Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
1da177e4 LT |
21 | #include "xfs_log.h" |
22 | #include "xfs_trans.h" | |
a844f451 NS |
23 | #include "xfs_sb.h" |
24 | #include "xfs_ag.h" | |
1da177e4 LT |
25 | #include "xfs_mount.h" |
26 | #include "xfs_error.h" | |
27 | #include "xfs_log_priv.h" | |
28 | #include "xfs_buf_item.h" | |
a844f451 | 29 | #include "xfs_bmap_btree.h" |
1da177e4 | 30 | #include "xfs_alloc_btree.h" |
a844f451 | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 | 32 | #include "xfs_log_recover.h" |
1da177e4 | 33 | #include "xfs_trans_priv.h" |
a844f451 NS |
34 | #include "xfs_dinode.h" |
35 | #include "xfs_inode.h" | |
0b1b213f | 36 | #include "xfs_trace.h" |
f661f1e0 | 37 | #include "xfs_fsops.h" |
1da177e4 | 38 | |
eb01c9cd | 39 | kmem_zone_t *xfs_log_ticket_zone; |
1da177e4 | 40 | |
1da177e4 | 41 | /* Local miscellaneous function prototypes */ |
ad223e60 MT |
42 | STATIC int |
43 | xlog_commit_record( | |
44 | struct xlog *log, | |
45 | struct xlog_ticket *ticket, | |
46 | struct xlog_in_core **iclog, | |
47 | xfs_lsn_t *commitlsnp); | |
48 | ||
9a8d2fdb MT |
49 | STATIC struct xlog * |
50 | xlog_alloc_log( | |
51 | struct xfs_mount *mp, | |
52 | struct xfs_buftarg *log_target, | |
53 | xfs_daddr_t blk_offset, | |
54 | int num_bblks); | |
ad223e60 MT |
55 | STATIC int |
56 | xlog_space_left( | |
57 | struct xlog *log, | |
58 | atomic64_t *head); | |
9a8d2fdb MT |
59 | STATIC int |
60 | xlog_sync( | |
61 | struct xlog *log, | |
62 | struct xlog_in_core *iclog); | |
63 | STATIC void | |
64 | xlog_dealloc_log( | |
65 | struct xlog *log); | |
1da177e4 LT |
66 | |
67 | /* local state machine functions */ | |
68 | STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int); | |
9a8d2fdb MT |
69 | STATIC void |
70 | xlog_state_do_callback( | |
71 | struct xlog *log, | |
72 | int aborted, | |
73 | struct xlog_in_core *iclog); | |
74 | STATIC int | |
75 | xlog_state_get_iclog_space( | |
76 | struct xlog *log, | |
77 | int len, | |
78 | struct xlog_in_core **iclog, | |
79 | struct xlog_ticket *ticket, | |
80 | int *continued_write, | |
81 | int *logoffsetp); | |
82 | STATIC int | |
83 | xlog_state_release_iclog( | |
84 | struct xlog *log, | |
85 | struct xlog_in_core *iclog); | |
86 | STATIC void | |
87 | xlog_state_switch_iclogs( | |
88 | struct xlog *log, | |
89 | struct xlog_in_core *iclog, | |
90 | int eventual_size); | |
91 | STATIC void | |
92 | xlog_state_want_sync( | |
93 | struct xlog *log, | |
94 | struct xlog_in_core *iclog); | |
1da177e4 | 95 | |
ad223e60 MT |
96 | STATIC void |
97 | xlog_grant_push_ail( | |
9a8d2fdb MT |
98 | struct xlog *log, |
99 | int need_bytes); | |
100 | STATIC void | |
101 | xlog_regrant_reserve_log_space( | |
102 | struct xlog *log, | |
103 | struct xlog_ticket *ticket); | |
104 | STATIC void | |
105 | xlog_ungrant_log_space( | |
106 | struct xlog *log, | |
107 | struct xlog_ticket *ticket); | |
1da177e4 | 108 | |
cfcbbbd0 | 109 | #if defined(DEBUG) |
9a8d2fdb MT |
110 | STATIC void |
111 | xlog_verify_dest_ptr( | |
112 | struct xlog *log, | |
113 | char *ptr); | |
ad223e60 MT |
114 | STATIC void |
115 | xlog_verify_grant_tail( | |
9a8d2fdb MT |
116 | struct xlog *log); |
117 | STATIC void | |
118 | xlog_verify_iclog( | |
119 | struct xlog *log, | |
120 | struct xlog_in_core *iclog, | |
121 | int count, | |
122 | boolean_t syncing); | |
123 | STATIC void | |
124 | xlog_verify_tail_lsn( | |
125 | struct xlog *log, | |
126 | struct xlog_in_core *iclog, | |
127 | xfs_lsn_t tail_lsn); | |
1da177e4 LT |
128 | #else |
129 | #define xlog_verify_dest_ptr(a,b) | |
3f336c6f | 130 | #define xlog_verify_grant_tail(a) |
1da177e4 LT |
131 | #define xlog_verify_iclog(a,b,c,d) |
132 | #define xlog_verify_tail_lsn(a,b,c) | |
133 | #endif | |
134 | ||
9a8d2fdb MT |
135 | STATIC int |
136 | xlog_iclogs_empty( | |
137 | struct xlog *log); | |
1da177e4 | 138 | |
dd954c69 | 139 | static void |
663e496a | 140 | xlog_grant_sub_space( |
ad223e60 MT |
141 | struct xlog *log, |
142 | atomic64_t *head, | |
143 | int bytes) | |
dd954c69 | 144 | { |
d0eb2f38 DC |
145 | int64_t head_val = atomic64_read(head); |
146 | int64_t new, old; | |
a69ed03c | 147 | |
d0eb2f38 DC |
148 | do { |
149 | int cycle, space; | |
a69ed03c | 150 | |
d0eb2f38 | 151 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 152 | |
d0eb2f38 DC |
153 | space -= bytes; |
154 | if (space < 0) { | |
155 | space += log->l_logsize; | |
156 | cycle--; | |
157 | } | |
158 | ||
159 | old = head_val; | |
160 | new = xlog_assign_grant_head_val(cycle, space); | |
161 | head_val = atomic64_cmpxchg(head, old, new); | |
162 | } while (head_val != old); | |
dd954c69 CH |
163 | } |
164 | ||
165 | static void | |
663e496a | 166 | xlog_grant_add_space( |
ad223e60 MT |
167 | struct xlog *log, |
168 | atomic64_t *head, | |
169 | int bytes) | |
dd954c69 | 170 | { |
d0eb2f38 DC |
171 | int64_t head_val = atomic64_read(head); |
172 | int64_t new, old; | |
a69ed03c | 173 | |
d0eb2f38 DC |
174 | do { |
175 | int tmp; | |
176 | int cycle, space; | |
a69ed03c | 177 | |
d0eb2f38 | 178 | xlog_crack_grant_head_val(head_val, &cycle, &space); |
a69ed03c | 179 | |
d0eb2f38 DC |
180 | tmp = log->l_logsize - space; |
181 | if (tmp > bytes) | |
182 | space += bytes; | |
183 | else { | |
184 | space = bytes - tmp; | |
185 | cycle++; | |
186 | } | |
187 | ||
188 | old = head_val; | |
189 | new = xlog_assign_grant_head_val(cycle, space); | |
190 | head_val = atomic64_cmpxchg(head, old, new); | |
191 | } while (head_val != old); | |
dd954c69 | 192 | } |
a69ed03c | 193 | |
c303c5b8 CH |
194 | STATIC void |
195 | xlog_grant_head_init( | |
196 | struct xlog_grant_head *head) | |
197 | { | |
198 | xlog_assign_grant_head(&head->grant, 1, 0); | |
199 | INIT_LIST_HEAD(&head->waiters); | |
200 | spin_lock_init(&head->lock); | |
201 | } | |
202 | ||
a79bf2d7 CH |
203 | STATIC void |
204 | xlog_grant_head_wake_all( | |
205 | struct xlog_grant_head *head) | |
206 | { | |
207 | struct xlog_ticket *tic; | |
208 | ||
209 | spin_lock(&head->lock); | |
210 | list_for_each_entry(tic, &head->waiters, t_queue) | |
211 | wake_up_process(tic->t_task); | |
212 | spin_unlock(&head->lock); | |
213 | } | |
214 | ||
e179840d CH |
215 | static inline int |
216 | xlog_ticket_reservation( | |
ad223e60 | 217 | struct xlog *log, |
e179840d CH |
218 | struct xlog_grant_head *head, |
219 | struct xlog_ticket *tic) | |
9f9c19ec | 220 | { |
e179840d CH |
221 | if (head == &log->l_write_head) { |
222 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
223 | return tic->t_unit_res; | |
224 | } else { | |
9f9c19ec | 225 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) |
e179840d | 226 | return tic->t_unit_res * tic->t_cnt; |
9f9c19ec | 227 | else |
e179840d | 228 | return tic->t_unit_res; |
9f9c19ec | 229 | } |
9f9c19ec CH |
230 | } |
231 | ||
232 | STATIC bool | |
e179840d | 233 | xlog_grant_head_wake( |
ad223e60 | 234 | struct xlog *log, |
e179840d | 235 | struct xlog_grant_head *head, |
9f9c19ec CH |
236 | int *free_bytes) |
237 | { | |
238 | struct xlog_ticket *tic; | |
239 | int need_bytes; | |
240 | ||
e179840d CH |
241 | list_for_each_entry(tic, &head->waiters, t_queue) { |
242 | need_bytes = xlog_ticket_reservation(log, head, tic); | |
9f9c19ec CH |
243 | if (*free_bytes < need_bytes) |
244 | return false; | |
9f9c19ec | 245 | |
e179840d CH |
246 | *free_bytes -= need_bytes; |
247 | trace_xfs_log_grant_wake_up(log, tic); | |
14a7235f | 248 | wake_up_process(tic->t_task); |
9f9c19ec CH |
249 | } |
250 | ||
251 | return true; | |
252 | } | |
253 | ||
254 | STATIC int | |
23ee3df3 | 255 | xlog_grant_head_wait( |
ad223e60 | 256 | struct xlog *log, |
23ee3df3 | 257 | struct xlog_grant_head *head, |
9f9c19ec CH |
258 | struct xlog_ticket *tic, |
259 | int need_bytes) | |
260 | { | |
23ee3df3 | 261 | list_add_tail(&tic->t_queue, &head->waiters); |
9f9c19ec CH |
262 | |
263 | do { | |
264 | if (XLOG_FORCED_SHUTDOWN(log)) | |
265 | goto shutdown; | |
266 | xlog_grant_push_ail(log, need_bytes); | |
267 | ||
14a7235f | 268 | __set_current_state(TASK_UNINTERRUPTIBLE); |
23ee3df3 | 269 | spin_unlock(&head->lock); |
14a7235f | 270 | |
9f9c19ec | 271 | XFS_STATS_INC(xs_sleep_logspace); |
9f9c19ec | 272 | |
14a7235f CH |
273 | trace_xfs_log_grant_sleep(log, tic); |
274 | schedule(); | |
9f9c19ec CH |
275 | trace_xfs_log_grant_wake(log, tic); |
276 | ||
23ee3df3 | 277 | spin_lock(&head->lock); |
9f9c19ec CH |
278 | if (XLOG_FORCED_SHUTDOWN(log)) |
279 | goto shutdown; | |
23ee3df3 | 280 | } while (xlog_space_left(log, &head->grant) < need_bytes); |
9f9c19ec CH |
281 | |
282 | list_del_init(&tic->t_queue); | |
283 | return 0; | |
284 | shutdown: | |
285 | list_del_init(&tic->t_queue); | |
286 | return XFS_ERROR(EIO); | |
287 | } | |
288 | ||
42ceedb3 CH |
289 | /* |
290 | * Atomically get the log space required for a log ticket. | |
291 | * | |
292 | * Once a ticket gets put onto head->waiters, it will only return after the | |
293 | * needed reservation is satisfied. | |
294 | * | |
295 | * This function is structured so that it has a lock free fast path. This is | |
296 | * necessary because every new transaction reservation will come through this | |
297 | * path. Hence any lock will be globally hot if we take it unconditionally on | |
298 | * every pass. | |
299 | * | |
300 | * As tickets are only ever moved on and off head->waiters under head->lock, we | |
301 | * only need to take that lock if we are going to add the ticket to the queue | |
302 | * and sleep. We can avoid taking the lock if the ticket was never added to | |
303 | * head->waiters because the t_queue list head will be empty and we hold the | |
304 | * only reference to it so it can safely be checked unlocked. | |
305 | */ | |
306 | STATIC int | |
307 | xlog_grant_head_check( | |
ad223e60 | 308 | struct xlog *log, |
42ceedb3 CH |
309 | struct xlog_grant_head *head, |
310 | struct xlog_ticket *tic, | |
311 | int *need_bytes) | |
312 | { | |
313 | int free_bytes; | |
314 | int error = 0; | |
315 | ||
316 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); | |
317 | ||
318 | /* | |
319 | * If there are other waiters on the queue then give them a chance at | |
320 | * logspace before us. Wake up the first waiters, if we do not wake | |
321 | * up all the waiters then go to sleep waiting for more free space, | |
322 | * otherwise try to get some space for this transaction. | |
323 | */ | |
324 | *need_bytes = xlog_ticket_reservation(log, head, tic); | |
325 | free_bytes = xlog_space_left(log, &head->grant); | |
326 | if (!list_empty_careful(&head->waiters)) { | |
327 | spin_lock(&head->lock); | |
328 | if (!xlog_grant_head_wake(log, head, &free_bytes) || | |
329 | free_bytes < *need_bytes) { | |
330 | error = xlog_grant_head_wait(log, head, tic, | |
331 | *need_bytes); | |
332 | } | |
333 | spin_unlock(&head->lock); | |
334 | } else if (free_bytes < *need_bytes) { | |
335 | spin_lock(&head->lock); | |
336 | error = xlog_grant_head_wait(log, head, tic, *need_bytes); | |
337 | spin_unlock(&head->lock); | |
338 | } | |
339 | ||
340 | return error; | |
341 | } | |
342 | ||
0adba536 CH |
343 | static void |
344 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
345 | { | |
346 | tic->t_res_num = 0; | |
347 | tic->t_res_arr_sum = 0; | |
348 | tic->t_res_num_ophdrs = 0; | |
349 | } | |
350 | ||
351 | static void | |
352 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
353 | { | |
354 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
355 | /* add to overflow and start again */ | |
356 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
357 | tic->t_res_num = 0; | |
358 | tic->t_res_arr_sum = 0; | |
359 | } | |
360 | ||
361 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
362 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
363 | tic->t_res_arr_sum += len; | |
364 | tic->t_res_num++; | |
365 | } | |
dd954c69 | 366 | |
9006fb91 CH |
367 | /* |
368 | * Replenish the byte reservation required by moving the grant write head. | |
369 | */ | |
370 | int | |
371 | xfs_log_regrant( | |
372 | struct xfs_mount *mp, | |
373 | struct xlog_ticket *tic) | |
374 | { | |
ad223e60 | 375 | struct xlog *log = mp->m_log; |
9006fb91 CH |
376 | int need_bytes; |
377 | int error = 0; | |
378 | ||
379 | if (XLOG_FORCED_SHUTDOWN(log)) | |
380 | return XFS_ERROR(EIO); | |
381 | ||
382 | XFS_STATS_INC(xs_try_logspace); | |
383 | ||
384 | /* | |
385 | * This is a new transaction on the ticket, so we need to change the | |
386 | * transaction ID so that the next transaction has a different TID in | |
387 | * the log. Just add one to the existing tid so that we can see chains | |
388 | * of rolling transactions in the log easily. | |
389 | */ | |
390 | tic->t_tid++; | |
391 | ||
392 | xlog_grant_push_ail(log, tic->t_unit_res); | |
393 | ||
394 | tic->t_curr_res = tic->t_unit_res; | |
395 | xlog_tic_reset_res(tic); | |
396 | ||
397 | if (tic->t_cnt > 0) | |
398 | return 0; | |
399 | ||
400 | trace_xfs_log_regrant(log, tic); | |
401 | ||
402 | error = xlog_grant_head_check(log, &log->l_write_head, tic, | |
403 | &need_bytes); | |
404 | if (error) | |
405 | goto out_error; | |
406 | ||
407 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
408 | trace_xfs_log_regrant_exit(log, tic); | |
409 | xlog_verify_grant_tail(log); | |
410 | return 0; | |
411 | ||
412 | out_error: | |
413 | /* | |
414 | * If we are failing, make sure the ticket doesn't have any current | |
415 | * reservations. We don't want to add this back when the ticket/ | |
416 | * transaction gets cancelled. | |
417 | */ | |
418 | tic->t_curr_res = 0; | |
419 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
420 | return error; | |
421 | } | |
422 | ||
423 | /* | |
424 | * Reserve log space and return a ticket corresponding the reservation. | |
425 | * | |
426 | * Each reservation is going to reserve extra space for a log record header. | |
427 | * When writes happen to the on-disk log, we don't subtract the length of the | |
428 | * log record header from any reservation. By wasting space in each | |
429 | * reservation, we prevent over allocation problems. | |
430 | */ | |
431 | int | |
432 | xfs_log_reserve( | |
433 | struct xfs_mount *mp, | |
434 | int unit_bytes, | |
435 | int cnt, | |
436 | struct xlog_ticket **ticp, | |
437 | __uint8_t client, | |
438 | bool permanent, | |
439 | uint t_type) | |
440 | { | |
ad223e60 | 441 | struct xlog *log = mp->m_log; |
9006fb91 CH |
442 | struct xlog_ticket *tic; |
443 | int need_bytes; | |
444 | int error = 0; | |
445 | ||
446 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); | |
447 | ||
448 | if (XLOG_FORCED_SHUTDOWN(log)) | |
449 | return XFS_ERROR(EIO); | |
450 | ||
451 | XFS_STATS_INC(xs_try_logspace); | |
452 | ||
453 | ASSERT(*ticp == NULL); | |
454 | tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, | |
455 | KM_SLEEP | KM_MAYFAIL); | |
456 | if (!tic) | |
457 | return XFS_ERROR(ENOMEM); | |
458 | ||
459 | tic->t_trans_type = t_type; | |
460 | *ticp = tic; | |
461 | ||
462 | xlog_grant_push_ail(log, tic->t_unit_res * tic->t_cnt); | |
463 | ||
464 | trace_xfs_log_reserve(log, tic); | |
465 | ||
466 | error = xlog_grant_head_check(log, &log->l_reserve_head, tic, | |
467 | &need_bytes); | |
468 | if (error) | |
469 | goto out_error; | |
470 | ||
471 | xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); | |
472 | xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); | |
473 | trace_xfs_log_reserve_exit(log, tic); | |
474 | xlog_verify_grant_tail(log); | |
475 | return 0; | |
476 | ||
477 | out_error: | |
478 | /* | |
479 | * If we are failing, make sure the ticket doesn't have any current | |
480 | * reservations. We don't want to add this back when the ticket/ | |
481 | * transaction gets cancelled. | |
482 | */ | |
483 | tic->t_curr_res = 0; | |
484 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
485 | return error; | |
486 | } | |
487 | ||
488 | ||
1da177e4 LT |
489 | /* |
490 | * NOTES: | |
491 | * | |
492 | * 1. currblock field gets updated at startup and after in-core logs | |
493 | * marked as with WANT_SYNC. | |
494 | */ | |
495 | ||
496 | /* | |
497 | * This routine is called when a user of a log manager ticket is done with | |
498 | * the reservation. If the ticket was ever used, then a commit record for | |
499 | * the associated transaction is written out as a log operation header with | |
500 | * no data. The flag XLOG_TIC_INITED is set when the first write occurs with | |
501 | * a given ticket. If the ticket was one with a permanent reservation, then | |
502 | * a few operations are done differently. Permanent reservation tickets by | |
503 | * default don't release the reservation. They just commit the current | |
504 | * transaction with the belief that the reservation is still needed. A flag | |
505 | * must be passed in before permanent reservations are actually released. | |
506 | * When these type of tickets are not released, they need to be set into | |
507 | * the inited state again. By doing this, a start record will be written | |
508 | * out when the next write occurs. | |
509 | */ | |
510 | xfs_lsn_t | |
35a8a72f CH |
511 | xfs_log_done( |
512 | struct xfs_mount *mp, | |
513 | struct xlog_ticket *ticket, | |
514 | struct xlog_in_core **iclog, | |
515 | uint flags) | |
1da177e4 | 516 | { |
ad223e60 | 517 | struct xlog *log = mp->m_log; |
35a8a72f | 518 | xfs_lsn_t lsn = 0; |
1da177e4 | 519 | |
1da177e4 LT |
520 | if (XLOG_FORCED_SHUTDOWN(log) || |
521 | /* | |
522 | * If nothing was ever written, don't write out commit record. | |
523 | * If we get an error, just continue and give back the log ticket. | |
524 | */ | |
525 | (((ticket->t_flags & XLOG_TIC_INITED) == 0) && | |
55b66332 | 526 | (xlog_commit_record(log, ticket, iclog, &lsn)))) { |
1da177e4 LT |
527 | lsn = (xfs_lsn_t) -1; |
528 | if (ticket->t_flags & XLOG_TIC_PERM_RESERV) { | |
529 | flags |= XFS_LOG_REL_PERM_RESERV; | |
530 | } | |
531 | } | |
532 | ||
533 | ||
534 | if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 || | |
535 | (flags & XFS_LOG_REL_PERM_RESERV)) { | |
0b1b213f CH |
536 | trace_xfs_log_done_nonperm(log, ticket); |
537 | ||
1da177e4 | 538 | /* |
c41564b5 | 539 | * Release ticket if not permanent reservation or a specific |
1da177e4 LT |
540 | * request has been made to release a permanent reservation. |
541 | */ | |
542 | xlog_ungrant_log_space(log, ticket); | |
cc09c0dc | 543 | xfs_log_ticket_put(ticket); |
1da177e4 | 544 | } else { |
0b1b213f CH |
