Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 : * All Rights Reserved.
5 : */
6 : #include "xfs.h"
7 : #include "xfs_fs.h"
8 : #include "xfs_shared.h"
9 : #include "xfs_format.h"
10 : #include "xfs_log_format.h"
11 : #include "xfs_trans_resv.h"
12 : #include "xfs_mount.h"
13 : #include "xfs_errortag.h"
14 : #include "xfs_error.h"
15 : #include "xfs_trans.h"
16 : #include "xfs_trans_priv.h"
17 : #include "xfs_log.h"
18 : #include "xfs_log_priv.h"
19 : #include "xfs_trace.h"
20 : #include "xfs_sysfs.h"
21 : #include "xfs_sb.h"
22 : #include "xfs_health.h"
23 :
24 : struct kmem_cache *xfs_log_ticket_cache;
25 :
26 : /* Local miscellaneous function prototypes */
27 : STATIC struct xlog *
28 : xlog_alloc_log(
29 : struct xfs_mount *mp,
30 : struct xfs_buftarg *log_target,
31 : xfs_daddr_t blk_offset,
32 : int num_bblks);
33 : STATIC int
34 : xlog_space_left(
35 : struct xlog *log,
36 : atomic64_t *head);
37 : STATIC void
38 : xlog_dealloc_log(
39 : struct xlog *log);
40 :
41 : /* local state machine functions */
42 : STATIC void xlog_state_done_syncing(
43 : struct xlog_in_core *iclog);
44 : STATIC void xlog_state_do_callback(
45 : struct xlog *log);
46 : STATIC int
47 : xlog_state_get_iclog_space(
48 : struct xlog *log,
49 : int len,
50 : struct xlog_in_core **iclog,
51 : struct xlog_ticket *ticket,
52 : int *logoffsetp);
53 : STATIC void
54 : xlog_grant_push_ail(
55 : struct xlog *log,
56 : int need_bytes);
57 : STATIC void
58 : xlog_sync(
59 : struct xlog *log,
60 : struct xlog_in_core *iclog,
61 : struct xlog_ticket *ticket);
62 : #if defined(DEBUG)
63 : STATIC void
64 : xlog_verify_grant_tail(
65 : struct xlog *log);
66 : STATIC void
67 : xlog_verify_iclog(
68 : struct xlog *log,
69 : struct xlog_in_core *iclog,
70 : int count);
71 : STATIC void
72 : xlog_verify_tail_lsn(
73 : struct xlog *log,
74 : struct xlog_in_core *iclog);
75 : #else
76 : #define xlog_verify_grant_tail(a)
77 : #define xlog_verify_iclog(a,b,c)
78 : #define xlog_verify_tail_lsn(a,b)
79 : #endif
80 :
81 : STATIC int
82 : xlog_iclogs_empty(
83 : struct xlog *log);
84 :
85 : static int
86 : xfs_log_cover(struct xfs_mount *);
87 :
88 : /*
89 : * We need to make sure the buffer pointer returned is naturally aligned for the
90 : * biggest basic data type we put into it. We have already accounted for this
91 : * padding when sizing the buffer.
92 : *
93 : * However, this padding does not get written into the log, and hence we have to
94 : * track the space used by the log vectors separately to prevent log space hangs
95 : * due to inaccurate accounting (i.e. a leak) of the used log space through the
96 : * CIL context ticket.
97 : *
98 : * We also add space for the xlog_op_header that describes this region in the
99 : * log. This prepends the data region we return to the caller to copy their data
100 : * into, so do all the static initialisation of the ophdr now. Because the ophdr
101 : * is not 8 byte aligned, we have to be careful to ensure that we align the
102 : * start of the buffer such that the region we return to the call is 8 byte
103 : * aligned and packed against the tail of the ophdr.
104 : */
105 : void *
106 18176211333 : xlog_prepare_iovec(
107 : struct xfs_log_vec *lv,
108 : struct xfs_log_iovec **vecp,
109 : uint type)
110 : {
111 18176211333 : struct xfs_log_iovec *vec = *vecp;
112 18176211333 : struct xlog_op_header *oph;
113 18176211333 : uint32_t len;
114 18176211333 : void *buf;
115 :
116 18176211333 : if (vec) {
117 10168465281 : ASSERT(vec - lv->lv_iovecp < lv->lv_niovecs);
118 10168465281 : vec++;
119 : } else {
120 8007746052 : vec = &lv->lv_iovecp[0];
121 : }
122 :
123 18176211333 : len = lv->lv_buf_len + sizeof(struct xlog_op_header);
124 18176211333 : if (!IS_ALIGNED(len, sizeof(uint64_t))) {
125 18096840587 : lv->lv_buf_len = round_up(len, sizeof(uint64_t)) -
126 : sizeof(struct xlog_op_header);
127 : }
128 :
129 18176211333 : vec->i_type = type;
130 18176211333 : vec->i_addr = lv->lv_buf + lv->lv_buf_len;
131 :
132 18176211333 : oph = vec->i_addr;
133 18176211333 : oph->oh_clientid = XFS_TRANSACTION;
134 18176211333 : oph->oh_res2 = 0;
135 18176211333 : oph->oh_flags = 0;
136 :
137 18176211333 : buf = vec->i_addr + sizeof(struct xlog_op_header);
138 18176211333 : ASSERT(IS_ALIGNED((unsigned long)buf, sizeof(uint64_t)));
139 :
140 18176211333 : *vecp = vec;
141 18176211333 : return buf;
142 : }
143 :
144 : static void
145 4465307897 : xlog_grant_sub_space(
146 : struct xlog *log,
147 : atomic64_t *head,
148 : int bytes)
149 : {
150 4465307897 : int64_t head_val = atomic64_read(head);
151 4469252142 : int64_t new, old;
152 :
153 4469252142 : do {
154 4469252142 : int cycle, space;
155 :
156 4469252142 : xlog_crack_grant_head_val(head_val, &cycle, &space);
157 :
158 4469252142 : space -= bytes;
159 4469252142 : if (space < 0) {
160 35218130 : space += log->l_logsize;
161 35218130 : cycle--;
162 : }
163 :
164 4469252142 : old = head_val;
165 4469252142 : new = xlog_assign_grant_head_val(cycle, space);
166 4469252142 : head_val = atomic64_cmpxchg(head, old, new);
167 4470639200 : } while (head_val != old);
168 4466694955 : }
169 :
170 : static void
171 3387955543 : xlog_grant_add_space(
172 : struct xlog *log,
173 : atomic64_t *head,
174 : int bytes)
175 : {
176 3387955543 : int64_t head_val = atomic64_read(head);
177 3394743478 : int64_t new, old;
178 :
179 3394743478 : do {
180 3394743478 : int tmp;
181 3394743478 : int cycle, space;
182 :
183 3394743478 : xlog_crack_grant_head_val(head_val, &cycle, &space);
184 :
185 3394743478 : tmp = log->l_logsize - space;
186 3394743478 : if (tmp > bytes)
187 3359492833 : space += bytes;
188 : else {
189 35250645 : space = bytes - tmp;
190 35250645 : cycle++;
191 : }
192 :
193 3394743478 : old = head_val;
194 3394743478 : new = xlog_assign_grant_head_val(cycle, space);
195 3394743478 : head_val = atomic64_cmpxchg(head, old, new);
196 3397965368 : } while (head_val != old);
197 3391177433 : }
198 :
199 : STATIC void
200 118578 : xlog_grant_head_init(
201 : struct xlog_grant_head *head)
202 : {
203 118578 : xlog_assign_grant_head(&head->grant, 1, 0);
204 118578 : INIT_LIST_HEAD(&head->waiters);
205 118578 : spin_lock_init(&head->lock);
206 118578 : }
207 :
208 : STATIC void
209 26256 : xlog_grant_head_wake_all(
210 : struct xlog_grant_head *head)
211 : {
212 26256 : struct xlog_ticket *tic;
213 :
214 26256 : spin_lock(&head->lock);
215 26256 : list_for_each_entry(tic, &head->waiters, t_queue)
216 0 : wake_up_process(tic->t_task);
217 26256 : spin_unlock(&head->lock);
218 26256 : }
219 :
220 : static inline int
221 2096508637 : xlog_ticket_reservation(
222 : struct xlog *log,
223 : struct xlog_grant_head *head,
224 : struct xlog_ticket *tic)
225 : {
226 2096508637 : if (head == &log->l_write_head) {
227 355092408 : ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
228 355092408 : return tic->t_unit_res;
229 : }
230 :
231 1741416229 : if (tic->t_flags & XLOG_TIC_PERM_RESERV)
232 1318886098 : return tic->t_unit_res * tic->t_cnt;
233 :
234 422530131 : return tic->t_unit_res;
235 : }
236 :
237 : STATIC bool
238 11338291 : xlog_grant_head_wake(
239 : struct xlog *log,
240 : struct xlog_grant_head *head,
241 : int *free_bytes)
242 : {
243 11338291 : struct xlog_ticket *tic;
244 11338291 : int need_bytes;
245 11338291 : bool woken_task = false;
246 :
247 410437924 : list_for_each_entry(tic, &head->waiters, t_queue) {
248 :
249 : /*
250 : * There is a chance that the size of the CIL checkpoints in
251 : * progress at the last AIL push target calculation resulted in
252 : * limiting the target to the log head (l_last_sync_lsn) at the
253 : * time. This may not reflect where the log head is now as the
254 : * CIL checkpoints may have completed.
255 : *
256 : * Hence when we are woken here, it may be that the head of the
257 : * log that has moved rather than the tail. As the tail didn't
258 : * move, there still won't be space available for the
259 : * reservation we require. However, if the AIL has already
260 : * pushed to the target defined by the old log head location, we
261 : * will hang here waiting for something else to update the AIL
262 : * push target.
263 : *
264 : * Therefore, if there isn't space to wake the first waiter on
265 : * the grant head, we need to push the AIL again to ensure the
266 : * target reflects both the current log tail and log head
267 : * position before we wait for the tail to move again.
268 : */
269 :
270 408481924 : need_bytes = xlog_ticket_reservation(log, head, tic);
271 408481924 : if (*free_bytes < need_bytes) {
272 9382291 : if (!woken_task)
273 600690 : xlog_grant_push_ail(log, need_bytes);
274 9382291 : return false;
275 : }
276 :
277 399099633 : *free_bytes -= need_bytes;
278 399099633 : trace_xfs_log_grant_wake_up(log, tic);
279 399099633 : wake_up_process(tic->t_task);
280 399099633 : woken_task = true;
281 : }
282 :
283 : return true;
284 : }
285 :
286 : STATIC int
287 4911129 : xlog_grant_head_wait(
288 : struct xlog *log,
289 : struct xlog_grant_head *head,
290 : struct xlog_ticket *tic,
291 : int need_bytes) __releases(&head->lock)
292 : __acquires(&head->lock)
293 : {
294 4911129 : list_add_tail(&tic->t_queue, &head->waiters);
295 :
296 4914073 : do {
297 9828146 : if (xlog_is_shutdown(log))
298 0 : goto shutdown;
299 4914073 : xlog_grant_push_ail(log, need_bytes);
300 :
301 4914073 : __set_current_state(TASK_UNINTERRUPTIBLE);
302 4914073 : spin_unlock(&head->lock);
303 :
304 4914040 : XFS_STATS_INC(log->l_mp, xs_sleep_logspace);
305 :
306 4914038 : trace_xfs_log_grant_sleep(log, tic);
307 4913959 : schedule();
308 4910283 : trace_xfs_log_grant_wake(log, tic);
309 :
310 4905221 : spin_lock(&head->lock);
311 9828146 : if (xlog_is_shutdown(log))
312 0 : goto shutdown;
313 4914073 : } while (xlog_space_left(log, &head->grant) < need_bytes);
314 :
315 4911129 : list_del_init(&tic->t_queue);
316 4911129 : return 0;
317 0 : shutdown:
318 0 : list_del_init(&tic->t_queue);
319 0 : return -EIO;
320 : }
321 :
322 : /*
323 : * Atomically get the log space required for a log ticket.
324 : *
325 : * Once a ticket gets put onto head->waiters, it will only return after the
326 : * needed reservation is satisfied.
327 : *
328 : * This function is structured so that it has a lock free fast path. This is
329 : * necessary because every new transaction reservation will come through this
330 : * path. Hence any lock will be globally hot if we take it unconditionally on
331 : * every pass.
332 : *
333 : * As tickets are only ever moved on and off head->waiters under head->lock, we
334 : * only need to take that lock if we are going to add the ticket to the queue
335 : * and sleep. We can avoid taking the lock if the ticket was never added to
336 : * head->waiters because the t_queue list head will be empty and we hold the
337 : * only reference to it so it can safely be checked unlocked.
338 : */
339 : STATIC int
340 1689211772 : xlog_grant_head_check(
341 : struct xlog *log,
342 : struct xlog_grant_head *head,
343 : struct xlog_ticket *tic,
344 : int *need_bytes)
345 : {
346 1689211772 : int free_bytes;
347 1689211772 : int error = 0;
348 :
349 3378423544 : ASSERT(!xlog_in_recovery(log));
350 :
351 : /*
352 : * If there are other waiters on the queue then give them a chance at
353 : * logspace before us. Wake up the first waiters, if we do not wake
354 : * up all the waiters then go to sleep waiting for more free space,
355 : * otherwise try to get some space for this transaction.
356 : */
357 1689211772 : *need_bytes = xlog_ticket_reservation(log, head, tic);
358 1687899681 : free_bytes = xlog_space_left(log, &head->grant);
359 1688197399 : if (!list_empty_careful(&head->waiters)) {
360 5269238 : spin_lock(&head->lock);
361 5560249 : if (!xlog_grant_head_wake(log, head, &free_bytes) ||
362 763208 : free_bytes < *need_bytes) {
363 4875730 : error = xlog_grant_head_wait(log, head, tic,
364 : *need_bytes);
365 : }
366 5560249 : spin_unlock(&head->lock);
367 1683586293 : } else if (free_bytes < *need_bytes) {
368 35074 : spin_lock(&head->lock);
369 35399 : error = xlog_grant_head_wait(log, head, tic, *need_bytes);
370 35399 : spin_unlock(&head->lock);
371 : }
372 :
373 1689146503 : return error;
374 : }
375 :
376 : bool
377 186407 : xfs_log_writable(
378 : struct xfs_mount *mp)
379 : {
380 : /*
381 : * Do not write to the log on norecovery mounts, if the data or log
382 : * devices are read-only, or if the filesystem is shutdown. Read-only
383 : * mounts allow internal writes for log recovery and unmount purposes,
384 : * so don't restrict that case.
385 : */
386 186407 : if (xfs_has_norecovery(mp))
387 : return false;
388 186299 : if (xfs_readonly_buftarg(mp->m_ddev_targp))
389 : return false;
390 186269 : if (xfs_readonly_buftarg(mp->m_log->l_targ))
391 : return false;
392 372538 : if (xlog_is_shutdown(mp->m_log))
393 26221 : return false;
394 : return true;
395 : }
396 :
397 : /*
398 : * Replenish the byte reservation required by moving the grant write head.
399 : */
400 : int
401 891546458 : xfs_log_regrant(
402 : struct xfs_mount *mp,
403 : struct xlog_ticket *tic)
404 : {
405 891546458 : struct xlog *log = mp->m_log;
406 891546458 : int need_bytes;
407 891546458 : int error = 0;
408 :
409 1783092916 : if (xlog_is_shutdown(log))
410 : return -EIO;
411 :
412 891546428 : XFS_STATS_INC(mp, xs_try_logspace);
413 :
414 : /*
415 : * This is a new transaction on the ticket, so we need to change the
416 : * transaction ID so that the next transaction has a different TID in
417 : * the log. Just add one to the existing tid so that we can see chains
418 : * of rolling transactions in the log easily.
419 : */
420 891525082 : tic->t_tid++;
421 :
422 891525082 : xlog_grant_push_ail(log, tic->t_unit_res);
423 :
424 891546219 : tic->t_curr_res = tic->t_unit_res;
425 891546219 : if (tic->t_cnt > 0)
426 : return 0;
427 :
428 355105365 : trace_xfs_log_regrant(log, tic);
429 :
430 355093813 : error = xlog_grant_head_check(log, &log->l_write_head, tic,
431 : &need_bytes);
432 355102426 : if (error)
433 0 : goto out_error;
434 :
435 355102426 : xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
436 355118553 : trace_xfs_log_regrant_exit(log, tic);
437 355115491 : xlog_verify_grant_tail(log);
438 355115491 : return 0;
439 :
440 : out_error:
441 : /*
442 : * If we are failing, make sure the ticket doesn't have any current
443 : * reservations. We don't want to add this back when the ticket/
444 : * transaction gets cancelled.
445 : */
446 0 : tic->t_curr_res = 0;
447 0 : tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
448 0 : return error;
449 : }
450 :
451 : /*
452 : * Reserve log space and return a ticket corresponding to the reservation.
453 : *
454 : * Each reservation is going to reserve extra space for a log record header.
455 : * When writes happen to the on-disk log, we don't subtract the length of the
456 : * log record header from any reservation. By wasting space in each
457 : * reservation, we prevent over allocation problems.
458 : */
459 : int
460 1333996338 : xfs_log_reserve(
461 : struct xfs_mount *mp,
462 : int unit_bytes,
463 : int cnt,
464 : struct xlog_ticket **ticp,
465 : bool permanent)
466 : {
467 1333996338 : struct xlog *log = mp->m_log;
468 1333996338 : struct xlog_ticket *tic;
469 1333996338 : int need_bytes;
470 1333996338 : int error = 0;
471 :
472 2667992676 : if (xlog_is_shutdown(log))
473 : return -EIO;
474 :
475 1333993747 : XFS_STATS_INC(mp, xs_try_logspace);
476 :
477 1333680141 : ASSERT(*ticp == NULL);
478 1333680141 : tic = xlog_ticket_alloc(log, unit_bytes, cnt, permanent);
479 1334050978 : *ticp = tic;
480 :
481 1334050978 : xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt
482 : : tic->t_unit_res);
483 :
484 1333952268 : trace_xfs_log_reserve(log, tic);
485 :
486 1333571533 : error = xlog_grant_head_check(log, &log->l_reserve_head, tic,
487 : &need_bytes);
488 1334067847 : if (error)
489 0 : goto out_error;
490 :
491 1334067847 : xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes);
492 1336528561 : xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
493 1336744559 : trace_xfs_log_reserve_exit(log, tic);
494 1336337039 : xlog_verify_grant_tail(log);
495 1336337039 : return 0;
496 :
497 : out_error:
498 : /*
499 : * If we are failing, make sure the ticket doesn't have any current
500 : * reservations. We don't want to add this back when the ticket/
501 : * transaction gets cancelled.
502 : */
503 0 : tic->t_curr_res = 0;
504 0 : tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
505 0 : return error;
506 : }
507 :
508 : /*
509 : * Run all the pending iclog callbacks and wake log force waiters and iclog
510 : * space waiters so they can process the newly set shutdown state. We really
511 : * don't care what order we process callbacks here because the log is shut down
512 : * and so state cannot change on disk anymore. However, we cannot wake waiters
513 : * until the callbacks have been processed because we may be in unmount and
514 : * we must ensure that all AIL operations the callbacks perform have completed
515 : * before we tear down the AIL.
516 : *
517 : * We avoid processing actively referenced iclogs so that we don't run callbacks
518 : * while the iclog owner might still be preparing the iclog for IO submssion.
519 : * These will be caught by xlog_state_iclog_release() and call this function
520 : * again to process any callbacks that may have been added to that iclog.
