]>
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" |
1da177e4 | 26 | |
eb01c9cd | 27 | kmem_zone_t *xfs_log_ticket_zone; |
1da177e4 | 28 | |
1da177e4 | 29 | /* Local miscellaneous function prototypes */ |
ad223e60 MT |
30 | STATIC int |
31 | xlog_commit_record( | |
32 | struct xlog *log, | |
33 | struct xlog_ticket *ticket, | |
34 | struct xlog_in_core **iclog, | |
35 | xfs_lsn_t *commitlsnp); | |
36 | ||
9a8d2fdb MT |
37 | STATIC struct xlog * |
38 | xlog_alloc_log( | |
39 | struct xfs_mount *mp, | |
40 | struct xfs_buftarg *log_target, | |
41 | xfs_daddr_t blk_offset, | |
42 | int num_bblks); | |
ad223e60 MT |
43 | STATIC int |
44 | xlog_space_left( | |
45 | struct xlog *log, | |
46 | atomic64_t *head); | |
9a8d2fdb MT |
47 | STATIC int |
48 | xlog_sync( | |
49 | struct xlog *log, | |
50 | struct xlog_in_core *iclog); | |
51 | STATIC void | |
52 | xlog_dealloc_log( | |
53 | struct xlog *log); | |
1da177e4 LT |
54 | |
55 | /* local state machine functions */ | |
56 | STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int); | |
9a8d2fdb MT |
57 | STATIC void |
58 | xlog_state_do_callback( | |
59 | struct xlog *log, | |
60 | int aborted, | |
61 | struct xlog_in_core *iclog); | |
62 | STATIC int | |
63 | xlog_state_get_iclog_space( | |
64 | struct xlog *log, | |
65 | int len, | |
66 | struct xlog_in_core **iclog, | |
67 | struct xlog_ticket *ticket, | |
68 | int *continued_write, | |
69 | int *logoffsetp); | |
70 | STATIC int | |
71 | xlog_state_release_iclog( | |
72 | struct xlog *log, | |
73 | struct xlog_in_core *iclog); | |
74 | STATIC void | |
75 | xlog_state_switch_iclogs( | |
76 | struct xlog *log, | |
77 | struct xlog_in_core *iclog, | |
78 | int eventual_size); | |
79 | STATIC void | |
80 | xlog_state_want_sync( | |
81 | struct xlog *log, | |
82 | struct xlog_in_core *iclog); | |
1da177e4 | 83 | |
ad223e60 MT |
84 | STATIC void |
85 | xlog_grant_push_ail( | |
9a8d2fdb MT |
86 | struct xlog *log, |
87 | int need_bytes); | |
88 | STATIC void | |
89 | xlog_regrant_reserve_log_space( | |
90 | struct xlog *log, | |
91 | struct xlog_ticket *ticket); | |
92 | STATIC void | |
93 | xlog_ungrant_log_space( | |
94 | struct xlog *log, | |
95 | struct xlog_ticket *ticket); | |
1da177e4 | 96 | |
cfcbbbd0 | 97 | #if defined(DEBUG) |
9a8d2fdb MT |
98 | STATIC void |
99 | xlog_verify_dest_ptr( | |
100 | struct xlog *log, | |
5809d5e0 | 101 | void *ptr); |
ad223e60 MT |
102 | STATIC void |
103 | xlog_verify_grant_tail( | |
9a8d2fdb MT |
104 | struct xlog *log); |
105 | STATIC void | |
106 | xlog_verify_iclog( | |
107 | struct xlog *log, | |
108 | struct xlog_in_core *iclog, | |
109 | int count, | |
667a9291 | 110 | bool syncing); |
9a8d2fdb MT |
111 | STATIC void |
112 | xlog_verify_tail_lsn( | |
113 | struct xlog *log, | |
114 | struct xlog_in_core *iclog, | |
115 | xfs_lsn_t tail_lsn); | |
1da177e4 LT |
116 | #else |
117 | #define xlog_verify_dest_ptr(a,b) | |
3f336c6f | 118 | #define xlog_verify_grant_tail(a) |
1da177e4 LT |
119 | #define xlog_verify_iclog(a,b,c,d) |
120 | #define xlog_verify_tail_lsn(a,b,c) | |
121 | #endif | |
122 | ||
9a8d2fdb MT |
123 | STATIC int |
124 | xlog_iclogs_empty( | |
125 | struct xlog *log); | |
1da177e4 | 126 | |
dd954c69 | 127 | static void |
663e496a | 128 | xlog_grant_sub_space( |
ad223e60 MT |
129 | struct xlog *log, |
130 | atomic64_t *head, | |
131 | int bytes) | |
dd954c69 | 132 | { |
d0eb2f38 DC |
133 | int64_t head_val = atomic64_read(head); |
134 | int64_t new, old; | |
a69ed03c | 135 | |
d0eb2f38 DC |
136 | do { |
137 | int cycle, space; | |
a69ed03c | 138 | |
d0eb2f38 | 139 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 140 | |
d0eb2f38 DC |
141 | space -= bytes; |
142 | if (space < 0) { | |
143 | space += log->l_logsize; | |
144 | cycle--; | |
145 | } | |
146 | ||
147 | old = head_val; | |
148 | new = xlog_assign_grant_head_val(cycle, space); | |
149 | head_val = atomic64_cmpxchg(head, old, new); | |
150 | } while (head_val != old); | |
dd954c69 CH |
151 | } |
152 | ||
153 | static void | |
663e496a | 154 | xlog_grant_add_space( |
ad223e60 MT |
155 | struct xlog *log, |
156 | atomic64_t *head, | |
157 | int bytes) | |
dd954c69 | 158 | { |
d0eb2f38 DC |
159 | int64_t head_val = atomic64_read(head); |
160 | int64_t new, old; | |
a69ed03c | 161 | |
d0eb2f38 DC |
162 | do { |
163 | int tmp; | |
164 | int cycle, space; | |
a69ed03c | 165 | |
d0eb2f38 | 166 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 167 | |
d0eb2f38 DC |
168 | tmp = log->l_logsize - space; |
169 | if (tmp > bytes) | |
170 | space += bytes; | |
171 | else { | |
172 | space = bytes - tmp; | |
173 | cycle++; | |
174 | } | |
175 | ||
176 | old = head_val; | |
177 | new = xlog_assign_grant_head_val(cycle, space); | |
178 | head_val = atomic64_cmpxchg(head, old, new); | |
179 | } while (head_val != old); | |
dd954c69 | 180 | } |
a69ed03c | 181 | |
c303c5b8 CH |
182 | STATIC void |
183 | xlog_grant_head_init( | |
184 | struct xlog_grant_head *head) | |
185 | { | |
186 | xlog_assign_grant_head(&head->grant, 1, 0); | |
187 | INIT_LIST_HEAD(&head->waiters); | |
188 | spin_lock_init(&head->lock); | |
189 | } | |
190 | ||
a79bf2d7 CH |
191 | STATIC void |
192 | xlog_grant_head_wake_all( | |
193 | struct xlog_grant_head *head) | |
194 | { | |
195 | struct xlog_ticket *tic; | |
196 | ||
197 | spin_lock(&head->lock); | |
198 | list_for_each_entry(tic, &head->waiters, t_queue) | |
199 | wake_up_process(tic->t_task); | |
200 | spin_unlock(&head->lock); | |
201 | } | |
202 | ||
e179840d CH |
203 | static inline int |
204 | xlog_ticket_reservation( | |
ad223e60 | 205 | struct xlog *log, |
e179840d CH |
206 | struct xlog_grant_head *head, |
207 | struct xlog_ticket *tic) | |
9f9c19ec | 208 | { |
e179840d CH |
209 | if (head == &log->l_write_head) { |
210 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
211 | return tic->t_unit_res; | |
212 | } else { | |
9f9c19ec | 213 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) |
e179840d | 214 | return tic->t_unit_res * tic->t_cnt; |
9f9c19ec | 215 | else |
e179840d | 216 | return tic->t_unit_res; |
9f9c19ec | 217 | } |
9f9c19ec CH |
218 | } |
219 | ||
220 | STATIC bool | |
e179840d | 221 | xlog_grant_head_wake( |
ad223e60 | 222 | struct xlog *log, |
e179840d | 223 | struct xlog_grant_head *head, |
9f9c19ec CH |
224 | int *free_bytes) |
225 | { | |
226 | struct xlog_ticket *tic; | |
227 | int need_bytes; | |
228 | ||
e179840d CH |
229 | list_for_each_entry(tic, &head->waiters, t_queue) { |
230 | need_bytes = xlog_ticket_reservation(log, head, tic); | |
9f9c19ec CH |
231 | if (*free_bytes < need_bytes) |
232 | return false; | |
9f9c19ec | 233 | |
e179840d CH |
234 | *free_bytes -= need_bytes; |
235 | trace_xfs_log_grant_wake_up(log, tic); | |
14a7235f | 236 | wake_up_process(tic->t_task); |
9f9c19ec CH |
237 | } |
238 | ||
239 | return true; | |
240 | } | |
241 | ||
242 | STATIC int | |
23ee3df3 | 243 | xlog_grant_head_wait( |
ad223e60 | 244 | struct xlog *log, |
23ee3df3 | 245 | struct xlog_grant_head *head, |
9f9c19ec | 246 | struct xlog_ticket *tic, |
a30b0367 DC |
247 | int need_bytes) __releases(&head->lock) |
248 | __acquires(&head->lock) | |
9f9c19ec | 249 | { |
23ee3df3 | 250 | list_add_tail(&tic->t_queue, &head->waiters); |
9f9c19ec CH |
251 | |
252 | do { | |
253 | if (XLOG_FORCED_SHUTDOWN(log)) | |
254 | goto shutdown; | |
255 | xlog_grant_push_ail(log, need_bytes); | |
256 | ||
14a7235f | 257 | __set_current_state(TASK_UNINTERRUPTIBLE); |
23ee3df3 | 258 | spin_unlock(&head->lock); |
14a7235f | 259 | |
ff6d6af2 | 260 | XFS_STATS_INC(log->l_mp, xs_sleep_logspace); |
9f9c19ec | 261 | |
14a7235f CH |
262 | trace_xfs_log_grant_sleep(log, tic); |
263 | schedule(); | |
9f9c19ec CH |
264 | trace_xfs_log_grant_wake(log, tic); |
265 | ||
23ee3df3 | 266 | spin_lock(&head->lock); |
9f9c19ec CH |
267 | if (XLOG_FORCED_SHUTDOWN(log)) |
268 | goto shutdown; | |
23ee3df3 | 269 | } while (xlog_space_left(log, &head->grant) < need_bytes); |
9f9c19ec CH |
270 | |
271 | list_del_init(&tic->t_queue); | |
272 | return 0; | |
273 | shutdown: | |
274 | list_del_init(&tic->t_queue); | |
2451337d | 275 | return -EIO; |
9f9c19ec CH |
276 | } |
277 | ||
42ceedb3 CH |
278 | /* |
279 | * Atomically get the log space required for a log ticket. | |
280 | * | |
281 | * Once a ticket gets put onto head->waiters, it will only return after the | |
282 | * needed reservation is satisfied. | |
283 | * | |
284 | * This function is structured so that it has a lock free fast path. This is | |
285 | * necessary because every new transaction reservation will come through this | |
286 | * path. Hence any lock will be globally hot if we take it unconditionally on | |
287 | * every pass. | |
288 | * | |
289 | * As tickets are only ever moved on and off head->waiters under head->lock, we | |
290 | * only need to take that lock if we are going to add the ticket to the queue | |
291 | * and sleep. We can avoid taking the lock if the ticket was never added to | |
292 | * head->waiters because the t_queue list head will be empty and we hold the | |
293 | * only reference to it so it can safely be checked unlocked. | |
294 | */ | |
295 | STATIC int | |
296 | xlog_grant_head_check( | |
ad223e60 | 297 | struct xlog *log, |
42ceedb3 CH |
298 | struct xlog_grant_head *head, |
299 | struct xlog_ticket *tic, | |
300 | int *need_bytes) | |
301 | { | |
302 | int free_bytes; | |
303 | int error = 0; | |
304 | ||
305 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); | |
306 | ||
307 | /* | |
308 | * If there are other waiters on the queue then give them a chance at | |
309 | * logspace before us. Wake up the first waiters, if we do not wake | |
310 | * up all the waiters then go to sleep waiting for more free space, | |
311 | * otherwise try to get some space for this transaction. | |
312 | */ | |
313 | *need_bytes = xlog_ticket_reservation(log, head, tic); | |
314 | free_bytes = xlog_space_left(log, &head->grant); | |
315 | if (!list_empty_careful(&head->waiters)) { | |
316 | spin_lock(&head->lock); | |
317 | if (!xlog_grant_head_wake(log, head, &free_bytes) || | |
318 | free_bytes < *need_bytes) { | |
319 | error = xlog_grant_head_wait(log, head, tic, | |
320 | *need_bytes); | |
321 | } | |
322 | spin_unlock(&head->lock); | |
323 | } else if (free_bytes < *need_bytes) { | |
324 | spin_lock(&head->lock); | |
325 | error = xlog_grant_head_wait(log, head, tic, *need_bytes); | |
326 | spin_unlock(&head->lock); | |
327 | } | |
328 | ||
329 | return error; | |
330 | } | |
331 | ||
0adba536 CH |
332 | static void |
333 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
334 | { | |
335 | tic->t_res_num = 0; | |
336 | tic->t_res_arr_sum = 0; | |
337 | tic->t_res_num_ophdrs = 0; | |
338 | } | |
339 | ||
340 | static void | |
341 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
342 | { | |
343 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
344 | /* add to overflow and start again */ | |
345 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
346 | tic->t_res_num = 0; | |
347 | tic->t_res_arr_sum = 0; | |
348 | } | |
349 | ||
350 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
351 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
352 | tic->t_res_arr_sum += len; | |
353 | tic->t_res_num++; | |
354 | } | |
dd954c69 | 355 | |
9006fb91 CH |
356 | /* |
357 | * Replenish the byte reservation required by moving the grant write head. | |
358 | */ | |
359 | int | |
360 | xfs_log_regrant( | |
361 | struct xfs_mount *mp, | |
362 | struct xlog_ticket *tic) | |
363 | { | |
ad223e60 | 364 | struct xlog *log = mp->m_log; |
9006fb91 CH |
365 | int need_bytes; |
366 | int error = 0; | |
367 | ||
368 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 369 | return -EIO; |
9006fb91 | 370 | |
ff6d6af2 | 371 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
372 | |
373 | /* | |
374 | * This is a new transaction on the ticket, so we need to change the | |
375 | * transaction ID so that the next transaction has a different TID in | |
376 | * the log. Just add one to the existing tid so that we can see chains | |
377 | * of rolling transactions in the log easily. | |
378 | */ | |
379 | tic->t_tid++; | |
380 | ||
381 | xlog_grant_push_ail(log, tic->t_unit_res); | |
382 | ||
383 | tic->t_curr_res = tic->t_unit_res; | |
384 | xlog_tic_reset_res(tic); | |
385 | ||
386 | if (tic->t_cnt > 0) | |
387 | return 0; | |
388 | ||
389 | trace_xfs_log_regrant(log, tic); | |
390 | ||
391 | error = xlog_grant_head_check(log, &log->l_write_head, tic, | |
392 | &need_bytes); | |
393 | if (error) | |
394 | goto out_error; | |
395 | ||
396 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
397 | trace_xfs_log_regrant_exit(log, tic); | |
398 | xlog_verify_grant_tail(log); | |
399 | return 0; | |
400 | ||
401 | out_error: | |
402 | /* | |
403 | * If we are failing, make sure the ticket doesn't have any current | |
404 | * reservations. We don't want to add this back when the ticket/ | |
405 | * transaction gets cancelled. | |
406 | */ | |
407 | tic->t_curr_res = 0; | |
408 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
409 | return error; | |
410 | } | |
411 | ||
412 | /* | |
413 | * Reserve log space and return a ticket corresponding the reservation. | |
414 | * | |
415 | * Each reservation is going to reserve extra space for a log record header. | |
416 | * When writes happen to the on-disk log, we don't subtract the length of the | |
417 | * log record header from any reservation. By wasting space in each | |
418 | * reservation, we prevent over allocation problems. | |
419 | */ | |
420 | int | |
421 | xfs_log_reserve( | |
422 | struct xfs_mount *mp, | |
423 | int unit_bytes, | |
424 | int cnt, | |
425 | struct xlog_ticket **ticp, | |
c8ce540d | 426 | uint8_t client, |
710b1e2c | 427 | bool permanent) |
9006fb91 | 428 | { |
ad223e60 | 429 | struct xlog *log = mp->m_log; |
9006fb91 CH |
430 | struct xlog_ticket *tic; |
431 | int need_bytes; | |
432 | int error = 0; | |
433 | ||
434 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); | |
435 | ||
436 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2451337d | 437 | return -EIO; |
9006fb91 | 438 | |
ff6d6af2 | 439 | XFS_STATS_INC(mp, xs_try_logspace); |
9006fb91 CH |
440 | |
441 | ASSERT(*ticp == NULL); | |
442 | tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, | |
443 | KM_SLEEP | KM_MAYFAIL); | |
444 | if (!tic) | |
2451337d | 445 | return -ENOMEM; |
9006fb91 | 446 | |
9006fb91 CH |
447 | *ticp = tic; |
448 | ||
437a255a DC |
449 | xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt |
450 | : tic->t_unit_res); | |
9006fb91 CH |
451 | |
452 | trace_xfs_log_reserve(log, tic); | |
453 | ||
454 | error = xlog_grant_head_check(log, &log->l_reserve_head, tic, | |
455 | &need_bytes); | |
456 | if (error) | |
457 | goto out_error; | |
458 | ||
459 | xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); | |
460 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
461 | trace_xfs_log_reserve_exit(log, tic); | |
462 | xlog_verify_grant_tail(log); | |
463 | return 0; | |
464 | ||
465 | out_error: | |
466 | /* | |
467 | * If we are failing, make sure the ticket doesn't have any current | |
468 | * reservations. We don't want to add this back when the ticket/ | |
469 | * transaction gets cancelled. | |
470 | */ | |
471 | tic->t_curr_res = 0; | |
472 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
473 | return error; | |
474 | } | |
475 | ||
476 | ||
1da177e4 LT |
477 | /* |
478 | * NOTES: | |
479 | * | |
480 | * 1. currblock field gets updated at startup and after in-core logs | |
481 | * marked as with WANT_SYNC. | |
482 | */ | |
483 | ||
484 | /* | |
485 | * This routine is called when a user of a log manager ticket is done with | |
486 | * the reservation. If the ticket was ever used, then a commit record for | |
487 | * the associated transaction is written out as a log operation header with | |
488 | * no data. The flag XLOG_TIC_INITED is set when the first write occurs with | |
489 | * a given ticket. If the ticket was one with a permanent reservation, then | |
490 | * a few operations are done differently. Permanent reservation tickets by | |
491 | * default don't release the reservation. They just commit the current | |
492 | * transaction with the belief that the reservation is still needed. A flag | |
493 | * must be passed in before permanent reservations are actually released. | |
494 | * When these type of tickets are not released, they need to be set into | |
495 | * the inited state again. By doing this, a start record will be written | |
496 | * out when the next write occurs. | |
497 | */ | |
498 | xfs_lsn_t | |
35a8a72f CH |
499 | xfs_log_done( |
500 | struct xfs_mount *mp, | |
501 | struct xlog_ticket *ticket, | |
502 | struct xlog_in_core **iclog, | |
f78c3901 | 503 | bool regrant) |
1da177e4 | 504 | { |
ad223e60 | 505 | struct xlog *log = mp->m_log; |
35a8a72f | 506 | xfs_lsn_t lsn = 0; |
1da177e4 | 507 | |
1da177e4 LT |
508 | if (XLOG_FORCED_SHUTDOWN(log) || |
509 | /* | |
510 | * If nothing was ever written, don't write out commit record. | |
511 | * If we get an error, just continue and give back the log ticket. | |
512 | */ | |
513 | (((ticket->t_flags & XLOG_TIC_INITED) == 0) && | |
55b66332 | 514 | (xlog_commit_record(log, ticket, iclog, &lsn)))) { |
1da177e4 | 515 | lsn = (xfs_lsn_t) -1; |
f78c3901 | 516 | regrant = false; |
1da177e4 LT |
517 | } |
518 | ||
519 | ||
f78c3901 | 520 | if (!regrant) { |
0b1b213f CH |
521 | trace_xfs_log_done_nonperm(log, ticket); |
522 | ||
1da177e4 | 523 | /* |
c41564b5 | 524 | * Release ticket if not permanent reservation or a specific |
1da177e4 LT |
525 | * request has been made to release a permanent reservation. |
526 | */ | |
527 | xlog_ungrant_log_space(log, ticket); | |
1da177e4 | 528 | } else { |
0b1b213f CH |
529 | trace_xfs_log_done_perm(log, ticket); |
530 | ||
1da177e4 | 531 | xlog_regrant_reserve_log_space(log, ticket); |
c6a7b0f8 LM |
532 | /* If this ticket was a permanent reservation and we aren't |
533 | * trying to release it, reset the inited flags; so next time | |
534 | * we write, a start record will be written out. | |
535 | */ | |
1da177e4 | 536 | ticket->t_flags |= XLOG_TIC_INITED; |
c6a7b0f8 | 537 | } |
1da177e4 | 538 | |
f78c3901 | 539 | xfs_log_ticket_put(ticket); |
1da177e4 | 540 | return lsn; |
35a8a72f | 541 | } |
1da177e4 | 542 | |
1da177e4 LT |
543 | /* |
544 | * Attaches a new iclog I/O completion callback routine during | |
545 | * transaction commit. If the log is in error state, a non-zero | |
546 | * return code is handed back and the caller is responsible for | |
547 | * executing the callback at an appropriate time. | |
548 | */ | |
549 | int | |
35a8a72f | 550 | xfs_log_notify( |
35a8a72f CH |
551 | struct xlog_in_core *iclog, |
552 | xfs_log_callback_t *cb) | |
1da177e4 | 553 | { |
b22cd72c | 554 | int abortflg; |
1da177e4 | 555 | |
114d23aa | 556 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
557 | abortflg = (iclog->ic_state & XLOG_STATE_IOERROR); |
558 | if (!abortflg) { | |
559 | ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) || | |
560 | (iclog->ic_state == XLOG_STATE_WANT_SYNC)); | |
561 | cb->cb_next = NULL; | |
562 | *(iclog->ic_callback_tail) = cb; | |
563 | iclog->ic_callback_tail = &(cb->cb_next); | |
564 | } | |
114d23aa | 565 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 | 566 | return abortflg; |
35a8a72f | 567 | } |
1da177e4 LT |
568 | |
569 | int | |
35a8a72f CH |
570 | xfs_log_release_iclog( |
571 | struct xfs_mount *mp, | |
572 | struct xlog_in_core *iclog) | |
1da177e4 | 573 | { |
35a8a72f | 574 | if (xlog_state_release_iclog(mp->m_log, iclog)) { |
7d04a335 | 575 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
2451337d | 576 | return -EIO; |
1da177e4 LT |
577 | } |
578 | ||
579 | return 0; | |
580 | } | |
581 | ||
1da177e4 LT |
582 | /* |
583 | * Mount a log filesystem | |
584 | * | |
585 | * mp - ubiquitous xfs mount point structure | |
586 | * log_target - buftarg of on-disk log device | |
587 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
588 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
589 | * | |
590 | * Return error or zero. | |
591 | */ | |
592 | int | |
249a8c11 DC |
593 | xfs_log_mount( |
594 | xfs_mount_t *mp, | |
595 | xfs_buftarg_t *log_target, | |
596 | xfs_daddr_t blk_offset, | |
