = 0; | |
330 | tic->t_res_num_ophdrs = 0; | |
331 | } | |
332 | ||
333 | static void | |
334 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
335 | { | |
336 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
337 | /* add to overflow and start again */ | |
338 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
339 | tic->t_res_num = 0; | |
340 | tic->t_res_arr_sum = 0; | |
341 | } | |
342 | ||
343 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
344 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
345 | tic->t_res_arr_sum += len; | |
346 | tic->t_res_num++; | |
347 | } | |
dd954c69 | 348 | |
9006fb91 CH |
349 | /* |
350 | * Replenish the byte reservation required by moving the grant write head. | |
351 | */ | |
352 | int | |
353 | xfs_log_regrant( | |
354 | struct xfs_mount *mp, | |
355 | struct xlog_ticket *tic) | |
356 | { | |
ad223e60 | 357 | struct xlog *log = mp->m_log; |
9006fb91 CH |
358 | int need_bytes; |
359 | int error = 0; | |
360 | ||
361 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 362 | return -EIO; |
9006fb91 | 363 | |
ff6d6af2 | 364 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
365 | |
366 | /* | |
367 | * This is a new transaction on the ticket, so we need to change the | |
368 | * transaction ID so that the next transaction has a different TID in | |
369 | * the log. Just add one to the existing tid so that we can see chains | |
370 | * of rolling transactions in the log easily. | |
371 | */ | |
372 | tic->t_tid++; | |
373 | ||
374 | xlog_grant_push_ail(log, tic->t_unit_res); | |
375 | ||
376 | tic->t_curr_res = tic->t_unit_res; | |
377 | xlog_tic_reset_res(tic); | |
378 | ||
379 | if (tic->t_cnt > 0) | |
380 | return 0; | |
381 | ||
382 | trace_xfs_log_regrant(log, tic); | |
383 | ||
384 | error = xlog_grant_head_check(log, &log->l_write_head, tic, | |
385 | &need_bytes); | |
386 | if (error) | |
387 | goto out_error; | |
388 | ||
389 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
390 | trace_xfs_log_regrant_exit(log, tic); | |
391 | xlog_verify_grant_tail(log); | |
392 | return 0; | |
393 | ||
394 | out_error: | |
395 | /* | |
396 | * If we are failing, make sure the ticket doesn't have any current | |
397 | * reservations. We don't want to add this back when the ticket/ | |
398 | * transaction gets cancelled. | |
399 | */ | |
400 | tic->t_curr_res = 0; | |
401 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
402 | return error; | |
403 | } | |
404 | ||
405 | /* | |
a0e336ba | 406 | * Reserve log space and return a ticket corresponding to the reservation. |
9006fb91 CH |
407 | * |
408 | * Each reservation is going to reserve extra space for a log record header. | |
409 | * When writes happen to the on-disk log, we don't subtract the length of the | |
410 | * log record header from any reservation. By wasting space in each | |
411 | * reservation, we prevent over allocation problems. | |
412 | */ | |
413 | int | |
414 | xfs_log_reserve( | |
415 | struct xfs_mount *mp, | |
416 | int unit_bytes, | |
417 | int cnt, | |
418 | struct xlog_ticket **ticp, | |
c8ce540d | 419 | uint8_t client, |
710b1e2c | 420 | bool permanent) |
9006fb91 | 421 | { |
ad223e60 | 422 | struct xlog *log = mp->m_log; |
9006fb91 CH |
423 | struct xlog_ticket *tic; |
424 | int need_bytes; | |
425 | int error = 0; | |
426 | ||
427 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); | |
428 | ||
429 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 430 | return -EIO; |
9006fb91 | 431 | |
ff6d6af2 | 432 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
433 | |
434 | ASSERT(*ticp == NULL); | |
435 | tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, | |
436 | KM_SLEEP | KM_MAYFAIL); | |
437 | if (!tic) | |
2451337d | 438 | return -ENOMEM; |
9006fb91 | 439 | |
9006fb91 CH |
440 | *ticp = tic; |
441 | ||
437a255a DC |
442 | xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt |
443 | : tic->t_unit_res); | |
9006fb91 CH |
444 | |
445 | trace_xfs_log_reserve(log, tic); | |
446 | ||
447 | error = xlog_grant_head_check(log, &log->l_reserve_head, tic, | |
448 | &need_bytes); | |
449 | if (error) | |
450 | goto out_error; | |
451 | ||
452 | xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); | |
453 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
454 | trace_xfs_log_reserve_exit(log, tic); | |
455 | xlog_verify_grant_tail(log); | |
456 | return 0; | |
457 | ||
458 | out_error: | |
459 | /* | |
460 | * If we are failing, make sure the ticket doesn't have any current | |
461 | * reservations. We don't want to add this back when the ticket/ | |
462 | * transaction gets cancelled. | |
463 | */ | |
464 | tic->t_curr_res = 0; | |
465 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
466 | return error; | |
467 | } | |
468 | ||
469 | ||
1da177e4 LT |
470 | /* |
471 | * NOTES: | |
472 | * | |
473 | * 1. currblock field gets updated at startup and after in-core logs | |
474 | * marked as with WANT_SYNC. | |
475 | */ | |
476 | ||
477 | /* | |
478 | * This routine is called when a user of a log manager ticket is done with | |
479 | * the reservation. If the ticket was ever used, then a commit record for | |
480 | * the associated transaction is written out as a log operation header with | |
481 | * no data. The flag XLOG_TIC_INITED is set when the first write occurs with | |
482 | * a given ticket. If the ticket was one with a permanent reservation, then | |
483 | * a few operations are done differently. Permanent reservation tickets by | |
484 | * default don't release the reservation. They just commit the current | |
485 | * transaction with the belief that the reservation is still needed. A flag | |
486 | * must be passed in before permanent reservations are actually released. | |
487 | * When these type of tickets are not released, they need to be set into | |
488 | * the inited state again. By doing this, a start record will be written | |
489 | * out when the next write occurs. | |
490 | */ | |
491 | xfs_lsn_t | |
35a8a72f CH |
492 | xfs_log_done( |
493 | struct xfs_mount *mp, | |
494 | struct xlog_ticket *ticket, | |
495 | struct xlog_in_core **iclog, | |
f78c3901 | 496 | bool regrant) |
1da177e4 | 497 | { |
ad223e60 | 498 | struct xlog *log = mp->m_log; |
35a8a72f | 499 | xfs_lsn_t lsn = 0; |
1da177e4 | 500 | |
1da177e4 LT |
501 | if (XLOG_FORCED_SHUTDOWN(log) || |
502 | /* | |
503 | * If nothing was ever written, don't write out commit record. | |
504 | * If we get an error, just continue and give back the log ticket. | |
505 | */ | |
506 | (((ticket->t_flags & XLOG_TIC_INITED) == 0) && | |
55b66332 | 507 | (xlog_commit_record(log, ticket, iclog, &lsn)))) { |
1da177e4 | 508 | lsn = (xfs_lsn_t) -1; |
f78c3901 | 509 | regrant = false; |
1da177e4 LT |
510 | } |
511 | ||
512 | ||
f78c3901 | 513 | if (!regrant) { |
0b1b213f CH |
514 | trace_xfs_log_done_nonperm(log, ticket); |
515 | ||
1da177e4 | 516 | /* |
c41564b5 | 517 | * Release ticket if not permanent reservation or a specific |
1da177e4 LT |
518 | * request has been made to release a permanent reservation. |
519 | */ | |
520 | xlog_ungrant_log_space(log, ticket); | |
1da177e4 | 521 | } else { |
0b1b213f CH |
522 | trace_xfs_log_done_perm(log, ticket); |
523 | ||
1da177e4 | 524 | xlog_regrant_reserve_log_space(log, ticket); |
c6a7b0f8 LM |
525 | /* If this ticket was a permanent reservation and we aren't |
526 | * trying to release it, reset the inited flags; so next time | |
527 | * we write, a start record will be written out. | |
528 | */ | |
1da177e4 | 529 | ticket->t_flags |= XLOG_TIC_INITED; |
c6a7b0f8 | 530 | } |
1da177e4 | 531 | |
f78c3901 | 532 | xfs_log_ticket_put(ticket); |
1da177e4 | 533 | return lsn; |
35a8a72f | 534 | } |
1da177e4 | 535 | |
1da177e4 LT |
536 | /* |
537 | * Attaches a new iclog I/O completion callback routine during | |
538 | * transaction commit. If the log is in error state, a non-zero | |
539 | * return code is handed back and the caller is responsible for | |
540 | * executing the callback at an appropriate time. | |
541 | */ | |
542 | int | |
35a8a72f | 543 | xfs_log_notify( |
35a8a72f CH |
544 | struct xlog_in_core *iclog, |
545 | xfs_log_callback_t *cb) | |
1da177e4 | 546 | { |
b22cd72c | 547 | int abortflg; |
1da177e4 | 548 | |
114d23aa | 549 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
550 | abortflg = (iclog->ic_state & XLOG_STATE_IOERROR); |
551 | if (!abortflg) { | |
552 | ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) || | |
553 | (iclog->ic_state == XLOG_STATE_WANT_SYNC)); | |
554 | cb->cb_next = NULL; | |
555 | *(iclog->ic_callback_tail) = cb; | |
556 | iclog->ic_callback_tail = &(cb->cb_next); | |
557 | } | |
114d23aa | 558 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 | 559 | return abortflg; |
35a8a72f | 560 | } |
1da177e4 LT |
561 | |
562 | int | |
35a8a72f CH |
563 | xfs_log_release_iclog( |
564 | struct xfs_mount *mp, | |
565 | struct xlog_in_core *iclog) | |
1da177e4 | 566 | { |
35a8a72f | 567 | if (xlog_state_release_iclog(mp->m_log, iclog)) { |
7d04a335 | 568 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
2451337d | 569 | return -EIO; |
1da177e4 LT |
570 | } |
571 | ||
572 | return 0; | |
573 | } | |
574 | ||
1da177e4 LT |
575 | /* |
576 | * Mount a log filesystem | |
577 | * | |
578 | * mp - ubiquitous xfs mount point structure | |
579 | * log_target - buftarg of on-disk log device | |
580 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
581 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
582 | * | |
583 | * Return error or zero. | |
584 | */ | |
585 | int | |
249a8c11 DC |
586 | xfs_log_mount( |
587 | xfs_mount_t *mp, | |
588 | xfs_buftarg_t *log_target, | |
589 | xfs_daddr_t blk_offset, | |
590 | int num_bblks) | |
1da177e4 | 591 | { |
9c92ee20 | 592 | bool fatal = xfs_sb_version_hascrc(&mp->m_sb); |
3e7b91cf JL |
593 | int error = 0; |
594 | int min_logfsbs; | |
249a8c11 | 595 | |
c99d609a DC |
596 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
597 | xfs_notice(mp, "Mounting V%d Filesystem", | |
598 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
599 | } else { | |
a0fa2b67 | 600 | xfs_notice(mp, |
c99d609a DC |
601 | "Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.", |
602 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
bd186aa9 | 603 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
604 | } |
605 | ||
606 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e | 607 | if (IS_ERR(mp->m_log)) { |
2451337d | 608 | error = PTR_ERR(mp->m_log); |
644c3567 DC |
609 | goto out; |
610 | } | |
1da177e4 | 611 | |
3e7b91cf JL |
612 | /* |
613 | * Validate the given log space and drop a critical message via syslog | |
614 | * if the log size is too small that would lead to some unexpected | |
615 | * situations in transaction log space reservation stage. | |
616 | * | |
617 | * Note: we can't just reject the mount if the validation fails. This | |
618 | * would mean that people would have to downgrade their kernel just to | |
619 | * remedy the situation as there is no way to grow the log (short of | |
620 | * black magic surgery with xfs_db). | |
621 | * | |
622 | * We can, however, reject mounts for CRC format filesystems, as the | |
623 | * mkfs binary being used to make the filesystem should never create a | |
624 | * filesystem with a log that is too small. | |
625 | */ | |
626 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
627 | ||
628 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
629 | xfs_warn(mp, | |
630 | "Log size %d blocks too small, minimum size is %d blocks", | |
631 | mp->m_sb.sb_logblocks, min_logfsbs); | |
2451337d | 632 | error = -EINVAL; |
3e7b91cf JL |
633 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { |
634 | xfs_warn(mp, | |
635 | "Log size %d blocks too large, maximum size is %lld blocks", | |
636 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
2451337d | 637 | error = -EINVAL; |
3e7b91cf JL |
638 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { |
639 | xfs_warn(mp, | |
640 | "log size %lld bytes too large, maximum size is %lld bytes", | |
641 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
642 | XFS_MAX_LOG_BYTES); | |
2451337d | 643 | error = -EINVAL; |
9c92ee20 DW |
644 | } else if (mp->m_sb.sb_logsunit > 1 && |
645 | mp->m_sb.sb_logsunit % mp->m_sb.sb_blocksize) { | |
646 | xfs_warn(mp, | |
647 | "log stripe unit %u bytes must be a multiple of block size", | |
648 | mp->m_sb.sb_logsunit); | |
649 | error = -EINVAL; | |
650 | fatal = true; | |
3e7b91cf JL |
651 | } |
652 | if (error) { | |
9c92ee20 DW |
653 | /* |
654 | * Log check errors are always fatal on v5; or whenever bad | |
655 | * metadata leads to a crash. | |
656 | */ | |
657 | if (fatal) { | |
3e7b91cf JL |
658 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); |
659 | ASSERT(0); | |
660 | goto out_free_log; | |
661 | } | |
f41febd2 | 662 | xfs_crit(mp, "Log size out of supported range."); |
3e7b91cf | 663 | xfs_crit(mp, |
f41febd2 | 664 | "Continuing onwards, but if log hangs are experienced then please report this message in the bug report."); |
3e7b91cf JL |
665 | } |
666 | ||
249a8c11 DC |
667 | /* |
668 | * Initialize the AIL now we have a log. | |
669 | */ | |
249a8c11 DC |
670 | error = xfs_trans_ail_init(mp); |
671 | if (error) { | |
a0fa2b67 | 672 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 673 | goto out_free_log; |
249a8c11 | 674 | } |
a9c21c1b | 675 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 676 | |
1da177e4 LT |
677 | /* |
678 | * skip log recovery on a norecovery mount. pretend it all | |
679 | * just worked. | |
680 | */ | |
681 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 682 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
683 | |
684 | if (readonly) | |
bd186aa9 | 685 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 686 | |
65be6054 | 687 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
688 | |
689 | if (readonly) | |
bd186aa9 | 690 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 691 | if (error) { |
a0fa2b67 DC |
692 | xfs_warn(mp, "log mount/recovery failed: error %d", |
693 | error); | |
f0b2efad | 694 | xlog_recover_cancel(mp->m_log); |
26430752 | 695 | goto out_destroy_ail; |
1da177e4 LT |
696 | } |
697 | } | |
698 | ||
baff4e44 BF |
699 | error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj, |
700 | "log"); | |
701 | if (error) | |
702 | goto out_destroy_ail; | |
703 | ||
1da177e4 LT |
704 | /* Normal transactions can now occur */ |
705 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
706 | ||
71e330b5 DC |
707 | /* |
708 | * Now the log has been fully initialised and we know were our | |
709 | * space grant counters are, we can initialise the permanent ticket | |
710 | * needed for delayed logging to work. | |
711 | */ | |
712 | xlog_cil_init_post_recovery(mp->m_log); | |
713 | ||
1da177e4 | 714 | return 0; |
26430752 CH |
715 | |
716 | out_destroy_ail: | |
717 | xfs_trans_ail_destroy(mp); | |
718 | out_free_log: | |
719 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 720 | out: |
249a8c11 | 721 | return error; |
26430752 | 722 | } |
1da177e4 LT |
723 | |
724 | /* | |
f661f1e0 DC |
725 | * Finish the recovery of the file system. This is separate from the |
726 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
727 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
728 | * here. | |
1da177e4 | 729 | * |
f661f1e0 DC |
730 | * If we finish recovery successfully, start the background log work. If we are |
731 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
732 | * it. | |
1da177e4 LT |
733 | */ |
734 | int | |
f0b2efad BF |
735 | xfs_log_mount_finish( |
736 | struct xfs_mount *mp) | |
1da177e4 | 737 | { |
f661f1e0 | 738 | int error = 0; |
6f4a1eef | 739 | bool readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
f1b92bbc | 740 | bool recovered = mp->m_log->l_flags & XLOG_RECOVERY_NEEDED; |
1da177e4 | 741 | |
f0b2efad | 742 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) { |
bd186aa9 | 743 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
f0b2efad | 744 | return 0; |
6f4a1eef ES |
745 | } else if (readonly) { |
746 | /* Allow unlinked processing to proceed */ | |
747 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
1da177e4 LT |
748 | } |
749 | ||
8204f8dd DW |
750 | /* |
751 | * During the second phase of log recovery, we need iget and | |
752 | * iput to behave like they do for an active filesystem. | |
753 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
754 | * of inodes before we're done replaying log items on those | |
755 | * inodes. Turn it off immediately after recovery finishes | |
756 | * so that we don't leak the quota inodes if subsequent mount | |
757 | * activities fail. | |
799ea9e9 DW |
758 | * |
759 | * We let all inodes involved in redo item processing end up on | |
760 | * the LRU instead of being evicted immediately so that if we do | |
761 | * something to an unlinked inode, the irele won't cause | |
762 | * premature truncation and freeing of the inode, which results | |
763 | * in log recovery failure. We have to evict the unreferenced | |
1751e8a6 | 764 | * lru inodes after clearing SB_ACTIVE because we don't |
799ea9e9 DW |
765 | * otherwise clean up the lru if there's a subsequent failure in |
766 | * xfs_mountfs, which leads to us leaking the inodes if nothing | |
767 | * else (e.g. quotacheck) references the inodes before the | |
768 | * mount failure occurs. | |
8204f8dd | 769 | */ |
1751e8a6 | 770 | mp->m_super->s_flags |= SB_ACTIVE; |
f0b2efad BF |
771 | error = xlog_recover_finish(mp->m_log); |
772 | if (!error) | |
773 | xfs_log_work_queue(mp); | |
1751e8a6 | 774 | mp->m_super->s_flags &= ~SB_ACTIVE; |
799ea9e9 | 775 | evict_inodes(mp->m_super); |
f0b2efad | 776 | |
f1b92bbc BF |
777 | /* |
778 | * Drain the buffer LRU after log recovery. This is required for v4 | |
779 | * filesystems to avoid leaving around buffers with NULL verifier ops, | |
780 | * but we do it unconditionally to make sure we're always in a clean | |
781 | * cache state after mount. | |
782 | * | |
783 | * Don't push in the error case because the AIL may have pending intents | |
784 | * that aren't removed until recovery is cancelled. | |
785 | */ | |
786 | if (!error && recovered) { | |
787 | xfs_log_force(mp, XFS_LOG_SYNC); | |
788 | xfs_ail_push_all_sync(mp->m_ail); | |
789 | } | |
790 | xfs_wait_buftarg(mp->m_ddev_targp); | |
791 | ||
6f4a1eef ES |
792 | if (readonly) |
793 | mp->m_flags |= XFS_MOUNT_RDONLY; | |
794 | ||
f0b2efad BF |
795 | return error; |
796 | } | |
797 | ||
798 | /* | |
799 | * The mount has failed. Cancel the recovery if it hasn't completed and destroy | |
800 | * the log. | |
801 | */ | |
802 | int | |
803 | xfs_log_mount_cancel( | |