545 | trace_xfs_log_done_perm(log, ticket); |
546 | ||
1da177e4 | 547 | xlog_regrant_reserve_log_space(log, ticket); |
c6a7b0f8 LM |
548 | /* If this ticket was a permanent reservation and we aren't |
549 | * trying to release it, reset the inited flags; so next time | |
550 | * we write, a start record will be written out. | |
551 | */ | |
1da177e4 | 552 | ticket->t_flags |= XLOG_TIC_INITED; |
c6a7b0f8 | 553 | } |
1da177e4 LT |
554 | |
555 | return lsn; | |
35a8a72f | 556 | } |
1da177e4 | 557 | |
1da177e4 LT |
558 | /* |
559 | * Attaches a new iclog I/O completion callback routine during | |
560 | * transaction commit. If the log is in error state, a non-zero | |
561 | * return code is handed back and the caller is responsible for | |
562 | * executing the callback at an appropriate time. | |
563 | */ | |
564 | int | |
35a8a72f CH |
565 | xfs_log_notify( |
566 | struct xfs_mount *mp, | |
567 | struct xlog_in_core *iclog, | |
568 | xfs_log_callback_t *cb) | |
1da177e4 | 569 | { |
b22cd72c | 570 | int abortflg; |
1da177e4 | 571 | |
114d23aa | 572 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
573 | abortflg = (iclog->ic_state & XLOG_STATE_IOERROR); |
574 | if (!abortflg) { | |
575 | ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) || | |
576 | (iclog->ic_state == XLOG_STATE_WANT_SYNC)); | |
577 | cb->cb_next = NULL; | |
578 | *(iclog->ic_callback_tail) = cb; | |
579 | iclog->ic_callback_tail = &(cb->cb_next); | |
580 | } | |
114d23aa | 581 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 | 582 | return abortflg; |
35a8a72f | 583 | } |
1da177e4 LT |
584 | |
585 | int | |
35a8a72f CH |
586 | xfs_log_release_iclog( |
587 | struct xfs_mount *mp, | |
588 | struct xlog_in_core *iclog) | |
1da177e4 | 589 | { |
35a8a72f | 590 | if (xlog_state_release_iclog(mp->m_log, iclog)) { |
7d04a335 | 591 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 592 | return EIO; |
1da177e4 LT |
593 | } |
594 | ||
595 | return 0; | |
596 | } | |
597 | ||
1da177e4 LT |
598 | /* |
599 | * Mount a log filesystem | |
600 | * | |
601 | * mp - ubiquitous xfs mount point structure | |
602 | * log_target - buftarg of on-disk log device | |
603 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
604 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
605 | * | |
606 | * Return error or zero. | |
607 | */ | |
608 | int | |
249a8c11 DC |
609 | xfs_log_mount( |
610 | xfs_mount_t *mp, | |
611 | xfs_buftarg_t *log_target, | |
612 | xfs_daddr_t blk_offset, | |
613 | int num_bblks) | |
1da177e4 | 614 | { |
249a8c11 DC |
615 | int error; |
616 | ||
1da177e4 | 617 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) |
a0fa2b67 | 618 | xfs_notice(mp, "Mounting Filesystem"); |
1da177e4 | 619 | else { |
a0fa2b67 DC |
620 | xfs_notice(mp, |
621 | "Mounting filesystem in no-recovery mode. Filesystem will be inconsistent."); | |
bd186aa9 | 622 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
623 | } |
624 | ||
625 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e DC |
626 | if (IS_ERR(mp->m_log)) { |
627 | error = -PTR_ERR(mp->m_log); | |
644c3567 DC |
628 | goto out; |
629 | } | |
1da177e4 | 630 | |
249a8c11 DC |
631 | /* |
632 | * Initialize the AIL now we have a log. | |
633 | */ | |
249a8c11 DC |
634 | error = xfs_trans_ail_init(mp); |
635 | if (error) { | |
a0fa2b67 | 636 | xfs_warn(mp, "AIL initialisation failed: error %d", error); |
26430752 | 637 | goto out_free_log; |
249a8c11 | 638 | } |
a9c21c1b | 639 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 640 | |
1da177e4 LT |
641 | /* |
642 | * skip log recovery on a norecovery mount. pretend it all | |
643 | * just worked. | |
644 | */ | |
645 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 646 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
647 | |
648 | if (readonly) | |
bd186aa9 | 649 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 650 | |
65be6054 | 651 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
652 | |
653 | if (readonly) | |
bd186aa9 | 654 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 655 | if (error) { |
a0fa2b67 DC |
656 | xfs_warn(mp, "log mount/recovery failed: error %d", |
657 | error); | |
26430752 | 658 | goto out_destroy_ail; |
1da177e4 LT |
659 | } |
660 | } | |
661 | ||
662 | /* Normal transactions can now occur */ | |
663 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
664 | ||
71e330b5 DC |
665 | /* |
666 | * Now the log has been fully initialised and we know were our | |
667 | * space grant counters are, we can initialise the permanent ticket | |
668 | * needed for delayed logging to work. | |
669 | */ | |
670 | xlog_cil_init_post_recovery(mp->m_log); | |
671 | ||
1da177e4 | 672 | return 0; |
26430752 CH |
673 | |
674 | out_destroy_ail: | |
675 | xfs_trans_ail_destroy(mp); | |
676 | out_free_log: | |
677 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 678 | out: |
249a8c11 | 679 | return error; |
26430752 | 680 | } |
1da177e4 LT |
681 | |
682 | /* | |
f661f1e0 DC |
683 | * Finish the recovery of the file system. This is separate from the |
684 | * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read | |
685 | * in the root and real-time bitmap inodes between calling xfs_log_mount() and | |
686 | * here. | |
1da177e4 | 687 | * |
f661f1e0 DC |
688 | * If we finish recovery successfully, start the background log work. If we are |
689 | * not doing recovery, then we have a RO filesystem and we don't need to start | |
690 | * it. | |
1da177e4 LT |
691 | */ |
692 | int | |
4249023a | 693 | xfs_log_mount_finish(xfs_mount_t *mp) |
1da177e4 | 694 | { |
f661f1e0 | 695 | int error = 0; |
1da177e4 | 696 | |
f661f1e0 | 697 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { |
4249023a | 698 | error = xlog_recover_finish(mp->m_log); |
f661f1e0 DC |
699 | if (!error) |
700 | xfs_log_work_queue(mp); | |
701 | } else { | |
bd186aa9 | 702 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
703 | } |
704 | ||
f661f1e0 | 705 | |
1da177e4 LT |
706 | return error; |
707 | } | |
708 | ||
1da177e4 LT |
709 | /* |
710 | * Final log writes as part of unmount. | |
711 | * | |
712 | * Mark the filesystem clean as unmount happens. Note that during relocation | |
713 | * this routine needs to be executed as part of source-bag while the | |
714 | * deallocation must not be done until source-end. | |
715 | */ | |
716 | ||
717 | /* | |
718 | * Unmount record used to have a string "Unmount filesystem--" in the | |
719 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
720 | * We just write the magic number now since that particular field isn't | |
721 | * currently architecture converted and "nUmount" is a bit foo. | |
722 | * As far as I know, there weren't any dependencies on the old behaviour. | |
723 | */ | |
724 | ||
725 | int | |
726 | xfs_log_unmount_write(xfs_mount_t *mp) | |
727 | { | |
9a8d2fdb | 728 | struct xlog *log = mp->m_log; |
1da177e4 LT |
729 | xlog_in_core_t *iclog; |
730 | #ifdef DEBUG | |
731 | xlog_in_core_t *first_iclog; | |
732 | #endif | |
35a8a72f | 733 | xlog_ticket_t *tic = NULL; |
1da177e4 LT |
734 | xfs_lsn_t lsn; |
735 | int error; | |
1da177e4 | 736 | |
1da177e4 LT |
737 | /* |
738 | * Don't write out unmount record on read-only mounts. | |
739 | * Or, if we are doing a forced umount (typically because of IO errors). | |
740 | */ | |
bd186aa9 | 741 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
742 | return 0; |
743 | ||
a14a348b | 744 | error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
b911ca04 | 745 | ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log))); |
1da177e4 LT |
746 | |
747 | #ifdef DEBUG | |
748 | first_iclog = iclog = log->l_iclog; | |
749 | do { | |
750 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
751 | ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE); | |
752 | ASSERT(iclog->ic_offset == 0); | |
753 | } | |
754 | iclog = iclog->ic_next; | |
755 | } while (iclog != first_iclog); | |
756 | #endif | |
757 | if (! (XLOG_FORCED_SHUTDOWN(log))) { | |
955e47ad TS |
758 | error = xfs_log_reserve(mp, 600, 1, &tic, |
759 | XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE); | |
1da177e4 | 760 | if (!error) { |
55b66332 DC |
761 | /* the data section must be 32 bit size aligned */ |
762 | struct { | |
763 | __uint16_t magic; | |
764 | __uint16_t pad1; | |
765 | __uint32_t pad2; /* may as well make it 64 bits */ | |
766 | } magic = { | |
767 | .magic = XLOG_UNMOUNT_TYPE, | |
768 | }; | |
769 | struct xfs_log_iovec reg = { | |
4e0d5f92 | 770 | .i_addr = &magic, |
55b66332 DC |
771 | .i_len = sizeof(magic), |
772 | .i_type = XLOG_REG_TYPE_UNMOUNT, | |
773 | }; | |
774 | struct xfs_log_vec vec = { | |
775 | .lv_niovecs = 1, | |
776 | .lv_iovecp = ®, | |
777 | }; | |
778 | ||
3948659e | 779 | /* remove inited flag, and account for space used */ |
55b66332 | 780 | tic->t_flags = 0; |
3948659e | 781 | tic->t_curr_res -= sizeof(magic); |
55b66332 | 782 | error = xlog_write(log, &vec, tic, &lsn, |
1da177e4 LT |
783 | NULL, XLOG_UNMOUNT_TRANS); |
784 | /* | |
785 | * At this point, we're umounting anyway, | |
786 | * so there's no point in transitioning log state | |
787 | * to IOERROR. Just continue... | |
788 | */ | |
789 | } | |
790 | ||
a0fa2b67 DC |
791 | if (error) |
792 | xfs_alert(mp, "%s: unmount record failed", __func__); | |
1da177e4 LT |
793 | |
794 | ||
b22cd72c | 795 | spin_lock(&log->l_icloglock); |
1da177e4 | 796 | iclog = log->l_iclog; |
155cc6b7 | 797 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 798 | xlog_state_want_sync(log, iclog); |
39e2defe | 799 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 800 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 801 | |
b22cd72c | 802 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
803 | if (!(iclog->ic_state == XLOG_STATE_ACTIVE || |
804 | iclog->ic_state == XLOG_STATE_DIRTY)) { | |
805 | if (!XLOG_FORCED_SHUTDOWN(log)) { | |
eb40a875 DC |
806 | xlog_wait(&iclog->ic_force_wait, |
807 | &log->l_icloglock); | |
1da177e4 | 808 | } else { |
b22cd72c | 809 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
810 | } |
811 | } else { | |
b22cd72c | 812 | spin_unlock(&log->l_icloglock); |
1da177e4 | 813 | } |
955e47ad | 814 | if (tic) { |
0b1b213f | 815 | trace_xfs_log_umount_write(log, tic); |
955e47ad | 816 | xlog_ungrant_log_space(log, tic); |
cc09c0dc | 817 | xfs_log_ticket_put(tic); |
955e47ad | 818 | } |
1da177e4 LT |
819 | } else { |
820 | /* | |
821 | * We're already in forced_shutdown mode, couldn't | |
822 | * even attempt to write out the unmount transaction. | |
823 | * | |
824 | * Go through the motions of sync'ing and releasing | |
825 | * the iclog, even though no I/O will actually happen, | |
c41564b5 | 826 | * we need to wait for other log I/Os that may already |
1da177e4 LT |
827 | * be in progress. Do this as a separate section of |
828 | * code so we'll know if we ever get stuck here that | |
829 | * we're in this odd situation of trying to unmount | |
830 | * a file system that went into forced_shutdown as | |
831 | * the result of an unmount.. | |
832 | */ | |
b22cd72c | 833 | spin_lock(&log->l_icloglock); |
1da177e4 | 834 | iclog = log->l_iclog; |
155cc6b7 | 835 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 LT |
836 | |
837 | xlog_state_want_sync(log, iclog); | |
39e2defe | 838 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 839 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 840 | |
b22cd72c | 841 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
842 | |
843 | if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE | |
844 | || iclog->ic_state == XLOG_STATE_DIRTY | |
845 | || iclog->ic_state == XLOG_STATE_IOERROR) ) { | |
846 | ||
eb40a875 DC |
847 | xlog_wait(&iclog->ic_force_wait, |
848 | &log->l_icloglock); | |
1da177e4 | 849 | } else { |
b22cd72c | 850 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
851 | } |
852 | } | |
853 | ||
1bb7d6b5 | 854 | return error; |
1da177e4 LT |
855 | } /* xfs_log_unmount_write */ |
856 | ||
857 | /* | |
cf2931db | 858 | * Shut down and release the AIL and Log. |
249a8c11 | 859 | * |
cf2931db DC |
860 | * During unmount, we need to ensure we flush all the dirty metadata objects |
861 | * from the AIL so that the log is empty before we write the unmount record to | |
862 | * the log. | |
863 | * | |
864 | * To do this, we first need to shut down the background log work so it is not | |
865 | * trying to cover the log as we clean up. We then need to unpin all objects in | |
866 | * the log so we can then flush them out. Once they have completed their IO and | |
867 | * run the callbacks removing themselves from the AIL, we can write the unmount | |
868 | * record, tear down the AIL and finally free the log. | |
1da177e4 LT |
869 | */ |
870 | void | |
21b699c8 | 871 | xfs_log_unmount(xfs_mount_t *mp) |
1da177e4 | 872 | { |
f661f1e0 | 873 | cancel_delayed_work_sync(&mp->m_log->l_work); |
cf2931db DC |
874 | xfs_log_force(mp, XFS_LOG_SYNC); |
875 | ||
876 | /* | |
877 | * The superblock buffer is uncached and while xfs_ail_push_all_sync() | |
878 | * will push it, xfs_wait_buftarg() will not wait for it. Further, | |
879 | * xfs_buf_iowait() cannot be used because it was pushed with the | |
880 | * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for | |
881 | * the IO to complete. | |
882 | */ | |
883 | xfs_ail_push_all_sync(mp->m_ail); | |
884 | xfs_wait_buftarg(mp->m_ddev_targp); | |
885 | xfs_buf_lock(mp->m_sb_bp); | |
886 | xfs_buf_unlock(mp->m_sb_bp); | |
887 | ||
888 | xfs_log_unmount_write(mp); | |
889 | ||
249a8c11 | 890 | xfs_trans_ail_destroy(mp); |
c41564b5 | 891 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
892 | } |
893 | ||
43f5efc5 DC |
894 | void |
895 | xfs_log_item_init( | |
896 | struct xfs_mount *mp, | |
897 | struct xfs_log_item *item, | |
898 | int type, | |
272e42b2 | 899 | const struct xfs_item_ops *ops) |
43f5efc5 DC |
900 | { |
901 | item->li_mountp = mp; | |
902 | item->li_ailp = mp->m_ail; | |
903 | item->li_type = type; | |
904 | item->li_ops = ops; | |
71e330b5 DC |
905 | item->li_lv = NULL; |
906 | ||
907 | INIT_LIST_HEAD(&item->li_ail); | |
908 | INIT_LIST_HEAD(&item->li_cil); | |
43f5efc5 DC |
909 | } |
910 | ||
09a423a3 CH |
911 | /* |
912 | * Wake up processes waiting for log space after we have moved the log tail. | |
09a423a3 | 913 | */ |
1da177e4 | 914 | void |
09a423a3 | 915 | xfs_log_space_wake( |
cfb7cdca | 916 | struct xfs_mount *mp) |
1da177e4 | 917 | { |
ad223e60 | 918 | struct xlog *log = mp->m_log; |
cfb7cdca | 919 | int free_bytes; |
1da177e4 | 920 | |
1da177e4 LT |
921 | if (XLOG_FORCED_SHUTDOWN(log)) |
922 | return; | |
1da177e4 | 923 | |
28496968 | 924 | if (!list_empty_careful(&log->l_write_head.waiters)) { |
09a423a3 CH |
925 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
926 | ||
28496968 CH |
927 | spin_lock(&log->l_write_head.lock); |
928 | free_bytes = xlog_space_left(log, &log->l_write_head.grant); | |
e179840d | 929 | xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); |
28496968 | 930 | spin_unlock(&log->l_write_head.lock); |
1da177e4 | 931 | } |
10547941 | 932 | |
28496968 | 933 | if (!list_empty_careful(&log->l_reserve_head.waiters)) { |
09a423a3 CH |
934 | ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); |
935 | ||
28496968 CH |
936 | spin_lock(&log->l_reserve_head.lock); |
937 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); | |
e179840d | 938 | xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); |
28496968 | 939 | spin_unlock(&log->l_reserve_head.lock); |
1da177e4 | 940 | } |
3f16b985 | 941 | } |
1da177e4 LT |
942 | |
943 | /* | |
944 | * Determine if we have a transaction that has gone to disk | |
b6f8dd49 DC |
945 | * that needs to be covered. To begin the transition to the idle state |
946 | * firstly the log needs to be idle (no AIL and nothing in the iclogs). | |
947 | * If we are then in a state where covering is needed, the caller is informed | |
948 | * that dummy transactions are required to move the log into the idle state. | |
949 | * | |
950 | * Because this is called as part of the sync process, we should also indicate | |
951 | * that dummy transactions should be issued in anything but the covered or | |
952 | * idle states. This ensures that the log tail is accurately reflected in | |
953 | * the log at the end of the sync, hence if a crash occurrs avoids replay | |
954 | * of transactions where the metadata is already on disk. | |
1da177e4 LT |
955 | */ |
956 | int | |
957 | xfs_log_need_covered(xfs_mount_t *mp) | |
958 | { | |
27d8d5fe | 959 | int needed = 0; |
9a8d2fdb | 960 | struct xlog *log = mp->m_log; |
1da177e4 | 961 | |
92821e2b | 962 | if (!xfs_fs_writable(mp)) |
1da177e4 LT |
963 | return 0; |
964 | ||
b22cd72c | 965 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
966 | switch (log->l_covered_state) { |
967 | case XLOG_STATE_COVER_DONE: | |
968 | case XLOG_STATE_COVER_DONE2: | |
969 | case XLOG_STATE_COVER_IDLE: | |
970 | break; | |
971 | case XLOG_STATE_COVER_NEED: | |
972 | case XLOG_STATE_COVER_NEED2: | |
fd074841 | 973 | if (!xfs_ail_min_lsn(log->l_ailp) && |
b6f8dd49 DC |
974 | xlog_iclogs_empty(log)) { |
975 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) | |
976 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
977 | else | |
978 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1da177e4 | 979 | } |
b6f8dd49 DC |
980 | /* FALLTHRU */ |
981 | default: | |
1da177e4 | 982 | needed = 1; |
b6f8dd49 | 983 | break; |
1da177e4 | 984 | } |
b22cd72c | 985 | spin_unlock(&log->l_icloglock); |
014c2544 | 986 | return needed; |
1da177e4 LT |
987 | } |
988 | ||
09a423a3 | 989 | /* |
1da177e4 LT |
990 | * We may be holding the log iclog lock upon entering this routine. |
991 | */ | |
992 | xfs_lsn_t | |
1c304625 | 993 | xlog_assign_tail_lsn_locked( |
1c3cb9ec | 994 | struct xfs_mount *mp) |
1da177e4 | 995 | { |
ad223e60 | 996 | struct xlog *log = mp->m_log; |
1c304625 CH |
997 | struct xfs_log_item *lip; |
998 | xfs_lsn_t tail_lsn; | |
999 | ||
1000 | assert_spin_locked(&mp->m_ail->xa_lock); | |
1da177e4 | 1001 | |
09a423a3 CH |
1002 | /* |
1003 | * To make sure we always have a valid LSN for the log tail we keep | |
1004 | * track of the last LSN which was committed in log->l_last_sync_lsn, | |
1c304625 | 1005 | * and use that when the AIL was empty. |
09a423a3 | 1006 | */ |
1c304625 CH |
1007 | lip = xfs_ail_min(mp->m_ail); |
1008 | if (lip) | |
1009 | tail_lsn = lip->li_lsn; | |
1010 | else | |
84f3c683 | 1011 | tail_lsn = atomic64_read(&log->l_last_sync_lsn); |
