~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/xfs/xfs_log.c

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
  3  * All Rights Reserved.
  4  *
  5  * This program is free software; you can redistribute it and/or
  6  * modify it under the terms of the GNU General Public License as
  7  * published by the Free Software Foundation.
  8  *
  9  * This program is distributed in the hope that it would be useful,
 10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12  * GNU General Public License for more details.
 13  *
 14  * You should have received a copy of the GNU General Public License
 15  * along with this program; if not, write the Free Software Foundation,
 16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 17  */
 18 #include "xfs.h"
 19 #include "xfs_fs.h"
 20 #include "xfs_shared.h"
 21 #include "xfs_format.h"
 22 #include "xfs_log_format.h"
 23 #include "xfs_trans_resv.h"
 24 #include "xfs_sb.h"
 25 #include "xfs_ag.h"
 26 #include "xfs_mount.h"
 27 #include "xfs_error.h"
 28 #include "xfs_trans.h"
 29 #include "xfs_trans_priv.h"
 30 #include "xfs_log.h"
 31 #include "xfs_log_priv.h"
 32 #include "xfs_log_recover.h"
 33 #include "xfs_inode.h"
 34 #include "xfs_trace.h"
 35 #include "xfs_fsops.h"
 36 #include "xfs_cksum.h"
 37 
 38 kmem_zone_t     *xfs_log_ticket_zone;
 39 
 40 /* Local miscellaneous function prototypes */
 41 STATIC int
 42 xlog_commit_record(
 43         struct xlog             *log,
 44         struct xlog_ticket      *ticket,
 45         struct xlog_in_core     **iclog,
 46         xfs_lsn_t               *commitlsnp);
 47 
 48 STATIC struct xlog *
 49 xlog_alloc_log(
 50         struct xfs_mount        *mp,
 51         struct xfs_buftarg      *log_target,
 52         xfs_daddr_t             blk_offset,
 53         int                     num_bblks);
 54 STATIC int
 55 xlog_space_left(
 56         struct xlog             *log,
 57         atomic64_t              *head);
 58 STATIC int
 59 xlog_sync(
 60         struct xlog             *log,
 61         struct xlog_in_core     *iclog);
 62 STATIC void
 63 xlog_dealloc_log(
 64         struct xlog             *log);
 65 
 66 /* local state machine functions */
 67 STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
 68 STATIC void
 69 xlog_state_do_callback(
 70         struct xlog             *log,
 71         int                     aborted,
 72         struct xlog_in_core     *iclog);
 73 STATIC int
 74 xlog_state_get_iclog_space(
 75         struct xlog             *log,
 76         int                     len,
 77         struct xlog_in_core     **iclog,
 78         struct xlog_ticket      *ticket,
 79         int                     *continued_write,
 80         int                     *logoffsetp);
 81 STATIC int
 82 xlog_state_release_iclog(
 83         struct xlog             *log,
 84         struct xlog_in_core     *iclog);
 85 STATIC void
 86 xlog_state_switch_iclogs(
 87         struct xlog             *log,
 88         struct xlog_in_core     *iclog,
 89         int                     eventual_size);
 90 STATIC void
 91 xlog_state_want_sync(
 92         struct xlog             *log,
 93         struct xlog_in_core     *iclog);
 94 
 95 STATIC void
 96 xlog_grant_push_ail(
 97         struct xlog             *log,
 98         int                     need_bytes);
 99 STATIC void
100 xlog_regrant_reserve_log_space(
101         struct xlog             *log,
102         struct xlog_ticket      *ticket);
103 STATIC void
104 xlog_ungrant_log_space(
105         struct xlog             *log,
106         struct xlog_ticket      *ticket);
107 
108 #if defined(DEBUG)
109 STATIC void
110 xlog_verify_dest_ptr(
111         struct xlog             *log,
112         char                    *ptr);
113 STATIC void
114 xlog_verify_grant_tail(
115         struct xlog *log);
116 STATIC void
117 xlog_verify_iclog(
118         struct xlog             *log,
119         struct xlog_in_core     *iclog,
120         int                     count,
121         bool                    syncing);
122 STATIC void
123 xlog_verify_tail_lsn(
124         struct xlog             *log,
125         struct xlog_in_core     *iclog,
126         xfs_lsn_t               tail_lsn);
127 #else
128 #define xlog_verify_dest_ptr(a,b)
129 #define xlog_verify_grant_tail(a)
130 #define xlog_verify_iclog(a,b,c,d)
131 #define xlog_verify_tail_lsn(a,b,c)
132 #endif
133 
134 STATIC int
135 xlog_iclogs_empty(
136         struct xlog             *log);
137 
138 static void
139 xlog_grant_sub_space(
140         struct xlog             *log,
141         atomic64_t              *head,
142         int                     bytes)
143 {
144         int64_t head_val = atomic64_read(head);
145         int64_t new, old;
146 
147         do {
148                 int     cycle, space;
149 
150                 xlog_crack_grant_head_val(head_val, &cycle, &space);
151 
152                 space -= bytes;
153                 if (space < 0) {
154                         space += log->l_logsize;
155                         cycle--;
156                 }
157 
158                 old = head_val;
159                 new = xlog_assign_grant_head_val(cycle, space);
160                 head_val = atomic64_cmpxchg(head, old, new);
161         } while (head_val != old);
162 }
163 
164 static void
165 xlog_grant_add_space(
166         struct xlog             *log,
167         atomic64_t              *head,
168         int                     bytes)
169 {