804 | struct xfs_mount *mp) | |
805 | { | |
806 | int error; | |
807 | ||
808 | error = xlog_recover_cancel(mp->m_log); | |
809 | xfs_log_unmount(mp); | |
f661f1e0 | 810 | |
1da177e4 LT |
811 | return error; |
812 | } | |
813 | ||
1da177e4 LT |
814 | /* |
815 | * Final log writes as part of unmount. | |
816 | * | |
817 | * Mark the filesystem clean as unmount happens. Note that during relocation | |
818 | * this routine needs to be executed as part of source-bag while the | |
819 | * deallocation must not be done until source-end. | |
820 | */ | |
821 | ||
53235f22 DW |
822 | /* Actually write the unmount record to disk. */ |
823 | static void | |
824 | xfs_log_write_unmount_record( | |
825 | struct xfs_mount *mp) | |
826 | { | |
827 | /* the data section must be 32 bit size aligned */ | |
828 | struct xfs_unmount_log_format magic = { | |
829 | .magic = XLOG_UNMOUNT_TYPE, | |
830 | }; | |
831 | struct xfs_log_iovec reg = { | |
832 | .i_addr = &magic, | |
833 | .i_len = sizeof(magic), | |
834 | .i_type = XLOG_REG_TYPE_UNMOUNT, | |
835 | }; | |
836 | struct xfs_log_vec vec = { | |
837 | .lv_niovecs = 1, | |
838 | .lv_iovecp = ®, | |
839 | }; | |
840 | struct xlog *log = mp->m_log; | |
841 | struct xlog_in_core *iclog; | |
842 | struct xlog_ticket *tic = NULL; | |
843 | xfs_lsn_t lsn; | |
f467cad9 | 844 | uint flags = XLOG_UNMOUNT_TRANS; |
53235f22 DW |
845 | int error; |
846 | ||
847 | error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0); | |
848 | if (error) | |
849 | goto out_err; | |
850 | ||
f467cad9 DW |
851 | /* |
852 | * If we think the summary counters are bad, clear the unmount header | |
853 | * flag in the unmount record so that the summary counters will be | |
854 | * recalculated during log recovery at next mount. Refer to | |
855 | * xlog_check_unmount_rec for more details. | |
856 | */ | |
39353ff6 | 857 | if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp, |
f467cad9 DW |
858 | XFS_ERRTAG_FORCE_SUMMARY_RECALC)) { |
859 | xfs_alert(mp, "%s: will fix summary counters at next mount", | |
860 | __func__); | |
861 | flags &= ~XLOG_UNMOUNT_TRANS; | |
862 | } | |
863 | ||
53235f22 DW |
864 | /* remove inited flag, and account for space used */ |
865 | tic->t_flags = 0; | |
866 | tic->t_curr_res -= sizeof(magic); | |
f467cad9 | 867 | error = xlog_write(log, &vec, tic, &lsn, NULL, flags); |
53235f22 DW |
868 | /* |
869 | * At this point, we're umounting anyway, so there's no point in | |
870 | * transitioning log state to IOERROR. Just continue... | |
871 | */ | |
872 | out_err: | |
873 | if (error) | |
874 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
875 | ||
876 | spin_lock(&log->l_icloglock); | |
877 | iclog = log->l_iclog; | |
878 | atomic_inc(&iclog->ic_refcnt); | |
879 | xlog_state_want_sync(log, iclog); | |
880 | spin_unlock(&log->l_icloglock); | |
881 | error = xlog_state_release_iclog(log, iclog); | |
882 | ||
883 | spin_lock(&log->l_icloglock); | |
884 | switch (iclog->ic_state) { | |
885 | default: | |
886 | if (!XLOG_FORCED_SHUTDOWN(log)) { | |
887 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
888 | break; | |
889 | } | |
890 | /* fall through */ | |
891 | case XLOG_STATE_ACTIVE: | |
892 | case XLOG_STATE_DIRTY: | |
893 | spin_unlock(&log->l_icloglock); | |
894 | break; | |
895 | } | |
896 | ||
897 | if (tic) { | |
898 | trace_xfs_log_umount_write(log, tic); | |
899 | xlog_ungrant_log_space(log, tic); | |
900 | xfs_log_ticket_put(tic); | |
901 | } | |
902 | } | |
903 | ||
1da177e4 LT |
904 | /* |
905 | * Unmount record used to have a string "Unmount filesystem--" in the | |
906 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
907 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 908 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
909 | * As far as I know, there weren't any dependencies on the old behaviour. |
910 | */ | |
911 | ||
0d5a75e9 | 912 | static int |
1da177e4 LT |
913 | xfs_log_unmount_write(xfs_mount_t *mp) |
914 | { | |
9a8d2fdb | 915 | struct xlog *log = mp->m_log; |
1da177e4 LT |
916 | xlog_in_core_t *iclog; |
917 | #ifdef DEBUG | |
918 | xlog_in_core_t *first_iclog; | |
919 | #endif | |
1da177e4 | 920 | int error; |
1da177e4 | 921 | |
1da177e4 | 922 | /* |
757a69ef | 923 | * Don't write out unmount record on norecovery mounts or ro devices. |
1da177e4 LT |
924 | * Or, if we are doing a forced umount (typically because of IO errors). |
925 | */ | |
757a69ef | 926 | if (mp->m_flags & XFS_MOUNT_NORECOVERY || |
2d15d2c0 | 927 | xfs_readonly_buftarg(log->l_targ)) { |
757a69ef | 928 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 | 929 | return 0; |
757a69ef | 930 | } |
1da177e4 | 931 | |
60e5bb78 | 932 | error = xfs_log_force(mp, XFS_LOG_SYNC); |
b911ca04 | 933 | ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log))); |
1da177e4 LT |
934 | |
935 | #ifdef DEBUG | |
936 | first_iclog = iclog = log->l_iclog; | |
937 | do { | |
938 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
939 | ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE); | |
940 | ASSERT(iclog->ic_offset == 0); | |
941 | } | |
942 | iclog = iclog->ic_next; | |
943 | } while (iclog != first_iclog); | |
944 | #endif | |
945 | if (! (XLOG_FORCED_SHUTDOWN(log))) { | |
53235f22 | 946 | xfs_log_write_unmount_record(mp); |
1da177e4 LT |
947 | } else { |
948 | /* | |
949 | * We're already in forced_shutdown mode, couldn't | |
950 | * even attempt to write out the unmount transaction. | |
951 | * | |
952 | * Go through the motions of sync'ing and releasing | |
953 | * the iclog, even though no I/O will actually happen, | |
c41564b5 | 954 | * we need to wait for other log I/Os that may already |
1da177e4 LT |
955 | * be in progress. Do this as a separate section of |
956 | * code so we'll know if we ever get stuck here that | |
957 | * we're in this odd situation of trying to unmount | |
958 | * a file system that went into forced_shutdown as | |
959 | * the result of an unmount.. | |
960 | */ | |
b22cd72c | 961 | spin_lock(&log->l_icloglock); |
1da177e4 | 962 | iclog = log->l_iclog; |
155cc6b7 | 963 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 LT |
964 | |
965 | xlog_state_want_sync(log, iclog); | |
39e2defe | 966 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 967 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 968 | |
b22cd72c | 969 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
970 | |
971 | if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE | |
972 | || iclog->ic_state == XLOG_STATE_DIRTY | |
973 | || iclog->ic_state == XLOG_STATE_IOERROR) ) { | |
974 | ||
eb40a875 DC |
975 | xlog_wait(&iclog->ic_force_wait, |
976 | &log->l_icloglock); | |
1da177e4 | 977 | } else { |
b22cd72c | 978 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
979 | } |
980 | } | |
981 | ||
1bb7d6b5 | 982 | return error; |
1da177e4 LT |
983 | } /* xfs_log_unmount_write */ |
984 | ||
985 | /* | |
c75921a7 | 986 | * Empty the log for unmount/freeze. |
cf2931db DC |
987 | * |
988 | * To do this, we first need to shut down the background log work so it is not | |
989 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
990 | * the log so we can then flush them out. Once they have completed their IO and | |
991 | * run the callbacks removing themselves from the AIL, we can write the unmount | |
c75921a7 | 992 | * record. |
1da177e4 LT |
993 | */ |
994 | void | |
c75921a7 DC |
995 | xfs_log_quiesce( |
996 | struct xfs_mount *mp) | |
1da177e4 | 997 | { |
f661f1e0 | 998 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
999 | xfs_log_force(mp, XFS_LOG_SYNC); |
1000 | ||
1001 | /* | |
1002 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
1003 | * will push it, xfs_wait_buftarg() will not wait for it. Further, | |
1004 | * xfs_buf_iowait() cannot be used because it was pushed with the | |
1005 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
1006 | * the IO to complete. | |
1007 | */ | |
1008 | xfs_ail_push_all_sync(mp->m_ail); | |
1009 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1010 | xfs_buf_lock(mp->m_sb_bp); | |
1011 | xfs_buf_unlock(mp->m_sb_bp); | |
1012 | ||
1013 | xfs_log_unmount_write(mp); | |
c75921a7 DC |
1014 | } |
1015 | ||
1016 | /* | |
1017 | * Shut down and release the AIL and Log. | |
1018 | * | |
1019 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
1020 | * from the AIL so that the log is empty before we write the unmount record to | |
1021 | * the log. Once this is done, we can tear down the AIL and the log. | |
1022 | */ | |
1023 | void | |
1024 | xfs_log_unmount( | |
1025 | struct xfs_mount *mp) | |
1026 | { | |
1027 | xfs_log_quiesce(mp); | |
cf2931db | 1028 | |
249a8c11 | 1029 | xfs_trans_ail_destroy(mp); |
baff4e44 BF |
1030 | |
1031 | xfs_sysfs_del(&mp->m_log->l_kobj); | |
1032 | ||
c41564b5 | 1033 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
1034 | } |
1035 | ||
43f5efc5 DC |
1036 | void |
1037 | xfs_log_item_init( | |
1038 | struct xfs_mount *mp, | |
1039 | struct xfs_log_item *item, | |
1040 | int type, | |
272e42b2 | 1041 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
1042 | { |
1043 | item->li_mountp = mp; | |
1044 | item->li_ailp = mp->m_ail; | |
1045 | item->li_type = type; | |
1046 | item->li_ops = ops; | |
71e330b5 DC |
1047 | item->li_lv = NULL; |
1048 | ||
1049 | INIT_LIST_HEAD(&item->li_ail); | |
1050 | INIT_LIST_HEAD(&item->li_cil); | |
643c8c05 | 1051 | INIT_LIST_HEAD(&item->li_bio_list); |
e6631f85 | 1052 | INIT_LIST_HEAD(&item->li_trans); |
43f5efc5 DC |
1053 | } |
1054 | ||
09a423a3 CH |
1055 | /* |
1056 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 1057 | */ |
1da177e4 | 1058 | void |
09a423a3 | 1059 | xfs_log_space_wake( |
cfb7cdca | 1060 | struct xfs_mount *mp) |
1da177e4 | 1061 | { |
ad223e60 | 1062 | struct xlog *log = mp->m_log; |
cfb7cdca | 1063 | int free_bytes; |
1da177e4 | 1064 | |
1da177e4 LT |
1065 | if (XLOG_FORCED_SHUTDOWN(log)) |
1066 | return; | |
1da177e4 | 1067 | |
28496968 | 1068 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
1069 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1070 | ||
28496968 CH |
1071 | spin_lock(&log->l_write_head.lock); |
1072 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 1073 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 1074 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 1075 | } |
10547941 | 1076 | |
28496968 | 1077 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
1078 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1079 | ||
28496968 CH |
1080 | spin_lock(&log->l_reserve_head.lock); |
1081 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 1082 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 1083 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 1084 | } |
3f16b985 | 1085 | } |
1da177e4 LT |
1086 | |
1087 | /* | |
2c6e24ce DC |
1088 | * Determine if we have a transaction that has gone to disk that needs to be |
1089 | * covered. To begin the transition to the idle state firstly the log needs to | |
1090 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1091 | * we start attempting to cover the log. | |
b6f8dd49 | 1092 | * |
2c6e24ce DC |
1093 | * Only if we are then in a state where covering is needed, the caller is |
1094 | * informed that dummy transactions are required to move the log into the idle | |
1095 | * state. | |
1096 | * | |
1097 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1098 | * cover the log as we may be in a situation where there isn't log space | |
1099 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1100 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1101 | * there's no point in running a dummy transaction at this point because we | |
1102 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 | 1103 | */ |
0d5a75e9 | 1104 | static int |
1da177e4 LT |
1105 | xfs_log_need_covered(xfs_mount_t *mp) |
1106 | { | |
9a8d2fdb | 1107 | struct xlog *log = mp->m_log; |
2c6e24ce | 1108 | int needed = 0; |
1da177e4 | 1109 | |
91ee575f | 1110 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
1da177e4 LT |
1111 | return 0; |
1112 | ||
2c6e24ce DC |
1113 | if (!xlog_cil_empty(log)) |
1114 | return 0; | |
1115 | ||
b22cd72c | 1116 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1117 | switch (log->l_covered_state) { |
1118 | case XLOG_STATE_COVER_DONE: | |
1119 | case XLOG_STATE_COVER_DONE2: | |
1120 | case XLOG_STATE_COVER_IDLE: | |
1121 | break; | |
1122 | case XLOG_STATE_COVER_NEED: | |
1123 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1124 | if (xfs_ail_min_lsn(log->l_ailp)) |
1125 | break; | |
1126 | if (!xlog_iclogs_empty(log)) | |
1127 | break; | |
1128 | ||
1129 | needed = 1; | |
1130 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) | |
1131 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1132 | else | |
1133 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1134 | break; | |
b6f8dd49 | 1135 | default: |
1da177e4 | 1136 | needed = 1; |
b6f8dd49 | 1137 | break; |
1da177e4 | 1138 | } |
b22cd72c | 1139 | spin_unlock(&log->l_icloglock); |
014c2544 | 1140 | return needed; |
1da177e4 LT |
1141 | } |
1142 | ||
09a423a3 | 1143 | /* |
1da177e4 LT |
1144 | * We may be holding the log iclog lock upon entering this routine. |
1145 | */ | |
1146 | xfs_lsn_t | |
1c304625 | 1147 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1148 | struct xfs_mount *mp) |
1da177e4 | 1149 | { |
ad223e60 | 1150 | struct xlog *log = mp->m_log; |
1c304625 CH |
1151 | struct xfs_log_item *lip; |
1152 | xfs_lsn_t tail_lsn; | |
1153 | ||
57e80956 | 1154 | assert_spin_locked(&mp->m_ail->ail_lock); |
1da177e4 | 1155 | |
09a423a3 CH |
1156 | /* |
1157 | * To make sure we always have a valid LSN for the log tail we keep | |
1158 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1159 | * and use that when the AIL was empty. |
09a423a3 | 1160 | */ |
1c304625 CH |
1161 | lip = xfs_ail_min(mp->m_ail); |
1162 | if (lip) | |
1163 | tail_lsn = lip->li_lsn; | |
1164 | else | |
84f3c683 | 1165 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
750b9c90 | 1166 | trace_xfs_log_assign_tail_lsn(log, tail_lsn); |
1c3cb9ec | 1167 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1168 | return tail_lsn; |
1c3cb9ec | 1169 | } |
1da177e4 | 1170 | |
1c304625 CH |
1171 | xfs_lsn_t |
1172 | xlog_assign_tail_lsn( | |
1173 | struct xfs_mount *mp) | |
1174 | { | |
1175 | xfs_lsn_t tail_lsn; | |
1176 | ||
57e80956 | 1177 | spin_lock(&mp->m_ail->ail_lock); |
1c304625 | 1178 | tail_lsn = xlog_assign_tail_lsn_locked(mp); |
57e80956 | 1179 | spin_unlock(&mp->m_ail->ail_lock); |
1c304625 CH |
1180 | |
1181 | return tail_lsn; | |
1182 | } | |
1183 | ||
1da177e4 LT |
1184 | /* |
1185 | * Return the space in the log between the tail and the head. The head | |
1186 | * is passed in the cycle/bytes formal parms. In the special case where | |
1187 | * the reserve head has wrapped passed the tail, this calculation is no | |
1188 | * longer valid. In this case, just return 0 which means there is no space | |
1189 | * in the log. This works for all places where this function is called | |
1190 | * with the reserve head. Of course, if the write head were to ever | |
1191 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1192 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1193 | * | |
1194 | * This code also handles the case where the reservation head is behind | |
1195 | * the tail. The details of this case are described below, but the end | |
1196 | * result is that we return the size of the log as the amount of space left. | |
1197 | */ | |
a8272ce0 | 1198 | STATIC int |
a69ed03c | 1199 | xlog_space_left( |
ad223e60 | 1200 | struct xlog *log, |
c8a09ff8 | 1201 | atomic64_t *head) |
1da177e4 | 1202 | { |
a69ed03c DC |
1203 | int free_bytes; |
1204 | int tail_bytes; | |
1205 | int tail_cycle; | |
1206 | int head_cycle; | |
1207 | int head_bytes; | |
1da177e4 | 1208 | |
a69ed03c | 1209 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1210 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1211 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1212 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1213 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1214 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1215 | return 0; |
a69ed03c DC |
1216 | else if (tail_cycle < head_cycle) { |
1217 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1218 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1219 | } else { |
1220 | /* | |
1221 | * The reservation head is behind the tail. | |
1222 | * In this case we just want to return the size of the | |
1223 | * log as the amount of space left. | |
1224 | */ | |
f41febd2 | 1225 | xfs_alert(log->l_mp, "xlog_space_left: head behind tail"); |
a0fa2b67 | 1226 | xfs_alert(log->l_mp, |
f41febd2 JP |
1227 | " tail_cycle = %d, tail_bytes = %d", |
1228 | tail_cycle, tail_bytes); | |
1229 | xfs_alert(log->l_mp, | |
1230 | " GH cycle = %d, GH bytes = %d", | |
1231 | head_cycle, head_bytes); | |
1da177e4 LT |
1232 | ASSERT(0); |
1233 | free_bytes = log->l_logsize; | |
1234 | } | |
1235 | return free_bytes; | |
a69ed03c | 1236 | } |
1da177e4 LT |
1237 | |
1238 | ||
0d5a75e9 | 1239 | static void |
79b54d9b CH |
1240 | xlog_ioend_work( |
1241 | struct work_struct *work) | |
1da177e4 | 1242 | { |
79b54d9b CH |
1243 | struct xlog_in_core *iclog = |
1244 | container_of(work, struct xlog_in_core, ic_end_io_work); | |
1245 | struct xlog *log = iclog->ic_log; | |
d15cbf2f | 1246 | bool aborted = false; |
79b54d9b | 1247 | int error; |
1da177e4 | 1248 | |
79b54d9b | 1249 | error = blk_status_to_errno(iclog->ic_bio.bi_status); |
366fc4b8 CH |
1250 | #ifdef DEBUG |
1251 | /* treat writes with injected CRC errors as failed */ | |
1252 | if (iclog->ic_fail_crc) | |
79b54d9b | 1253 | error = -EIO; |
366fc4b8 CH |
1254 | #endif |
1255 | ||
1da177e4 | 1256 | /* |
366fc4b8 | 1257 | * Race to shutdown the filesystem if we see an error. |
1da177e4 | 1258 | */ |
79b54d9b CH |
1259 | if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) { |
1260 | xfs_alert(log->l_mp, "log I/O error %d", error); | |
1261 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); | |
1da177e4 LT |
1262 | /* |
1263 | * This flag will be propagated to the trans-committed | |
1264 | * callback routines to let them know that the log-commit | |
1265 | * didn't succeed. | |
1266 | */ | |
d15cbf2f | 1267 | aborted = true; |
1da177e4 | 1268 | } else if (iclog->ic_state & XLOG_STATE_IOERROR) { |
d15cbf2f | 1269 | aborted = true; |
1da177e4 | 1270 | } |
3db296f3 | 1271 | |
1da177e4 | 1272 | xlog_state_done_syncing(iclog, aborted); |
79b54d9b | 1273 | bio_uninit(&iclog->ic_bio); |
9c23eccc | 1274 | |
3db296f3 | 1275 | /* |
79b54d9b CH |
1276 | * Drop the lock to signal that we are done. Nothing references the |