1c3cb9ec | 1012 | atomic64_set(&log->l_tail_lsn, tail_lsn); |
1da177e4 | 1013 | return tail_lsn; |
1c3cb9ec | 1014 | } |
1da177e4 | 1015 | |
1c304625 CH |
1016 | xfs_lsn_t |
1017 | xlog_assign_tail_lsn( | |
1018 | struct xfs_mount *mp) | |
1019 | { | |
1020 | xfs_lsn_t tail_lsn; | |
1021 | ||
1022 | spin_lock(&mp->m_ail->xa_lock); | |
1023 | tail_lsn = xlog_assign_tail_lsn_locked(mp); | |
1024 | spin_unlock(&mp->m_ail->xa_lock); | |
1025 | ||
1026 | return tail_lsn; | |
1027 | } | |
1028 | ||
1da177e4 LT |
1029 | /* |
1030 | * Return the space in the log between the tail and the head. The head | |
1031 | * is passed in the cycle/bytes formal parms. In the special case where | |
1032 | * the reserve head has wrapped passed the tail, this calculation is no | |
1033 | * longer valid. In this case, just return 0 which means there is no space | |
1034 | * in the log. This works for all places where this function is called | |
1035 | * with the reserve head. Of course, if the write head were to ever | |
1036 | * wrap the tail, we should blow up. Rather than catch this case here, | |
1037 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
1038 | * | |
1039 | * This code also handles the case where the reservation head is behind | |
1040 | * the tail. The details of this case are described below, but the end | |
1041 | * result is that we return the size of the log as the amount of space left. | |
1042 | */ | |
a8272ce0 | 1043 | STATIC int |
a69ed03c | 1044 | xlog_space_left( |
ad223e60 | 1045 | struct xlog *log, |
c8a09ff8 | 1046 | atomic64_t *head) |
1da177e4 | 1047 | { |
a69ed03c DC |
1048 | int free_bytes; |
1049 | int tail_bytes; | |
1050 | int tail_cycle; | |
1051 | int head_cycle; | |
1052 | int head_bytes; | |
1da177e4 | 1053 | |
a69ed03c | 1054 | xlog_crack_grant_head(head, &head_cycle, &head_bytes); |
1c3cb9ec DC |
1055 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); |
1056 | tail_bytes = BBTOB(tail_bytes); | |
a69ed03c DC |
1057 | if (tail_cycle == head_cycle && head_bytes >= tail_bytes) |
1058 | free_bytes = log->l_logsize - (head_bytes - tail_bytes); | |
1059 | else if (tail_cycle + 1 < head_cycle) | |
1da177e4 | 1060 | return 0; |
a69ed03c DC |
1061 | else if (tail_cycle < head_cycle) { |
1062 | ASSERT(tail_cycle == (head_cycle - 1)); | |
1063 | free_bytes = tail_bytes - head_bytes; | |
1da177e4 LT |
1064 | } else { |
1065 | /* | |
1066 | * The reservation head is behind the tail. | |
1067 | * In this case we just want to return the size of the | |
1068 | * log as the amount of space left. | |
1069 | */ | |
a0fa2b67 | 1070 | xfs_alert(log->l_mp, |
1da177e4 LT |
1071 | "xlog_space_left: head behind tail\n" |
1072 | " tail_cycle = %d, tail_bytes = %d\n" | |
1073 | " GH cycle = %d, GH bytes = %d", | |
a69ed03c | 1074 | tail_cycle, tail_bytes, head_cycle, head_bytes); |
1da177e4 LT |
1075 | ASSERT(0); |
1076 | free_bytes = log->l_logsize; | |
1077 | } | |
1078 | return free_bytes; | |
a69ed03c | 1079 | } |
1da177e4 LT |
1080 | |
1081 | ||
1082 | /* | |
1083 | * Log function which is called when an io completes. | |
1084 | * | |
1085 | * The log manager needs its own routine, in order to control what | |
1086 | * happens with the buffer after the write completes. | |
1087 | */ | |
1088 | void | |
1089 | xlog_iodone(xfs_buf_t *bp) | |
1090 | { | |
9a8d2fdb MT |
1091 | struct xlog_in_core *iclog = bp->b_fspriv; |
1092 | struct xlog *l = iclog->ic_log; | |
1093 | int aborted = 0; | |
1da177e4 LT |
1094 | |
1095 | /* | |
1096 | * Race to shutdown the filesystem if we see an error. | |
1097 | */ | |
5a52c2a5 | 1098 | if (XFS_TEST_ERROR((xfs_buf_geterror(bp)), l->l_mp, |
1da177e4 | 1099 | XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) { |
901796af | 1100 | xfs_buf_ioerror_alert(bp, __func__); |
c867cb61 | 1101 | xfs_buf_stale(bp); |
7d04a335 | 1102 | xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 LT |
1103 | /* |
1104 | * This flag will be propagated to the trans-committed | |
1105 | * callback routines to let them know that the log-commit | |
1106 | * didn't succeed. | |
1107 | */ | |
1108 | aborted = XFS_LI_ABORTED; | |
1109 | } else if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
1110 | aborted = XFS_LI_ABORTED; | |
1111 | } | |
3db296f3 DC |
1112 | |
1113 | /* log I/O is always issued ASYNC */ | |
1114 | ASSERT(XFS_BUF_ISASYNC(bp)); | |
1da177e4 | 1115 | xlog_state_done_syncing(iclog, aborted); |
3db296f3 DC |
1116 | /* |
1117 | * do not reference the buffer (bp) here as we could race | |
1118 | * with it being freed after writing the unmount record to the | |
1119 | * log. | |
1120 | */ | |
1121 | ||
1da177e4 LT |
1122 | } /* xlog_iodone */ |
1123 | ||
1da177e4 LT |
1124 | /* |
1125 | * Return size of each in-core log record buffer. | |
1126 | * | |
9da096fd | 1127 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
1128 | * |
1129 | * If the filesystem blocksize is too large, we may need to choose a | |
1130 | * larger size since the directory code currently logs entire blocks. | |
1131 | */ | |
1132 | ||
1133 | STATIC void | |
9a8d2fdb MT |
1134 | xlog_get_iclog_buffer_size( |
1135 | struct xfs_mount *mp, | |
1136 | struct xlog *log) | |
1da177e4 LT |
1137 | { |
1138 | int size; | |
1139 | int xhdrs; | |
1140 | ||
1cb51258 ES |
1141 | if (mp->m_logbufs <= 0) |
1142 | log->l_iclog_bufs = XLOG_MAX_ICLOGS; | |
1143 | else | |
cfcbbbd0 | 1144 | log->l_iclog_bufs = mp->m_logbufs; |
1da177e4 LT |
1145 | |
1146 | /* | |
1147 | * Buffer size passed in from mount system call. | |
1148 | */ | |
cfcbbbd0 | 1149 | if (mp->m_logbsize > 0) { |
1da177e4 LT |
1150 | size = log->l_iclog_size = mp->m_logbsize; |
1151 | log->l_iclog_size_log = 0; | |
1152 | while (size != 1) { | |
1153 | log->l_iclog_size_log++; | |
1154 | size >>= 1; | |
1155 | } | |
1156 | ||
62118709 | 1157 | if (xfs_sb_version_haslogv2(&mp->m_sb)) { |
9da096fd MP |
1158 | /* # headers = size / 32k |
1159 | * one header holds cycles from 32k of data | |
1da177e4 LT |
1160 | */ |
1161 | ||
1162 | xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE; | |
1163 | if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE) | |
1164 | xhdrs++; | |
1165 | log->l_iclog_hsize = xhdrs << BBSHIFT; | |
1166 | log->l_iclog_heads = xhdrs; | |
1167 | } else { | |
1168 | ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE); | |
1169 | log->l_iclog_hsize = BBSIZE; | |
1170 | log->l_iclog_heads = 1; | |
1171 | } | |
cfcbbbd0 | 1172 | goto done; |
1da177e4 LT |
1173 | } |
1174 | ||
9da096fd | 1175 | /* All machines use 32kB buffers by default. */ |
1cb51258 ES |
1176 | log->l_iclog_size = XLOG_BIG_RECORD_BSIZE; |
1177 | log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT; | |
1da177e4 LT |
1178 | |
1179 | /* the default log size is 16k or 32k which is one header sector */ | |
1180 | log->l_iclog_hsize = BBSIZE; | |
1181 | log->l_iclog_heads = 1; | |
1182 | ||
7153f8ba CH |
1183 | done: |
1184 | /* are we being asked to make the sizes selected above visible? */ | |
cfcbbbd0 NS |
1185 | if (mp->m_logbufs == 0) |
1186 | mp->m_logbufs = log->l_iclog_bufs; | |
1187 | if (mp->m_logbsize == 0) | |
1188 | mp->m_logbsize = log->l_iclog_size; | |
1da177e4 LT |
1189 | } /* xlog_get_iclog_buffer_size */ |
1190 | ||
1191 | ||
f661f1e0 DC |
1192 | void |
1193 | xfs_log_work_queue( | |
1194 | struct xfs_mount *mp) | |
1195 | { | |
1196 | queue_delayed_work(xfs_syncd_wq, &mp->m_log->l_work, | |
1197 | msecs_to_jiffies(xfs_syncd_centisecs * 10)); | |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * Every sync period we need to unpin all items in the AIL and push them to | |
1202 | * disk. If there is nothing dirty, then we might need to cover the log to | |
1203 | * indicate that the filesystem is idle. | |
1204 | */ | |
1205 | void | |
1206 | xfs_log_worker( | |
1207 | struct work_struct *work) | |
1208 | { | |
1209 | struct xlog *log = container_of(to_delayed_work(work), | |
1210 | struct xlog, l_work); | |
1211 | struct xfs_mount *mp = log->l_mp; | |
1212 | ||
1213 | /* dgc: errors ignored - not fatal and nowhere to report them */ | |
1214 | if (xfs_log_need_covered(mp)) | |
1215 | xfs_fs_log_dummy(mp); | |
1216 | else | |
1217 | xfs_log_force(mp, 0); | |
1218 | ||
1219 | /* start pushing all the metadata that is currently dirty */ | |
1220 | xfs_ail_push_all(mp->m_ail); | |
1221 | ||
1222 | /* queue us up again */ | |
1223 | xfs_log_work_queue(mp); | |
1224 | } | |
1225 | ||
1da177e4 LT |
1226 | /* |
1227 | * This routine initializes some of the log structure for a given mount point. | |
1228 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1229 | * some other stuff may be filled in too. | |
1230 | */ | |
9a8d2fdb MT |
1231 | STATIC struct xlog * |
1232 | xlog_alloc_log( | |
1233 | struct xfs_mount *mp, | |
1234 | struct xfs_buftarg *log_target, | |
1235 | xfs_daddr_t blk_offset, | |
1236 | int num_bblks) | |
1da177e4 | 1237 | { |
9a8d2fdb | 1238 | struct xlog *log; |
1da177e4 LT |
1239 | xlog_rec_header_t *head; |
1240 | xlog_in_core_t **iclogp; | |
1241 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1242 | xfs_buf_t *bp; | |
1243 | int i; | |
a6cb767e | 1244 | int error = ENOMEM; |
69ce58f0 | 1245 | uint log2_size = 0; |
1da177e4 | 1246 | |
9a8d2fdb | 1247 | log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); |
a6cb767e | 1248 | if (!log) { |
a0fa2b67 | 1249 | xfs_warn(mp, "Log allocation failed: No memory!"); |
a6cb767e DC |
1250 | goto out; |
1251 | } | |
1da177e4 LT |
1252 | |
1253 | log->l_mp = mp; | |
1254 | log->l_targ = log_target; | |
1255 | log->l_logsize = BBTOB(num_bblks); | |
1256 | log->l_logBBstart = blk_offset; | |
1257 | log->l_logBBsize = num_bblks; | |
1258 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1259 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
f661f1e0 | 1260 | INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); |
1da177e4 LT |
1261 | |
1262 | log->l_prev_block = -1; | |
1da177e4 | 1263 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1c3cb9ec DC |
1264 | xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); |
1265 | xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); | |
1da177e4 | 1266 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ |
c303c5b8 CH |
1267 | |
1268 | xlog_grant_head_init(&log->l_reserve_head); | |
1269 | xlog_grant_head_init(&log->l_write_head); | |
1da177e4 | 1270 | |
a6cb767e | 1271 | error = EFSCORRUPTED; |
62118709 | 1272 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1273 | log2_size = mp->m_sb.sb_logsectlog; |
1274 | if (log2_size < BBSHIFT) { | |
a0fa2b67 DC |
1275 | xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", |
1276 | log2_size, BBSHIFT); | |
a6cb767e DC |
1277 | goto out_free_log; |
1278 | } | |
1279 | ||
69ce58f0 AE |
1280 | log2_size -= BBSHIFT; |
1281 | if (log2_size > mp->m_sectbb_log) { | |
a0fa2b67 DC |
1282 | xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", |
1283 | log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1284 | goto out_free_log; |
1285 | } | |
69ce58f0 AE |
1286 | |
1287 | /* for larger sector sizes, must have v2 or external log */ | |
1288 | if (log2_size && log->l_logBBstart > 0 && | |
1289 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
a0fa2b67 DC |
1290 | xfs_warn(mp, |
1291 | "log sector size (0x%x) invalid for configuration.", | |
1292 | log2_size); | |
a6cb767e DC |
1293 | goto out_free_log; |
1294 | } | |
1da177e4 | 1295 | } |
69ce58f0 | 1296 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1297 | |
1298 | xlog_get_iclog_buffer_size(mp, log); | |
1299 | ||
a6cb767e | 1300 | error = ENOMEM; |
e70b73f8 | 1301 | bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0); |
644c3567 DC |
1302 | if (!bp) |
1303 | goto out_free_log; | |
cb669ca5 | 1304 | bp->b_iodone = xlog_iodone; |
0c842ad4 | 1305 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
1306 | log->l_xbuf = bp; |
1307 | ||
007c61c6 | 1308 | spin_lock_init(&log->l_icloglock); |
eb40a875 | 1309 | init_waitqueue_head(&log->l_flush_wait); |
1da177e4 | 1310 | |
1da177e4 LT |
1311 | iclogp = &log->l_iclog; |
1312 | /* | |
1313 | * The amount of memory to allocate for the iclog structure is | |
1314 | * rather funky due to the way the structure is defined. It is | |
1315 | * done this way so that we can use different sizes for machines | |
1316 | * with different amounts of memory. See the definition of | |
1317 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1318 | */ | |
1da177e4 LT |
1319 | ASSERT(log->l_iclog_size >= 4096); |
1320 | for (i=0; i < log->l_iclog_bufs; i++) { | |
644c3567 DC |
1321 | *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL); |
1322 | if (!*iclogp) | |
1323 | goto out_free_iclog; | |
1324 | ||
1da177e4 | 1325 | iclog = *iclogp; |
1da177e4 LT |
1326 | iclog->ic_prev = prev_iclog; |
1327 | prev_iclog = iclog; | |
1fa40b01 | 1328 | |
686865f7 | 1329 | bp = xfs_buf_get_uncached(mp->m_logdev_targp, |
e70b73f8 | 1330 | BTOBB(log->l_iclog_size), 0); |
644c3567 DC |
1331 | if (!bp) |
1332 | goto out_free_iclog; | |
c8da0faf | 1333 | |
cb669ca5 | 1334 | bp->b_iodone = xlog_iodone; |
1fa40b01 | 1335 | iclog->ic_bp = bp; |
b28708d6 | 1336 | iclog->ic_data = bp->b_addr; |
4679b2d3 | 1337 | #ifdef DEBUG |
1da177e4 | 1338 | log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header); |
4679b2d3 | 1339 | #endif |
1da177e4 LT |
1340 | head = &iclog->ic_header; |
1341 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1342 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1343 | head->h_version = cpu_to_be32( | |
62118709 | 1344 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1345 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1346 | /* new fields */ |
b53e675d | 1347 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1348 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1349 | ||
4e94b71b | 1350 | iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize; |
1da177e4 LT |
1351 | iclog->ic_state = XLOG_STATE_ACTIVE; |
1352 | iclog->ic_log = log; | |
114d23aa DC |
1353 | atomic_set(&iclog->ic_refcnt, 0); |
1354 | spin_lock_init(&iclog->ic_callback_lock); | |
1da177e4 | 1355 | iclog->ic_callback_tail = &(iclog->ic_callback); |
b28708d6 | 1356 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 | 1357 | |
0c842ad4 | 1358 | ASSERT(xfs_buf_islocked(iclog->ic_bp)); |
eb40a875 DC |
1359 | init_waitqueue_head(&iclog->ic_force_wait); |
1360 | init_waitqueue_head(&iclog->ic_write_wait); | |
1da177e4 LT |
1361 | |
1362 | iclogp = &iclog->ic_next; | |
1363 | } | |
1364 | *iclogp = log->l_iclog; /* complete ring */ | |
1365 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1366 | ||
71e330b5 DC |
1367 | error = xlog_cil_init(log); |
1368 | if (error) | |
1369 | goto out_free_iclog; | |
1da177e4 | 1370 | return log; |
644c3567 DC |
1371 | |
1372 | out_free_iclog: | |
1373 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1374 | prev_iclog = iclog->ic_next; | |
eb40a875 | 1375 | if (iclog->ic_bp) |
644c3567 | 1376 | xfs_buf_free(iclog->ic_bp); |
644c3567 DC |
1377 | kmem_free(iclog); |
1378 | } | |
1379 | spinlock_destroy(&log->l_icloglock); | |
644c3567 DC |
1380 | xfs_buf_free(log->l_xbuf); |
1381 | out_free_log: | |
1382 | kmem_free(log); | |
a6cb767e DC |
1383 | out: |
1384 | return ERR_PTR(-error); | |
1da177e4 LT |
1385 | } /* xlog_alloc_log */ |
1386 | ||
1387 | ||
1388 | /* | |
1389 | * Write out the commit record of a transaction associated with the given | |
1390 | * ticket. Return the lsn of the commit record. | |
1391 | */ | |
1392 | STATIC int | |
55b66332 | 1393 | xlog_commit_record( |
ad223e60 | 1394 | struct xlog *log, |
55b66332 DC |
1395 | struct xlog_ticket *ticket, |
1396 | struct xlog_in_core **iclog, | |
1397 | xfs_lsn_t *commitlsnp) | |
1da177e4 | 1398 | { |
55b66332 DC |
1399 | struct xfs_mount *mp = log->l_mp; |
1400 | int error; | |
1401 | struct xfs_log_iovec reg = { | |
1402 | .i_addr = NULL, | |
1403 | .i_len = 0, | |
1404 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1405 | }; | |
1406 | struct xfs_log_vec vec = { | |
1407 | .lv_niovecs = 1, | |
1408 | .lv_iovecp = ®, | |
1409 | }; | |
1da177e4 LT |
1410 | |
1411 | ASSERT_ALWAYS(iclog); | |
55b66332 DC |
1412 | error = xlog_write(log, &vec, ticket, commitlsnp, iclog, |
1413 | XLOG_COMMIT_TRANS); | |
1414 | if (error) | |
7d04a335 | 1415 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1416 | return error; |
55b66332 | 1417 | } |
1da177e4 LT |
1418 | |
1419 | /* | |
1420 | * Push on the buffer cache code if we ever use more than 75% of the on-disk | |
1421 | * log space. This code pushes on the lsn which would supposedly free up | |
1422 | * the 25% which we want to leave free. We may need to adopt a policy which | |
1423 | * pushes on an lsn which is further along in the log once we reach the high | |
1424 | * water mark. In this manner, we would be creating a low water mark. | |
1425 | */ | |
a8272ce0 | 1426 | STATIC void |
2ced19cb | 1427 | xlog_grant_push_ail( |
ad223e60 | 1428 | struct xlog *log, |
2ced19cb | 1429 | int need_bytes) |
1da177e4 | 1430 | { |
2ced19cb | 1431 | xfs_lsn_t threshold_lsn = 0; |
84f3c683 | 1432 | xfs_lsn_t last_sync_lsn; |
2ced19cb DC |
1433 | int free_blocks; |
1434 | int free_bytes; | |
1435 | int threshold_block; | |
1436 | int threshold_cycle; | |
1437 | int free_threshold; | |
1438 | ||
1439 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1440 | ||
28496968 | 1441 | free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); |
2ced19cb DC |
1442 | free_blocks = BTOBBT(free_bytes); |
1443 | ||
1444 | /* | |
1445 | * Set the threshold for the minimum number of free blocks in the | |
1446 | * log to the maximum of what the caller needs, one quarter of the | |
1447 | * log, and 256 blocks. | |
1448 | */ | |
1449 | free_threshold = BTOBB(need_bytes); | |
1450 | free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2)); | |
1451 | free_threshold = MAX(free_threshold, 256); | |
1452 | if (free_blocks >= free_threshold) | |
1453 | return; | |
1454 | ||
1c3cb9ec DC |
1455 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, |
1456 | &threshold_block); | |
1457 | threshold_block += free_threshold; | |
1da177e4 | 1458 | if (threshold_block >= log->l_logBBsize) { |
2ced19cb DC |
1459 | threshold_block -= log->l_logBBsize; |
1460 | threshold_cycle += 1; | |
1da177e4 | 1461 | } |
2ced19cb DC |
1462 | threshold_lsn = xlog_assign_lsn(threshold_cycle, |
1463 | threshold_block); | |
1464 | /* | |
1465 | * Don't pass in an lsn greater than the lsn of the last | |