521 : */
522 : static void
523 15411 : xlog_state_shutdown_callbacks(
524 : struct xlog *log)
525 : {
526 15411 : struct xlog_in_core *iclog;
527 15411 : LIST_HEAD(cb_list);
528 :
529 15411 : iclog = log->l_iclog;
530 123288 : do {
531 123288 : if (atomic_read(&iclog->ic_refcnt)) {
532 : /* Reference holder will re-run iclog callbacks. */
533 2283 : continue;
534 : }
535 121005 : list_splice_init(&iclog->ic_callbacks, &cb_list);
536 121005 : spin_unlock(&log->l_icloglock);
537 :
538 121005 : xlog_cil_process_committed(&cb_list);
539 :
540 121005 : spin_lock(&log->l_icloglock);
541 121005 : wake_up_all(&iclog->ic_write_wait);
542 121005 : wake_up_all(&iclog->ic_force_wait);
543 123288 : } while ((iclog = iclog->ic_next) != log->l_iclog);
544 :
545 15411 : wake_up_all(&log->l_flush_wait);
546 15411 : }
547 :
548 : /*
549 : * Flush iclog to disk if this is the last reference to the given iclog and the
550 : * it is in the WANT_SYNC state.
551 : *
552 : * If XLOG_ICL_NEED_FUA is already set on the iclog, we need to ensure that the
553 : * log tail is updated correctly. NEED_FUA indicates that the iclog will be
554 : * written to stable storage, and implies that a commit record is contained
555 : * within the iclog. We need to ensure that the log tail does not move beyond
556 : * the tail that the first commit record in the iclog ordered against, otherwise
557 : * correct recovery of that checkpoint becomes dependent on future operations
558 : * performed on this iclog.
559 : *
560 : * Hence if NEED_FUA is set and the current iclog tail lsn is empty, write the
561 : * current tail into iclog. Once the iclog tail is set, future operations must
562 : * not modify it, otherwise they potentially violate ordering constraints for
563 : * the checkpoint commit that wrote the initial tail lsn value. The tail lsn in
564 : * the iclog will get zeroed on activation of the iclog after sync, so we
565 : * always capture the tail lsn on the iclog on the first NEED_FUA release
566 : * regardless of the number of active reference counts on this iclog.
567 : */
568 : int
569 36119341 : xlog_state_release_iclog(
570 : struct xlog *log,
571 : struct xlog_in_core *iclog,
572 : struct xlog_ticket *ticket)
573 : {
574 36119341 : xfs_lsn_t tail_lsn;
575 36119341 : bool last_ref;
576 :
577 36119341 : lockdep_assert_held(&log->l_icloglock);
578 :
579 36119341 : trace_xlog_iclog_release(iclog, _RET_IP_);
580 : /*
581 : * Grabbing the current log tail needs to be atomic w.r.t. the writing
582 : * of the tail LSN into the iclog so we guarantee that the log tail does
583 : * not move between the first time we know that the iclog needs to be
584 : * made stable and when we eventually submit it.
585 : */
586 36119341 : if ((iclog->ic_state == XLOG_STATE_WANT_SYNC ||
587 14913113 : (iclog->ic_flags & XLOG_ICL_NEED_FUA)) &&
588 25915987 : !iclog->ic_header.h_tail_lsn) {
589 21166947 : tail_lsn = xlog_assign_tail_lsn(log->l_mp);
590 21166946 : iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
591 : }
592 :
593 36119340 : last_ref = atomic_dec_and_test(&iclog->ic_refcnt);
594 :
595 72238690 : if (xlog_is_shutdown(log)) {
596 : /*
597 : * If there are no more references to this iclog, process the
598 : * pending iclog callbacks that were waiting on the release of
599 : * this iclog.
600 : */
601 2283 : if (last_ref)
602 2283 : xlog_state_shutdown_callbacks(log);
603 2283 : return -EIO;
604 : }
605 :
606 36117062 : if (!last_ref)
607 : return 0;
608 :
609 30973106 : if (iclog->ic_state != XLOG_STATE_WANT_SYNC) {
610 9808491 : ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
611 9808491 : return 0;
612 : }
613 :
614 21164615 : iclog->ic_state = XLOG_STATE_SYNCING;
615 21164615 : xlog_verify_tail_lsn(log, iclog);
616 21164614 : trace_xlog_iclog_syncing(iclog, _RET_IP_);
617 :
618 21164612 : spin_unlock(&log->l_icloglock);
619 21164616 : xlog_sync(log, iclog, ticket);
620 21164572 : spin_lock(&log->l_icloglock);
621 21164572 : return 0;
622 : }
623 :
624 : /*
625 : * Mount a log filesystem
626 : *
627 : * mp - ubiquitous xfs mount point structure
628 : * log_target - buftarg of on-disk log device
629 : * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
630 : * num_bblocks - Number of BBSIZE blocks in on-disk log
631 : *
632 : * Return error or zero.
633 : */
634 : int
635 59289 : xfs_log_mount(
636 : xfs_mount_t *mp,
637 : xfs_buftarg_t *log_target,
638 : xfs_daddr_t blk_offset,
639 : int num_bblks)
640 : {
641 59289 : struct xlog *log;
642 59289 : int error = 0;
643 59289 : int min_logfsbs;
644 :
645 59289 : if (!xfs_has_norecovery(mp)) {
646 59235 : xfs_notice(mp, "Mounting V%d Filesystem %pU",
647 : XFS_SB_VERSION_NUM(&mp->m_sb),
648 : &mp->m_sb.sb_uuid);
649 : } else {
650 54 : xfs_notice(mp,
651 : "Mounting V%d filesystem %pU in no-recovery mode. Filesystem will be inconsistent.",
652 : XFS_SB_VERSION_NUM(&mp->m_sb),
653 : &mp->m_sb.sb_uuid);
654 108 : ASSERT(xfs_is_readonly(mp));
655 : }
656 :
657 59289 : log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
658 59289 : if (IS_ERR(log)) {
659 0 : error = PTR_ERR(log);
660 0 : goto out;
661 : }
662 59289 : mp->m_log = log;
663 :
664 : /*
665 : * Now that we have set up the log and it's internal geometry
666 : * parameters, we can validate the given log space and drop a critical
667 : * message via syslog if the log size is too small. A log that is too
668 : * small can lead to unexpected situations in transaction log space
669 : * reservation stage. The superblock verifier has already validated all
670 : * the other log geometry constraints, so we don't have to check those
671 : * here.
672 : *
673 : * Note: For v4 filesystems, we can't just reject the mount if the
674 : * validation fails. This would mean that people would have to
675 : * downgrade their kernel just to remedy the situation as there is no
676 : * way to grow the log (short of black magic surgery with xfs_db).
677 : *
678 : * We can, however, reject mounts for V5 format filesystems, as the
679 : * mkfs binary being used to make the filesystem should never create a
680 : * filesystem with a log that is too small.
681 : */
682 59289 : min_logfsbs = xfs_log_calc_minimum_size(mp);
683 59289 : if (mp->m_sb.sb_logblocks < min_logfsbs) {
684 0 : xfs_warn(mp,
685 : "Log size %d blocks too small, minimum size is %d blocks",
686 : mp->m_sb.sb_logblocks, min_logfsbs);
687 :
688 : /*
689 : * Log check errors are always fatal on v5; or whenever bad
690 : * metadata leads to a crash.
691 : */
692 0 : if (xfs_has_crc(mp)) {
693 0 : xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!");
694 0 : ASSERT(0);
695 0 : error = -EINVAL;
696 0 : goto out_free_log;
697 : }
698 0 : xfs_crit(mp, "Log size out of supported range.");
699 0 : xfs_crit(mp,
700 : "Continuing onwards, but if log hangs are experienced then please report this message in the bug report.");
701 : }
702 :
703 : /*
704 : * Initialize the AIL now we have a log.
705 : */
706 59289 : error = xfs_trans_ail_init(mp);
707 59289 : if (error) {
708 0 : xfs_warn(mp, "AIL initialisation failed: error %d", error);
709 0 : goto out_free_log;
710 : }
711 59289 : log->l_ailp = mp->m_ail;
712 :
713 : /*
714 : * skip log recovery on a norecovery mount. pretend it all
715 : * just worked.
716 : */
717 59289 : if (!xfs_has_norecovery(mp)) {
718 : /*
719 : * log recovery ignores readonly state and so we need to clear
720 : * mount-based read only state so it can write to disk.
721 : */
722 59235 : bool readonly = test_and_clear_bit(XFS_OPSTATE_READONLY,
723 59235 : &mp->m_opstate);
724 59235 : error = xlog_recover(log);
725 59235 : if (readonly)
726 2177 : set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
727 59235 : if (error) {
728 42 : xfs_warn(mp, "log mount/recovery failed: error %d",
729 : error);
730 42 : xlog_recover_cancel(log);
731 42 : goto out_destroy_ail;
732 : }
733 : }
734 :
735 59247 : error = xfs_sysfs_init(&log->l_kobj, &xfs_log_ktype, &mp->m_kobj,
736 : "log");
737 59247 : if (error)
738 0 : goto out_destroy_ail;
739 :
740 : /* Normal transactions can now occur */
741 59247 : clear_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);
742 :
743 : /*
744 : * Now the log has been fully initialised and we know were our
745 : * space grant counters are, we can initialise the permanent ticket
746 : * needed for delayed logging to work.
747 : */
748 59247 : xlog_cil_init_post_recovery(log);
749 :
750 59247 : return 0;
751 :
752 42 : out_destroy_ail:
753 42 : xfs_trans_ail_destroy(mp);
754 42 : out_free_log:
755 42 : xlog_dealloc_log(log);
756 : out:
757 : return error;
758 : }
759 :
760 : /*
761 : * Finish the recovery of the file system. This is separate from the
762 : * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read
763 : * in the root and real-time bitmap inodes between calling xfs_log_mount() and
764 : * here.
765 : *
766 : * If we finish recovery successfully, start the background log work. If we are
767 : * not doing recovery, then we have a RO filesystem and we don't need to start
768 : * it.
769 : */
770 : int
771 59202 : xfs_log_mount_finish(
772 : struct xfs_mount *mp)
773 : {
774 59202 : struct xlog *log = mp->m_log;
775 59202 : bool readonly;
776 59202 : int error = 0;
777 :
778 59202 : if (xfs_has_norecovery(mp)) {
779 94 : ASSERT(xfs_is_readonly(mp));
780 47 : return 0;
781 : }
782 :
783 : /*
784 : * log recovery ignores readonly state and so we need to clear
785 : * mount-based read only state so it can write to disk.
786 : */
787 59155 : readonly = test_and_clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
788 :
789 : /*
790 : * During the second phase of log recovery, we need iget and
791 : * iput to behave like they do for an active filesystem.
792 : * xfs_fs_drop_inode needs to be able to prevent the deletion
793 : * of inodes before we're done replaying log items on those
794 : * inodes. Turn it off immediately after recovery finishes
795 : * so that we don't leak the quota inodes if subsequent mount
796 : * activities fail.
797 : *
798 : * We let all inodes involved in redo item processing end up on
799 : * the LRU instead of being evicted immediately so that if we do
800 : * something to an unlinked inode, the irele won't cause
801 : * premature truncation and freeing of the inode, which results
802 : * in log recovery failure. We have to evict the unreferenced
803 : * lru inodes after clearing SB_ACTIVE because we don't
804 : * otherwise clean up the lru if there's a subsequent failure in
805 : * xfs_mountfs, which leads to us leaking the inodes if nothing
806 : * else (e.g. quotacheck) references the inodes before the
807 : * mount failure occurs.
808 : */
809 59155 : mp->m_super->s_flags |= SB_ACTIVE;
810 59155 : xfs_log_work_queue(mp);
811 118310 : if (xlog_recovery_needed(log))
812 13723 : error = xlog_recover_finish(log);
813 59155 : mp->m_super->s_flags &= ~SB_ACTIVE;
814 59155 : evict_inodes(mp->m_super);
815 :
816 : /*
817 : * Drain the buffer LRU after log recovery. This is required for v4
818 : * filesystems to avoid leaving around buffers with NULL verifier ops,
819 : * but we do it unconditionally to make sure we're always in a clean
820 : * cache state after mount.
821 : *
822 : * Don't push in the error case because the AIL may have pending intents
823 : * that aren't removed until recovery is cancelled.
824 : */
825 118310 : if (xlog_recovery_needed(log)) {
826 13723 : if (!error) {
827 13717 : xfs_log_force(mp, XFS_LOG_SYNC);
828 13717 : xfs_ail_push_all_sync(mp->m_ail);
829 : }
830 15349 : xfs_notice(mp, "Ending recovery (logdev: %s)",
831 : mp->m_logname ? mp->m_logname : "internal");
832 : } else {
833 45432 : xfs_info(mp, "Ending clean mount");
834 : }
835 59155 : xfs_buftarg_drain(mp->m_ddev_targp);
836 :
837 59155 : clear_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate);
838 59155 : if (readonly)
839 2159 : set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
840 :
841 : /* Make sure the log is dead if we're returning failure. */
842 59161 : ASSERT(!error || xlog_is_shutdown(log));
843 :
844 : return error;
845 : }
846 :
847 : /*
848 : * The mount has failed. Cancel the recovery if it hasn't completed and destroy
849 : * the log.
850 : */
851 : void
852 148 : xfs_log_mount_cancel(
853 : struct xfs_mount *mp)
854 : {
855 148 : xlog_recover_cancel(mp->m_log);
856 148 : xfs_log_unmount(mp);
857 148 : }
858 :
859 : /*
860 : * Flush out the iclog to disk ensuring that device caches are flushed and
861 : * the iclog hits stable storage before any completion waiters are woken.
862 : */
863 : static inline int
864 4661619 : xlog_force_iclog(
865 : struct xlog_in_core *iclog)
866 : {
867 4661619 : atomic_inc(&iclog->ic_refcnt);
868 4661621 : iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA;
869 4661621 : if (iclog->ic_state == XLOG_STATE_ACTIVE)
870 4661621 : xlog_state_switch_iclogs(iclog->ic_log, iclog, 0);
871 4661620 : return xlog_state_release_iclog(iclog->ic_log, iclog, NULL);
872 : }
873 :
874 : /*
875 : * Cycle all the iclogbuf locks to make sure all log IO completion
876 : * is done before we tear down these buffers.
877 : */
878 : static void
879 59254 : xlog_wait_iclog_completion(struct xlog *log)
880 : {
881 59254 : int i;
882 59254 : struct xlog_in_core *iclog = log->l_iclog;
883 :
884 533226 : for (i = 0; i < log->l_iclog_bufs; i++) {
885 473972 : down(&iclog->ic_sema);
886 473972 : up(&iclog->ic_sema);
887 473972 : iclog = iclog->ic_next;
888 : }
889 59254 : }
890 :
891 : /*
892 : * Wait for the iclog and all prior iclogs to be written disk as required by the
893 : * log force state machine. Waiting on ic_force_wait ensures iclog completions
894 : * have been ordered and callbacks run before we are woken here, hence
895 : * guaranteeing that all the iclogs up to this one are on stable storage.
896 : */
897 : int
898 8939147 : xlog_wait_on_iclog(
899 : struct xlog_in_core *iclog)
900 : __releases(iclog->ic_log->l_icloglock)
901 : {
902 8939147 : struct xlog *log = iclog->ic_log;
903 :
904 8939147 : trace_xlog_iclog_wait_on(iclog, _RET_IP_);
905 17878284 : if (!xlog_is_shutdown(log) &&
906 8939076 : iclog->ic_state != XLOG_STATE_ACTIVE &&
907 : iclog->ic_state != XLOG_STATE_DIRTY) {
908 7276278 : XFS_STATS_INC(log->l_mp, xs_log_force_sleep);
909 7276277 : xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
910 : } else {
911 1662864 : spin_unlock(&log->l_icloglock);
912 : }
913 :
914 17876344 : if (xlog_is_shutdown(log))
915 5229 : return -EIO;
916 : return 0;
917 : }
918 :
919 : /*
920 : * Write out an unmount record using the ticket provided. We have to account for
921 : * the data space used in the unmount ticket as this write is not done from a
922 : * transaction context that has already done the accounting for us.
923 : */
924 : static int
925 48695 : xlog_write_unmount_record(
926 : struct xlog *log,
927 : struct xlog_ticket *ticket)
928 : {
929 48695 : struct {
930 : struct xlog_op_header ophdr;
931 : struct xfs_unmount_log_format ulf;
932 48695 : } unmount_rec = {
933 : .ophdr = {
934 : .oh_clientid = XFS_LOG,
935 48695 : .oh_tid = cpu_to_be32(ticket->t_tid),
936 : .oh_flags = XLOG_UNMOUNT_TRANS,
937 : },
938 : .ulf = {
939 : .magic = XLOG_UNMOUNT_TYPE,
940 : },
941 : };
942 48695 : struct xfs_log_iovec reg = {
943 : .i_addr = &unmount_rec,
944 : .i_len = sizeof(unmount_rec),
945 : .i_type = XLOG_REG_TYPE_UNMOUNT,
946 : };
947 48695 : struct xfs_log_vec vec = {
948 : .lv_niovecs = 1,
949 : .lv_iovecp = ®,
950 : };
951 48695 : LIST_HEAD(lv_chain);
952 48695 : list_add(&vec.lv_list, &lv_chain);
953 :
954 48695 : BUILD_BUG_ON((sizeof(struct xlog_op_header) +
955 : sizeof(struct xfs_unmount_log_format)) !=
956 : sizeof(unmount_rec));
957 :
958 : /* account for space used by record data */
959 48695 : ticket->t_curr_res -= sizeof(unmount_rec);
960 :
961 48695 : return xlog_write(log, NULL, &lv_chain, ticket, reg.i_len);
962 : }
963 :
964 : /*
965 : * Mark the filesystem clean by writing an unmount record to the head of the
966 : * log.
967 : */
968 : static void
969 48695 : xlog_unmount_write(
970 : struct xlog *log)
971 : {
972 48695 : struct xfs_mount *mp = log->l_mp;
973 48695 : struct xlog_in_core *iclog;
974 48695 : struct xlog_ticket *tic = NULL;
975 48695 : int error;
976 :
977 48695 : error = xfs_log_reserve(mp, 600, 1, &tic, 0);
978 48695 : if (error)
979 0 : goto out_err;
980 :
981 48695 : error = xlog_write_unmount_record(log, tic);
982 : /*
983 : * At this point, we're umounting anyway, so there's no point in
984 : * transitioning log state to shutdown. Just continue...
985 : */
986 48695 : out_err:
987 48695 : if (error)
988 0 : xfs_alert(mp, "%s: unmount record failed", __func__);
989 :
990 48695 : spin_lock(&log->l_icloglock);
991 48695 : iclog = log->l_iclog;
992 48695 : error = xlog_force_iclog(iclog);
993 48695 : xlog_wait_on_iclog(iclog);
994 :
995 48695 : if (tic) {
996 48695 : trace_xfs_log_umount_write(log, tic);
997 48695 : xfs_log_ticket_ungrant(log, tic);
998 : }
999 48695 : }
1000 :
1001 : static void
1002 48695 : xfs_log_unmount_verify_iclog(
1003 : struct xlog *log)
1004 : {
1005 48695 : struct xlog_in_core *iclog = log->l_iclog;
1006 :
1007 389500 : do {
1008 389500 : ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
1009 389500 : ASSERT(iclog->ic_offset == 0);
1010 389500 : } while ((iclog = iclog->ic_next) != log->l_iclog);
1011 48695 : }
1012 :
1013 : /*
1014 : * Unmount record used to have a string "Unmount filesystem--" in the
1015 : * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
1016 : * We just write the magic number now since that particular field isn't
1017 : * currently architecture converted and "Unmount" is a bit foo.
1018 : * As far as I know, there weren't any dependencies on the old behaviour.
1019 : */
1020 : static void
1021 61986 : xfs_log_unmount_write(
1022 : struct xfs_mount *mp)
1023 : {
1024 61986 : struct xlog *log = mp->m_log;
1025 :
1026 61986 : if (!xfs_log_writable(mp))
1027 : return;
1028 :
1029 48778 : xfs_log_force(mp, XFS_LOG_SYNC);
1030 :
1031 97556 : if (xlog_is_shutdown(log))
1032 : return;
1033 :
1034 : /*
1035 : * If we think the summary counters are bad, avoid writing the unmount
1036 : * record to force log recovery at next mount, after which the summary
1037 : * counters will be recalculated. Refer to xlog_check_unmount_rec for
1038 : * more details.