1277 | * iclog after this, so an unmount waiting on this lock can now tear it | |
1278 | * down safely. As such, it is unsafe to reference the iclog after the | |
1279 | * unlock as we could race with it being freed. | |
3db296f3 | 1280 | */ |
79b54d9b | 1281 | up(&iclog->ic_sema); |
c3f8fc73 | 1282 | } |
1da177e4 | 1283 | |
1da177e4 LT |
1284 | /* |
1285 | * Return size of each in-core log record buffer. | |
1286 | * | |
9da096fd | 1287 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1288 | * |
1289 | * If the filesystem blocksize is too large, we may need to choose a | |
1290 | * larger size since the directory code currently logs entire blocks. | |
1291 | */ | |
1da177e4 | 1292 | STATIC void |
9a8d2fdb MT |
1293 | xlog_get_iclog_buffer_size( |
1294 | struct xfs_mount *mp, | |
1295 | struct xlog *log) | |
1da177e4 | 1296 | { |
1cb51258 | 1297 | if (mp->m_logbufs <= 0) |
4f62282a CH |
1298 | mp->m_logbufs = XLOG_MAX_ICLOGS; |
1299 | if (mp->m_logbsize <= 0) | |
1300 | mp->m_logbsize = XLOG_BIG_RECORD_BSIZE; | |
1301 | ||
1302 | log->l_iclog_bufs = mp->m_logbufs; | |
1303 | log->l_iclog_size = mp->m_logbsize; | |
1da177e4 LT |
1304 | |
1305 | /* | |
4f62282a | 1306 | * # headers = size / 32k - one header holds cycles from 32k of data. |
1da177e4 | 1307 | */ |
4f62282a CH |
1308 | log->l_iclog_heads = |
1309 | DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE); | |
1310 | log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT; | |
1311 | } | |
1da177e4 | 1312 | |
f661f1e0 DC |
1313 | void |
1314 | xfs_log_work_queue( | |
1315 | struct xfs_mount *mp) | |
1316 | { | |
696a5620 | 1317 | queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1318 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Every sync period we need to unpin all items in the AIL and push them to | |
1323 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1324 | * indicate that the filesystem is idle. | |
1325 | */ | |
0d5a75e9 | 1326 | static void |
f661f1e0 DC |
1327 | xfs_log_worker( |
1328 | struct work_struct *work) | |
1329 | { | |
1330 | struct xlog *log = container_of(to_delayed_work(work), | |
1331 | struct xlog, l_work); | |
1332 | struct xfs_mount *mp = log->l_mp; | |
1333 | ||
1334 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
61e63ecb DC |
1335 | if (xfs_log_need_covered(mp)) { |
1336 | /* | |
1337 | * Dump a transaction into the log that contains no real change. | |
1338 | * This is needed to stamp the current tail LSN into the log | |
1339 | * during the covering operation. | |
1340 | * | |
1341 | * We cannot use an inode here for this - that will push dirty | |
1342 | * state back up into the VFS and then periodic inode flushing | |
1343 | * will prevent log covering from making progress. Hence we | |
1344 | * synchronously log the superblock instead to ensure the | |
1345 | * superblock is immediately unpinned and can be written back. | |
1346 | */ | |
1347 | xfs_sync_sb(mp, true); | |
1348 | } else | |
f661f1e0 DC |
1349 | xfs_log_force(mp, 0); |
1350 | ||
1351 | /* start pushing all the metadata that is currently dirty */ | |
1352 | xfs_ail_push_all(mp->m_ail); | |
1353 | ||
1354 | /* queue us up again */ | |
1355 | xfs_log_work_queue(mp); | |
1356 | } | |
1357 | ||
1da177e4 LT |
1358 | /* |
1359 | * This routine initializes some of the log structure for a given mount point. | |
1360 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1361 | * some other stuff may be filled in too. | |
1362 | */ | |
9a8d2fdb MT |
1363 | STATIC struct xlog * |
1364 | xlog_alloc_log( | |
1365 | struct xfs_mount *mp, | |
1366 | struct xfs_buftarg *log_target, | |
1367 | xfs_daddr_t blk_offset, | |
1368 | int num_bblks) | |
1da177e4 | 1369 | { |
9a8d2fdb | 1370 | struct xlog *log; |
1da177e4 LT |
1371 | xlog_rec_header_t *head; |
1372 | xlog_in_core_t **iclogp; | |
1373 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1da177e4 | 1374 | int i; |
2451337d | 1375 | int error = -ENOMEM; |
69ce58f0 | 1376 | uint log2_size = 0; |
1da177e4 | 1377 | |
9a8d2fdb | 1378 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1379 | if (!log) { |
a0fa2b67 | 1380 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1381 | goto out; |
1382 | } | |
1da177e4 LT |
1383 | |
1384 | log->l_mp = mp; | |
1385 | log->l_targ = log_target; | |
1386 | log->l_logsize = BBTOB(num_bblks); | |
1387 | log->l_logBBstart = blk_offset; | |
1388 | log->l_logBBsize = num_bblks; | |
1389 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1390 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1391 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1392 | |
1393 | log->l_prev_block = -1; | |
1da177e4 | 1394 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1395 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1396 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1397 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 CH |
1398 | |
1399 | xlog_grant_head_init(&log->l_reserve_head); | |
1400 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1401 | |
2451337d | 1402 | error = -EFSCORRUPTED; |
62118709 | 1403 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1404 | log2_size = mp->m_sb.sb_logsectlog; |
1405 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1406 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1407 | log2_size, BBSHIFT); | |
a6cb767e DC |
1408 | goto out_free_log; |
1409 | } | |
1410 | ||
69ce58f0 AE |
1411 | log2_size -= BBSHIFT; |
1412 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1413 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1414 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1415 | goto out_free_log; |
1416 | } | |
69ce58f0 AE |
1417 | |
1418 | /* for larger sector sizes, must have v2 or external log */ | |
1419 | if (log2_size && log->l_logBBstart > 0 && | |
1420 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1421 | xfs_warn(mp, |
1422 | "log sector size (0x%x) invalid for configuration.", | |
1423 | log2_size); | |
a6cb767e DC |
1424 | goto out_free_log; |
1425 | } | |
1da177e4 | 1426 | } |
69ce58f0 | 1427 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1428 | |
1429 | xlog_get_iclog_buffer_size(mp, log); | |
1430 | ||
007c61c6 | 1431 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1432 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1433 | |
1da177e4 LT |
1434 | iclogp = &log->l_iclog; |
1435 | /* | |
1436 | * The amount of memory to allocate for the iclog structure is | |
1437 | * rather funky due to the way the structure is defined. It is | |
1438 | * done this way so that we can use different sizes for machines | |
1439 | * with different amounts of memory. See the definition of | |
1440 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1441 | */ | |
1da177e4 | 1442 | ASSERT(log->l_iclog_size >= 4096); |
79b54d9b CH |
1443 | for (i = 0; i < log->l_iclog_bufs; i++) { |
1444 | size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE); | |
1445 | ||
1446 | iclog = kmem_zalloc(sizeof(*iclog) + bvec_size, KM_MAYFAIL); | |
1447 | if (!iclog) | |
644c3567 DC |
1448 | goto out_free_iclog; |
1449 | ||
79b54d9b | 1450 | *iclogp = iclog; |
1da177e4 LT |
1451 | iclog->ic_prev = prev_iclog; |
1452 | prev_iclog = iclog; | |
1fa40b01 | 1453 | |
79b54d9b CH |
1454 | iclog->ic_data = kmem_alloc_large(log->l_iclog_size, |
1455 | KM_MAYFAIL); | |
1456 | if (!iclog->ic_data) | |
644c3567 | 1457 | goto out_free_iclog; |
4679b2d3 | 1458 | #ifdef DEBUG |
5809d5e0 | 1459 | log->l_iclog_bak[i] = &iclog->ic_header; |
4679b2d3 | 1460 | #endif |
1da177e4 LT |
1461 | head = &iclog->ic_header; |
1462 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1463 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1464 | head->h_version = cpu_to_be32( | |
62118709 | 1465 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1466 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1467 | /* new fields */ |
b53e675d | 1468 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1469 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1470 | ||
79b54d9b | 1471 | iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize; |
1da177e4 LT |
1472 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1473 | iclog->ic_log = log; | |
114d23aa DC |
1474 | atomic_set(&iclog->ic_refcnt, 0); |
1475 | spin_lock_init(&iclog->ic_callback_lock); | |
1da177e4 | 1476 | iclog->ic_callback_tail = &(iclog->ic_callback); |
b28708d6 | 1477 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1478 | |
eb40a875 DC |
1479 | init_waitqueue_head(&iclog->ic_force_wait); |
1480 | init_waitqueue_head(&iclog->ic_write_wait); | |
79b54d9b CH |
1481 | INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work); |
1482 | sema_init(&iclog->ic_sema, 1); | |
1da177e4 LT |
1483 | |
1484 | iclogp = &iclog->ic_next; | |
1485 | } | |
1486 | *iclogp = log->l_iclog; /* complete ring */ | |
1487 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1488 | ||
1058d0f5 CH |
1489 | log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s", |
1490 | WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_HIGHPRI, 0, | |
1491 | mp->m_fsname); | |
1492 | if (!log->l_ioend_workqueue) | |
1493 | goto out_free_iclog; | |
1494 | ||
71e330b5 DC |
1495 | error = xlog_cil_init(log); |
1496 | if (error) | |
1058d0f5 | 1497 | goto out_destroy_workqueue; |
1da177e4 | 1498 | return log; |
644c3567 | 1499 | |
1058d0f5 CH |
1500 | out_destroy_workqueue: |
1501 | destroy_workqueue(log->l_ioend_workqueue); | |
644c3567 DC |
1502 | out_free_iclog: |
1503 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1504 | prev_iclog = iclog->ic_next; | |
79b54d9b | 1505 | kmem_free(iclog->ic_data); |
644c3567 DC |
1506 | kmem_free(iclog); |
1507 | } | |
644c3567 DC |
1508 | out_free_log: |
1509 | kmem_free(log); | |
a6cb767e | 1510 | out: |
2451337d | 1511 | return ERR_PTR(error); |
1da177e4 LT |
1512 | } /* xlog_alloc_log */ |
1513 | ||
1514 | ||
1515 | /* | |
1516 | * Write out the commit record of a transaction associated with the given | |
1517 | * ticket. Return the lsn of the commit record. | |
1518 | */ | |
1519 | STATIC int | |
55b66332 | 1520 | xlog_commit_record( |
ad223e60 | 1521 | struct xlog *log, |
55b66332 DC |
1522 | struct xlog_ticket *ticket, |
1523 | struct xlog_in_core **iclog, | |
1524 | xfs_lsn_t *commitlsnp) | |
1da177e4 | 1525 | { |
55b66332 DC |
1526 | struct xfs_mount *mp = log->l_mp; |
1527 | int error; | |
1528 | struct xfs_log_iovec reg = { | |
1529 | .i_addr = NULL, | |
1530 | .i_len = 0, | |
1531 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1532 | }; | |
1533 | struct xfs_log_vec vec = { | |
1534 | .lv_niovecs = 1, | |
1535 | .lv_iovecp = ®, | |
1536 | }; | |
1da177e4 LT |
1537 | |
1538 | ASSERT_ALWAYS(iclog); | |
55b66332 DC |
1539 | error = xlog_write(log, &vec, ticket, commitlsnp, iclog, |
1540 | XLOG_COMMIT_TRANS); | |
1541 | if (error) | |
7d04a335 | 1542 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1543 | return error; |
55b66332 | 1544 | } |
1da177e4 LT |
1545 | |
1546 | /* | |
1547 | * Push on the buffer cache code if we ever use more than 75% of the on-disk | |
1548 | * log space. This code pushes on the lsn which would supposedly free up | |
1549 | * the 25% which we want to leave free. We may need to adopt a policy which | |
1550 | * pushes on an lsn which is further along in the log once we reach the high | |
1551 | * water mark. In this manner, we would be creating a low water mark. | |
1552 | */ | |
a8272ce0 | 1553 | STATIC void |
2ced19cb | 1554 | xlog_grant_push_ail( |
ad223e60 | 1555 | struct xlog *log, |
2ced19cb | 1556 | int need_bytes) |
1da177e4 | 1557 | { |
2ced19cb | 1558 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1559 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1560 | int free_blocks; |
1561 | int free_bytes; | |
1562 | int threshold_block; | |
1563 | int threshold_cycle; | |
1564 | int free_threshold; | |
1565 | ||
1566 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1567 | ||
28496968 | 1568 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1569 | free_blocks = BTOBBT(free_bytes); |
1570 | ||
1571 | /* | |
1572 | * Set the threshold for the minimum number of free blocks in the | |
1573 | * log to the maximum of what the caller needs, one quarter of the | |
1574 | * log, and 256 blocks. | |
1575 | */ | |
1576 | free_threshold = BTOBB(need_bytes); | |
9bb54cb5 DC |
1577 | free_threshold = max(free_threshold, (log->l_logBBsize >> 2)); |
1578 | free_threshold = max(free_threshold, 256); | |
2ced19cb DC |
1579 | if (free_blocks >= free_threshold) |
1580 | return; | |
1581 | ||
1c3cb9ec DC |
1582 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1583 | &threshold_block); | |
1584 | threshold_block += free_threshold; | |
1da177e4 | 1585 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1586 | threshold_block -= log->l_logBBsize; |
1587 | threshold_cycle += 1; | |
1da177e4 | 1588 | } |
2ced19cb DC |
1589 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1590 | threshold_block); | |
1591 | /* | |
1592 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1593 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1594 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1595 | */ |
84f3c683 DC |
1596 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1597 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1598 | threshold_lsn = last_sync_lsn; | |
2ced19cb DC |
1599 | |
1600 | /* | |
1601 | * Get the transaction layer to kick the dirty buffers out to | |
1602 | * disk asynchronously. No point in trying to do this if | |
1603 | * the filesystem is shutting down. | |
1604 | */ | |
1605 | if (!XLOG_FORCED_SHUTDOWN(log)) | |
fd074841 | 1606 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1607 | } |
1da177e4 | 1608 | |
0e446be4 CH |
1609 | /* |
1610 | * Stamp cycle number in every block | |
1611 | */ | |
1612 | STATIC void | |
1613 | xlog_pack_data( | |
1614 | struct xlog *log, | |
1615 | struct xlog_in_core *iclog, | |
1616 | int roundoff) | |
1617 | { | |
1618 | int i, j, k; | |
1619 | int size = iclog->ic_offset + roundoff; | |
1620 | __be32 cycle_lsn; | |
b2a922cd | 1621 | char *dp; |
0e446be4 CH |
1622 | |
1623 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1624 | ||
1625 | dp = iclog->ic_datap; | |
1626 | for (i = 0; i < BTOBB(size); i++) { | |
1627 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1628 | break; | |
1629 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1630 | *(__be32 *)dp = cycle_lsn; | |
1631 | dp += BBSIZE; | |
1632 | } | |
1633 | ||
1634 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1635 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1636 | ||
1637 | for ( ; i < BTOBB(size); i++) { | |
1638 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1639 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1640 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1641 | *(__be32 *)dp = cycle_lsn; | |
1642 | dp += BBSIZE; | |
1643 | } | |
1644 | ||
1645 | for (i = 1; i < log->l_iclog_heads; i++) | |
1646 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1647 | } | |
1648 | } | |
1649 | ||
1650 | /* | |
1651 | * Calculate the checksum for a log buffer. | |
1652 | * | |
1653 | * This is a little more complicated than it should be because the various | |
1654 | * headers and the actual data are non-contiguous. | |
1655 | */ | |
f9668a09 | 1656 | __le32 |
0e446be4 CH |
1657 | xlog_cksum( |
1658 | struct xlog *log, | |
1659 | struct xlog_rec_header *rhead, | |
1660 | char *dp, | |
1661 | int size) | |
1662 | { | |
c8ce540d | 1663 | uint32_t crc; |
0e446be4 CH |
1664 | |
1665 | /* first generate the crc for the record header ... */ | |
cae028df | 1666 | crc = xfs_start_cksum_update((char *)rhead, |
0e446be4 CH |
1667 | sizeof(struct xlog_rec_header), |
1668 | offsetof(struct xlog_rec_header, h_crc)); | |
1669 | ||
1670 | /* ... then for additional cycle data for v2 logs ... */ | |
1671 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1672 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1673 | int i; | |
a3f20014 | 1674 | int xheads; |
0e446be4 | 1675 | |
a3f20014 BF |
1676 | xheads = size / XLOG_HEADER_CYCLE_SIZE; |
1677 | if (size % XLOG_HEADER_CYCLE_SIZE) | |
1678 | xheads++; | |
0e446be4 | 1679 | |
a3f20014 | 1680 | for (i = 1; i < xheads; i++) { |
0e446be4 CH |
1681 | crc = crc32c(crc, &xhdr[i].hic_xheader, |
1682 | sizeof(struct xlog_rec_ext_header)); | |
1683 | } | |
1684 | } | |
1685 | ||
1686 | /* ... and finally for the payload */ | |
1687 | crc = crc32c(crc, dp, size); | |
1688 | ||
1689 | return xfs_end_cksum(crc); | |
1690 | } | |
1691 | ||
79b54d9b CH |
1692 | static void |
1693 | xlog_bio_end_io( | |
1694 | struct bio *bio) | |
1695 | { | |
1696 | struct xlog_in_core *iclog = bio->bi_private; | |
1697 | ||
1058d0f5 | 1698 | queue_work(iclog->ic_log->l_ioend_workqueue, |
79b54d9b CH |
1699 | &iclog->ic_end_io_work); |
1700 | } | |
1701 | ||
1702 | static void | |
1703 | xlog_map_iclog_data( | |
1704 | struct bio *bio, | |
1705 | void *data, | |
1706 | size_t count) | |
1707 | { | |
1708 | do { | |
1709 | struct page *page = kmem_to_page(data); | |
1710 | unsigned int off = offset_in_page(data); | |
1711 | size_t len = min_t(size_t, count, PAGE_SIZE - off); | |
1712 | ||
1713 | WARN_ON_ONCE(bio_add_page(bio, page, len, off) != len); | |
1714 | ||
1715 | data += len; | |
1716 | count -= len; | |
1717 | } while (count); | |
1718 | } | |
1719 | ||
94860a30 CH |
1720 | STATIC void |
1721 | xlog_write_iclog( | |
1722 | struct xlog *log, | |
1723 | struct xlog_in_core *iclog, | |
94860a30 | 1724 | uint64_t bno, |
79b54d9b | 1725 | unsigned int count, |
94860a30 | 1726 | bool need_flush) |
873ff550 | 1727 | { |
94860a30 | 1728 | ASSERT(bno < log->l_logBBsize); |
94860a30 CH |
1729 | |
1730 | /* | |
1731 | * We lock the iclogbufs here so that we can serialise against I/O | |
1732 | * completion during unmount. We might be processing a shutdown | |
1733 | * triggered during unmount, and that can occur asynchronously to the | |
1734 | * unmount thread, and hence we need to ensure that completes before | |
1735 | * tearing down the iclogbufs. Hence we need to hold the buffer lock | |
1736 | * across the log IO to archieve that. | |
1737 | */ | |
79b54d9b | 1738 | down(&iclog->ic_sema); |
94860a30 | 1739 | if (unlikely(iclog->ic_state & XLOG_STATE_IOERROR)) { |
873ff550 CH |
1740 | /* |
1741 | * It would seem logical to return EIO here, but we rely on | |
1742 | * the log state machine to propagate I/O errors instead of | |
79b54d9b CH |
1743 | * doing it here. We kick of the state machine and unlock |
1744 | * the buffer manually, the code needs to be kept in sync | |
1745 | * with the I/O completion path. | |
873ff550 | 1746 | */ |
79b54d9b CH |
1747 | xlog_state_done_syncing(iclog, XFS_LI_ABORTED); |