84f3c683 DC |
1466 | * log record known to be on disk. Use a snapshot of the last sync lsn |
1467 | * so that it doesn't change between the compare and the set. | |
1da177e4 | 1468 | */ |
84f3c683 DC |
1469 | last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); |
1470 | if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) | |
1471 | threshold_lsn = last_sync_lsn; | |
2ced19cb DC |
1472 | |
1473 | /* | |
1474 | * Get the transaction layer to kick the dirty buffers out to | |
1475 | * disk asynchronously. No point in trying to do this if | |
1476 | * the filesystem is shutting down. | |
1477 | */ | |
1478 | if (!XLOG_FORCED_SHUTDOWN(log)) | |
fd074841 | 1479 | xfs_ail_push(log->l_ailp, threshold_lsn); |
2ced19cb | 1480 | } |
1da177e4 | 1481 | |
873ff550 CH |
1482 | /* |
1483 | * The bdstrat callback function for log bufs. This gives us a central | |
1484 | * place to trap bufs in case we get hit by a log I/O error and need to | |
1485 | * shutdown. Actually, in practice, even when we didn't get a log error, | |
1486 | * we transition the iclogs to IOERROR state *after* flushing all existing | |
1487 | * iclogs to disk. This is because we don't want anymore new transactions to be | |
1488 | * started or completed afterwards. | |
1489 | */ | |
1490 | STATIC int | |
1491 | xlog_bdstrat( | |
1492 | struct xfs_buf *bp) | |
1493 | { | |
adadbeef | 1494 | struct xlog_in_core *iclog = bp->b_fspriv; |
873ff550 | 1495 | |
873ff550 | 1496 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
5a52c2a5 | 1497 | xfs_buf_ioerror(bp, EIO); |
c867cb61 | 1498 | xfs_buf_stale(bp); |
1a1a3e97 | 1499 | xfs_buf_ioend(bp, 0); |
873ff550 CH |
1500 | /* |
1501 | * It would seem logical to return EIO here, but we rely on | |
1502 | * the log state machine to propagate I/O errors instead of | |
1503 | * doing it here. | |
1504 | */ | |
1505 | return 0; | |
1506 | } | |
1507 | ||
873ff550 CH |
1508 | xfs_buf_iorequest(bp); |
1509 | return 0; | |
1510 | } | |
1da177e4 LT |
1511 | |
1512 | /* | |
1513 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1514 | * fashion. Previously, we should have moved the current iclog | |
1515 | * ptr in the log to point to the next available iclog. This allows further | |
1516 | * write to continue while this code syncs out an iclog ready to go. | |
1517 | * Before an in-core log can be written out, the data section must be scanned | |
1518 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1519 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1520 | * cycle count because there in an implicit assumption that drives will | |
1521 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1522 | * we can't have part of a 512 byte block written and part not written. By | |
1523 | * tagging each block, we will know which blocks are valid when recovering | |
1524 | * after an unclean shutdown. | |
1525 | * | |
1526 | * This routine is single threaded on the iclog. No other thread can be in | |
1527 | * this routine with the same iclog. Changing contents of iclog can there- | |
1528 | * fore be done without grabbing the state machine lock. Updating the global | |
1529 | * log will require grabbing the lock though. | |
1530 | * | |
1531 | * The entire log manager uses a logical block numbering scheme. Only | |
1532 | * log_sync (and then only bwrite()) know about the fact that the log may | |
1533 | * not start with block zero on a given device. The log block start offset | |
1534 | * is added immediately before calling bwrite(). | |
1535 | */ | |
1536 | ||
a8272ce0 | 1537 | STATIC int |
9a8d2fdb MT |
1538 | xlog_sync( |
1539 | struct xlog *log, | |
1540 | struct xlog_in_core *iclog) | |
1da177e4 LT |
1541 | { |
1542 | xfs_caddr_t dptr; /* pointer to byte sized element */ | |
1543 | xfs_buf_t *bp; | |
b53e675d | 1544 | int i; |
1da177e4 LT |
1545 | uint count; /* byte count of bwrite */ |
1546 | uint count_init; /* initial count before roundup */ | |
1547 | int roundoff; /* roundoff to BB or stripe */ | |
1548 | int split = 0; /* split write into two regions */ | |
1549 | int error; | |
62118709 | 1550 | int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb); |
1da177e4 LT |
1551 | |
1552 | XFS_STATS_INC(xs_log_writes); | |
155cc6b7 | 1553 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
1554 | |
1555 | /* Add for LR header */ | |
1556 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
1557 | ||
1558 | /* Round out the log write size */ | |
1559 | if (v2 && log->l_mp->m_sb.sb_logsunit > 1) { | |
1560 | /* we have a v2 stripe unit to use */ | |
1561 | count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init)); | |
1562 | } else { | |
1563 | count = BBTOB(BTOBB(count_init)); | |
1564 | } | |
1565 | roundoff = count - count_init; | |
1566 | ASSERT(roundoff >= 0); | |
1567 | ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && | |
1568 | roundoff < log->l_mp->m_sb.sb_logsunit) | |
1569 | || | |
1570 | (log->l_mp->m_sb.sb_logsunit <= 1 && | |
1571 | roundoff < BBTOB(1))); | |
1572 | ||
1573 | /* move grant heads by roundoff in sync */ | |
28496968 CH |
1574 | xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); |
1575 | xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); | |
1da177e4 LT |
1576 | |
1577 | /* put cycle number in every block */ | |
1578 | xlog_pack_data(log, iclog, roundoff); | |
1579 | ||
1580 | /* real byte length */ | |
1581 | if (v2) { | |
b53e675d CH |
1582 | iclog->ic_header.h_len = |
1583 | cpu_to_be32(iclog->ic_offset + roundoff); | |
1da177e4 | 1584 | } else { |
b53e675d CH |
1585 | iclog->ic_header.h_len = |
1586 | cpu_to_be32(iclog->ic_offset); | |
1da177e4 LT |
1587 | } |
1588 | ||
f5faad79 | 1589 | bp = iclog->ic_bp; |
b53e675d | 1590 | XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn))); |
1da177e4 LT |
1591 | |
1592 | XFS_STATS_ADD(xs_log_blocks, BTOBB(count)); | |
1593 | ||
1594 | /* Do we need to split this write into 2 parts? */ | |
1595 | if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) { | |
1596 | split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp))); | |
1597 | count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)); | |
1598 | iclog->ic_bwritecnt = 2; /* split into 2 writes */ | |
1599 | } else { | |
1600 | iclog->ic_bwritecnt = 1; | |
1601 | } | |
aa0e8833 | 1602 | bp->b_io_length = BTOBB(count); |
adadbeef | 1603 | bp->b_fspriv = iclog; |
f5faad79 | 1604 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 | 1605 | XFS_BUF_ASYNC(bp); |
1d5ae5df | 1606 | bp->b_flags |= XBF_SYNCIO; |
651701d7 | 1607 | |
a27a263b | 1608 | if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) { |
e163cbde CH |
1609 | bp->b_flags |= XBF_FUA; |
1610 | ||
a27a263b | 1611 | /* |
e163cbde CH |
1612 | * Flush the data device before flushing the log to make |
1613 | * sure all meta data written back from the AIL actually made | |
1614 | * it to disk before stamping the new log tail LSN into the | |
1615 | * log buffer. For an external log we need to issue the | |
1616 | * flush explicitly, and unfortunately synchronously here; | |
1617 | * for an internal log we can simply use the block layer | |
1618 | * state machine for preflushes. | |
a27a263b CH |
1619 | */ |
1620 | if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp) | |
1621 | xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp); | |
e163cbde CH |
1622 | else |
1623 | bp->b_flags |= XBF_FLUSH; | |
a27a263b | 1624 | } |
1da177e4 LT |
1625 | |
1626 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1627 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1628 | ||
1629 | xlog_verify_iclog(log, iclog, count, B_TRUE); | |
1630 | ||
1631 | /* account for log which doesn't start at block #0 */ | |
1632 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); | |
1633 | /* | |
1634 | * Don't call xfs_bwrite here. We do log-syncs even when the filesystem | |
1635 | * is shutting down. | |
1636 | */ | |
1637 | XFS_BUF_WRITE(bp); | |
1638 | ||
901796af CH |
1639 | error = xlog_bdstrat(bp); |
1640 | if (error) { | |
1641 | xfs_buf_ioerror_alert(bp, "xlog_sync"); | |
014c2544 | 1642 | return error; |
1da177e4 LT |
1643 | } |
1644 | if (split) { | |
f5faad79 | 1645 | bp = iclog->ic_log->l_xbuf; |
1da177e4 | 1646 | XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */ |
02fe03d9 CS |
1647 | xfs_buf_associate_memory(bp, |
1648 | (char *)&iclog->ic_header + count, split); | |
adadbeef | 1649 | bp->b_fspriv = iclog; |
f5faad79 | 1650 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 | 1651 | XFS_BUF_ASYNC(bp); |
1d5ae5df | 1652 | bp->b_flags |= XBF_SYNCIO; |
f538d4da | 1653 | if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) |
e163cbde | 1654 | bp->b_flags |= XBF_FUA; |
62926044 | 1655 | dptr = bp->b_addr; |
1da177e4 LT |
1656 | /* |
1657 | * Bump the cycle numbers at the start of each block | |
1658 | * since this part of the buffer is at the start of | |
1659 | * a new cycle. Watch out for the header magic number | |
1660 | * case, though. | |
1661 | */ | |
b53e675d | 1662 | for (i = 0; i < split; i += BBSIZE) { |
413d57c9 | 1663 | be32_add_cpu((__be32 *)dptr, 1); |
b53e675d | 1664 | if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM) |
413d57c9 | 1665 | be32_add_cpu((__be32 *)dptr, 1); |
1da177e4 LT |
1666 | dptr += BBSIZE; |
1667 | } | |
1668 | ||
1669 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1670 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1671 | ||
c41564b5 | 1672 | /* account for internal log which doesn't start at block #0 */ |
1da177e4 LT |
1673 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); |
1674 | XFS_BUF_WRITE(bp); | |
901796af CH |
1675 | error = xlog_bdstrat(bp); |
1676 | if (error) { | |
1677 | xfs_buf_ioerror_alert(bp, "xlog_sync (split)"); | |
014c2544 | 1678 | return error; |
1da177e4 LT |
1679 | } |
1680 | } | |
014c2544 | 1681 | return 0; |
1da177e4 LT |
1682 | } /* xlog_sync */ |
1683 | ||
1684 | ||
1685 | /* | |
c41564b5 | 1686 | * Deallocate a log structure |
1da177e4 | 1687 | */ |
a8272ce0 | 1688 | STATIC void |
9a8d2fdb MT |
1689 | xlog_dealloc_log( |
1690 | struct xlog *log) | |
1da177e4 LT |
1691 | { |
1692 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1693 | int i; |
1694 | ||
71e330b5 DC |
1695 | xlog_cil_destroy(log); |
1696 | ||
44396476 DC |
1697 | /* |
1698 | * always need to ensure that the extra buffer does not point to memory | |
1699 | * owned by another log buffer before we free it. | |
1700 | */ | |
e70b73f8 | 1701 | xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size)); |
44396476 DC |
1702 | xfs_buf_free(log->l_xbuf); |
1703 | ||
1da177e4 LT |
1704 | iclog = log->l_iclog; |
1705 | for (i=0; i<log->l_iclog_bufs; i++) { | |
1da177e4 | 1706 | xfs_buf_free(iclog->ic_bp); |
1da177e4 | 1707 | next_iclog = iclog->ic_next; |
f0e2d93c | 1708 | kmem_free(iclog); |
1da177e4 LT |
1709 | iclog = next_iclog; |
1710 | } | |
1da177e4 | 1711 | spinlock_destroy(&log->l_icloglock); |
1da177e4 | 1712 | |
1da177e4 | 1713 | log->l_mp->m_log = NULL; |
f0e2d93c | 1714 | kmem_free(log); |
c41564b5 | 1715 | } /* xlog_dealloc_log */ |
1da177e4 LT |
1716 | |
1717 | /* | |
1718 | * Update counters atomically now that memcpy is done. | |
1719 | */ | |
1720 | /* ARGSUSED */ | |
1721 | static inline void | |
9a8d2fdb MT |
1722 | xlog_state_finish_copy( |
1723 | struct xlog *log, | |
1724 | struct xlog_in_core *iclog, | |
1725 | int record_cnt, | |
1726 | int copy_bytes) | |
1da177e4 | 1727 | { |
b22cd72c | 1728 | spin_lock(&log->l_icloglock); |
1da177e4 | 1729 | |
413d57c9 | 1730 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 LT |
1731 | iclog->ic_offset += copy_bytes; |
1732 | ||
b22cd72c | 1733 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
1734 | } /* xlog_state_finish_copy */ |
1735 | ||
1736 | ||
1737 | ||
1738 | ||
7e9c6396 TS |
1739 | /* |
1740 | * print out info relating to regions written which consume | |
1741 | * the reservation | |
1742 | */ | |
71e330b5 DC |
1743 | void |
1744 | xlog_print_tic_res( | |
1745 | struct xfs_mount *mp, | |
1746 | struct xlog_ticket *ticket) | |
7e9c6396 TS |
1747 | { |
1748 | uint i; | |
1749 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
1750 | ||
1751 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
1752 | static char *res_type_str[XLOG_REG_TYPE_MAX] = { | |
1753 | "bformat", | |
1754 | "bchunk", | |
1755 | "efi_format", | |
1756 | "efd_format", | |
1757 | "iformat", | |
1758 | "icore", | |
1759 | "iext", | |
1760 | "ibroot", | |
1761 | "ilocal", | |
1762 | "iattr_ext", | |
1763 | "iattr_broot", | |
1764 | "iattr_local", | |
1765 | "qformat", | |
1766 | "dquot", | |
1767 | "quotaoff", | |
1768 | "LR header", | |
1769 | "unmount", | |
1770 | "commit", | |
1771 | "trans header" | |
1772 | }; | |
1773 | static char *trans_type_str[XFS_TRANS_TYPE_MAX] = { | |
1774 | "SETATTR_NOT_SIZE", | |
1775 | "SETATTR_SIZE", | |
1776 | "INACTIVE", | |
1777 | "CREATE", | |
1778 | "CREATE_TRUNC", | |
1779 | "TRUNCATE_FILE", | |
1780 | "REMOVE", | |
1781 | "LINK", | |
1782 | "RENAME", | |
1783 | "MKDIR", | |
1784 | "RMDIR", | |
1785 | "SYMLINK", | |
1786 | "SET_DMATTRS", | |
1787 | "GROWFS", | |
1788 | "STRAT_WRITE", | |
1789 | "DIOSTRAT", | |
1790 | "WRITE_SYNC", | |
1791 | "WRITEID", | |
1792 | "ADDAFORK", | |
1793 | "ATTRINVAL", | |
1794 | "ATRUNCATE", | |
1795 | "ATTR_SET", | |
1796 | "ATTR_RM", | |
1797 | "ATTR_FLAG", | |
1798 | "CLEAR_AGI_BUCKET", | |
1799 | "QM_SBCHANGE", | |
1800 | "DUMMY1", | |
1801 | "DUMMY2", | |
1802 | "QM_QUOTAOFF", | |
1803 | "QM_DQALLOC", | |
1804 | "QM_SETQLIM", | |
1805 | "QM_DQCLUSTER", | |
1806 | "QM_QINOCREATE", | |
1807 | "QM_QUOTAOFF_END", | |
1808 | "SB_UNIT", | |
1809 | "FSYNC_TS", | |
1810 | "GROWFSRT_ALLOC", | |
1811 | "GROWFSRT_ZERO", | |
1812 | "GROWFSRT_FREE", | |
1813 | "SWAPEXT" | |
1814 | }; | |
1815 | ||
a0fa2b67 | 1816 | xfs_warn(mp, |
93b8a585 | 1817 | "xlog_write: reservation summary:\n" |
a0fa2b67 DC |
1818 | " trans type = %s (%u)\n" |
1819 | " unit res = %d bytes\n" | |
1820 | " current res = %d bytes\n" | |
1821 | " total reg = %u bytes (o/flow = %u bytes)\n" | |
1822 | " ophdrs = %u (ophdr space = %u bytes)\n" | |
1823 | " ophdr + reg = %u bytes\n" | |
1824 | " num regions = %u\n", | |
1825 | ((ticket->t_trans_type <= 0 || | |
1826 | ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ? | |
1827 | "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]), | |
1828 | ticket->t_trans_type, | |
1829 | ticket->t_unit_res, | |
1830 | ticket->t_curr_res, | |
1831 | ticket->t_res_arr_sum, ticket->t_res_o_flow, | |
1832 | ticket->t_res_num_ophdrs, ophdr_spc, | |
1833 | ticket->t_res_arr_sum + | |
1834 | ticket->t_res_o_flow + ophdr_spc, | |
1835 | ticket->t_res_num); | |
7e9c6396 TS |
1836 | |
1837 | for (i = 0; i < ticket->t_res_num; i++) { | |
a0fa2b67 DC |
1838 | uint r_type = ticket->t_res_arr[i].r_type; |
1839 | xfs_warn(mp, "region[%u]: %s - %u bytes\n", i, | |
7e9c6396 TS |
1840 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? |
1841 | "bad-rtype" : res_type_str[r_type-1]), | |
1842 | ticket->t_res_arr[i].r_len); | |
1843 | } | |
169a7b07 | 1844 | |
a0fa2b67 | 1845 | xfs_alert_tag(mp, XFS_PTAG_LOGRES, |
93b8a585 | 1846 | "xlog_write: reservation ran out. Need to up reservation"); |
169a7b07 | 1847 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
7e9c6396 | 1848 | } |
7e9c6396 | 1849 | |
b5203cd0 DC |
1850 | /* |
1851 | * Calculate the potential space needed by the log vector. Each region gets | |
1852 | * its own xlog_op_header_t and may need to be double word aligned. | |
1853 | */ | |
1854 | static int | |
1855 | xlog_write_calc_vec_length( | |
1856 | struct xlog_ticket *ticket, | |
55b66332 | 1857 | struct xfs_log_vec *log_vector) |
b5203cd0 | 1858 | { |
55b66332 | 1859 | struct xfs_log_vec *lv; |
b5203cd0 DC |
1860 | int headers = 0; |
1861 | int len = 0; | |
1862 | int i; | |
1863 | ||
1864 | /* acct for start rec of xact */ | |
1865 | if (ticket->t_flags & XLOG_TIC_INITED) | |
1866 | headers++; | |
1867 | ||
55b66332 DC |
1868 | for (lv = log_vector; lv; lv = lv->lv_next) { |
1869 | headers += lv->lv_niovecs; | |
1870 | ||
1871 | for (i = 0; i < lv->lv_niovecs; i++) { | |
1872 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 1873 | |
55b66332 DC |
1874 | len += vecp->i_len; |
1875 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
1876 | } | |
b5203cd0 DC |
1877 | } |
1878 | ||
1879 | ticket->t_res_num_ophdrs += headers; | |
1880 | len += headers * sizeof(struct xlog_op_header); | |
1881 | ||
1882 | return len; | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * If first write for transaction, insert start record We can't be trying to | |
1887 | * commit if we are inited. We can't have any "partial_copy" if we are inited. | |
1888 | */ | |
1889 | static int | |
1890 | xlog_write_start_rec( | |
e6b1f273 | 1891 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
1892 | struct xlog_ticket *ticket) |
1893 | { | |
b5203cd0 DC |
1894 | if (!(ticket->t_flags & XLOG_TIC_INITED)) |
1895 | return 0; | |
1896 | ||
1897 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); | |
1898 | ophdr->oh_clientid = ticket->t_clientid; | |
1899 | ophdr->oh_len = 0; | |
1900 | ophdr->oh_flags = XLOG_START_TRANS; | |
1901 | ophdr->oh_res2 = 0; | |
1902 | ||
1903 | ticket->t_flags &= ~XLOG_TIC_INITED; | |
1904 | ||
1905 | return sizeof(struct xlog_op_header); | |
1906 | } | |
1907 | ||
1908 | static xlog_op_header_t * | |
1909 | xlog_write_setup_ophdr( | |
ad223e60 | 1910 | struct xlog *log, |
e6b1f273 | 1911 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
1912 | struct xlog_ticket *ticket, |
1913 | uint flags) | |
1914 | { | |
b5203cd0 DC |
1915 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
1916 | ophdr->oh_clientid = ticket->t_clientid; | |
1917 | ophdr->oh_res2 = 0; | |
1918 | ||
1919 | /* are we copying a commit or unmount record? */ | |
1920 | ophdr->oh_flags = flags; | |
1921 | ||
1922 | /* | |
1923 | * We've seen logs corrupted with bad transaction client ids. This | |
1924 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
1925 | * and shut down the filesystem. | |
1926 | */ | |
1927 | switch (ophdr->oh_clientid) { | |
1928 | case XFS_TRANSACTION: | |
1929 | case XFS_VOLUME: | |
1930 | case XFS_LOG: | |
1931 | break; | |
1932 | default: | |
a0fa2b67 | 1933 | xfs_warn(log->l_mp, |
b5203cd0 DC |
1934 | "Bad XFS transaction clientid 0x%x in ticket 0x%p", |
1935 | ophdr->oh_clientid, ticket); | |
1936 | return NULL; | |
1937 | } | |
1938 | ||