170         int64_t head_val = atomic64_read(head);
171         int64_t new, old;
172 
173         do {
174                 int             tmp;
175                 int             cycle, space;
176 
177                 xlog_crack_grant_head_val(head_val, &cycle, &space);
178 
179                 tmp = log->l_logsize - space;
180                 if (tmp > bytes)
181                         space += bytes;
182                 else {
183                         space = bytes - tmp;
184                         cycle++;
185                 }
186 
187                 old = head_val;
188                 new = xlog_assign_grant_head_val(cycle, space);
189                 head_val = atomic64_cmpxchg(head, old, new);
190         } while (head_val != old);
191 }
192 
193 STATIC void
194 xlog_grant_head_init(
195         struct xlog_grant_head  *head)
196 {
197         xlog_assign_grant_head(&head->grant, 1, 0);
198         INIT_LIST_HEAD(&head->waiters);
199         spin_lock_init(&head->lock);
200 }
201 
202 STATIC void
203 xlog_grant_head_wake_all(
204         struct xlog_grant_head  *head)
205 {
206         struct xlog_ticket      *tic;
207 
208         spin_lock(&head->lock);
209         list_for_each_entry(tic, &head->waiters, t_queue)
210                 wake_up_process(tic->t_task);
211         spin_unlock(&head->lock);
212 }
213 
214 static inline int
215 xlog_ticket_reservation(
216         struct xlog             *log,
217         struct xlog_grant_head  *head,
218         struct xlog_ticket      *tic)
219 {
220         if (head == &log->l_write_head) {
221                 ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
222                 return tic->t_unit_res;
223         } else {
224                 if (tic->t_flags & XLOG_TIC_PERM_RESERV)
225                         return tic->t_unit_res * tic->t_cnt;
226                 else
227                         return tic->t_unit_res;
228         }
229 }
230 
231 STATIC bool
232 xlog_grant_head_wake(
233         struct xlog             *log,
234         struct xlog_grant_head  *head,
235         int                     *free_bytes)
236 {
237         struct xlog_ticket      *tic;
238         int                     need_bytes;
239 
240         list_for_each_entry(tic, &head->waiters, t_queue) {
241                 need_bytes = xlog_ticket_reservation(log, head, tic);
242                 if (*free_bytes < need_bytes)
243                         return false;
244 
245                 *free_bytes -= need_bytes;
246                 trace_xfs_log_grant_wake_up(log, tic);
247                 wake_up_process(tic->t_task);
248         }
249 
250         return true;
251 }
252 
253 STATIC int
254 xlog_grant_head_wait(
255         struct xlog             *log,
256         struct xlog_grant_head  *head,
257         struct xlog_ticket      *tic,
258         int                     need_bytes) __releases(&head->lock)
259                                             __acquires(&head->lock)
260 {
261         list_add_tail(&tic->t_queue, &head->waiters);
262 
263         do {
264                 if (XLOG_FORCED_SHUTDOWN(log))
265                         goto shutdown;
266                 xlog_grant_push_ail(log, need_bytes);
267 
268                 __set_current_state(TASK_UNINTERRUPTIBLE);
269                 spin_unlock(&head->lock);
270 
271                 XFS_STATS_INC(xs_sleep_logspace);
272 
273                 trace_xfs_log_grant_sleep(log, tic);
274                 schedule();
275                 trace_xfs_log_grant_wake(log, tic);
276 
277                 spin_lock(&head->lock);
278                 if (XLOG_FORCED_SHUTDOWN(log))
279                         goto shutdown;
280         } while (xlog_space_left(log, &head->grant) < need_bytes);
281 
282         list_del_init(&tic->t_queue);
283         return 0;
284 shutdown:
285         list_del_init(&tic->t_queue);
286         return XFS_ERROR(EIO);
287 }
288 
289 /*
290  * Atomically get the log space required for a log ticket.
291  *
292  * Once a ticket gets put onto head->waiters, it will only return after the
293  * needed reservation is satisfied.
294  *
295  * This function is structured so that it has a lock free fast path. This is
296  * necessary because every new transaction reservation will come through this
297  * path. Hence any lock will be globally hot if we take it unconditionally on
298  * every pass.
299  *
300  * As tickets are only ever moved on and off head->waiters under head->lock, we
301  * only need to take that lock if we are going to add the ticket to the queue
302  * and sleep. We can avoid taking the lock if the ticket was never added to
303  * head->waiters because the t_queue list head will be empty and we hold the
304  * only reference to it so it can safely be checked unlocked.
305  */
306 STATIC int
307 xlog_grant_head_check(
308         struct xlog             *log,
309         struct xlog_grant_head  *head,
310         struct xlog_ticket      *tic,
311         int                     *need_bytes)
312 {
313         int                     free_bytes;
314         int                     error = 0;
315 
316         ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
317 
318         /*
319          * If there are other waiters on the queue then give them a chance at
320          * logspace before us.  Wake up the first waiters, if we do not wake
321          * up all the waiters then go to sleep waiting for more free space,
322          * otherwise try to get some space for this transaction.