1748 | up(&iclog->ic_sema); | |
94860a30 | 1749 | return; |
873ff550 CH |
1750 | } |
1751 | ||
79b54d9b CH |
1752 | iclog->ic_io_size = count; |
1753 | ||
1754 | bio_init(&iclog->ic_bio, iclog->ic_bvec, howmany(count, PAGE_SIZE)); | |
1755 | bio_set_dev(&iclog->ic_bio, log->l_targ->bt_bdev); | |
1756 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno; | |
1757 | iclog->ic_bio.bi_end_io = xlog_bio_end_io; | |
1758 | iclog->ic_bio.bi_private = iclog; | |
1759 | iclog->ic_bio.bi_opf = REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_FUA; | |
1760 | if (need_flush) | |
1761 | iclog->ic_bio.bi_opf |= REQ_PREFLUSH; | |
1762 | ||
1763 | xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, iclog->ic_io_size); | |
1764 | if (is_vmalloc_addr(iclog->ic_data)) | |
1765 | flush_kernel_vmap_range(iclog->ic_data, iclog->ic_io_size); | |
1766 | ||
1767 | /* | |
1768 | * If this log buffer would straddle the end of the log we will have | |
1769 | * to split it up into two bios, so that we can continue at the start. | |
1770 | */ | |
1771 | if (bno + BTOBB(count) > log->l_logBBsize) { | |
1772 | struct bio *split; | |
1773 | ||
1774 | split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno, | |
1775 | GFP_NOIO, &fs_bio_set); | |
1776 | bio_chain(split, &iclog->ic_bio); | |
1777 | submit_bio(split); | |
1778 | ||
1779 | /* restart at logical offset zero for the remainder */ | |
1780 | iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart; | |
1781 | } | |
1782 | ||
1783 | submit_bio(&iclog->ic_bio); | |
873ff550 | 1784 | } |
1da177e4 | 1785 | |
56933848 CH |
1786 | /* |
1787 | * We need to bump cycle number for the part of the iclog that is | |
1788 | * written to the start of the log. Watch out for the header magic | |
1789 | * number case, though. | |
1790 | */ | |
79b54d9b | 1791 | static void |
56933848 CH |
1792 | xlog_split_iclog( |
1793 | struct xlog *log, | |
1794 | void *data, | |
1795 | uint64_t bno, | |
1796 | unsigned int count) | |
1797 | { | |
1798 | unsigned int split_offset = BBTOB(log->l_logBBsize - bno); | |
1799 | unsigned int i; | |
1800 | ||
1801 | for (i = split_offset; i < count; i += BBSIZE) { | |
1802 | uint32_t cycle = get_unaligned_be32(data + i); | |
1803 | ||
1804 | if (++cycle == XLOG_HEADER_MAGIC_NUM) | |
1805 | cycle++; | |
1806 | put_unaligned_be32(cycle, data + i); | |
1807 | } | |
56933848 CH |
1808 | } |
1809 | ||
db0a6faf CH |
1810 | static int |
1811 | xlog_calc_iclog_size( | |
1812 | struct xlog *log, | |
1813 | struct xlog_in_core *iclog, | |
1814 | uint32_t *roundoff) | |
1815 | { | |
1816 | uint32_t count_init, count; | |
1817 | bool use_lsunit; | |
1818 | ||
1819 | use_lsunit = xfs_sb_version_haslogv2(&log->l_mp->m_sb) && | |
1820 | log->l_mp->m_sb.sb_logsunit > 1; | |
1821 | ||
1822 | /* Add for LR header */ | |
1823 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
1824 | ||
1825 | /* Round out the log write size */ | |
1826 | if (use_lsunit) { | |
1827 | /* we have a v2 stripe unit to use */ | |
1828 | count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init)); | |
1829 | } else { | |
1830 | count = BBTOB(BTOBB(count_init)); | |
1831 | } | |
1832 | ||
1833 | ASSERT(count >= count_init); | |
1834 | *roundoff = count - count_init; | |
1835 | ||
1836 | if (use_lsunit) | |
1837 | ASSERT(*roundoff < log->l_mp->m_sb.sb_logsunit); | |
1838 | else | |
1839 | ASSERT(*roundoff < BBTOB(1)); | |
1840 | return count; | |
1841 | } | |
1842 | ||
1da177e4 LT |
1843 | /* |
1844 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1845 | * fashion. Previously, we should have moved the current iclog | |
1846 | * ptr in the log to point to the next available iclog. This allows further | |
1847 | * write to continue while this code syncs out an iclog ready to go. | |
1848 | * Before an in-core log can be written out, the data section must be scanned | |
1849 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1850 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1851 | * cycle count because there in an implicit assumption that drives will | |
1852 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1853 | * we can't have part of a 512 byte block written and part not written. By | |
1854 | * tagging each block, we will know which blocks are valid when recovering | |
1855 | * after an unclean shutdown. | |
1856 | * | |
1857 | * This routine is single threaded on the iclog. No other thread can be in | |
1858 | * this routine with the same iclog. Changing contents of iclog can there- | |
1859 | * fore be done without grabbing the state machine lock. Updating the global | |
1860 | * log will require grabbing the lock though. | |
1861 | * | |
1862 | * The entire log manager uses a logical block numbering scheme. Only | |
94860a30 CH |
1863 | * xlog_write_iclog knows about the fact that the log may not start with |
1864 | * block zero on a given device. | |
1da177e4 | 1865 | */ |
94860a30 | 1866 | STATIC void |
9a8d2fdb MT |
1867 | xlog_sync( |
1868 | struct xlog *log, | |
1869 | struct xlog_in_core *iclog) | |
1da177e4 | 1870 | { |
db0a6faf CH |
1871 | unsigned int count; /* byte count of bwrite */ |
1872 | unsigned int roundoff; /* roundoff to BB or stripe */ | |
1873 | uint64_t bno; | |
db0a6faf | 1874 | unsigned int size; |
79b54d9b | 1875 | bool need_flush = true, split = false; |
1da177e4 | 1876 | |
155cc6b7 | 1877 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 | 1878 | |
db0a6faf | 1879 | count = xlog_calc_iclog_size(log, iclog, &roundoff); |
1da177e4 LT |
1880 | |
1881 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1882 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1883 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1884 | |
1885 | /* put cycle number in every block */ | |
1886 | xlog_pack_data(log, iclog, roundoff); | |
1887 | ||
1888 | /* real byte length */ | |
0e446be4 | 1889 | size = iclog->ic_offset; |
db0a6faf | 1890 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) |
0e446be4 CH |
1891 | size += roundoff; |
1892 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 1893 | |
9b0489c1 | 1894 | XFS_STATS_INC(log->l_mp, xs_log_writes); |
ff6d6af2 | 1895 | XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count)); |
1da177e4 | 1896 | |
94860a30 CH |
1897 | bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)); |
1898 | ||
1da177e4 | 1899 | /* Do we need to split this write into 2 parts? */ |
79b54d9b CH |
1900 | if (bno + BTOBB(count) > log->l_logBBsize) { |
1901 | xlog_split_iclog(log, &iclog->ic_header, bno, count); | |
1902 | split = true; | |
1903 | } | |
0e446be4 CH |
1904 | |
1905 | /* calculcate the checksum */ | |
1906 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
1907 | iclog->ic_datap, size); | |
609adfc2 BF |
1908 | /* |
1909 | * Intentionally corrupt the log record CRC based on the error injection | |
1910 | * frequency, if defined. This facilitates testing log recovery in the | |
1911 | * event of torn writes. Hence, set the IOABORT state to abort the log | |
1912 | * write on I/O completion and shutdown the fs. The subsequent mount | |
1913 | * detects the bad CRC and attempts to recover. | |
1914 | */ | |
366fc4b8 | 1915 | #ifdef DEBUG |
3e88a007 | 1916 | if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) { |
e2a64192 | 1917 | iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA); |
366fc4b8 | 1918 | iclog->ic_fail_crc = true; |
609adfc2 BF |
1919 | xfs_warn(log->l_mp, |
1920 | "Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.", | |
1921 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1922 | } | |
366fc4b8 | 1923 | #endif |
0e446be4 | 1924 | |
2291dab2 DC |
1925 | /* |
1926 | * Flush the data device before flushing the log to make sure all meta | |
1927 | * data written back from the AIL actually made it to disk before | |
1928 | * stamping the new log tail LSN into the log buffer. For an external | |
1929 | * log we need to issue the flush explicitly, and unfortunately | |
1930 | * synchronously here; for an internal log we can simply use the block | |
1931 | * layer state machine for preflushes. | |
1932 | */ | |
2d15d2c0 | 1933 | if (log->l_targ != log->l_mp->m_ddev_targp || split) { |
2291dab2 | 1934 | xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp); |
94860a30 CH |
1935 | need_flush = false; |
1936 | } | |
1da177e4 | 1937 | |
abca1f33 | 1938 | xlog_verify_iclog(log, iclog, count); |
79b54d9b | 1939 | xlog_write_iclog(log, iclog, bno, count, need_flush); |
94860a30 | 1940 | } |
1da177e4 | 1941 | |
1da177e4 | 1942 | /* |
c41564b5 | 1943 | * Deallocate a log structure |
1da177e4 | 1944 | */ |
a8272ce0 | 1945 | STATIC void |
9a8d2fdb MT |
1946 | xlog_dealloc_log( |
1947 | struct xlog *log) | |
1da177e4 LT |
1948 | { |
1949 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1950 | int i; |
1951 | ||
71e330b5 DC |
1952 | xlog_cil_destroy(log); |
1953 | ||
44396476 | 1954 | /* |
9c23eccc DC |
1955 | * Cycle all the iclogbuf locks to make sure all log IO completion |
1956 | * is done before we tear down these buffers. | |
1957 | */ | |
1958 | iclog = log->l_iclog; | |
1959 | for (i = 0; i < log->l_iclog_bufs; i++) { | |
79b54d9b CH |
1960 | down(&iclog->ic_sema); |
1961 | up(&iclog->ic_sema); | |
9c23eccc DC |
1962 | iclog = iclog->ic_next; |
1963 | } | |
1964 | ||
1da177e4 | 1965 | iclog = log->l_iclog; |
9c23eccc | 1966 | for (i = 0; i < log->l_iclog_bufs; i++) { |
1da177e4 | 1967 | next_iclog = iclog->ic_next; |
79b54d9b | 1968 | kmem_free(iclog->ic_data); |
f0e2d93c | 1969 | kmem_free(iclog); |
1da177e4 LT |
1970 | iclog = next_iclog; |
1971 | } | |
1da177e4 | 1972 | |
1da177e4 | 1973 | log->l_mp->m_log = NULL; |
1058d0f5 | 1974 | destroy_workqueue(log->l_ioend_workqueue); |
f0e2d93c | 1975 | kmem_free(log); |
c41564b5 | 1976 | } /* xlog_dealloc_log */ |
1da177e4 LT |
1977 | |
1978 | /* | |
1979 | * Update counters atomically now that memcpy is done. | |
1980 | */ | |
1981 | /* ARGSUSED */ | |
1982 | static inline void | |
9a8d2fdb MT |
1983 | xlog_state_finish_copy( |
1984 | struct xlog *log, | |
1985 | struct xlog_in_core *iclog, | |
1986 | int record_cnt, | |
1987 | int copy_bytes) | |
1da177e4 | 1988 | { |
b22cd72c | 1989 | spin_lock(&log->l_icloglock); |
1da177e4 | 1990 | |
413d57c9 | 1991 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 LT |
1992 | iclog->ic_offset += copy_bytes; |
1993 | ||
b22cd72c | 1994 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
1995 | } /* xlog_state_finish_copy */ |
1996 | ||
1997 | ||
1998 | ||
1999 | ||
7e9c6396 TS |
2000 | /* |
2001 | * print out info relating to regions written which consume | |
2002 | * the reservation | |
2003 | */ | |
71e330b5 DC |
2004 | void |
2005 | xlog_print_tic_res( | |
2006 | struct xfs_mount *mp, | |
2007 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
2008 | { |
2009 | uint i; | |
2010 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
2011 | ||
2012 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
5110cd82 | 2013 | #define REG_TYPE_STR(type, str) [XLOG_REG_TYPE_##type] = str |
d31d7185 | 2014 | static char *res_type_str[] = { |
5110cd82 DW |
2015 | REG_TYPE_STR(BFORMAT, "bformat"), |
2016 | REG_TYPE_STR(BCHUNK, "bchunk"), | |
2017 | REG_TYPE_STR(EFI_FORMAT, "efi_format"), | |
2018 | REG_TYPE_STR(EFD_FORMAT, "efd_format"), | |
2019 | REG_TYPE_STR(IFORMAT, "iformat"), | |
2020 | REG_TYPE_STR(ICORE, "icore"), | |
2021 | REG_TYPE_STR(IEXT, "iext"), | |
2022 | REG_TYPE_STR(IBROOT, "ibroot"), | |
2023 | REG_TYPE_STR(ILOCAL, "ilocal"), | |
2024 | REG_TYPE_STR(IATTR_EXT, "iattr_ext"), | |
2025 | REG_TYPE_STR(IATTR_BROOT, "iattr_broot"), | |
2026 | REG_TYPE_STR(IATTR_LOCAL, "iattr_local"), | |
2027 | REG_TYPE_STR(QFORMAT, "qformat"), | |
2028 | REG_TYPE_STR(DQUOT, "dquot"), | |
2029 | REG_TYPE_STR(QUOTAOFF, "quotaoff"), | |
2030 | REG_TYPE_STR(LRHEADER, "LR header"), | |
2031 | REG_TYPE_STR(UNMOUNT, "unmount"), | |
2032 | REG_TYPE_STR(COMMIT, "commit"), | |
2033 | REG_TYPE_STR(TRANSHDR, "trans header"), | |
d31d7185 DW |
2034 | REG_TYPE_STR(ICREATE, "inode create"), |
2035 | REG_TYPE_STR(RUI_FORMAT, "rui_format"), | |
2036 | REG_TYPE_STR(RUD_FORMAT, "rud_format"), | |
2037 | REG_TYPE_STR(CUI_FORMAT, "cui_format"), | |
2038 | REG_TYPE_STR(CUD_FORMAT, "cud_format"), | |
2039 | REG_TYPE_STR(BUI_FORMAT, "bui_format"), | |
2040 | REG_TYPE_STR(BUD_FORMAT, "bud_format"), | |
7e9c6396 | 2041 | }; |
d31d7185 | 2042 | BUILD_BUG_ON(ARRAY_SIZE(res_type_str) != XLOG_REG_TYPE_MAX + 1); |
5110cd82 | 2043 | #undef REG_TYPE_STR |
7e9c6396 | 2044 | |
7d2d5653 | 2045 | xfs_warn(mp, "ticket reservation summary:"); |
f41febd2 JP |
2046 | xfs_warn(mp, " unit res = %d bytes", |
2047 | ticket->t_unit_res); | |
2048 | xfs_warn(mp, " current res = %d bytes", | |
2049 | ticket->t_curr_res); | |
2050 | xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)", | |
2051 | ticket->t_res_arr_sum, ticket->t_res_o_flow); | |
2052 | xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)", | |
2053 | ticket->t_res_num_ophdrs, ophdr_spc); | |
2054 | xfs_warn(mp, " ophdr + reg = %u bytes", | |
2055 | ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc); | |
2056 | xfs_warn(mp, " num regions = %u", | |
2057 | ticket->t_res_num); | |
7e9c6396 TS |
2058 | |
2059 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 2060 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 2061 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 | 2062 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
5110cd82 | 2063 | "bad-rtype" : res_type_str[r_type]), |
7e9c6396 TS |
2064 | ticket->t_res_arr[i].r_len); |
2065 | } | |
2066 | } | |
7e9c6396 | 2067 | |
d4ca1d55 BF |
2068 | /* |
2069 | * Print a summary of the transaction. | |
2070 | */ | |
2071 | void | |
2072 | xlog_print_trans( | |
e6631f85 | 2073 | struct xfs_trans *tp) |
d4ca1d55 | 2074 | { |
e6631f85 DC |
2075 | struct xfs_mount *mp = tp->t_mountp; |
2076 | struct xfs_log_item *lip; | |
d4ca1d55 BF |
2077 | |
2078 | /* dump core transaction and ticket info */ | |
2079 | xfs_warn(mp, "transaction summary:"); | |
2c8f6265 BF |
2080 | xfs_warn(mp, " log res = %d", tp->t_log_res); |
2081 | xfs_warn(mp, " log count = %d", tp->t_log_count); | |
2082 | xfs_warn(mp, " flags = 0x%x", tp->t_flags); | |
d4ca1d55 BF |
2083 | |
2084 | xlog_print_tic_res(mp, tp->t_ticket); | |
2085 | ||
2086 | /* dump each log item */ | |
e6631f85 | 2087 | list_for_each_entry(lip, &tp->t_items, li_trans) { |
d4ca1d55 BF |
2088 | struct xfs_log_vec *lv = lip->li_lv; |
2089 | struct xfs_log_iovec *vec; | |
2090 | int i; | |
2091 | ||
2092 | xfs_warn(mp, "log item: "); | |
2093 | xfs_warn(mp, " type = 0x%x", lip->li_type); | |
22525c17 | 2094 | xfs_warn(mp, " flags = 0x%lx", lip->li_flags); |
d4ca1d55 BF |
2095 | if (!lv) |
2096 | continue; | |
2097 | xfs_warn(mp, " niovecs = %d", lv->lv_niovecs); | |
2098 | xfs_warn(mp, " size = %d", lv->lv_size); | |
2099 | xfs_warn(mp, " bytes = %d", lv->lv_bytes); | |
2100 | xfs_warn(mp, " buf len = %d", lv->lv_buf_len); | |
2101 | ||
2102 | /* dump each iovec for the log item */ | |
2103 | vec = lv->lv_iovecp; | |
2104 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2105 | int dumplen = min(vec->i_len, 32); | |
2106 | ||
2107 | xfs_warn(mp, " iovec[%d]", i); | |
2108 | xfs_warn(mp, " type = 0x%x", vec->i_type); | |
2109 | xfs_warn(mp, " len = %d", vec->i_len); | |
2110 | xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i); | |
244e3dea | 2111 | xfs_hex_dump(vec->i_addr, dumplen); |
d4ca1d55 BF |
2112 | |
2113 | vec++; | |
2114 | } | |
2115 | } | |
2116 | } | |
2117 | ||
b5203cd0 DC |
2118 | /* |
2119 | * Calculate the potential space needed by the log vector. Each region gets | |
2120 | * its own xlog_op_header_t and may need to be double word aligned. | |
2121 | */ | |
2122 | static int | |
2123 | xlog_write_calc_vec_length( | |
2124 | struct xlog_ticket *ticket, | |
55b66332 | 2125 | struct xfs_log_vec *log_vector) |
b5203cd0 | 2126 | { |
55b66332 | 2127 | struct xfs_log_vec *lv; |
b5203cd0 DC |
2128 | int headers = 0; |
2129 | int len = 0; | |
2130 | int i; | |
2131 | ||
2132 | /* acct for start rec of xact */ | |
2133 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2134 | headers++; | |
2135 | ||
55b66332 | 2136 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2137 | /* we don't write ordered log vectors */ |
2138 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2139 | continue; | |
2140 | ||
55b66332 DC |
2141 | headers += lv->lv_niovecs; |
2142 | ||
2143 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2144 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2145 | |
55b66332 DC |
2146 | len += vecp->i_len; |
2147 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2148 | } | |
b5203cd0 DC |
2149 | } |
2150 | ||
2151 | ticket->t_res_num_ophdrs += headers; | |
2152 | len += headers * sizeof(struct xlog_op_header); | |
2153 | ||
2154 | return len; | |
2155 | } | |
2156 | ||
2157 | /* | |
2158 | * If first write for transaction, insert start record We can't be trying to | |
2159 | * commit if we are inited. We can't have any "partial_copy" if we are inited. | |
2160 | */ | |
2161 | static int | |
2162 | xlog_write_start_rec( | |
e6b1f273 | 2163 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2164 | struct xlog_ticket *ticket) |
2165 | { | |
b5203cd0 DC |
2166 | if (!(ticket->t_flags & XLOG_TIC_INITED)) |
2167 | return 0; | |
2168 | ||
2169 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); | |
2170 | ophdr->oh_clientid = ticket->t_clientid; | |
2171 | ophdr->oh_len = 0; | |
2172 | ophdr->oh_flags = XLOG_START_TRANS; | |
2173 | ophdr->oh_res2 = 0; | |
2174 | ||
2175 | ticket->t_flags &= ~XLOG_TIC_INITED; | |
2176 | ||
2177 | return sizeof(struct xlog_op_header); | |
2178 | } | |
2179 | ||
2180 | static xlog_op_header_t * | |
2181 | xlog_write_setup_ophdr( | |
ad223e60 | 2182 | struct xlog *log, |
e6b1f273 | 2183 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2184 | struct xlog_ticket *ticket, |
2185 | uint flags) | |
2186 | { | |
b5203cd0 DC |
2187 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2188 | ophdr->oh_clientid = ticket->t_clientid; | |
2189 | ophdr->oh_res2 = 0; | |
2190 | ||
2191 | /* are we copying a commit or unmount record? */ | |
2192 | ophdr->oh_flags = flags; | |
2193 | ||
2194 | /* | |
2195 | * We've seen logs corrupted with bad transaction client ids. This | |
2196 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2197 | * and shut down the filesystem. | |
2198 | */ | |
2199 | switch (ophdr->oh_clientid) { | |
2200 | case XFS_TRANSACTION: | |
2201 | case XFS_VOLUME: | |
2202 | case XFS_LOG: | |
2203 | break; | |
2204 | default: | |
a0fa2b67 | 2205 | xfs_warn(log->l_mp, |
c9690043 | 2206 | "Bad XFS transaction clientid 0x%x in ticket "PTR_FMT, |