1939 | return ophdr; | |
1940 | } | |
1941 | ||
1942 | /* | |
1943 | * Set up the parameters of the region copy into the log. This has | |
1944 | * to handle region write split across multiple log buffers - this | |
1945 | * state is kept external to this function so that this code can | |
1946 | * can be written in an obvious, self documenting manner. | |
1947 | */ | |
1948 | static int | |
1949 | xlog_write_setup_copy( | |
1950 | struct xlog_ticket *ticket, | |
1951 | struct xlog_op_header *ophdr, | |
1952 | int space_available, | |
1953 | int space_required, | |
1954 | int *copy_off, | |
1955 | int *copy_len, | |
1956 | int *last_was_partial_copy, | |
1957 | int *bytes_consumed) | |
1958 | { | |
1959 | int still_to_copy; | |
1960 | ||
1961 | still_to_copy = space_required - *bytes_consumed; | |
1962 | *copy_off = *bytes_consumed; | |
1963 | ||
1964 | if (still_to_copy <= space_available) { | |
1965 | /* write of region completes here */ | |
1966 | *copy_len = still_to_copy; | |
1967 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
1968 | if (*last_was_partial_copy) | |
1969 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
1970 | *last_was_partial_copy = 0; | |
1971 | *bytes_consumed = 0; | |
1972 | return 0; | |
1973 | } | |
1974 | ||
1975 | /* partial write of region, needs extra log op header reservation */ | |
1976 | *copy_len = space_available; | |
1977 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
1978 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
1979 | if (*last_was_partial_copy) | |
1980 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
1981 | *bytes_consumed += *copy_len; | |
1982 | (*last_was_partial_copy)++; | |
1983 | ||
1984 | /* account for new log op header */ | |
1985 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
1986 | ticket->t_res_num_ophdrs++; | |
1987 | ||
1988 | return sizeof(struct xlog_op_header); | |
1989 | } | |
1990 | ||
1991 | static int | |
1992 | xlog_write_copy_finish( | |
ad223e60 | 1993 | struct xlog *log, |
b5203cd0 DC |
1994 | struct xlog_in_core *iclog, |
1995 | uint flags, | |
1996 | int *record_cnt, | |
1997 | int *data_cnt, | |
1998 | int *partial_copy, | |
1999 | int *partial_copy_len, | |
2000 | int log_offset, | |
2001 | struct xlog_in_core **commit_iclog) | |
2002 | { | |
2003 | if (*partial_copy) { | |
2004 | /* | |
2005 | * This iclog has already been marked WANT_SYNC by | |
2006 | * xlog_state_get_iclog_space. | |
2007 | */ | |
2008 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2009 | *record_cnt = 0; | |
2010 | *data_cnt = 0; | |
2011 | return xlog_state_release_iclog(log, iclog); | |
2012 | } | |
2013 | ||
2014 | *partial_copy = 0; | |
2015 | *partial_copy_len = 0; | |
2016 | ||
2017 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
2018 | /* no more space in this iclog - push it. */ | |
2019 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
2020 | *record_cnt = 0; | |
2021 | *data_cnt = 0; | |
2022 | ||
2023 | spin_lock(&log->l_icloglock); | |
2024 | xlog_state_want_sync(log, iclog); | |
2025 | spin_unlock(&log->l_icloglock); | |
2026 | ||
2027 | if (!commit_iclog) | |
2028 | return xlog_state_release_iclog(log, iclog); | |
2029 | ASSERT(flags & XLOG_COMMIT_TRANS); | |
2030 | *commit_iclog = iclog; | |
2031 | } | |
2032 | ||
2033 | return 0; | |
2034 | } | |
2035 | ||
1da177e4 LT |
2036 | /* |
2037 | * Write some region out to in-core log | |
2038 | * | |
2039 | * This will be called when writing externally provided regions or when | |
2040 | * writing out a commit record for a given transaction. | |
2041 | * | |
2042 | * General algorithm: | |
2043 | * 1. Find total length of this write. This may include adding to the | |
2044 | * lengths passed in. | |
2045 | * 2. Check whether we violate the tickets reservation. | |
2046 | * 3. While writing to this iclog | |
2047 | * A. Reserve as much space in this iclog as can get | |
2048 | * B. If this is first write, save away start lsn | |
2049 | * C. While writing this region: | |
2050 | * 1. If first write of transaction, write start record | |
2051 | * 2. Write log operation header (header per region) | |
2052 | * 3. Find out if we can fit entire region into this iclog | |
2053 | * 4. Potentially, verify destination memcpy ptr | |
2054 | * 5. Memcpy (partial) region | |
2055 | * 6. If partial copy, release iclog; otherwise, continue | |
2056 | * copying more regions into current iclog | |
2057 | * 4. Mark want sync bit (in simulation mode) | |
2058 | * 5. Release iclog for potential flush to on-disk log. | |
2059 | * | |
2060 | * ERRORS: | |
2061 | * 1. Panic if reservation is overrun. This should never happen since | |
2062 | * reservation amounts are generated internal to the filesystem. | |
2063 | * NOTES: | |
2064 | * 1. Tickets are single threaded data structures. | |
2065 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
2066 | * syncing routine. When a single log_write region needs to span | |
2067 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
2068 | * on all log operation writes which don't contain the end of the | |
2069 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
2070 | * operation which contains the end of the continued log_write region. | |
2071 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
2072 | * we don't really know exactly how much space will be used. As a result, | |
2073 | * we don't update ic_offset until the end when we know exactly how many | |
2074 | * bytes have been written out. | |
2075 | */ | |
71e330b5 | 2076 | int |
35a8a72f | 2077 | xlog_write( |
ad223e60 | 2078 | struct xlog *log, |
55b66332 | 2079 | struct xfs_log_vec *log_vector, |
35a8a72f CH |
2080 | struct xlog_ticket *ticket, |
2081 | xfs_lsn_t *start_lsn, | |
2082 | struct xlog_in_core **commit_iclog, | |
2083 | uint flags) | |
1da177e4 | 2084 | { |
99428ad0 | 2085 | struct xlog_in_core *iclog = NULL; |
55b66332 DC |
2086 | struct xfs_log_iovec *vecp; |
2087 | struct xfs_log_vec *lv; | |
99428ad0 CH |
2088 | int len; |
2089 | int index; | |
2090 | int partial_copy = 0; | |
2091 | int partial_copy_len = 0; | |
2092 | int contwr = 0; | |
2093 | int record_cnt = 0; | |
2094 | int data_cnt = 0; | |
2095 | int error; | |
2096 | ||
2097 | *start_lsn = 0; | |
2098 | ||
55b66332 | 2099 | len = xlog_write_calc_vec_length(ticket, log_vector); |
71e330b5 | 2100 | |
93b8a585 CH |
2101 | /* |
2102 | * Region headers and bytes are already accounted for. | |
2103 | * We only need to take into account start records and | |
2104 | * split regions in this function. | |
2105 | */ | |
2106 | if (ticket->t_flags & XLOG_TIC_INITED) | |
2107 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
2108 | ||
2109 | /* | |
2110 | * Commit record headers need to be accounted for. These | |
2111 | * come in as separate writes so are easy to detect. | |
2112 | */ | |
2113 | if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS)) | |
2114 | ticket->t_curr_res -= sizeof(xlog_op_header_t); | |
71e330b5 DC |
2115 | |
2116 | if (ticket->t_curr_res < 0) | |
55b66332 | 2117 | xlog_print_tic_res(log->l_mp, ticket); |
1da177e4 | 2118 | |
55b66332 DC |
2119 | index = 0; |
2120 | lv = log_vector; | |
2121 | vecp = lv->lv_iovecp; | |
2122 | while (lv && index < lv->lv_niovecs) { | |
e6b1f273 | 2123 | void *ptr; |
99428ad0 | 2124 | int log_offset; |
1da177e4 | 2125 | |
99428ad0 CH |
2126 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
2127 | &contwr, &log_offset); | |
2128 | if (error) | |
2129 | return error; | |
1da177e4 | 2130 | |
99428ad0 | 2131 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 2132 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 2133 | |
99428ad0 CH |
2134 | /* start_lsn is the first lsn written to. That's all we need. */ |
2135 | if (!*start_lsn) | |
2136 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
b5203cd0 | 2137 | |
99428ad0 CH |
2138 | /* |
2139 | * This loop writes out as many regions as can fit in the amount | |
2140 | * of space which was allocated by xlog_state_get_iclog_space(). | |
2141 | */ | |
55b66332 DC |
2142 | while (lv && index < lv->lv_niovecs) { |
2143 | struct xfs_log_iovec *reg = &vecp[index]; | |
99428ad0 CH |
2144 | struct xlog_op_header *ophdr; |
2145 | int start_rec_copy; | |
2146 | int copy_len; | |
2147 | int copy_off; | |
2148 | ||
55b66332 | 2149 | ASSERT(reg->i_len % sizeof(__int32_t) == 0); |
e6b1f273 | 2150 | ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0); |
99428ad0 CH |
2151 | |
2152 | start_rec_copy = xlog_write_start_rec(ptr, ticket); | |
2153 | if (start_rec_copy) { | |
2154 | record_cnt++; | |
e6b1f273 | 2155 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2156 | start_rec_copy); |
2157 | } | |
b5203cd0 | 2158 | |
99428ad0 CH |
2159 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags); |
2160 | if (!ophdr) | |
2161 | return XFS_ERROR(EIO); | |
2162 | ||
e6b1f273 | 2163 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
2164 | sizeof(struct xlog_op_header)); |
2165 | ||
2166 | len += xlog_write_setup_copy(ticket, ophdr, | |
2167 | iclog->ic_size-log_offset, | |
55b66332 | 2168 | reg->i_len, |
99428ad0 CH |
2169 | ©_off, ©_len, |
2170 | &partial_copy, | |
2171 | &partial_copy_len); | |
2172 | xlog_verify_dest_ptr(log, ptr); | |
2173 | ||
2174 | /* copy region */ | |
2175 | ASSERT(copy_len >= 0); | |
e6b1f273 CH |
2176 | memcpy(ptr, reg->i_addr + copy_off, copy_len); |
2177 | xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len); | |
99428ad0 CH |
2178 | |
2179 | copy_len += start_rec_copy + sizeof(xlog_op_header_t); | |
2180 | record_cnt++; | |
2181 | data_cnt += contwr ? copy_len : 0; | |
2182 | ||
2183 | error = xlog_write_copy_finish(log, iclog, flags, | |
2184 | &record_cnt, &data_cnt, | |
2185 | &partial_copy, | |
2186 | &partial_copy_len, | |
2187 | log_offset, | |
2188 | commit_iclog); | |
2189 | if (error) | |
2190 | return error; | |
2191 | ||
2192 | /* | |
2193 | * if we had a partial copy, we need to get more iclog | |
2194 | * space but we don't want to increment the region | |
2195 | * index because there is still more is this region to | |
2196 | * write. | |
2197 | * | |
2198 | * If we completed writing this region, and we flushed | |
2199 | * the iclog (indicated by resetting of the record | |
2200 | * count), then we also need to get more log space. If | |
2201 | * this was the last record, though, we are done and | |
2202 | * can just return. | |
2203 | */ | |
2204 | if (partial_copy) | |
2205 | break; | |
2206 | ||
55b66332 DC |
2207 | if (++index == lv->lv_niovecs) { |
2208 | lv = lv->lv_next; | |
2209 | index = 0; | |
2210 | if (lv) | |
2211 | vecp = lv->lv_iovecp; | |
2212 | } | |
99428ad0 | 2213 | if (record_cnt == 0) { |
55b66332 | 2214 | if (!lv) |
99428ad0 CH |
2215 | return 0; |
2216 | break; | |
2217 | } | |
2218 | } | |
2219 | } | |
2220 | ||
2221 | ASSERT(len == 0); | |
2222 | ||
2223 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); | |
2224 | if (!commit_iclog) | |
2225 | return xlog_state_release_iclog(log, iclog); | |
1da177e4 | 2226 | |
1da177e4 LT |
2227 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2228 | *commit_iclog = iclog; | |
2229 | return 0; | |
99428ad0 | 2230 | } |
1da177e4 LT |
2231 | |
2232 | ||
2233 | /***************************************************************************** | |
2234 | * | |
2235 | * State Machine functions | |
2236 | * | |
2237 | ***************************************************************************** | |
2238 | */ | |
2239 | ||
2240 | /* Clean iclogs starting from the head. This ordering must be | |
2241 | * maintained, so an iclog doesn't become ACTIVE beyond one that | |
2242 | * is SYNCING. This is also required to maintain the notion that we use | |
12017faf | 2243 | * a ordered wait queue to hold off would be writers to the log when every |
1da177e4 LT |
2244 | * iclog is trying to sync to disk. |
2245 | * | |
2246 | * State Change: DIRTY -> ACTIVE | |
2247 | */ | |
ba0f32d4 | 2248 | STATIC void |
9a8d2fdb MT |
2249 | xlog_state_clean_log( |
2250 | struct xlog *log) | |
1da177e4 LT |
2251 | { |
2252 | xlog_in_core_t *iclog; | |
2253 | int changed = 0; | |
2254 | ||
2255 | iclog = log->l_iclog; | |
2256 | do { | |
2257 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
2258 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2259 | iclog->ic_offset = 0; | |
114d23aa | 2260 | ASSERT(iclog->ic_callback == NULL); |
1da177e4 LT |
2261 | /* |
2262 | * If the number of ops in this iclog indicate it just | |
2263 | * contains the dummy transaction, we can | |
2264 | * change state into IDLE (the second time around). | |
2265 | * Otherwise we should change the state into | |
2266 | * NEED a dummy. | |
2267 | * We don't need to cover the dummy. | |
2268 | */ | |
2269 | if (!changed && | |
b53e675d CH |
2270 | (be32_to_cpu(iclog->ic_header.h_num_logops) == |
2271 | XLOG_COVER_OPS)) { | |
1da177e4 LT |
2272 | changed = 1; |
2273 | } else { | |
2274 | /* | |
2275 | * We have two dirty iclogs so start over | |
2276 | * This could also be num of ops indicates | |
2277 | * this is not the dummy going out. | |
2278 | */ | |
2279 | changed = 2; | |
2280 | } | |
2281 | iclog->ic_header.h_num_logops = 0; | |
2282 | memset(iclog->ic_header.h_cycle_data, 0, | |
2283 | sizeof(iclog->ic_header.h_cycle_data)); | |
2284 | iclog->ic_header.h_lsn = 0; | |
2285 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) | |
2286 | /* do nothing */; | |
2287 | else | |
2288 | break; /* stop cleaning */ | |
2289 | iclog = iclog->ic_next; | |
2290 | } while (iclog != log->l_iclog); | |
2291 | ||
2292 | /* log is locked when we are called */ | |
2293 | /* | |
2294 | * Change state for the dummy log recording. | |
2295 | * We usually go to NEED. But we go to NEED2 if the changed indicates | |
2296 | * we are done writing the dummy record. | |
2297 | * If we are done with the second dummy recored (DONE2), then | |
2298 | * we go to IDLE. | |
2299 | */ | |
2300 | if (changed) { | |
2301 | switch (log->l_covered_state) { | |
2302 | case XLOG_STATE_COVER_IDLE: | |
2303 | case XLOG_STATE_COVER_NEED: | |
2304 | case XLOG_STATE_COVER_NEED2: | |
2305 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2306 | break; | |
2307 | ||
2308 | case XLOG_STATE_COVER_DONE: | |
2309 | if (changed == 1) | |
2310 | log->l_covered_state = XLOG_STATE_COVER_NEED2; | |
2311 | else | |
2312 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2313 | break; | |
2314 | ||
2315 | case XLOG_STATE_COVER_DONE2: | |
2316 | if (changed == 1) | |
2317 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
2318 | else | |
2319 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2320 | break; | |
2321 | ||
2322 | default: | |
2323 | ASSERT(0); | |
2324 | } | |
2325 | } | |
2326 | } /* xlog_state_clean_log */ | |
2327 | ||
2328 | STATIC xfs_lsn_t | |
2329 | xlog_get_lowest_lsn( | |
9a8d2fdb | 2330 | struct xlog *log) |
1da177e4 LT |
2331 | { |
2332 | xlog_in_core_t *lsn_log; | |
2333 | xfs_lsn_t lowest_lsn, lsn; | |
2334 | ||
2335 | lsn_log = log->l_iclog; | |
2336 | lowest_lsn = 0; | |
2337 | do { | |
2338 | if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) { | |
b53e675d | 2339 | lsn = be64_to_cpu(lsn_log->ic_header.h_lsn); |
1da177e4 LT |
2340 | if ((lsn && !lowest_lsn) || |
2341 | (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) { | |
2342 | lowest_lsn = lsn; | |
2343 | } | |
2344 | } | |
2345 | lsn_log = lsn_log->ic_next; | |
2346 | } while (lsn_log != log->l_iclog); | |
014c2544 | 2347 | return lowest_lsn; |
1da177e4 LT |
2348 | } |
2349 | ||
2350 | ||
2351 | STATIC void | |
2352 | xlog_state_do_callback( | |
9a8d2fdb MT |
2353 | struct xlog *log, |
2354 | int aborted, | |
2355 | struct xlog_in_core *ciclog) | |
1da177e4 LT |
2356 | { |
2357 | xlog_in_core_t *iclog; | |
2358 | xlog_in_core_t *first_iclog; /* used to know when we've | |
2359 | * processed all iclogs once */ | |
2360 | xfs_log_callback_t *cb, *cb_next; | |
2361 | int flushcnt = 0; | |
2362 | xfs_lsn_t lowest_lsn; | |
2363 | int ioerrors; /* counter: iclogs with errors */ | |
2364 | int loopdidcallbacks; /* flag: inner loop did callbacks*/ | |
2365 | int funcdidcallbacks; /* flag: function did callbacks */ | |
2366 | int repeats; /* for issuing console warnings if | |
2367 | * looping too many times */ | |
d748c623 | 2368 | int wake = 0; |
1da177e4 | 2369 | |
b22cd72c | 2370 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2371 | first_iclog = iclog = log->l_iclog; |
2372 | ioerrors = 0; | |
2373 | funcdidcallbacks = 0; | |
2374 | repeats = 0; | |
2375 | ||
2376 | do { | |
2377 | /* | |
2378 | * Scan all iclogs starting with the one pointed to by the | |
2379 | * log. Reset this starting point each time the log is | |
2380 | * unlocked (during callbacks). | |
2381 | * | |
2382 | * Keep looping through iclogs until one full pass is made | |
2383 | * without running any callbacks. | |
2384 | */ | |
2385 | first_iclog = log->l_iclog; | |
2386 | iclog = log->l_iclog; | |
2387 | loopdidcallbacks = 0; | |
2388 | repeats++; | |
2389 | ||
2390 | do { | |
2391 | ||
2392 | /* skip all iclogs in the ACTIVE & DIRTY states */ | |
2393 | if (iclog->ic_state & | |
2394 | (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) { | |
2395 | iclog = iclog->ic_next; | |
2396 | continue; | |
2397 | } | |
2398 | ||
2399 | /* | |
2400 | * Between marking a filesystem SHUTDOWN and stopping | |
2401 | * the log, we do flush all iclogs to disk (if there | |
2402 | * wasn't a log I/O error). So, we do want things to | |
2403 | * go smoothly in case of just a SHUTDOWN w/o a | |
2404 | * LOG_IO_ERROR. | |
2405 | */ | |
2406 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
2407 | /* | |
2408 | * Can only perform callbacks in order. Since | |
2409 | * this iclog is not in the DONE_SYNC/ | |
2410 | * DO_CALLBACK state, we skip the rest and | |
2411 | * just try to clean up. If we set our iclog | |
2412 | * to DO_CALLBACK, we will not process it when | |
2413 | * we retry since a previous iclog is in the | |
2414 | * CALLBACK and the state cannot change since | |
b22cd72c | 2415 | * we are holding the l_icloglock. |
1da177e4 LT |
2416 | */ |
2417 | if (!(iclog->ic_state & | |
2418 | (XLOG_STATE_DONE_SYNC | | |
2419 | XLOG_STATE_DO_CALLBACK))) { | |
2420 | if (ciclog && (ciclog->ic_state == | |
2421 | XLOG_STATE_DONE_SYNC)) { | |
2422 | ciclog->ic_state = XLOG_STATE_DO_CALLBACK; | |
2423 | } | |
2424 | break; | |
2425 | } | |
2426 | /* | |
2427 | * We now have an iclog that is in either the | |