323          */
324         *need_bytes = xlog_ticket_reservation(log, head, tic);
325         free_bytes = xlog_space_left(log, &head->grant);
326         if (!list_empty_careful(&head->waiters)) {
327                 spin_lock(&head->lock);
328                 if (!xlog_grant_head_wake(log, head, &free_bytes) ||
329                     free_bytes < *need_bytes) {
330                         error = xlog_grant_head_wait(log, head, tic,
331                                                      *need_bytes);
332                 }
333                 spin_unlock(&head->lock);
334         } else if (free_bytes < *need_bytes) {
335                 spin_lock(&head->lock);
336                 error = xlog_grant_head_wait(log, head, tic, *need_bytes);
337                 spin_unlock(&head->lock);
338         }
339 
340         return error;
341 }
342 
343 static void
344 xlog_tic_reset_res(xlog_ticket_t *tic)
345 {
346         tic->t_res_num = 0;
347         tic->t_res_arr_sum = 0;
348         tic->t_res_num_ophdrs = 0;
349 }
350 
351 static void
352 xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
353 {
354         if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
355                 /* add to overflow and start again */
356                 tic->t_res_o_flow += tic->t_res_arr_sum;
357                 tic->t_res_num = 0;
358                 tic->t_res_arr_sum = 0;
359         }
360 
361         tic->t_res_arr[tic->t_res_num].r_len = len;
362         tic->t_res_arr[tic->t_res_num].r_type = type;
363         tic->t_res_arr_sum += len;
364         tic->t_res_num++;
365 }
366 
367 /*
368  * Replenish the byte reservation required by moving the grant write head.
369  */
370 int
371 xfs_log_regrant(
372         struct xfs_mount        *mp,
373         struct xlog_ticket      *tic)
374 {
375         struct xlog             *log = mp->m_log;
376         int                     need_bytes;
377         int                     error = 0;
378 
379         if (XLOG_FORCED_SHUTDOWN(log))
380                 return XFS_ERROR(EIO);
381 
382         XFS_STATS_INC(xs_try_logspace);
383 
384         /*
385          * This is a new transaction on the ticket, so we need to change the
386          * transaction ID so that the next transaction has a different TID in
387          * the log. Just add one to the existing tid so that we can see chains
388          * of rolling transactions in the log easily.
389          */
390         tic->t_tid++;
391 
392         xlog_grant_push_ail(log, tic->t_unit_res);
393 
394         tic->t_curr_res = tic->t_unit_res;
395         xlog_tic_reset_res(tic);
396 
397         if (tic->t_cnt > 0)
398                 return 0;
399 
400         trace_xfs_log_regrant(log, tic);
401 
402         error = xlog_grant_head_check(log, &log->l_write_head, tic,
403                                       &need_bytes);
404         if (error)
405                 goto out_error;
406 
407         xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
408         trace_xfs_log_regrant_exit(log, tic);
409         xlog_verify_grant_tail(log);
410         return 0;
411 
412 out_error:
413         /*
414          * If we are failing, make sure the ticket doesn't have any current
415          * reservations.  We don't want to add this back when the ticket/
416          * transaction gets cancelled.
417          */
418         tic->t_curr_res = 0;
419         tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
420         return error;
421 }
422 
423 /*
424  * Reserve log space and return a ticket corresponding the reservation.
425  *
426  * Each reservation is going to reserve extra space for a log record header.
427  * When writes happen to the on-disk log, we don't subtract the length of the
428  * log record header from any reservation.  By wasting space in each
429  * reservation, we prevent over allocation problems.
430  */
431 int
432 xfs_log_reserve(
433         struct xfs_mount        *mp,
434         int                     unit_bytes,
435         int                     cnt,
436         struct xlog_ticket      **ticp,
437         __uint8_t               client,
438         bool                    permanent,
439         uint                    t_type)
440 {
441         struct xlog             *log = mp->m_log;
442         struct xlog_ticket      *tic;
443         int                     need_bytes;
444         int                     error = 0;
445 
446         ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
447 
448         if (XLOG_FORCED_SHUTDOWN(log))
449                 return XFS_ERROR(EIO);
450 
451         XFS_STATS_INC(xs_try_logspace);
452 
453         ASSERT(*ticp == NULL);
454         tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent,
455                                 KM_SLEEP | KM_MAYFAIL);
456         if (!tic)
457                 return XFS_ERROR(ENOMEM);
458 
459         tic->t_trans_type = t_type;
460         *ticp = tic;
461 
462         xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt
463                                             : tic->t_unit_res);
464 
465         trace_xfs_log_reserve(log, tic);
466 
467         error = xlog_grant_head_check(log, &log->l_reserve_head, tic,
468                                       &need_bytes);
469         if (error)
470                 goto out_error;
471 
472         xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes);
473         xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
474         trace_xfs_log_reserve_exit(log, tic);
475         xlog_verify_grant_tail(log);
476         return 0;
477 
478 out_error:
479         /*
480          * If we are failing, make sure the ticket doesn't have any current
481          * reservations.  We don't want to add this back when the ticket/
482          * transaction gets cancelled.
483          */
484         tic->t_curr_res = 0;
485         tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
486         return error;
487 }
488 
489 
490 /*
491  * NOTES:
492  *
493  *      1. currblock field gets updated at startup and after in-core logs
494  *              marked as with WANT_SYNC.
495  */
496 
497 /*
498  * This routine is called when a user of a log manager ticket is done with
499  * the reservation.  If the ticket was ever used, then a commit record for
500  * the associated transaction is written out as a log operation header with
501  * no data.  The flag XLOG_TIC_INITED is set when the first write occurs with
502  * a given ticket.  If the ticket was one with a permanent reservation, then
503  * a few operations are done differently.  Permanent reservation tickets by
504  * default don't release the reservation.  They just commit the current
505  * transaction with the belief that the reservation is still needed.  A flag
506  * must be passed in before permanent reservations are actually released.