1039 : */
1040 48778 : if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp,
1041 : XFS_ERRTAG_FORCE_SUMMARY_RECALC)) {
1042 83 : xfs_alert(mp, "%s: will fix summary counters at next mount",
1043 : __func__);
1044 83 : return;
1045 : }
1046 :
1047 48695 : xfs_log_unmount_verify_iclog(log);
1048 48695 : xlog_unmount_write(log);
1049 : }
1050 :
1051 : /*
1052 : * Empty the log for unmount/freeze.
1053 : *
1054 : * To do this, we first need to shut down the background log work so it is not
1055 : * trying to cover the log as we clean up. We then need to unpin all objects in
1056 : * the log so we can then flush them out. Once they have completed their IO and
1057 : * run the callbacks removing themselves from the AIL, we can cover the log.
1058 : */
1059 : int
1060 124421 : xfs_log_quiesce(
1061 : struct xfs_mount *mp)
1062 : {
1063 : /*
1064 : * Clear log incompat features since we're quiescing the log. Report
1065 : * failures, though it's not fatal to have a higher log feature
1066 : * protection level than the log contents actually require.
1067 : */
1068 124421 : if (xfs_clear_incompat_log_features(mp)) {
1069 983 : int error;
1070 :
1071 983 : error = xfs_sync_sb(mp, false);
1072 983 : if (error)
1073 0 : xfs_warn(mp,
1074 : "Failed to clear log incompat features on quiesce");
1075 : }
1076 :
1077 124421 : cancel_delayed_work_sync(&mp->m_log->l_work);
1078 124421 : xfs_log_force(mp, XFS_LOG_SYNC);
1079 :
1080 : /*
1081 : * The superblock buffer is uncached and while xfs_ail_push_all_sync()
1082 : * will push it, xfs_buftarg_wait() will not wait for it. Further,
1083 : * xfs_buf_iowait() cannot be used because it was pushed with the
1084 : * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for
1085 : * the IO to complete.
1086 : */
1087 124421 : xfs_ail_push_all_sync(mp->m_ail);
1088 124421 : xfs_buftarg_wait(mp->m_ddev_targp);
1089 124421 : xfs_buf_lock(mp->m_sb_bp);
1090 124421 : xfs_buf_unlock(mp->m_sb_bp);
1091 :
1092 124421 : return xfs_log_cover(mp);
1093 : }
1094 :
1095 : void
1096 2732 : xfs_log_clean(
1097 : struct xfs_mount *mp)
1098 : {
1099 2732 : xfs_log_quiesce(mp);
1100 61986 : xfs_log_unmount_write(mp);
1101 2732 : }
1102 :
1103 : /*
1104 : * Shut down and release the AIL and Log.
1105 : *
1106 : * During unmount, we need to ensure we flush all the dirty metadata objects
1107 : * from the AIL so that the log is empty before we write the unmount record to
1108 : * the log. Once this is done, we can tear down the AIL and the log.
1109 : */
1110 : void
1111 59254 : xfs_log_unmount(
1112 : struct xfs_mount *mp)
1113 : {
1114 59254 : xfs_log_clean(mp);
1115 :
1116 : /*
1117 : * If shutdown has come from iclog IO context, the log
1118 : * cleaning will have been skipped and so we need to wait
1119 : * for the iclog to complete shutdown processing before we
1120 : * tear anything down.
1121 : */
1122 59254 : xlog_wait_iclog_completion(mp->m_log);
1123 :
1124 59254 : xfs_buftarg_drain(mp->m_ddev_targp);
1125 :
1126 59254 : xfs_trans_ail_destroy(mp);
1127 :
1128 59254 : xfs_sysfs_del(&mp->m_log->l_kobj);
1129 :
1130 59254 : xlog_dealloc_log(mp->m_log);
1131 59254 : }
1132 :
1133 : void
1134 8098260611 : xfs_log_item_init(
1135 : struct xfs_mount *mp,
1136 : struct xfs_log_item *item,
1137 : int type,
1138 : const struct xfs_item_ops *ops)
1139 : {
1140 8098260611 : item->li_log = mp->m_log;
1141 8098260611 : item->li_ailp = mp->m_ail;
1142 8098260611 : item->li_type = type;
1143 8098260611 : item->li_ops = ops;
1144 8098260611 : item->li_lv = NULL;
1145 :
1146 8098260611 : INIT_LIST_HEAD(&item->li_ail);
1147 8098260611 : INIT_LIST_HEAD(&item->li_cil);
1148 8098260611 : INIT_LIST_HEAD(&item->li_bio_list);
1149 8098260611 : INIT_LIST_HEAD(&item->li_trans);
1150 8098260611 : }
1151 :
1152 : /*
1153 : * Wake up processes waiting for log space after we have moved the log tail.
1154 : */
1155 : void
1156 1342760154 : xfs_log_space_wake(
1157 : struct xfs_mount *mp)
1158 : {
1159 1342760154 : struct xlog *log = mp->m_log;
1160 1342760154 : int free_bytes;
1161 :
1162 2685520308 : if (xlog_is_shutdown(log))
1163 67890 : return;
1164 :
1165 1342692264 : if (!list_empty_careful(&log->l_write_head.waiters)) {
1166 402 : ASSERT(!xlog_in_recovery(log));
1167 :
1168 201 : spin_lock(&log->l_write_head.lock);
1169 201 : free_bytes = xlog_space_left(log, &log->l_write_head.grant);
1170 201 : xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);
1171 201 : spin_unlock(&log->l_write_head.lock);
1172 : }
1173 :
1174 1342815300 : if (!list_empty_careful(&log->l_reserve_head.waiters)) {
1175 11537576 : ASSERT(!xlog_in_recovery(log));
1176 :
1177 5768788 : spin_lock(&log->l_reserve_head.lock);
1178 5777841 : free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
1179 5777841 : xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);
1180 5777841 : spin_unlock(&log->l_reserve_head.lock);
1181 : }
1182 : }
1183 :
1184 : /*
1185 : * Determine if we have a transaction that has gone to disk that needs to be
1186 : * covered. To begin the transition to the idle state firstly the log needs to
1187 : * be idle. That means the CIL, the AIL and the iclogs needs to be empty before
1188 : * we start attempting to cover the log.
1189 : *
1190 : * Only if we are then in a state where covering is needed, the caller is
1191 : * informed that dummy transactions are required to move the log into the idle
1192 : * state.
1193 : *
1194 : * If there are any items in the AIl or CIL, then we do not want to attempt to
1195 : * cover the log as we may be in a situation where there isn't log space
1196 : * available to run a dummy transaction and this can lead to deadlocks when the
1197 : * tail of the log is pinned by an item that is modified in the CIL. Hence
1198 : * there's no point in running a dummy transaction at this point because we
1199 : * can't start trying to idle the log until both the CIL and AIL are empty.
1200 : */
1201 : static bool
1202 326325 : xfs_log_need_covered(
1203 : struct xfs_mount *mp)
1204 : {
1205 326325 : struct xlog *log = mp->m_log;
1206 326325 : bool needed = false;
1207 :
1208 326325 : if (!xlog_cil_empty(log))
1209 : return false;
1210 :
1211 321037 : spin_lock(&log->l_icloglock);
1212 321037 : switch (log->l_covered_state) {
1213 : case XLOG_STATE_COVER_DONE:
1214 : case XLOG_STATE_COVER_DONE2:
1215 : case XLOG_STATE_COVER_IDLE:
1216 : break;
1217 183882 : case XLOG_STATE_COVER_NEED:
1218 : case XLOG_STATE_COVER_NEED2:
1219 183882 : if (xfs_ail_min_lsn(log->l_ailp))
1220 : break;
1221 183137 : if (!xlog_iclogs_empty(log))
1222 : break;
1223 :
1224 183136 : needed = true;
1225 183136 : if (log->l_covered_state == XLOG_STATE_COVER_NEED)
1226 91586 : log->l_covered_state = XLOG_STATE_COVER_DONE;
1227 : else
1228 91550 : log->l_covered_state = XLOG_STATE_COVER_DONE2;
1229 : break;
1230 0 : default:
1231 0 : needed = true;
1232 0 : break;
1233 : }
1234 321037 : spin_unlock(&log->l_icloglock);
1235 321037 : return needed;
1236 : }
1237 :
1238 : /*
1239 : * Explicitly cover the log. This is similar to background log covering but
1240 : * intended for usage in quiesce codepaths. The caller is responsible to ensure
1241 : * the log is idle and suitable for covering. The CIL, iclog buffers and AIL
1242 : * must all be empty.
1243 : */
1244 : static int
1245 124421 : xfs_log_cover(
1246 : struct xfs_mount *mp)
1247 : {
1248 124421 : int error = 0;
1249 124421 : bool need_covered;
1250 :
1251 131554 : ASSERT((xlog_cil_empty(mp->m_log) && xlog_iclogs_empty(mp->m_log) &&
1252 : !xfs_ail_min_lsn(mp->m_log->l_ailp)) ||
1253 : xlog_is_shutdown(mp->m_log));
1254 :
1255 124421 : if (!xfs_log_writable(mp))
1256 : return 0;
1257 :
1258 : /*
1259 : * xfs_log_need_covered() is not idempotent because it progresses the
1260 : * state machine if the log requires covering. Therefore, we must call
1261 : * this function once and use the result until we've issued an sb sync.
1262 : * Do so first to make that abundantly clear.
1263 : *
1264 : * Fall into the covering sequence if the log needs covering or the
1265 : * mount has lazy superblock accounting to sync to disk. The sb sync
1266 : * used for covering accumulates the in-core counters, so covering
1267 : * handles this for us.
1268 : */
1269 111270 : need_covered = xfs_log_need_covered(mp);
1270 111270 : if (!need_covered && !xfs_has_lazysbcount(mp))
1271 : return 0;
1272 :
1273 : /*
1274 : * To cover the log, commit the superblock twice (at most) in
1275 : * independent checkpoints. The first serves as a reference for the
1276 : * tail pointer. The sync transaction and AIL push empties the AIL and
1277 : * updates the in-core tail to the LSN of the first checkpoint. The
1278 : * second commit updates the on-disk tail with the in-core LSN,
1279 : * covering the log. Push the AIL one more time to leave it empty, as
1280 : * we found it.
1281 : */
1282 202646 : do {
1283 202646 : error = xfs_sync_sb(mp, true);
1284 202646 : if (error)
1285 : break;
1286 202611 : xfs_ail_push_all_sync(mp->m_ail);
1287 202611 : } while (xfs_log_need_covered(mp));
1288 :
1289 : return error;
1290 : }
1291 :
1292 : /*
1293 : * We may be holding the log iclog lock upon entering this routine.
1294 : */
1295 : xfs_lsn_t
1296 22139851 : xlog_assign_tail_lsn_locked(
1297 : struct xfs_mount *mp)
1298 : {
1299 22139851 : struct xlog *log = mp->m_log;
1300 22139851 : struct xfs_log_item *lip;
1301 22139851 : xfs_lsn_t tail_lsn;
1302 :
1303 22139851 : assert_spin_locked(&mp->m_ail->ail_lock);
1304 :
1305 : /*
1306 : * To make sure we always have a valid LSN for the log tail we keep
1307 : * track of the last LSN which was committed in log->l_last_sync_lsn,
1308 : * and use that when the AIL was empty.
1309 : */
1310 22139851 : lip = xfs_ail_min(mp->m_ail);
1311 21036758 : if (lip)
1312 21036758 : tail_lsn = lip->li_lsn;
1313 : else
1314 1103093 : tail_lsn = atomic64_read(&log->l_last_sync_lsn);
1315 22139851 : trace_xfs_log_assign_tail_lsn(log, tail_lsn);
1316 22139852 : atomic64_set(&log->l_tail_lsn, tail_lsn);
1317 22139852 : return tail_lsn;
1318 : }
1319 :
1320 : xfs_lsn_t
1321 21180669 : xlog_assign_tail_lsn(
1322 : struct xfs_mount *mp)
1323 : {
1324 21180669 : xfs_lsn_t tail_lsn;
1325 :
1326 21180669 : spin_lock(&mp->m_ail->ail_lock);
1327 21180671 : tail_lsn = xlog_assign_tail_lsn_locked(mp);
1328 21180668 : spin_unlock(&mp->m_ail->ail_lock);
1329 :
1330 21180670 : return tail_lsn;
1331 : }
1332 :
1333 : /*
1334 : * Return the space in the log between the tail and the head. The head
1335 : * is passed in the cycle/bytes formal parms. In the special case where
1336 : * the reserve head has wrapped passed the tail, this calculation is no
1337 : * longer valid. In this case, just return 0 which means there is no space
1338 : * in the log. This works for all places where this function is called
1339 : * with the reserve head. Of course, if the write head were to ever
1340 : * wrap the tail, we should blow up. Rather than catch this case here,
1341 : * we depend on other ASSERTions in other parts of the code. XXXmiken
1342 : *
1343 : * If reservation head is behind the tail, we have a problem. Warn about it,
1344 : * but then treat it as if the log is empty.
1345 : *
1346 : * If the log is shut down, the head and tail may be invalid or out of whack, so
1347 : * shortcut invalidity asserts in this case so that we don't trigger them
1348 : * falsely.
1349 : */
1350 : STATIC int
1351 3934034633 : xlog_space_left(
1352 : struct xlog *log,
1353 : atomic64_t *head)
1354 : {
1355 3934034633 : int tail_bytes;
1356 3934034633 : int tail_cycle;
1357 3934034633 : int head_cycle;
1358 3934034633 : int head_bytes;
1359 :
1360 3934034633 : xlog_crack_grant_head(head, &head_cycle, &head_bytes);
1361 3934034633 : xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
1362 3934034633 : tail_bytes = BBTOB(tail_bytes);
1363 3934034633 : if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
1364 3174639351 : return log->l_logsize - (head_bytes - tail_bytes);
1365 759395282 : if (tail_cycle + 1 < head_cycle)
1366 : return 0;
1367 :
1368 : /* Ignore potential inconsistency when shutdown. */
1369 1518790392 : if (xlog_is_shutdown(log))
1370 0 : return log->l_logsize;
1371 :
1372 759395196 : if (tail_cycle < head_cycle) {
1373 759395196 : ASSERT(tail_cycle == (head_cycle - 1));
1374 759395196 : return tail_bytes - head_bytes;
1375 : }
1376 :
1377 : /*
1378 : * The reservation head is behind the tail. In this case we just want to
1379 : * return the size of the log as the amount of space left.
1380 : */
1381 0 : xfs_alert(log->l_mp, "xlog_space_left: head behind tail");
1382 0 : xfs_alert(log->l_mp, " tail_cycle = %d, tail_bytes = %d",
1383 : tail_cycle, tail_bytes);
1384 0 : xfs_alert(log->l_mp, " GH cycle = %d, GH bytes = %d",
1385 : head_cycle, head_bytes);
1386 0 : ASSERT(0);
1387 0 : return log->l_logsize;
1388 : }
1389 :
1390 :
1391 : static void
1392 21164609 : xlog_ioend_work(
1393 : struct work_struct *work)
1394 : {
1395 21164609 : struct xlog_in_core *iclog =
1396 21164609 : container_of(work, struct xlog_in_core, ic_end_io_work);
1397 21164609 : struct xlog *log = iclog->ic_log;
1398 21164609 : int error;
1399 :
1400 21164609 : error = blk_status_to_errno(iclog->ic_bio.bi_status);
1401 : #ifdef DEBUG
1402 : /* treat writes with injected CRC errors as failed */
1403 21164609 : if (iclog->ic_fail_crc)
1404 : error = -EIO;
1405 : #endif
1406 :
1407 : /*
1408 : * Race to shutdown the filesystem if we see an error.
1409 : */
1410 21164552 : if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) {
1411 12506 : xfs_alert(log->l_mp, "log I/O error %d", error);
1412 12506 : xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
1413 : }
1414 :
1415 21164609 : xlog_state_done_syncing(iclog);
1416 21164609 : bio_uninit(&iclog->ic_bio);
1417 :
1418 : /*
1419 : * Drop the lock to signal that we are done. Nothing references the
1420 : * iclog after this, so an unmount waiting on this lock can now tear it
1421 : * down safely. As such, it is unsafe to reference the iclog after the
1422 : * unlock as we could race with it being freed.
1423 : */
1424 21164609 : up(&iclog->ic_sema);
1425 21164609 : }
1426 :
1427 : /*
1428 : * Return size of each in-core log record buffer.
1429 : *
1430 : * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
1431 : *
1432 : * If the filesystem blocksize is too large, we may need to choose a
1433 : * larger size since the directory code currently logs entire blocks.
1434 : */
1435 : STATIC void
1436 59289 : xlog_get_iclog_buffer_size(
1437 : struct xfs_mount *mp,
1438 : struct xlog *log)
1439 : {
1440 59289 : if (mp->m_logbufs <= 0)
1441 59269 : mp->m_logbufs = XLOG_MAX_ICLOGS;
1442 59289 : if (mp->m_logbsize <= 0)
1443 58222 : mp->m_logbsize = XLOG_BIG_RECORD_BSIZE;
1444 :
1445 59289 : log->l_iclog_bufs = mp->m_logbufs;
1446 59289 : log->l_iclog_size = mp->m_logbsize;
1447 :
1448 : /*
1449 : * # headers = size / 32k - one header holds cycles from 32k of data.
1450 : */
1451 59289 : log->l_iclog_heads =
1452 59289 : DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE);
1453 59289 : log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT;
1454 59289 : }
1455 :
1456 : void
1457 134569 : xfs_log_work_queue(
1458 : struct xfs_mount *mp)
1459 : {
1460 134569 : queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work,
1461 134569 : msecs_to_jiffies(xfs_syncd_centisecs * 10));
1462 134569 : }
1463 :
1464 : /*
1465 : * Clear the log incompat flags if we have the opportunity.
1466 : *
1467 : * This only happens if we're about to log the second dummy transaction as part
1468 : * of covering the log and we can get the log incompat feature usage lock.
1469 : */
1470 : static inline void
1471 304 : xlog_clear_incompat(
1472 : struct xlog *log)
1473 : {
1474 304 : struct xfs_mount *mp = log->l_mp;
1475 :
1476 304 : if (!xfs_sb_has_incompat_log_feature(&mp->m_sb,
1477 : XFS_SB_FEAT_INCOMPAT_LOG_ALL))
1478 : return;
1479 :
1480 1 : if (log->l_covered_state != XLOG_STATE_COVER_DONE2)
1481 : return;
1482 :
1483 0 : if (!down_write_trylock(&log->l_incompat_users))
1484 : return;
1485 :
1486 0 : xfs_clear_incompat_log_features(mp);
1487 0 : up_write(&log->l_incompat_users);
1488 : }
1489 :
1490 : /*
1491 : * Every sync period we need to unpin all items in the AIL and push them to
1492 : * disk. If there is nothing dirty, then we might need to cover the log to
1493 : * indicate that the filesystem is idle.
1494 : */
1495 : static void
1496 12451 : xfs_log_worker(
1497 : struct work_struct *work)
1498 : {
1499 12451 : struct xlog *log = container_of(to_delayed_work(work),
1500 : struct xlog, l_work);
1501 12451 : struct xfs_mount *mp = log->l_mp;
1502 :
1503 : /* dgc: errors ignored - not fatal and nowhere to report them */
1504 12451 : if (xfs_fs_writable(mp, SB_FREEZE_WRITE) && xfs_log_need_covered(mp)) {
1505 : /*
1506 : * Dump a transaction into the log that contains no real change.
1507 : * This is needed to stamp the current tail LSN into the log
1508 : * during the covering operation.
1509 : *
1510 : * We cannot use an inode here for this - that will push dirty
1511 : * state back up into the VFS and then periodic inode flushing
1512 : * will prevent log covering from making progress. Hence we
1513 : * synchronously log the superblock instead to ensure the
1514 : * superblock is immediately unpinned and can be written back.