2428 | * DO_CALLBACK or DONE_SYNC states. The other | |
2429 | * states (WANT_SYNC, SYNCING, or CALLBACK were | |
2430 | * caught by the above if and are going to | |
2431 | * clean (i.e. we aren't doing their callbacks) | |
2432 | * see the above if. | |
2433 | */ | |
2434 | ||
2435 | /* | |
2436 | * We will do one more check here to see if we | |
2437 | * have chased our tail around. | |
2438 | */ | |
2439 | ||
2440 | lowest_lsn = xlog_get_lowest_lsn(log); | |
b53e675d CH |
2441 | if (lowest_lsn && |
2442 | XFS_LSN_CMP(lowest_lsn, | |
84f3c683 | 2443 | be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { |
1da177e4 LT |
2444 | iclog = iclog->ic_next; |
2445 | continue; /* Leave this iclog for | |
2446 | * another thread */ | |
2447 | } | |
2448 | ||
2449 | iclog->ic_state = XLOG_STATE_CALLBACK; | |
2450 | ||
1da177e4 | 2451 | |
84f3c683 DC |
2452 | /* |
2453 | * update the last_sync_lsn before we drop the | |
2454 | * icloglock to ensure we are the only one that | |
2455 | * can update it. | |
1da177e4 | 2456 | */ |
84f3c683 DC |
2457 | ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), |
2458 | be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); | |
2459 | atomic64_set(&log->l_last_sync_lsn, | |
2460 | be64_to_cpu(iclog->ic_header.h_lsn)); | |
1da177e4 | 2461 | |
84f3c683 | 2462 | } else |
1da177e4 | 2463 | ioerrors++; |
84f3c683 DC |
2464 | |
2465 | spin_unlock(&log->l_icloglock); | |
1da177e4 | 2466 | |
114d23aa DC |
2467 | /* |
2468 | * Keep processing entries in the callback list until | |
2469 | * we come around and it is empty. We need to | |
2470 | * atomically see that the list is empty and change the | |
2471 | * state to DIRTY so that we don't miss any more | |
2472 | * callbacks being added. | |
2473 | */ | |
2474 | spin_lock(&iclog->ic_callback_lock); | |
2475 | cb = iclog->ic_callback; | |
4b80916b | 2476 | while (cb) { |
1da177e4 LT |
2477 | iclog->ic_callback_tail = &(iclog->ic_callback); |
2478 | iclog->ic_callback = NULL; | |
114d23aa | 2479 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2480 | |
2481 | /* perform callbacks in the order given */ | |
4b80916b | 2482 | for (; cb; cb = cb_next) { |
1da177e4 LT |
2483 | cb_next = cb->cb_next; |
2484 | cb->cb_func(cb->cb_arg, aborted); | |
2485 | } | |
114d23aa | 2486 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
2487 | cb = iclog->ic_callback; |
2488 | } | |
2489 | ||
2490 | loopdidcallbacks++; | |
2491 | funcdidcallbacks++; | |
2492 | ||
114d23aa | 2493 | spin_lock(&log->l_icloglock); |
4b80916b | 2494 | ASSERT(iclog->ic_callback == NULL); |
114d23aa | 2495 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2496 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) |
2497 | iclog->ic_state = XLOG_STATE_DIRTY; | |
2498 | ||
2499 | /* | |
2500 | * Transition from DIRTY to ACTIVE if applicable. | |
2501 | * NOP if STATE_IOERROR. | |
2502 | */ | |
2503 | xlog_state_clean_log(log); | |
2504 | ||
2505 | /* wake up threads waiting in xfs_log_force() */ | |
eb40a875 | 2506 | wake_up_all(&iclog->ic_force_wait); |
1da177e4 LT |
2507 | |
2508 | iclog = iclog->ic_next; | |
2509 | } while (first_iclog != iclog); | |
a3c6685e NS |
2510 | |
2511 | if (repeats > 5000) { | |
2512 | flushcnt += repeats; | |
2513 | repeats = 0; | |
a0fa2b67 | 2514 | xfs_warn(log->l_mp, |
a3c6685e | 2515 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2516 | __func__, flushcnt); |
1da177e4 LT |
2517 | } |
2518 | } while (!ioerrors && loopdidcallbacks); | |
2519 | ||
2520 | /* | |
2521 | * make one last gasp attempt to see if iclogs are being left in | |
2522 | * limbo.. | |
2523 | */ | |
2524 | #ifdef DEBUG | |
2525 | if (funcdidcallbacks) { | |
2526 | first_iclog = iclog = log->l_iclog; | |
2527 | do { | |
2528 | ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK); | |
2529 | /* | |
2530 | * Terminate the loop if iclogs are found in states | |
2531 | * which will cause other threads to clean up iclogs. | |
2532 | * | |
2533 | * SYNCING - i/o completion will go through logs | |
2534 | * DONE_SYNC - interrupt thread should be waiting for | |
b22cd72c | 2535 | * l_icloglock |
1da177e4 LT |
2536 | * IOERROR - give up hope all ye who enter here |
2537 | */ | |
2538 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
2539 | iclog->ic_state == XLOG_STATE_SYNCING || | |
2540 | iclog->ic_state == XLOG_STATE_DONE_SYNC || | |
2541 | iclog->ic_state == XLOG_STATE_IOERROR ) | |
2542 | break; | |
2543 | iclog = iclog->ic_next; | |
2544 | } while (first_iclog != iclog); | |
2545 | } | |
2546 | #endif | |
2547 | ||
d748c623 MW |
2548 | if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) |
2549 | wake = 1; | |
b22cd72c | 2550 | spin_unlock(&log->l_icloglock); |
d748c623 MW |
2551 | |
2552 | if (wake) | |
eb40a875 | 2553 | wake_up_all(&log->l_flush_wait); |
d748c623 | 2554 | } |
1da177e4 LT |
2555 | |
2556 | ||
2557 | /* | |
2558 | * Finish transitioning this iclog to the dirty state. | |
2559 | * | |
2560 | * Make sure that we completely execute this routine only when this is | |
2561 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2562 | * when we reach the end of the physical log, get turned into 2 separate | |
2563 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2564 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2565 | * the second completion goes through. | |
2566 | * | |
2567 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2568 | * global state machine log lock. |
1da177e4 | 2569 | */ |
a8272ce0 | 2570 | STATIC void |
1da177e4 LT |
2571 | xlog_state_done_syncing( |
2572 | xlog_in_core_t *iclog, | |
2573 | int aborted) | |
2574 | { | |
9a8d2fdb | 2575 | struct xlog *log = iclog->ic_log; |
1da177e4 | 2576 | |
b22cd72c | 2577 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2578 | |
2579 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING || | |
2580 | iclog->ic_state == XLOG_STATE_IOERROR); | |
155cc6b7 | 2581 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
2582 | ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2); |
2583 | ||
2584 | ||
2585 | /* | |
2586 | * If we got an error, either on the first buffer, or in the case of | |
2587 | * split log writes, on the second, we mark ALL iclogs STATE_IOERROR, | |
2588 | * and none should ever be attempted to be written to disk | |
2589 | * again. | |
2590 | */ | |
2591 | if (iclog->ic_state != XLOG_STATE_IOERROR) { | |
2592 | if (--iclog->ic_bwritecnt == 1) { | |
b22cd72c | 2593 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2594 | return; |
2595 | } | |
2596 | iclog->ic_state = XLOG_STATE_DONE_SYNC; | |
2597 | } | |
2598 | ||
2599 | /* | |
2600 | * Someone could be sleeping prior to writing out the next | |
2601 | * iclog buffer, we wake them all, one will get to do the | |
2602 | * I/O, the others get to wait for the result. | |
2603 | */ | |
eb40a875 | 2604 | wake_up_all(&iclog->ic_write_wait); |
b22cd72c | 2605 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2606 | xlog_state_do_callback(log, aborted, iclog); /* also cleans log */ |
2607 | } /* xlog_state_done_syncing */ | |
2608 | ||
2609 | ||
2610 | /* | |
2611 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2612 | * sleep. We wait on the flush queue on the head iclog as that should be |
2613 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2614 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2615 | * |
2616 | * The in-core logs are used in a circular fashion. They are not used | |
2617 | * out-of-order even when an iclog past the head is free. | |
2618 | * | |
2619 | * return: | |
2620 | * * log_offset where xlog_write() can start writing into the in-core | |
2621 | * log's data space. | |
2622 | * * in-core log pointer to which xlog_write() should write. | |
2623 | * * boolean indicating this is a continued write to an in-core log. | |
2624 | * If this is the last write, then the in-core log's offset field | |
2625 | * needs to be incremented, depending on the amount of data which | |
2626 | * is copied. | |
2627 | */ | |
a8272ce0 | 2628 | STATIC int |
9a8d2fdb MT |
2629 | xlog_state_get_iclog_space( |
2630 | struct xlog *log, | |
2631 | int len, | |
2632 | struct xlog_in_core **iclogp, | |
2633 | struct xlog_ticket *ticket, | |
2634 | int *continued_write, | |
2635 | int *logoffsetp) | |
1da177e4 | 2636 | { |
1da177e4 LT |
2637 | int log_offset; |
2638 | xlog_rec_header_t *head; | |
2639 | xlog_in_core_t *iclog; | |
2640 | int error; | |
2641 | ||
2642 | restart: | |
b22cd72c | 2643 | spin_lock(&log->l_icloglock); |
1da177e4 | 2644 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2645 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2646 | return XFS_ERROR(EIO); |
2647 | } | |
2648 | ||
2649 | iclog = log->l_iclog; | |
d748c623 | 2650 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
1da177e4 | 2651 | XFS_STATS_INC(xs_log_noiclogs); |
d748c623 MW |
2652 | |
2653 | /* Wait for log writes to have flushed */ | |
eb40a875 | 2654 | xlog_wait(&log->l_flush_wait, &log->l_icloglock); |
1da177e4 LT |
2655 | goto restart; |
2656 | } | |
d748c623 | 2657 | |
1da177e4 LT |
2658 | head = &iclog->ic_header; |
2659 | ||
155cc6b7 | 2660 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
2661 | log_offset = iclog->ic_offset; |
2662 | ||
2663 | /* On the 1st write to an iclog, figure out lsn. This works | |
2664 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
2665 | * committing to. If the offset is set, that's how many blocks | |
2666 | * must be written. | |
2667 | */ | |
2668 | if (log_offset == 0) { | |
2669 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 2670 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
2671 | log->l_iclog_hsize, |
2672 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
2673 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
2674 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 2675 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
2676 | ASSERT(log->l_curr_block >= 0); |
2677 | } | |
2678 | ||
2679 | /* If there is enough room to write everything, then do it. Otherwise, | |
2680 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
2681 | * bit is on, so this will get flushed out. Don't update ic_offset | |
2682 | * until you know exactly how many bytes get copied. Therefore, wait | |
2683 | * until later to update ic_offset. | |
2684 | * | |
2685 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
2686 | * can fit into remaining data section. | |
2687 | */ | |
2688 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
2689 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2690 | ||
49641f1a DC |
2691 | /* |
2692 | * If I'm the only one writing to this iclog, sync it to disk. | |
2693 | * We need to do an atomic compare and decrement here to avoid | |
2694 | * racing with concurrent atomic_dec_and_lock() calls in | |
2695 | * xlog_state_release_iclog() when there is more than one | |
2696 | * reference to the iclog. | |
2697 | */ | |
2698 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) { | |
2699 | /* we are the only one */ | |
b22cd72c | 2700 | spin_unlock(&log->l_icloglock); |
49641f1a DC |
2701 | error = xlog_state_release_iclog(log, iclog); |
2702 | if (error) | |
014c2544 | 2703 | return error; |
1da177e4 | 2704 | } else { |
b22cd72c | 2705 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2706 | } |
2707 | goto restart; | |
2708 | } | |
2709 | ||
2710 | /* Do we have enough room to write the full amount in the remainder | |
2711 | * of this iclog? Or must we continue a write on the next iclog and | |
2712 | * mark this iclog as completely taken? In the case where we switch | |
2713 | * iclogs (to mark it taken), this particular iclog will release/sync | |
2714 | * to disk in xlog_write(). | |
2715 | */ | |
2716 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
2717 | *continued_write = 0; | |
2718 | iclog->ic_offset += len; | |
2719 | } else { | |
2720 | *continued_write = 1; | |
2721 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2722 | } | |
2723 | *iclogp = iclog; | |
2724 | ||
2725 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 2726 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2727 | |
2728 | *logoffsetp = log_offset; | |
2729 | return 0; | |
2730 | } /* xlog_state_get_iclog_space */ | |
2731 | ||
1da177e4 LT |
2732 | /* The first cnt-1 times through here we don't need to |
2733 | * move the grant write head because the permanent | |
2734 | * reservation has reserved cnt times the unit amount. | |
2735 | * Release part of current permanent unit reservation and | |
2736 | * reset current reservation to be one units worth. Also | |
2737 | * move grant reservation head forward. | |
2738 | */ | |
2739 | STATIC void | |
9a8d2fdb MT |
2740 | xlog_regrant_reserve_log_space( |
2741 | struct xlog *log, | |
2742 | struct xlog_ticket *ticket) | |
1da177e4 | 2743 | { |
0b1b213f CH |
2744 | trace_xfs_log_regrant_reserve_enter(log, ticket); |
2745 | ||
1da177e4 LT |
2746 | if (ticket->t_cnt > 0) |
2747 | ticket->t_cnt--; | |
2748 | ||
28496968 | 2749 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, |
a69ed03c | 2750 | ticket->t_curr_res); |
28496968 | 2751 | xlog_grant_sub_space(log, &log->l_write_head.grant, |
a69ed03c | 2752 | ticket->t_curr_res); |
1da177e4 | 2753 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2754 | xlog_tic_reset_res(ticket); |
0b1b213f CH |
2755 | |
2756 | trace_xfs_log_regrant_reserve_sub(log, ticket); | |
2757 | ||
1da177e4 | 2758 | /* just return if we still have some of the pre-reserved space */ |
d0eb2f38 | 2759 | if (ticket->t_cnt > 0) |
1da177e4 | 2760 | return; |
1da177e4 | 2761 | |
28496968 | 2762 | xlog_grant_add_space(log, &log->l_reserve_head.grant, |
a69ed03c | 2763 | ticket->t_unit_res); |
0b1b213f CH |
2764 | |
2765 | trace_xfs_log_regrant_reserve_exit(log, ticket); | |
2766 | ||
1da177e4 | 2767 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2768 | xlog_tic_reset_res(ticket); |
1da177e4 LT |
2769 | } /* xlog_regrant_reserve_log_space */ |
2770 | ||
2771 | ||
2772 | /* | |
2773 | * Give back the space left from a reservation. | |
2774 | * | |
2775 | * All the information we need to make a correct determination of space left | |
2776 | * is present. For non-permanent reservations, things are quite easy. The | |
2777 | * count should have been decremented to zero. We only need to deal with the | |
2778 | * space remaining in the current reservation part of the ticket. If the | |
2779 | * ticket contains a permanent reservation, there may be left over space which | |
2780 | * needs to be released. A count of N means that N-1 refills of the current | |
2781 | * reservation can be done before we need to ask for more space. The first | |
2782 | * one goes to fill up the first current reservation. Once we run out of | |
2783 | * space, the count will stay at zero and the only space remaining will be | |
2784 | * in the current reservation field. | |
2785 | */ | |
2786 | STATIC void | |
9a8d2fdb MT |
2787 | xlog_ungrant_log_space( |
2788 | struct xlog *log, | |
2789 | struct xlog_ticket *ticket) | |
1da177e4 | 2790 | { |
663e496a DC |
2791 | int bytes; |
2792 | ||
1da177e4 LT |
2793 | if (ticket->t_cnt > 0) |
2794 | ticket->t_cnt--; | |
2795 | ||
0b1b213f | 2796 | trace_xfs_log_ungrant_enter(log, ticket); |
0b1b213f | 2797 | trace_xfs_log_ungrant_sub(log, ticket); |
1da177e4 | 2798 | |
663e496a DC |
2799 | /* |
2800 | * If this is a permanent reservation ticket, we may be able to free | |
1da177e4 LT |
2801 | * up more space based on the remaining count. |
2802 | */ | |
663e496a | 2803 | bytes = ticket->t_curr_res; |
1da177e4 LT |
2804 | if (ticket->t_cnt > 0) { |
2805 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
663e496a | 2806 | bytes += ticket->t_unit_res*ticket->t_cnt; |
1da177e4 LT |
2807 | } |
2808 | ||
28496968 CH |
2809 | xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); |
2810 | xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); | |
663e496a | 2811 | |
0b1b213f CH |
2812 | trace_xfs_log_ungrant_exit(log, ticket); |
2813 | ||
cfb7cdca | 2814 | xfs_log_space_wake(log->l_mp); |
09a423a3 | 2815 | } |
1da177e4 | 2816 | |
1da177e4 LT |
2817 | /* |
2818 | * Flush iclog to disk if this is the last reference to the given iclog and | |
2819 | * the WANT_SYNC bit is set. | |
2820 | * | |
2821 | * When this function is entered, the iclog is not necessarily in the | |
2822 | * WANT_SYNC state. It may be sitting around waiting to get filled. | |
2823 | * | |
2824 | * | |
2825 | */ | |
a8272ce0 | 2826 | STATIC int |
b589334c | 2827 | xlog_state_release_iclog( |
9a8d2fdb MT |
2828 | struct xlog *log, |
2829 | struct xlog_in_core *iclog) | |
1da177e4 | 2830 | { |
1da177e4 LT |
2831 | int sync = 0; /* do we sync? */ |
2832 | ||
155cc6b7 DC |
2833 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
2834 | return XFS_ERROR(EIO); | |
2835 | ||
2836 | ASSERT(atomic_read(&iclog->ic_refcnt) > 0); | |
2837 | if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) | |
2838 | return 0; | |
2839 | ||
1da177e4 | 2840 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
b22cd72c | 2841 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2842 | return XFS_ERROR(EIO); |
2843 | } | |
1da177e4 LT |
2844 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE || |
2845 | iclog->ic_state == XLOG_STATE_WANT_SYNC); | |
2846 | ||
155cc6b7 | 2847 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { |
b589334c | 2848 | /* update tail before writing to iclog */ |
1c3cb9ec | 2849 | xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp); |
1da177e4 LT |
2850 | sync++; |
2851 | iclog->ic_state = XLOG_STATE_SYNCING; | |
1c3cb9ec DC |
2852 | iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); |
2853 | xlog_verify_tail_lsn(log, iclog, tail_lsn); | |
1da177e4 LT |
2854 | /* cycle incremented when incrementing curr_block */ |
2855 | } | |
b22cd72c | 2856 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2857 | |
2858 | /* | |
2859 | * We let the log lock go, so it's possible that we hit a log I/O | |
c41564b5 | 2860 | * error or some other SHUTDOWN condition that marks the iclog |
1da177e4 LT |
2861 | * as XLOG_STATE_IOERROR before the bwrite. However, we know that |
2862 | * this iclog has consistent data, so we ignore IOERROR | |
2863 | * flags after this point. | |
2864 | */ | |
b589334c | 2865 | if (sync) |
1da177e4 | 2866 | return xlog_sync(log, iclog); |
014c2544 | 2867 | return 0; |
1da177e4 LT |