b5203cd0 DC |
2207 | ophdr->oh_clientid, ticket); |
2208 | return NULL; | |
2209 | } | |
2210 | ||
2211 | return ophdr; | |
2212 | } | |
2213 | ||
2214 | /* | |
2215 | * Set up the parameters of the region copy into the log. This has | |
2216 | * to handle region write split across multiple log buffers - this | |
2217 | * state is kept external to this function so that this code can | |
ac0e300f | 2218 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2219 | */ |
2220 | static int | |
2221 | xlog_write_setup_copy( | |
2222 | struct xlog_ticket *ticket, | |
2223 | struct xlog_op_header *ophdr, | |
2224 | int space_available, | |
2225 | int space_required, | |
2226 | int *copy_off, | |
2227 | int *copy_len, | |
2228 | int *last_was_partial_copy, | |
2229 | int *bytes_consumed) | |
2230 | { | |
2231 | int still_to_copy; | |
2232 | ||
2233 | still_to_copy = space_required - *bytes_consumed; | |
2234 | *copy_off = *bytes_consumed; | |
2235 | ||
2236 | if (still_to_copy <= space_available) { | |
2237 | /* write of region completes here */ | |
2238 | *copy_len = still_to_copy; | |
2239 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2240 | if (*last_was_partial_copy) | |
2241 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2242 | *last_was_partial_copy = 0; | |
2243 | *bytes_consumed = 0; | |
2244 | return 0; | |
2245 | } | |
2246 | ||
2247 | /* partial write of region, needs extra log op header reservation */ | |
2248 | *copy_len = space_available; | |
2249 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2250 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2251 | if (*last_was_partial_copy) | |
2252 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2253 | *bytes_consumed += *copy_len; | |
2254 | (*last_was_partial_copy)++; | |
2255 | ||
2256 | /* account for new log op header */ | |
2257 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2258 | ticket->t_res_num_ophdrs++; | |
2259 | ||
2260 | return sizeof(struct xlog_op_header); | |
2261 | } | |
2262 | ||
2263 | static int | |
2264 | xlog_write_copy_finish( | |
ad223e60 | 2265 | struct xlog *log, |
b5203cd0 DC |
2266 | struct xlog_in_core *iclog, |
2267 | uint flags, | |
2268 | int *record_cnt, | |
2269 | int *data_cnt, | |
2270 | int *partial_copy, | |
2271 | int *partial_copy_len, | |
2272 | int log_offset, | |
2273 | struct xlog_in_core **commit_iclog) | |
2274 | { | |
2275 | if (*partial_copy) { | |
2276 | /* | |
2277 | * This iclog has already been marked WANT_SYNC by | |
2278 | * xlog_state_get_iclog_space. | |
2279 | */ | |
2280 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2281 | *record_cnt = 0; | |
2282 | *data_cnt = 0; | |
2283 | return xlog_state_release_iclog(log, iclog); | |
2284 | } | |
2285 | ||
2286 | *partial_copy = 0; | |
2287 | *partial_copy_len = 0; | |
2288 | ||
2289 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2290 | /* no more space in this iclog - push it. */ | |
2291 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2292 | *record_cnt = 0; | |
2293 | *data_cnt = 0; | |
2294 | ||
2295 | spin_lock(&log->l_icloglock); | |
2296 | xlog_state_want_sync(log, iclog); | |
2297 | spin_unlock(&log->l_icloglock); | |
2298 | ||
2299 | if (!commit_iclog) | |
2300 | return xlog_state_release_iclog(log, iclog); | |
2301 | ASSERT(flags & XLOG_COMMIT_TRANS); | |
2302 | *commit_iclog = iclog; | |
2303 | } | |
2304 | ||
2305 | return 0; | |
2306 | } | |
2307 | ||
1da177e4 LT |
2308 | /* |
2309 | * Write some region out to in-core log | |
2310 | * | |
2311 | * This will be called when writing externally provided regions or when | |
2312 | * writing out a commit record for a given transaction. | |
2313 | * | |
2314 | * General algorithm: | |
2315 | * 1. Find total length of this write. This may include adding to the | |
2316 | * lengths passed in. | |
2317 | * 2. Check whether we violate the tickets reservation. | |
2318 | * 3. While writing to this iclog | |
2319 | * A. Reserve as much space in this iclog as can get | |
2320 | * B. If this is first write, save away start lsn | |
2321 | * C. While writing this region: | |
2322 | * 1. If first write of transaction, write start record | |
2323 | * 2. Write log operation header (header per region) | |
2324 | * 3. Find out if we can fit entire region into this iclog | |
2325 | * 4. Potentially, verify destination memcpy ptr | |
2326 | * 5. Memcpy (partial) region | |
2327 | * 6. If partial copy, release iclog; otherwise, continue | |
2328 | * copying more regions into current iclog | |
2329 | * 4. Mark want sync bit (in simulation mode) | |
2330 | * 5. Release iclog for potential flush to on-disk log. | |
2331 | * | |
2332 | * ERRORS: | |
2333 | * 1. Panic if reservation is overrun. This should never happen since | |
2334 | * reservation amounts are generated internal to the filesystem. | |
2335 | * NOTES: | |
2336 | * 1. Tickets are single threaded data structures. | |
2337 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2338 | * syncing routine. When a single log_write region needs to span | |
2339 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2340 | * on all log operation writes which don't contain the end of the | |
2341 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2342 | * operation which contains the end of the continued log_write region. | |
2343 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2344 | * we don't really know exactly how much space will be used. As a result, | |
2345 | * we don't update ic_offset until the end when we know exactly how many | |
2346 | * bytes have been written out. | |
2347 | */ | |
71e330b5 | 2348 | int |
35a8a72f | 2349 | xlog_write( |
ad223e60 | 2350 | struct xlog *log, |
55b66332 | 2351 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2352 | struct xlog_ticket *ticket, |
2353 | xfs_lsn_t *start_lsn, | |
2354 | struct xlog_in_core **commit_iclog, | |
2355 | uint flags) | |
1da177e4 | 2356 | { |
99428ad0 | 2357 | struct xlog_in_core *iclog = NULL; |
55b66332 DC |
2358 | struct xfs_log_iovec *vecp; |
2359 | struct xfs_log_vec *lv; | |
99428ad0 CH |
2360 | int len; |
2361 | int index; | |
2362 | int partial_copy = 0; | |
2363 | int partial_copy_len = 0; | |
2364 | int contwr = 0; | |
2365 | int record_cnt = 0; | |
2366 | int data_cnt = 0; | |
2367 | int error; | |
2368 | ||
2369 | *start_lsn = 0; | |
2370 | ||
55b66332 | 2371 | len = xlog_write_calc_vec_length(ticket, log_vector); |
71e330b5 | 2372 | |
93b8a585 CH |
2373 | /* |
2374 | * Region headers and bytes are already accounted for. | |
2375 | * We only need to take into account start records and | |
2376 | * split regions in this function. | |
2377 | */ | |
2378 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2379 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
2380 | ||
2381 | /* | |
2382 | * Commit record headers need to be accounted for. These | |
2383 | * come in as separate writes so are easy to detect. | |
2384 | */ | |
2385 | if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS)) | |
2386 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
71e330b5 | 2387 | |
7d2d5653 BF |
2388 | if (ticket->t_curr_res < 0) { |
2389 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
2390 | "ctx ticket reservation ran out. Need to up reservation"); | |
55b66332 | 2391 | xlog_print_tic_res(log->l_mp, ticket); |
7d2d5653 BF |
2392 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
2393 | } | |
1da177e4 | 2394 | |
55b66332 DC |
2395 | index = 0; |
2396 | lv = log_vector; | |
2397 | vecp = lv->lv_iovecp; | |
fd63875c | 2398 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2399 | void *ptr; |
99428ad0 | 2400 | int log_offset; |
1da177e4 | 2401 | |
99428ad0 CH |
2402 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2403 | &contwr, &log_offset); | |
2404 | if (error) | |
2405 | return error; | |
1da177e4 | 2406 | |
99428ad0 | 2407 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2408 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2409 | |
99428ad0 CH |
2410 | /* start_lsn is the first lsn written to. That's all we need. */ |
2411 | if (!*start_lsn) | |
2412 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
b5203cd0 | 2413 | |
99428ad0 CH |
2414 | /* |
2415 | * This loop writes out as many regions as can fit in the amount | |
2416 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2417 | */ | |
fd63875c DC |
2418 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2419 | struct xfs_log_iovec *reg; | |
99428ad0 CH |
2420 | struct xlog_op_header *ophdr; |
2421 | int start_rec_copy; | |
2422 | int copy_len; | |
2423 | int copy_off; | |
fd63875c DC |
2424 | bool ordered = false; |
2425 | ||
2426 | /* ordered log vectors have no regions to write */ | |
2427 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2428 | ASSERT(lv->lv_niovecs == 0); | |
2429 | ordered = true; | |
2430 | goto next_lv; | |
2431 | } | |
99428ad0 | 2432 | |
fd63875c | 2433 | reg = &vecp[index]; |
c8ce540d DW |
2434 | ASSERT(reg->i_len % sizeof(int32_t) == 0); |
2435 | ASSERT((unsigned long)ptr % sizeof(int32_t) == 0); | |
99428ad0 CH |
2436 | |
2437 | start_rec_copy = xlog_write_start_rec(ptr, ticket); | |
2438 | if (start_rec_copy) { | |
2439 | record_cnt++; | |
e6b1f273 | 2440 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2441 | start_rec_copy); |
2442 | } | |
b5203cd0 | 2443 | |
99428ad0 CH |
2444 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags); |
2445 | if (!ophdr) | |
2451337d | 2446 | return -EIO; |
99428ad0 | 2447 | |
e6b1f273 | 2448 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2449 | sizeof(struct xlog_op_header)); |
2450 | ||
2451 | len += xlog_write_setup_copy(ticket, ophdr, | |
2452 | iclog->ic_size-log_offset, | |
55b66332 | 2453 | reg->i_len, |
99428ad0 CH |
2454 | ©_off, ©_len, |
2455 | &partial_copy, | |
2456 | &partial_copy_len); | |
2457 | xlog_verify_dest_ptr(log, ptr); | |
2458 | ||
91f9f5fe ES |
2459 | /* |
2460 | * Copy region. | |
2461 | * | |
2462 | * Unmount records just log an opheader, so can have | |
2463 | * empty payloads with no data region to copy. Hence we | |
2464 | * only copy the payload if the vector says it has data | |
2465 | * to copy. | |
2466 | */ | |
99428ad0 | 2467 | ASSERT(copy_len >= 0); |
91f9f5fe ES |
2468 | if (copy_len > 0) { |
2469 | memcpy(ptr, reg->i_addr + copy_off, copy_len); | |
2470 | xlog_write_adv_cnt(&ptr, &len, &log_offset, | |
2471 | copy_len); | |
2472 | } | |
99428ad0 CH |
2473 | copy_len += start_rec_copy + sizeof(xlog_op_header_t); |
2474 | record_cnt++; | |
2475 | data_cnt += contwr ? copy_len : 0; | |
2476 | ||
2477 | error = xlog_write_copy_finish(log, iclog, flags, | |
2478 | &record_cnt, &data_cnt, | |
2479 | &partial_copy, | |
2480 | &partial_copy_len, | |
2481 | log_offset, | |
2482 | commit_iclog); | |
2483 | if (error) | |
2484 | return error; | |
2485 | ||
2486 | /* | |
2487 | * if we had a partial copy, we need to get more iclog | |
2488 | * space but we don't want to increment the region | |
2489 | * index because there is still more is this region to | |
2490 | * write. | |
2491 | * | |
2492 | * If we completed writing this region, and we flushed | |
2493 | * the iclog (indicated by resetting of the record | |
2494 | * count), then we also need to get more log space. If | |
2495 | * this was the last record, though, we are done and | |
2496 | * can just return. | |
2497 | */ | |
2498 | if (partial_copy) | |
2499 | break; | |
2500 | ||
55b66332 | 2501 | if (++index == lv->lv_niovecs) { |
fd63875c | 2502 | next_lv: |
55b66332 DC |
2503 | lv = lv->lv_next; |
2504 | index = 0; | |
2505 | if (lv) | |
2506 | vecp = lv->lv_iovecp; | |
2507 | } | |
749f24f3 | 2508 | if (record_cnt == 0 && !ordered) { |
55b66332 | 2509 | if (!lv) |
99428ad0 CH |
2510 | return 0; |
2511 | break; | |
2512 | } | |
2513 | } | |
2514 | } | |
2515 | ||
2516 | ASSERT(len == 0); | |
2517 | ||
2518 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); | |
2519 | if (!commit_iclog) | |
2520 | return xlog_state_release_iclog(log, iclog); | |
1da177e4 | 2521 | |
1da177e4 LT |
2522 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2523 | *commit_iclog = iclog; | |
2524 | return 0; | |
99428ad0 | 2525 | } |
1da177e4 LT |
2526 | |
2527 | ||
2528 | /***************************************************************************** | |
2529 | * | |
2530 | * State Machine functions | |
2531 | * | |
2532 | ***************************************************************************** | |
2533 | */ | |
2534 | ||
2535 | /* Clean iclogs starting from the head. This ordering must be | |
2536 | * maintained, so an iclog doesn't become ACTIVE beyond one that | |
2537 | * is SYNCING. This is also required to maintain the notion that we use | |
12017faf | 2538 | * a ordered wait queue to hold off would be writers to the log when every |
1da177e4 LT |
2539 | * iclog is trying to sync to disk. |
2540 | * | |
2541 | * State Change: DIRTY -> ACTIVE | |
2542 | */ | |
ba0f32d4 | 2543 | STATIC void |
9a8d2fdb MT |
2544 | xlog_state_clean_log( |
2545 | struct xlog *log) | |
1da177e4 LT |
2546 | { |
2547 | xlog_in_core_t *iclog; | |
2548 | int changed = 0; | |
2549 | ||
2550 | iclog = log->l_iclog; | |
2551 | do { | |
2552 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
2553 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2554 | iclog->ic_offset = 0; | |
114d23aa | 2555 | ASSERT(iclog->ic_callback == NULL); |
1da177e4 LT |
2556 | /* |
2557 | * If the number of ops in this iclog indicate it just | |
2558 | * contains the dummy transaction, we can | |
2559 | * change state into IDLE (the second time around). | |
2560 | * Otherwise we should change the state into | |
2561 | * NEED a dummy. | |
2562 | * We don't need to cover the dummy. | |
2563 | */ | |
2564 | if (!changed && | |
b53e675d CH |
2565 | (be32_to_cpu(iclog->ic_header.h_num_logops) == |
2566 | XLOG_COVER_OPS)) { | |
1da177e4 LT |
2567 | changed = 1; |
2568 | } else { | |
2569 | /* | |
2570 | * We have two dirty iclogs so start over | |
2571 | * This could also be num of ops indicates | |
2572 | * this is not the dummy going out. | |
2573 | */ | |
2574 | changed = 2; | |
2575 | } | |
2576 | iclog->ic_header.h_num_logops = 0; | |
2577 | memset(iclog->ic_header.h_cycle_data, 0, | |
2578 | sizeof(iclog->ic_header.h_cycle_data)); | |
2579 | iclog->ic_header.h_lsn = 0; | |
2580 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) | |
2581 | /* do nothing */; | |
2582 | else | |
2583 | break; /* stop cleaning */ | |
2584 | iclog = iclog->ic_next; | |
2585 | } while (iclog != log->l_iclog); | |
2586 | ||
2587 | /* log is locked when we are called */ | |
2588 | /* | |
2589 | * Change state for the dummy log recording. | |
2590 | * We usually go to NEED. But we go to NEED2 if the changed indicates | |
2591 | * we are done writing the dummy record. | |
2592 | * If we are done with the second dummy recored (DONE2), then | |
2593 | * we go to IDLE. | |
2594 | */ | |
2595 | if (changed) { | |
2596 | switch (log->l_covered_state) { | |
2597 | case XLOG_STATE_COVER_IDLE: | |
2598 | case XLOG_STATE_COVER_NEED: | |
2599 | case XLOG_STATE_COVER_NEED2: | |
2600 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2601 | break; | |
2602 | ||
2603 | case XLOG_STATE_COVER_DONE: | |
2604 | if (changed == 1) | |
2605 | log->l_covered_state = XLOG_STATE_COVER_NEED2; | |
2606 | else | |
2607 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2608 | break; | |
2609 | ||
2610 | case XLOG_STATE_COVER_DONE2: | |
2611 | if (changed == 1) | |
2612 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
2613 | else | |
2614 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2615 | break; | |
2616 | ||
2617 | default: | |
2618 | ASSERT(0); | |
2619 | } | |
2620 | } | |
2621 | } /* xlog_state_clean_log */ | |
2622 | ||
2623 | STATIC xfs_lsn_t | |
2624 | xlog_get_lowest_lsn( | |
9bff3132 | 2625 | struct xlog *log) |
1da177e4 | 2626 | { |
9bff3132 CH |
2627 | struct xlog_in_core *iclog = log->l_iclog; |
2628 | xfs_lsn_t lowest_lsn = 0, lsn; | |
1da177e4 | 2629 | |
1da177e4 | 2630 | do { |
9bff3132 CH |
2631 | if (iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY)) |
2632 | continue; | |
2633 | ||
2634 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
2635 | if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0) | |
1da177e4 | 2636 | lowest_lsn = lsn; |
9bff3132 CH |
2637 | } while ((iclog = iclog->ic_next) != log->l_iclog); |
2638 | ||
014c2544 | 2639 | return lowest_lsn; |
1da177e4 LT |
2640 | } |
2641 | ||
1da177e4 LT |
2642 | STATIC void |
2643 | xlog_state_do_callback( | |
9a8d2fdb | 2644 | struct xlog *log, |
d15cbf2f | 2645 | bool aborted, |
9a8d2fdb | 2646 | struct xlog_in_core *ciclog) |
1da177e4 LT |
2647 | { |
2648 | xlog_in_core_t *iclog; | |
2649 | xlog_in_core_t *first_iclog; /* used to know when we've | |
2650 | * processed all iclogs once */ | |
2651 | xfs_log_callback_t *cb, *cb_next; | |
2652 | int flushcnt = 0; | |
2653 | xfs_lsn_t lowest_lsn; | |
2654 | int ioerrors; /* counter: iclogs with errors */ | |
2655 | int loopdidcallbacks; /* flag: inner loop did callbacks*/ | |
2656 | int funcdidcallbacks; /* flag: function did callbacks */ | |
2657 | int repeats; /* for issuing console warnings if | |
2658 | * looping too many times */ | |
d748c623 | 2659 | int wake = 0; |
1da177e4 | 2660 | |
b22cd72c | 2661 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2662 | first_iclog = iclog = log->l_iclog; |
2663 | ioerrors = 0; | |
2664 | funcdidcallbacks = 0; | |
2665 | repeats = 0; | |
2666 | ||
2667 | do { | |
2668 | /* | |
2669 | * Scan all iclogs starting with the one pointed to by the | |
2670 | * log. Reset this starting point each time the log is | |
2671 | * unlocked (during callbacks). | |
2672 | * | |
2673 | * Keep looping through iclogs until one full pass is made | |
2674 | * without running any callbacks. | |
2675 | */ | |
2676 | first_iclog = log->l_iclog; | |
2677 | iclog = log->l_iclog; | |
2678 | loopdidcallbacks = 0; | |
2679 | repeats++; | |
2680 | ||
2681 | do { | |
2682 | ||
2683 | /* skip all iclogs in the ACTIVE & DIRTY states */ | |
2684 | if (iclog->ic_state & | |
2685 | (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) { | |
2686 | iclog = iclog->ic_next; | |
2687 | continue; | |
2688 | } | |
2689 | ||
2690 | /* | |
2691 | * Between marking a filesystem SHUTDOWN and stopping | |
2692 | * the log, we do flush all iclogs to disk (if there | |
2693 | * wasn't a log I/O error). So, we do want things to | |
2694 | * go smoothly in case of just a SHUTDOWN w/o a | |
2695 | * LOG_IO_ERROR. | |
2696 | */ | |
2697 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
2698 | /* | |
2699 | * Can only perform callbacks in order. Since | |
2700 | * this iclog is not in the DONE_SYNC/ | |