1515 : */
1516 304 : xlog_clear_incompat(log);
1517 304 : xfs_sync_sb(mp, true);
1518 : } else
1519 12149 : xfs_log_force(mp, 0);
1520 :
1521 : /* start pushing all the metadata that is currently dirty */
1522 12456 : xfs_ail_push_all(mp->m_ail);
1523 :
1524 : /* queue us up again */
1525 12456 : xfs_log_work_queue(mp);
1526 12456 : }
1527 :
1528 : /*
1529 : * This routine initializes some of the log structure for a given mount point.
1530 : * Its primary purpose is to fill in enough, so recovery can occur. However,
1531 : * some other stuff may be filled in too.
1532 : */
1533 : STATIC struct xlog *
1534 59289 : xlog_alloc_log(
1535 : struct xfs_mount *mp,
1536 : struct xfs_buftarg *log_target,
1537 : xfs_daddr_t blk_offset,
1538 : int num_bblks)
1539 : {
1540 59289 : struct xlog *log;
1541 59289 : xlog_rec_header_t *head;
1542 59289 : xlog_in_core_t **iclogp;
1543 59289 : xlog_in_core_t *iclog, *prev_iclog=NULL;
1544 59289 : int i;
1545 59289 : int error = -ENOMEM;
1546 59289 : uint log2_size = 0;
1547 :
1548 59289 : log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL);
1549 59289 : if (!log) {
1550 0 : xfs_warn(mp, "Log allocation failed: No memory!");
1551 0 : goto out;
1552 : }
1553 :
1554 59289 : log->l_mp = mp;
1555 59289 : log->l_targ = log_target;
1556 59289 : log->l_logsize = BBTOB(num_bblks);
1557 59289 : log->l_logBBstart = blk_offset;
1558 59289 : log->l_logBBsize = num_bblks;
1559 59289 : log->l_covered_state = XLOG_STATE_COVER_IDLE;
1560 59289 : set_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);
1561 59289 : INIT_DELAYED_WORK(&log->l_work, xfs_log_worker);
1562 :
1563 59289 : log->l_prev_block = -1;
1564 : /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1565 59289 : xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
1566 59289 : xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
1567 59289 : log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
1568 :
1569 59289 : if (xfs_has_logv2(mp) && mp->m_sb.sb_logsunit > 1)
1570 39798 : log->l_iclog_roundoff = mp->m_sb.sb_logsunit;
1571 : else
1572 19491 : log->l_iclog_roundoff = BBSIZE;
1573 :
1574 59289 : xlog_grant_head_init(&log->l_reserve_head);
1575 59289 : xlog_grant_head_init(&log->l_write_head);
1576 :
1577 59289 : error = -EFSCORRUPTED;
1578 59289 : if (xfs_has_sector(mp)) {
1579 56967 : log2_size = mp->m_sb.sb_logsectlog;
1580 56967 : if (log2_size < BBSHIFT) {
1581 0 : xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
1582 : log2_size, BBSHIFT);
1583 0 : goto out_free_log;
1584 : }
1585 :
1586 56967 : log2_size -= BBSHIFT;
1587 56967 : if (log2_size > mp->m_sectbb_log) {
1588 0 : xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
1589 : log2_size, mp->m_sectbb_log);
1590 0 : goto out_free_log;
1591 : }
1592 :
1593 : /* for larger sector sizes, must have v2 or external log */
1594 56967 : if (log2_size && log->l_logBBstart > 0 &&
1595 : !xfs_has_logv2(mp)) {
1596 0 : xfs_warn(mp,
1597 : "log sector size (0x%x) invalid for configuration.",
1598 : log2_size);
1599 0 : goto out_free_log;
1600 : }
1601 : }
1602 56967 : log->l_sectBBsize = 1 << log2_size;
1603 :
1604 59289 : init_rwsem(&log->l_incompat_users);
1605 :
1606 59289 : xlog_get_iclog_buffer_size(mp, log);
1607 :
1608 59289 : spin_lock_init(&log->l_icloglock);
1609 59289 : init_waitqueue_head(&log->l_flush_wait);
1610 :
1611 59289 : iclogp = &log->l_iclog;
1612 : /*
1613 : * The amount of memory to allocate for the iclog structure is
1614 : * rather funky due to the way the structure is defined. It is
1615 : * done this way so that we can use different sizes for machines
1616 : * with different amounts of memory. See the definition of
1617 : * xlog_in_core_t in xfs_log_priv.h for details.
1618 : */
1619 59289 : ASSERT(log->l_iclog_size >= 4096);
1620 533541 : for (i = 0; i < log->l_iclog_bufs; i++) {
1621 474252 : size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE) *
1622 : sizeof(struct bio_vec);
1623 :
1624 474252 : iclog = kmem_zalloc(sizeof(*iclog) + bvec_size, KM_MAYFAIL);
1625 474252 : if (!iclog)
1626 0 : goto out_free_iclog;
1627 :
1628 474252 : *iclogp = iclog;
1629 474252 : iclog->ic_prev = prev_iclog;
1630 474252 : prev_iclog = iclog;
1631 :
1632 474252 : iclog->ic_data = kvzalloc(log->l_iclog_size,
1633 : GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1634 474252 : if (!iclog->ic_data)
1635 0 : goto out_free_iclog;
1636 474252 : head = &iclog->ic_header;
1637 474252 : memset(head, 0, sizeof(xlog_rec_header_t));
1638 474252 : head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1639 474252 : head->h_version = cpu_to_be32(
1640 : xfs_has_logv2(log->l_mp) ? 2 : 1);
1641 474252 : head->h_size = cpu_to_be32(log->l_iclog_size);
1642 : /* new fields */
1643 474252 : head->h_fmt = cpu_to_be32(XLOG_FMT);
1644 948504 : memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1645 :
1646 474252 : iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize;
1647 474252 : iclog->ic_state = XLOG_STATE_ACTIVE;
1648 474252 : iclog->ic_log = log;
1649 474252 : atomic_set(&iclog->ic_refcnt, 0);
1650 474252 : INIT_LIST_HEAD(&iclog->ic_callbacks);
1651 474252 : iclog->ic_datap = (void *)iclog->ic_data + log->l_iclog_hsize;
1652 :
1653 474252 : init_waitqueue_head(&iclog->ic_force_wait);
1654 474252 : init_waitqueue_head(&iclog->ic_write_wait);
1655 474252 : INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work);
1656 474252 : sema_init(&iclog->ic_sema, 1);
1657 :
1658 474252 : iclogp = &iclog->ic_next;
1659 : }
1660 59289 : *iclogp = log->l_iclog; /* complete ring */
1661 59289 : log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1662 :
1663 118578 : log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s",
1664 : XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM |
1665 : WQ_HIGHPRI),
1666 59289 : 0, mp->m_super->s_id);
1667 59289 : if (!log->l_ioend_workqueue)
1668 0 : goto out_free_iclog;
1669 :
1670 59289 : error = xlog_cil_init(log);
1671 59289 : if (error)
1672 0 : goto out_destroy_workqueue;
1673 : return log;
1674 :
1675 : out_destroy_workqueue:
1676 0 : destroy_workqueue(log->l_ioend_workqueue);
1677 0 : out_free_iclog:
1678 0 : for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1679 0 : prev_iclog = iclog->ic_next;
1680 0 : kmem_free(iclog->ic_data);
1681 0 : kmem_free(iclog);
1682 0 : if (prev_iclog == log->l_iclog)
1683 : break;
1684 : }
1685 0 : out_free_log:
1686 0 : kmem_free(log);
1687 0 : out:
1688 0 : return ERR_PTR(error);
1689 : } /* xlog_alloc_log */
1690 :
1691 : /*
1692 : * Compute the LSN that we'd need to push the log tail towards in order to have
1693 : * (a) enough on-disk log space to log the number of bytes specified, (b) at
1694 : * least 25% of the log space free, and (c) at least 256 blocks free. If the
1695 : * log free space already meets all three thresholds, this function returns
1696 : * NULLCOMMITLSN.
1697 : */
1698 : xfs_lsn_t
1699 2236322518 : xlog_grant_push_threshold(
1700 : struct xlog *log,
1701 : int need_bytes)
1702 : {
1703 2236322518 : xfs_lsn_t threshold_lsn = 0;
1704 2236322518 : xfs_lsn_t last_sync_lsn;
1705 2236322518 : int free_blocks;
1706 2236322518 : int free_bytes;
1707 2236322518 : int threshold_block;
1708 2236322518 : int threshold_cycle;
1709 2236322518 : int free_threshold;
1710 :
1711 2236322518 : ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
1712 :
1713 2236322518 : free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
1714 2236288284 : free_blocks = BTOBBT(free_bytes);
1715 :
1716 : /*
1717 : * Set the threshold for the minimum number of free blocks in the
1718 : * log to the maximum of what the caller needs, one quarter of the
1719 : * log, and 256 blocks.
1720 : */
1721 2236288284 : free_threshold = BTOBB(need_bytes);
1722 2236288284 : free_threshold = max(free_threshold, (log->l_logBBsize >> 2));
1723 2236288284 : free_threshold = max(free_threshold, 256);
1724 2236288284 : if (free_blocks >= free_threshold)
1725 : return NULLCOMMITLSN;
1726 :
1727 20324805 : xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
1728 : &threshold_block);
1729 20324805 : threshold_block += free_threshold;
1730 20324805 : if (threshold_block >= log->l_logBBsize) {
1731 3100307 : threshold_block -= log->l_logBBsize;
1732 3100307 : threshold_cycle += 1;
1733 : }
1734 20324805 : threshold_lsn = xlog_assign_lsn(threshold_cycle,
1735 : threshold_block);
1736 : /*
1737 : * Don't pass in an lsn greater than the lsn of the last
1738 : * log record known to be on disk. Use a snapshot of the last sync lsn
1739 : * so that it doesn't change between the compare and the set.
1740 : */
1741 20324805 : last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
1742 20324805 : if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
1743 : threshold_lsn = last_sync_lsn;
1744 :
1745 : return threshold_lsn;
1746 : }
1747 :
1748 : /*
1749 : * Push the tail of the log if we need to do so to maintain the free log space
1750 : * thresholds set out by xlog_grant_push_threshold. We may need to adopt a
1751 : * policy which pushes on an lsn which is further along in the log once we
1752 : * reach the high water mark. In this manner, we would be creating a low water
1753 : * mark.
1754 : */
1755 : STATIC void
1756 2236648497 : xlog_grant_push_ail(
1757 : struct xlog *log,
1758 : int need_bytes)
1759 : {
1760 2236648497 : xfs_lsn_t threshold_lsn;
1761 :
1762 2236648497 : threshold_lsn = xlog_grant_push_threshold(log, need_bytes);
1763 2256109528 : if (threshold_lsn == NULLCOMMITLSN || xlog_is_shutdown(log))
1764 : return;
1765 :
1766 : /*
1767 : * Get the transaction layer to kick the dirty buffers out to
1768 : * disk asynchronously. No point in trying to do this if
1769 : * the filesystem is shutting down.
1770 : */
1771 20313560 : xfs_ail_push(log->l_ailp, threshold_lsn);
1772 : }
1773 :
1774 : /*
1775 : * Stamp cycle number in every block
1776 : */
1777 : STATIC void
1778 21164488 : xlog_pack_data(
1779 : struct xlog *log,
1780 : struct xlog_in_core *iclog,
1781 : int roundoff)
1782 : {
1783 21164488 : int i, j, k;
1784 21164488 : int size = iclog->ic_offset + roundoff;
1785 21164488 : __be32 cycle_lsn;
1786 21164488 : char *dp;
1787 :
1788 21164488 : cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
1789 :
1790 21164488 : dp = iclog->ic_datap;
1791 1140251103 : for (i = 0; i < BTOBB(size); i++) {
1792 1119098502 : if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE))
1793 : break;
1794 1119086524 : iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
1795 1119086615 : *(__be32 *)dp = cycle_lsn;
1796 1119086615 : dp += BBSIZE;
1797 : }
1798 :
1799 21164579 : if (xfs_has_logv2(log->l_mp)) {
1800 21164519 : xlog_in_core_2_t *xhdr = iclog->ic_data;
1801 :
1802 24861191 : for ( ; i < BTOBB(size); i++) {
1803 3696778 : j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1804 3696778 : k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1805 3696778 : xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
1806 3696672 : *(__be32 *)dp = cycle_lsn;
1807 3696672 : dp += BBSIZE;
1808 : }
1809 :
1810 21237883 : for (i = 1; i < log->l_iclog_heads; i++)
1811 73469 : xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
1812 : }
1813 21164474 : }
1814 :
1815 : /*
1816 : * Calculate the checksum for a log buffer.
1817 : *
1818 : * This is a little more complicated than it should be because the various
1819 : * headers and the actual data are non-contiguous.
1820 : */
1821 : __le32
1822 25929798 : xlog_cksum(
1823 : struct xlog *log,
1824 : struct xlog_rec_header *rhead,
1825 : char *dp,
1826 : int size)
1827 : {
1828 25929798 : uint32_t crc;
1829 :
1830 : /* first generate the crc for the record header ... */
1831 25929798 : crc = xfs_start_cksum_update((char *)rhead,
1832 : sizeof(struct xlog_rec_header),
1833 : offsetof(struct xlog_rec_header, h_crc));
1834 :
1835 : /* ... then for additional cycle data for v2 logs ... */
1836 25929344 : if (xfs_has_logv2(log->l_mp)) {
1837 25929210 : union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead;
1838 25929210 : int i;
1839 25929210 : int xheads;
1840 :
1841 25929210 : xheads = DIV_ROUND_UP(size, XLOG_HEADER_CYCLE_SIZE);
1842 :
1843 26028684 : for (i = 1; i < xheads; i++) {
1844 99492 : crc = crc32c(crc, &xhdr[i].hic_xheader,
1845 : sizeof(struct xlog_rec_ext_header));
1846 : }
1847 : }
1848 :
1849 : /* ... and finally for the payload */
1850 25929354 : crc = crc32c(crc, dp, size);
1851 :
1852 25929251 : return xfs_end_cksum(crc);
1853 : }
1854 :
1855 : static void
1856 21164609 : xlog_bio_end_io(
1857 : struct bio *bio)
1858 : {
1859 21164609 : struct xlog_in_core *iclog = bio->bi_private;
1860 :
1861 21164609 : queue_work(iclog->ic_log->l_ioend_workqueue,
1862 : &iclog->ic_end_io_work);
1863 21164609 : }
1864 :
1865 : static int
1866 21163621 : xlog_map_iclog_data(
1867 : struct bio *bio,
1868 : void *data,
1869 : size_t count)
1870 : {
1871 143647126 : do {
1872 143647126 : struct page *page = kmem_to_page(data);
1873 143646174 : unsigned int off = offset_in_page(data);
1874 143646174 : size_t len = min_t(size_t, count, PAGE_SIZE - off);
1875 :
1876 143646174 : if (bio_add_page(bio, page, len, off) != len)
1877 : return -EIO;
1878 :
1879 143647909 : data += len;
1880 143647909 : count -= len;
1881 143647909 : } while (count);
1882 :
1883 : return 0;
1884 : }
1885 :
1886 : STATIC void
1887 21164181 : xlog_write_iclog(
1888 : struct xlog *log,
1889 : struct xlog_in_core *iclog,
1890 : uint64_t bno,
1891 : unsigned int count)
1892 : {
1893 21164181 : ASSERT(bno < log->l_logBBsize);
1894 21164181 : trace_xlog_iclog_write(iclog, _RET_IP_);
1895 :
1896 : /*
1897 : * We lock the iclogbufs here so that we can serialise against I/O
1898 : * completion during unmount. We might be processing a shutdown
1899 : * triggered during unmount, and that can occur asynchronously to the
1900 : * unmount thread, and hence we need to ensure that completes before
1901 : * tearing down the iclogbufs. Hence we need to hold the buffer lock
1902 : * across the log IO to archieve that.
1903 : */
1904 21163897 : down(&iclog->ic_sema);
1905 42327788 : if (xlog_is_shutdown(log)) {
1906 : /*
1907 : * It would seem logical to return EIO here, but we rely on
1908 : * the log state machine to propagate I/O errors instead of
1909 : * doing it here. We kick of the state machine and unlock
1910 : * the buffer manually, the code needs to be kept in sync
1911 : * with the I/O completion path.
1912 : */
1913 10 : xlog_state_done_syncing(iclog);
1914 10 : up(&iclog->ic_sema);
1915 10 : return;
1916 : }
1917 :
1918 : /*
1919 : * We use REQ_SYNC | REQ_IDLE here to tell the block layer the are more
1920 : * IOs coming immediately after this one. This prevents the block layer
1921 : * writeback throttle from throttling log writes behind background
1922 : * metadata writeback and causing priority inversions.
1923 : */
1924 21163884 : bio_init(&iclog->ic_bio, log->l_targ->bt_bdev, iclog->ic_bvec,
1925 21163884 : howmany(count, PAGE_SIZE),
1926 : REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_IDLE);
1927 21163750 : iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno;
1928 21163750 : iclog->ic_bio.bi_end_io = xlog_bio_end_io;
1929 21163750 : iclog->ic_bio.bi_private = iclog;
1930 :
1931 21163750 : if (iclog->ic_flags & XLOG_ICL_NEED_FLUSH) {
1932 6697777 : iclog->ic_bio.bi_opf |= REQ_PREFLUSH;
1933 : /*
1934 : * For external log devices, we also need to flush the data
1935 : * device cache first to ensure all metadata writeback covered
1936 : * by the LSN in this iclog is on stable storage. This is slow,
1937 : * but it *must* complete before we issue the external log IO.
1938 : *
1939 : * If the flush fails, we cannot conclude that past metadata
1940 : * writeback from the log succeeded. Repeating the flush is
1941 : * not possible, hence we must shut down with log IO error to
1942 : * avoid shutdown re-entering this path and erroring out again.
1943 : */
1944 8349055 : if (log->l_targ != log->l_mp->m_ddev_targp &&
1945 1651285 : blkdev_issue_flush(log->l_mp->m_ddev_targp->bt_bdev)) {
1946 0 : xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
1947 0 : return;
1948 : }
1949 : }
1950 21163743 : if (iclog->ic_flags & XLOG_ICL_NEED_FUA)
1951 5117202 : iclog->ic_bio.bi_opf |= REQ_FUA;
1952 :
1953 21163743 : iclog->ic_flags &= ~(XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA);
1954 :
1955 21163743 : if (xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, count)) {
1956 0 : xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
1957 0 : return;
1958 : }
1959 21164261 : if (is_vmalloc_addr(iclog->ic_data))
1960 : flush_kernel_vmap_range(iclog->ic_data, count);
1961 :
1962 : /*
1963 : * If this log buffer would straddle the end of the log we will have
1964 : * to split it up into two bios, so that we can continue at the start.
1965 : */
1966 21164267 : if (bno + BTOBB(count) > log->l_logBBsize) {
1967 6899 : struct bio *split;
1968 :
1969 6899 : split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno,
1970 : GFP_NOIO, &fs_bio_set);
1971 6899 : bio_chain(split, &iclog->ic_bio);
1972 6899 : submit_bio(split);
1973 :
1974 : /* restart at logical offset zero for the remainder */
1975 6899 : iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart;
1976 : }
1977 :
1978 21164267 : submit_bio(&iclog->ic_bio);
1979 : }
1980 :
1981 : /*
1982 : * We need to bump cycle number for the part of the iclog that is
1983 : * written to the start of the log. Watch out for the header magic
1984 : * number case, though.