2868 | } /* xlog_state_release_iclog */ |
2869 | ||
2870 | ||
2871 | /* | |
2872 | * This routine will mark the current iclog in the ring as WANT_SYNC | |
2873 | * and move the current iclog pointer to the next iclog in the ring. | |
2874 | * When this routine is called from xlog_state_get_iclog_space(), the | |
2875 | * exact size of the iclog has not yet been determined. All we know is | |
2876 | * that every data block. We have run out of space in this log record. | |
2877 | */ | |
2878 | STATIC void | |
9a8d2fdb MT |
2879 | xlog_state_switch_iclogs( |
2880 | struct xlog *log, | |
2881 | struct xlog_in_core *iclog, | |
2882 | int eventual_size) | |
1da177e4 LT |
2883 | { |
2884 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
2885 | if (!eventual_size) | |
2886 | eventual_size = iclog->ic_offset; | |
2887 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 2888 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
2889 | log->l_prev_block = log->l_curr_block; |
2890 | log->l_prev_cycle = log->l_curr_cycle; | |
2891 | ||
2892 | /* roll log?: ic_offset changed later */ | |
2893 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
2894 | ||
2895 | /* Round up to next log-sunit */ | |
62118709 | 2896 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
1da177e4 LT |
2897 | log->l_mp->m_sb.sb_logsunit > 1) { |
2898 | __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit); | |
2899 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); | |
2900 | } | |
2901 | ||
2902 | if (log->l_curr_block >= log->l_logBBsize) { | |
2903 | log->l_curr_cycle++; | |
2904 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
2905 | log->l_curr_cycle++; | |
2906 | log->l_curr_block -= log->l_logBBsize; | |
2907 | ASSERT(log->l_curr_block >= 0); | |
2908 | } | |
2909 | ASSERT(iclog == log->l_iclog); | |
2910 | log->l_iclog = iclog->ic_next; | |
2911 | } /* xlog_state_switch_iclogs */ | |
2912 | ||
1da177e4 LT |
2913 | /* |
2914 | * Write out all data in the in-core log as of this exact moment in time. | |
2915 | * | |
2916 | * Data may be written to the in-core log during this call. However, | |
2917 | * we don't guarantee this data will be written out. A change from past | |
2918 | * implementation means this routine will *not* write out zero length LRs. | |
2919 | * | |
2920 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
2921 | * If we determine there is no flushable data, we just return. There is no | |
2922 | * flushable data if: | |
2923 | * | |
2924 | * 1. the current iclog is active and has no data; the previous iclog | |
2925 | * is in the active or dirty state. | |
2926 | * 2. the current iclog is drity, and the previous iclog is in the | |
2927 | * active or dirty state. | |
2928 | * | |
12017faf | 2929 | * We may sleep if: |
1da177e4 LT |
2930 | * |
2931 | * 1. the current iclog is not in the active nor dirty state. | |
2932 | * 2. the current iclog dirty, and the previous iclog is not in the | |
2933 | * active nor dirty state. | |
2934 | * 3. the current iclog is active, and there is another thread writing | |
2935 | * to this particular iclog. | |
2936 | * 4. a) the current iclog is active and has no other writers | |
2937 | * b) when we return from flushing out this iclog, it is still | |
2938 | * not in the active nor dirty state. | |
2939 | */ | |
a14a348b CH |
2940 | int |
2941 | _xfs_log_force( | |
2942 | struct xfs_mount *mp, | |
2943 | uint flags, | |
2944 | int *log_flushed) | |
1da177e4 | 2945 | { |
ad223e60 | 2946 | struct xlog *log = mp->m_log; |
a14a348b CH |
2947 | struct xlog_in_core *iclog; |
2948 | xfs_lsn_t lsn; | |
2949 | ||
2950 | XFS_STATS_INC(xs_log_force); | |
1da177e4 | 2951 | |
93b8a585 | 2952 | xlog_cil_force(log); |
71e330b5 | 2953 | |
b22cd72c | 2954 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2955 | |
2956 | iclog = log->l_iclog; | |
2957 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 2958 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2959 | return XFS_ERROR(EIO); |
2960 | } | |
2961 | ||
2962 | /* If the head iclog is not active nor dirty, we just attach | |
2963 | * ourselves to the head and go to sleep. | |
2964 | */ | |
2965 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
2966 | iclog->ic_state == XLOG_STATE_DIRTY) { | |
2967 | /* | |
2968 | * If the head is dirty or (active and empty), then | |
2969 | * we need to look at the previous iclog. If the previous | |
2970 | * iclog is active or dirty we are done. There is nothing | |
2971 | * to sync out. Otherwise, we attach ourselves to the | |
2972 | * previous iclog and go to sleep. | |
2973 | */ | |
2974 | if (iclog->ic_state == XLOG_STATE_DIRTY || | |
155cc6b7 DC |
2975 | (atomic_read(&iclog->ic_refcnt) == 0 |
2976 | && iclog->ic_offset == 0)) { | |
1da177e4 LT |
2977 | iclog = iclog->ic_prev; |
2978 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
2979 | iclog->ic_state == XLOG_STATE_DIRTY) | |
2980 | goto no_sleep; | |
2981 | else | |
2982 | goto maybe_sleep; | |
2983 | } else { | |
155cc6b7 | 2984 | if (atomic_read(&iclog->ic_refcnt) == 0) { |
1da177e4 LT |
2985 | /* We are the only one with access to this |
2986 | * iclog. Flush it out now. There should | |
2987 | * be a roundoff of zero to show that someone | |
2988 | * has already taken care of the roundoff from | |
2989 | * the previous sync. | |
2990 | */ | |
155cc6b7 | 2991 | atomic_inc(&iclog->ic_refcnt); |
b53e675d | 2992 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
1da177e4 | 2993 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 2994 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2995 | |
2996 | if (xlog_state_release_iclog(log, iclog)) | |
2997 | return XFS_ERROR(EIO); | |
a14a348b CH |
2998 | |
2999 | if (log_flushed) | |
3000 | *log_flushed = 1; | |
b22cd72c | 3001 | spin_lock(&log->l_icloglock); |
b53e675d | 3002 | if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn && |
1da177e4 LT |
3003 | iclog->ic_state != XLOG_STATE_DIRTY) |
3004 | goto maybe_sleep; | |
3005 | else | |
3006 | goto no_sleep; | |
3007 | } else { | |
3008 | /* Someone else is writing to this iclog. | |
3009 | * Use its call to flush out the data. However, | |
3010 | * the other thread may not force out this LR, | |
3011 | * so we mark it WANT_SYNC. | |
3012 | */ | |
3013 | xlog_state_switch_iclogs(log, iclog, 0); | |
3014 | goto maybe_sleep; | |
3015 | } | |
3016 | } | |
3017 | } | |
3018 | ||
3019 | /* By the time we come around again, the iclog could've been filled | |
3020 | * which would give it another lsn. If we have a new lsn, just | |
3021 | * return because the relevant data has been flushed. | |
3022 | */ | |
3023 | maybe_sleep: | |
3024 | if (flags & XFS_LOG_SYNC) { | |
3025 | /* | |
3026 | * We must check if we're shutting down here, before | |
b22cd72c | 3027 | * we wait, while we're holding the l_icloglock. |
1da177e4 LT |
3028 | * Then we check again after waking up, in case our |
3029 | * sleep was disturbed by a bad news. | |
3030 | */ | |
3031 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3032 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3033 | return XFS_ERROR(EIO); |
3034 | } | |
3035 | XFS_STATS_INC(xs_log_force_sleep); | |
eb40a875 | 3036 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); |
1da177e4 LT |
3037 | /* |
3038 | * No need to grab the log lock here since we're | |
3039 | * only deciding whether or not to return EIO | |
3040 | * and the memory read should be atomic. | |
3041 | */ | |
3042 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3043 | return XFS_ERROR(EIO); | |
a14a348b CH |
3044 | if (log_flushed) |
3045 | *log_flushed = 1; | |
1da177e4 LT |
3046 | } else { |
3047 | ||
3048 | no_sleep: | |
b22cd72c | 3049 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3050 | } |
3051 | return 0; | |
a14a348b | 3052 | } |
1da177e4 | 3053 | |
a14a348b CH |
3054 | /* |
3055 | * Wrapper for _xfs_log_force(), to be used when caller doesn't care | |
3056 | * about errors or whether the log was flushed or not. This is the normal | |
3057 | * interface to use when trying to unpin items or move the log forward. | |
3058 | */ | |
3059 | void | |
3060 | xfs_log_force( | |
3061 | xfs_mount_t *mp, | |
3062 | uint flags) | |
3063 | { | |
3064 | int error; | |
3065 | ||
14c26c6a | 3066 | trace_xfs_log_force(mp, 0); |
a14a348b | 3067 | error = _xfs_log_force(mp, flags, NULL); |
a0fa2b67 DC |
3068 | if (error) |
3069 | xfs_warn(mp, "%s: error %d returned.", __func__, error); | |
a14a348b | 3070 | } |
1da177e4 LT |
3071 | |
3072 | /* | |
a14a348b | 3073 | * Force the in-core log to disk for a specific LSN. |
1da177e4 LT |
3074 | * |
3075 | * Find in-core log with lsn. | |
3076 | * If it is in the DIRTY state, just return. | |
3077 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3078 | * state and go to sleep or return. | |
3079 | * If it is in any other state, go to sleep or return. | |
3080 | * | |
a14a348b CH |
3081 | * Synchronous forces are implemented with a signal variable. All callers |
3082 | * to force a given lsn to disk will wait on a the sv attached to the | |
3083 | * specific in-core log. When given in-core log finally completes its | |
3084 | * write to disk, that thread will wake up all threads waiting on the | |
3085 | * sv. | |
1da177e4 | 3086 | */ |
a14a348b CH |
3087 | int |
3088 | _xfs_log_force_lsn( | |
3089 | struct xfs_mount *mp, | |
3090 | xfs_lsn_t lsn, | |
3091 | uint flags, | |
3092 | int *log_flushed) | |
1da177e4 | 3093 | { |
ad223e60 | 3094 | struct xlog *log = mp->m_log; |
a14a348b CH |
3095 | struct xlog_in_core *iclog; |
3096 | int already_slept = 0; | |
1da177e4 | 3097 | |
a14a348b | 3098 | ASSERT(lsn != 0); |
1da177e4 | 3099 | |
a14a348b | 3100 | XFS_STATS_INC(xs_log_force); |
1da177e4 | 3101 | |
93b8a585 CH |
3102 | lsn = xlog_cil_force_lsn(log, lsn); |
3103 | if (lsn == NULLCOMMITLSN) | |
3104 | return 0; | |
71e330b5 | 3105 | |
a14a348b CH |
3106 | try_again: |
3107 | spin_lock(&log->l_icloglock); | |
3108 | iclog = log->l_iclog; | |
3109 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3110 | spin_unlock(&log->l_icloglock); |
a14a348b | 3111 | return XFS_ERROR(EIO); |
1da177e4 LT |
3112 | } |
3113 | ||
a14a348b CH |
3114 | do { |
3115 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { | |
3116 | iclog = iclog->ic_next; | |
3117 | continue; | |
3118 | } | |
3119 | ||
3120 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
3121 | spin_unlock(&log->l_icloglock); | |
3122 | return 0; | |
3123 | } | |
3124 | ||
3125 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3126 | /* | |
3127 | * We sleep here if we haven't already slept (e.g. | |
3128 | * this is the first time we've looked at the correct | |
3129 | * iclog buf) and the buffer before us is going to | |
3130 | * be sync'ed. The reason for this is that if we | |
3131 | * are doing sync transactions here, by waiting for | |
3132 | * the previous I/O to complete, we can allow a few | |
3133 | * more transactions into this iclog before we close | |
3134 | * it down. | |
3135 | * | |
3136 | * Otherwise, we mark the buffer WANT_SYNC, and bump | |
3137 | * up the refcnt so we can release the log (which | |
3138 | * drops the ref count). The state switch keeps new | |
3139 | * transaction commits from using this buffer. When | |
3140 | * the current commits finish writing into the buffer, | |
3141 | * the refcount will drop to zero and the buffer will | |
3142 | * go out then. | |
3143 | */ | |
3144 | if (!already_slept && | |
3145 | (iclog->ic_prev->ic_state & | |
3146 | (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) { | |
3147 | ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); | |
3148 | ||
3149 | XFS_STATS_INC(xs_log_force_sleep); | |
3150 | ||
eb40a875 DC |
3151 | xlog_wait(&iclog->ic_prev->ic_write_wait, |
3152 | &log->l_icloglock); | |
a14a348b CH |
3153 | if (log_flushed) |
3154 | *log_flushed = 1; | |
3155 | already_slept = 1; | |
3156 | goto try_again; | |
3157 | } | |
155cc6b7 | 3158 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 3159 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 3160 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3161 | if (xlog_state_release_iclog(log, iclog)) |
3162 | return XFS_ERROR(EIO); | |
a14a348b CH |
3163 | if (log_flushed) |
3164 | *log_flushed = 1; | |
b22cd72c | 3165 | spin_lock(&log->l_icloglock); |
1da177e4 | 3166 | } |
1da177e4 | 3167 | |
a14a348b CH |
3168 | if ((flags & XFS_LOG_SYNC) && /* sleep */ |
3169 | !(iclog->ic_state & | |
3170 | (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) { | |
3171 | /* | |
3172 | * Don't wait on completion if we know that we've | |
3173 | * gotten a log write error. | |
3174 | */ | |
3175 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
3176 | spin_unlock(&log->l_icloglock); | |
3177 | return XFS_ERROR(EIO); | |
3178 | } | |
3179 | XFS_STATS_INC(xs_log_force_sleep); | |
eb40a875 | 3180 | xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); |
a14a348b CH |
3181 | /* |
3182 | * No need to grab the log lock here since we're | |
3183 | * only deciding whether or not to return EIO | |
3184 | * and the memory read should be atomic. | |
3185 | */ | |
3186 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3187 | return XFS_ERROR(EIO); | |
1da177e4 | 3188 | |
a14a348b CH |
3189 | if (log_flushed) |
3190 | *log_flushed = 1; | |
3191 | } else { /* just return */ | |
b22cd72c | 3192 | spin_unlock(&log->l_icloglock); |
1da177e4 | 3193 | } |
1da177e4 | 3194 | |
a14a348b CH |
3195 | return 0; |
3196 | } while (iclog != log->l_iclog); | |
1da177e4 | 3197 | |
a14a348b CH |
3198 | spin_unlock(&log->l_icloglock); |
3199 | return 0; | |
3200 | } | |
3201 | ||
3202 | /* | |
3203 | * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care | |
3204 | * about errors or whether the log was flushed or not. This is the normal | |
3205 | * interface to use when trying to unpin items or move the log forward. | |
3206 | */ | |
3207 | void | |
3208 | xfs_log_force_lsn( | |
3209 | xfs_mount_t *mp, | |
3210 | xfs_lsn_t lsn, | |
3211 | uint flags) | |
3212 | { | |
3213 | int error; | |
1da177e4 | 3214 | |
14c26c6a | 3215 | trace_xfs_log_force(mp, lsn); |
a14a348b | 3216 | error = _xfs_log_force_lsn(mp, lsn, flags, NULL); |
a0fa2b67 DC |
3217 | if (error) |
3218 | xfs_warn(mp, "%s: error %d returned.", __func__, error); | |
a14a348b | 3219 | } |
1da177e4 LT |
3220 | |
3221 | /* | |
3222 | * Called when we want to mark the current iclog as being ready to sync to | |
3223 | * disk. | |
3224 | */ | |
a8272ce0 | 3225 | STATIC void |
9a8d2fdb MT |
3226 | xlog_state_want_sync( |
3227 | struct xlog *log, | |
3228 | struct xlog_in_core *iclog) | |
1da177e4 | 3229 | { |
a8914f3a | 3230 | assert_spin_locked(&log->l_icloglock); |
1da177e4 LT |
3231 | |
3232 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3233 | xlog_state_switch_iclogs(log, iclog, 0); | |
3234 | } else { | |
3235 | ASSERT(iclog->ic_state & | |
3236 | (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR)); | |
3237 | } | |
39e2defe | 3238 | } |
1da177e4 LT |
3239 | |
3240 | ||
3241 | /***************************************************************************** | |
3242 | * | |
3243 | * TICKET functions | |
3244 | * | |
3245 | ***************************************************************************** | |
3246 | */ | |
3247 | ||
3248 | /* | |
9da096fd | 3249 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3250 | */ |
cc09c0dc DC |
3251 | void |
3252 | xfs_log_ticket_put( | |
3253 | xlog_ticket_t *ticket) | |
1da177e4 | 3254 | { |
cc09c0dc | 3255 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
eb40a875 | 3256 | if (atomic_dec_and_test(&ticket->t_ref)) |
cc09c0dc | 3257 | kmem_zone_free(xfs_log_ticket_zone, ticket); |
cc09c0dc | 3258 | } |
1da177e4 | 3259 | |
cc09c0dc DC |
3260 | xlog_ticket_t * |
3261 | xfs_log_ticket_get( | |
3262 | xlog_ticket_t *ticket) | |
3263 | { | |
3264 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3265 | atomic_inc(&ticket->t_ref); | |
3266 | return ticket; | |
3267 | } | |
1da177e4 LT |
3268 | |
3269 | /* | |
eb01c9cd | 3270 | * Allocate and initialise a new log ticket. |
1da177e4 | 3271 | */ |
9a8d2fdb | 3272 | struct xlog_ticket * |
9b9fc2b7 | 3273 | xlog_ticket_alloc( |
ad223e60 | 3274 | struct xlog *log, |
9b9fc2b7 DC |
3275 | int unit_bytes, |
3276 | int cnt, | |
3277 | char client, | |
9006fb91 | 3278 | bool permanent, |
77ba7877 | 3279 | xfs_km_flags_t alloc_flags) |
1da177e4 | 3280 | { |
9b9fc2b7 | 3281 | struct xlog_ticket *tic; |
1da177e4 | 3282 | uint num_headers; |
9b9fc2b7 | 3283 | int iclog_space; |
1da177e4 | 3284 | |
3383ca57 | 3285 | tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags); |
eb01c9cd DC |
3286 | if (!tic) |
3287 | return NULL; | |
1da177e4 LT |
3288 | |
3289 | /* | |
3290 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3291 | * in the log. A unit in this case is the amount of space for one | |
3292 | * of these log operations. Normal reservations have a cnt of 1 | |
3293 | * and their unit amount is the total amount of space required. | |
3294 | * | |
3295 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3296 | * which occupy space in the on-disk log. |
3297 | * | |
3298 | * Normal form of a transaction is: | |
3299 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3300 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3301 | * | |
3302 | * We need to account for all the leadup data and trailer data | |
3303 | * around the transaction data. | |
3304 | * And then we need to account for the worst case in terms of using | |
3305 | * more space. | |
3306 | * The worst case will happen if: | |
3307 | * - the placement of the transaction happens to be such that the | |
3308 | * roundoff is at its maximum | |
3309 | * - the transaction data is synced before the commit record is synced | |
3310 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3311 | * Therefore the commit record is in its own Log Record. | |
3312 | * This can happen as the commit record is called with its | |
3313 | * own region to xlog_write(). | |
3314 | * This then means that in the worst case, roundoff can happen for | |
3315 | * the commit-rec as well. | |
3316 | * The commit-rec is smaller than padding in this scenario and so it is | |
3317 | * not added separately. | |
1da177e4 LT |
3318 | */ |
3319 | ||
32fb9b57 TS |
3320 | /* for trans header */ |
3321 | unit_bytes += sizeof(xlog_op_header_t); | |
3322 | unit_bytes += sizeof(xfs_trans_header_t); | |
3323 | ||