597 | int num_bblks) | |
1da177e4 | 598 | { |
9c92ee20 | 599 | bool fatal = xfs_sb_version_hascrc(&mp->m_sb); |
3e7b91cf JL |
600 | int error = 0; |
601 | int min_logfsbs; | |
249a8c11 | 602 | |
c99d609a DC |
603 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
604 | xfs_notice(mp, "Mounting V%d Filesystem", | |
605 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
606 | } else { | |
a0fa2b67 | 607 | xfs_notice(mp, |
c99d609a DC |
608 | "Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.", |
609 | XFS_SB_VERSION_NUM(&mp->m_sb)); | |
bd186aa9 | 610 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
611 | } |
612 | ||
613 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e | 614 | if (IS_ERR(mp->m_log)) { |
2451337d | 615 | error = PTR_ERR(mp->m_log); |
644c3567 DC |
616 | goto out; |
617 | } | |
1da177e4 | 618 | |
3e7b91cf JL |
619 | /* |
620 | * Validate the given log space and drop a critical message via syslog | |
621 | * if the log size is too small that would lead to some unexpected | |
622 | * situations in transaction log space reservation stage. | |
623 | * | |
624 | * Note: we can't just reject the mount if the validation fails. This | |
625 | * would mean that people would have to downgrade their kernel just to | |
626 | * remedy the situation as there is no way to grow the log (short of | |
627 | * black magic surgery with xfs_db). | |
628 | * | |
629 | * We can, however, reject mounts for CRC format filesystems, as the | |
630 | * mkfs binary being used to make the filesystem should never create a | |
631 | * filesystem with a log that is too small. | |
632 | */ | |
633 | min_logfsbs = xfs_log_calc_minimum_size(mp); | |
634 | ||
635 | if (mp->m_sb.sb_logblocks < min_logfsbs) { | |
636 | xfs_warn(mp, | |
637 | "Log size %d blocks too small, minimum size is %d blocks", | |
638 | mp->m_sb.sb_logblocks, min_logfsbs); | |
2451337d | 639 | error = -EINVAL; |
3e7b91cf JL |
640 | } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { |
641 | xfs_warn(mp, | |
642 | "Log size %d blocks too large, maximum size is %lld blocks", | |
643 | mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); | |
2451337d | 644 | error = -EINVAL; |
3e7b91cf JL |
645 | } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { |
646 | xfs_warn(mp, | |
647 | "log size %lld bytes too large, maximum size is %lld bytes", | |
648 | XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), | |
649 | XFS_MAX_LOG_BYTES); | |
2451337d | 650 | error = -EINVAL; |
9c92ee20 DW |
651 | } else if (mp->m_sb.sb_logsunit > 1 && |
652 | mp->m_sb.sb_logsunit % mp->m_sb.sb_blocksize) { | |
653 | xfs_warn(mp, | |
654 | "log stripe unit %u bytes must be a multiple of block size", | |
655 | mp->m_sb.sb_logsunit); | |
656 | error = -EINVAL; | |
657 | fatal = true; | |
3e7b91cf JL |
658 | } |
659 | if (error) { | |
9c92ee20 DW |
660 | /* |
661 | * Log check errors are always fatal on v5; or whenever bad | |
662 | * metadata leads to a crash. | |
663 | */ | |
664 | if (fatal) { | |
3e7b91cf JL |
665 | xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); |
666 | ASSERT(0); | |
667 | goto out_free_log; | |
668 | } | |
f41febd2 | 669 | xfs_crit(mp, "Log size out of supported range."); |
3e7b91cf | 670 | xfs_crit(mp, |
f41febd2 | 671 | "Continuing onwards, but if log hangs are experienced then please report this message in the bug report."); |
3e7b91cf JL |
672 | } |
673 | ||
249a8c11 DC |
674 | /* |
675 | * Initialize the AIL now we have a log. | |
676 | */ | |
249a8c11 DC |
677 | error = xfs_trans_ail_init(mp); |
678 | if (error) { | |
a0fa2b67 | 679 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 680 | goto out_free_log; |
249a8c11 | 681 | } |
a9c21c1b | 682 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 683 | |
1da177e4 LT |
684 | /* |
685 | * skip log recovery on a norecovery mount. pretend it all | |
686 | * just worked. | |
687 | */ | |
688 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 689 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
690 | |
691 | if (readonly) | |
bd186aa9 | 692 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 693 | |
65be6054 | 694 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
695 | |
696 | if (readonly) | |
bd186aa9 | 697 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 698 | if (error) { |
a0fa2b67 DC |
699 | xfs_warn(mp, "log mount/recovery failed: error %d", |
700 | error); | |
f0b2efad | 701 | xlog_recover_cancel(mp->m_log); |
26430752 | 702 | goto out_destroy_ail; |
1da177e4 LT |
703 | } |
704 | } | |
705 | ||
baff4e44 BF |
706 | error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj, |
707 | "log"); | |
708 | if (error) | |
709 | goto out_destroy_ail; | |
710 | ||
1da177e4 LT |
711 | /* Normal transactions can now occur */ |
712 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
713 | ||
71e330b5 DC |
714 | /* |
715 | * Now the log has been fully initialised and we know were our | |
716 | * space grant counters are, we can initialise the permanent ticket | |
717 | * needed for delayed logging to work. | |
718 | */ | |
719 | xlog_cil_init_post_recovery(mp->m_log); | |
720 | ||
1da177e4 | 721 | return 0; |
26430752 CH |
722 | |
723 | out_destroy_ail: | |
724 | xfs_trans_ail_destroy(mp); | |
725 | out_free_log: | |
726 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 727 | out: |
249a8c11 | 728 | return error; |
26430752 | 729 | } |
1da177e4 LT |
730 | |
731 | /* | |
f661f1e0 DC |
732 | * Finish the recovery of the file system. This is separate from the |
733 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
734 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
735 | * here. | |
1da177e4 | 736 | * |
f661f1e0 DC |
737 | * If we finish recovery successfully, start the background log work. If we are |
738 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
739 | * it. | |
1da177e4 LT |
740 | */ |
741 | int | |
f0b2efad BF |
742 | xfs_log_mount_finish( |
743 | struct xfs_mount *mp) | |
1da177e4 | 744 | { |
f661f1e0 | 745 | int error = 0; |
6f4a1eef | 746 | bool readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
f1b92bbc | 747 | bool recovered = mp->m_log->l_flags & XLOG_RECOVERY_NEEDED; |
1da177e4 | 748 | |
f0b2efad | 749 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) { |
bd186aa9 | 750 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
f0b2efad | 751 | return 0; |
6f4a1eef ES |
752 | } else if (readonly) { |
753 | /* Allow unlinked processing to proceed */ | |
754 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
1da177e4 LT |
755 | } |
756 | ||
8204f8dd DW |
757 | /* |
758 | * During the second phase of log recovery, we need iget and | |
759 | * iput to behave like they do for an active filesystem. | |
760 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
761 | * of inodes before we're done replaying log items on those | |
762 | * inodes. Turn it off immediately after recovery finishes | |
763 | * so that we don't leak the quota inodes if subsequent mount | |
764 | * activities fail. | |
799ea9e9 DW |
765 | * |
766 | * We let all inodes involved in redo item processing end up on | |
767 | * the LRU instead of being evicted immediately so that if we do | |
768 | * something to an unlinked inode, the irele won't cause | |
769 | * premature truncation and freeing of the inode, which results | |
770 | * in log recovery failure. We have to evict the unreferenced | |
1751e8a6 | 771 | * lru inodes after clearing SB_ACTIVE because we don't |
799ea9e9 DW |
772 | * otherwise clean up the lru if there's a subsequent failure in |
773 | * xfs_mountfs, which leads to us leaking the inodes if nothing | |
774 | * else (e.g. quotacheck) references the inodes before the | |
775 | * mount failure occurs. | |
8204f8dd | 776 | */ |
1751e8a6 | 777 | mp->m_super->s_flags |= SB_ACTIVE; |
f0b2efad BF |
778 | error = xlog_recover_finish(mp->m_log); |
779 | if (!error) | |
780 | xfs_log_work_queue(mp); | |
1751e8a6 | 781 | mp->m_super->s_flags &= ~SB_ACTIVE; |
799ea9e9 | 782 | evict_inodes(mp->m_super); |
f0b2efad | 783 | |
f1b92bbc BF |
784 | /* |
785 | * Drain the buffer LRU after log recovery. This is required for v4 | |
786 | * filesystems to avoid leaving around buffers with NULL verifier ops, | |
787 | * but we do it unconditionally to make sure we're always in a clean | |
788 | * cache state after mount. | |
789 | * | |
790 | * Don't push in the error case because the AIL may have pending intents | |
791 | * that aren't removed until recovery is cancelled. | |
792 | */ | |
793 | if (!error && recovered) { | |
794 | xfs_log_force(mp, XFS_LOG_SYNC); | |
795 | xfs_ail_push_all_sync(mp->m_ail); | |
796 | } | |
797 | xfs_wait_buftarg(mp->m_ddev_targp); | |
798 | ||
6f4a1eef ES |
799 | if (readonly) |
800 | mp->m_flags |= XFS_MOUNT_RDONLY; | |
801 | ||
f0b2efad BF |
802 | return error; |
803 | } | |
804 | ||
805 | /* | |
806 | * The mount has failed. Cancel the recovery if it hasn't completed and destroy | |
807 | * the log. | |
808 | */ | |
809 | int | |
810 | xfs_log_mount_cancel( | |
811 | struct xfs_mount *mp) | |
812 | { | |
813 | int error; | |
814 | ||
815 | error = xlog_recover_cancel(mp->m_log); | |
816 | xfs_log_unmount(mp); | |
f661f1e0 | 817 | |
1da177e4 LT |
818 | return error; |
819 | } | |
820 | ||
1da177e4 LT |
821 | /* |
822 | * Final log writes as part of unmount. | |
823 | * | |
824 | * Mark the filesystem clean as unmount happens. Note that during relocation | |
825 | * this routine needs to be executed as part of source-bag while the | |
826 | * deallocation must not be done until source-end. | |
827 | */ | |
828 | ||
53235f22 DW |
829 | /* Actually write the unmount record to disk. */ |
830 | static void | |
831 | xfs_log_write_unmount_record( | |
832 | struct xfs_mount *mp) | |
833 | { | |
834 | /* the data section must be 32 bit size aligned */ | |
835 | struct xfs_unmount_log_format magic = { | |
836 | .magic = XLOG_UNMOUNT_TYPE, | |
837 | }; | |
838 | struct xfs_log_iovec reg = { | |
839 | .i_addr = &magic, | |
840 | .i_len = sizeof(magic), | |
841 | .i_type = XLOG_REG_TYPE_UNMOUNT, | |
842 | }; | |
843 | struct xfs_log_vec vec = { | |
844 | .lv_niovecs = 1, | |
845 | .lv_iovecp = ®, | |
846 | }; | |
847 | struct xlog *log = mp->m_log; | |
848 | struct xlog_in_core *iclog; | |
849 | struct xlog_ticket *tic = NULL; | |
850 | xfs_lsn_t lsn; | |
851 | int error; | |
852 | ||
853 | error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0); | |
854 | if (error) | |
855 | goto out_err; | |
856 | ||
857 | /* remove inited flag, and account for space used */ | |
858 | tic->t_flags = 0; | |
859 | tic->t_curr_res -= sizeof(magic); | |
860 | error = xlog_write(log, &vec, tic, &lsn, NULL, XLOG_UNMOUNT_TRANS); | |
861 | /* | |
862 | * At this point, we're umounting anyway, so there's no point in | |
863 | * transitioning log state to IOERROR. Just continue... | |
864 | */ | |
865 | out_err: | |
866 | if (error) | |
867 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
868 | ||
869 | spin_lock(&log->l_icloglock); | |
870 | iclog = log->l_iclog; | |
871 | atomic_inc(&iclog->ic_refcnt); | |
872 | xlog_state_want_sync(log, iclog); | |
873 | spin_unlock(&log->l_icloglock); | |
874 | error = xlog_state_release_iclog(log, iclog); | |
875 | ||
876 | spin_lock(&log->l_icloglock); | |
877 | switch (iclog->ic_state) { | |
878 | default: | |
879 | if (!XLOG_FORCED_SHUTDOWN(log)) { | |
880 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
881 | break; | |
882 | } | |
883 | /* fall through */ | |
884 | case XLOG_STATE_ACTIVE: | |
885 | case XLOG_STATE_DIRTY: | |
886 | spin_unlock(&log->l_icloglock); | |
887 | break; | |
888 | } | |
889 | ||
890 | if (tic) { | |
891 | trace_xfs_log_umount_write(log, tic); | |
892 | xlog_ungrant_log_space(log, tic); | |
893 | xfs_log_ticket_put(tic); | |
894 | } | |
895 | } | |
896 | ||
1da177e4 LT |
897 | /* |
898 | * Unmount record used to have a string "Unmount filesystem--" in the | |
899 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
900 | * We just write the magic number now since that particular field isn't | |
8e159e72 | 901 | * currently architecture converted and "Unmount" is a bit foo. |
1da177e4 LT |
902 | * As far as I know, there weren't any dependencies on the old behaviour. |
903 | */ | |
904 | ||
0d5a75e9 | 905 | static int |
1da177e4 LT |
906 | xfs_log_unmount_write(xfs_mount_t *mp) |
907 | { | |
9a8d2fdb | 908 | struct xlog *log = mp->m_log; |
1da177e4 LT |
909 | xlog_in_core_t *iclog; |
910 | #ifdef DEBUG | |
911 | xlog_in_core_t *first_iclog; | |
912 | #endif | |
1da177e4 | 913 | int error; |
1da177e4 | 914 | |
1da177e4 | 915 | /* |
757a69ef | 916 | * Don't write out unmount record on norecovery mounts or ro devices. |
1da177e4 LT |
917 | * Or, if we are doing a forced umount (typically because of IO errors). |
918 | */ | |
757a69ef ES |
919 | if (mp->m_flags & XFS_MOUNT_NORECOVERY || |
920 | xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) { | |
921 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); | |
1da177e4 | 922 | return 0; |
757a69ef | 923 | } |
1da177e4 | 924 | |
60e5bb78 | 925 | error = xfs_log_force(mp, XFS_LOG_SYNC); |
b911ca04 | 926 | ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log))); |
1da177e4 LT |
927 | |
928 | #ifdef DEBUG | |
929 | first_iclog = iclog = log->l_iclog; | |
930 | do { | |
931 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
932 | ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE); | |
933 | ASSERT(iclog->ic_offset == 0); | |
934 | } | |
935 | iclog = iclog->ic_next; | |
936 | } while (iclog != first_iclog); | |
937 | #endif | |
938 | if (! (XLOG_FORCED_SHUTDOWN(log))) { | |
53235f22 | 939 | xfs_log_write_unmount_record(mp); |
1da177e4 LT |
940 | } else { |
941 | /* | |
942 | * We're already in forced_shutdown mode, couldn't | |
943 | * even attempt to write out the unmount transaction. | |
944 | * | |
945 | * Go through the motions of sync'ing and releasing | |
946 | * the iclog, even though no I/O will actually happen, | |
c41564b5 | 947 | * we need to wait for other log I/Os that may already |
1da177e4 LT |
948 | * be in progress. Do this as a separate section of |
949 | * code so we'll know if we ever get stuck here that | |
950 | * we're in this odd situation of trying to unmount | |
951 | * a file system that went into forced_shutdown as | |
952 | * the result of an unmount.. | |
953 | */ | |
b22cd72c | 954 | spin_lock(&log->l_icloglock); |
1da177e4 | 955 | iclog = log->l_iclog; |
155cc6b7 | 956 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 LT |
957 | |
958 | xlog_state_want_sync(log, iclog); | |
39e2defe | 959 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 960 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 961 | |
b22cd72c | 962 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
963 | |
964 | if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE | |
965 | || iclog->ic_state == XLOG_STATE_DIRTY | |
966 | || iclog->ic_state == XLOG_STATE_IOERROR) ) { | |
967 | ||
eb40a875 DC |
968 | xlog_wait(&iclog->ic_force_wait, |
969 | &log->l_icloglock); | |
1da177e4 | 970 | } else { |
b22cd72c | 971 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
972 | } |
973 | } | |
974 | ||
1bb7d6b5 | 975 | return error; |
1da177e4 LT |
976 | } /* xfs_log_unmount_write */ |
977 | ||
978 | /* | |
c75921a7 | 979 | * Empty the log for unmount/freeze. |
cf2931db DC |
980 | * |
981 | * To do this, we first need to shut down the background log work so it is not | |
982 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
983 | * the log so we can then flush them out. Once they have completed their IO and | |
984 | * run the callbacks removing themselves from the AIL, we can write the unmount | |
c75921a7 | 985 | * record. |
1da177e4 LT |
986 | */ |
987 | void | |
c75921a7 DC |
988 | xfs_log_quiesce( |
989 | struct xfs_mount *mp) | |
1da177e4 | 990 | { |
f661f1e0 | 991 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
992 | xfs_log_force(mp, XFS_LOG_SYNC); |
993 | ||
994 | /* | |
995 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
996 | * will push it, xfs_wait_buftarg() will not wait for it. Further, | |
997 | * xfs_buf_iowait() cannot be used because it was pushed with the | |
998 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
999 | * the IO to complete. | |
1000 | */ | |
1001 | xfs_ail_push_all_sync(mp->m_ail); | |
1002 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1003 | xfs_buf_lock(mp->m_sb_bp); | |
1004 | xfs_buf_unlock(mp->m_sb_bp); | |
1005 | ||
1006 | xfs_log_unmount_write(mp); | |
c75921a7 DC |
1007 | } |
1008 | ||
1009 | /* | |
1010 | * Shut down and release the AIL and Log. | |
1011 | * | |
1012 | * During unmount, we need to ensure we flush all the dirty metadata objects | |
1013 | * from the AIL so that the log is empty before we write the unmount record to | |
1014 | * the log. Once this is done, we can tear down the AIL and the log. | |
1015 | */ | |
1016 | void | |
1017 | xfs_log_unmount( | |
1018 | struct xfs_mount *mp) | |
1019 | { | |
1020 | xfs_log_quiesce(mp); | |
cf2931db | 1021 | |
249a8c11 | 1022 | xfs_trans_ail_destroy(mp); |
baff4e44 BF |
1023 | |
1024 | xfs_sysfs_del(&mp->m_log->l_kobj); | |
1025 | ||
c41564b5 | 1026 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
1027 | } |
1028 | ||
43f5efc5 DC |
1029 | void |
1030 | xfs_log_item_init( | |
1031 | struct xfs_mount *mp, | |
1032 | struct xfs_log_item *item, | |
1033 | int type, | |
272e42b2 | 1034 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
1035 | { |
1036 | item->li_mountp = mp; | |
1037 | item->li_ailp = mp->m_ail; | |
1038 | item->li_type = type; | |
1039 | item->li_ops = ops; | |
71e330b5 DC |
1040 | item->li_lv = NULL; |
1041 | ||
1042 | INIT_LIST_HEAD(&item->li_ail); | |
1043 | INIT_LIST_HEAD(&item->li_cil); | |
643c8c05 | 1044 | INIT_LIST_HEAD(&item->li_bio_list); |
e6631f85 | 1045 | INIT_LIST_HEAD(&item->li_trans); |
43f5efc5 DC |
1046 | } |
1047 | ||
09a423a3 CH |
1048 | /* |
1049 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 1050 | */ |
1da177e4 | 1051 | void |
09a423a3 | 1052 | xfs_log_space_wake( |
cfb7cdca | 1053 | struct xfs_mount *mp) |
1da177e4 | 1054 | { |
ad223e60 | 1055 | struct xlog *log = mp->m_log; |
cfb7cdca | 1056 | int free_bytes; |
1da177e4 | 1057 | |
1da177e4 LT |
1058 | if (XLOG_FORCED_SHUTDOWN(log)) |
1059 | return; | |
1da177e4 | 1060 | |
28496968 | 1061 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
1062 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1063 | ||
28496968 CH |
1064 | spin_lock(&log->l_write_head.lock); |
1065 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 1066 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 1067 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 1068 | } |
10547941 | 1069 | |
28496968 | 1070 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
1071 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
1072 | ||
28496968 CH |
1073 | spin_lock(&log->l_reserve_head.lock); |
1074 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 1075 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 1076 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 1077 | } |
3f16b985 | 1078 | } |
1da177e4 LT |
1079 | |
1080 | /* | |
2c6e24ce DC |
1081 | * Determine if we have a transaction that has gone to disk that needs to be |
1082 | * covered. To begin the transition to the idle state firstly the log needs to | |
1083 | * be idle. That means the CIL, the AIL and the iclogs needs to be empty before | |
1084 | * we start attempting to cover the log. | |
b6f8dd49 | 1085 | * |
2c6e24ce DC |
1086 | * Only if we are then in a state where covering is needed, the caller is |
1087 | * informed that dummy transactions are required to move the log into the idle | |
1088 | * state. | |
1089 | * | |
1090 | * If there are any items in the AIl or CIL, then we do not want to attempt to | |
1091 | * cover the log as we may be in a situation where there isn't log space | |
1092 | * available to run a dummy transaction and this can lead to deadlocks when the | |
1093 | * tail of the log is pinned by an item that is modified in the CIL. Hence | |
1094 | * there's no point in running a dummy transaction at this point because we | |
1095 | * can't start trying to idle the log until both the CIL and AIL are empty. | |
1da177e4 | 1096 | */ |