507  * When these type of tickets are not released, they need to be set into
508  * the inited state again.  By doing this, a start record will be written
509  * out when the next write occurs.
510  */
511 xfs_lsn_t
512 xfs_log_done(
513         struct xfs_mount        *mp,
514         struct xlog_ticket      *ticket,
515         struct xlog_in_core     **iclog,
516         uint                    flags)
517 {
518         struct xlog             *log = mp->m_log;
519         xfs_lsn_t               lsn = 0;
520 
521         if (XLOG_FORCED_SHUTDOWN(log) ||
522             /*
523              * If nothing was ever written, don't write out commit record.
524              * If we get an error, just continue and give back the log ticket.
525              */
526             (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
527              (xlog_commit_record(log, ticket, iclog, &lsn)))) {
528                 lsn = (xfs_lsn_t) -1;
529                 if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
530                         flags |= XFS_LOG_REL_PERM_RESERV;
531                 }
532         }
533 
534 
535         if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
536             (flags & XFS_LOG_REL_PERM_RESERV)) {
537                 trace_xfs_log_done_nonperm(log, ticket);
538 
539                 /*
540                  * Release ticket if not permanent reservation or a specific
541                  * request has been made to release a permanent reservation.
542                  */
543                 xlog_ungrant_log_space(log, ticket);
544                 xfs_log_ticket_put(ticket);
545         } else {
546                 trace_xfs_log_done_perm(log, ticket);
547 
548                 xlog_regrant_reserve_log_space(log, ticket);
549                 /* If this ticket was a permanent reservation and we aren't
550                  * trying to release it, reset the inited flags; so next time
551                  * we write, a start record will be written out.
552                  */
553                 ticket->t_flags |= XLOG_TIC_INITED;
554         }
555 
556         return lsn;
557 }
558 
559 /*
560  * Attaches a new iclog I/O completion callback routine during
561  * transaction commit.  If the log is in error state, a non-zero
562  * return code is handed back and the caller is responsible for
563  * executing the callback at an appropriate time.
564  */
565 int
566 xfs_log_notify(
567         struct xfs_mount        *mp,
568         struct xlog_in_core     *iclog,
569         xfs_log_callback_t      *cb)
570 {
571         int     abortflg;
572 
573         spin_lock(&iclog->ic_callback_lock);
574         abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
575         if (!abortflg) {
576                 ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
577                               (iclog->ic_state == XLOG_STATE_WANT_SYNC));
578                 cb->cb_next = NULL;
579                 *(iclog->ic_callback_tail) = cb;
580                 iclog->ic_callback_tail = &(cb->cb_next);
581         }
582         spin_unlock(&iclog->ic_callback_lock);
583         return abortflg;
584 }
585 
586 int
587 xfs_log_release_iclog(
588         struct xfs_mount        *mp,
589         struct xlog_in_core     *iclog)
590 {
591         if (xlog_state_release_iclog(mp->m_log, iclog)) {
592                 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
593                 return EIO;
594         }
595 
596         return 0;
597 }
598 
599 /*
600  * Mount a log filesystem
601  *
602  * mp           - ubiquitous xfs mount point structure
603  * log_target   - buftarg of on-disk log device
604  * blk_offset   - Start block # where block size is 512 bytes (BBSIZE)
605  * num_bblocks  - Number of BBSIZE blocks in on-disk log
606  *
607  * Return error or zero.
608  */
609 int
610 xfs_log_mount(
611         xfs_mount_t     *mp,
612         xfs_buftarg_t   *log_target,
613         xfs_daddr_t     blk_offset,
614         int             num_bblks)
615 {
616         int             error = 0;
617         int             min_logfsbs;
618 
619         if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
620                 xfs_notice(mp, "Mounting Filesystem");
621         else {
622                 xfs_notice(mp,
623 "Mounting filesystem in no-recovery mode.  Filesystem will be inconsistent.");
624                 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
625         }
626 
627         mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
628         if (IS_ERR(mp->m_log)) {
629                 error = -PTR_ERR(mp->m_log);
630                 goto out;
631         }
632 
633         /*
634          * Validate the given log space and drop a critical message via syslog
635          * if the log size is too small that would lead to some unexpected
636          * situations in transaction log space reservation stage.
637          *
638          * Note: we can't just reject the mount if the validation fails.  This
639          * would mean that people would have to downgrade their kernel just to
640          * remedy the situation as there is no way to grow the log (short of
641          * black magic surgery with xfs_db).
642          *
643          * We can, however, reject mounts for CRC format filesystems, as the
644          * mkfs binary being used to make the filesystem should never create a
645          * filesystem with a log that is too small.
646          */
647         min_logfsbs = xfs_log_calc_minimum_size(mp);
648 
649         if (mp->m_sb.sb_logblocks < min_logfsbs) {
650                 xfs_warn(mp,
651                 "Log size %d blocks too small, minimum size is %d blocks",
652                          mp->m_sb.sb_logblocks, min_logfsbs);
653                 error = EINVAL;
654         } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) {
655                 xfs_warn(mp,
656                 "Log size %d blocks too large, maximum size is %lld blocks",
657                          mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS);
658                 error = EINVAL;
659         } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) {
660                 xfs_warn(mp,
661                 "log size %lld bytes too large, maximum size is %lld bytes",
662                          XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks),
663                          XFS_MAX_LOG_BYTES);
664                 error = EINVAL;
665         }
666         if (error) {
667                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
668                         xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!");
669                         ASSERT(0);
670                         goto out_free_log;
671                 }
672                 xfs_crit(mp,
673 "Log size out of supported range. Continuing onwards, but if log hangs are\n"
674 "experienced then please report this message in the bug report.");
675         }
676 
677         /*
678          * Initialize the AIL now we have a log.