1da177e4 | 3324 | /* for start-rec */ |
32fb9b57 TS |
3325 | unit_bytes += sizeof(xlog_op_header_t); |
3326 | ||
9b9fc2b7 DC |
3327 | /* |
3328 | * for LR headers - the space for data in an iclog is the size minus | |
3329 | * the space used for the headers. If we use the iclog size, then we | |
3330 | * undercalculate the number of headers required. | |
3331 | * | |
3332 | * Furthermore - the addition of op headers for split-recs might | |
3333 | * increase the space required enough to require more log and op | |
3334 | * headers, so take that into account too. | |
3335 | * | |
3336 | * IMPORTANT: This reservation makes the assumption that if this | |
3337 | * transaction is the first in an iclog and hence has the LR headers | |
3338 | * accounted to it, then the remaining space in the iclog is | |
3339 | * exclusively for this transaction. i.e. if the transaction is larger | |
3340 | * than the iclog, it will be the only thing in that iclog. | |
3341 | * Fundamentally, this means we must pass the entire log vector to | |
3342 | * xlog_write to guarantee this. | |
3343 | */ | |
3344 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3345 | num_headers = howmany(unit_bytes, iclog_space); | |
3346 | ||
3347 | /* for split-recs - ophdrs added when data split over LRs */ | |
3348 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3349 | ||
3350 | /* add extra header reservations if we overrun */ | |
3351 | while (!num_headers || | |
3352 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3353 | unit_bytes += sizeof(xlog_op_header_t); | |
3354 | num_headers++; | |
3355 | } | |
32fb9b57 | 3356 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3357 | |
32fb9b57 TS |
3358 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3359 | unit_bytes += log->l_iclog_hsize; | |
3360 | ||
32fb9b57 | 3361 | /* for roundoff padding for transaction data and one for commit record */ |
62118709 | 3362 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
32fb9b57 | 3363 | log->l_mp->m_sb.sb_logsunit > 1) { |
1da177e4 | 3364 | /* log su roundoff */ |
32fb9b57 | 3365 | unit_bytes += 2*log->l_mp->m_sb.sb_logsunit; |
1da177e4 LT |
3366 | } else { |
3367 | /* BB roundoff */ | |
32fb9b57 | 3368 | unit_bytes += 2*BBSIZE; |
1da177e4 LT |
3369 | } |
3370 | ||
cc09c0dc | 3371 | atomic_set(&tic->t_ref, 1); |
14a7235f | 3372 | tic->t_task = current; |
10547941 | 3373 | INIT_LIST_HEAD(&tic->t_queue); |
1da177e4 LT |
3374 | tic->t_unit_res = unit_bytes; |
3375 | tic->t_curr_res = unit_bytes; | |
3376 | tic->t_cnt = cnt; | |
3377 | tic->t_ocnt = cnt; | |
f9837107 | 3378 | tic->t_tid = random32(); |
1da177e4 LT |
3379 | tic->t_clientid = client; |
3380 | tic->t_flags = XLOG_TIC_INITED; | |
7e9c6396 | 3381 | tic->t_trans_type = 0; |
9006fb91 | 3382 | if (permanent) |
1da177e4 | 3383 | tic->t_flags |= XLOG_TIC_PERM_RESERV; |
1da177e4 | 3384 | |
0adba536 | 3385 | xlog_tic_reset_res(tic); |
7e9c6396 | 3386 | |
1da177e4 | 3387 | return tic; |
cc09c0dc | 3388 | } |
1da177e4 LT |
3389 | |
3390 | ||
3391 | /****************************************************************************** | |
3392 | * | |
3393 | * Log debug routines | |
3394 | * | |
3395 | ****************************************************************************** | |
3396 | */ | |
cfcbbbd0 | 3397 | #if defined(DEBUG) |
1da177e4 LT |
3398 | /* |
3399 | * Make sure that the destination ptr is within the valid data region of | |
3400 | * one of the iclogs. This uses backup pointers stored in a different | |
3401 | * part of the log in case we trash the log structure. | |
3402 | */ | |
3403 | void | |
e6b1f273 | 3404 | xlog_verify_dest_ptr( |
ad223e60 | 3405 | struct xlog *log, |
e6b1f273 | 3406 | char *ptr) |
1da177e4 LT |
3407 | { |
3408 | int i; | |
3409 | int good_ptr = 0; | |
3410 | ||
e6b1f273 CH |
3411 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3412 | if (ptr >= log->l_iclog_bak[i] && | |
3413 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3414 | good_ptr++; |
3415 | } | |
e6b1f273 CH |
3416 | |
3417 | if (!good_ptr) | |
a0fa2b67 | 3418 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
e6b1f273 | 3419 | } |
1da177e4 | 3420 | |
da8a1a4a DC |
3421 | /* |
3422 | * Check to make sure the grant write head didn't just over lap the tail. If | |
3423 | * the cycles are the same, we can't be overlapping. Otherwise, make sure that | |
3424 | * the cycles differ by exactly one and check the byte count. | |
3425 | * | |
3426 | * This check is run unlocked, so can give false positives. Rather than assert | |
3427 | * on failures, use a warn-once flag and a panic tag to allow the admin to | |
3428 | * determine if they want to panic the machine when such an error occurs. For | |
3429 | * debug kernels this will have the same effect as using an assert but, unlinke | |
3430 | * an assert, it can be turned off at runtime. | |
3431 | */ | |
3f336c6f DC |
3432 | STATIC void |
3433 | xlog_verify_grant_tail( | |
ad223e60 | 3434 | struct xlog *log) |
3f336c6f | 3435 | { |
1c3cb9ec | 3436 | int tail_cycle, tail_blocks; |
a69ed03c | 3437 | int cycle, space; |
3f336c6f | 3438 | |
28496968 | 3439 | xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); |
1c3cb9ec DC |
3440 | xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); |
3441 | if (tail_cycle != cycle) { | |
da8a1a4a DC |
3442 | if (cycle - 1 != tail_cycle && |
3443 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3444 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3445 | "%s: cycle - 1 != tail_cycle", __func__); | |
3446 | log->l_flags |= XLOG_TAIL_WARN; | |
3447 | } | |
3448 | ||
3449 | if (space > BBTOB(tail_blocks) && | |
3450 | !(log->l_flags & XLOG_TAIL_WARN)) { | |
3451 | xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, | |
3452 | "%s: space > BBTOB(tail_blocks)", __func__); | |
3453 | log->l_flags |= XLOG_TAIL_WARN; | |
3454 | } | |
3f336c6f DC |
3455 | } |
3456 | } | |
3457 | ||
1da177e4 LT |
3458 | /* check if it will fit */ |
3459 | STATIC void | |
9a8d2fdb MT |
3460 | xlog_verify_tail_lsn( |
3461 | struct xlog *log, | |
3462 | struct xlog_in_core *iclog, | |
3463 | xfs_lsn_t tail_lsn) | |
1da177e4 LT |
3464 | { |
3465 | int blocks; | |
3466 | ||
3467 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3468 | blocks = | |
3469 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3470 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
a0fa2b67 | 3471 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3472 | } else { |
3473 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3474 | ||
3475 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
a0fa2b67 | 3476 | xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); |
1da177e4 LT |
3477 | |
3478 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3479 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
a0fa2b67 | 3480 | xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); |
1da177e4 LT |
3481 | } |
3482 | } /* xlog_verify_tail_lsn */ | |
3483 | ||
3484 | /* | |
3485 | * Perform a number of checks on the iclog before writing to disk. | |
3486 | * | |
3487 | * 1. Make sure the iclogs are still circular | |
3488 | * 2. Make sure we have a good magic number | |
3489 | * 3. Make sure we don't have magic numbers in the data | |
3490 | * 4. Check fields of each log operation header for: | |
3491 | * A. Valid client identifier | |
3492 | * B. tid ptr value falls in valid ptr space (user space code) | |
3493 | * C. Length in log record header is correct according to the | |
3494 | * individual operation headers within record. | |
3495 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3496 | * log, check the preceding blocks of the physical log to make sure all | |
3497 | * the cycle numbers agree with the current cycle number. | |
3498 | */ | |
3499 | STATIC void | |
9a8d2fdb MT |
3500 | xlog_verify_iclog( |
3501 | struct xlog *log, | |
3502 | struct xlog_in_core *iclog, | |
3503 | int count, | |
3504 | boolean_t syncing) | |
1da177e4 LT |
3505 | { |
3506 | xlog_op_header_t *ophead; | |
3507 | xlog_in_core_t *icptr; | |
3508 | xlog_in_core_2_t *xhdr; | |
3509 | xfs_caddr_t ptr; | |
3510 | xfs_caddr_t base_ptr; | |
3511 | __psint_t field_offset; | |
3512 | __uint8_t clientid; | |
3513 | int len, i, j, k, op_len; | |
3514 | int idx; | |
1da177e4 LT |
3515 | |
3516 | /* check validity of iclog pointers */ | |
b22cd72c | 3517 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3518 | icptr = log->l_iclog; |
3519 | for (i=0; i < log->l_iclog_bufs; i++) { | |
4b80916b | 3520 | if (icptr == NULL) |
a0fa2b67 | 3521 | xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); |
1da177e4 LT |
3522 | icptr = icptr->ic_next; |
3523 | } | |
3524 | if (icptr != log->l_iclog) | |
a0fa2b67 | 3525 | xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); |
b22cd72c | 3526 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3527 | |
3528 | /* check log magic numbers */ | |
69ef921b | 3529 | if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 | 3530 | xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); |
1da177e4 | 3531 | |
b53e675d CH |
3532 | ptr = (xfs_caddr_t) &iclog->ic_header; |
3533 | for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count; | |
1da177e4 | 3534 | ptr += BBSIZE) { |
69ef921b | 3535 | if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) |
a0fa2b67 DC |
3536 | xfs_emerg(log->l_mp, "%s: unexpected magic num", |
3537 | __func__); | |
1da177e4 LT |
3538 | } |
3539 | ||
3540 | /* check fields */ | |
b53e675d | 3541 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
1da177e4 LT |
3542 | ptr = iclog->ic_datap; |
3543 | base_ptr = ptr; | |
3544 | ophead = (xlog_op_header_t *)ptr; | |
b28708d6 | 3545 | xhdr = iclog->ic_data; |
1da177e4 LT |
3546 | for (i = 0; i < len; i++) { |
3547 | ophead = (xlog_op_header_t *)ptr; | |
3548 | ||
3549 | /* clientid is only 1 byte */ | |
3550 | field_offset = (__psint_t) | |
3551 | ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr); | |
3552 | if (syncing == B_FALSE || (field_offset & 0x1ff)) { | |
3553 | clientid = ophead->oh_clientid; | |
3554 | } else { | |
3555 | idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap); | |
3556 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3557 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3558 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3559 | clientid = xlog_get_client_id( |
3560 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3561 | } else { |
03bea6fe CH |
3562 | clientid = xlog_get_client_id( |
3563 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3564 | } |
3565 | } | |
3566 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
a0fa2b67 DC |
3567 | xfs_warn(log->l_mp, |
3568 | "%s: invalid clientid %d op 0x%p offset 0x%lx", | |
3569 | __func__, clientid, ophead, | |
3570 | (unsigned long)field_offset); | |
1da177e4 LT |
3571 | |
3572 | /* check length */ | |
3573 | field_offset = (__psint_t) | |
3574 | ((xfs_caddr_t)&(ophead->oh_len) - base_ptr); | |
3575 | if (syncing == B_FALSE || (field_offset & 0x1ff)) { | |
67fcb7bf | 3576 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 LT |
3577 | } else { |
3578 | idx = BTOBBT((__psint_t)&ophead->oh_len - | |
3579 | (__psint_t)iclog->ic_datap); | |
3580 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3581 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3582 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3583 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3584 | } else { |
b53e675d | 3585 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3586 | } |
3587 | } | |
3588 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3589 | } | |
3590 | } /* xlog_verify_iclog */ | |
cfcbbbd0 | 3591 | #endif |
1da177e4 LT |
3592 | |
3593 | /* | |
b22cd72c | 3594 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3595 | */ |
3596 | STATIC int | |
3597 | xlog_state_ioerror( | |
9a8d2fdb | 3598 | struct xlog *log) |
1da177e4 LT |
3599 | { |
3600 | xlog_in_core_t *iclog, *ic; | |
3601 | ||
3602 | iclog = log->l_iclog; | |
3603 | if (! (iclog->ic_state & XLOG_STATE_IOERROR)) { | |
3604 | /* | |
3605 | * Mark all the incore logs IOERROR. | |
3606 | * From now on, no log flushes will result. | |
3607 | */ | |
3608 | ic = iclog; | |
3609 | do { | |
3610 | ic->ic_state = XLOG_STATE_IOERROR; | |
3611 | ic = ic->ic_next; | |
3612 | } while (ic != iclog); | |
014c2544 | 3613 | return 0; |
1da177e4 LT |
3614 | } |
3615 | /* | |
3616 | * Return non-zero, if state transition has already happened. | |
3617 | */ | |
014c2544 | 3618 | return 1; |
1da177e4 LT |
3619 | } |
3620 | ||
3621 | /* | |
3622 | * This is called from xfs_force_shutdown, when we're forcibly | |
3623 | * shutting down the filesystem, typically because of an IO error. | |
3624 | * Our main objectives here are to make sure that: | |
3625 | * a. the filesystem gets marked 'SHUTDOWN' for all interested | |
3626 | * parties to find out, 'atomically'. | |
3627 | * b. those who're sleeping on log reservations, pinned objects and | |
3628 | * other resources get woken up, and be told the bad news. | |
3629 | * c. nothing new gets queued up after (a) and (b) are done. | |
3630 | * d. if !logerror, flush the iclogs to disk, then seal them off | |
3631 | * for business. | |
9da1ab18 DC |
3632 | * |
3633 | * Note: for delayed logging the !logerror case needs to flush the regions | |
3634 | * held in memory out to the iclogs before flushing them to disk. This needs | |
3635 | * to be done before the log is marked as shutdown, otherwise the flush to the | |
3636 | * iclogs will fail. | |
1da177e4 LT |
3637 | */ |
3638 | int | |
3639 | xfs_log_force_umount( | |
3640 | struct xfs_mount *mp, | |
3641 | int logerror) | |
3642 | { | |
9a8d2fdb | 3643 | struct xlog *log; |
1da177e4 | 3644 | int retval; |
1da177e4 LT |
3645 | |
3646 | log = mp->m_log; | |
3647 | ||
3648 | /* | |
3649 | * If this happens during log recovery, don't worry about | |
3650 | * locking; the log isn't open for business yet. | |
3651 | */ | |
3652 | if (!log || | |
3653 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3654 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 CH |
3655 | if (mp->m_sb_bp) |
3656 | XFS_BUF_DONE(mp->m_sb_bp); | |
014c2544 | 3657 | return 0; |
1da177e4 LT |
3658 | } |
3659 | ||
3660 | /* | |
3661 | * Somebody could've already done the hard work for us. | |
3662 | * No need to get locks for this. | |
3663 | */ | |
3664 | if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) { | |
3665 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); | |
014c2544 | 3666 | return 1; |
1da177e4 LT |
3667 | } |
3668 | retval = 0; | |
9da1ab18 DC |
3669 | |
3670 | /* | |
3671 | * Flush the in memory commit item list before marking the log as | |
3672 | * being shut down. We need to do it in this order to ensure all the | |
3673 | * completed transactions are flushed to disk with the xfs_log_force() | |
3674 | * call below. | |
3675 | */ | |
93b8a585 | 3676 | if (!logerror) |
a44f13ed | 3677 | xlog_cil_force(log); |
9da1ab18 | 3678 | |
1da177e4 | 3679 | /* |
3f16b985 DC |
3680 | * mark the filesystem and the as in a shutdown state and wake |
3681 | * everybody up to tell them the bad news. | |
1da177e4 | 3682 | */ |
b22cd72c | 3683 | spin_lock(&log->l_icloglock); |
1da177e4 | 3684 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 CH |
3685 | if (mp->m_sb_bp) |
3686 | XFS_BUF_DONE(mp->m_sb_bp); | |
3687 | ||
1da177e4 LT |
3688 | /* |
3689 | * This flag is sort of redundant because of the mount flag, but | |
3690 | * it's good to maintain the separation between the log and the rest | |
3691 | * of XFS. | |
3692 | */ | |
3693 | log->l_flags |= XLOG_IO_ERROR; | |
3694 | ||
3695 | /* | |
3696 | * If we hit a log error, we want to mark all the iclogs IOERROR | |
3697 | * while we're still holding the loglock. | |
3698 | */ | |
3699 | if (logerror) | |
3700 | retval = xlog_state_ioerror(log); | |
b22cd72c | 3701 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3702 | |
3703 | /* | |
10547941 DC |
3704 | * We don't want anybody waiting for log reservations after this. That |
3705 | * means we have to wake up everybody queued up on reserveq as well as | |
3706 | * writeq. In addition, we make sure in xlog_{re}grant_log_space that | |
3707 | * we don't enqueue anything once the SHUTDOWN flag is set, and this | |
3f16b985 | 3708 | * action is protected by the grant locks. |
1da177e4 | 3709 | */ |
a79bf2d7 CH |
3710 | xlog_grant_head_wake_all(&log->l_reserve_head); |
3711 | xlog_grant_head_wake_all(&log->l_write_head); | |
1da177e4 | 3712 | |
a14a348b | 3713 | if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) { |
1da177e4 LT |
3714 | ASSERT(!logerror); |
3715 | /* | |
3716 | * Force the incore logs to disk before shutting the | |
3717 | * log down completely. | |
3718 | */ | |
a14a348b CH |
3719 | _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
3720 | ||
b22cd72c | 3721 | spin_lock(&log->l_icloglock); |
1da177e4 | 3722 | retval = xlog_state_ioerror(log); |
b22cd72c | 3723 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3724 | } |
3725 | /* | |
3726 | * Wake up everybody waiting on xfs_log_force. | |
3727 | * Callback all log item committed functions as if the | |
3728 | * log writes were completed. | |
3729 | */ | |
3730 | xlog_state_do_callback(log, XFS_LI_ABORTED, NULL); | |
3731 | ||
3732 | #ifdef XFSERRORDEBUG | |
3733 | { | |
3734 | xlog_in_core_t *iclog; | |
3735 | ||
b22cd72c | 3736 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3737 | iclog = log->l_iclog; |
3738 | do { | |
3739 | ASSERT(iclog->ic_callback == 0); | |
3740 | iclog = iclog->ic_next; | |
3741 | } while (iclog != log->l_iclog); | |
b22cd72c | 3742 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3743 | } |
3744 | #endif | |
3745 | /* return non-zero if log IOERROR transition had already happened */ | |
014c2544 | 3746 | return retval; |
1da177e4 LT |
3747 | } |
3748 | ||
ba0f32d4 | 3749 | STATIC int |
9a8d2fdb MT |
3750 | xlog_iclogs_empty( |
3751 | struct xlog *log) | |
1da177e4 LT |
3752 | { |
3753 | xlog_in_core_t *iclog; | |
3754 | ||
3755 | iclog = log->l_iclog; | |
3756 | do { | |
3757 | /* endianness does not matter here, zero is zero in | |
3758 | * any language. | |
3759 | */ | |
3760 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3761 | return 0; |
1da177e4 LT |
3762 | iclog = iclog->ic_next; |
3763 | } while (iclog != log->l_iclog); | |
014c2544 | 3764 | return 1; |
1da177e4 | 3765 | } |
f661f1e0 | 3766 |