0d5a75e9 | 1097 | static int |
1da177e4 LT |
1098 | xfs_log_need_covered(xfs_mount_t *mp) |
1099 | { | |
9a8d2fdb | 1100 | struct xlog *log = mp->m_log; |
2c6e24ce | 1101 | int needed = 0; |
1da177e4 | 1102 | |
91ee575f | 1103 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
1da177e4 LT |
1104 | return 0; |
1105 | ||
2c6e24ce DC |
1106 | if (!xlog_cil_empty(log)) |
1107 | return 0; | |
1108 | ||
b22cd72c | 1109 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
1110 | switch (log->l_covered_state) { |
1111 | case XLOG_STATE_COVER_DONE: | |
1112 | case XLOG_STATE_COVER_DONE2: | |
1113 | case XLOG_STATE_COVER_IDLE: | |
1114 | break; | |
1115 | case XLOG_STATE_COVER_NEED: | |
1116 | case XLOG_STATE_COVER_NEED2: | |
2c6e24ce DC |
1117 | if (xfs_ail_min_lsn(log->l_ailp)) |
1118 | break; | |
1119 | if (!xlog_iclogs_empty(log)) | |
1120 | break; | |
1121 | ||
1122 | needed = 1; | |
1123 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) | |
1124 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
1125 | else | |
1126 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1127 | break; | |
b6f8dd49 | 1128 | default: |
1da177e4 | 1129 | needed = 1; |
b6f8dd49 | 1130 | break; |
1da177e4 | 1131 | } |
b22cd72c | 1132 | spin_unlock(&log->l_icloglock); |
014c2544 | 1133 | return needed; |
1da177e4 LT |
1134 | } |
1135 | ||
09a423a3 | 1136 | /* |
1da177e4 LT |
1137 | * We may be holding the log iclog lock upon entering this routine. |
1138 | */ | |
1139 | xfs_lsn_t | |
1c304625 | 1140 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 1141 | struct xfs_mount *mp) |
1da177e4 | 1142 | { |
ad223e60 | 1143 | struct xlog *log = mp->m_log; |
1c304625 CH |
1144 | struct xfs_log_item *lip; |
1145 | xfs_lsn_t tail_lsn; | |
1146 | ||
57e80956 | 1147 | assert_spin_locked(&mp->m_ail->ail_lock); |
1da177e4 | 1148 | |
09a423a3 CH |
1149 | /* |
1150 | * To make sure we always have a valid LSN for the log tail we keep | |
1151 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1152 | * and use that when the AIL was empty. |
09a423a3 | 1153 | */ |
1c304625 CH |
1154 | lip = xfs_ail_min(mp->m_ail); |
1155 | if (lip) | |
1156 | tail_lsn = lip->li_lsn; | |
1157 | else | |
84f3c683 | 1158 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
750b9c90 | 1159 | trace_xfs_log_assign_tail_lsn(log, tail_lsn); |
1c3cb9ec | 1160 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1161 | return tail_lsn; |
1c3cb9ec | 1162 | } |
1da177e4 | 1163 | |
1c304625 CH |
1164 | xfs_lsn_t |
1165 | xlog_assign_tail_lsn( | |
1166 | struct xfs_mount *mp) | |
1167 | { | |
1168 | xfs_lsn_t tail_lsn; | |
1169 | ||
57e80956 | 1170 | spin_lock(&mp->m_ail->ail_lock); |
1c304625 | 1171 | tail_lsn = xlog_assign_tail_lsn_locked(mp); |
57e80956 | 1172 | spin_unlock(&mp->m_ail->ail_lock); |
1c304625 CH |
1173 | |
1174 | return tail_lsn; | |
1175 | } | |
1176 | ||
1da177e4 LT |
1177 | /* |
1178 | * Return the space in the log between the tail and the head. The head | |
1179 | * is passed in the cycle/bytes formal parms. In the special case where | |
1180 | * the reserve head has wrapped passed the tail, this calculation is no | |
1181 | * longer valid. In this case, just return 0 which means there is no space | |
1182 | * in the log. This works for all places where this function is called | |
1183 | * with the reserve head. Of course, if the write head were to ever | |
1184 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1185 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1186 | * | |
1187 | * This code also handles the case where the reservation head is behind | |
1188 | * the tail. The details of this case are described below, but the end | |
1189 | * result is that we return the size of the log as the amount of space left. | |
1190 | */ | |
a8272ce0 | 1191 | STATIC int |
a69ed03c | 1192 | xlog_space_left( |
ad223e60 | 1193 | struct xlog *log, |
c8a09ff8 | 1194 | atomic64_t *head) |
1da177e4 | 1195 | { |
a69ed03c DC |
1196 | int free_bytes; |
1197 | int tail_bytes; | |
1198 | int tail_cycle; | |
1199 | int head_cycle; | |
1200 | int head_bytes; | |
1da177e4 | 1201 | |
a69ed03c | 1202 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1203 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1204 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1205 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1206 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1207 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1208 | return 0; |
a69ed03c DC |
1209 | else if (tail_cycle < head_cycle) { |
1210 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1211 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1212 | } else { |
1213 | /* | |
1214 | * The reservation head is behind the tail. | |
1215 | * In this case we just want to return the size of the | |
1216 | * log as the amount of space left. | |
1217 | */ | |
f41febd2 | 1218 | xfs_alert(log->l_mp, "xlog_space_left: head behind tail"); |
a0fa2b67 | 1219 | xfs_alert(log->l_mp, |
f41febd2 JP |
1220 | " tail_cycle = %d, tail_bytes = %d", |
1221 | tail_cycle, tail_bytes); | |
1222 | xfs_alert(log->l_mp, | |
1223 | " GH cycle = %d, GH bytes = %d", | |
1224 | head_cycle, head_bytes); | |
1da177e4 LT |
1225 | ASSERT(0); |
1226 | free_bytes = log->l_logsize; | |
1227 | } | |
1228 | return free_bytes; | |
a69ed03c | 1229 | } |
1da177e4 LT |
1230 | |
1231 | ||
1232 | /* | |
1233 | * Log function which is called when an io completes. | |
1234 | * | |
1235 | * The log manager needs its own routine, in order to control what | |
1236 | * happens with the buffer after the write completes. | |
1237 | */ | |
0d5a75e9 | 1238 | static void |
1da177e4 LT |
1239 | xlog_iodone(xfs_buf_t *bp) |
1240 | { | |
fb1755a6 | 1241 | struct xlog_in_core *iclog = bp->b_log_item; |
9a8d2fdb MT |
1242 | struct xlog *l = iclog->ic_log; |
1243 | int aborted = 0; | |
1da177e4 LT |
1244 | |
1245 | /* | |
609adfc2 BF |
1246 | * Race to shutdown the filesystem if we see an error or the iclog is in |
1247 | * IOABORT state. The IOABORT state is only set in DEBUG mode to inject | |
1248 | * CRC errors into log recovery. | |
1da177e4 | 1249 | */ |
9e24cfd0 | 1250 | if (XFS_TEST_ERROR(bp->b_error, l->l_mp, XFS_ERRTAG_IODONE_IOERR) || |
609adfc2 BF |
1251 | iclog->ic_state & XLOG_STATE_IOABORT) { |
1252 | if (iclog->ic_state & XLOG_STATE_IOABORT) | |
1253 | iclog->ic_state &= ~XLOG_STATE_IOABORT; | |
1254 | ||
901796af | 1255 | xfs_buf_ioerror_alert(bp, __func__); |
c867cb61 | 1256 | xfs_buf_stale(bp); |
7d04a335 | 1257 | xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 LT |
1258 | /* |
1259 | * This flag will be propagated to the trans-committed | |
1260 | * callback routines to let them know that the log-commit | |
1261 | * didn't succeed. | |
1262 | */ | |
1263 | aborted = XFS_LI_ABORTED; | |
1264 | } else if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
1265 | aborted = XFS_LI_ABORTED; | |
1266 | } | |
3db296f3 DC |
1267 | |
1268 | /* log I/O is always issued ASYNC */ | |
1157b32c | 1269 | ASSERT(bp->b_flags & XBF_ASYNC); |
1da177e4 | 1270 | xlog_state_done_syncing(iclog, aborted); |
9c23eccc | 1271 | |
3db296f3 | 1272 | /* |
9c23eccc DC |
1273 | * drop the buffer lock now that we are done. Nothing references |
1274 | * the buffer after this, so an unmount waiting on this lock can now | |
1275 | * tear it down safely. As such, it is unsafe to reference the buffer | |
1276 | * (bp) after the unlock as we could race with it being freed. | |
3db296f3 | 1277 | */ |
9c23eccc | 1278 | xfs_buf_unlock(bp); |
c3f8fc73 | 1279 | } |
1da177e4 | 1280 | |
1da177e4 LT |
1281 | /* |
1282 | * Return size of each in-core log record buffer. | |
1283 | * | |
9da096fd | 1284 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1285 | * |
1286 | * If the filesystem blocksize is too large, we may need to choose a | |
1287 | * larger size since the directory code currently logs entire blocks. | |
1288 | */ | |
1289 | ||
1290 | STATIC void | |
9a8d2fdb MT |
1291 | xlog_get_iclog_buffer_size( |
1292 | struct xfs_mount *mp, | |
1293 | struct xlog *log) | |
1da177e4 LT |
1294 | { |
1295 | int size; | |
1296 | int xhdrs; | |
1297 | ||
1cb51258 ES |
1298 | if (mp->m_logbufs <= 0) |
1299 | log->l_iclog_bufs = XLOG_MAX_ICLOGS; | |
1300 | else | |
cfcbbbd0 | 1301 | log->l_iclog_bufs = mp->m_logbufs; |
1da177e4 LT |
1302 | |
1303 | /* | |
1304 | * Buffer size passed in from mount system call. | |
1305 | */ | |
cfcbbbd0 | 1306 | if (mp->m_logbsize > 0) { |
1da177e4 LT |
1307 | size = log->l_iclog_size = mp->m_logbsize; |
1308 | log->l_iclog_size_log = 0; | |
1309 | while (size != 1) { | |
1310 | log->l_iclog_size_log++; | |
1311 | size >>= 1; | |
1312 | } | |
1313 | ||
62118709 | 1314 | if (xfs_sb_version_haslogv2(&mp->m_sb)) { |
9da096fd MP |
1315 | /* # headers = size / 32k |
1316 | * one header holds cycles from 32k of data | |
1da177e4 LT |
1317 | */ |
1318 | ||
1319 | xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE; | |
1320 | if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE) | |
1321 | xhdrs++; | |
1322 | log->l_iclog_hsize = xhdrs << BBSHIFT; | |
1323 | log->l_iclog_heads = xhdrs; | |
1324 | } else { | |
1325 | ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE); | |
1326 | log->l_iclog_hsize = BBSIZE; | |
1327 | log->l_iclog_heads = 1; | |
1328 | } | |
cfcbbbd0 | 1329 | goto done; |
1da177e4 LT |
1330 | } |
1331 | ||
9da096fd | 1332 | /* All machines use 32kB buffers by default. */ |
1cb51258 ES |
1333 | log->l_iclog_size = XLOG_BIG_RECORD_BSIZE; |
1334 | log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT; | |
1da177e4 LT |
1335 | |
1336 | /* the default log size is 16k or 32k which is one header sector */ | |
1337 | log->l_iclog_hsize = BBSIZE; | |
1338 | log->l_iclog_heads = 1; | |
1339 | ||
7153f8ba CH |
1340 | done: |
1341 | /* are we being asked to make the sizes selected above visible? */ | |
cfcbbbd0 NS |
1342 | if (mp->m_logbufs == 0) |
1343 | mp->m_logbufs = log->l_iclog_bufs; | |
1344 | if (mp->m_logbsize == 0) | |
1345 | mp->m_logbsize = log->l_iclog_size; | |
1da177e4 LT |
1346 | } /* xlog_get_iclog_buffer_size */ |
1347 | ||
1348 | ||
f661f1e0 DC |
1349 | void |
1350 | xfs_log_work_queue( | |
1351 | struct xfs_mount *mp) | |
1352 | { | |
696a5620 | 1353 | queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work, |
f661f1e0 DC |
1354 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); |
1355 | } | |
1356 | ||
1357 | /* | |
1358 | * Every sync period we need to unpin all items in the AIL and push them to | |
1359 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1360 | * indicate that the filesystem is idle. | |
1361 | */ | |
0d5a75e9 | 1362 | static void |
f661f1e0 DC |
1363 | xfs_log_worker( |
1364 | struct work_struct *work) | |
1365 | { | |
1366 | struct xlog *log = container_of(to_delayed_work(work), | |
1367 | struct xlog, l_work); | |
1368 | struct xfs_mount *mp = log->l_mp; | |
1369 | ||
1370 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
61e63ecb DC |
1371 | if (xfs_log_need_covered(mp)) { |
1372 | /* | |
1373 | * Dump a transaction into the log that contains no real change. | |
1374 | * This is needed to stamp the current tail LSN into the log | |
1375 | * during the covering operation. | |
1376 | * | |
1377 | * We cannot use an inode here for this - that will push dirty | |
1378 | * state back up into the VFS and then periodic inode flushing | |
1379 | * will prevent log covering from making progress. Hence we | |
1380 | * synchronously log the superblock instead to ensure the | |
1381 | * superblock is immediately unpinned and can be written back. | |
1382 | */ | |
1383 | xfs_sync_sb(mp, true); | |
1384 | } else | |
f661f1e0 DC |
1385 | xfs_log_force(mp, 0); |
1386 | ||
1387 | /* start pushing all the metadata that is currently dirty */ | |
1388 | xfs_ail_push_all(mp->m_ail); | |
1389 | ||
1390 | /* queue us up again */ | |
1391 | xfs_log_work_queue(mp); | |
1392 | } | |
1393 | ||
1da177e4 LT |
1394 | /* |
1395 | * This routine initializes some of the log structure for a given mount point. | |
1396 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1397 | * some other stuff may be filled in too. | |
1398 | */ | |
9a8d2fdb MT |
1399 | STATIC struct xlog * |
1400 | xlog_alloc_log( | |
1401 | struct xfs_mount *mp, | |
1402 | struct xfs_buftarg *log_target, | |
1403 | xfs_daddr_t blk_offset, | |
1404 | int num_bblks) | |
1da177e4 | 1405 | { |
9a8d2fdb | 1406 | struct xlog *log; |
1da177e4 LT |
1407 | xlog_rec_header_t *head; |
1408 | xlog_in_core_t **iclogp; | |
1409 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1410 | xfs_buf_t *bp; | |
1411 | int i; | |
2451337d | 1412 | int error = -ENOMEM; |
69ce58f0 | 1413 | uint log2_size = 0; |
1da177e4 | 1414 | |
9a8d2fdb | 1415 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1416 | if (!log) { |
a0fa2b67 | 1417 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1418 | goto out; |
1419 | } | |
1da177e4 LT |
1420 | |
1421 | log->l_mp = mp; | |
1422 | log->l_targ = log_target; | |
1423 | log->l_logsize = BBTOB(num_bblks); | |
1424 | log->l_logBBstart = blk_offset; | |
1425 | log->l_logBBsize = num_bblks; | |
1426 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1427 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1428 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1429 | |
1430 | log->l_prev_block = -1; | |
1da177e4 | 1431 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1432 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1433 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1434 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 CH |
1435 | |
1436 | xlog_grant_head_init(&log->l_reserve_head); | |
1437 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1438 | |
2451337d | 1439 | error = -EFSCORRUPTED; |
62118709 | 1440 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1441 | log2_size = mp->m_sb.sb_logsectlog; |
1442 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1443 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1444 | log2_size, BBSHIFT); | |
a6cb767e DC |
1445 | goto out_free_log; |
1446 | } | |
1447 | ||
69ce58f0 AE |
1448 | log2_size -= BBSHIFT; |
1449 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1450 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1451 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1452 | goto out_free_log; |
1453 | } | |
69ce58f0 AE |
1454 | |
1455 | /* for larger sector sizes, must have v2 or external log */ | |
1456 | if (log2_size && log->l_logBBstart > 0 && | |
1457 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1458 | xfs_warn(mp, |
1459 | "log sector size (0x%x) invalid for configuration.", | |
1460 | log2_size); | |
a6cb767e DC |
1461 | goto out_free_log; |
1462 | } | |
1da177e4 | 1463 | } |
69ce58f0 | 1464 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1465 | |
1466 | xlog_get_iclog_buffer_size(mp, log); | |
1467 | ||
400b9d88 DC |
1468 | /* |
1469 | * Use a NULL block for the extra log buffer used during splits so that | |
1470 | * it will trigger errors if we ever try to do IO on it without first | |
1471 | * having set it up properly. | |
1472 | */ | |
2451337d | 1473 | error = -ENOMEM; |
400b9d88 | 1474 | bp = xfs_buf_alloc(mp->m_logdev_targp, XFS_BUF_DADDR_NULL, |
c891c30a | 1475 | BTOBB(log->l_iclog_size), XBF_NO_IOACCT); |
644c3567 DC |
1476 | if (!bp) |
1477 | goto out_free_log; | |
9c23eccc DC |
1478 | |
1479 | /* | |
1480 | * The iclogbuf buffer locks are held over IO but we are not going to do | |
1481 | * IO yet. Hence unlock the buffer so that the log IO path can grab it | |
1482 | * when appropriately. | |
1483 | */ | |
0c842ad4 | 1484 | ASSERT(xfs_buf_islocked(bp)); |
9c23eccc DC |
1485 | xfs_buf_unlock(bp); |
1486 | ||
96ab7954 BF |
1487 | /* use high priority wq for log I/O completion */ |
1488 | bp->b_ioend_wq = mp->m_log_workqueue; | |
9c23eccc | 1489 | bp->b_iodone = xlog_iodone; |
1da177e4 LT |
1490 | log->l_xbuf = bp; |
1491 | ||
007c61c6 | 1492 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1493 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1494 | |
1da177e4 LT |
1495 | iclogp = &log->l_iclog; |
1496 | /* | |
1497 | * The amount of memory to allocate for the iclog structure is | |
1498 | * rather funky due to the way the structure is defined. It is | |
1499 | * done this way so that we can use different sizes for machines | |
1500 | * with different amounts of memory. See the definition of | |
1501 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1502 | */ | |
1da177e4 LT |
1503 | ASSERT(log->l_iclog_size >= 4096); |
1504 | for (i=0; i < log->l_iclog_bufs; i++) { | |
644c3567 DC |
1505 | *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL); |
1506 | if (!*iclogp) | |
1507 | goto out_free_iclog; | |
1508 | ||
1da177e4 | 1509 | iclog = *iclogp; |
1da177e4 LT |
1510 | iclog->ic_prev = prev_iclog; |
1511 | prev_iclog = iclog; | |
1fa40b01 | 1512 | |
686865f7 | 1513 | bp = xfs_buf_get_uncached(mp->m_logdev_targp, |
c891c30a BF |
1514 | BTOBB(log->l_iclog_size), |
1515 | XBF_NO_IOACCT); | |
644c3567 DC |
1516 | if (!bp) |
1517 | goto out_free_iclog; | |
c8da0faf | 1518 | |
9c23eccc DC |
1519 | ASSERT(xfs_buf_islocked(bp)); |
1520 | xfs_buf_unlock(bp); | |
1521 | ||
96ab7954 BF |
1522 | /* use high priority wq for log I/O completion */ |
1523 | bp->b_ioend_wq = mp->m_log_workqueue; | |
cb669ca5 | 1524 | bp->b_iodone = xlog_iodone; |
1fa40b01 | 1525 | iclog->ic_bp = bp; |
b28708d6 | 1526 | iclog->ic_data = bp->b_addr; |
4679b2d3 | 1527 | #ifdef DEBUG |
5809d5e0 | 1528 | log->l_iclog_bak[i] = &iclog->ic_header; |
4679b2d3 | 1529 | #endif |
1da177e4 LT |
1530 | head = &iclog->ic_header; |
1531 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1532 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1533 | head->h_version = cpu_to_be32( | |
62118709 | 1534 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1535 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1536 | /* new fields */ |
b53e675d | 1537 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1538 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1539 | ||
4e94b71b | 1540 | iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize; |
1da177e4 LT |
1541 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1542 | iclog->ic_log = log; | |
114d23aa DC |
1543 | atomic_set(&iclog->ic_refcnt, 0); |
1544 | spin_lock_init(&iclog->ic_callback_lock); | |
1da177e4 | 1545 | iclog->ic_callback_tail = &(iclog->ic_callback); |
b28708d6 | 1546 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1547 | |
eb40a875 DC |
1548 | init_waitqueue_head(&iclog->ic_force_wait); |
1549 | init_waitqueue_head(&iclog->ic_write_wait); | |
1da177e4 LT |
1550 | |
1551 | iclogp = &iclog->ic_next; | |
1552 | } | |
1553 | *iclogp = log->l_iclog; /* complete ring */ | |
1554 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1555 | ||
71e330b5 DC |
1556 | error = xlog_cil_init(log); |
1557 | if (error) | |
1558 | goto out_free_iclog; | |
1da177e4 | 1559 | return log; |
644c3567 DC |
1560 | |
1561 | out_free_iclog: | |
1562 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1563 | prev_iclog = iclog->ic_next; | |
eb40a875 | 1564 | if (iclog->ic_bp) |
644c3567 | 1565 | xfs_buf_free(iclog->ic_bp); |
644c3567 DC |
1566 | kmem_free(iclog); |
1567 | } | |
1568 | spinlock_destroy(&log->l_icloglock); | |
644c3567 DC |
1569 | xfs_buf_free(log->l_xbuf); |
1570 | out_free_log: | |
1571 | kmem_free(log); | |
a6cb767e | 1572 | out: |
2451337d | 1573 | return ERR_PTR(error); |
1da177e4 LT |
1574 | } /* xlog_alloc_log */ |
1575 | ||
1576 | ||
1577 | /* | |
1578 | * Write out the commit record of a transaction associated with the given | |
1579 | * ticket. Return the lsn of the commit record. | |
1580 | */ | |
1581 | STATIC int | |
55b66332 | 1582 | xlog_commit_record( |
ad223e60 | 1583 | struct xlog *log, |
55b66332 DC |
1584 | struct xlog_ticket *ticket, |
1585 | struct xlog_in_core **iclog, | |
1586 | xfs_lsn_t *commitlsnp) | |
1da177e4 | 1587 | { |
55b66332 DC |
1588 | struct xfs_mount *mp = log->l_mp; |
1589 | int error; | |
1590 | struct xfs_log_iovec reg = { | |
1591 | .i_addr = NULL, | |
1592 | .i_len = 0, | |
1593 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1594 | }; | |
1595 | struct xfs_log_vec vec = { | |
1596 | .lv_niovecs = 1, | |
1597 | .lv_iovecp = ®, | |
1598 | }; | |