2701 | * DO_CALLBACK state, we skip the rest and | |
2702 | * just try to clean up. If we set our iclog | |
2703 | * to DO_CALLBACK, we will not process it when | |
2704 | * we retry since a previous iclog is in the | |
2705 | * CALLBACK and the state cannot change since | |
b22cd72c | 2706 | * we are holding the l_icloglock. |
1da177e4 LT |
2707 | */ |
2708 | if (!(iclog->ic_state & | |
2709 | (XLOG_STATE_DONE_SYNC | | |
2710 | XLOG_STATE_DO_CALLBACK))) { | |
2711 | if (ciclog && (ciclog->ic_state == | |
2712 | XLOG_STATE_DONE_SYNC)) { | |
2713 | ciclog->ic_state = XLOG_STATE_DO_CALLBACK; | |
2714 | } | |
2715 | break; | |
2716 | } | |
2717 | /* | |
2718 | * We now have an iclog that is in either the | |
2719 | * DO_CALLBACK or DONE_SYNC states. The other | |
2720 | * states (WANT_SYNC, SYNCING, or CALLBACK were | |
2721 | * caught by the above if and are going to | |
2722 | * clean (i.e. we aren't doing their callbacks) | |
2723 | * see the above if. | |
2724 | */ | |
2725 | ||
2726 | /* | |
2727 | * We will do one more check here to see if we | |
2728 | * have chased our tail around. | |
2729 | */ | |
2730 | ||
2731 | lowest_lsn = xlog_get_lowest_lsn(log); | |
b53e675d CH |
2732 | if (lowest_lsn && |
2733 | XFS_LSN_CMP(lowest_lsn, | |
84f3c683 | 2734 | be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { |
1da177e4 LT |
2735 | iclog = iclog->ic_next; |
2736 | continue; /* Leave this iclog for | |
2737 | * another thread */ | |
2738 | } | |
2739 | ||
2740 | iclog->ic_state = XLOG_STATE_CALLBACK; | |
2741 | ||
1da177e4 | 2742 | |
84f3c683 | 2743 | /* |
d35e88fa DC |
2744 | * Completion of a iclog IO does not imply that |
2745 | * a transaction has completed, as transactions | |
2746 | * can be large enough to span many iclogs. We | |
2747 | * cannot change the tail of the log half way | |
2748 | * through a transaction as this may be the only | |
2749 | * transaction in the log and moving th etail to | |
2750 | * point to the middle of it will prevent | |
2751 | * recovery from finding the start of the | |
2752 | * transaction. Hence we should only update the | |
2753 | * last_sync_lsn if this iclog contains | |
2754 | * transaction completion callbacks on it. | |
2755 | * | |
2756 | * We have to do this before we drop the | |
84f3c683 DC |
2757 | * icloglock to ensure we are the only one that |
2758 | * can update it. | |
1da177e4 | 2759 | */ |
84f3c683 DC |
2760 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), |
2761 | be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); | |
d35e88fa DC |
2762 | if (iclog->ic_callback) |
2763 | atomic64_set(&log->l_last_sync_lsn, | |
2764 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1da177e4 | 2765 | |
84f3c683 | 2766 | } else |
1da177e4 | 2767 | ioerrors++; |
84f3c683 DC |
2768 | |
2769 | spin_unlock(&log->l_icloglock); | |
1da177e4 | 2770 | |
114d23aa DC |
2771 | /* |
2772 | * Keep processing entries in the callback list until | |
2773 | * we come around and it is empty. We need to | |
2774 | * atomically see that the list is empty and change the | |
2775 | * state to DIRTY so that we don't miss any more | |
2776 | * callbacks being added. | |
2777 | */ | |
2778 | spin_lock(&iclog->ic_callback_lock); | |
2779 | cb = iclog->ic_callback; | |
4b80916b | 2780 | while (cb) { |
1da177e4 LT |
2781 | iclog->ic_callback_tail = &(iclog->ic_callback); |
2782 | iclog->ic_callback = NULL; | |
114d23aa | 2783 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2784 | |
2785 | /* perform callbacks in the order given */ | |
4b80916b | 2786 | for (; cb; cb = cb_next) { |
1da177e4 LT |
2787 | cb_next = cb->cb_next; |
2788 | cb->cb_func(cb->cb_arg, aborted); | |
2789 | } | |
114d23aa | 2790 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
2791 | cb = iclog->ic_callback; |
2792 | } | |
2793 | ||
2794 | loopdidcallbacks++; | |
2795 | funcdidcallbacks++; | |
2796 | ||
114d23aa | 2797 | spin_lock(&log->l_icloglock); |
4b80916b | 2798 | ASSERT(iclog->ic_callback == NULL); |
114d23aa | 2799 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2800 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) |
2801 | iclog->ic_state = XLOG_STATE_DIRTY; | |
2802 | ||
2803 | /* | |
2804 | * Transition from DIRTY to ACTIVE if applicable. | |
2805 | * NOP if STATE_IOERROR. | |
2806 | */ | |
2807 | xlog_state_clean_log(log); | |
2808 | ||
2809 | /* wake up threads waiting in xfs_log_force() */ | |
eb40a875 | 2810 | wake_up_all(&iclog->ic_force_wait); |
1da177e4 LT |
2811 | |
2812 | iclog = iclog->ic_next; | |
2813 | } while (first_iclog != iclog); | |
a3c6685e NS |
2814 | |
2815 | if (repeats > 5000) { | |
2816 | flushcnt += repeats; | |
2817 | repeats = 0; | |
a0fa2b67 | 2818 | xfs_warn(log->l_mp, |
a3c6685e | 2819 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2820 | __func__, flushcnt); |
1da177e4 LT |
2821 | } |
2822 | } while (!ioerrors && loopdidcallbacks); | |
2823 | ||
609adfc2 | 2824 | #ifdef DEBUG |
1da177e4 | 2825 | /* |
609adfc2 BF |
2826 | * Make one last gasp attempt to see if iclogs are being left in limbo. |
2827 | * If the above loop finds an iclog earlier than the current iclog and | |
2828 | * in one of the syncing states, the current iclog is put into | |
2829 | * DO_CALLBACK and the callbacks are deferred to the completion of the | |
2830 | * earlier iclog. Walk the iclogs in order and make sure that no iclog | |
2831 | * is in DO_CALLBACK unless an earlier iclog is in one of the syncing | |
2832 | * states. | |
2833 | * | |
2834 | * Note that SYNCING|IOABORT is a valid state so we cannot just check | |
2835 | * for ic_state == SYNCING. | |
1da177e4 | 2836 | */ |
1da177e4 LT |
2837 | if (funcdidcallbacks) { |
2838 | first_iclog = iclog = log->l_iclog; | |
2839 | do { | |
2840 | ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK); | |
2841 | /* | |
2842 | * Terminate the loop if iclogs are found in states | |
2843 | * which will cause other threads to clean up iclogs. | |
2844 | * | |
2845 | * SYNCING - i/o completion will go through logs | |
2846 | * DONE_SYNC - interrupt thread should be waiting for | |
b22cd72c | 2847 | * l_icloglock |
1da177e4 LT |
2848 | * IOERROR - give up hope all ye who enter here |
2849 | */ | |
2850 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
609adfc2 | 2851 | iclog->ic_state & XLOG_STATE_SYNCING || |
1da177e4 LT |
2852 | iclog->ic_state == XLOG_STATE_DONE_SYNC || |
2853 | iclog->ic_state == XLOG_STATE_IOERROR ) | |
2854 | break; | |
2855 | iclog = iclog->ic_next; | |
2856 | } while (first_iclog != iclog); | |
2857 | } | |
2858 | #endif | |
2859 | ||
d748c623 MW |
2860 | if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) |
2861 | wake = 1; | |
b22cd72c | 2862 | spin_unlock(&log->l_icloglock); |
d748c623 MW |
2863 | |
2864 | if (wake) | |
eb40a875 | 2865 | wake_up_all(&log->l_flush_wait); |
d748c623 | 2866 | } |
1da177e4 LT |
2867 | |
2868 | ||
2869 | /* | |
2870 | * Finish transitioning this iclog to the dirty state. | |
2871 | * | |
2872 | * Make sure that we completely execute this routine only when this is | |
2873 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2874 | * when we reach the end of the physical log, get turned into 2 separate | |
2875 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2876 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2877 | * the second completion goes through. | |
2878 | * | |
2879 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2880 | * global state machine log lock. |
1da177e4 | 2881 | */ |
a8272ce0 | 2882 | STATIC void |
1da177e4 | 2883 | xlog_state_done_syncing( |
d15cbf2f CH |
2884 | struct xlog_in_core *iclog, |
2885 | bool aborted) | |
1da177e4 | 2886 | { |
d15cbf2f | 2887 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2888 | |
b22cd72c | 2889 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2890 | |
2891 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING || | |
2892 | iclog->ic_state == XLOG_STATE_IOERROR); | |
155cc6b7 | 2893 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
2894 | |
2895 | /* | |
2896 | * If we got an error, either on the first buffer, or in the case of | |
2897 | * split log writes, on the second, we mark ALL iclogs STATE_IOERROR, | |
2898 | * and none should ever be attempted to be written to disk | |
2899 | * again. | |
2900 | */ | |
79b54d9b | 2901 | if (iclog->ic_state != XLOG_STATE_IOERROR) |
1da177e4 | 2902 | iclog->ic_state = XLOG_STATE_DONE_SYNC; |
1da177e4 LT |
2903 | |
2904 | /* | |
2905 | * Someone could be sleeping prior to writing out the next | |
2906 | * iclog buffer, we wake them all, one will get to do the | |
2907 | * I/O, the others get to wait for the result. | |
2908 | */ | |
eb40a875 | 2909 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2910 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2911 | xlog_state_do_callback(log, aborted, iclog); /* also cleans log */ |
2912 | } /* xlog_state_done_syncing */ | |
2913 | ||
2914 | ||
2915 | /* | |
2916 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2917 | * sleep. We wait on the flush queue on the head iclog as that should be |
2918 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2919 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2920 | * |
2921 | * The in-core logs are used in a circular fashion. They are not used | |
2922 | * out-of-order even when an iclog past the head is free. | |
2923 | * | |
2924 | * return: | |
2925 | * * log_offset where xlog_write() can start writing into the in-core | |
2926 | * log's data space. | |
2927 | * * in-core log pointer to which xlog_write() should write. | |
2928 | * * boolean indicating this is a continued write to an in-core log. | |
2929 | * If this is the last write, then the in-core log's offset field | |
2930 | * needs to be incremented, depending on the amount of data which | |
2931 | * is copied. | |
2932 | */ | |
a8272ce0 | 2933 | STATIC int |
9a8d2fdb MT |
2934 | xlog_state_get_iclog_space( |
2935 | struct xlog *log, | |
2936 | int len, | |
2937 | struct xlog_in_core **iclogp, | |
2938 | struct xlog_ticket *ticket, | |
2939 | int *continued_write, | |
2940 | int *logoffsetp) | |
1da177e4 | 2941 | { |
1da177e4 LT |
2942 | int log_offset; |
2943 | xlog_rec_header_t *head; | |
2944 | xlog_in_core_t *iclog; | |
2945 | int error; | |
2946 | ||
2947 | restart: | |
b22cd72c | 2948 | spin_lock(&log->l_icloglock); |
1da177e4 | 2949 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2950 | spin_unlock(&log->l_icloglock); |
2451337d | 2951 | return -EIO; |
1da177e4 LT |
2952 | } |
2953 | ||
2954 | iclog = log->l_iclog; | |
d748c623 | 2955 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
ff6d6af2 | 2956 | XFS_STATS_INC(log->l_mp, xs_log_noiclogs); |
d748c623 MW |
2957 | |
2958 | /* Wait for log writes to have flushed */ | |
eb40a875 | 2959 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
2960 | goto restart; |
2961 | } | |
d748c623 | 2962 | |
1da177e4 LT |
2963 | head = &iclog->ic_header; |
2964 | ||
155cc6b7 | 2965 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
2966 | log_offset = iclog->ic_offset; |
2967 | ||
2968 | /* On the 1st write to an iclog, figure out lsn. This works | |
2969 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
2970 | * committing to. If the offset is set, that's how many blocks | |
2971 | * must be written. | |
2972 | */ | |
2973 | if (log_offset == 0) { | |
2974 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 2975 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
2976 | log->l_iclog_hsize, |
2977 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
2978 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
2979 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 2980 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
2981 | ASSERT(log->l_curr_block >= 0); |
2982 | } | |
2983 | ||
2984 | /* If there is enough room to write everything, then do it. Otherwise, | |
2985 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
2986 | * bit is on, so this will get flushed out. Don't update ic_offset | |
2987 | * until you know exactly how many bytes get copied. Therefore, wait | |
2988 | * until later to update ic_offset. | |
2989 | * | |
2990 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
2991 | * can fit into remaining data section. | |
2992 | */ | |
2993 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
2994 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2995 | ||
49641f1a DC |
2996 | /* |
2997 | * If I'm the only one writing to this iclog, sync it to disk. | |
2998 | * We need to do an atomic compare and decrement here to avoid | |
2999 | * racing with concurrent atomic_dec_and_lock() calls in | |
3000 | * xlog_state_release_iclog() when there is more than one | |
3001 | * reference to the iclog. | |
3002 | */ | |
3003 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) { | |
3004 | /* we are the only one */ | |
b22cd72c | 3005 | spin_unlock(&log->l_icloglock); |
49641f1a DC |
3006 | error = xlog_state_release_iclog(log, iclog); |
3007 | if (error) | |
014c2544 | 3008 | return error; |
1da177e4 | 3009 | } else { |
b22cd72c | 3010 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3011 | } |
3012 | goto restart; | |
3013 | } | |
3014 | ||
3015 | /* Do we have enough room to write the full amount in the remainder | |
3016 | * of this iclog? Or must we continue a write on the next iclog and | |
3017 | * mark this iclog as completely taken? In the case where we switch | |
3018 | * iclogs (to mark it taken), this particular iclog will release/sync | |
3019 | * to disk in xlog_write(). | |
3020 | */ | |
3021 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
3022 | *continued_write = 0; | |
3023 | iclog->ic_offset += len; | |
3024 | } else { | |
3025 | *continued_write = 1; | |
3026 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
3027 | } | |
3028 | *iclogp = iclog; | |
3029 | ||
3030 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 3031 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3032 | |
3033 | *logoffsetp = log_offset; | |
3034 | return 0; | |
3035 | } /* xlog_state_get_iclog_space */ | |
3036 | ||
1da177e4 LT |
3037 | /* The first cnt-1 times through here we don't need to |
3038 | * move the grant write head because the permanent | |
3039 | * reservation has reserved cnt times the unit amount. | |
3040 | * Release part of current permanent unit reservation and | |
3041 | * reset current reservation to be one units worth. Also | |
3042 | * move grant reservation head forward. | |
3043 | */ | |
3044 | STATIC void | |
9a8d2fdb MT |
3045 | xlog_regrant_reserve_log_space( |
3046 | struct xlog *log, | |
3047 | struct xlog_ticket *ticket) | |
1da177e4 | 3048 | { |
0b1b213f CH |
3049 | trace_xfs_log_regrant_reserve_enter(log, ticket); |
3050 | ||
1da177e4 LT |
3051 | if (ticket->t_cnt > 0) |
3052 | ticket->t_cnt--; | |
3053 | ||
28496968 | 3054 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3055 | ticket->t_curr_res); |
28496968 | 3056 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 3057 | ticket->t_curr_res); |
1da177e4 | 3058 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3059 | xlog_tic_reset_res(ticket); |
0b1b213f CH |
3060 | |
3061 | trace_xfs_log_regrant_reserve_sub(log, ticket); | |
3062 | ||
1da177e4 | 3063 | /* just return if we still have some of the pre-reserved space */ |
d0eb2f38 | 3064 | if (ticket->t_cnt > 0) |
1da177e4 | 3065 | return; |
1da177e4 | 3066 | |
28496968 | 3067 | xlog_grant_add_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3068 | ticket->t_unit_res); |
0b1b213f CH |
3069 | |
3070 | trace_xfs_log_regrant_reserve_exit(log, ticket); | |
3071 | ||
1da177e4 | 3072 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3073 | xlog_tic_reset_res(ticket); |
1da177e4 LT |
3074 | } /* xlog_regrant_reserve_log_space */ |
3075 | ||
3076 | ||
3077 | /* | |
3078 | * Give back the space left from a reservation. | |
3079 | * | |
3080 | * All the information we need to make a correct determination of space left | |
3081 | * is present. For non-permanent reservations, things are quite easy. The | |
3082 | * count should have been decremented to zero. We only need to deal with the | |
3083 | * space remaining in the current reservation part of the ticket. If the | |
3084 | * ticket contains a permanent reservation, there may be left over space which | |
3085 | * needs to be released. A count of N means that N-1 refills of the current | |
3086 | * reservation can be done before we need to ask for more space. The first | |
3087 | * one goes to fill up the first current reservation. Once we run out of | |
3088 | * space, the count will stay at zero and the only space remaining will be | |
3089 | * in the current reservation field. | |
3090 | */ | |
3091 | STATIC void | |
9a8d2fdb MT |
3092 | xlog_ungrant_log_space( |
3093 | struct xlog *log, | |
3094 | struct xlog_ticket *ticket) | |
1da177e4 | 3095 | { |
663e496a DC |
3096 | int bytes; |
3097 | ||
1da177e4 LT |
3098 | if (ticket->t_cnt > 0) |
3099 | ticket->t_cnt--; | |
3100 | ||
0b1b213f | 3101 | trace_xfs_log_ungrant_enter(log, ticket); |
0b1b213f | 3102 | trace_xfs_log_ungrant_sub(log, ticket); |
1da177e4 | 3103 | |
663e496a DC |
3104 | /* |
3105 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
3106 | * up more space based on the remaining count. |
3107 | */ | |
663e496a | 3108 | bytes = ticket->t_curr_res; |
1da177e4 LT |
3109 | if (ticket->t_cnt > 0) { |
3110 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 3111 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
3112 | } |
3113 | ||
28496968 CH |
3114 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
3115 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 3116 | |
0b1b213f CH |
3117 | trace_xfs_log_ungrant_exit(log, ticket); |
3118 | ||
cfb7cdca | 3119 | xfs_log_space_wake(log->l_mp); |
09a423a3 | 3120 | } |
1da177e4 | 3121 | |
1da177e4 LT |
3122 | /* |
3123 | * Flush iclog to disk if this is the last reference to the given iclog and | |
3124 | * the WANT_SYNC bit is set. | |
3125 | * | |
3126 | * When this function is entered, the iclog is not necessarily in the | |
3127 | * WANT_SYNC state. It may be sitting around waiting to get filled. | |
3128 | * | |
3129 | * | |
3130 | */ | |
a8272ce0 | 3131 | STATIC int |
b589334c | 3132 | xlog_state_release_iclog( |
9a8d2fdb MT |
3133 | struct xlog *log, |
3134 | struct xlog_in_core *iclog) | |
1da177e4 | 3135 | { |
1da177e4 LT |
3136 | int sync = 0; /* do we sync? */ |
3137 | ||
155cc6b7 | 3138 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
2451337d | 3139 | return -EIO; |
155cc6b7 DC |
3140 | |
3141 | ASSERT(atomic_read(&iclog->ic_refcnt) > 0); | |
3142 | if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) | |