1985 : */
1986 : static void
1987 6899 : xlog_split_iclog(
1988 : struct xlog *log,
1989 : void *data,
1990 : uint64_t bno,
1991 : unsigned int count)
1992 : {
1993 6899 : unsigned int split_offset = BBTOB(log->l_logBBsize - bno);
1994 6899 : unsigned int i;
1995 :
1996 220151 : for (i = split_offset; i < count; i += BBSIZE) {
1997 213252 : uint32_t cycle = get_unaligned_be32(data + i);
1998 :
1999 213252 : if (++cycle == XLOG_HEADER_MAGIC_NUM)
2000 0 : cycle++;
2001 213252 : put_unaligned_be32(cycle, data + i);
2002 : }
2003 6899 : }
2004 :
2005 : static int
2006 21164550 : xlog_calc_iclog_size(
2007 : struct xlog *log,
2008 : struct xlog_in_core *iclog,
2009 : uint32_t *roundoff)
2010 : {
2011 21164550 : uint32_t count_init, count;
2012 :
2013 : /* Add for LR header */
2014 21164550 : count_init = log->l_iclog_hsize + iclog->ic_offset;
2015 21164550 : count = roundup(count_init, log->l_iclog_roundoff);
2016 :
2017 21164550 : *roundoff = count - count_init;
2018 :
2019 21164550 : ASSERT(count >= count_init);
2020 21164550 : ASSERT(*roundoff < log->l_iclog_roundoff);
2021 21164550 : return count;
2022 : }
2023 :
2024 : /*
2025 : * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
2026 : * fashion. Previously, we should have moved the current iclog
2027 : * ptr in the log to point to the next available iclog. This allows further
2028 : * write to continue while this code syncs out an iclog ready to go.
2029 : * Before an in-core log can be written out, the data section must be scanned
2030 : * to save away the 1st word of each BBSIZE block into the header. We replace
2031 : * it with the current cycle count. Each BBSIZE block is tagged with the
2032 : * cycle count because there in an implicit assumption that drives will
2033 : * guarantee that entire 512 byte blocks get written at once. In other words,
2034 : * we can't have part of a 512 byte block written and part not written. By
2035 : * tagging each block, we will know which blocks are valid when recovering
2036 : * after an unclean shutdown.
2037 : *
2038 : * This routine is single threaded on the iclog. No other thread can be in
2039 : * this routine with the same iclog. Changing contents of iclog can there-
2040 : * fore be done without grabbing the state machine lock. Updating the global
2041 : * log will require grabbing the lock though.
2042 : *
2043 : * The entire log manager uses a logical block numbering scheme. Only
2044 : * xlog_write_iclog knows about the fact that the log may not start with
2045 : * block zero on a given device.
2046 : */
2047 : STATIC void
2048 21164591 : xlog_sync(
2049 : struct xlog *log,
2050 : struct xlog_in_core *iclog,
2051 : struct xlog_ticket *ticket)
2052 : {
2053 21164591 : unsigned int count; /* byte count of bwrite */
2054 21164591 : unsigned int roundoff; /* roundoff to BB or stripe */
2055 21164591 : uint64_t bno;
2056 21164591 : unsigned int size;
2057 :
2058 21164591 : ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
2059 21164591 : trace_xlog_iclog_sync(iclog, _RET_IP_);
2060 :
2061 21164560 : count = xlog_calc_iclog_size(log, iclog, &roundoff);
2062 :
2063 : /*
2064 : * If we have a ticket, account for the roundoff via the ticket
2065 : * reservation to avoid touching the hot grant heads needlessly.
2066 : * Otherwise, we have to move grant heads directly.
2067 : */
2068 21164552 : if (ticket) {
2069 16521869 : ticket->t_curr_res -= roundoff;
2070 : } else {
2071 4642683 : xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
2072 4642687 : xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
2073 : }
2074 :
2075 : /* put cycle number in every block */
2076 21164555 : xlog_pack_data(log, iclog, roundoff);
2077 :
2078 : /* real byte length */
2079 21164467 : size = iclog->ic_offset;
2080 21164467 : if (xfs_has_logv2(log->l_mp))
2081 21164420 : size += roundoff;
2082 21164467 : iclog->ic_header.h_len = cpu_to_be32(size);
2083 :
2084 21164467 : XFS_STATS_INC(log->l_mp, xs_log_writes);
2085 21164516 : XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count));
2086 :
2087 21164538 : bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn));
2088 :
2089 : /* Do we need to split this write into 2 parts? */
2090 21164538 : if (bno + BTOBB(count) > log->l_logBBsize)
2091 6899 : xlog_split_iclog(log, &iclog->ic_header, bno, count);
2092 :
2093 : /* calculcate the checksum */
2094 42328435 : iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
2095 21164538 : iclog->ic_datap, size);
2096 : /*
2097 : * Intentionally corrupt the log record CRC based on the error injection
2098 : * frequency, if defined. This facilitates testing log recovery in the
2099 : * event of torn writes. Hence, set the IOABORT state to abort the log
2100 : * write on I/O completion and shutdown the fs. The subsequent mount
2101 : * detects the bad CRC and attempts to recover.
2102 : */
2103 : #ifdef DEBUG
2104 21163897 : if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) {
2105 58 : iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA);
2106 58 : iclog->ic_fail_crc = true;
2107 58 : xfs_warn(log->l_mp,
2108 : "Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.",
2109 : be64_to_cpu(iclog->ic_header.h_lsn));
2110 : }
2111 : #endif
2112 21163709 : xlog_verify_iclog(log, iclog, count);
2113 21164543 : xlog_write_iclog(log, iclog, bno, count);
2114 21164582 : }
2115 :
2116 : /*
2117 : * Deallocate a log structure
2118 : */
2119 : STATIC void
2120 59296 : xlog_dealloc_log(
2121 : struct xlog *log)
2122 : {
2123 59296 : xlog_in_core_t *iclog, *next_iclog;
2124 59296 : int i;
2125 :
2126 : /*
2127 : * Destroy the CIL after waiting for iclog IO completion because an
2128 : * iclog EIO error will try to shut down the log, which accesses the
2129 : * CIL to wake up the waiters.
2130 : */
2131 59296 : xlog_cil_destroy(log);
2132 :
2133 59296 : iclog = log->l_iclog;
2134 533604 : for (i = 0; i < log->l_iclog_bufs; i++) {
2135 474308 : next_iclog = iclog->ic_next;
2136 474308 : kmem_free(iclog->ic_data);
2137 474308 : kmem_free(iclog);
2138 474308 : iclog = next_iclog;
2139 : }
2140 :
2141 59296 : log->l_mp->m_log = NULL;
2142 59296 : destroy_workqueue(log->l_ioend_workqueue);
2143 59296 : kmem_free(log);
2144 59296 : }
2145 :
2146 : /*
2147 : * Update counters atomically now that memcpy is done.
2148 : */
2149 : static inline void
2150 : xlog_state_finish_copy(
2151 : struct xlog *log,
2152 : struct xlog_in_core *iclog,
2153 : int record_cnt,
2154 : int copy_bytes)
2155 : {
2156 26357370 : lockdep_assert_held(&log->l_icloglock);
2157 :
2158 26357370 : be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
2159 26357370 : iclog->ic_offset += copy_bytes;
2160 : }
2161 :
2162 : /*
2163 : * print out info relating to regions written which consume
2164 : * the reservation
2165 : */
2166 : void
2167 2 : xlog_print_tic_res(
2168 : struct xfs_mount *mp,
2169 : struct xlog_ticket *ticket)
2170 : {
2171 2 : xfs_warn(mp, "ticket reservation summary:");
2172 2 : xfs_warn(mp, " unit res = %d bytes", ticket->t_unit_res);
2173 2 : xfs_warn(mp, " current res = %d bytes", ticket->t_curr_res);
2174 2 : xfs_warn(mp, " original count = %d", ticket->t_ocnt);
2175 2 : xfs_warn(mp, " remaining count = %d", ticket->t_cnt);
2176 2 : }
2177 :
2178 : /*
2179 : * Print a summary of the transaction.
2180 : */
2181 : void
2182 0 : xlog_print_trans(
2183 : struct xfs_trans *tp)
2184 : {
2185 0 : struct xfs_mount *mp = tp->t_mountp;
2186 0 : struct xfs_log_item *lip;
2187 :
2188 : /* dump core transaction and ticket info */
2189 0 : xfs_warn(mp, "transaction summary:");
2190 0 : xfs_warn(mp, " log res = %d", tp->t_log_res);
2191 0 : xfs_warn(mp, " log count = %d", tp->t_log_count);
2192 0 : xfs_warn(mp, " flags = 0x%x", tp->t_flags);
2193 :
2194 0 : xlog_print_tic_res(mp, tp->t_ticket);
2195 :
2196 : /* dump each log item */
2197 0 : list_for_each_entry(lip, &tp->t_items, li_trans) {
2198 0 : struct xfs_log_vec *lv = lip->li_lv;
2199 0 : struct xfs_log_iovec *vec;
2200 0 : int i;
2201 :
2202 0 : xfs_warn(mp, "log item: ");
2203 0 : xfs_warn(mp, " type = 0x%x", lip->li_type);
2204 0 : xfs_warn(mp, " flags = 0x%lx", lip->li_flags);
2205 0 : if (!lv)
2206 0 : continue;
2207 0 : xfs_warn(mp, " niovecs = %d", lv->lv_niovecs);
2208 0 : xfs_warn(mp, " size = %d", lv->lv_size);
2209 0 : xfs_warn(mp, " bytes = %d", lv->lv_bytes);
2210 0 : xfs_warn(mp, " buf len = %d", lv->lv_buf_len);
2211 :
2212 : /* dump each iovec for the log item */
2213 0 : vec = lv->lv_iovecp;
2214 0 : for (i = 0; i < lv->lv_niovecs; i++) {
2215 0 : int dumplen = min(vec->i_len, 32);
2216 :
2217 0 : xfs_warn(mp, " iovec[%d]", i);
2218 0 : xfs_warn(mp, " type = 0x%x", vec->i_type);
2219 0 : xfs_warn(mp, " len = %d", vec->i_len);
2220 0 : xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i);
2221 0 : xfs_hex_dump(vec->i_addr, dumplen);
2222 :
2223 0 : vec++;
2224 : }
2225 : }
2226 0 : }
2227 :
2228 : static inline void
2229 1696085551 : xlog_write_iovec(
2230 : struct xlog_in_core *iclog,
2231 : uint32_t *log_offset,
2232 : void *data,
2233 : uint32_t write_len,
2234 : int *bytes_left,
2235 : uint32_t *record_cnt,
2236 : uint32_t *data_cnt)
2237 : {
2238 1696085551 : ASSERT(*log_offset < iclog->ic_log->l_iclog_size);
2239 1696085551 : ASSERT(*log_offset % sizeof(int32_t) == 0);
2240 1696085551 : ASSERT(write_len % sizeof(int32_t) == 0);
2241 :
2242 3392171102 : memcpy(iclog->ic_datap + *log_offset, data, write_len);
2243 1696085551 : *log_offset += write_len;
2244 1696085551 : *bytes_left -= write_len;
2245 1696085551 : (*record_cnt)++;
2246 1696085551 : *data_cnt += write_len;
2247 1696085551 : }
2248 :
2249 : /*
2250 : * Write log vectors into a single iclog which is guaranteed by the caller
2251 : * to have enough space to write the entire log vector into.
2252 : */
2253 : static void
2254 643906091 : xlog_write_full(
2255 : struct xfs_log_vec *lv,
2256 : struct xlog_ticket *ticket,
2257 : struct xlog_in_core *iclog,
2258 : uint32_t *log_offset,
2259 : uint32_t *len,
2260 : uint32_t *record_cnt,
2261 : uint32_t *data_cnt)
2262 : {
2263 643906091 : int index;
2264 :
2265 643906091 : ASSERT(*log_offset + *len <= iclog->ic_size ||
2266 : iclog->ic_state == XLOG_STATE_WANT_SYNC);
2267 :
2268 : /*
2269 : * Ordered log vectors have no regions to write so this
2270 : * loop will naturally skip them.
2271 : */
2272 2280698967 : for (index = 0; index < lv->lv_niovecs; index++) {
2273 1636751150 : struct xfs_log_iovec *reg = &lv->lv_iovecp[index];
2274 1636751150 : struct xlog_op_header *ophdr = reg->i_addr;
2275 :
2276 1636751150 : ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2277 1636751150 : xlog_write_iovec(iclog, log_offset, reg->i_addr,
2278 1636751150 : reg->i_len, len, record_cnt, data_cnt);
2279 : }
2280 643947817 : }
2281 :
2282 : static int
2283 16111174 : xlog_write_get_more_iclog_space(
2284 : struct xlog_ticket *ticket,
2285 : struct xlog_in_core **iclogp,
2286 : uint32_t *log_offset,
2287 : uint32_t len,
2288 : uint32_t *record_cnt,
2289 : uint32_t *data_cnt)
2290 : {
2291 16111174 : struct xlog_in_core *iclog = *iclogp;
2292 16111174 : struct xlog *log = iclog->ic_log;
2293 16111174 : int error;
2294 :
2295 16111174 : spin_lock(&log->l_icloglock);
2296 16111221 : ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC);
2297 16111221 : xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
2298 16111221 : error = xlog_state_release_iclog(log, iclog, ticket);
2299 16111218 : spin_unlock(&log->l_icloglock);
2300 16111218 : if (error)
2301 : return error;
2302 :
2303 16111214 : error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
2304 : log_offset);
2305 16111211 : if (error)
2306 : return error;
2307 16108140 : *record_cnt = 0;
2308 16108140 : *data_cnt = 0;
2309 16108140 : *iclogp = iclog;
2310 16108140 : return 0;
2311 : }
2312 :
2313 : /*
2314 : * Write log vectors into a single iclog which is smaller than the current chain
2315 : * length. We write until we cannot fit a full record into the remaining space
2316 : * and then stop. We return the log vector that is to be written that cannot
2317 : * wholly fit in the iclog.
2318 : */
2319 : static int
2320 16110394 : xlog_write_partial(
2321 : struct xfs_log_vec *lv,
2322 : struct xlog_ticket *ticket,
2323 : struct xlog_in_core **iclogp,
2324 : uint32_t *log_offset,
2325 : uint32_t *len,
2326 : uint32_t *record_cnt,
2327 : uint32_t *data_cnt)
2328 : {
2329 16110394 : struct xlog_in_core *iclog = *iclogp;
2330 16110394 : struct xlog_op_header *ophdr;
2331 16110394 : int index = 0;
2332 16110394 : uint32_t rlen;
2333 16110394 : int error;
2334 :
2335 : /* walk the logvec, copying until we run out of space in the iclog */
2336 60318540 : for (index = 0; index < lv->lv_niovecs; index++) {
2337 44211237 : struct xfs_log_iovec *reg = &lv->lv_iovecp[index];
2338 44211237 : uint32_t reg_offset = 0;
2339 :
2340 : /*
2341 : * The first region of a continuation must have a non-zero
2342 : * length otherwise log recovery will just skip over it and
2343 : * start recovering from the next opheader it finds. Because we
2344 : * mark the next opheader as a continuation, recovery will then
2345 : * incorrectly add the continuation to the previous region and
2346 : * that breaks stuff.
2347 : *
2348 : * Hence if there isn't space for region data after the
2349 : * opheader, then we need to start afresh with a new iclog.
2350 : */
2351 44211237 : if (iclog->ic_size - *log_offset <=
2352 : sizeof(struct xlog_op_header)) {
2353 764610 : error = xlog_write_get_more_iclog_space(ticket,
2354 : &iclog, log_offset, *len, record_cnt,
2355 : data_cnt);
2356 764610 : if (error)
2357 98 : return error;
2358 : }
2359 :
2360 44211139 : ophdr = reg->i_addr;
2361 44211139 : rlen = min_t(uint32_t, reg->i_len, iclog->ic_size - *log_offset);
2362 :
2363 44211139 : ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2364 44211139 : ophdr->oh_len = cpu_to_be32(rlen - sizeof(struct xlog_op_header));
2365 44211139 : if (rlen != reg->i_len)
2366 15345878 : ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
2367 :
2368 44211139 : xlog_write_iovec(iclog, log_offset, reg->i_addr,
2369 : rlen, len, record_cnt, data_cnt);
2370 :
2371 : /* If we wrote the whole region, move to the next. */
2372 44210978 : if (rlen == reg->i_len)
2373 28865244 : continue;
2374 :
2375 : /*
2376 : * We now have a partially written iovec, but it can span
2377 : * multiple iclogs so we loop here. First we release the iclog
2378 : * we currently have, then we get a new iclog and add a new
2379 : * opheader. Then we continue copying from where we were until
2380 : * we either complete the iovec or fill the iclog. If we
2381 : * complete the iovec, then we increment the index and go right
2382 : * back to the top of the outer loop. if we fill the iclog, we
2383 : * run the inner loop again.
2384 : *
2385 : * This is complicated by the tail of a region using all the
2386 : * space in an iclog and hence requiring us to release the iclog
2387 : * and get a new one before returning to the outer loop. We must
2388 : * always guarantee that we exit this inner loop with at least
2389 : * space for log transaction opheaders left in the current
2390 : * iclog, hence we cannot just terminate the loop at the end
2391 : * of the of the continuation. So we loop while there is no
2392 : * space left in the current iclog, and check for the end of the
2393 : * continuation after getting a new iclog.
2394 : */
2395 15346404 : do {
2396 : /*
2397 : * Ensure we include the continuation opheader in the
2398 : * space we need in the new iclog by adding that size
2399 : * to the length we require. This continuation opheader
2400 : * needs to be accounted to the ticket as the space it
2401 : * consumes hasn't been accounted to the lv we are
2402 : * writing.
2403 : */
2404 15346404 : error = xlog_write_get_more_iclog_space(ticket,
2405 : &iclog, log_offset,
2406 15346404 : *len + sizeof(struct xlog_op_header),
2407 : record_cnt, data_cnt);
2408 15346553 : if (error)
2409 2978 : return error;
2410 :
2411 15343575 : ophdr = iclog->ic_datap + *log_offset;
2412 15343575 : ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2413 15343575 : ophdr->oh_clientid = XFS_TRANSACTION;
2414 15343575 : ophdr->oh_res2 = 0;
2415 15343575 : ophdr->oh_flags = XLOG_WAS_CONT_TRANS;
2416 :
2417 15343575 : ticket->t_curr_res -= sizeof(struct xlog_op_header);
2418 15343575 : *log_offset += sizeof(struct xlog_op_header);
2419 15343575 : *data_cnt += sizeof(struct xlog_op_header);
2420 :
2421 : /*
2422 : * If rlen fits in the iclog, then end the region
2423 : * continuation. Otherwise we're going around again.
2424 : */
2425 15343575 : reg_offset += rlen;
2426 15343575 : rlen = reg->i_len - reg_offset;
2427 15343575 : if (rlen <= iclog->ic_size - *log_offset)
2428 15342919 : ophdr->oh_flags |= XLOG_END_TRANS;
2429 : else
2430 656 : ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
2431 :
2432 15343575 : rlen = min_t(uint32_t, rlen, iclog->ic_size - *log_offset);
2433 15343575 : ophdr->oh_len = cpu_to_be32(rlen);
2434 :
2435 15343575 : xlog_write_iovec(iclog, log_offset,
2436 15343575 : reg->i_addr + reg_offset,
2437 : rlen, len, record_cnt, data_cnt);
2438 :
2439 15343572 : } while (ophdr->oh_flags & XLOG_CONTINUE_TRANS);
2440 : }
2441 :
2442 : /*
2443 : * No more iovecs remain in this logvec so return the next log vec to
2444 : * the caller so it can go back to fast path copying.
2445 : */
2446 16107303 : *iclogp = iclog;
2447 16107303 : return 0;
2448 : }
2449 :
2450 : /*
2451 : * Write some region out to in-core log
2452 : *
2453 : * This will be called when writing externally provided regions or when
2454 : * writing out a commit record for a given transaction.
2455 : *
2456 : * General algorithm:
2457 : * 1. Find total length of this write. This may include adding to the
2458 : * lengths passed in.
2459 : * 2. Check whether we violate the tickets reservation.
2460 : * 3. While writing to this iclog
2461 : * A. Reserve as much space in this iclog as can get
2462 : * B. If this is first write, save away start lsn
2463 : * C. While writing this region:
2464 : * 1. If first write of transaction, write start record
2465 : * 2. Write log operation header (header per region)
2466 : * 3. Find out if we can fit entire region into this iclog
2467 : * 4. Potentially, verify destination memcpy ptr
2468 : * 5. Memcpy (partial) region
2469 : * 6. If partial copy, release iclog; otherwise, continue
2470 : * copying more regions into current iclog
2471 : * 4. Mark want sync bit (in simulation mode)
2472 : * 5. Release iclog for potential flush to on-disk log.