1da177e4 LT |
1599 | |
1600 | ASSERT_ALWAYS(iclog); | |
55b66332 DC |
1601 | error = xlog_write(log, &vec, ticket, commitlsnp, iclog, |
1602 | XLOG_COMMIT_TRANS); | |
1603 | if (error) | |
7d04a335 | 1604 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1605 | return error; |
55b66332 | 1606 | } |
1da177e4 LT |
1607 | |
1608 | /* | |
1609 | * Push on the buffer cache code if we ever use more than 75% of the on-disk | |
1610 | * log space. This code pushes on the lsn which would supposedly free up | |
1611 | * the 25% which we want to leave free. We may need to adopt a policy which | |
1612 | * pushes on an lsn which is further along in the log once we reach the high | |
1613 | * water mark. In this manner, we would be creating a low water mark. | |
1614 | */ | |
a8272ce0 | 1615 | STATIC void |
2ced19cb | 1616 | xlog_grant_push_ail( |
ad223e60 | 1617 | struct xlog *log, |
2ced19cb | 1618 | int need_bytes) |
1da177e4 | 1619 | { |
2ced19cb | 1620 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1621 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1622 | int free_blocks; |
1623 | int free_bytes; | |
1624 | int threshold_block; | |
1625 | int threshold_cycle; | |
1626 | int free_threshold; | |
1627 | ||
1628 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1629 | ||
28496968 | 1630 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1631 | free_blocks = BTOBBT(free_bytes); |
1632 | ||
1633 | /* | |
1634 | * Set the threshold for the minimum number of free blocks in the | |
1635 | * log to the maximum of what the caller needs, one quarter of the | |
1636 | * log, and 256 blocks. | |
1637 | */ | |
1638 | free_threshold = BTOBB(need_bytes); | |
9bb54cb5 DC |
1639 | free_threshold = max(free_threshold, (log->l_logBBsize >> 2)); |
1640 | free_threshold = max(free_threshold, 256); | |
2ced19cb DC |
1641 | if (free_blocks >= free_threshold) |
1642 | return; | |
1643 | ||
1c3cb9ec DC |
1644 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1645 | &threshold_block); | |
1646 | threshold_block += free_threshold; | |
1da177e4 | 1647 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1648 | threshold_block -= log->l_logBBsize; |
1649 | threshold_cycle += 1; | |
1da177e4 | 1650 | } |
2ced19cb DC |
1651 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1652 | threshold_block); | |
1653 | /* | |
1654 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1655 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1656 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1657 | */ |
84f3c683 DC |
1658 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1659 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1660 | threshold_lsn = last_sync_lsn; | |
2ced19cb DC |
1661 | |
1662 | /* | |
1663 | * Get the transaction layer to kick the dirty buffers out to | |
1664 | * disk asynchronously. No point in trying to do this if | |
1665 | * the filesystem is shutting down. | |
1666 | */ | |
1667 | if (!XLOG_FORCED_SHUTDOWN(log)) | |
fd074841 | 1668 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1669 | } |
1da177e4 | 1670 | |
0e446be4 CH |
1671 | /* |
1672 | * Stamp cycle number in every block | |
1673 | */ | |
1674 | STATIC void | |
1675 | xlog_pack_data( | |
1676 | struct xlog *log, | |
1677 | struct xlog_in_core *iclog, | |
1678 | int roundoff) | |
1679 | { | |
1680 | int i, j, k; | |
1681 | int size = iclog->ic_offset + roundoff; | |
1682 | __be32 cycle_lsn; | |
b2a922cd | 1683 | char *dp; |
0e446be4 CH |
1684 | |
1685 | cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); | |
1686 | ||
1687 | dp = iclog->ic_datap; | |
1688 | for (i = 0; i < BTOBB(size); i++) { | |
1689 | if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) | |
1690 | break; | |
1691 | iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; | |
1692 | *(__be32 *)dp = cycle_lsn; | |
1693 | dp += BBSIZE; | |
1694 | } | |
1695 | ||
1696 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1697 | xlog_in_core_2_t *xhdr = iclog->ic_data; | |
1698 | ||
1699 | for ( ; i < BTOBB(size); i++) { | |
1700 | j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1701 | k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
1702 | xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; | |
1703 | *(__be32 *)dp = cycle_lsn; | |
1704 | dp += BBSIZE; | |
1705 | } | |
1706 | ||
1707 | for (i = 1; i < log->l_iclog_heads; i++) | |
1708 | xhdr[i].hic_xheader.xh_cycle = cycle_lsn; | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | /* | |
1713 | * Calculate the checksum for a log buffer. | |
1714 | * | |
1715 | * This is a little more complicated than it should be because the various | |
1716 | * headers and the actual data are non-contiguous. | |
1717 | */ | |
f9668a09 | 1718 | __le32 |
0e446be4 CH |
1719 | xlog_cksum( |
1720 | struct xlog *log, | |
1721 | struct xlog_rec_header *rhead, | |
1722 | char *dp, | |
1723 | int size) | |
1724 | { | |
c8ce540d | 1725 | uint32_t crc; |
0e446be4 CH |
1726 | |
1727 | /* first generate the crc for the record header ... */ | |
cae028df | 1728 | crc = xfs_start_cksum_update((char *)rhead, |
0e446be4 CH |
1729 | sizeof(struct xlog_rec_header), |
1730 | offsetof(struct xlog_rec_header, h_crc)); | |
1731 | ||
1732 | /* ... then for additional cycle data for v2 logs ... */ | |
1733 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { | |
1734 | union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; | |
1735 | int i; | |
a3f20014 | 1736 | int xheads; |
0e446be4 | 1737 | |
a3f20014 BF |
1738 | xheads = size / XLOG_HEADER_CYCLE_SIZE; |
1739 | if (size % XLOG_HEADER_CYCLE_SIZE) | |
1740 | xheads++; | |
0e446be4 | 1741 | |
a3f20014 | 1742 | for (i = 1; i < xheads; i++) { |
0e446be4 CH |
1743 | crc = crc32c(crc, &xhdr[i].hic_xheader, |
1744 | sizeof(struct xlog_rec_ext_header)); | |
1745 | } | |
1746 | } | |
1747 | ||
1748 | /* ... and finally for the payload */ | |
1749 | crc = crc32c(crc, dp, size); | |
1750 | ||
1751 | return xfs_end_cksum(crc); | |
1752 | } | |
1753 | ||
873ff550 CH |
1754 | /* |
1755 | * The bdstrat callback function for log bufs. This gives us a central | |
1756 | * place to trap bufs in case we get hit by a log I/O error and need to | |
1757 | * shutdown. Actually, in practice, even when we didn't get a log error, | |
1758 | * we transition the iclogs to IOERROR state *after* flushing all existing | |
1759 | * iclogs to disk. This is because we don't want anymore new transactions to be | |
1760 | * started or completed afterwards. | |
9c23eccc DC |
1761 | * |
1762 | * We lock the iclogbufs here so that we can serialise against IO completion | |
1763 | * during unmount. We might be processing a shutdown triggered during unmount, | |
1764 | * and that can occur asynchronously to the unmount thread, and hence we need to | |
1765 | * ensure that completes before tearing down the iclogbufs. Hence we need to | |
1766 | * hold the buffer lock across the log IO to acheive that. | |
873ff550 CH |
1767 | */ |
1768 | STATIC int | |
1769 | xlog_bdstrat( | |
1770 | struct xfs_buf *bp) | |
1771 | { | |
fb1755a6 | 1772 | struct xlog_in_core *iclog = bp->b_log_item; |
873ff550 | 1773 | |
9c23eccc | 1774 | xfs_buf_lock(bp); |
873ff550 | 1775 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
2451337d | 1776 | xfs_buf_ioerror(bp, -EIO); |
c867cb61 | 1777 | xfs_buf_stale(bp); |
e8aaba9a | 1778 | xfs_buf_ioend(bp); |
873ff550 CH |
1779 | /* |
1780 | * It would seem logical to return EIO here, but we rely on | |
1781 | * the log state machine to propagate I/O errors instead of | |
9c23eccc DC |
1782 | * doing it here. Similarly, IO completion will unlock the |
1783 | * buffer, so we don't do it here. | |
873ff550 CH |
1784 | */ |
1785 | return 0; | |
1786 | } | |
1787 | ||
595bff75 | 1788 | xfs_buf_submit(bp); |
873ff550 CH |
1789 | return 0; |
1790 | } | |
1da177e4 LT |
1791 | |
1792 | /* | |
1793 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1794 | * fashion. Previously, we should have moved the current iclog | |
1795 | * ptr in the log to point to the next available iclog. This allows further | |
1796 | * write to continue while this code syncs out an iclog ready to go. | |
1797 | * Before an in-core log can be written out, the data section must be scanned | |
1798 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1799 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1800 | * cycle count because there in an implicit assumption that drives will | |
1801 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1802 | * we can't have part of a 512 byte block written and part not written. By | |
1803 | * tagging each block, we will know which blocks are valid when recovering | |
1804 | * after an unclean shutdown. | |
1805 | * | |
1806 | * This routine is single threaded on the iclog. No other thread can be in | |
1807 | * this routine with the same iclog. Changing contents of iclog can there- | |
1808 | * fore be done without grabbing the state machine lock. Updating the global | |
1809 | * log will require grabbing the lock though. | |
1810 | * | |
1811 | * The entire log manager uses a logical block numbering scheme. Only | |
1812 | * log_sync (and then only bwrite()) know about the fact that the log may | |
1813 | * not start with block zero on a given device. The log block start offset | |
1814 | * is added immediately before calling bwrite(). | |
1815 | */ | |
1816 | ||
a8272ce0 | 1817 | STATIC int |
9a8d2fdb MT |
1818 | xlog_sync( |
1819 | struct xlog *log, | |
1820 | struct xlog_in_core *iclog) | |
1da177e4 | 1821 | { |
1da177e4 | 1822 | xfs_buf_t *bp; |
b53e675d | 1823 | int i; |
1da177e4 LT |
1824 | uint count; /* byte count of bwrite */ |
1825 | uint count_init; /* initial count before roundup */ | |
1826 | int roundoff; /* roundoff to BB or stripe */ | |
1827 | int split = 0; /* split write into two regions */ | |
1828 | int error; | |
62118709 | 1829 | int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb); |
0e446be4 | 1830 | int size; |
1da177e4 | 1831 | |
ff6d6af2 | 1832 | XFS_STATS_INC(log->l_mp, xs_log_writes); |
155cc6b7 | 1833 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
1834 | |
1835 | /* Add for LR header */ | |
1836 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
1837 | ||
1838 | /* Round out the log write size */ | |
1839 | if (v2 && log->l_mp->m_sb.sb_logsunit > 1) { | |
1840 | /* we have a v2 stripe unit to use */ | |
1841 | count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init)); | |
1842 | } else { | |
1843 | count = BBTOB(BTOBB(count_init)); | |
1844 | } | |
1845 | roundoff = count - count_init; | |
1846 | ASSERT(roundoff >= 0); | |
1847 | ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && | |
1848 | roundoff < log->l_mp->m_sb.sb_logsunit) | |
1849 | || | |
1850 | (log->l_mp->m_sb.sb_logsunit <= 1 && | |
1851 | roundoff < BBTOB(1))); | |
1852 | ||
1853 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1854 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1855 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1856 | |
1857 | /* put cycle number in every block */ | |
1858 | xlog_pack_data(log, iclog, roundoff); | |
1859 | ||
1860 | /* real byte length */ | |
0e446be4 CH |
1861 | size = iclog->ic_offset; |
1862 | if (v2) | |
1863 | size += roundoff; | |
1864 | iclog->ic_header.h_len = cpu_to_be32(size); | |
1da177e4 | 1865 | |
f5faad79 | 1866 | bp = iclog->ic_bp; |
b53e675d | 1867 | XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn))); |
1da177e4 | 1868 | |
ff6d6af2 | 1869 | XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count)); |
1da177e4 LT |
1870 | |
1871 | /* Do we need to split this write into 2 parts? */ | |
1872 | if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) { | |
0e446be4 CH |
1873 | char *dptr; |
1874 | ||
1da177e4 LT |
1875 | split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp))); |
1876 | count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)); | |
0e446be4 CH |
1877 | iclog->ic_bwritecnt = 2; |
1878 | ||
1879 | /* | |
1880 | * Bump the cycle numbers at the start of each block in the | |
1881 | * part of the iclog that ends up in the buffer that gets | |
1882 | * written to the start of the log. | |
1883 | * | |
1884 | * Watch out for the header magic number case, though. | |
1885 | */ | |
1886 | dptr = (char *)&iclog->ic_header + count; | |
1887 | for (i = 0; i < split; i += BBSIZE) { | |
c8ce540d | 1888 | uint32_t cycle = be32_to_cpu(*(__be32 *)dptr); |
0e446be4 CH |
1889 | if (++cycle == XLOG_HEADER_MAGIC_NUM) |
1890 | cycle++; | |
1891 | *(__be32 *)dptr = cpu_to_be32(cycle); | |
1892 | ||
1893 | dptr += BBSIZE; | |
1894 | } | |
1da177e4 LT |
1895 | } else { |
1896 | iclog->ic_bwritecnt = 1; | |
1897 | } | |
0e446be4 CH |
1898 | |
1899 | /* calculcate the checksum */ | |
1900 | iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, | |
1901 | iclog->ic_datap, size); | |
609adfc2 BF |
1902 | /* |
1903 | * Intentionally corrupt the log record CRC based on the error injection | |
1904 | * frequency, if defined. This facilitates testing log recovery in the | |
1905 | * event of torn writes. Hence, set the IOABORT state to abort the log | |
1906 | * write on I/O completion and shutdown the fs. The subsequent mount | |
1907 | * detects the bad CRC and attempts to recover. | |
1908 | */ | |
3e88a007 | 1909 | if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) { |
e2a64192 | 1910 | iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA); |
609adfc2 BF |
1911 | iclog->ic_state |= XLOG_STATE_IOABORT; |
1912 | xfs_warn(log->l_mp, | |
1913 | "Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.", | |
1914 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1915 | } | |
0e446be4 | 1916 | |
aa0e8833 | 1917 | bp->b_io_length = BTOBB(count); |
fb1755a6 | 1918 | bp->b_log_item = iclog; |
2291dab2 DC |
1919 | bp->b_flags &= ~XBF_FLUSH; |
1920 | bp->b_flags |= (XBF_ASYNC | XBF_SYNCIO | XBF_WRITE | XBF_FUA); | |
651701d7 | 1921 | |
2291dab2 DC |
1922 | /* |
1923 | * Flush the data device before flushing the log to make sure all meta | |
1924 | * data written back from the AIL actually made it to disk before | |
1925 | * stamping the new log tail LSN into the log buffer. For an external | |
1926 | * log we need to issue the flush explicitly, and unfortunately | |
1927 | * synchronously here; for an internal log we can simply use the block | |
1928 | * layer state machine for preflushes. | |
1929 | */ | |
1930 | if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp) | |
1931 | xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp); | |
1932 | else | |
1933 | bp->b_flags |= XBF_FLUSH; | |
1da177e4 LT |
1934 | |
1935 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1936 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1937 | ||
003fd6c8 | 1938 | xlog_verify_iclog(log, iclog, count, true); |
1da177e4 LT |
1939 | |
1940 | /* account for log which doesn't start at block #0 */ | |
1941 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); | |
b68c0821 | 1942 | |
1da177e4 LT |
1943 | /* |
1944 | * Don't call xfs_bwrite here. We do log-syncs even when the filesystem | |
1945 | * is shutting down. | |
1946 | */ | |
901796af CH |
1947 | error = xlog_bdstrat(bp); |
1948 | if (error) { | |
1949 | xfs_buf_ioerror_alert(bp, "xlog_sync"); | |
014c2544 | 1950 | return error; |
1da177e4 LT |
1951 | } |
1952 | if (split) { | |
f5faad79 | 1953 | bp = iclog->ic_log->l_xbuf; |
1da177e4 | 1954 | XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */ |
02fe03d9 CS |
1955 | xfs_buf_associate_memory(bp, |
1956 | (char *)&iclog->ic_header + count, split); | |
fb1755a6 | 1957 | bp->b_log_item = iclog; |
2291dab2 DC |
1958 | bp->b_flags &= ~XBF_FLUSH; |
1959 | bp->b_flags |= (XBF_ASYNC | XBF_SYNCIO | XBF_WRITE | XBF_FUA); | |
1da177e4 LT |
1960 | |
1961 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1962 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1963 | ||
c41564b5 | 1964 | /* account for internal log which doesn't start at block #0 */ |
1da177e4 | 1965 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); |
901796af CH |
1966 | error = xlog_bdstrat(bp); |
1967 | if (error) { | |
1968 | xfs_buf_ioerror_alert(bp, "xlog_sync (split)"); | |
014c2544 | 1969 | return error; |
1da177e4 LT |
1970 | } |
1971 | } | |
014c2544 | 1972 | return 0; |
1da177e4 LT |
1973 | } /* xlog_sync */ |
1974 | ||
1da177e4 | 1975 | /* |
c41564b5 | 1976 | * Deallocate a log structure |
1da177e4 | 1977 | */ |
a8272ce0 | 1978 | STATIC void |
9a8d2fdb MT |
1979 | xlog_dealloc_log( |
1980 | struct xlog *log) | |
1da177e4 LT |
1981 | { |
1982 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1983 | int i; |
1984 | ||
71e330b5 DC |
1985 | xlog_cil_destroy(log); |
1986 | ||
44396476 | 1987 | /* |
9c23eccc DC |
1988 | * Cycle all the iclogbuf locks to make sure all log IO completion |
1989 | * is done before we tear down these buffers. | |
1990 | */ | |
1991 | iclog = log->l_iclog; | |
1992 | for (i = 0; i < log->l_iclog_bufs; i++) { | |
1993 | xfs_buf_lock(iclog->ic_bp); | |
1994 | xfs_buf_unlock(iclog->ic_bp); | |
1995 | iclog = iclog->ic_next; | |
1996 | } | |
1997 | ||
1998 | /* | |
1999 | * Always need to ensure that the extra buffer does not point to memory | |
2000 | * owned by another log buffer before we free it. Also, cycle the lock | |
2001 | * first to ensure we've completed IO on it. | |
44396476 | 2002 | */ |
9c23eccc DC |
2003 | xfs_buf_lock(log->l_xbuf); |
2004 | xfs_buf_unlock(log->l_xbuf); | |
e70b73f8 | 2005 | xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size)); |
44396476 DC |
2006 | xfs_buf_free(log->l_xbuf); |
2007 | ||
1da177e4 | 2008 | iclog = log->l_iclog; |
9c23eccc | 2009 | for (i = 0; i < log->l_iclog_bufs; i++) { |
1da177e4 | 2010 | xfs_buf_free(iclog->ic_bp); |
1da177e4 | 2011 | next_iclog = iclog->ic_next; |
f0e2d93c | 2012 | kmem_free(iclog); |
1da177e4 LT |
2013 | iclog = next_iclog; |
2014 | } | |
1da177e4 | 2015 | spinlock_destroy(&log->l_icloglock); |
1da177e4 | 2016 | |
1da177e4 | 2017 | log->l_mp->m_log = NULL; |
f0e2d93c | 2018 | kmem_free(log); |
c41564b5 | 2019 | } /* xlog_dealloc_log */ |
1da177e4 LT |
2020 | |
2021 | /* | |
2022 | * Update counters atomically now that memcpy is done. | |
2023 | */ | |
2024 | /* ARGSUSED */ | |
2025 | static inline void | |
9a8d2fdb MT |
2026 | xlog_state_finish_copy( |
2027 | struct xlog *log, | |
2028 | struct xlog_in_core *iclog, | |
2029 | int record_cnt, | |
2030 | int copy_bytes) | |
1da177e4 | 2031 | { |
b22cd72c | 2032 | spin_lock(&log->l_icloglock); |
1da177e4 | 2033 | |
413d57c9 | 2034 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 LT |
2035 | iclog->ic_offset += copy_bytes; |
2036 | ||
b22cd72c | 2037 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2038 | } /* xlog_state_finish_copy */ |
2039 | ||
2040 | ||
2041 | ||
2042 | ||
7e9c6396 TS |
2043 | /* |
2044 | * print out info relating to regions written which consume | |
2045 | * the reservation | |
2046 | */ | |
71e330b5 DC |
2047 | void |
2048 | xlog_print_tic_res( | |
2049 | struct xfs_mount *mp, | |
2050 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
2051 | { |
2052 | uint i; | |
2053 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
2054 | ||
2055 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
5110cd82 DW |
2056 | #define REG_TYPE_STR(type, str) [XLOG_REG_TYPE_##type] = str |
2057 | static char *res_type_str[XLOG_REG_TYPE_MAX + 1] = { | |
2058 | REG_TYPE_STR(BFORMAT, "bformat"), | |
2059 | REG_TYPE_STR(BCHUNK, "bchunk"), | |
2060 | REG_TYPE_STR(EFI_FORMAT, "efi_format"), | |
2061 | REG_TYPE_STR(EFD_FORMAT, "efd_format"), | |
2062 | REG_TYPE_STR(IFORMAT, "iformat"), | |
2063 | REG_TYPE_STR(ICORE, "icore"), | |
2064 | REG_TYPE_STR(IEXT, "iext"), | |
2065 | REG_TYPE_STR(IBROOT, "ibroot"), | |
2066 | REG_TYPE_STR(ILOCAL, "ilocal"), | |
2067 | REG_TYPE_STR(IATTR_EXT, "iattr_ext"), | |
2068 | REG_TYPE_STR(IATTR_BROOT, "iattr_broot"), | |
2069 | REG_TYPE_STR(IATTR_LOCAL, "iattr_local"), | |
2070 | REG_TYPE_STR(QFORMAT, "qformat"), | |
2071 | REG_TYPE_STR(DQUOT, "dquot"), | |
2072 | REG_TYPE_STR(QUOTAOFF, "quotaoff"), | |
2073 | REG_TYPE_STR(LRHEADER, "LR header"), | |
2074 | REG_TYPE_STR(UNMOUNT, "unmount"), | |
2075 | REG_TYPE_STR(COMMIT, "commit"), | |
2076 | REG_TYPE_STR(TRANSHDR, "trans header"), | |
2077 | REG_TYPE_STR(ICREATE, "inode create") | |