679          */
680         error = xfs_trans_ail_init(mp);
681         if (error) {
682                 xfs_warn(mp, "AIL initialisation failed: error %d", error);
683                 goto out_free_log;
684         }
685         mp->m_log->l_ailp = mp->m_ail;
686 
687         /*
688          * skip log recovery on a norecovery mount.  pretend it all
689          * just worked.
690          */
691         if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
692                 int     readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
693 
694                 if (readonly)
695                         mp->m_flags &= ~XFS_MOUNT_RDONLY;
696 
697                 error = xlog_recover(mp->m_log);
698 
699                 if (readonly)
700                         mp->m_flags |= XFS_MOUNT_RDONLY;
701                 if (error) {
702                         xfs_warn(mp, "log mount/recovery failed: error %d",
703                                 error);
704                         goto out_destroy_ail;
705                 }
706         }
707 
708         /* Normal transactions can now occur */
709         mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
710 
711         /*
712          * Now the log has been fully initialised and we know were our
713          * space grant counters are, we can initialise the permanent ticket
714          * needed for delayed logging to work.
715          */
716         xlog_cil_init_post_recovery(mp->m_log);
717 
718         return 0;
719 
720 out_destroy_ail:
721         xfs_trans_ail_destroy(mp);
722 out_free_log:
723         xlog_dealloc_log(mp->m_log);
724 out:
725         return error;
726 }
727 
728 /*
729  * Finish the recovery of the file system.  This is separate from the
730  * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read
731  * in the root and real-time bitmap inodes between calling xfs_log_mount() and
732  * here.
733  *
734  * If we finish recovery successfully, start the background log work. If we are
735  * not doing recovery, then we have a RO filesystem and we don't need to start
736  * it.
737  */
738 int
739 xfs_log_mount_finish(xfs_mount_t *mp)
740 {
741         int     error = 0;
742 
743         if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
744                 error = xlog_recover_finish(mp->m_log);
745                 if (!error)
746                         xfs_log_work_queue(mp);
747         } else {
748                 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
749         }
750 
751 
752         return error;
753 }
754 
755 /*
756  * Final log writes as part of unmount.
757  *
758  * Mark the filesystem clean as unmount happens.  Note that during relocation
759  * this routine needs to be executed as part of source-bag while the
760  * deallocation must not be done until source-end.
761  */
762 
763 /*
764  * Unmount record used to have a string "Unmount filesystem--" in the
765  * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
766  * We just write the magic number now since that particular field isn't
767  * currently architecture converted and "Unmount" is a bit foo.
768  * As far as I know, there weren't any dependencies on the old behaviour.
769  */
770 
771 int
772 xfs_log_unmount_write(xfs_mount_t *mp)
773 {
774         struct xlog      *log = mp->m_log;
775         xlog_in_core_t   *iclog;
776 #ifdef DEBUG
777         xlog_in_core_t   *first_iclog;
778 #endif
779         xlog_ticket_t   *tic = NULL;
780         xfs_lsn_t        lsn;
781         int              error;
782 
783         /*
784          * Don't write out unmount record on read-only mounts.
785          * Or, if we are doing a forced umount (typically because of IO errors).
786          */
787         if (mp->m_flags & XFS_MOUNT_RDONLY)
788                 return 0;
789 
790         error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
791         ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
792 
793 #ifdef DEBUG
794         first_iclog = iclog = log->l_iclog;
795         do {
796                 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
797                         ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
798                         ASSERT(iclog->ic_offset == 0);
799                 }
800                 iclog = iclog->ic_next;
801         } while (iclog != first_iclog);
802 #endif
803         if (! (XLOG_FORCED_SHUTDOWN(log))) {
804                 error = xfs_log_reserve(mp, 600, 1, &tic,
805                                         XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
806                 if (!error) {
807                         /* the data section must be 32 bit size aligned */
808                         struct {
809                             __uint16_t magic;
810                             __uint16_t pad1;
811                             __uint32_t pad2; /* may as well make it 64 bits */
812                         } magic = {
813                                 .magic = XLOG_UNMOUNT_TYPE,
814                         };
815                         struct xfs_log_iovec reg = {
816                                 .i_addr = &magic,
817                                 .i_len = sizeof(magic),
818                                 .i_type = XLOG_REG_TYPE_UNMOUNT,
819                         };
820                         struct xfs_log_vec vec = {
821                                 .lv_niovecs = 1,
822                                 .lv_iovecp = &reg,
823                         };
824 
825                         /* remove inited flag, and account for space used */
826                         tic->t_flags = 0;
827                         tic->t_curr_res -= sizeof(magic);
828                         error = xlog_write(log, &vec, tic, &lsn,
829                                            NULL, XLOG_UNMOUNT_TRANS);
830                         /*
831                          * At this point, we're umounting anyway,
832                          * so there's no point in transitioning log state
833                          * to IOERROR. Just continue...