2473 : *
2474 : * ERRORS:
2475 : * 1. Panic if reservation is overrun. This should never happen since
2476 : * reservation amounts are generated internal to the filesystem.
2477 : * NOTES:
2478 : * 1. Tickets are single threaded data structures.
2479 : * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
2480 : * syncing routine. When a single log_write region needs to span
2481 : * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
2482 : * on all log operation writes which don't contain the end of the
2483 : * region. The XLOG_END_TRANS bit is used for the in-core log
2484 : * operation which contains the end of the continued log_write region.
2485 : * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
2486 : * we don't really know exactly how much space will be used. As a result,
2487 : * we don't update ic_offset until the end when we know exactly how many
2488 : * bytes have been written out.
2489 : */
2490 : int
2491 10249224 : xlog_write(
2492 : struct xlog *log,
2493 : struct xfs_cil_ctx *ctx,
2494 : struct list_head *lv_chain,
2495 : struct xlog_ticket *ticket,
2496 : uint32_t len)
2497 :
2498 : {
2499 10249224 : struct xlog_in_core *iclog = NULL;
2500 10249224 : struct xfs_log_vec *lv;
2501 10249224 : uint32_t record_cnt = 0;
2502 10249224 : uint32_t data_cnt = 0;
2503 10249224 : int error = 0;
2504 10249224 : int log_offset;
2505 :
2506 10249224 : if (ticket->t_curr_res < 0) {
2507 2 : xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
2508 : "ctx ticket reservation ran out. Need to up reservation");
2509 2 : xlog_print_tic_res(log->l_mp, ticket);
2510 2 : xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
2511 : }
2512 :
2513 10249224 : error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
2514 : &log_offset);
2515 10249226 : if (error)
2516 : return error;
2517 :
2518 10249224 : ASSERT(log_offset <= iclog->ic_size - 1);
2519 :
2520 : /*
2521 : * If we have a context pointer, pass it the first iclog we are
2522 : * writing to so it can record state needed for iclog write
2523 : * ordering.
2524 : */
2525 10249224 : if (ctx)
2526 10200529 : xlog_cil_set_ctx_write_state(ctx, iclog);
2527 :
2528 670260961 : list_for_each_entry(lv, lv_chain, lv_list) {
2529 : /*
2530 : * If the entire log vec does not fit in the iclog, punt it to
2531 : * the partial copy loop which can handle this case.
2532 : */
2533 660014813 : if (lv->lv_niovecs &&
2534 658802053 : lv->lv_bytes > iclog->ic_size - log_offset) {
2535 16110402 : error = xlog_write_partial(lv, ticket, &iclog,
2536 : &log_offset, &len, &record_cnt,
2537 : &data_cnt);
2538 16110347 : if (error) {
2539 : /*
2540 : * We have no iclog to release, so just return
2541 : * the error immediately.
2542 : */
2543 3076 : return error;
2544 : }
2545 : } else {
2546 643904411 : xlog_write_full(lv, ticket, iclog, &log_offset,
2547 : &len, &record_cnt, &data_cnt);
2548 : }
2549 : }
2550 10246148 : ASSERT(len == 0);
2551 :
2552 : /*
2553 : * We've already been guaranteed that the last writes will fit inside
2554 : * the current iclog, and hence it will already have the space used by
2555 : * those writes accounted to it. Hence we do not need to update the
2556 : * iclog with the number of bytes written here.
2557 : */
2558 10246148 : spin_lock(&log->l_icloglock);
2559 10246149 : xlog_state_finish_copy(log, iclog, record_cnt, 0);
2560 10246149 : error = xlog_state_release_iclog(log, iclog, ticket);
2561 10246146 : spin_unlock(&log->l_icloglock);
2562 :
2563 10246146 : return error;
2564 : }
2565 :
2566 : static void
2567 21151092 : xlog_state_activate_iclog(
2568 : struct xlog_in_core *iclog,
2569 : int *iclogs_changed)
2570 : {
2571 42302184 : ASSERT(list_empty_careful(&iclog->ic_callbacks));
2572 21151092 : trace_xlog_iclog_activate(iclog, _RET_IP_);
2573 :
2574 : /*
2575 : * If the number of ops in this iclog indicate it just contains the
2576 : * dummy transaction, we can change state into IDLE (the second time
2577 : * around). Otherwise we should change the state into NEED a dummy.
2578 : * We don't need to cover the dummy.
2579 : */
2580 21151092 : if (*iclogs_changed == 0 &&
2581 21151092 : iclog->ic_header.h_num_logops == cpu_to_be32(XLOG_COVER_OPS)) {
2582 247121 : *iclogs_changed = 1;
2583 : } else {
2584 : /*
2585 : * We have two dirty iclogs so start over. This could also be
2586 : * num of ops indicating this is not the dummy going out.
2587 : */
2588 20903971 : *iclogs_changed = 2;
2589 : }
2590 :
2591 21151092 : iclog->ic_state = XLOG_STATE_ACTIVE;
2592 21151092 : iclog->ic_offset = 0;
2593 21151092 : iclog->ic_header.h_num_logops = 0;
2594 21151092 : memset(iclog->ic_header.h_cycle_data, 0,
2595 : sizeof(iclog->ic_header.h_cycle_data));
2596 21151092 : iclog->ic_header.h_lsn = 0;
2597 21151092 : iclog->ic_header.h_tail_lsn = 0;
2598 21151092 : }
2599 :
2600 : /*
2601 : * Loop through all iclogs and mark all iclogs currently marked DIRTY as
2602 : * ACTIVE after iclog I/O has completed.
2603 : */
2604 : static void
2605 21151092 : xlog_state_activate_iclogs(
2606 : struct xlog *log,
2607 : int *iclogs_changed)
2608 : {
2609 21151092 : struct xlog_in_core *iclog = log->l_iclog;
2610 :
2611 114038359 : do {
2612 114038359 : if (iclog->ic_state == XLOG_STATE_DIRTY)
2613 21151092 : xlog_state_activate_iclog(iclog, iclogs_changed);
2614 : /*
2615 : * The ordering of marking iclogs ACTIVE must be maintained, so
2616 : * an iclog doesn't become ACTIVE beyond one that is SYNCING.
2617 : */
2618 92887267 : else if (iclog->ic_state != XLOG_STATE_ACTIVE)
2619 : break;
2620 98858046 : } while ((iclog = iclog->ic_next) != log->l_iclog);
2621 21151092 : }
2622 :
2623 : static int
2624 21151092 : xlog_covered_state(
2625 : int prev_state,
2626 : int iclogs_changed)
2627 : {
2628 : /*
2629 : * We go to NEED for any non-covering writes. We go to NEED2 if we just
2630 : * wrote the first covering record (DONE). We go to IDLE if we just
2631 : * wrote the second covering record (DONE2) and remain in IDLE until a
2632 : * non-covering write occurs.
2633 : */
2634 21151092 : switch (prev_state) {
2635 173036 : case XLOG_STATE_COVER_IDLE:
2636 173036 : if (iclogs_changed == 1)
2637 24301 : return XLOG_STATE_COVER_IDLE;
2638 : fallthrough;
2639 : case XLOG_STATE_COVER_NEED:
2640 : case XLOG_STATE_COVER_NEED2:
2641 : break;
2642 91584 : case XLOG_STATE_COVER_DONE:
2643 91584 : if (iclogs_changed == 1)
2644 91584 : return XLOG_STATE_COVER_NEED2;
2645 : break;
2646 91550 : case XLOG_STATE_COVER_DONE2:
2647 91550 : if (iclogs_changed == 1)
2648 91550 : return XLOG_STATE_COVER_IDLE;
2649 : break;
2650 0 : default:
2651 0 : ASSERT(0);
2652 : }
2653 :
2654 : return XLOG_STATE_COVER_NEED;
2655 : }
2656 :
2657 : STATIC void
2658 21151092 : xlog_state_clean_iclog(
2659 : struct xlog *log,
2660 : struct xlog_in_core *dirty_iclog)
2661 : {
2662 21151092 : int iclogs_changed = 0;
2663 :
2664 21151092 : trace_xlog_iclog_clean(dirty_iclog, _RET_IP_);
2665 :
2666 21151092 : dirty_iclog->ic_state = XLOG_STATE_DIRTY;
2667 :
2668 21151092 : xlog_state_activate_iclogs(log, &iclogs_changed);
2669 21151092 : wake_up_all(&dirty_iclog->ic_force_wait);
2670 :
2671 21151092 : if (iclogs_changed) {
2672 21151092 : log->l_covered_state = xlog_covered_state(log->l_covered_state,
2673 : iclogs_changed);
2674 : }
2675 21151092 : }
2676 :
2677 : STATIC xfs_lsn_t
2678 21151092 : xlog_get_lowest_lsn(
2679 : struct xlog *log)
2680 : {
2681 21151092 : struct xlog_in_core *iclog = log->l_iclog;
2682 21151092 : xfs_lsn_t lowest_lsn = 0, lsn;
2683 :
2684 169208616 : do {
2685 169208616 : if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2686 : iclog->ic_state == XLOG_STATE_DIRTY)
2687 77791391 : continue;
2688 :
2689 91417225 : lsn = be64_to_cpu(iclog->ic_header.h_lsn);
2690 91417225 : if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0)
2691 : lowest_lsn = lsn;
2692 169208616 : } while ((iclog = iclog->ic_next) != log->l_iclog);
2693 :
2694 21151092 : return lowest_lsn;
2695 : }
2696 :
2697 : /*
2698 : * Completion of a iclog IO does not imply that a transaction has completed, as
2699 : * transactions can be large enough to span many iclogs. We cannot change the
2700 : * tail of the log half way through a transaction as this may be the only
2701 : * transaction in the log and moving the tail to point to the middle of it
2702 : * will prevent recovery from finding the start of the transaction. Hence we
2703 : * should only update the last_sync_lsn if this iclog contains transaction
2704 : * completion callbacks on it.
2705 : *
2706 : * We have to do this before we drop the icloglock to ensure we are the only one
2707 : * that can update it.
2708 : *
2709 : * If we are moving the last_sync_lsn forwards, we also need to ensure we kick
2710 : * the reservation grant head pushing. This is due to the fact that the push
2711 : * target is bound by the current last_sync_lsn value. Hence if we have a large
2712 : * amount of log space bound up in this committing transaction then the
2713 : * last_sync_lsn value may be the limiting factor preventing tail pushing from
2714 : * freeing space in the log. Hence once we've updated the last_sync_lsn we
2715 : * should push the AIL to ensure the push target (and hence the grant head) is
2716 : * no longer bound by the old log head location and can move forwards and make
2717 : * progress again.
2718 : */
2719 : static void
2720 21151092 : xlog_state_set_callback(
2721 : struct xlog *log,
2722 : struct xlog_in_core *iclog,
2723 : xfs_lsn_t header_lsn)
2724 : {
2725 21151092 : trace_xlog_iclog_callback(iclog, _RET_IP_);
2726 21151092 : iclog->ic_state = XLOG_STATE_CALLBACK;
2727 :
2728 42302184 : ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
2729 : header_lsn) <= 0);
2730 :
2731 21151092 : if (list_empty_careful(&iclog->ic_callbacks))
2732 : return;
2733 :
2734 5065696 : atomic64_set(&log->l_last_sync_lsn, header_lsn);
2735 5065696 : xlog_grant_push_ail(log, 0);
2736 : }
2737 :
2738 : /*
2739 : * Return true if we need to stop processing, false to continue to the next
2740 : * iclog. The caller will need to run callbacks if the iclog is returned in the
2741 : * XLOG_STATE_CALLBACK state.
2742 : */
2743 : static bool
2744 200503546 : xlog_state_iodone_process_iclog(
2745 : struct xlog *log,
2746 : struct xlog_in_core *iclog)
2747 : {
2748 200503546 : xfs_lsn_t lowest_lsn;
2749 200503546 : xfs_lsn_t header_lsn;
2750 :
2751 200503546 : switch (iclog->ic_state) {
2752 : case XLOG_STATE_ACTIVE:
2753 : case XLOG_STATE_DIRTY:
2754 : /*
2755 : * Skip all iclogs in the ACTIVE & DIRTY states:
2756 : */
2757 : return false;
2758 21151092 : case XLOG_STATE_DONE_SYNC:
2759 : /*
2760 : * Now that we have an iclog that is in the DONE_SYNC state, do
2761 : * one more check here to see if we have chased our tail around.
2762 : * If this is not the lowest lsn iclog, then we will leave it
2763 : * for another completion to process.
2764 : */
2765 21151092 : header_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
2766 21151092 : lowest_lsn = xlog_get_lowest_lsn(log);
2767 21151092 : if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, header_lsn) < 0)
2768 : return false;
2769 21151092 : xlog_state_set_callback(log, iclog, header_lsn);
2770 21151092 : return false;
2771 25599197 : default:
2772 : /*
2773 : * Can only perform callbacks in order. Since this iclog is not
2774 : * in the DONE_SYNC state, we skip the rest and just try to
2775 : * clean up.
2776 : */
2777 25599197 : return true;
2778 : }
2779 : }
2780 :
2781 : /*
2782 : * Loop over all the iclogs, running attached callbacks on them. Return true if
2783 : * we ran any callbacks, indicating that we dropped the icloglock. We don't need
2784 : * to handle transient shutdown state here at all because
2785 : * xlog_state_shutdown_callbacks() will be run to do the necessary shutdown
2786 : * cleanup of the callbacks.
2787 : */
2788 : static bool
2789 37572131 : xlog_state_do_iclog_callbacks(
2790 : struct xlog *log)
2791 : __releases(&log->l_icloglock)
2792 : __acquires(&log->l_icloglock)
2793 : {
2794 37572131 : struct xlog_in_core *first_iclog = log->l_iclog;
2795 37572131 : struct xlog_in_core *iclog = first_iclog;
2796 37572131 : bool ran_callback = false;
2797 :
2798 200503546 : do {
2799 200503546 : LIST_HEAD(cb_list);
2800 :
2801 200503546 : if (xlog_state_iodone_process_iclog(log, iclog))
2802 : break;
2803 174904349 : if (iclog->ic_state != XLOG_STATE_CALLBACK) {
2804 153753257 : iclog = iclog->ic_next;
2805 153753257 : continue;
2806 : }
2807 21151092 : list_splice_init(&iclog->ic_callbacks, &cb_list);
2808 21151092 : spin_unlock(&log->l_icloglock);
2809 :
2810 21151092 : trace_xlog_iclog_callbacks_start(iclog, _RET_IP_);
2811 21151092 : xlog_cil_process_committed(&cb_list);
2812 21151092 : trace_xlog_iclog_callbacks_done(iclog, _RET_IP_);
2813 21151092 : ran_callback = true;
2814 :
2815 21151092 : spin_lock(&log->l_icloglock);
2816 21151092 : xlog_state_clean_iclog(log, iclog);
2817 21151092 : iclog = iclog->ic_next;
2818 174904349 : } while (iclog != first_iclog);
2819 :
2820 37572131 : return ran_callback;
2821 : }
2822 :
2823 :
2824 : /*
2825 : * Loop running iclog completion callbacks until there are no more iclogs in a
2826 : * state that can run callbacks.
2827 : */
2828 : STATIC void
2829 21164619 : xlog_state_do_callback(
2830 : struct xlog *log)
2831 : {
2832 21164619 : int flushcnt = 0;
2833 21164619 : int repeats = 0;
2834 :
2835 21164619 : spin_lock(&log->l_icloglock);
2836 37572131 : while (xlog_state_do_iclog_callbacks(log)) {
2837 32815050 : if (xlog_is_shutdown(log))
2838 : break;
2839 :
2840 16407512 : if (++repeats > 5000) {
2841 0 : flushcnt += repeats;
2842 0 : repeats = 0;
2843 0 : xfs_warn(log->l_mp,
2844 : "%s: possible infinite loop (%d iterations)",
2845 : __func__, flushcnt);
2846 : }
2847 : }
2848 :
2849 21164619 : if (log->l_iclog->ic_state == XLOG_STATE_ACTIVE)
2850 18305092 : wake_up_all(&log->l_flush_wait);
2851 :
2852 21164619 : spin_unlock(&log->l_icloglock);
2853 21164619 : }
2854 :
2855 :
2856 : /*
2857 : * Finish transitioning this iclog to the dirty state.
2858 : *
2859 : * Callbacks could take time, so they are done outside the scope of the
2860 : * global state machine log lock.
2861 : */
2862 : STATIC void
2863 21164619 : xlog_state_done_syncing(
2864 : struct xlog_in_core *iclog)
2865 : {
2866 21164619 : struct xlog *log = iclog->ic_log;
2867 :
2868 21164619 : spin_lock(&log->l_icloglock);
2869 21164619 : ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
2870 21164619 : trace_xlog_iclog_sync_done(iclog, _RET_IP_);
2871 :
2872 : /*
2873 : * If we got an error, either on the first buffer, or in the case of
2874 : * split log writes, on the second, we shut down the file system and
2875 : * no iclogs should ever be attempted to be written to disk again.
2876 : */
2877 42329238 : if (!xlog_is_shutdown(log)) {
2878 21151131 : ASSERT(iclog->ic_state == XLOG_STATE_SYNCING);
2879 21151131 : iclog->ic_state = XLOG_STATE_DONE_SYNC;
2880 : }
2881 :
2882 : /*
2883 : * Someone could be sleeping prior to writing out the next
2884 : * iclog buffer, we wake them all, one will get to do the
2885 : * I/O, the others get to wait for the result.
2886 : */
2887 21164619 : wake_up_all(&iclog->ic_write_wait);
2888 21164619 : spin_unlock(&log->l_icloglock);
2889 21164619 : xlog_state_do_callback(log);
2890 21164619 : }
2891 :
2892 : /*
2893 : * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2894 : * sleep. We wait on the flush queue on the head iclog as that should be
2895 : * the first iclog to complete flushing. Hence if all iclogs are syncing,
2896 : * we will wait here and all new writes will sleep until a sync completes.
2897 : *
2898 : * The in-core logs are used in a circular fashion. They are not used
2899 : * out-of-order even when an iclog past the head is free.
2900 : *
2901 : * return:
2902 : * * log_offset where xlog_write() can start writing into the in-core
2903 : * log's data space.
2904 : * * in-core log pointer to which xlog_write() should write.
2905 : * * boolean indicating this is a continued write to an in-core log.
2906 : * If this is the last write, then the in-core log's offset field
2907 : * needs to be incremented, depending on the amount of data which
2908 : * is copied.
2909 : */
2910 : STATIC int
2911 26360417 : xlog_state_get_iclog_space(
2912 : struct xlog *log,
2913 : int len,
2914 : struct xlog_in_core **iclogp,
2915 : struct xlog_ticket *ticket,
2916 : int *logoffsetp)
2917 : {
2918 30622507 : int log_offset;
2919 30622507 : xlog_rec_header_t *head;
2920 30622507 : xlog_in_core_t *iclog;
2921 :
2922 : restart:
2923 30622507 : spin_lock(&log->l_icloglock);
2924 61245258 : if (xlog_is_shutdown(log)) {
2925 3073 : spin_unlock(&log->l_icloglock);
2926 3073 : return -EIO;
2927 : }
2928 :
2929 30619556 : iclog = log->l_iclog;
2930 30619556 : if (iclog->ic_state != XLOG_STATE_ACTIVE) {
2931 4260544 : XFS_STATS_INC(log->l_mp, xs_log_noiclogs);
2932 :
2933 : /* Wait for log writes to have flushed */
2934 4260544 : xlog_wait(&log->l_flush_wait, &log->l_icloglock);
2935 4260444 : goto restart;
2936 : }
2937 :
2938 26359012 : head = &iclog->ic_header;
2939 :
2940 26359012 : atomic_inc(&iclog->ic_refcnt); /* prevents sync */
2941 26359012 : log_offset = iclog->ic_offset;
2942 :
2943 26359012 : trace_xlog_iclog_get_space(iclog, _RET_IP_);
2944 :
2945 : /* On the 1st write to an iclog, figure out lsn. This works
2946 : * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2947 : * committing to. If the offset is set, that's how many blocks
2948 : * must be written.