7e9c6396 | 2078 | }; |
5110cd82 | 2079 | #undef REG_TYPE_STR |
7e9c6396 | 2080 | |
7d2d5653 | 2081 | xfs_warn(mp, "ticket reservation summary:"); |
f41febd2 JP |
2082 | xfs_warn(mp, " unit res = %d bytes", |
2083 | ticket->t_unit_res); | |
2084 | xfs_warn(mp, " current res = %d bytes", | |
2085 | ticket->t_curr_res); | |
2086 | xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)", | |
2087 | ticket->t_res_arr_sum, ticket->t_res_o_flow); | |
2088 | xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)", | |
2089 | ticket->t_res_num_ophdrs, ophdr_spc); | |
2090 | xfs_warn(mp, " ophdr + reg = %u bytes", | |
2091 | ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc); | |
2092 | xfs_warn(mp, " num regions = %u", | |
2093 | ticket->t_res_num); | |
7e9c6396 TS |
2094 | |
2095 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 | 2096 | uint r_type = ticket->t_res_arr[i].r_type; |
08e96e1a | 2097 | xfs_warn(mp, "region[%u]: %s - %u bytes", i, |
7e9c6396 | 2098 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
5110cd82 | 2099 | "bad-rtype" : res_type_str[r_type]), |
7e9c6396 TS |
2100 | ticket->t_res_arr[i].r_len); |
2101 | } | |
2102 | } | |
7e9c6396 | 2103 | |
d4ca1d55 BF |
2104 | /* |
2105 | * Print a summary of the transaction. | |
2106 | */ | |
2107 | void | |
2108 | xlog_print_trans( | |
e6631f85 | 2109 | struct xfs_trans *tp) |
d4ca1d55 | 2110 | { |
e6631f85 DC |
2111 | struct xfs_mount *mp = tp->t_mountp; |
2112 | struct xfs_log_item *lip; | |
d4ca1d55 BF |
2113 | |
2114 | /* dump core transaction and ticket info */ | |
2115 | xfs_warn(mp, "transaction summary:"); | |
2c8f6265 BF |
2116 | xfs_warn(mp, " log res = %d", tp->t_log_res); |
2117 | xfs_warn(mp, " log count = %d", tp->t_log_count); | |
2118 | xfs_warn(mp, " flags = 0x%x", tp->t_flags); | |
d4ca1d55 BF |
2119 | |
2120 | xlog_print_tic_res(mp, tp->t_ticket); | |
2121 | ||
2122 | /* dump each log item */ | |
e6631f85 | 2123 | list_for_each_entry(lip, &tp->t_items, li_trans) { |
d4ca1d55 BF |
2124 | struct xfs_log_vec *lv = lip->li_lv; |
2125 | struct xfs_log_iovec *vec; | |
2126 | int i; | |
2127 | ||
2128 | xfs_warn(mp, "log item: "); | |
2129 | xfs_warn(mp, " type = 0x%x", lip->li_type); | |
22525c17 | 2130 | xfs_warn(mp, " flags = 0x%lx", lip->li_flags); |
d4ca1d55 BF |
2131 | if (!lv) |
2132 | continue; | |
2133 | xfs_warn(mp, " niovecs = %d", lv->lv_niovecs); | |
2134 | xfs_warn(mp, " size = %d", lv->lv_size); | |
2135 | xfs_warn(mp, " bytes = %d", lv->lv_bytes); | |
2136 | xfs_warn(mp, " buf len = %d", lv->lv_buf_len); | |
2137 | ||
2138 | /* dump each iovec for the log item */ | |
2139 | vec = lv->lv_iovecp; | |
2140 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2141 | int dumplen = min(vec->i_len, 32); | |
2142 | ||
2143 | xfs_warn(mp, " iovec[%d]", i); | |
2144 | xfs_warn(mp, " type = 0x%x", vec->i_type); | |
2145 | xfs_warn(mp, " len = %d", vec->i_len); | |
2146 | xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i); | |
244e3dea | 2147 | xfs_hex_dump(vec->i_addr, dumplen); |
d4ca1d55 BF |
2148 | |
2149 | vec++; | |
2150 | } | |
2151 | } | |
2152 | } | |
2153 | ||
b5203cd0 DC |
2154 | /* |
2155 | * Calculate the potential space needed by the log vector. Each region gets | |
2156 | * its own xlog_op_header_t and may need to be double word aligned. | |
2157 | */ | |
2158 | static int | |
2159 | xlog_write_calc_vec_length( | |
2160 | struct xlog_ticket *ticket, | |
55b66332 | 2161 | struct xfs_log_vec *log_vector) |
b5203cd0 | 2162 | { |
55b66332 | 2163 | struct xfs_log_vec *lv; |
b5203cd0 DC |
2164 | int headers = 0; |
2165 | int len = 0; | |
2166 | int i; | |
2167 | ||
2168 | /* acct for start rec of xact */ | |
2169 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2170 | headers++; | |
2171 | ||
55b66332 | 2172 | for (lv = log_vector; lv; lv = lv->lv_next) { |
fd63875c DC |
2173 | /* we don't write ordered log vectors */ |
2174 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) | |
2175 | continue; | |
2176 | ||
55b66332 DC |
2177 | headers += lv->lv_niovecs; |
2178 | ||
2179 | for (i = 0; i < lv->lv_niovecs; i++) { | |
2180 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 2181 | |
55b66332 DC |
2182 | len += vecp->i_len; |
2183 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
2184 | } | |
b5203cd0 DC |
2185 | } |
2186 | ||
2187 | ticket->t_res_num_ophdrs += headers; | |
2188 | len += headers * sizeof(struct xlog_op_header); | |
2189 | ||
2190 | return len; | |
2191 | } | |
2192 | ||
2193 | /* | |
2194 | * If first write for transaction, insert start record We can't be trying to | |
2195 | * commit if we are inited. We can't have any "partial_copy" if we are inited. | |
2196 | */ | |
2197 | static int | |
2198 | xlog_write_start_rec( | |
e6b1f273 | 2199 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2200 | struct xlog_ticket *ticket) |
2201 | { | |
b5203cd0 DC |
2202 | if (!(ticket->t_flags & XLOG_TIC_INITED)) |
2203 | return 0; | |
2204 | ||
2205 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); | |
2206 | ophdr->oh_clientid = ticket->t_clientid; | |
2207 | ophdr->oh_len = 0; | |
2208 | ophdr->oh_flags = XLOG_START_TRANS; | |
2209 | ophdr->oh_res2 = 0; | |
2210 | ||
2211 | ticket->t_flags &= ~XLOG_TIC_INITED; | |
2212 | ||
2213 | return sizeof(struct xlog_op_header); | |
2214 | } | |
2215 | ||
2216 | static xlog_op_header_t * | |
2217 | xlog_write_setup_ophdr( | |
ad223e60 | 2218 | struct xlog *log, |
e6b1f273 | 2219 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
2220 | struct xlog_ticket *ticket, |
2221 | uint flags) | |
2222 | { | |
b5203cd0 DC |
2223 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
2224 | ophdr->oh_clientid = ticket->t_clientid; | |
2225 | ophdr->oh_res2 = 0; | |
2226 | ||
2227 | /* are we copying a commit or unmount record? */ | |
2228 | ophdr->oh_flags = flags; | |
2229 | ||
2230 | /* | |
2231 | * We've seen logs corrupted with bad transaction client ids. This | |
2232 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
2233 | * and shut down the filesystem. | |
2234 | */ | |
2235 | switch (ophdr->oh_clientid) { | |
2236 | case XFS_TRANSACTION: | |
2237 | case XFS_VOLUME: | |
2238 | case XFS_LOG: | |
2239 | break; | |
2240 | default: | |
a0fa2b67 | 2241 | xfs_warn(log->l_mp, |
c9690043 | 2242 | "Bad XFS transaction clientid 0x%x in ticket "PTR_FMT, |
b5203cd0 DC |
2243 | ophdr->oh_clientid, ticket); |
2244 | return NULL; | |
2245 | } | |
2246 | ||
2247 | return ophdr; | |
2248 | } | |
2249 | ||
2250 | /* | |
2251 | * Set up the parameters of the region copy into the log. This has | |
2252 | * to handle region write split across multiple log buffers - this | |
2253 | * state is kept external to this function so that this code can | |
ac0e300f | 2254 | * be written in an obvious, self documenting manner. |
b5203cd0 DC |
2255 | */ |
2256 | static int | |
2257 | xlog_write_setup_copy( | |
2258 | struct xlog_ticket *ticket, | |
2259 | struct xlog_op_header *ophdr, | |
2260 | int space_available, | |
2261 | int space_required, | |
2262 | int *copy_off, | |
2263 | int *copy_len, | |
2264 | int *last_was_partial_copy, | |
2265 | int *bytes_consumed) | |
2266 | { | |
2267 | int still_to_copy; | |
2268 | ||
2269 | still_to_copy = space_required - *bytes_consumed; | |
2270 | *copy_off = *bytes_consumed; | |
2271 | ||
2272 | if (still_to_copy <= space_available) { | |
2273 | /* write of region completes here */ | |
2274 | *copy_len = still_to_copy; | |
2275 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2276 | if (*last_was_partial_copy) | |
2277 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
2278 | *last_was_partial_copy = 0; | |
2279 | *bytes_consumed = 0; | |
2280 | return 0; | |
2281 | } | |
2282 | ||
2283 | /* partial write of region, needs extra log op header reservation */ | |
2284 | *copy_len = space_available; | |
2285 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
2286 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
2287 | if (*last_was_partial_copy) | |
2288 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
2289 | *bytes_consumed += *copy_len; | |
2290 | (*last_was_partial_copy)++; | |
2291 | ||
2292 | /* account for new log op header */ | |
2293 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
2294 | ticket->t_res_num_ophdrs++; | |
2295 | ||
2296 | return sizeof(struct xlog_op_header); | |
2297 | } | |
2298 | ||
2299 | static int | |
2300 | xlog_write_copy_finish( | |
ad223e60 | 2301 | struct xlog *log, |
b5203cd0 DC |
2302 | struct xlog_in_core *iclog, |
2303 | uint flags, | |
2304 | int *record_cnt, | |
2305 | int *data_cnt, | |
2306 | int *partial_copy, | |
2307 | int *partial_copy_len, | |
2308 | int log_offset, | |
2309 | struct xlog_in_core **commit_iclog) | |
2310 | { | |
2311 | if (*partial_copy) { | |
2312 | /* | |
2313 | * This iclog has already been marked WANT_SYNC by | |
2314 | * xlog_state_get_iclog_space. | |
2315 | */ | |
2316 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2317 | *record_cnt = 0; | |
2318 | *data_cnt = 0; | |
2319 | return xlog_state_release_iclog(log, iclog); | |
2320 | } | |
2321 | ||
2322 | *partial_copy = 0; | |
2323 | *partial_copy_len = 0; | |
2324 | ||
2325 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2326 | /* no more space in this iclog - push it. */ | |
2327 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2328 | *record_cnt = 0; | |
2329 | *data_cnt = 0; | |
2330 | ||
2331 | spin_lock(&log->l_icloglock); | |
2332 | xlog_state_want_sync(log, iclog); | |
2333 | spin_unlock(&log->l_icloglock); | |
2334 | ||
2335 | if (!commit_iclog) | |
2336 | return xlog_state_release_iclog(log, iclog); | |
2337 | ASSERT(flags & XLOG_COMMIT_TRANS); | |
2338 | *commit_iclog = iclog; | |
2339 | } | |
2340 | ||
2341 | return 0; | |
2342 | } | |
2343 | ||
1da177e4 LT |
2344 | /* |
2345 | * Write some region out to in-core log | |
2346 | * | |
2347 | * This will be called when writing externally provided regions or when | |
2348 | * writing out a commit record for a given transaction. | |
2349 | * | |
2350 | * General algorithm: | |
2351 | * 1. Find total length of this write. This may include adding to the | |
2352 | * lengths passed in. | |
2353 | * 2. Check whether we violate the tickets reservation. | |
2354 | * 3. While writing to this iclog | |
2355 | * A. Reserve as much space in this iclog as can get | |
2356 | * B. If this is first write, save away start lsn | |
2357 | * C. While writing this region: | |
2358 | * 1. If first write of transaction, write start record | |
2359 | * 2. Write log operation header (header per region) | |
2360 | * 3. Find out if we can fit entire region into this iclog | |
2361 | * 4. Potentially, verify destination memcpy ptr | |
2362 | * 5. Memcpy (partial) region | |
2363 | * 6. If partial copy, release iclog; otherwise, continue | |
2364 | * copying more regions into current iclog | |
2365 | * 4. Mark want sync bit (in simulation mode) | |
2366 | * 5. Release iclog for potential flush to on-disk log. | |
2367 | * | |
2368 | * ERRORS: | |
2369 | * 1. Panic if reservation is overrun. This should never happen since | |
2370 | * reservation amounts are generated internal to the filesystem. | |
2371 | * NOTES: | |
2372 | * 1. Tickets are single threaded data structures. | |
2373 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2374 | * syncing routine. When a single log_write region needs to span | |
2375 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2376 | * on all log operation writes which don't contain the end of the | |
2377 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2378 | * operation which contains the end of the continued log_write region. | |
2379 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2380 | * we don't really know exactly how much space will be used. As a result, | |
2381 | * we don't update ic_offset until the end when we know exactly how many | |
2382 | * bytes have been written out. | |
2383 | */ | |
71e330b5 | 2384 | int |
35a8a72f | 2385 | xlog_write( |
ad223e60 | 2386 | struct xlog *log, |
55b66332 | 2387 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2388 | struct xlog_ticket *ticket, |
2389 | xfs_lsn_t *start_lsn, | |
2390 | struct xlog_in_core **commit_iclog, | |
2391 | uint flags) | |
1da177e4 | 2392 | { |
99428ad0 | 2393 | struct xlog_in_core *iclog = NULL; |
55b66332 DC |
2394 | struct xfs_log_iovec *vecp; |
2395 | struct xfs_log_vec *lv; | |
99428ad0 CH |
2396 | int len; |
2397 | int index; | |
2398 | int partial_copy = 0; | |
2399 | int partial_copy_len = 0; | |
2400 | int contwr = 0; | |
2401 | int record_cnt = 0; | |
2402 | int data_cnt = 0; | |
2403 | int error; | |
2404 | ||
2405 | *start_lsn = 0; | |
2406 | ||
55b66332 | 2407 | len = xlog_write_calc_vec_length(ticket, log_vector); |
71e330b5 | 2408 | |
93b8a585 CH |
2409 | /* |
2410 | * Region headers and bytes are already accounted for. | |
2411 | * We only need to take into account start records and | |
2412 | * split regions in this function. | |
2413 | */ | |
2414 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2415 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
2416 | ||
2417 | /* | |
2418 | * Commit record headers need to be accounted for. These | |
2419 | * come in as separate writes so are easy to detect. | |
2420 | */ | |
2421 | if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS)) | |
2422 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
71e330b5 | 2423 | |
7d2d5653 BF |
2424 | if (ticket->t_curr_res < 0) { |
2425 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
2426 | "ctx ticket reservation ran out. Need to up reservation"); | |
55b66332 | 2427 | xlog_print_tic_res(log->l_mp, ticket); |
7d2d5653 BF |
2428 | xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR); |
2429 | } | |
1da177e4 | 2430 | |
55b66332 DC |
2431 | index = 0; |
2432 | lv = log_vector; | |
2433 | vecp = lv->lv_iovecp; | |
fd63875c | 2434 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
e6b1f273 | 2435 | void *ptr; |
99428ad0 | 2436 | int log_offset; |
1da177e4 | 2437 | |
99428ad0 CH |
2438 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2439 | &contwr, &log_offset); | |
2440 | if (error) | |
2441 | return error; | |
1da177e4 | 2442 | |
99428ad0 | 2443 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2444 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2445 | |
99428ad0 CH |
2446 | /* start_lsn is the first lsn written to. That's all we need. */ |
2447 | if (!*start_lsn) | |
2448 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
b5203cd0 | 2449 | |
99428ad0 CH |
2450 | /* |
2451 | * This loop writes out as many regions as can fit in the amount | |
2452 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2453 | */ | |
fd63875c DC |
2454 | while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { |
2455 | struct xfs_log_iovec *reg; | |
99428ad0 CH |
2456 | struct xlog_op_header *ophdr; |
2457 | int start_rec_copy; | |
2458 | int copy_len; | |
2459 | int copy_off; | |
fd63875c DC |
2460 | bool ordered = false; |
2461 | ||
2462 | /* ordered log vectors have no regions to write */ | |
2463 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
2464 | ASSERT(lv->lv_niovecs == 0); | |
2465 | ordered = true; | |
2466 | goto next_lv; | |
2467 | } | |
99428ad0 | 2468 | |
fd63875c | 2469 | reg = &vecp[index]; |
c8ce540d DW |
2470 | ASSERT(reg->i_len % sizeof(int32_t) == 0); |
2471 | ASSERT((unsigned long)ptr % sizeof(int32_t) == 0); | |
99428ad0 CH |
2472 | |
2473 | start_rec_copy = xlog_write_start_rec(ptr, ticket); | |
2474 | if (start_rec_copy) { | |
2475 | record_cnt++; | |
e6b1f273 | 2476 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2477 | start_rec_copy); |
2478 | } | |
b5203cd0 | 2479 | |
99428ad0 CH |
2480 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags); |
2481 | if (!ophdr) | |
2451337d | 2482 | return -EIO; |
99428ad0 | 2483 | |
e6b1f273 | 2484 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2485 | sizeof(struct xlog_op_header)); |
2486 | ||
2487 | len += xlog_write_setup_copy(ticket, ophdr, | |
2488 | iclog->ic_size-log_offset, | |
55b66332 | 2489 | reg->i_len, |
99428ad0 CH |
2490 | ©_off, ©_len, |
2491 | &partial_copy, | |
2492 | &partial_copy_len); | |
2493 | xlog_verify_dest_ptr(log, ptr); | |
2494 | ||
91f9f5fe ES |
2495 | /* |
2496 | * Copy region. | |
2497 | * | |
2498 | * Unmount records just log an opheader, so can have | |
2499 | * empty payloads with no data region to copy. Hence we | |
2500 | * only copy the payload if the vector says it has data | |
2501 | * to copy. | |
2502 | */ | |
99428ad0 | 2503 | ASSERT(copy_len >= 0); |
91f9f5fe ES |
2504 | if (copy_len > 0) { |
2505 | memcpy(ptr, reg->i_addr + copy_off, copy_len); | |
2506 | xlog_write_adv_cnt(&ptr, &len, &log_offset, | |
2507 | copy_len); | |
2508 | } | |
99428ad0 CH |
2509 | copy_len += start_rec_copy + sizeof(xlog_op_header_t); |
2510 | record_cnt++; | |
2511 | data_cnt += contwr ? copy_len : 0; | |
2512 | ||
2513 | error = xlog_write_copy_finish(log, iclog, flags, | |
2514 | &record_cnt, &data_cnt, | |
2515 | &partial_copy, | |
2516 | &partial_copy_len, | |
2517 | log_offset, | |
2518 | commit_iclog); | |
2519 | if (error) | |
2520 | return error; | |
2521 | ||
2522 | /* | |
2523 | * if we had a partial copy, we need to get more iclog | |
2524 | * space but we don't want to increment the region | |
2525 | * index because there is still more is this region to | |
2526 | * write. | |
2527 | * | |
2528 | * If we completed writing this region, and we flushed | |
2529 | * the iclog (indicated by resetting of the record | |
2530 | * count), then we also need to get more log space. If | |
2531 | * this was the last record, though, we are done and | |
2532 | * can just return. | |
2533 | */ | |
2534 | if (partial_copy) | |
2535 | break; | |
2536 | ||
55b66332 | 2537 | if (++index == lv->lv_niovecs) { |
fd63875c | 2538 | next_lv: |
55b66332 DC |
2539 | lv = lv->lv_next; |
2540 | index = 0; | |
2541 | if (lv) | |
2542 | vecp = lv->lv_iovecp; | |
2543 | } | |
749f24f3 | 2544 | if (record_cnt == 0 && !ordered) { |
55b66332 | 2545 | if (!lv) |
99428ad0 CH |
2546 | return 0; |
2547 | break; | |
2548 | } | |
2549 | } | |
2550 | } | |
2551 | ||
2552 | ASSERT(len == 0); | |
2553 | ||
2554 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); | |
2555 | if (!commit_iclog) | |
2556 | return xlog_state_release_iclog(log, iclog); | |
1da177e4 | 2557 | |
1da177e4 LT |
2558 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2559 | *commit_iclog = iclog; | |
2560 | return 0; | |
99428ad0 | 2561 | } |
1da177e4 LT |
2562 | |
2563 | ||
2564 | /***************************************************************************** | |
2565 | * | |
2566 | * State Machine functions | |
2567 | * | |
2568 | ***************************************************************************** | |
2569 | */ | |
2570 | ||
2571 | /* Clean iclogs starting from the head. This ordering must be | |
2572 | * maintained, so an iclog doesn't become ACTIVE beyond one that | |
2573 | * is SYNCING. This is also required to maintain the notion that we use | |
12017faf | 2574 | * a ordered wait queue to hold off would be writers to the log when every |
1da177e4 LT |
2575 | * iclog is trying to sync to disk. |
2576 | * | |
2577 | * State Change: DIRTY -> ACTIVE | |
2578 | */ | |
ba0f32d4 | 2579 | STATIC void |
9a8d2fdb MT |
2580 | xlog_state_clean_log( |
2581 | struct xlog *log) | |
1da177e4 LT |
2582 | { |
2583 | xlog_in_core_t *iclog; | |
2584 | int changed = 0; | |
2585 | ||
2586 | iclog = log->l_iclog; | |
2587 | do { | |
2588 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
2589 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2590 | iclog->ic_offset = 0; | |
114d23aa | 2591 | ASSERT(iclog->ic_callback == NULL); |
1da177e4 LT |
2592 | /* |
2593 | * If the number of ops in this iclog indicate it just | |
2594 | * contains the dummy transaction, we can | |
2595 | * change state into IDLE (the second time around). | |
2596 | * Otherwise we should change the state into | |
2597 | * NEED a dummy. | |
2598 | * We don't need to cover the dummy. | |
2599 | */ | |
2600 | if (!changed && | |
b53e675d CH |
2601 | (be32_to_cpu(iclog->ic_header.h_num_logops) == |
2602 | XLOG_COVER_OPS)) { | |
1da177e4 LT |
2603 | changed = 1; |
2604 | } else { | |
2605 | /* | |
2606 | * We have two dirty iclogs so start over | |