834                          */
835                 }
836 
837                 if (error)
838                         xfs_alert(mp, "%s: unmount record failed", __func__);
839 
840 
841                 spin_lock(&log->l_icloglock);
842                 iclog = log->l_iclog;
843                 atomic_inc(&iclog->ic_refcnt);
844                 xlog_state_want_sync(log, iclog);
845                 spin_unlock(&log->l_icloglock);
846                 error = xlog_state_release_iclog(log, iclog);
847 
848                 spin_lock(&log->l_icloglock);
849                 if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
850                       iclog->ic_state == XLOG_STATE_DIRTY)) {
851                         if (!XLOG_FORCED_SHUTDOWN(log)) {
852                                 xlog_wait(&iclog->ic_force_wait,
853                                                         &log->l_icloglock);
854                         } else {
855                                 spin_unlock(&log->l_icloglock);
856                         }
857                 } else {
858                         spin_unlock(&log->l_icloglock);
859                 }
860                 if (tic) {
861                         trace_xfs_log_umount_write(log, tic);
862                         xlog_ungrant_log_space(log, tic);
863                         xfs_log_ticket_put(tic);
864                 }
865         } else {
866                 /*
867                  * We're already in forced_shutdown mode, couldn't
868                  * even attempt to write out the unmount transaction.
869                  *
870                  * Go through the motions of sync'ing and releasing
871                  * the iclog, even though no I/O will actually happen,
872                  * we need to wait for other log I/Os that may already
873                  * be in progress.  Do this as a separate section of
874                  * code so we'll know if we ever get stuck here that
875                  * we're in this odd situation of trying to unmount
876                  * a file system that went into forced_shutdown as
877                  * the result of an unmount..
878                  */
879                 spin_lock(&log->l_icloglock);
880                 iclog = log->l_iclog;
881                 atomic_inc(&iclog->ic_refcnt);
882 
883                 xlog_state_want_sync(log, iclog);
884                 spin_unlock(&log->l_icloglock);
885                 error =  xlog_state_release_iclog(log, iclog);
886 
887                 spin_lock(&log->l_icloglock);
888 
889                 if ( ! (   iclog->ic_state == XLOG_STATE_ACTIVE
890                         || iclog->ic_state == XLOG_STATE_DIRTY
891                         || iclog->ic_state == XLOG_STATE_IOERROR) ) {
892 
893                                 xlog_wait(&iclog->ic_force_wait,
894                                                         &log->l_icloglock);
895                 } else {
896                         spin_unlock(&log->l_icloglock);
897                 }
898         }
899 
900         return error;
901 }       /* xfs_log_unmount_write */
902 
903 /*
904  * Empty the log for unmount/freeze.
905  *
906  * To do this, we first need to shut down the background log work so it is not
907  * trying to cover the log as we clean up. We then need to unpin all objects in
908  * the log so we can then flush them out. Once they have completed their IO and
909  * run the callbacks removing themselves from the AIL, we can write the unmount
910  * record.
911  */
912 void
913 xfs_log_quiesce(
914         struct xfs_mount        *mp)
915 {
916         cancel_delayed_work_sync(&mp->m_log->l_work);
917         xfs_log_force(mp, XFS_LOG_SYNC);
918 
919         /*
920          * The superblock buffer is uncached and while xfs_ail_push_all_sync()
921          * will push it, xfs_wait_buftarg() will not wait for it. Further,
922          * xfs_buf_iowait() cannot be used because it was pushed with the
923          * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for
924          * the IO to complete.
925          */
926         xfs_ail_push_all_sync(mp->m_ail);
927         xfs_wait_buftarg(mp->m_ddev_targp);
928         xfs_buf_lock(mp->m_sb_bp);
929         xfs_buf_unlock(mp->m_sb_bp);
930 
931         xfs_log_unmount_write(mp);
932 }
933 
934 /*
935  * Shut down and release the AIL and Log.
936  *
937  * During unmount, we need to ensure we flush all the dirty metadata objects
938  * from the AIL so that the log is empty before we write the unmount record to
939  * the log. Once this is done, we can tear down the AIL and the log.
940  */
941 void
942 xfs_log_unmount(
943         struct xfs_mount        *mp)
944 {
945         xfs_log_quiesce(mp);
946 
947         xfs_trans_ail_destroy(mp);
948         xlog_dealloc_log(mp->m_log);
949 }
950 
951 void
952 xfs_log_item_init(
953         struct xfs_mount        *mp,
954         struct xfs_log_item     *item,
955         int                     type,
956         const struct xfs_item_ops *ops)
957 {
958         item->li_mountp = mp;
959         item->li_ailp = mp->m_ail;
960         item->li_type = type;
961         item->li_ops = ops;
962         item->li_lv = NULL;
963 
964         INIT_LIST_HEAD(&item->li_ail);
965         INIT_LIST_HEAD(&item->li_cil);
966 }
967 
968 /*
969  * Wake up processes waiting for log space after we have moved the log tail.
970  */
971 void
972 xfs_log_space_wake(
973         struct xfs_mount        *mp)
974 {
975         struct xlog             *log = mp->m_log;
976         int                     free_bytes;
977 
978         if (XLOG_FORCED_SHUTDOWN(log))
979                 return;
980 
981         if (!list_empty_careful(&log->l_write_head.waiters)) {
982                 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
983 
984                 spin_lock(&log->l_write_head.lock);
985                 free_bytes = xlog_space_left(log, &log->l_write_head.grant);
986                 xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);
987                 spin_unlock(&log->l_write_head.lock);
988         }
989 
990         if (!list_empty_careful(&log->l_reserve_head.waiters)) {
991                 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
992 
993                 spin_lock(&log->l_reserve_head.lock);
994                 free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
995                 xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);
996                 spin_unlock(&log->l_reserve_head.lock);
997         }
998 }
999 
1000 /*
1001  * Determine if we have a transaction that has gone to disk that needs to be
1002  * covered. To begin the transition to the idle state firstly the log needs to
1003  * be idle. That means the CIL, the AIL and the iclogs needs to be empty before
1004  * we start attempting to cover the log.