3143 | return 0; | |
3144 | ||
1da177e4 | 3145 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
b22cd72c | 3146 | spin_unlock(&log->l_icloglock); |
2451337d | 3147 | return -EIO; |
1da177e4 | 3148 | } |
1da177e4 LT |
3149 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE || |
3150 | iclog->ic_state == XLOG_STATE_WANT_SYNC); | |
3151 | ||
155cc6b7 | 3152 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { |
b589334c | 3153 | /* update tail before writing to iclog */ |
1c3cb9ec | 3154 | xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp); |
1da177e4 LT |
3155 | sync++; |
3156 | iclog->ic_state = XLOG_STATE_SYNCING; | |
1c3cb9ec DC |
3157 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); |
3158 | xlog_verify_tail_lsn(log, iclog, tail_lsn); | |
1da177e4 LT |
3159 | /* cycle incremented when incrementing curr_block */ |
3160 | } | |
b22cd72c | 3161 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3162 | |
3163 | /* | |
3164 | * We let the log lock go, so it's possible that we hit a log I/O | |
c41564b5 | 3165 | * error or some other SHUTDOWN condition that marks the iclog |
1da177e4 LT |
3166 | * as XLOG_STATE_IOERROR before the bwrite. However, we know that |
3167 | * this iclog has consistent data, so we ignore IOERROR | |
3168 | * flags after this point. | |
3169 | */ | |
b589334c | 3170 | if (sync) |
94860a30 | 3171 | xlog_sync(log, iclog); |
014c2544 | 3172 | return 0; |
1da177e4 LT |
3173 | } /* xlog_state_release_iclog */ |
3174 | ||
3175 | ||
3176 | /* | |
3177 | * This routine will mark the current iclog in the ring as WANT_SYNC | |
3178 | * and move the current iclog pointer to the next iclog in the ring. | |
3179 | * When this routine is called from xlog_state_get_iclog_space(), the | |
3180 | * exact size of the iclog has not yet been determined. All we know is | |
3181 | * that every data block. We have run out of space in this log record. | |
3182 | */ | |
3183 | STATIC void | |
9a8d2fdb MT |
3184 | xlog_state_switch_iclogs( |
3185 | struct xlog *log, | |
3186 | struct xlog_in_core *iclog, | |
3187 | int eventual_size) | |
1da177e4 LT |
3188 | { |
3189 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
3190 | if (!eventual_size) | |
3191 | eventual_size = iclog->ic_offset; | |
3192 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3193 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3194 | log->l_prev_block = log->l_curr_block; |
3195 | log->l_prev_cycle = log->l_curr_cycle; | |
3196 | ||
3197 | /* roll log?: ic_offset changed later */ | |
3198 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3199 | ||
3200 | /* Round up to next log-sunit */ | |
62118709 | 3201 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
1da177e4 | 3202 | log->l_mp->m_sb.sb_logsunit > 1) { |
c8ce540d | 3203 | uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit); |
1da177e4 LT |
3204 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); |
3205 | } | |
3206 | ||
3207 | if (log->l_curr_block >= log->l_logBBsize) { | |
a45086e2 BF |
3208 | /* |
3209 | * Rewind the current block before the cycle is bumped to make | |
3210 | * sure that the combined LSN never transiently moves forward | |
3211 | * when the log wraps to the next cycle. This is to support the | |
3212 | * unlocked sample of these fields from xlog_valid_lsn(). Most | |
3213 | * other cases should acquire l_icloglock. | |
3214 | */ | |
3215 | log->l_curr_block -= log->l_logBBsize; | |
3216 | ASSERT(log->l_curr_block >= 0); | |
3217 | smp_wmb(); | |
1da177e4 LT |
3218 | log->l_curr_cycle++; |
3219 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3220 | log->l_curr_cycle++; | |
1da177e4 LT |
3221 | } |
3222 | ASSERT(iclog == log->l_iclog); | |
3223 | log->l_iclog = iclog->ic_next; | |
3224 | } /* xlog_state_switch_iclogs */ | |
3225 | ||
1da177e4 LT |
3226 | /* |
3227 | * Write out all data in the in-core log as of this exact moment in time. | |
3228 | * | |
3229 | * Data may be written to the in-core log during this call. However, | |
3230 | * we don't guarantee this data will be written out. A change from past | |
3231 | * implementation means this routine will *not* write out zero length LRs. | |
3232 | * | |
3233 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3234 | * If we determine there is no flushable data, we just return. There is no | |
3235 | * flushable data if: | |
3236 | * | |
3237 | * 1. the current iclog is active and has no data; the previous iclog | |
3238 | * is in the active or dirty state. | |
3239 | * 2. the current iclog is drity, and the previous iclog is in the | |
3240 | * active or dirty state. | |
3241 | * | |
12017faf | 3242 | * We may sleep if: |
1da177e4 LT |
3243 | * |
3244 | * 1. the current iclog is not in the active nor dirty state. | |
3245 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3246 | * active nor dirty state. | |
3247 | * 3. the current iclog is active, and there is another thread writing | |
3248 | * to this particular iclog. | |
3249 | * 4. a) the current iclog is active and has no other writers | |
3250 | * b) when we return from flushing out this iclog, it is still | |
3251 | * not in the active nor dirty state. | |
3252 | */ | |
a14a348b | 3253 | int |
60e5bb78 | 3254 | xfs_log_force( |
a14a348b | 3255 | struct xfs_mount *mp, |
60e5bb78 | 3256 | uint flags) |
1da177e4 | 3257 | { |
ad223e60 | 3258 | struct xlog *log = mp->m_log; |
a14a348b CH |
3259 | struct xlog_in_core *iclog; |
3260 | xfs_lsn_t lsn; | |
3261 | ||
ff6d6af2 | 3262 | XFS_STATS_INC(mp, xs_log_force); |
60e5bb78 | 3263 | trace_xfs_log_force(mp, 0, _RET_IP_); |
1da177e4 | 3264 | |
93b8a585 | 3265 | xlog_cil_force(log); |
71e330b5 | 3266 | |
b22cd72c | 3267 | spin_lock(&log->l_icloglock); |
1da177e4 | 3268 | iclog = log->l_iclog; |
e6b96570 CH |
3269 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3270 | goto out_error; | |
1da177e4 | 3271 | |
e6b96570 CH |
3272 | if (iclog->ic_state == XLOG_STATE_DIRTY || |
3273 | (iclog->ic_state == XLOG_STATE_ACTIVE && | |
3274 | atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) { | |
1da177e4 | 3275 | /* |
e6b96570 CH |
3276 | * If the head is dirty or (active and empty), then we need to |
3277 | * look at the previous iclog. | |
3278 | * | |
3279 | * If the previous iclog is active or dirty we are done. There | |
3280 | * is nothing to sync out. Otherwise, we attach ourselves to the | |
1da177e4 LT |
3281 | * previous iclog and go to sleep. |
3282 | */ | |
e6b96570 CH |
3283 | iclog = iclog->ic_prev; |
3284 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3285 | iclog->ic_state == XLOG_STATE_DIRTY) | |
3286 | goto out_unlock; | |
3287 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3288 | if (atomic_read(&iclog->ic_refcnt) == 0) { | |
3289 | /* | |
3290 | * We are the only one with access to this iclog. | |
3291 | * | |
3292 | * Flush it out now. There should be a roundoff of zero | |
3293 | * to show that someone has already taken care of the | |
3294 | * roundoff from the previous sync. | |
3295 | */ | |
3296 | atomic_inc(&iclog->ic_refcnt); | |
3297 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
3298 | xlog_state_switch_iclogs(log, iclog, 0); | |
3299 | spin_unlock(&log->l_icloglock); | |
a14a348b | 3300 | |
e6b96570 CH |
3301 | if (xlog_state_release_iclog(log, iclog)) |
3302 | return -EIO; | |
1da177e4 | 3303 | |
e6b96570 CH |
3304 | spin_lock(&log->l_icloglock); |
3305 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn || | |
3306 | iclog->ic_state == XLOG_STATE_DIRTY) | |
3307 | goto out_unlock; | |
3308 | } else { | |
3309 | /* | |
3310 | * Someone else is writing to this iclog. | |
3311 | * | |
3312 | * Use its call to flush out the data. However, the | |
3313 | * other thread may not force out this LR, so we mark | |
3314 | * it WANT_SYNC. | |
3315 | */ | |
3316 | xlog_state_switch_iclogs(log, iclog, 0); | |
1da177e4 | 3317 | } |
e6b96570 | 3318 | } else { |
1da177e4 | 3319 | /* |
e6b96570 CH |
3320 | * If the head iclog is not active nor dirty, we just attach |
3321 | * ourselves to the head and go to sleep if necessary. | |
1da177e4 | 3322 | */ |
e6b96570 | 3323 | ; |
1da177e4 | 3324 | } |
e6b96570 CH |
3325 | |
3326 | if (!(flags & XFS_LOG_SYNC)) | |
3327 | goto out_unlock; | |
3328 | ||
3329 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3330 | goto out_error; | |
3331 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3332 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
3333 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3334 | return -EIO; | |
1da177e4 | 3335 | return 0; |
e6b96570 CH |
3336 | |
3337 | out_unlock: | |
3338 | spin_unlock(&log->l_icloglock); | |
3339 | return 0; | |
3340 | out_error: | |
3341 | spin_unlock(&log->l_icloglock); | |
3342 | return -EIO; | |
a14a348b | 3343 | } |
1da177e4 | 3344 | |
3e4da466 CH |
3345 | static int |
3346 | __xfs_log_force_lsn( | |
a14a348b CH |
3347 | struct xfs_mount *mp, |
3348 | xfs_lsn_t lsn, | |
3349 | uint flags, | |
3e4da466 CH |
3350 | int *log_flushed, |
3351 | bool already_slept) | |
1da177e4 | 3352 | { |
ad223e60 | 3353 | struct xlog *log = mp->m_log; |
a14a348b | 3354 | struct xlog_in_core *iclog; |
71e330b5 | 3355 | |
a14a348b CH |
3356 | spin_lock(&log->l_icloglock); |
3357 | iclog = log->l_iclog; | |
93806299 CH |
3358 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3359 | goto out_error; | |
1da177e4 | 3360 | |
93806299 CH |
3361 | while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { |
3362 | iclog = iclog->ic_next; | |
3363 | if (iclog == log->l_iclog) | |
3364 | goto out_unlock; | |
3365 | } | |
a14a348b | 3366 | |
93806299 CH |
3367 | if (iclog->ic_state == XLOG_STATE_DIRTY) |
3368 | goto out_unlock; | |
a14a348b | 3369 | |
93806299 CH |
3370 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3371 | /* | |
3372 | * We sleep here if we haven't already slept (e.g. this is the | |
3373 | * first time we've looked at the correct iclog buf) and the | |
3374 | * buffer before us is going to be sync'ed. The reason for this | |
3375 | * is that if we are doing sync transactions here, by waiting | |
3376 | * for the previous I/O to complete, we can allow a few more | |
3377 | * transactions into this iclog before we close it down. | |
3378 | * | |
3379 | * Otherwise, we mark the buffer WANT_SYNC, and bump up the | |
3380 | * refcnt so we can release the log (which drops the ref count). | |
3381 | * The state switch keeps new transaction commits from using | |
3382 | * this buffer. When the current commits finish writing into | |
3383 | * the buffer, the refcount will drop to zero and the buffer | |
3384 | * will go out then. | |
3385 | */ | |
3386 | if (!already_slept && | |
3387 | (iclog->ic_prev->ic_state & | |
3388 | (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) { | |
3389 | ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); | |
a14a348b | 3390 | |
93806299 | 3391 | XFS_STATS_INC(mp, xs_log_force_sleep); |
a14a348b | 3392 | |
93806299 CH |
3393 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3394 | &log->l_icloglock); | |
3e4da466 | 3395 | return -EAGAIN; |
1da177e4 | 3396 | } |
93806299 CH |
3397 | atomic_inc(&iclog->ic_refcnt); |
3398 | xlog_state_switch_iclogs(log, iclog, 0); | |
3399 | spin_unlock(&log->l_icloglock); | |
3400 | if (xlog_state_release_iclog(log, iclog)) | |
3401 | return -EIO; | |
3402 | if (log_flushed) | |
3403 | *log_flushed = 1; | |
3404 | spin_lock(&log->l_icloglock); | |
3405 | } | |
1da177e4 | 3406 | |
93806299 CH |
3407 | if (!(flags & XFS_LOG_SYNC) || |
3408 | (iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) | |
3409 | goto out_unlock; | |
1da177e4 | 3410 | |
93806299 CH |
3411 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3412 | goto out_error; | |
3413 | ||
3414 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3415 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
3416 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3417 | return -EIO; | |
3418 | return 0; | |
1da177e4 | 3419 | |
93806299 | 3420 | out_unlock: |
a14a348b CH |
3421 | spin_unlock(&log->l_icloglock); |
3422 | return 0; | |
93806299 CH |
3423 | out_error: |
3424 | spin_unlock(&log->l_icloglock); | |
3425 | return -EIO; | |
a14a348b CH |
3426 | } |
3427 | ||
3e4da466 CH |
3428 | /* |
3429 | * Force the in-core log to disk for a specific LSN. | |
3430 | * | |
3431 | * Find in-core log with lsn. | |
3432 | * If it is in the DIRTY state, just return. | |
3433 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3434 | * state and go to sleep or return. | |
3435 | * If it is in any other state, go to sleep or return. | |
3436 | * | |
3437 | * Synchronous forces are implemented with a wait queue. All callers trying | |
3438 | * to force a given lsn to disk must wait on the queue attached to the | |
3439 | * specific in-core log. When given in-core log finally completes its write | |
3440 | * to disk, that thread will wake up all threads waiting on the queue. | |
3441 | */ | |
3442 | int | |
3443 | xfs_log_force_lsn( | |
3444 | struct xfs_mount *mp, | |
3445 | xfs_lsn_t lsn, | |
3446 | uint flags, | |
3447 | int *log_flushed) | |
3448 | { | |
3449 | int ret; | |
3450 | ASSERT(lsn != 0); | |
3451 | ||
3452 | XFS_STATS_INC(mp, xs_log_force); | |
3453 | trace_xfs_log_force(mp, lsn, _RET_IP_); | |
3454 | ||
3455 | lsn = xlog_cil_force_lsn(mp->m_log, lsn); | |
3456 | if (lsn == NULLCOMMITLSN) | |
3457 | return 0; | |
3458 | ||
3459 | ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, false); | |
3460 | if (ret == -EAGAIN) | |
3461 | ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, true); | |
3462 | return ret; | |
3463 | } | |
3464 | ||
1da177e4 LT |
3465 | /* |
3466 | * Called when we want to mark the current iclog as being ready to sync to | |
3467 | * disk. | |
3468 | */ | |
a8272ce0 | 3469 | STATIC void |
9a8d2fdb MT |
3470 | xlog_state_want_sync( |
3471 | struct xlog *log, | |
3472 | struct xlog_in_core *iclog) | |
1da177e4 | 3473 | { |
a8914f3a | 3474 | assert_spin_locked(&log->l_icloglock); |
1da177e4 LT |
3475 | |
3476 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3477 | xlog_state_switch_iclogs(log, iclog, 0); | |
3478 | } else { | |
3479 | ASSERT(iclog->ic_state & | |
3480 | (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR)); | |
3481 | } | |
39e2defe | 3482 | } |
1da177e4 LT |
3483 | |
3484 | ||
3485 | /***************************************************************************** | |
3486 | * | |
3487 | * TICKET functions | |
3488 | * | |
3489 | ***************************************************************************** | |
3490 | */ | |
3491 | ||
3492 | /* | |
9da096fd | 3493 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3494 | */ |
cc09c0dc DC |
3495 | void |
3496 | xfs_log_ticket_put( | |
3497 | xlog_ticket_t *ticket) | |
1da177e4 | 3498 | { |
cc09c0dc | 3499 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3500 | if (atomic_dec_and_test(&ticket->t_ref)) |
cc09c0dc | 3501 | kmem_zone_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3502 | } |
1da177e4 | 3503 | |
cc09c0dc DC |
3504 | xlog_ticket_t * |
3505 | xfs_log_ticket_get( | |
3506 | xlog_ticket_t *ticket) | |
3507 | { | |
3508 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3509 | atomic_inc(&ticket->t_ref); | |
3510 | return ticket; | |
3511 | } | |
1da177e4 LT |
3512 | |
3513 | /* | |
e773fc93 JL |
3514 | * Figure out the total log space unit (in bytes) that would be |
3515 | * required for a log ticket. | |
1da177e4 | 3516 | */ |
e773fc93 JL |
3517 | int |
3518 | xfs_log_calc_unit_res( | |
3519 | struct xfs_mount *mp, | |
3520 | int unit_bytes) | |
1da177e4 | 3521 | { |
e773fc93 JL |
3522 | struct xlog *log = mp->m_log; |
3523 | int iclog_space; | |
3524 | uint num_headers; | |
1da177e4 LT |
3525 | |
3526 | /* | |
3527 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3528 | * in the log. A unit in this case is the amount of space for one | |
3529 | * of these log operations. Normal reservations have a cnt of 1 | |
3530 | * and their unit amount is the total amount of space required. | |
3531 | * | |
3532 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3533 | * which occupy space in the on-disk log. |
3534 | * | |
3535 | * Normal form of a transaction is: | |
3536 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3537 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3538 | * | |
3539 | * We need to account for all the leadup data and trailer data | |
3540 | * around the transaction data. | |
3541 | * And then we need to account for the worst case in terms of using | |
3542 | * more space. | |
3543 | * The worst case will happen if: | |
3544 | * - the placement of the transaction happens to be such that the | |
3545 | * roundoff is at its maximum | |
3546 | * - the transaction data is synced before the commit record is synced | |
3547 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3548 | * Therefore the commit record is in its own Log Record. | |
3549 | * This can happen as the commit record is called with its | |
3550 | * own region to xlog_write(). | |
3551 | * This then means that in the worst case, roundoff can happen for | |
3552 | * the commit-rec as well. | |
3553 | * The commit-rec is smaller than padding in this scenario and so it is | |
3554 | * not added separately. | |
1da177e4 LT |
3555 | */ |
3556 | ||
32fb9b57 TS |
3557 | /* for trans header */ |
3558 | unit_bytes += sizeof(xlog_op_header_t); | |
3559 | unit_bytes += sizeof(xfs_trans_header_t); | |
3560 | ||
1da177e4 | 3561 | /* for start-rec */ |
32fb9b57 TS |
3562 | unit_bytes += sizeof(xlog_op_header_t); |
3563 | ||
9b9fc2b7 DC |
3564 | /* |
3565 | * for LR headers - the space for data in an iclog is the size minus | |
3566 | * the space used for the headers. If we use the iclog size, then we | |
3567 | * undercalculate the number of headers required. | |
3568 | * | |
3569 | * Furthermore - the addition of op headers for split-recs might | |
3570 | * increase the space required enough to require more log and op | |
3571 | * headers, so take that into account too. | |
3572 | * | |
3573 | * IMPORTANT: This reservation makes the assumption that if this | |
3574 | * transaction is the first in an iclog and hence has the LR headers | |
3575 | * accounted to it, then the remaining space in the iclog is | |
3576 | * exclusively for this transaction. i.e. if the transaction is larger | |
3577 | * than the iclog, it will be the only thing in that iclog. | |
3578 | * Fundamentally, this means we must pass the entire log vector to | |
3579 | * xlog_write to guarantee this. | |
3580 | */ | |
3581 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3582 | num_headers = howmany(unit_bytes, iclog_space); | |
3583 | ||