2949 : */
2950 26359000 : if (log_offset == 0) {
2951 21166950 : ticket->t_curr_res -= log->l_iclog_hsize;
2952 21166950 : head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2953 21166950 : head->h_lsn = cpu_to_be64(
2954 : xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
2955 21166950 : ASSERT(log->l_curr_block >= 0);
2956 : }
2957 :
2958 : /* If there is enough room to write everything, then do it. Otherwise,
2959 : * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2960 : * bit is on, so this will get flushed out. Don't update ic_offset
2961 : * until you know exactly how many bytes get copied. Therefore, wait
2962 : * until later to update ic_offset.
2963 : *
2964 : * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2965 : * can fit into remaining data section.
2966 : */
2967 26359000 : if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2968 1646 : int error = 0;
2969 :
2970 1646 : xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2971 :
2972 : /*
2973 : * If we are the only one writing to this iclog, sync it to
2974 : * disk. We need to do an atomic compare and decrement here to
2975 : * avoid racing with concurrent atomic_dec_and_lock() calls in
2976 : * xlog_state_release_iclog() when there is more than one
2977 : * reference to the iclog.
2978 : */
2979 3292 : if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1))
2980 1631 : error = xlog_state_release_iclog(log, iclog, ticket);
2981 1646 : spin_unlock(&log->l_icloglock);
2982 1646 : if (error)
2983 0 : return error;
2984 1646 : goto restart;
2985 : }
2986 :
2987 : /* Do we have enough room to write the full amount in the remainder
2988 : * of this iclog? Or must we continue a write on the next iclog and
2989 : * mark this iclog as completely taken? In the case where we switch
2990 : * iclogs (to mark it taken), this particular iclog will release/sync
2991 : * to disk in xlog_write().
2992 : */
2993 26357354 : if (len <= iclog->ic_size - iclog->ic_offset)
2994 10246139 : iclog->ic_offset += len;
2995 : else
2996 16111215 : xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2997 26357352 : *iclogp = iclog;
2998 :
2999 26357352 : ASSERT(iclog->ic_offset <= iclog->ic_size);
3000 26357352 : spin_unlock(&log->l_icloglock);
3001 :
3002 26357365 : *logoffsetp = log_offset;
3003 26357365 : return 0;
3004 : }
3005 :
3006 : /*
3007 : * The first cnt-1 times a ticket goes through here we don't need to move the
3008 : * grant write head because the permanent reservation has reserved cnt times the
3009 : * unit amount. Release part of current permanent unit reservation and reset
3010 : * current reservation to be one units worth. Also move grant reservation head
3011 : * forward.
3012 : */
3013 : void
3014 891564670 : xfs_log_ticket_regrant(
3015 : struct xlog *log,
3016 : struct xlog_ticket *ticket)
3017 : {
3018 891564670 : trace_xfs_log_ticket_regrant(log, ticket);
3019 :
3020 891550664 : if (ticket->t_cnt > 0)
3021 553790371 : ticket->t_cnt--;
3022 :
3023 891550664 : xlog_grant_sub_space(log, &log->l_reserve_head.grant,
3024 : ticket->t_curr_res);
3025 891621241 : xlog_grant_sub_space(log, &log->l_write_head.grant,
3026 : ticket->t_curr_res);
3027 891617061 : ticket->t_curr_res = ticket->t_unit_res;
3028 :
3029 891617061 : trace_xfs_log_ticket_regrant_sub(log, ticket);
3030 :
3031 : /* just return if we still have some of the pre-reserved space */
3032 891606589 : if (!ticket->t_cnt) {
3033 355116165 : xlog_grant_add_space(log, &log->l_reserve_head.grant,
3034 : ticket->t_unit_res);
3035 355115425 : trace_xfs_log_ticket_regrant_exit(log, ticket);
3036 :
3037 355112438 : ticket->t_curr_res = ticket->t_unit_res;
3038 : }
3039 :
3040 891602862 : xfs_log_ticket_put(ticket);
3041 891616150 : }
3042 :
3043 : /*
3044 : * Give back the space left from a reservation.
3045 : *
3046 : * All the information we need to make a correct determination of space left
3047 : * is present. For non-permanent reservations, things are quite easy. The
3048 : * count should have been decremented to zero. We only need to deal with the
3049 : * space remaining in the current reservation part of the ticket. If the
3050 : * ticket contains a permanent reservation, there may be left over space which
3051 : * needs to be released. A count of N means that N-1 refills of the current
3052 : * reservation can be done before we need to ask for more space. The first
3053 : * one goes to fill up the first current reservation. Once we run out of
3054 : * space, the count will stay at zero and the only space remaining will be
3055 : * in the current reservation field.
3056 : */
3057 : void
3058 1341205616 : xfs_log_ticket_ungrant(
3059 : struct xlog *log,
3060 : struct xlog_ticket *ticket)
3061 : {
3062 1341205616 : int bytes;
3063 :
3064 1341205616 : trace_xfs_log_ticket_ungrant(log, ticket);
3065 :
3066 1340925373 : if (ticket->t_cnt > 0)
3067 1162785061 : ticket->t_cnt--;
3068 :
3069 1340925373 : trace_xfs_log_ticket_ungrant_sub(log, ticket);
3070 :
3071 : /*
3072 : * If this is a permanent reservation ticket, we may be able to free
3073 : * up more space based on the remaining count.
3074 : */
3075 1340936951 : bytes = ticket->t_curr_res;
3076 1340936951 : if (ticket->t_cnt > 0) {
3077 1110313349 : ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
3078 1110313349 : bytes += ticket->t_unit_res*ticket->t_cnt;
3079 : }
3080 :
3081 1340936951 : xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes);
3082 1342370691 : xlog_grant_sub_space(log, &log->l_write_head.grant, bytes);
3083 :
3084 1342203545 : trace_xfs_log_ticket_ungrant_exit(log, ticket);
3085 :
3086 1342041859 : xfs_log_space_wake(log->l_mp);
3087 1341700738 : xfs_log_ticket_put(ticket);
3088 1341830162 : }
3089 :
3090 : /*
3091 : * This routine will mark the current iclog in the ring as WANT_SYNC and move
3092 : * the current iclog pointer to the next iclog in the ring.
3093 : */
3094 : void
3095 21165151 : xlog_state_switch_iclogs(
3096 : struct xlog *log,
3097 : struct xlog_in_core *iclog,
3098 : int eventual_size)
3099 : {
3100 21165151 : ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
3101 21165151 : assert_spin_locked(&log->l_icloglock);
3102 21165151 : trace_xlog_iclog_switch(iclog, _RET_IP_);
3103 :
3104 21165149 : if (!eventual_size)
3105 5052291 : eventual_size = iclog->ic_offset;
3106 21165149 : iclog->ic_state = XLOG_STATE_WANT_SYNC;
3107 21165149 : iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
3108 21165149 : log->l_prev_block = log->l_curr_block;
3109 21165149 : log->l_prev_cycle = log->l_curr_cycle;
3110 :
3111 : /* roll log?: ic_offset changed later */
3112 21165149 : log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
3113 :
3114 : /* Round up to next log-sunit */
3115 21165149 : if (log->l_iclog_roundoff > BBSIZE) {
3116 18133560 : uint32_t sunit_bb = BTOBB(log->l_iclog_roundoff);
3117 18133560 : log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
3118 : }
3119 :
3120 21165149 : if (log->l_curr_block >= log->l_logBBsize) {
3121 : /*
3122 : * Rewind the current block before the cycle is bumped to make
3123 : * sure that the combined LSN never transiently moves forward
3124 : * when the log wraps to the next cycle. This is to support the
3125 : * unlocked sample of these fields from xlog_valid_lsn(). Most
3126 : * other cases should acquire l_icloglock.
3127 : */
3128 8021 : log->l_curr_block -= log->l_logBBsize;
3129 8021 : ASSERT(log->l_curr_block >= 0);
3130 8021 : smp_wmb();
3131 8021 : log->l_curr_cycle++;
3132 8021 : if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
3133 0 : log->l_curr_cycle++;
3134 : }
3135 21165149 : ASSERT(iclog == log->l_iclog);
3136 21165149 : log->l_iclog = iclog->ic_next;
3137 21165149 : }
3138 :
3139 : /*
3140 : * Force the iclog to disk and check if the iclog has been completed before
3141 : * xlog_force_iclog() returns. This can happen on synchronous (e.g.
3142 : * pmem) or fast async storage because we drop the icloglock to issue the IO.
3143 : * If completion has already occurred, tell the caller so that it can avoid an
3144 : * unnecessary wait on the iclog.
3145 : */
3146 : static int
3147 4612925 : xlog_force_and_check_iclog(
3148 : struct xlog_in_core *iclog,
3149 : bool *completed)
3150 : {
3151 4612925 : xfs_lsn_t lsn = be64_to_cpu(iclog->ic_header.h_lsn);
3152 4612925 : int error;
3153 :
3154 4612925 : *completed = false;
3155 4612925 : error = xlog_force_iclog(iclog);
3156 4612925 : if (error)
3157 : return error;
3158 :
3159 : /*
3160 : * If the iclog has already been completed and reused the header LSN
3161 : * will have been rewritten by completion
3162 : */
3163 4612925 : if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn)
3164 124 : *completed = true;
3165 : return 0;
3166 : }
3167 :
3168 : /*
3169 : * Write out all data in the in-core log as of this exact moment in time.
3170 : *
3171 : * Data may be written to the in-core log during this call. However,
3172 : * we don't guarantee this data will be written out. A change from past
3173 : * implementation means this routine will *not* write out zero length LRs.
3174 : *
3175 : * Basically, we try and perform an intelligent scan of the in-core logs.
3176 : * If we determine there is no flushable data, we just return. There is no
3177 : * flushable data if:
3178 : *
3179 : * 1. the current iclog is active and has no data; the previous iclog
3180 : * is in the active or dirty state.
3181 : * 2. the current iclog is drity, and the previous iclog is in the
3182 : * active or dirty state.
3183 : *
3184 : * We may sleep if:
3185 : *
3186 : * 1. the current iclog is not in the active nor dirty state.
3187 : * 2. the current iclog dirty, and the previous iclog is not in the
3188 : * active nor dirty state.
3189 : * 3. the current iclog is active, and there is another thread writing
3190 : * to this particular iclog.
3191 : * 4. a) the current iclog is active and has no other writers
3192 : * b) when we return from flushing out this iclog, it is still
3193 : * not in the active nor dirty state.
3194 : */
3195 : int
3196 3553699 : xfs_log_force(
3197 : struct xfs_mount *mp,
3198 : uint flags)
3199 : {
3200 3553699 : struct xlog *log = mp->m_log;
3201 3553699 : struct xlog_in_core *iclog;
3202 :
3203 3553699 : XFS_STATS_INC(mp, xs_log_force);
3204 3554622 : trace_xfs_log_force(mp, 0, _RET_IP_);
3205 :
3206 3554585 : xlog_cil_force(log);
3207 :
3208 3556628 : spin_lock(&log->l_icloglock);
3209 7113464 : if (xlog_is_shutdown(log))
3210 48024 : goto out_error;
3211 :
3212 3508708 : iclog = log->l_iclog;
3213 3508708 : trace_xlog_iclog_force(iclog, _RET_IP_);
3214 :
3215 3508705 : if (iclog->ic_state == XLOG_STATE_DIRTY ||
3216 3503604 : (iclog->ic_state == XLOG_STATE_ACTIVE &&
3217 3171212 : atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) {
3218 : /*
3219 : * If the head is dirty or (active and empty), then we need to
3220 : * look at the previous iclog.
3221 : *
3222 : * If the previous iclog is active or dirty we are done. There
3223 : * is nothing to sync out. Otherwise, we attach ourselves to the
3224 : * previous iclog and go to sleep.
3225 : */
3226 1920838 : iclog = iclog->ic_prev;
3227 1587867 : } else if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3228 1582775 : if (atomic_read(&iclog->ic_refcnt) == 0) {
3229 : /* We have exclusive access to this iclog. */
3230 1250383 : bool completed;
3231 :
3232 1250383 : if (xlog_force_and_check_iclog(iclog, &completed))
3233 0 : goto out_error;
3234 :
3235 1250384 : if (completed)
3236 53 : goto out_unlock;
3237 : } else {
3238 : /*
3239 : * Someone else is still writing to this iclog, so we
3240 : * need to ensure that when they release the iclog it
3241 : * gets synced immediately as we may be waiting on it.
3242 : */
3243 332392 : xlog_state_switch_iclogs(log, iclog, 0);
3244 : }
3245 : }
3246 :
3247 : /*
3248 : * The iclog we are about to wait on may contain the checkpoint pushed
3249 : * by the above xlog_cil_force() call, but it may not have been pushed
3250 : * to disk yet. Like the ACTIVE case above, we need to make sure caches
3251 : * are flushed when this iclog is written.
3252 : */
3253 3508653 : if (iclog->ic_state == XLOG_STATE_WANT_SYNC)
3254 355368 : iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA;
3255 :
3256 3508653 : if (flags & XFS_LOG_SYNC)
3257 3434749 : return xlog_wait_on_iclog(iclog);
3258 73904 : out_unlock:
3259 73957 : spin_unlock(&log->l_icloglock);
3260 73957 : return 0;
3261 48024 : out_error:
3262 48024 : spin_unlock(&log->l_icloglock);
3263 48024 : return -EIO;
3264 : }
3265 :
3266 : /*
3267 : * Force the log to a specific LSN.
3268 : *
3269 : * If an iclog with that lsn can be found:
3270 : * If it is in the DIRTY state, just return.
3271 : * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3272 : * state and go to sleep or return.
3273 : * If it is in any other state, go to sleep or return.
3274 : *
3275 : * Synchronous forces are implemented with a wait queue. All callers trying
3276 : * to force a given lsn to disk must wait on the queue attached to the
3277 : * specific in-core log. When given in-core log finally completes its write
3278 : * to disk, that thread will wake up all threads waiting on the queue.
3279 : */
3280 : static int
3281 4034812 : xlog_force_lsn(
3282 : struct xlog *log,
3283 : xfs_lsn_t lsn,
3284 : uint flags,
3285 : int *log_flushed,
3286 : bool already_slept)
3287 : {
3288 4034812 : struct xlog_in_core *iclog;
3289 4034812 : bool completed;
3290 :
3291 4034812 : spin_lock(&log->l_icloglock);
3292 8071264 : if (xlog_is_shutdown(log))
3293 310 : goto out_error;
3294 :
3295 4035322 : iclog = log->l_iclog;
3296 6725802 : while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3297 2692697 : trace_xlog_iclog_force_lsn(iclog, _RET_IP_);
3298 2692698 : iclog = iclog->ic_next;
3299 2692698 : if (iclog == log->l_iclog)
3300 2218 : goto out_unlock;
3301 : }
3302 :
3303 4033105 : switch (iclog->ic_state) {
3304 3567259 : case XLOG_STATE_ACTIVE:
3305 : /*
3306 : * We sleep here if we haven't already slept (e.g. this is the
3307 : * first time we've looked at the correct iclog buf) and the
3308 : * buffer before us is going to be sync'ed. The reason for this
3309 : * is that if we are doing sync transactions here, by waiting
3310 : * for the previous I/O to complete, we can allow a few more
3311 : * transactions into this iclog before we close it down.
3312 : *
3313 : * Otherwise, we mark the buffer WANT_SYNC, and bump up the
3314 : * refcnt so we can release the log (which drops the ref count).
3315 : * The state switch keeps new transaction commits from using
3316 : * this buffer. When the current commits finish writing into
3317 : * the buffer, the refcount will drop to zero and the buffer
3318 : * will go out then.
3319 : */
3320 3567259 : if (!already_slept &&
3321 3415688 : (iclog->ic_prev->ic_state == XLOG_STATE_WANT_SYNC ||
3322 : iclog->ic_prev->ic_state == XLOG_STATE_SYNCING)) {
3323 204717 : xlog_wait(&iclog->ic_prev->ic_write_wait,
3324 204717 : &log->l_icloglock);
3325 204717 : return -EAGAIN;
3326 : }
3327 3362542 : if (xlog_force_and_check_iclog(iclog, &completed))
3328 0 : goto out_error;
3329 3362542 : if (log_flushed)
3330 3096619 : *log_flushed = 1;
3331 3362542 : if (completed)
3332 71 : goto out_unlock;
3333 : break;
3334 5251 : case XLOG_STATE_WANT_SYNC:
3335 : /*
3336 : * This iclog may contain the checkpoint pushed by the
3337 : * xlog_cil_force_seq() call, but there are other writers still
3338 : * accessing it so it hasn't been pushed to disk yet. Like the
3339 : * ACTIVE case above, we need to make sure caches are flushed
3340 : * when this iclog is written.
3341 : */
3342 5251 : iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA;
3343 5251 : break;
3344 : default:
3345 : /*
3346 : * The entire checkpoint was written by the CIL force and is on
3347 : * its way to disk already. It will be stable when it
3348 : * completes, so we don't need to manipulate caches here at all.
3349 : * We just need to wait for completion if necessary.
3350 : */
3351 : break;
3352 : }
3353 :
3354 3828317 : if (flags & XFS_LOG_SYNC)
3355 3828317 : return xlog_wait_on_iclog(iclog);
3356 0 : out_unlock:
3357 2289 : spin_unlock(&log->l_icloglock);
3358 2289 : return 0;
3359 310 : out_error:
3360 310 : spin_unlock(&log->l_icloglock);
3361 310 : return -EIO;
3362 : }
3363 :
3364 : /*
3365 : * Force the log to a specific checkpoint sequence.
3366 : *
3367 : * First force the CIL so that all the required changes have been flushed to the
3368 : * iclogs. If the CIL force completed it will return a commit LSN that indicates
3369 : * the iclog that needs to be flushed to stable storage. If the caller needs
3370 : * a synchronous log force, we will wait on the iclog with the LSN returned by
3371 : * xlog_cil_force_seq() to be completed.
3372 : */
3373 : int
3374 3852433 : xfs_log_force_seq(
3375 : struct xfs_mount *mp,
3376 : xfs_csn_t seq,
3377 : uint flags,
3378 : int *log_flushed)
3379 : {
3380 3852433 : struct xlog *log = mp->m_log;
3381 3852433 : xfs_lsn_t lsn;
3382 3852433 : int ret;
3383 3852433 : ASSERT(seq != 0);
3384 :
3385 3852433 : XFS_STATS_INC(mp, xs_log_force);
3386 3852480 : trace_xfs_log_force(mp, seq, _RET_IP_);
3387 :
3388 3852420 : lsn = xlog_cil_force_seq(log, seq);
3389 3852408 : if (lsn == NULLCOMMITLSN)
3390 : return 0;
3391 :
3392 3830668 : ret = xlog_force_lsn(log, lsn, flags, log_flushed, false);
3393 3830340 : if (ret == -EAGAIN) {
3394 204711 : XFS_STATS_INC(mp, xs_log_force_sleep);
3395 204714 : ret = xlog_force_lsn(log, lsn, flags, log_flushed, true);
3396 : }
3397 : return ret;
3398 : }
3399 :
3400 : /*
3401 : * Free a used ticket when its refcount falls to zero.
3402 : */
3403 : void
3404 2233240352 : xfs_log_ticket_put(
3405 : xlog_ticket_t *ticket)
3406 : {
3407 2233240352 : ASSERT(atomic_read(&ticket->t_ref) > 0);
3408 2233240352 : if (atomic_dec_and_test(&ticket->t_ref))
3409 1342145357 : kmem_cache_free(xfs_log_ticket_cache, ticket);
3410 2233483117 : }
3411 :
3412 : xlog_ticket_t *
3413 891491499 : xfs_log_ticket_get(
3414 : xlog_ticket_t *ticket)
3415 : {
3416 891491499 : ASSERT(atomic_read(&ticket->t_ref) > 0);
3417 891491499 : atomic_inc(&ticket->t_ref);
3418 891616580 : return ticket;
3419 : }
3420 :
3421 : /*
3422 : * Figure out the total log space unit (in bytes) that would be
3423 : * required for a log ticket.