1005  *
1006  * Only if we are then in a state where covering is needed, the caller is
1007  * informed that dummy transactions are required to move the log into the idle
1008  * state.
1009  *
1010  * If there are any items in the AIl or CIL, then we do not want to attempt to
1011  * cover the log as we may be in a situation where there isn't log space
1012  * available to run a dummy transaction and this can lead to deadlocks when the
1013  * tail of the log is pinned by an item that is modified in the CIL.  Hence
1014  * there's no point in running a dummy transaction at this point because we
1015  * can't start trying to idle the log until both the CIL and AIL are empty.
1016  */
1017 int
1018 xfs_log_need_covered(xfs_mount_t *mp)
1019 {
1020         struct xlog     *log = mp->m_log;
1021         int             needed = 0;
1022 
1023         if (!xfs_fs_writable(mp))
1024                 return 0;
1025 
1026         if (!xlog_cil_empty(log))
1027                 return 0;
1028 
1029         spin_lock(&log->l_icloglock);
1030         switch (log->l_covered_state) {
1031         case XLOG_STATE_COVER_DONE:
1032         case XLOG_STATE_COVER_DONE2:
1033         case XLOG_STATE_COVER_IDLE:
1034                 break;
1035         case XLOG_STATE_COVER_NEED:
1036         case XLOG_STATE_COVER_NEED2:
1037                 if (xfs_ail_min_lsn(log->l_ailp))
1038                         break;
1039                 if (!xlog_iclogs_empty(log))
1040                         break;
1041 
1042                 needed = 1;
1043                 if (log->l_covered_state == XLOG_STATE_COVER_NEED)
1044                         log->l_covered_state = XLOG_STATE_COVER_DONE;
1045                 else
1046                         log->l_covered_state = XLOG_STATE_COVER_DONE2;
1047                 break;
1048         default:
1049                 needed = 1;
1050                 break;
1051         }
1052         spin_unlock(&log->l_icloglock);
1053         return needed;
1054 }
1055 
1056 /*
1057  * We may be holding the log iclog lock upon entering this routine.
1058  */
1059 xfs_lsn_t
1060 xlog_assign_tail_lsn_locked(
1061         struct xfs_mount        *mp)
1062 {
1063         struct xlog             *log = mp->m_log;
1064         struct xfs_log_item     *lip;
1065         xfs_lsn_t               tail_lsn;
1066 
1067         assert_spin_locked(&mp->m_ail->xa_lock);
1068 
1069         /*
1070          * To make sure we always have a valid LSN for the log tail we keep
1071          * track of the last LSN which was committed in log->l_last_sync_lsn,
1072          * and use that when the AIL was empty.
1073          */
1074         lip = xfs_ail_min(mp->m_ail);
1075         if (lip)
1076                 tail_lsn = lip->li_lsn;
1077         else
1078                 tail_lsn = atomic64_read(&log->l_last_sync_lsn);
1079         trace_xfs_log_assign_tail_lsn(log, tail_lsn);
1080         atomic64_set(&log->l_tail_lsn, tail_lsn);
1081         return tail_lsn;
1082 }
1083 
1084 xfs_lsn_t
1085 xlog_assign_tail_lsn(
1086         struct xfs_mount        *mp)
1087 {
1088         xfs_lsn_t               tail_lsn;
1089 
1090         spin_lock(&mp->m_ail->xa_lock);
1091         tail_lsn = xlog_assign_tail_lsn_locked(mp);
1092         spin_unlock(&mp->m_ail->xa_lock);
1093 
1094         return tail_lsn;
1095 }
1096 
1097 /*
1098  * Return the space in the log between the tail and the head.  The head
1099  * is passed in the cycle/bytes formal parms.  In the special case where
1100  * the reserve head has wrapped passed the tail, this calculation is no
1101  * longer valid.  In this case, just return 0 which means there is no space
1102  * in the log.  This works for all places where this function is called
1103  * with the reserve head.  Of course, if the write head were to ever
1104  * wrap the tail, we should blow up.  Rather than catch this case here,
1105  * we depend on other ASSERTions in other parts of the code.   XXXmiken
1106  *
1107  * This code also handles the case where the reservation head is behind
1108  * the tail.  The details of this case are described below, but the end
1109  * result is that we return the size of the log as the amount of space left.
1110  */
1111 STATIC int
1112 xlog_space_left(
1113         struct xlog     *log,
1114         atomic64_t      *head)
1115 {
1116         int             free_bytes;
1117         int             tail_bytes;
1118         int             tail_cycle;
1119         int             head_cycle;
1120         int             head_bytes;
1121 
1122         xlog_crack_grant_head(head, &head_cycle, &head_bytes);
1123         xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
1124         tail_bytes = BBTOB(tail_bytes);
1125         if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
1126                 free_bytes = log->l_logsize - (head_bytes - tail_bytes);
1127         else if (tail_cycle + 1 < head_cycle)
1128                 return 0;
1129         else if (tail_cycle < head_cycle) {
1130                 ASSERT(tail_cycle == (head_cycle - 1));
1131                 free_bytes = tail_bytes - head_bytes;
1132         } else {
1133                 /*
1134                  * The reservation head is behind the tail.
1135                  * In this case we just want to return the size of the
1136                  * log as the amount of space left.