3584 | /* for split-recs - ophdrs added when data split over LRs */ | |
3585 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3586 | ||
3587 | /* add extra header reservations if we overrun */ | |
3588 | while (!num_headers || | |
3589 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3590 | unit_bytes += sizeof(xlog_op_header_t); | |
3591 | num_headers++; | |
3592 | } | |
32fb9b57 | 3593 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3594 | |
32fb9b57 TS |
3595 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3596 | unit_bytes += log->l_iclog_hsize; | |
3597 | ||
32fb9b57 | 3598 | /* for roundoff padding for transaction data and one for commit record */ |
e773fc93 | 3599 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) { |
1da177e4 | 3600 | /* log su roundoff */ |
e773fc93 | 3601 | unit_bytes += 2 * mp->m_sb.sb_logsunit; |
1da177e4 LT |
3602 | } else { |
3603 | /* BB roundoff */ | |
e773fc93 | 3604 | unit_bytes += 2 * BBSIZE; |
1da177e4 LT |
3605 | } |
3606 | ||
e773fc93 JL |
3607 | return unit_bytes; |
3608 | } | |
3609 | ||
3610 | /* | |
3611 | * Allocate and initialise a new log ticket. | |
3612 | */ | |
3613 | struct xlog_ticket * | |
3614 | xlog_ticket_alloc( | |
3615 | struct xlog *log, | |
3616 | int unit_bytes, | |
3617 | int cnt, | |
3618 | char client, | |
3619 | bool permanent, | |
3620 | xfs_km_flags_t alloc_flags) | |
3621 | { | |
3622 | struct xlog_ticket *tic; | |
3623 | int unit_res; | |
3624 | ||
3625 | tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags); | |
3626 | if (!tic) | |
3627 | return NULL; | |
3628 | ||
3629 | unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes); | |
3630 | ||
cc09c0dc | 3631 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3632 | tic->t_task = current; |
10547941 | 3633 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3634 | tic->t_unit_res = unit_res; |
3635 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3636 | tic->t_cnt = cnt; |
3637 | tic->t_ocnt = cnt; | |
ecb3403d | 3638 | tic->t_tid = prandom_u32(); |
1da177e4 LT |
3639 | tic->t_clientid = client; |
3640 | tic->t_flags = XLOG_TIC_INITED; | |
9006fb91 | 3641 | if (permanent) |
1da177e4 | 3642 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3643 | |
0adba536 | 3644 | xlog_tic_reset_res(tic); |
7e9c6396 | 3645 | |
1da177e4 | 3646 | return tic; |
cc09c0dc | 3647 | } |
1da177e4 LT |
3648 | |
3649 | ||
3650 | /****************************************************************************** | |
3651 | * | |
3652 | * Log debug routines | |
3653 | * | |
3654 | ****************************************************************************** | |
3655 | */ | |
cfcbbbd0 | 3656 | #if defined(DEBUG) |
1da177e4 LT |
3657 | /* |
3658 | * Make sure that the destination ptr is within the valid data region of | |
3659 | * one of the iclogs. This uses backup pointers stored in a different | |
3660 | * part of the log in case we trash the log structure. | |
3661 | */ | |
181fdfe6 | 3662 | STATIC void |
e6b1f273 | 3663 | xlog_verify_dest_ptr( |
ad223e60 | 3664 | struct xlog *log, |
5809d5e0 | 3665 | void *ptr) |
1da177e4 LT |
3666 | { |
3667 | int i; | |
3668 | int good_ptr = 0; | |
3669 | ||
e6b1f273 CH |
3670 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3671 | if (ptr >= log->l_iclog_bak[i] && | |
3672 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3673 | good_ptr++; |
3674 | } | |
e6b1f273 CH |
3675 | |
3676 | if (!good_ptr) | |
a0fa2b67 | 3677 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3678 | } |
1da177e4 | 3679 | |
da8a1a4a DC |
3680 | /* |
3681 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3682 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3683 | * the cycles differ by exactly one and check the byte count. | |
3684 | * | |
3685 | * This check is run unlocked, so can give false positives. Rather than assert | |
3686 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3687 | * determine if they want to panic the machine when such an error occurs. For | |
3688 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3689 | * an assert, it can be turned off at runtime. | |
3690 | */ | |
3f336c6f DC |
3691 | STATIC void |
3692 | xlog_verify_grant_tail( | |
ad223e60 | 3693 | struct xlog *log) |
3f336c6f | 3694 | { |
1c3cb9ec | 3695 | int tail_cycle, tail_blocks; |
a69ed03c | 3696 | int cycle, space; |
3f336c6f | 3697 | |
28496968 | 3698 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3699 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3700 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3701 | if (cycle - 1 != tail_cycle && |
3702 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3703 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3704 | "%s: cycle - 1 != tail_cycle", __func__); | |
3705 | log->l_flags |= XLOG_TAIL_WARN; | |
3706 | } | |
3707 | ||
3708 | if (space > BBTOB(tail_blocks) && | |
3709 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3710 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3711 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3712 | log->l_flags |= XLOG_TAIL_WARN; | |
3713 | } | |
3f336c6f DC |
3714 | } |
3715 | } | |
3716 | ||
1da177e4 LT |
3717 | /* check if it will fit */ |
3718 | STATIC void | |
9a8d2fdb MT |
3719 | xlog_verify_tail_lsn( |
3720 | struct xlog *log, | |
3721 | struct xlog_in_core *iclog, | |
3722 | xfs_lsn_t tail_lsn) | |
1da177e4 LT |
3723 | { |
3724 | int blocks; | |
3725 | ||
3726 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3727 | blocks = | |
3728 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3729 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3730 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3731 | } else { |
3732 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3733 | ||
3734 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3735 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3736 | |
3737 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3738 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3739 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3740 | } |
3741 | } /* xlog_verify_tail_lsn */ | |
3742 | ||
3743 | /* | |
3744 | * Perform a number of checks on the iclog before writing to disk. | |
3745 | * | |
3746 | * 1. Make sure the iclogs are still circular | |
3747 | * 2. Make sure we have a good magic number | |
3748 | * 3. Make sure we don't have magic numbers in the data | |
3749 | * 4. Check fields of each log operation header for: | |
3750 | * A. Valid client identifier | |
3751 | * B. tid ptr value falls in valid ptr space (user space code) | |
3752 | * C. Length in log record header is correct according to the | |
3753 | * individual operation headers within record. | |
3754 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3755 | * log, check the preceding blocks of the physical log to make sure all | |
3756 | * the cycle numbers agree with the current cycle number. | |
3757 | */ | |
3758 | STATIC void | |
9a8d2fdb MT |
3759 | xlog_verify_iclog( |
3760 | struct xlog *log, | |
3761 | struct xlog_in_core *iclog, | |
abca1f33 | 3762 | int count) |
1da177e4 LT |
3763 | { |
3764 | xlog_op_header_t *ophead; | |
3765 | xlog_in_core_t *icptr; | |
3766 | xlog_in_core_2_t *xhdr; | |
5809d5e0 | 3767 | void *base_ptr, *ptr, *p; |
db9d67d6 | 3768 | ptrdiff_t field_offset; |
c8ce540d | 3769 | uint8_t clientid; |
1da177e4 LT |
3770 | int len, i, j, k, op_len; |
3771 | int idx; | |
1da177e4 LT |
3772 | |
3773 | /* check validity of iclog pointers */ | |
b22cd72c | 3774 | spin_lock(&log->l_icloglock); |
1da177e4 | 3775 | icptr = log->l_iclog; |
643f7c4e GB |
3776 | for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next) |
3777 | ASSERT(icptr); | |
3778 | ||
1da177e4 | 3779 | if (icptr != log->l_iclog) |
a0fa2b67 | 3780 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3781 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3782 | |
3783 | /* check log magic numbers */ | |
69ef921b | 3784 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3785 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3786 | |
5809d5e0 CH |
3787 | base_ptr = ptr = &iclog->ic_header; |
3788 | p = &iclog->ic_header; | |
3789 | for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) { | |
69ef921b | 3790 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3791 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3792 | __func__); | |
1da177e4 LT |
3793 | } |
3794 | ||
3795 | /* check fields */ | |
b53e675d | 3796 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
5809d5e0 CH |
3797 | base_ptr = ptr = iclog->ic_datap; |
3798 | ophead = ptr; | |
b28708d6 | 3799 | xhdr = iclog->ic_data; |
1da177e4 | 3800 | for (i = 0; i < len; i++) { |
5809d5e0 | 3801 | ophead = ptr; |
1da177e4 LT |
3802 | |
3803 | /* clientid is only 1 byte */ | |
5809d5e0 CH |
3804 | p = &ophead->oh_clientid; |
3805 | field_offset = p - base_ptr; | |
abca1f33 | 3806 | if (field_offset & 0x1ff) { |
1da177e4 LT |
3807 | clientid = ophead->oh_clientid; |
3808 | } else { | |
b2a922cd | 3809 | idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap); |
1da177e4 LT |
3810 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3811 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3812 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3813 | clientid = xlog_get_client_id( |
3814 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3815 | } else { |
03bea6fe CH |
3816 | clientid = xlog_get_client_id( |
3817 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3818 | } |
3819 | } | |
3820 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 | 3821 | xfs_warn(log->l_mp, |
c9690043 | 3822 | "%s: invalid clientid %d op "PTR_FMT" offset 0x%lx", |
a0fa2b67 DC |
3823 | __func__, clientid, ophead, |
3824 | (unsigned long)field_offset); | |
1da177e4 LT |
3825 | |
3826 | /* check length */ | |
5809d5e0 CH |
3827 | p = &ophead->oh_len; |
3828 | field_offset = p - base_ptr; | |
abca1f33 | 3829 | if (field_offset & 0x1ff) { |
67fcb7bf | 3830 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 | 3831 | } else { |
db9d67d6 CH |
3832 | idx = BTOBBT((uintptr_t)&ophead->oh_len - |
3833 | (uintptr_t)iclog->ic_datap); | |
1da177e4 LT |
3834 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3835 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3836 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3837 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3838 | } else { |
b53e675d | 3839 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3840 | } |
3841 | } | |
3842 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3843 | } | |
3844 | } /* xlog_verify_iclog */ | |
cfcbbbd0 | 3845 | #endif |
1da177e4 LT |
3846 | |
3847 | /* | |
b22cd72c | 3848 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3849 | */ |
3850 | STATIC int | |
3851 | xlog_state_ioerror( | |
9a8d2fdb | 3852 | struct xlog *log) |
1da177e4 LT |
3853 | { |
3854 | xlog_in_core_t *iclog, *ic; | |
3855 | ||
3856 | iclog = log->l_iclog; | |
3857 | if (! (iclog->ic_state & XLOG_STATE_IOERROR)) { | |
3858 | /* | |
3859 | * Mark all the incore logs IOERROR. | |
3860 | * From now on, no log flushes will result. | |
3861 | */ | |
3862 | ic = iclog; | |
3863 | do { | |
3864 | ic->ic_state = XLOG_STATE_IOERROR; | |
3865 | ic = ic->ic_next; | |
3866 | } while (ic != iclog); | |
014c2544 | 3867 | return 0; |
1da177e4 LT |
3868 | } |
3869 | /* | |
3870 | * Return non-zero, if state transition has already happened. | |
3871 | */ | |
014c2544 | 3872 | return 1; |
1da177e4 LT |
3873 | } |
3874 | ||
3875 | /* | |
3876 | * This is called from xfs_force_shutdown, when we're forcibly | |
3877 | * shutting down the filesystem, typically because of an IO error. | |
3878 | * Our main objectives here are to make sure that: | |
a870fe6d DC |
3879 | * a. if !logerror, flush the logs to disk. Anything modified |
3880 | * after this is ignored. | |
3881 | * b. the filesystem gets marked 'SHUTDOWN' for all interested | |
1da177e4 | 3882 | * parties to find out, 'atomically'. |
a870fe6d | 3883 | * c. those who're sleeping on log reservations, pinned objects and |
1da177e4 | 3884 | * other resources get woken up, and be told the bad news. |
a870fe6d | 3885 | * d. nothing new gets queued up after (b) and (c) are done. |
9da1ab18 | 3886 | * |
a870fe6d DC |
3887 | * Note: for the !logerror case we need to flush the regions held in memory out |
3888 | * to disk first. This needs to be done before the log is marked as shutdown, | |
3889 | * otherwise the iclog writes will fail. | |
1da177e4 LT |
3890 | */ |
3891 | int | |
3892 | xfs_log_force_umount( | |
3893 | struct xfs_mount *mp, | |
3894 | int logerror) | |
3895 | { | |
9a8d2fdb | 3896 | struct xlog *log; |
1da177e4 | 3897 | int retval; |
1da177e4 LT |
3898 | |
3899 | log = mp->m_log; | |
3900 | ||
3901 | /* | |
3902 | * If this happens during log recovery, don't worry about | |
3903 | * locking; the log isn't open for business yet. | |
3904 | */ | |
3905 | if (!log || | |
3906 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3907 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 | 3908 | if (mp->m_sb_bp) |
b0388bf1 | 3909 | mp->m_sb_bp->b_flags |= XBF_DONE; |
014c2544 | 3910 | return 0; |
1da177e4 LT |
3911 | } |
3912 | ||
3913 | /* | |
3914 | * Somebody could've already done the hard work for us. | |
3915 | * No need to get locks for this. | |
3916 | */ | |
3917 | if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) { | |
3918 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); | |
014c2544 | 3919 | return 1; |
1da177e4 | 3920 | } |
9da1ab18 DC |
3921 | |
3922 | /* | |
a870fe6d DC |
3923 | * Flush all the completed transactions to disk before marking the log |
3924 | * being shut down. We need to do it in this order to ensure that | |
3925 | * completed operations are safely on disk before we shut down, and that | |
3926 | * we don't have to issue any buffer IO after the shutdown flags are set | |
3927 | * to guarantee this. | |
9da1ab18 | 3928 | */ |
93b8a585 | 3929 | if (!logerror) |
60e5bb78 | 3930 | xfs_log_force(mp, XFS_LOG_SYNC); |
9da1ab18 | 3931 | |
1da177e4 | 3932 | /* |
3f16b985 DC |
3933 | * mark the filesystem and the as in a shutdown state and wake |
3934 | * everybody up to tell them the bad news. | |
1da177e4 | 3935 | */ |
b22cd72c | 3936 | spin_lock(&log->l_icloglock); |
1da177e4 | 3937 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 | 3938 | if (mp->m_sb_bp) |
b0388bf1 | 3939 | mp->m_sb_bp->b_flags |= XBF_DONE; |
bac8dca9 | 3940 | |
1da177e4 | 3941 | /* |
a870fe6d DC |
3942 | * Mark the log and the iclogs with IO error flags to prevent any |
3943 | * further log IO from being issued or completed. | |
1da177e4 LT |
3944 | */ |
3945 | log->l_flags |= XLOG_IO_ERROR; | |
a870fe6d | 3946 | retval = xlog_state_ioerror(log); |
b22cd72c | 3947 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3948 | |
3949 | /* | |
10547941 DC |
3950 | * We don't want anybody waiting for log reservations after this. That |
3951 | * means we have to wake up everybody queued up on reserveq as well as | |
3952 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3953 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3954 | * action is protected by the grant locks. |
1da177e4 | 3955 | */ |
a79bf2d7 CH |
3956 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3957 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3958 | |
1da177e4 | 3959 | /* |
ac983517 DC |
3960 | * Wake up everybody waiting on xfs_log_force. Wake the CIL push first |
3961 | * as if the log writes were completed. The abort handling in the log | |
3962 | * item committed callback functions will do this again under lock to | |
3963 | * avoid races. | |
1da177e4 | 3964 | */ |
ac983517 | 3965 | wake_up_all(&log->l_cilp->xc_commit_wait); |
d15cbf2f | 3966 | xlog_state_do_callback(log, true, NULL); |
1da177e4 LT |
3967 | |
3968 | #ifdef XFSERRORDEBUG | |
3969 | { | |
3970 | xlog_in_core_t *iclog; | |
3971 | ||
b22cd72c | 3972 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3973 | iclog = log->l_iclog; |
3974 | do { | |
3975 | ASSERT(iclog->ic_callback == 0); | |
3976 | iclog = iclog->ic_next; | |
3977 | } while (iclog != log->l_iclog); | |
b22cd72c | 3978 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3979 | } |
3980 | #endif | |
3981 | /* return non-zero if log IOERROR transition had already happened */ | |
014c2544 | 3982 | return retval; |
1da177e4 LT |
3983 | } |
3984 | ||
ba0f32d4 | 3985 | STATIC int |
9a8d2fdb MT |
3986 | xlog_iclogs_empty( |
3987 | struct xlog *log) | |
1da177e4 LT |
3988 | { |
3989 | xlog_in_core_t *iclog; | |
3990 | ||
3991 | iclog = log->l_iclog; | |
3992 | do { | |
3993 | /* endianness does not matter here, zero is zero in | |
3994 | * any language. | |
3995 | */ | |
3996 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3997 | return 0; |
1da177e4 LT |
3998 | iclog = iclog->ic_next; |
3999 | } while (iclog != log->l_iclog); | |
014c2544 | 4000 | return 1; |
1da177e4 | 4001 | } |
f661f1e0 | 4002 | |
a45086e2 BF |
4003 | /* |
4004 | * Verify that an LSN stamped into a piece of metadata is valid. This is | |
4005 | * intended for use in read verifiers on v5 superblocks. | |
4006 | */ | |
4007 | bool | |
4008 | xfs_log_check_lsn( | |
4009 | struct xfs_mount *mp, | |
4010 | xfs_lsn_t lsn) | |
4011 | { | |
4012 | struct xlog *log = mp->m_log; | |
4013 | bool valid; | |
4014 | ||
4015 | /* | |
4016 | * norecovery mode skips mount-time log processing and unconditionally | |
4017 | * resets the in-core LSN. We can't validate in this mode, but | |
4018 | * modifications are not allowed anyways so just return true. | |
4019 | */ | |
4020 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
4021 | return true; | |
4022 | ||
4023 | /* | |
4024 | * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is | |
4025 | * handled by recovery and thus safe to ignore here. | |
4026 | */ | |
4027 | if (lsn == NULLCOMMITLSN) | |
4028 | return true; | |
4029 | ||
4030 | valid = xlog_valid_lsn(mp->m_log, lsn); | |
4031 | ||
4032 | /* warn the user about what's gone wrong before verifier failure */ | |
4033 | if (!valid) { | |
4034 | spin_lock(&log->l_icloglock); | |
4035 | xfs_warn(mp, | |
4036 | "Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). " | |
4037 | "Please unmount and run xfs_repair (>= v4.3) to resolve.", | |
4038 | CYCLE_LSN(lsn), BLOCK_LSN(lsn), | |
4039 | log->l_curr_cycle, log->l_curr_block); | |
4040 | spin_unlock(&log->l_icloglock); | |
4041 | } | |
4042 | ||
4043 | return valid; | |
4044 | } | |
0c60d3aa DW |
4045 | |
4046 | bool | |
4047 | xfs_log_in_recovery( | |
4048 | struct xfs_mount *mp) | |
4049 | { | |
4050 | struct xlog *log = mp->m_log; | |
4051 | ||
4052 | return log->l_flags & XLOG_ACTIVE_RECOVERY; | |
4053 | } |