2607 | * This could also be num of ops indicates | |
2608 | * this is not the dummy going out. | |
2609 | */ | |
2610 | changed = 2; | |
2611 | } | |
2612 | iclog->ic_header.h_num_logops = 0; | |
2613 | memset(iclog->ic_header.h_cycle_data, 0, | |
2614 | sizeof(iclog->ic_header.h_cycle_data)); | |
2615 | iclog->ic_header.h_lsn = 0; | |
2616 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) | |
2617 | /* do nothing */; | |
2618 | else | |
2619 | break; /* stop cleaning */ | |
2620 | iclog = iclog->ic_next; | |
2621 | } while (iclog != log->l_iclog); | |
2622 | ||
2623 | /* log is locked when we are called */ | |
2624 | /* | |
2625 | * Change state for the dummy log recording. | |
2626 | * We usually go to NEED. But we go to NEED2 if the changed indicates | |
2627 | * we are done writing the dummy record. | |
2628 | * If we are done with the second dummy recored (DONE2), then | |
2629 | * we go to IDLE. | |
2630 | */ | |
2631 | if (changed) { | |
2632 | switch (log->l_covered_state) { | |
2633 | case XLOG_STATE_COVER_IDLE: | |
2634 | case XLOG_STATE_COVER_NEED: | |
2635 | case XLOG_STATE_COVER_NEED2: | |
2636 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2637 | break; | |
2638 | ||
2639 | case XLOG_STATE_COVER_DONE: | |
2640 | if (changed == 1) | |
2641 | log->l_covered_state = XLOG_STATE_COVER_NEED2; | |
2642 | else | |
2643 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2644 | break; | |
2645 | ||
2646 | case XLOG_STATE_COVER_DONE2: | |
2647 | if (changed == 1) | |
2648 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
2649 | else | |
2650 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2651 | break; | |
2652 | ||
2653 | default: | |
2654 | ASSERT(0); | |
2655 | } | |
2656 | } | |
2657 | } /* xlog_state_clean_log */ | |
2658 | ||
2659 | STATIC xfs_lsn_t | |
2660 | xlog_get_lowest_lsn( | |
9a8d2fdb | 2661 | struct xlog *log) |
1da177e4 LT |
2662 | { |
2663 | xlog_in_core_t *lsn_log; | |
2664 | xfs_lsn_t lowest_lsn, lsn; | |
2665 | ||
2666 | lsn_log = log->l_iclog; | |
2667 | lowest_lsn = 0; | |
2668 | do { | |
2669 | if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) { | |
b53e675d | 2670 | lsn = be64_to_cpu(lsn_log->ic_header.h_lsn); |
1da177e4 LT |
2671 | if ((lsn && !lowest_lsn) || |
2672 | (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) { | |
2673 | lowest_lsn = lsn; | |
2674 | } | |
2675 | } | |
2676 | lsn_log = lsn_log->ic_next; | |
2677 | } while (lsn_log != log->l_iclog); | |
014c2544 | 2678 | return lowest_lsn; |
1da177e4 LT |
2679 | } |
2680 | ||
2681 | ||
2682 | STATIC void | |
2683 | xlog_state_do_callback( | |
9a8d2fdb MT |
2684 | struct xlog *log, |
2685 | int aborted, | |
2686 | struct xlog_in_core *ciclog) | |
1da177e4 LT |
2687 | { |
2688 | xlog_in_core_t *iclog; | |
2689 | xlog_in_core_t *first_iclog; /* used to know when we've | |
2690 | * processed all iclogs once */ | |
2691 | xfs_log_callback_t *cb, *cb_next; | |
2692 | int flushcnt = 0; | |
2693 | xfs_lsn_t lowest_lsn; | |
2694 | int ioerrors; /* counter: iclogs with errors */ | |
2695 | int loopdidcallbacks; /* flag: inner loop did callbacks*/ | |
2696 | int funcdidcallbacks; /* flag: function did callbacks */ | |
2697 | int repeats; /* for issuing console warnings if | |
2698 | * looping too many times */ | |
d748c623 | 2699 | int wake = 0; |
1da177e4 | 2700 | |
b22cd72c | 2701 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2702 | first_iclog = iclog = log->l_iclog; |
2703 | ioerrors = 0; | |
2704 | funcdidcallbacks = 0; | |
2705 | repeats = 0; | |
2706 | ||
2707 | do { | |
2708 | /* | |
2709 | * Scan all iclogs starting with the one pointed to by the | |
2710 | * log. Reset this starting point each time the log is | |
2711 | * unlocked (during callbacks). | |
2712 | * | |
2713 | * Keep looping through iclogs until one full pass is made | |
2714 | * without running any callbacks. | |
2715 | */ | |
2716 | first_iclog = log->l_iclog; | |
2717 | iclog = log->l_iclog; | |
2718 | loopdidcallbacks = 0; | |
2719 | repeats++; | |
2720 | ||
2721 | do { | |
2722 | ||
2723 | /* skip all iclogs in the ACTIVE & DIRTY states */ | |
2724 | if (iclog->ic_state & | |
2725 | (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) { | |
2726 | iclog = iclog->ic_next; | |
2727 | continue; | |
2728 | } | |
2729 | ||
2730 | /* | |
2731 | * Between marking a filesystem SHUTDOWN and stopping | |
2732 | * the log, we do flush all iclogs to disk (if there | |
2733 | * wasn't a log I/O error). So, we do want things to | |
2734 | * go smoothly in case of just a SHUTDOWN w/o a | |
2735 | * LOG_IO_ERROR. | |
2736 | */ | |
2737 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
2738 | /* | |
2739 | * Can only perform callbacks in order. Since | |
2740 | * this iclog is not in the DONE_SYNC/ | |
2741 | * DO_CALLBACK state, we skip the rest and | |
2742 | * just try to clean up. If we set our iclog | |
2743 | * to DO_CALLBACK, we will not process it when | |
2744 | * we retry since a previous iclog is in the | |
2745 | * CALLBACK and the state cannot change since | |
b22cd72c | 2746 | * we are holding the l_icloglock. |
1da177e4 LT |
2747 | */ |
2748 | if (!(iclog->ic_state & | |
2749 | (XLOG_STATE_DONE_SYNC | | |
2750 | XLOG_STATE_DO_CALLBACK))) { | |
2751 | if (ciclog && (ciclog->ic_state == | |
2752 | XLOG_STATE_DONE_SYNC)) { | |
2753 | ciclog->ic_state = XLOG_STATE_DO_CALLBACK; | |
2754 | } | |
2755 | break; | |
2756 | } | |
2757 | /* | |
2758 | * We now have an iclog that is in either the | |
2759 | * DO_CALLBACK or DONE_SYNC states. The other | |
2760 | * states (WANT_SYNC, SYNCING, or CALLBACK were | |
2761 | * caught by the above if and are going to | |
2762 | * clean (i.e. we aren't doing their callbacks) | |
2763 | * see the above if. | |
2764 | */ | |
2765 | ||
2766 | /* | |
2767 | * We will do one more check here to see if we | |
2768 | * have chased our tail around. | |
2769 | */ | |
2770 | ||
2771 | lowest_lsn = xlog_get_lowest_lsn(log); | |
b53e675d CH |
2772 | if (lowest_lsn && |
2773 | XFS_LSN_CMP(lowest_lsn, | |
84f3c683 | 2774 | be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { |
1da177e4 LT |
2775 | iclog = iclog->ic_next; |
2776 | continue; /* Leave this iclog for | |
2777 | * another thread */ | |
2778 | } | |
2779 | ||
2780 | iclog->ic_state = XLOG_STATE_CALLBACK; | |
2781 | ||
1da177e4 | 2782 | |
84f3c683 | 2783 | /* |
d35e88fa DC |
2784 | * Completion of a iclog IO does not imply that |
2785 | * a transaction has completed, as transactions | |
2786 | * can be large enough to span many iclogs. We | |
2787 | * cannot change the tail of the log half way | |
2788 | * through a transaction as this may be the only | |
2789 | * transaction in the log and moving th etail to | |
2790 | * point to the middle of it will prevent | |
2791 | * recovery from finding the start of the | |
2792 | * transaction. Hence we should only update the | |
2793 | * last_sync_lsn if this iclog contains | |
2794 | * transaction completion callbacks on it. | |
2795 | * | |
2796 | * We have to do this before we drop the | |
84f3c683 DC |
2797 | * icloglock to ensure we are the only one that |
2798 | * can update it. | |
1da177e4 | 2799 | */ |
84f3c683 DC |
2800 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), |
2801 | be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); | |
d35e88fa DC |
2802 | if (iclog->ic_callback) |
2803 | atomic64_set(&log->l_last_sync_lsn, | |
2804 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1da177e4 | 2805 | |
84f3c683 | 2806 | } else |
1da177e4 | 2807 | ioerrors++; |
84f3c683 DC |
2808 | |
2809 | spin_unlock(&log->l_icloglock); | |
1da177e4 | 2810 | |
114d23aa DC |
2811 | /* |
2812 | * Keep processing entries in the callback list until | |
2813 | * we come around and it is empty. We need to | |
2814 | * atomically see that the list is empty and change the | |
2815 | * state to DIRTY so that we don't miss any more | |
2816 | * callbacks being added. | |
2817 | */ | |
2818 | spin_lock(&iclog->ic_callback_lock); | |
2819 | cb = iclog->ic_callback; | |
4b80916b | 2820 | while (cb) { |
1da177e4 LT |
2821 | iclog->ic_callback_tail = &(iclog->ic_callback); |
2822 | iclog->ic_callback = NULL; | |
114d23aa | 2823 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2824 | |
2825 | /* perform callbacks in the order given */ | |
4b80916b | 2826 | for (; cb; cb = cb_next) { |
1da177e4 LT |
2827 | cb_next = cb->cb_next; |
2828 | cb->cb_func(cb->cb_arg, aborted); | |
2829 | } | |
114d23aa | 2830 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
2831 | cb = iclog->ic_callback; |
2832 | } | |
2833 | ||
2834 | loopdidcallbacks++; | |
2835 | funcdidcallbacks++; | |
2836 | ||
114d23aa | 2837 | spin_lock(&log->l_icloglock); |
4b80916b | 2838 | ASSERT(iclog->ic_callback == NULL); |
114d23aa | 2839 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2840 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) |
2841 | iclog->ic_state = XLOG_STATE_DIRTY; | |
2842 | ||
2843 | /* | |
2844 | * Transition from DIRTY to ACTIVE if applicable. | |
2845 | * NOP if STATE_IOERROR. | |
2846 | */ | |
2847 | xlog_state_clean_log(log); | |
2848 | ||
2849 | /* wake up threads waiting in xfs_log_force() */ | |
eb40a875 | 2850 | wake_up_all(&iclog->ic_force_wait); |
1da177e4 LT |
2851 | |
2852 | iclog = iclog->ic_next; | |
2853 | } while (first_iclog != iclog); | |
a3c6685e NS |
2854 | |
2855 | if (repeats > 5000) { | |
2856 | flushcnt += repeats; | |
2857 | repeats = 0; | |
a0fa2b67 | 2858 | xfs_warn(log->l_mp, |
a3c6685e | 2859 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2860 | __func__, flushcnt); |
1da177e4 LT |
2861 | } |
2862 | } while (!ioerrors && loopdidcallbacks); | |
2863 | ||
609adfc2 | 2864 | #ifdef DEBUG |
1da177e4 | 2865 | /* |
609adfc2 BF |
2866 | * Make one last gasp attempt to see if iclogs are being left in limbo. |
2867 | * If the above loop finds an iclog earlier than the current iclog and | |
2868 | * in one of the syncing states, the current iclog is put into | |
2869 | * DO_CALLBACK and the callbacks are deferred to the completion of the | |
2870 | * earlier iclog. Walk the iclogs in order and make sure that no iclog | |
2871 | * is in DO_CALLBACK unless an earlier iclog is in one of the syncing | |
2872 | * states. | |
2873 | * | |
2874 | * Note that SYNCING|IOABORT is a valid state so we cannot just check | |
2875 | * for ic_state == SYNCING. | |
1da177e4 | 2876 | */ |
1da177e4 LT |
2877 | if (funcdidcallbacks) { |
2878 | first_iclog = iclog = log->l_iclog; | |
2879 | do { | |
2880 | ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK); | |
2881 | /* | |
2882 | * Terminate the loop if iclogs are found in states | |
2883 | * which will cause other threads to clean up iclogs. | |
2884 | * | |
2885 | * SYNCING - i/o completion will go through logs | |
2886 | * DONE_SYNC - interrupt thread should be waiting for | |
b22cd72c | 2887 | * l_icloglock |
1da177e4 LT |
2888 | * IOERROR - give up hope all ye who enter here |
2889 | */ | |
2890 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
609adfc2 | 2891 | iclog->ic_state & XLOG_STATE_SYNCING || |
1da177e4 LT |
2892 | iclog->ic_state == XLOG_STATE_DONE_SYNC || |
2893 | iclog->ic_state == XLOG_STATE_IOERROR ) | |
2894 | break; | |
2895 | iclog = iclog->ic_next; | |
2896 | } while (first_iclog != iclog); | |
2897 | } | |
2898 | #endif | |
2899 | ||
d748c623 MW |
2900 | if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) |
2901 | wake = 1; | |
b22cd72c | 2902 | spin_unlock(&log->l_icloglock); |
d748c623 MW |
2903 | |
2904 | if (wake) | |
eb40a875 | 2905 | wake_up_all(&log->l_flush_wait); |
d748c623 | 2906 | } |
1da177e4 LT |
2907 | |
2908 | ||
2909 | /* | |
2910 | * Finish transitioning this iclog to the dirty state. | |
2911 | * | |
2912 | * Make sure that we completely execute this routine only when this is | |
2913 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2914 | * when we reach the end of the physical log, get turned into 2 separate | |
2915 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2916 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2917 | * the second completion goes through. | |
2918 | * | |
2919 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2920 | * global state machine log lock. |
1da177e4 | 2921 | */ |
a8272ce0 | 2922 | STATIC void |
1da177e4 LT |
2923 | xlog_state_done_syncing( |
2924 | xlog_in_core_t *iclog, | |
2925 | int aborted) | |
2926 | { | |
9a8d2fdb | 2927 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2928 | |
b22cd72c | 2929 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2930 | |
2931 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING || | |
2932 | iclog->ic_state == XLOG_STATE_IOERROR); | |
155cc6b7 | 2933 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
2934 | ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2); |
2935 | ||
2936 | ||
2937 | /* | |
2938 | * If we got an error, either on the first buffer, or in the case of | |
2939 | * split log writes, on the second, we mark ALL iclogs STATE_IOERROR, | |
2940 | * and none should ever be attempted to be written to disk | |
2941 | * again. | |
2942 | */ | |
2943 | if (iclog->ic_state != XLOG_STATE_IOERROR) { | |
2944 | if (--iclog->ic_bwritecnt == 1) { | |
b22cd72c | 2945 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2946 | return; |
2947 | } | |
2948 | iclog->ic_state = XLOG_STATE_DONE_SYNC; | |
2949 | } | |
2950 | ||
2951 | /* | |
2952 | * Someone could be sleeping prior to writing out the next | |
2953 | * iclog buffer, we wake them all, one will get to do the | |
2954 | * I/O, the others get to wait for the result. | |
2955 | */ | |
eb40a875 | 2956 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2957 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2958 | xlog_state_do_callback(log, aborted, iclog); /* also cleans log */ |
2959 | } /* xlog_state_done_syncing */ | |
2960 | ||
2961 | ||
2962 | /* | |
2963 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2964 | * sleep. We wait on the flush queue on the head iclog as that should be |
2965 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2966 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2967 | * |
2968 | * The in-core logs are used in a circular fashion. They are not used | |
2969 | * out-of-order even when an iclog past the head is free. | |
2970 | * | |
2971 | * return: | |
2972 | * * log_offset where xlog_write() can start writing into the in-core | |
2973 | * log's data space. | |
2974 | * * in-core log pointer to which xlog_write() should write. | |
2975 | * * boolean indicating this is a continued write to an in-core log. | |
2976 | * If this is the last write, then the in-core log's offset field | |
2977 | * needs to be incremented, depending on the amount of data which | |
2978 | * is copied. | |
2979 | */ | |
a8272ce0 | 2980 | STATIC int |
9a8d2fdb MT |
2981 | xlog_state_get_iclog_space( |
2982 | struct xlog *log, | |
2983 | int len, | |
2984 | struct xlog_in_core **iclogp, | |
2985 | struct xlog_ticket *ticket, | |
2986 | int *continued_write, | |
2987 | int *logoffsetp) | |
1da177e4 | 2988 | { |
1da177e4 LT |
2989 | int log_offset; |
2990 | xlog_rec_header_t *head; | |
2991 | xlog_in_core_t *iclog; | |
2992 | int error; | |
2993 | ||
2994 | restart: | |
b22cd72c | 2995 | spin_lock(&log->l_icloglock); |
1da177e4 | 2996 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2997 | spin_unlock(&log->l_icloglock); |
2451337d | 2998 | return -EIO; |
1da177e4 LT |
2999 | } |
3000 | ||
3001 | iclog = log->l_iclog; | |
d748c623 | 3002 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
ff6d6af2 | 3003 | XFS_STATS_INC(log->l_mp, xs_log_noiclogs); |
d748c623 MW |
3004 | |
3005 | /* Wait for log writes to have flushed */ | |
eb40a875 | 3006 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
3007 | goto restart; |
3008 | } | |
d748c623 | 3009 | |
1da177e4 LT |
3010 | head = &iclog->ic_header; |
3011 | ||
155cc6b7 | 3012 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
3013 | log_offset = iclog->ic_offset; |
3014 | ||
3015 | /* On the 1st write to an iclog, figure out lsn. This works | |
3016 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
3017 | * committing to. If the offset is set, that's how many blocks | |
3018 | * must be written. | |
3019 | */ | |
3020 | if (log_offset == 0) { | |
3021 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 3022 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
3023 | log->l_iclog_hsize, |
3024 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
3025 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
3026 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 3027 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
3028 | ASSERT(log->l_curr_block >= 0); |
3029 | } | |
3030 | ||
3031 | /* If there is enough room to write everything, then do it. Otherwise, | |
3032 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
3033 | * bit is on, so this will get flushed out. Don't update ic_offset | |
3034 | * until you know exactly how many bytes get copied. Therefore, wait | |
3035 | * until later to update ic_offset. | |
3036 | * | |
3037 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
3038 | * can fit into remaining data section. | |
3039 | */ | |
3040 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
3041 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
3042 | ||
49641f1a DC |
3043 | /* |
3044 | * If I'm the only one writing to this iclog, sync it to disk. | |
3045 | * We need to do an atomic compare and decrement here to avoid | |
3046 | * racing with concurrent atomic_dec_and_lock() calls in | |
3047 | * xlog_state_release_iclog() when there is more than one | |
3048 | * reference to the iclog. | |
3049 | */ | |
3050 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) { | |
3051 | /* we are the only one */ | |
b22cd72c | 3052 | spin_unlock(&log->l_icloglock); |
49641f1a DC |
3053 | error = xlog_state_release_iclog(log, iclog); |
3054 | if (error) | |
014c2544 | 3055 | return error; |
1da177e4 | 3056 | } else { |
b22cd72c | 3057 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3058 | } |
3059 | goto restart; | |
3060 | } | |
3061 | ||
3062 | /* Do we have enough room to write the full amount in the remainder | |
3063 | * of this iclog? Or must we continue a write on the next iclog and | |
3064 | * mark this iclog as completely taken? In the case where we switch | |
3065 | * iclogs (to mark it taken), this particular iclog will release/sync | |
3066 | * to disk in xlog_write(). | |
3067 | */ | |
3068 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
3069 | *continued_write = 0; | |
3070 | iclog->ic_offset += len; | |
3071 | } else { | |
3072 | *continued_write = 1; | |
3073 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
3074 | } | |
3075 | *iclogp = iclog; | |
3076 | ||
3077 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 3078 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3079 | |
3080 | *logoffsetp = log_offset; | |
3081 | return 0; | |
3082 | } /* xlog_state_get_iclog_space */ | |
3083 | ||
1da177e4 LT |
3084 | /* The first cnt-1 times through here we don't need to |
3085 | * move the grant write head because the permanent | |
3086 | * reservation has reserved cnt times the unit amount. | |
3087 | * Release part of current permanent unit reservation and | |
3088 | * reset current reservation to be one units worth. Also | |
3089 | * move grant reservation head forward. | |
3090 | */ | |
3091 | STATIC void | |
9a8d2fdb MT |
3092 | xlog_regrant_reserve_log_space( |
3093 | struct xlog *log, | |
3094 | struct xlog_ticket *ticket) | |
1da177e4 | 3095 | { |
0b1b213f CH |
3096 | trace_xfs_log_regrant_reserve_enter(log, ticket); |
3097 | ||
1da177e4 LT |
3098 | if (ticket->t_cnt > 0) |
3099 | ticket->t_cnt--; | |
3100 | ||
28496968 | 3101 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3102 | ticket->t_curr_res); |
28496968 | 3103 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 3104 | ticket->t_curr_res); |
1da177e4 | 3105 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3106 | xlog_tic_reset_res(ticket); |
0b1b213f CH |
3107 | |
3108 | trace_xfs_log_regrant_reserve_sub(log, ticket); | |
3109 | ||
1da177e4 | 3110 | /* just return if we still have some of the pre-reserved space */ |