3424 : */
3425 : static int
3426 1339891990 : xlog_calc_unit_res(
3427 : struct xlog *log,
3428 : int unit_bytes,
3429 : int *niclogs)
3430 : {
3431 1339891990 : int iclog_space;
3432 1339891990 : uint num_headers;
3433 :
3434 : /*
3435 : * Permanent reservations have up to 'cnt'-1 active log operations
3436 : * in the log. A unit in this case is the amount of space for one
3437 : * of these log operations. Normal reservations have a cnt of 1
3438 : * and their unit amount is the total amount of space required.
3439 : *
3440 : * The following lines of code account for non-transaction data
3441 : * which occupy space in the on-disk log.
3442 : *
3443 : * Normal form of a transaction is:
3444 : * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3445 : * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3446 : *
3447 : * We need to account for all the leadup data and trailer data
3448 : * around the transaction data.
3449 : * And then we need to account for the worst case in terms of using
3450 : * more space.
3451 : * The worst case will happen if:
3452 : * - the placement of the transaction happens to be such that the
3453 : * roundoff is at its maximum
3454 : * - the transaction data is synced before the commit record is synced
3455 : * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3456 : * Therefore the commit record is in its own Log Record.
3457 : * This can happen as the commit record is called with its
3458 : * own region to xlog_write().
3459 : * This then means that in the worst case, roundoff can happen for
3460 : * the commit-rec as well.
3461 : * The commit-rec is smaller than padding in this scenario and so it is
3462 : * not added separately.
3463 : */
3464 :
3465 : /* for trans header */
3466 1339891990 : unit_bytes += sizeof(xlog_op_header_t);
3467 1339891990 : unit_bytes += sizeof(xfs_trans_header_t);
3468 :
3469 : /* for start-rec */
3470 1339891990 : unit_bytes += sizeof(xlog_op_header_t);
3471 :
3472 : /*
3473 : * for LR headers - the space for data in an iclog is the size minus
3474 : * the space used for the headers. If we use the iclog size, then we
3475 : * undercalculate the number of headers required.
3476 : *
3477 : * Furthermore - the addition of op headers for split-recs might
3478 : * increase the space required enough to require more log and op
3479 : * headers, so take that into account too.
3480 : *
3481 : * IMPORTANT: This reservation makes the assumption that if this
3482 : * transaction is the first in an iclog and hence has the LR headers
3483 : * accounted to it, then the remaining space in the iclog is
3484 : * exclusively for this transaction. i.e. if the transaction is larger
3485 : * than the iclog, it will be the only thing in that iclog.
3486 : * Fundamentally, this means we must pass the entire log vector to
3487 : * xlog_write to guarantee this.
3488 : */
3489 1339891990 : iclog_space = log->l_iclog_size - log->l_iclog_hsize;
3490 1339891990 : num_headers = howmany(unit_bytes, iclog_space);
3491 :
3492 : /* for split-recs - ophdrs added when data split over LRs */
3493 1339891990 : unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3494 :
3495 : /* add extra header reservations if we overrun */
3496 1339202093 : while (!num_headers ||
3497 1339202093 : howmany(unit_bytes, iclog_space) > num_headers) {
3498 44667 : unit_bytes += sizeof(xlog_op_header_t);
3499 44667 : num_headers++;
3500 : }
3501 1339891990 : unit_bytes += log->l_iclog_hsize * num_headers;
3502 :
3503 : /* for commit-rec LR header - note: padding will subsume the ophdr */
3504 1339891990 : unit_bytes += log->l_iclog_hsize;
3505 :
3506 : /* roundoff padding for transaction data and one for commit record */
3507 1339891990 : unit_bytes += 2 * log->l_iclog_roundoff;
3508 :
3509 1339891990 : if (niclogs)
3510 1339832701 : *niclogs = num_headers;
3511 1339891990 : return unit_bytes;
3512 : }
3513 :
3514 : int
3515 59289 : xfs_log_calc_unit_res(
3516 : struct xfs_mount *mp,
3517 : int unit_bytes)
3518 : {
3519 59289 : return xlog_calc_unit_res(mp->m_log, unit_bytes, NULL);
3520 : }
3521 :
3522 : /*
3523 : * Allocate and initialise a new log ticket.
3524 : */
3525 : struct xlog_ticket *
3526 1338995203 : xlog_ticket_alloc(
3527 : struct xlog *log,
3528 : int unit_bytes,
3529 : int cnt,
3530 : bool permanent)
3531 : {
3532 1338995203 : struct xlog_ticket *tic;
3533 1338995203 : int unit_res;
3534 :
3535 1338995203 : tic = kmem_cache_zalloc(xfs_log_ticket_cache, GFP_NOFS | __GFP_NOFAIL);
3536 :
3537 1339784027 : unit_res = xlog_calc_unit_res(log, unit_bytes, &tic->t_iclog_hdrs);
3538 :
3539 1340282486 : atomic_set(&tic->t_ref, 1);
3540 1340282486 : tic->t_task = current;
3541 1340282486 : INIT_LIST_HEAD(&tic->t_queue);
3542 1340282486 : tic->t_unit_res = unit_res;
3543 1340282486 : tic->t_curr_res = unit_res;
3544 1340282486 : tic->t_cnt = cnt;
3545 1340282486 : tic->t_ocnt = cnt;
3546 1340282486 : tic->t_tid = get_random_u32();
3547 1339637019 : if (permanent)
3548 1173485910 : tic->t_flags |= XLOG_TIC_PERM_RESERV;
3549 :
3550 1339637019 : return tic;
3551 : }
3552 :
3553 : #if defined(DEBUG)
3554 : /*
3555 : * Check to make sure the grant write head didn't just over lap the tail. If
3556 : * the cycles are the same, we can't be overlapping. Otherwise, make sure that
3557 : * the cycles differ by exactly one and check the byte count.
3558 : *
3559 : * This check is run unlocked, so can give false positives. Rather than assert
3560 : * on failures, use a warn-once flag and a panic tag to allow the admin to
3561 : * determine if they want to panic the machine when such an error occurs. For
3562 : * debug kernels this will have the same effect as using an assert but, unlinke
3563 : * an assert, it can be turned off at runtime.
3564 : */
3565 : STATIC void
3566 1691656436 : xlog_verify_grant_tail(
3567 : struct xlog *log)
3568 : {
3569 1691656436 : int tail_cycle, tail_blocks;
3570 1691656436 : int cycle, space;
3571 :
3572 1691656436 : xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space);
3573 1691656436 : xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
3574 1691656436 : if (tail_cycle != cycle) {
3575 341826657 : if (cycle - 1 != tail_cycle &&
3576 72 : !test_and_set_bit(XLOG_TAIL_WARN, &log->l_opstate)) {
3577 0 : xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3578 : "%s: cycle - 1 != tail_cycle", __func__);
3579 : }
3580 :
3581 341830126 : if (space > BBTOB(tail_blocks) &&
3582 3315 : !test_and_set_bit(XLOG_TAIL_WARN, &log->l_opstate)) {
3583 75 : xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3584 : "%s: space > BBTOB(tail_blocks)", __func__);
3585 : }
3586 : }
3587 1691656667 : }
3588 :
3589 : /* check if it will fit */
3590 : STATIC void
3591 21164614 : xlog_verify_tail_lsn(
3592 : struct xlog *log,
3593 : struct xlog_in_core *iclog)
3594 : {
3595 21164614 : xfs_lsn_t tail_lsn = be64_to_cpu(iclog->ic_header.h_tail_lsn);
3596 21164614 : int blocks;
3597 :
3598 21164614 : if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3599 14337520 : blocks =
3600 14337520 : log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3601 14337520 : if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
3602 0 : xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
3603 : } else {
3604 6827094 : ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3605 :
3606 6827094 : if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
3607 0 : xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
3608 :
3609 6827094 : blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3610 6827094 : if (blocks < BTOBB(iclog->ic_offset) + 1)
3611 0 : xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
3612 : }
3613 21164614 : }
3614 :
3615 : /*
3616 : * Perform a number of checks on the iclog before writing to disk.
3617 : *
3618 : * 1. Make sure the iclogs are still circular
3619 : * 2. Make sure we have a good magic number
3620 : * 3. Make sure we don't have magic numbers in the data
3621 : * 4. Check fields of each log operation header for:
3622 : * A. Valid client identifier
3623 : * B. tid ptr value falls in valid ptr space (user space code)
3624 : * C. Length in log record header is correct according to the
3625 : * individual operation headers within record.
3626 : * 5. When a bwrite will occur within 5 blocks of the front of the physical
3627 : * log, check the preceding blocks of the physical log to make sure all
3628 : * the cycle numbers agree with the current cycle number.
3629 : */
3630 : STATIC void
3631 21163542 : xlog_verify_iclog(
3632 : struct xlog *log,
3633 : struct xlog_in_core *iclog,
3634 : int count)
3635 : {
3636 21163542 : xlog_op_header_t *ophead;
3637 21163542 : xlog_in_core_t *icptr;
3638 21163542 : xlog_in_core_2_t *xhdr;
3639 21163542 : void *base_ptr, *ptr, *p;
3640 21163542 : ptrdiff_t field_offset;
3641 21163542 : uint8_t clientid;
3642 21163542 : int len, i, j, k, op_len;
3643 21163542 : int idx;
3644 :
3645 : /* check validity of iclog pointers */
3646 21163542 : spin_lock(&log->l_icloglock);
3647 21164616 : icptr = log->l_iclog;
3648 190481382 : for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next)
3649 169316764 : ASSERT(icptr);
3650 :
3651 21164618 : if (icptr != log->l_iclog)
3652 0 : xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
3653 21164618 : spin_unlock(&log->l_icloglock);
3654 :
3655 : /* check log magic numbers */
3656 21164618 : if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
3657 0 : xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
3658 :
3659 21164618 : base_ptr = ptr = &iclog->ic_header;
3660 21164618 : p = &iclog->ic_header;
3661 1144013189 : for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) {
3662 1122848595 : if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
3663 0 : xfs_emerg(log->l_mp, "%s: unexpected magic num",
3664 : __func__);
3665 : }
3666 :
3667 : /* check fields */
3668 21164594 : len = be32_to_cpu(iclog->ic_header.h_num_logops);
3669 21164594 : base_ptr = ptr = iclog->ic_datap;
3670 21164594 : ophead = ptr;
3671 21164594 : xhdr = iclog->ic_data;
3672 1717559824 : for (i = 0; i < len; i++) {
3673 1696395282 : ophead = ptr;
3674 :
3675 : /* clientid is only 1 byte */
3676 1696395282 : p = &ophead->oh_clientid;
3677 1696395282 : field_offset = p - base_ptr;
3678 1696395282 : if (field_offset & 0x1ff) {
3679 1683904396 : clientid = ophead->oh_clientid;
3680 : } else {
3681 12490886 : idx = BTOBBT((void *)&ophead->oh_clientid - iclog->ic_datap);
3682 12490886 : if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3683 15521 : j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3684 15521 : k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3685 15521 : clientid = xlog_get_client_id(
3686 15521 : xhdr[j].hic_xheader.xh_cycle_data[k]);
3687 : } else {
3688 12475370 : clientid = xlog_get_client_id(
3689 12475365 : iclog->ic_header.h_cycle_data[idx]);
3690 : }
3691 : }
3692 1696395287 : if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) {
3693 0 : xfs_warn(log->l_mp,
3694 : "%s: op %d invalid clientid %d op "PTR_FMT" offset 0x%lx",
3695 : __func__, i, clientid, ophead,
3696 : (unsigned long)field_offset);
3697 : }
3698 :
3699 : /* check length */
3700 1696395222 : p = &ophead->oh_len;
3701 1696395222 : field_offset = p - base_ptr;
3702 1696395222 : if (field_offset & 0x1ff) {
3703 1683648751 : op_len = be32_to_cpu(ophead->oh_len);
3704 : } else {
3705 12746471 : idx = BTOBBT((void *)&ophead->oh_len - iclog->ic_datap);
3706 12746471 : if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3707 15472 : j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3708 15472 : k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3709 15472 : op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
3710 : } else {
3711 12730999 : op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
3712 : }
3713 : }
3714 1696395230 : ptr += sizeof(xlog_op_header_t) + op_len;
3715 : }
3716 21164542 : }
3717 : #endif
3718 :
3719 : /*
3720 : * Perform a forced shutdown on the log.
3721 : *
3722 : * This can be called from low level log code to trigger a shutdown, or from the
3723 : * high level mount shutdown code when the mount shuts down.
3724 : *
3725 : * Our main objectives here are to make sure that:
3726 : * a. if the shutdown was not due to a log IO error, flush the logs to
3727 : * disk. Anything modified after this is ignored.
3728 : * b. the log gets atomically marked 'XLOG_IO_ERROR' for all interested
3729 : * parties to find out. Nothing new gets queued after this is done.
3730 : * c. Tasks sleeping on log reservations, pinned objects and
3731 : * other resources get woken up.
3732 : * d. The mount is also marked as shut down so that log triggered shutdowns
3733 : * still behave the same as if they called xfs_forced_shutdown().
3734 : *
3735 : * Return true if the shutdown cause was a log IO error and we actually shut the
3736 : * log down.
3737 : */
3738 : bool
3739 21650 : xlog_force_shutdown(
3740 : struct xlog *log,
3741 : uint32_t shutdown_flags)
3742 : {
3743 21650 : bool log_error = (shutdown_flags & SHUTDOWN_LOG_IO_ERROR);
3744 :
3745 21650 : if (!log)
3746 : return false;
3747 :
3748 : /*
3749 : * Flush all the completed transactions to disk before marking the log
3750 : * being shut down. We need to do this first as shutting down the log
3751 : * before the force will prevent the log force from flushing the iclogs
3752 : * to disk.
3753 : *
3754 : * When we are in recovery, there are no transactions to flush, and
3755 : * we don't want to touch the log because we don't want to perturb the
3756 : * current head/tail for future recovery attempts. Hence we need to
3757 : * avoid a log force in this case.
3758 : *
3759 : * If we are shutting down due to a log IO error, then we must avoid
3760 : * trying to write the log as that may just result in more IO errors and
3761 : * an endless shutdown/force loop.
3762 : */
3763 26070 : if (!log_error && !xlog_in_recovery(log))
3764 4420 : xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3765 :
3766 : /*
3767 : * Atomically set the shutdown state. If the shutdown state is already
3768 : * set, there someone else is performing the shutdown and so we are done
3769 : * here. This should never happen because we should only ever get called
3770 : * once by the first shutdown caller.
3771 : *
3772 : * Much of the log state machine transitions assume that shutdown state
3773 : * cannot change once they hold the log->l_icloglock. Hence we need to
3774 : * hold that lock here, even though we use the atomic test_and_set_bit()
3775 : * operation to set the shutdown state.
3776 : */
3777 21650 : spin_lock(&log->l_icloglock);
3778 21650 : if (test_and_set_bit(XLOG_IO_ERROR, &log->l_opstate)) {
3779 8522 : spin_unlock(&log->l_icloglock);
3780 8522 : return false;
3781 : }
3782 13128 : spin_unlock(&log->l_icloglock);
3783 :
3784 : /*
3785 : * If this log shutdown also sets the mount shutdown state, issue a
3786 : * shutdown warning message.
3787 : */
3788 13128 : if (!test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &log->l_mp->m_opstate)) {
3789 4014 : xfs_alert_tag(log->l_mp, XFS_PTAG_SHUTDOWN_LOGERROR,
3790 : "Filesystem has been shut down due to log error (0x%x).",
3791 : shutdown_flags);
3792 4014 : xfs_alert(log->l_mp,
3793 : "Please unmount the filesystem and rectify the problem(s).");
3794 4014 : if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
3795 0 : xfs_stack_trace();
3796 : }
3797 :
3798 : /*
3799 : * We don't want anybody waiting for log reservations after this. That
3800 : * means we have to wake up everybody queued up on reserveq as well as
3801 : * writeq. In addition, we make sure in xlog_{re}grant_log_space that
3802 : * we don't enqueue anything once the SHUTDOWN flag is set, and this
3803 : * action is protected by the grant locks.
3804 : */
3805 13128 : xlog_grant_head_wake_all(&log->l_reserve_head);
3806 13128 : xlog_grant_head_wake_all(&log->l_write_head);
3807 :
3808 : /*
3809 : * Wake up everybody waiting on xfs_log_force. Wake the CIL push first
3810 : * as if the log writes were completed. The abort handling in the log
3811 : * item committed callback functions will do this again under lock to
3812 : * avoid races.
3813 : */
3814 13128 : spin_lock(&log->l_cilp->xc_push_lock);
3815 13128 : wake_up_all(&log->l_cilp->xc_start_wait);
3816 13128 : wake_up_all(&log->l_cilp->xc_commit_wait);
3817 13128 : spin_unlock(&log->l_cilp->xc_push_lock);
3818 :
3819 13128 : spin_lock(&log->l_icloglock);
3820 13128 : xlog_state_shutdown_callbacks(log);
3821 13128 : spin_unlock(&log->l_icloglock);
3822 :
3823 13128 : wake_up_var(&log->l_opstate);
3824 13128 : return log_error;
3825 : }
3826 :
3827 : STATIC int
3828 307558 : xlog_iclogs_empty(
3829 : struct xlog *log)
3830 : {
3831 307558 : xlog_in_core_t *iclog;
3832 :
3833 307558 : iclog = log->l_iclog;
3834 2443687 : do {
3835 : /* endianness does not matter here, zero is zero in
3836 : * any language.
3837 : */
3838 2443687 : if (iclog->ic_header.h_num_logops)
3839 : return 0;
3840 2436553 : iclog = iclog->ic_next;
3841 2436553 : } while (iclog != log->l_iclog);
3842 : return 1;
3843 : }
3844 :
3845 : /*
3846 : * Verify that an LSN stamped into a piece of metadata is valid. This is
3847 : * intended for use in read verifiers on v5 superblocks.
3848 : */
3849 : bool
3850 155773640 : xfs_log_check_lsn(
3851 : struct xfs_mount *mp,
3852 : xfs_lsn_t lsn)
3853 : {
3854 155773640 : struct xlog *log = mp->m_log;
3855 155773640 : bool valid;
3856 :
3857 : /*
3858 : * norecovery mode skips mount-time log processing and unconditionally
3859 : * resets the in-core LSN. We can't validate in this mode, but
3860 : * modifications are not allowed anyways so just return true.
3861 : */
3862 155773640 : if (xfs_has_norecovery(mp))
3863 : return true;
3864 :
3865 : /*
3866 : * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is
3867 : * handled by recovery and thus safe to ignore here.
3868 : */
3869 155773007 : if (lsn == NULLCOMMITLSN)
3870 : return true;
3871 :
3872 155196848 : valid = xlog_valid_lsn(mp->m_log, lsn);
3873 :
3874 : /* warn the user about what's gone wrong before verifier failure */
3875 155262747 : if (!valid) {
3876 11 : spin_lock(&log->l_icloglock);
3877 11 : xfs_warn(mp,
3878 : "Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). "
3879 : "Please unmount and run xfs_repair (>= v4.3) to resolve.",
3880 : CYCLE_LSN(lsn), BLOCK_LSN(lsn),
3881 : log->l_curr_cycle, log->l_curr_block);
3882 11 : spin_unlock(&log->l_icloglock);
3883 : }
3884 :
3885 : return valid;
3886 : }
3887 :
3888 : /*
3889 : * Notify the log that we're about to start using a feature that is protected
3890 : * by a log incompat feature flag. This will prevent log covering from
3891 : * clearing those flags.
3892 : */
3893 : void
3894 2726859 : xlog_use_incompat_feat(
3895 : struct xlog *log)
3896 : {
3897 2726859 : down_read(&log->l_incompat_users);
3898 2731743 : }
3899 :
3900 : /* Notify the log that we've finished using log incompat features. */
3901 : void
3902 2732284 : xlog_drop_incompat_feat(
3903 : struct xlog *log)
3904 : {
3905 2732284 : up_read(&log->l_incompat_users);
3906 2734380 : }
|