d0eb2f38 | 3111 | if (ticket->t_cnt > 0) |
1da177e4 | 3112 | return; |
1da177e4 | 3113 | |
28496968 | 3114 | xlog_grant_add_space(log, &log->l_reserve_head.grant, |
a69ed03c | 3115 | ticket->t_unit_res); |
0b1b213f CH |
3116 | |
3117 | trace_xfs_log_regrant_reserve_exit(log, ticket); | |
3118 | ||
1da177e4 | 3119 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 3120 | xlog_tic_reset_res(ticket); |
1da177e4 LT |
3121 | } /* xlog_regrant_reserve_log_space */ |
3122 | ||
3123 | ||
3124 | /* | |
3125 | * Give back the space left from a reservation. | |
3126 | * | |
3127 | * All the information we need to make a correct determination of space left | |
3128 | * is present. For non-permanent reservations, things are quite easy. The | |
3129 | * count should have been decremented to zero. We only need to deal with the | |
3130 | * space remaining in the current reservation part of the ticket. If the | |
3131 | * ticket contains a permanent reservation, there may be left over space which | |
3132 | * needs to be released. A count of N means that N-1 refills of the current | |
3133 | * reservation can be done before we need to ask for more space. The first | |
3134 | * one goes to fill up the first current reservation. Once we run out of | |
3135 | * space, the count will stay at zero and the only space remaining will be | |
3136 | * in the current reservation field. | |
3137 | */ | |
3138 | STATIC void | |
9a8d2fdb MT |
3139 | xlog_ungrant_log_space( |
3140 | struct xlog *log, | |
3141 | struct xlog_ticket *ticket) | |
1da177e4 | 3142 | { |
663e496a DC |
3143 | int bytes; |
3144 | ||
1da177e4 LT |
3145 | if (ticket->t_cnt > 0) |
3146 | ticket->t_cnt--; | |
3147 | ||
0b1b213f | 3148 | trace_xfs_log_ungrant_enter(log, ticket); |
0b1b213f | 3149 | trace_xfs_log_ungrant_sub(log, ticket); |
1da177e4 | 3150 | |
663e496a DC |
3151 | /* |
3152 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
3153 | * up more space based on the remaining count. |
3154 | */ | |
663e496a | 3155 | bytes = ticket->t_curr_res; |
1da177e4 LT |
3156 | if (ticket->t_cnt > 0) { |
3157 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 3158 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
3159 | } |
3160 | ||
28496968 CH |
3161 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
3162 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 3163 | |
0b1b213f CH |
3164 | trace_xfs_log_ungrant_exit(log, ticket); |
3165 | ||
cfb7cdca | 3166 | xfs_log_space_wake(log->l_mp); |
09a423a3 | 3167 | } |
1da177e4 | 3168 | |
1da177e4 LT |
3169 | /* |
3170 | * Flush iclog to disk if this is the last reference to the given iclog and | |
3171 | * the WANT_SYNC bit is set. | |
3172 | * | |
3173 | * When this function is entered, the iclog is not necessarily in the | |
3174 | * WANT_SYNC state. It may be sitting around waiting to get filled. | |
3175 | * | |
3176 | * | |
3177 | */ | |
a8272ce0 | 3178 | STATIC int |
b589334c | 3179 | xlog_state_release_iclog( |
9a8d2fdb MT |
3180 | struct xlog *log, |
3181 | struct xlog_in_core *iclog) | |
1da177e4 | 3182 | { |
1da177e4 LT |
3183 | int sync = 0; /* do we sync? */ |
3184 | ||
155cc6b7 | 3185 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
2451337d | 3186 | return -EIO; |
155cc6b7 DC |
3187 | |
3188 | ASSERT(atomic_read(&iclog->ic_refcnt) > 0); | |
3189 | if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) | |
3190 | return 0; | |
3191 | ||
1da177e4 | 3192 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
b22cd72c | 3193 | spin_unlock(&log->l_icloglock); |
2451337d | 3194 | return -EIO; |
1da177e4 | 3195 | } |
1da177e4 LT |
3196 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE || |
3197 | iclog->ic_state == XLOG_STATE_WANT_SYNC); | |
3198 | ||
155cc6b7 | 3199 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { |
b589334c | 3200 | /* update tail before writing to iclog */ |
1c3cb9ec | 3201 | xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp); |
1da177e4 LT |
3202 | sync++; |
3203 | iclog->ic_state = XLOG_STATE_SYNCING; | |
1c3cb9ec DC |
3204 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); |
3205 | xlog_verify_tail_lsn(log, iclog, tail_lsn); | |
1da177e4 LT |
3206 | /* cycle incremented when incrementing curr_block */ |
3207 | } | |
b22cd72c | 3208 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3209 | |
3210 | /* | |
3211 | * We let the log lock go, so it's possible that we hit a log I/O | |
c41564b5 | 3212 | * error or some other SHUTDOWN condition that marks the iclog |
1da177e4 LT |
3213 | * as XLOG_STATE_IOERROR before the bwrite. However, we know that |
3214 | * this iclog has consistent data, so we ignore IOERROR | |
3215 | * flags after this point. | |
3216 | */ | |
b589334c | 3217 | if (sync) |
1da177e4 | 3218 | return xlog_sync(log, iclog); |
014c2544 | 3219 | return 0; |
1da177e4 LT |
3220 | } /* xlog_state_release_iclog */ |
3221 | ||
3222 | ||
3223 | /* | |
3224 | * This routine will mark the current iclog in the ring as WANT_SYNC | |
3225 | * and move the current iclog pointer to the next iclog in the ring. | |
3226 | * When this routine is called from xlog_state_get_iclog_space(), the | |
3227 | * exact size of the iclog has not yet been determined. All we know is | |
3228 | * that every data block. We have run out of space in this log record. | |
3229 | */ | |
3230 | STATIC void | |
9a8d2fdb MT |
3231 | xlog_state_switch_iclogs( |
3232 | struct xlog *log, | |
3233 | struct xlog_in_core *iclog, | |
3234 | int eventual_size) | |
1da177e4 LT |
3235 | { |
3236 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
3237 | if (!eventual_size) | |
3238 | eventual_size = iclog->ic_offset; | |
3239 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 3240 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
3241 | log->l_prev_block = log->l_curr_block; |
3242 | log->l_prev_cycle = log->l_curr_cycle; | |
3243 | ||
3244 | /* roll log?: ic_offset changed later */ | |
3245 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
3246 | ||
3247 | /* Round up to next log-sunit */ | |
62118709 | 3248 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
1da177e4 | 3249 | log->l_mp->m_sb.sb_logsunit > 1) { |
c8ce540d | 3250 | uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit); |
1da177e4 LT |
3251 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); |
3252 | } | |
3253 | ||
3254 | if (log->l_curr_block >= log->l_logBBsize) { | |
a45086e2 BF |
3255 | /* |
3256 | * Rewind the current block before the cycle is bumped to make | |
3257 | * sure that the combined LSN never transiently moves forward | |
3258 | * when the log wraps to the next cycle. This is to support the | |
3259 | * unlocked sample of these fields from xlog_valid_lsn(). Most | |
3260 | * other cases should acquire l_icloglock. | |
3261 | */ | |
3262 | log->l_curr_block -= log->l_logBBsize; | |
3263 | ASSERT(log->l_curr_block >= 0); | |
3264 | smp_wmb(); | |
1da177e4 LT |
3265 | log->l_curr_cycle++; |
3266 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
3267 | log->l_curr_cycle++; | |
1da177e4 LT |
3268 | } |
3269 | ASSERT(iclog == log->l_iclog); | |
3270 | log->l_iclog = iclog->ic_next; | |
3271 | } /* xlog_state_switch_iclogs */ | |
3272 | ||
1da177e4 LT |
3273 | /* |
3274 | * Write out all data in the in-core log as of this exact moment in time. | |
3275 | * | |
3276 | * Data may be written to the in-core log during this call. However, | |
3277 | * we don't guarantee this data will be written out. A change from past | |
3278 | * implementation means this routine will *not* write out zero length LRs. | |
3279 | * | |
3280 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
3281 | * If we determine there is no flushable data, we just return. There is no | |
3282 | * flushable data if: | |
3283 | * | |
3284 | * 1. the current iclog is active and has no data; the previous iclog | |
3285 | * is in the active or dirty state. | |
3286 | * 2. the current iclog is drity, and the previous iclog is in the | |
3287 | * active or dirty state. | |
3288 | * | |
12017faf | 3289 | * We may sleep if: |
1da177e4 LT |
3290 | * |
3291 | * 1. the current iclog is not in the active nor dirty state. | |
3292 | * 2. the current iclog dirty, and the previous iclog is not in the | |
3293 | * active nor dirty state. | |
3294 | * 3. the current iclog is active, and there is another thread writing | |
3295 | * to this particular iclog. | |
3296 | * 4. a) the current iclog is active and has no other writers | |
3297 | * b) when we return from flushing out this iclog, it is still | |
3298 | * not in the active nor dirty state. | |
3299 | */ | |
a14a348b | 3300 | int |
60e5bb78 | 3301 | xfs_log_force( |
a14a348b | 3302 | struct xfs_mount *mp, |
60e5bb78 | 3303 | uint flags) |
1da177e4 | 3304 | { |
ad223e60 | 3305 | struct xlog *log = mp->m_log; |
a14a348b CH |
3306 | struct xlog_in_core *iclog; |
3307 | xfs_lsn_t lsn; | |
3308 | ||
ff6d6af2 | 3309 | XFS_STATS_INC(mp, xs_log_force); |
60e5bb78 | 3310 | trace_xfs_log_force(mp, 0, _RET_IP_); |
1da177e4 | 3311 | |
93b8a585 | 3312 | xlog_cil_force(log); |
71e330b5 | 3313 | |
b22cd72c | 3314 | spin_lock(&log->l_icloglock); |
1da177e4 | 3315 | iclog = log->l_iclog; |
e6b96570 CH |
3316 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3317 | goto out_error; | |
1da177e4 | 3318 | |
e6b96570 CH |
3319 | if (iclog->ic_state == XLOG_STATE_DIRTY || |
3320 | (iclog->ic_state == XLOG_STATE_ACTIVE && | |
3321 | atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) { | |
1da177e4 | 3322 | /* |
e6b96570 CH |
3323 | * If the head is dirty or (active and empty), then we need to |
3324 | * look at the previous iclog. | |
3325 | * | |
3326 | * If the previous iclog is active or dirty we are done. There | |
3327 | * is nothing to sync out. Otherwise, we attach ourselves to the | |
1da177e4 LT |
3328 | * previous iclog and go to sleep. |
3329 | */ | |
e6b96570 CH |
3330 | iclog = iclog->ic_prev; |
3331 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3332 | iclog->ic_state == XLOG_STATE_DIRTY) | |
3333 | goto out_unlock; | |
3334 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3335 | if (atomic_read(&iclog->ic_refcnt) == 0) { | |
3336 | /* | |
3337 | * We are the only one with access to this iclog. | |
3338 | * | |
3339 | * Flush it out now. There should be a roundoff of zero | |
3340 | * to show that someone has already taken care of the | |
3341 | * roundoff from the previous sync. | |
3342 | */ | |
3343 | atomic_inc(&iclog->ic_refcnt); | |
3344 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
3345 | xlog_state_switch_iclogs(log, iclog, 0); | |
3346 | spin_unlock(&log->l_icloglock); | |
a14a348b | 3347 | |
e6b96570 CH |
3348 | if (xlog_state_release_iclog(log, iclog)) |
3349 | return -EIO; | |
1da177e4 | 3350 | |
e6b96570 CH |
3351 | spin_lock(&log->l_icloglock); |
3352 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn || | |
3353 | iclog->ic_state == XLOG_STATE_DIRTY) | |
3354 | goto out_unlock; | |
3355 | } else { | |
3356 | /* | |
3357 | * Someone else is writing to this iclog. | |
3358 | * | |
3359 | * Use its call to flush out the data. However, the | |
3360 | * other thread may not force out this LR, so we mark | |
3361 | * it WANT_SYNC. | |
3362 | */ | |
3363 | xlog_state_switch_iclogs(log, iclog, 0); | |
1da177e4 | 3364 | } |
e6b96570 | 3365 | } else { |
1da177e4 | 3366 | /* |
e6b96570 CH |
3367 | * If the head iclog is not active nor dirty, we just attach |
3368 | * ourselves to the head and go to sleep if necessary. | |
1da177e4 | 3369 | */ |
e6b96570 | 3370 | ; |
1da177e4 | 3371 | } |
e6b96570 CH |
3372 | |
3373 | if (!(flags & XFS_LOG_SYNC)) | |
3374 | goto out_unlock; | |
3375 | ||
3376 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3377 | goto out_error; | |
3378 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3379 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
3380 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3381 | return -EIO; | |
1da177e4 | 3382 | return 0; |
e6b96570 CH |
3383 | |
3384 | out_unlock: | |
3385 | spin_unlock(&log->l_icloglock); | |
3386 | return 0; | |
3387 | out_error: | |
3388 | spin_unlock(&log->l_icloglock); | |
3389 | return -EIO; | |
a14a348b | 3390 | } |
1da177e4 | 3391 | |
3e4da466 CH |
3392 | static int |
3393 | __xfs_log_force_lsn( | |
a14a348b CH |
3394 | struct xfs_mount *mp, |
3395 | xfs_lsn_t lsn, | |
3396 | uint flags, | |
3e4da466 CH |
3397 | int *log_flushed, |
3398 | bool already_slept) | |
1da177e4 | 3399 | { |
ad223e60 | 3400 | struct xlog *log = mp->m_log; |
a14a348b | 3401 | struct xlog_in_core *iclog; |
71e330b5 | 3402 | |
a14a348b CH |
3403 | spin_lock(&log->l_icloglock); |
3404 | iclog = log->l_iclog; | |
93806299 CH |
3405 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3406 | goto out_error; | |
1da177e4 | 3407 | |
93806299 CH |
3408 | while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { |
3409 | iclog = iclog->ic_next; | |
3410 | if (iclog == log->l_iclog) | |
3411 | goto out_unlock; | |
3412 | } | |
a14a348b | 3413 | |
93806299 CH |
3414 | if (iclog->ic_state == XLOG_STATE_DIRTY) |
3415 | goto out_unlock; | |
a14a348b | 3416 | |
93806299 CH |
3417 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { |
3418 | /* | |
3419 | * We sleep here if we haven't already slept (e.g. this is the | |
3420 | * first time we've looked at the correct iclog buf) and the | |
3421 | * buffer before us is going to be sync'ed. The reason for this | |
3422 | * is that if we are doing sync transactions here, by waiting | |
3423 | * for the previous I/O to complete, we can allow a few more | |
3424 | * transactions into this iclog before we close it down. | |
3425 | * | |
3426 | * Otherwise, we mark the buffer WANT_SYNC, and bump up the | |
3427 | * refcnt so we can release the log (which drops the ref count). | |
3428 | * The state switch keeps new transaction commits from using | |
3429 | * this buffer. When the current commits finish writing into | |
3430 | * the buffer, the refcount will drop to zero and the buffer | |
3431 | * will go out then. | |
3432 | */ | |
3433 | if (!already_slept && | |
3434 | (iclog->ic_prev->ic_state & | |
3435 | (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) { | |
3436 | ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); | |
a14a348b | 3437 | |
93806299 | 3438 | XFS_STATS_INC(mp, xs_log_force_sleep); |
a14a348b | 3439 | |
93806299 CH |
3440 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3441 | &log->l_icloglock); | |
3e4da466 | 3442 | return -EAGAIN; |
1da177e4 | 3443 | } |
93806299 CH |
3444 | atomic_inc(&iclog->ic_refcnt); |
3445 | xlog_state_switch_iclogs(log, iclog, 0); | |
3446 | spin_unlock(&log->l_icloglock); | |
3447 | if (xlog_state_release_iclog(log, iclog)) | |
3448 | return -EIO; | |
3449 | if (log_flushed) | |
3450 | *log_flushed = 1; | |
3451 | spin_lock(&log->l_icloglock); | |
3452 | } | |
1da177e4 | 3453 | |
93806299 CH |
3454 | if (!(flags & XFS_LOG_SYNC) || |
3455 | (iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) | |
3456 | goto out_unlock; | |
1da177e4 | 3457 | |
93806299 CH |
3458 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
3459 | goto out_error; | |
3460 | ||
3461 | XFS_STATS_INC(mp, xs_log_force_sleep); | |
3462 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); | |
3463 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3464 | return -EIO; | |
3465 | return 0; | |
1da177e4 | 3466 | |
93806299 | 3467 | out_unlock: |
a14a348b CH |
3468 | spin_unlock(&log->l_icloglock); |
3469 | return 0; | |
93806299 CH |
3470 | out_error: |
3471 | spin_unlock(&log->l_icloglock); | |
3472 | return -EIO; | |
a14a348b CH |
3473 | } |
3474 | ||
3e4da466 CH |
3475 | /* |
3476 | * Force the in-core log to disk for a specific LSN. | |
3477 | * | |
3478 | * Find in-core log with lsn. | |
3479 | * If it is in the DIRTY state, just return. | |
3480 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3481 | * state and go to sleep or return. | |
3482 | * If it is in any other state, go to sleep or return. | |
3483 | * | |
3484 | * Synchronous forces are implemented with a wait queue. All callers trying | |
3485 | * to force a given lsn to disk must wait on the queue attached to the | |
3486 | * specific in-core log. When given in-core log finally completes its write | |
3487 | * to disk, that thread will wake up all threads waiting on the queue. | |
3488 | */ | |
3489 | int | |
3490 | xfs_log_force_lsn( | |
3491 | struct xfs_mount *mp, | |
3492 | xfs_lsn_t lsn, | |
3493 | uint flags, | |
3494 | int *log_flushed) | |
3495 | { | |
3496 | int ret; | |
3497 | ASSERT(lsn != 0); | |
3498 | ||
3499 | XFS_STATS_INC(mp, xs_log_force); | |
3500 | trace_xfs_log_force(mp, lsn, _RET_IP_); | |
3501 | ||
3502 | lsn = xlog_cil_force_lsn(mp->m_log, lsn); | |
3503 | if (lsn == NULLCOMMITLSN) | |
3504 | return 0; | |
3505 | ||
3506 | ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, false); | |
3507 | if (ret == -EAGAIN) | |
3508 | ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, true); | |
3509 | return ret; | |
3510 | } | |
3511 | ||
1da177e4 LT |
3512 | /* |
3513 | * Called when we want to mark the current iclog as being ready to sync to | |
3514 | * disk. | |
3515 | */ | |
a8272ce0 | 3516 | STATIC void |
9a8d2fdb MT |
3517 | xlog_state_want_sync( |
3518 | struct xlog *log, | |
3519 | struct xlog_in_core *iclog) | |
1da177e4 | 3520 | { |
a8914f3a | 3521 | assert_spin_locked(&log->l_icloglock); |
1da177e4 LT |
3522 | |
3523 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3524 | xlog_state_switch_iclogs(log, iclog, 0); | |
3525 | } else { | |
3526 | ASSERT(iclog->ic_state & | |
3527 | (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR)); | |
3528 | } | |
39e2defe | 3529 | } |
1da177e4 LT |
3530 | |
3531 | ||
3532 | /***************************************************************************** | |
3533 | * | |
3534 | * TICKET functions | |
3535 | * | |
3536 | ***************************************************************************** | |
3537 | */ | |
3538 | ||
3539 | /* | |
9da096fd | 3540 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3541 | */ |
cc09c0dc DC |
3542 | void |
3543 | xfs_log_ticket_put( | |
3544 | xlog_ticket_t *ticket) | |
1da177e4 | 3545 | { |
cc09c0dc | 3546 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3547 | if (atomic_dec_and_test(&ticket->t_ref)) |
cc09c0dc | 3548 | kmem_zone_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3549 | } |
1da177e4 | 3550 | |
cc09c0dc DC |
3551 | xlog_ticket_t * |
3552 | xfs_log_ticket_get( | |
3553 | xlog_ticket_t *ticket) | |
3554 | { | |
3555 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3556 | atomic_inc(&ticket->t_ref); | |
3557 | return ticket; | |
3558 | } | |
1da177e4 LT |
3559 | |
3560 | /* | |
e773fc93 JL |
3561 | * Figure out the total log space unit (in bytes) that would be |
3562 | * required for a log ticket. | |
1da177e4 | 3563 | */ |
e773fc93 JL |
3564 | int |
3565 | xfs_log_calc_unit_res( | |
3566 | struct xfs_mount *mp, | |
3567 | int unit_bytes) | |
1da177e4 | 3568 | { |
e773fc93 JL |
3569 | struct xlog *log = mp->m_log; |
3570 | int iclog_space; | |
3571 | uint num_headers; | |
1da177e4 LT |
3572 | |
3573 | /* | |
3574 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3575 | * in the log. A unit in this case is the amount of space for one | |
3576 | * of these log operations. Normal reservations have a cnt of 1 | |
3577 | * and their unit amount is the total amount of space required. | |
3578 | * | |
3579 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3580 | * which occupy space in the on-disk log. |
3581 | * | |
3582 | * Normal form of a transaction is: | |
3583 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3584 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3585 | * | |
3586 | * We need to account for all the leadup data and trailer data | |
3587 | * around the transaction data. | |
3588 | * And then we need to account for the worst case in terms of using | |
3589 | * more space. | |
3590 | * The worst case will happen if: | |
3591 | * - the placement of the transaction happens to be such that the | |
3592 | * roundoff is at its maximum | |
3593 | * - the transaction data is synced before the commit record is synced | |
3594 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3595 | * Therefore the commit record is in its own Log Record. | |
3596 | * This can happen as the commit record is called with its | |
3597 | * own region to xlog_write(). | |
3598 | * This then means that in the worst case, roundoff can happen for | |
3599 | * the commit-rec as well. | |
3600 | * The commit-rec is smaller than padding in this scenario and so it is | |
3601 | * not added separately. | |
1da177e4 LT |
3602 | */ |
3603 | ||
32fb9b57 TS |
3604 | /* for trans header */ |
3605 | unit_bytes += sizeof(xlog_op_header_t); | |
3606 | unit_bytes += sizeof(xfs_trans_header_t); | |
3607 | ||
1da177e4 | 3608 | /* for start-rec */ |
32fb9b57 TS |
3609 | unit_bytes += sizeof(xlog_op_header_t); |
3610 | ||
9b9fc2b7 DC |
3611 | /* |
3612 | * for LR headers - the space for data in an iclog is the size minus | |
3613 | * the space used for the headers. If we use the iclog size, then we | |
3614 | * undercalculate the number of headers required. | |
3615 | * | |
3616 | * Furthermore - the addition of op headers for split-recs might | |
3617 | * increase the space required enough to require more log and op | |
3618 | * headers, so take that into account too. | |
3619 | * | |
3620 | * IMPORTANT: This reservation makes the assumption that if this | |
3621 | * transaction is the first in an iclog and hence has the LR headers | |
3622 | * accounted to it, then the remaining space in the iclog is | |
3623 | * exclusively for this transaction. i.e. if the transaction is larger | |
3624 | * than the iclog, it will be the only thing in that iclog. | |
3625 | * Fundamentally, this means we must pass the entire log vector to | |
3626 | * xlog_write to guarantee this. | |
3627 | */ | |
3628 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3629 | num_headers = howmany(unit_bytes, iclog_space); | |
3630 | ||
3631 | /* for split-recs - ophdrs added when data split over LRs */ | |
3632 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3633 | ||
3634 | /* add extra header reservations if we overrun */ | |
3635 | while (!num_headers || | |
3636 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3637 | unit_bytes += sizeof(xlog_op_header_t); | |
3638 | num_headers++; | |
3639 | } | |
32fb9b57 | 3640 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3641 | |
32fb9b57 TS |
3642 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3643 | unit_bytes += log->l_iclog_hsize; | |
3644 | ||
32fb9b57 | 3645 | /* for roundoff padding for transaction data and one for commit record */ |
e773fc93 | 3646 | if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) { |
1da177e4 | 3647 | /* log su roundoff */ |
e773fc93 | 3648 | unit_bytes += 2 * mp->m_sb.sb_logsunit; |
1da177e4 LT |
3649 | } else { |
3650 | /* BB roundoff */ | |
e773fc93 | 3651 | unit_bytes += 2 * BBSIZE; |
1da177e4 LT |
3652 | } |
3653 | ||
e773fc93 JL |
3654 | return unit_bytes; |
3655 | } | |
3656 | ||
3657 | /* | |
3658 | * Allocate and initialise a new log ticket. | |
3659 | */ | |
3660 | struct xlog_ticket * | |
3661 | xlog_ticket_alloc( | |
3662 | struct xlog *log, | |
3663 | int unit_bytes, | |
3664 | int cnt, | |
3665 | char client, | |
3666 | bool permanent, | |
3667 | xfs_km_flags_t alloc_flags) | |
3668 | { | |
3669 | struct xlog_ticket *tic; | |
3670 | int unit_res; | |
3671 | ||
3672 | tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags); | |
3673 | if (!tic) | |
3674 | return NULL; | |
3675 | ||
3676 | unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes); | |
3677 | ||
cc09c0dc | 3678 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3679 | tic->t_task = current; |
10547941 | 3680 | INIT_LIST_HEAD(&tic->t_queue); |
e773fc93 JL |
3681 | tic->t_unit_res = unit_res; |
3682 | tic->t_curr_res = unit_res; | |
1da177e4 LT |
3683 | tic->t_cnt = cnt; |
3684 | tic->t_ocnt = cnt; | |
ecb3403d | 3685 | tic->t_tid = prandom_u32(); |
1da177e4 LT |
3686 | tic->t_clientid = client; |
3687 | tic->t_flags = XLOG_TIC_INITED; | |
9006fb91 | 3688 | if (permanent) |
1da177e4 | 3689 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3690 | |
0adba536 | 3691 | xlog_tic_reset_res(tic); |
7e9c6396 | 3692 | |
1da177e4 | 3693 | return tic; |
cc09c0dc | 3694 | } |
1da177e4 LT |
3695 | |
3696 | ||
3697 | /****************************************************************************** | |
3698 | * | |
3699 | * Log debug routines | |
3700 | * | |
3701 | ****************************************************************************** | |
3702 | */ | |
cfcbbbd0 | 3703 | #if defined(DEBUG) |
1da177e4 LT |
3704 | /* |
3705 | * Make sure that the destination ptr is within the valid data region of | |
3706 | * one of the iclogs. This uses backup pointers stored in a different | |
3707 | * part of the log in case we trash the log structure. | |
3708 | */ | |
181fdfe6 | 3709 | STATIC void |
e6b1f273 | 3710 | xlog_verify_dest_ptr( |
ad223e60 | 3711 | struct xlog *log, |
5809d5e0 | 3712 | void *ptr) |
1da177e4 LT |
3713 | { |
3714 | int i; | |
3715 | int good_ptr = 0; | |
3716 | ||
e6b1f273 CH |
3717 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3718 | if (ptr >= log->l_iclog_bak[i] && | |
3719 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3720 | good_ptr++; |
3721 | } | |
e6b1f273 CH |
3722 | |
3723 | if (!good_ptr) | |
a0fa2b67 | 3724 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3725 | } |
1da177e4 | 3726 | |
da8a1a4a DC |
3727 | /* |
3728 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3729 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3730 | * the cycles differ by exactly one and check the byte count. | |
3731 | * | |
3732 | * This check is run unlocked, so can give false positives. Rather than assert | |
3733 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3734 | * determine if they want to panic the machine when such an error occurs. For | |
3735 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3736 | * an assert, it can be turned off at runtime. | |
3737 | */ | |
3f336c6f DC |
3738 | STATIC void |
3739 | xlog_verify_grant_tail( | |
ad223e60 | 3740 | struct xlog *log) |
3f336c6f | 3741 | { |
1c3cb9ec | 3742 | int tail_cycle, tail_blocks; |
a69ed03c | 3743 | int cycle, space; |
3f336c6f | 3744 | |
28496968 | 3745 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3746 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3747 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3748 | if (cycle - 1 != tail_cycle && |
3749 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3750 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3751 | "%s: cycle - 1 != tail_cycle", __func__); | |
3752 | log->l_flags |= XLOG_TAIL_WARN; | |
3753 | } | |
3754 | ||
3755 | if (space > BBTOB(tail_blocks) && | |
3756 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3757 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3758 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3759 | log->l_flags |= XLOG_TAIL_WARN; | |
3760 | } | |
3f336c6f DC |
3761 | } |
3762 | } | |
3763 | ||
1da177e4 LT |
3764 | /* check if it will fit */ |
3765 | STATIC void | |
9a8d2fdb MT |
3766 | xlog_verify_tail_lsn( |
3767 | struct xlog *log, | |
3768 | struct xlog_in_core *iclog, | |
3769 | xfs_lsn_t tail_lsn) | |
1da177e4 LT |
3770 | { |
3771 | int blocks; | |
3772 | ||
3773 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3774 | blocks = | |
3775 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3776 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3777 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3778 | } else { |
3779 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3780 | ||
3781 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3782 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3783 | |
3784 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3785 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3786 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3787 | } |
3788 | } /* xlog_verify_tail_lsn */ | |
3789 | ||
3790 | /* | |
3791 | * Perform a number of checks on the iclog before writing to disk. | |
3792 | * | |
3793 | * 1. Make sure the iclogs are still circular | |
3794 | * 2. Make sure we have a good magic number | |
3795 | * 3. Make sure we don't have magic numbers in the data | |
3796 | * 4. Check fields of each log operation header for: | |
3797 | * A. Valid client identifier | |
3798 | * B. tid ptr value falls in valid ptr space (user space code) | |
3799 | * C. Length in log record header is correct according to the | |
3800 | * individual operation headers within record. | |
3801 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3802 | * log, check the preceding blocks of the physical log to make sure all | |
3803 | * the cycle numbers agree with the current cycle number. | |
3804 | */ | |
3805 | STATIC void | |
9a8d2fdb MT |
3806 | xlog_verify_iclog( |
3807 | struct xlog *log, | |
3808 | struct xlog_in_core *iclog, | |
3809 | int count, | |
667a9291 | 3810 | bool syncing) |
1da177e4 LT |
3811 | { |
3812 | xlog_op_header_t *ophead; | |
3813 | xlog_in_core_t *icptr; | |
3814 | xlog_in_core_2_t *xhdr; | |
5809d5e0 | 3815 | void *base_ptr, *ptr, *p; |
db9d67d6 | 3816 | ptrdiff_t field_offset; |
c8ce540d | 3817 | uint8_t clientid; |
1da177e4 LT |
3818 | int len, i, j, k, op_len; |
3819 | int idx; | |
1da177e4 LT |
3820 | |
3821 | /* check validity of iclog pointers */ | |
b22cd72c | 3822 | spin_lock(&log->l_icloglock); |
1da177e4 | 3823 | icptr = log->l_iclog; |
643f7c4e GB |
3824 | for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next) |
3825 | ASSERT(icptr); | |
3826 | ||
1da177e4 | 3827 | if (icptr != log->l_iclog) |
a0fa2b67 | 3828 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3829 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3830 | |
3831 | /* check log magic numbers */ | |
69ef921b | 3832 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3833 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3834 | |
5809d5e0 CH |
3835 | base_ptr = ptr = &iclog->ic_header; |
3836 | p = &iclog->ic_header; | |
3837 | for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) { | |
69ef921b | 3838 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3839 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3840 | __func__); | |
1da177e4 LT |
3841 | } |
3842 | ||
3843 | /* check fields */ | |
b53e675d | 3844 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
5809d5e0 CH |
3845 | base_ptr = ptr = iclog->ic_datap; |
3846 | ophead = ptr; | |
b28708d6 | 3847 | xhdr = iclog->ic_data; |
1da177e4 | 3848 | for (i = 0; i < len; i++) { |
5809d5e0 | 3849 | ophead = ptr; |
1da177e4 LT |
3850 | |
3851 | /* clientid is only 1 byte */ | |
5809d5e0 CH |
3852 | p = &ophead->oh_clientid; |
3853 | field_offset = p - base_ptr; | |
667a9291 | 3854 | if (!syncing || (field_offset & 0x1ff)) { |
1da177e4 LT |
3855 | clientid = ophead->oh_clientid; |
3856 | } else { | |
b2a922cd | 3857 | idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap); |
1da177e4 LT |
3858 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3859 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3860 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3861 | clientid = xlog_get_client_id( |
3862 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3863 | } else { |
03bea6fe CH |
3864 | clientid = xlog_get_client_id( |
3865 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3866 | } |
3867 | } | |
3868 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 | 3869 | xfs_warn(log->l_mp, |
c9690043 | 3870 | "%s: invalid clientid %d op "PTR_FMT" offset 0x%lx", |
a0fa2b67 DC |
3871 | __func__, clientid, ophead, |
3872 | (unsigned long)field_offset); | |
1da177e4 LT |
3873 | |
3874 | /* check length */ | |
5809d5e0 CH |
3875 | p = &ophead->oh_len; |
3876 | field_offset = p - base_ptr; | |
667a9291 | 3877 | if (!syncing || (field_offset & 0x1ff)) { |
67fcb7bf | 3878 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 | 3879 | } else { |
db9d67d6 CH |
3880 | idx = BTOBBT((uintptr_t)&ophead->oh_len - |
3881 | (uintptr_t)iclog->ic_datap); | |
1da177e4 LT |
3882 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { |
3883 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3884 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3885 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3886 | } else { |
b53e675d | 3887 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3888 | } |
3889 | } | |
3890 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3891 | } | |
3892 | } /* xlog_verify_iclog */ | |
cfcbbbd0 | 3893 | #endif |
1da177e4 LT |
3894 | |
3895 | /* | |
b22cd72c | 3896 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3897 | */ |
3898 | STATIC int | |
3899 | xlog_state_ioerror( | |
9a8d2fdb | 3900 | struct xlog *log) |
1da177e4 LT |
3901 | { |
3902 | xlog_in_core_t *iclog, *ic; | |
3903 | ||
3904 | iclog = log->l_iclog; | |
3905 | if (! (iclog->ic_state & XLOG_STATE_IOERROR)) { | |
3906 | /* | |
3907 | * Mark all the incore logs IOERROR. | |
3908 | * From now on, no log flushes will result. | |
3909 | */ | |
3910 | ic = iclog; | |
3911 | do { | |
3912 | ic->ic_state = XLOG_STATE_IOERROR; | |
3913 | ic = ic->ic_next; | |
3914 | } while (ic != iclog); | |
014c2544 | 3915 | return 0; |
1da177e4 LT |
3916 | } |
3917 | /* | |
3918 | * Return non-zero, if state transition has already happened. | |
3919 | */ | |
014c2544 | 3920 | return 1; |
1da177e4 LT |
3921 | } |
3922 | ||
3923 | /* | |
3924 | * This is called from xfs_force_shutdown, when we're forcibly | |
3925 | * shutting down the filesystem, typically because of an IO error. | |
3926 | * Our main objectives here are to make sure that: | |
a870fe6d DC |
3927 | * a. if !logerror, flush the logs to disk. Anything modified |
3928 | * after this is ignored. | |
3929 | * b. the filesystem gets marked 'SHUTDOWN' for all interested | |
1da177e4 | 3930 | * parties to find out, 'atomically'. |
a870fe6d | 3931 | * c. those who're sleeping on log reservations, pinned objects and |
1da177e4 | 3932 | * other resources get woken up, and be told the bad news. |
a870fe6d | 3933 | * d. nothing new gets queued up after (b) and (c) are done. |
9da1ab18 | 3934 | * |
a870fe6d DC |
3935 | * Note: for the !logerror case we need to flush the regions held in memory out |
3936 | * to disk first. This needs to be done before the log is marked as shutdown, | |
3937 | * otherwise the iclog writes will fail. | |
1da177e4 LT |
3938 | */ |
3939 | int | |
3940 | xfs_log_force_umount( | |
3941 | struct xfs_mount *mp, | |
3942 | int logerror) | |
3943 | { | |
9a8d2fdb | 3944 | struct xlog *log; |
1da177e4 | 3945 | int retval; |
1da177e4 LT |
3946 | |
3947 | log = mp->m_log; | |
3948 | ||
3949 | /* | |
3950 | * If this happens during log recovery, don't worry about | |
3951 | * locking; the log isn't open for business yet. | |
3952 | */ | |
3953 | if (!log || | |
3954 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3955 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 | 3956 | if (mp->m_sb_bp) |
b0388bf1 | 3957 | mp->m_sb_bp->b_flags |= XBF_DONE; |
014c2544 | 3958 | return 0; |
1da177e4 LT |
3959 | } |
3960 | ||
3961 | /* | |
3962 | * Somebody could've already done the hard work for us. | |
3963 | * No need to get locks for this. | |
3964 | */ | |
3965 | if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) { | |
3966 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); | |
014c2544 | 3967 | return 1; |
1da177e4 | 3968 | } |
9da1ab18 DC |
3969 | |
3970 | /* | |
a870fe6d DC |
3971 | * Flush all the completed transactions to disk before marking the log |
3972 | * being shut down. We need to do it in this order to ensure that | |
3973 | * completed operations are safely on disk before we shut down, and that | |
3974 | * we don't have to issue any buffer IO after the shutdown flags are set | |
3975 | * to guarantee this. | |
9da1ab18 | 3976 | */ |
93b8a585 | 3977 | if (!logerror) |
60e5bb78 | 3978 | xfs_log_force(mp, XFS_LOG_SYNC); |
9da1ab18 | 3979 | |
1da177e4 | 3980 | /* |
3f16b985 DC |
3981 | * mark the filesystem and the as in a shutdown state and wake |
3982 | * everybody up to tell them the bad news. | |
1da177e4 | 3983 | */ |
b22cd72c | 3984 | spin_lock(&log->l_icloglock); |
1da177e4 | 3985 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 | 3986 | if (mp->m_sb_bp) |
b0388bf1 | 3987 | mp->m_sb_bp->b_flags |= XBF_DONE; |
bac8dca9 | 3988 | |
1da177e4 | 3989 | /* |
a870fe6d DC |
3990 | * Mark the log and the iclogs with IO error flags to prevent any |
3991 | * further log IO from being issued or completed. | |
1da177e4 LT |
3992 | */ |
3993 | log->l_flags |= XLOG_IO_ERROR; | |
a870fe6d | 3994 | retval = xlog_state_ioerror(log); |
b22cd72c | 3995 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3996 | |
3997 | /* | |
10547941 DC |
3998 | * We don't want anybody waiting for log reservations after this. That |
3999 | * means we have to wake up everybody queued up on reserveq as well as | |
4000 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
4001 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 4002 | * action is protected by the grant locks. |
1da177e4 | 4003 | */ |
a79bf2d7 CH |
4004 | xlog_grant_head_wake_all(&log->l_reserve_head); |
4005 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 4006 | |
1da177e4 | 4007 | /* |
ac983517 DC |
4008 | * Wake up everybody waiting on xfs_log_force. Wake the CIL push first |
4009 | * as if the log writes were completed. The abort handling in the log | |
4010 | * item committed callback functions will do this again under lock to | |
4011 | * avoid races. | |
1da177e4 | 4012 | */ |
ac983517 | 4013 | wake_up_all(&log->l_cilp->xc_commit_wait); |
1da177e4 LT |
4014 | xlog_state_do_callback(log, XFS_LI_ABORTED, NULL); |
4015 | ||
4016 | #ifdef XFSERRORDEBUG | |
4017 | { | |
4018 | xlog_in_core_t *iclog; | |
4019 | ||
b22cd72c | 4020 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
4021 | iclog = log->l_iclog; |
4022 | do { | |
4023 | ASSERT(iclog->ic_callback == 0); | |
4024 | iclog = iclog->ic_next; | |
4025 | } while (iclog != log->l_iclog); | |
b22cd72c | 4026 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
4027 | } |
4028 | #endif | |
4029 | /* return non-zero if log IOERROR transition had already happened */ | |
014c2544 | 4030 | return retval; |
1da177e4 LT |
4031 | } |
4032 | ||
ba0f32d4 | 4033 | STATIC int |
9a8d2fdb MT |
4034 | xlog_iclogs_empty( |
4035 | struct xlog *log) | |
1da177e4 LT |
4036 | { |
4037 | xlog_in_core_t *iclog; | |
4038 | ||
4039 | iclog = log->l_iclog; | |
4040 | do { | |
4041 | /* endianness does not matter here, zero is zero in | |
4042 | * any language. | |
4043 | */ | |
4044 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 4045 | return 0; |
1da177e4 LT |
4046 | iclog = iclog->ic_next; |
4047 | } while (iclog != log->l_iclog); | |
014c2544 | 4048 | return 1; |
1da177e4 | 4049 | } |
f661f1e0 | 4050 | |
a45086e2 BF |
4051 | /* |
4052 | * Verify that an LSN stamped into a piece of metadata is valid. This is | |
4053 | * intended for use in read verifiers on v5 superblocks. | |
4054 | */ | |
4055 | bool | |
4056 | xfs_log_check_lsn( | |
4057 | struct xfs_mount *mp, | |
4058 | xfs_lsn_t lsn) | |
4059 | { | |
4060 | struct xlog *log = mp->m_log; | |
4061 | bool valid; | |
4062 | ||
4063 | /* | |
4064 | * norecovery mode skips mount-time log processing and unconditionally | |
4065 | * resets the in-core LSN. We can't validate in this mode, but | |
4066 | * modifications are not allowed anyways so just return true. | |
4067 | */ | |
4068 | if (mp->m_flags & XFS_MOUNT_NORECOVERY) | |
4069 | return true; | |
4070 | ||
4071 | /* | |
4072 | * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is | |
4073 | * handled by recovery and thus safe to ignore here. | |
4074 | */ | |
4075 | if (lsn == NULLCOMMITLSN) | |
4076 | return true; | |
4077 | ||
4078 | valid = xlog_valid_lsn(mp->m_log, lsn); | |
4079 | ||
4080 | /* warn the user about what's gone wrong before verifier failure */ | |
4081 | if (!valid) { | |
4082 | spin_lock(&log->l_icloglock); | |
4083 | xfs_warn(mp, | |
4084 | "Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). " | |
4085 | "Please unmount and run xfs_repair (>= v4.3) to resolve.", | |
4086 | CYCLE_LSN(lsn), BLOCK_LSN(lsn), | |
4087 | log->l_curr_cycle, log->l_curr_block); | |
4088 | spin_unlock(&log->l_icloglock); | |
4089 | } | |
4090 | ||
4091 | return valid; | |
4092 | } |