1137                  */
1138                 xfs_alert(log->l_mp,
1139                         "xlog_space_left: head behind tail\n"
1140                         "  tail_cycle = %d, tail_bytes = %d\n"
1141                         "  GH   cycle = %d, GH   bytes = %d",
1142                         tail_cycle, tail_bytes, head_cycle, head_bytes);
1143                 ASSERT(0);
1144                 free_bytes = log->l_logsize;
1145         }
1146         return free_bytes;
1147 }
1148 
1149 
1150 /*
1151  * Log function which is called when an io completes.
1152  *
1153  * The log manager needs its own routine, in order to control what
1154  * happens with the buffer after the write completes.
1155  */
1156 void
1157 xlog_iodone(xfs_buf_t *bp)
1158 {
1159         struct xlog_in_core     *iclog = bp->b_fspriv;
1160         struct xlog             *l = iclog->ic_log;
1161         int                     aborted = 0;
1162 
1163         /*
1164          * Race to shutdown the filesystem if we see an error.
1165          */
1166         if (XFS_TEST_ERROR((xfs_buf_geterror(bp)), l->l_mp,
1167                         XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
1168                 xfs_buf_ioerror_alert(bp, __func__);
1169                 xfs_buf_stale(bp);
1170                 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
1171                 /*
1172                  * This flag will be propagated to the trans-committed
1173                  * callback routines to let them know that the log-commit
1174                  * didn't succeed.
1175                  */
1176                 aborted = XFS_LI_ABORTED;
1177         } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
1178                 aborted = XFS_LI_ABORTED;
1179         }
1180 
1181         /* log I/O is always issued ASYNC */
1182         ASSERT(XFS_BUF_ISASYNC(bp));
1183         xlog_state_done_syncing(iclog, aborted);
1184         /*
1185          * do not reference the buffer (bp) here as we could race
1186          * with it being freed after writing the unmount record to the
1187          * log.
1188          */
1189 }
1190 
1191 /*
1192  * Return size of each in-core log record buffer.
1193  *
1194  * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
1195  *
1196  * If the filesystem blocksize is too large, we may need to choose a
1197  * larger size since the directory code currently logs entire blocks.
1198  */
1199 
1200 STATIC void
1201 xlog_get_iclog_buffer_size(
1202         struct xfs_mount        *mp,
1203         struct xlog             *log)
1204 {
1205         int size;
1206         int xhdrs;
1207 
1208         if (mp->m_logbufs <= 0)
1209                 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
1210         else
1211                 log->l_iclog_bufs = mp->m_logbufs;
1212 
1213         /*
1214          * Buffer size passed in from mount system call.
1215          */
1216         if (mp->m_logbsize > 0) {
1217                 size = log->l_iclog_size = mp->m_logbsize;
1218                 log->l_iclog_size_log = 0;
1219                 while (size != 1) {
1220                         log->l_iclog_size_log++;
1221                         size >>= 1;
1222                 }
1223 
1224                 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1225                         /* # headers = size / 32k
1226                          * one header holds cycles from 32k of data
1227                          */
1228 
1229                         xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
1230                         if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
1231                                 xhdrs++;
1232                         log->l_iclog_hsize = xhdrs << BBSHIFT;
1233                         log->l_iclog_heads = xhdrs;
1234                 } else {
1235                         ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
1236                         log->l_iclog_hsize = BBSIZE;
1237                         log->l_iclog_heads = 1;
1238                 }
1239                 goto done;
1240         }
1241 
1242         /* All machines use 32kB buffers by default. */
1243         log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
1244         log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
1245 
1246         /* the default log size is 16k or 32k which is one header sector */
1247         log->l_iclog_hsize = BBSIZE;
1248         log->l_iclog_heads = 1;
1249 
1250 done:
1251         /* are we being asked to make the sizes selected above visible? */
1252         if (mp->m_logbufs == 0)
1253                 mp->m_logbufs = log->l_iclog_bufs;
1254         if (mp->m_logbsize == 0)
1255                 mp->m_logbsize = log->l_iclog_size;
1256 }       /* xlog_get_iclog_buffer_size */
1257 
1258 
1259 void
1260 xfs_log_work_queue(
1261         struct xfs_mount        *mp)
1262 {
1263         queue_delayed_work(mp->m_log_workqueue, &mp->m_log->l_work,
1264                                 msecs_to_jiffies(xfs_syncd_centisecs * 10));
1265 }
1266 
1267 /*
1268  * Every sync period we need to unpin all items in the AIL and push them to
1269  * disk. If there is nothing dirty, then we might need to cover the log to
1270  * indicate that the filesystem is idle.
1271  */
1272 void
1273 xfs_log_worker(
1274         struct work_struct      *work)
1275 {
1276         struct xlog             *log = container_of(to_delayed_work(work),
1277                                                 struct xlog, l_work);
1278         struct xfs_mount        *mp = log->l_mp;
1279 
1280         /* dgc: errors ignored - not fatal and nowhere to report them */
1281         if (xfs_log_need_covered(mp))
1282                 xfs_fs_log_dummy(mp);
1283         else
1284                 xfs_log_force(mp, 0);
1285 
1286         /* start pushing all the metadata that is currently dirty */
1287         xfs_ail_push_all(mp->m_ail);
1288 
1289         /* queue us up again */
1290         xfs_log_work_queue(mp);
1291 }
1292 
1293 /*
1294  * This routine initializes some of the log structure for a given mount point.
1295  * Its primary purpose is to fill in enough, so recovery can occur.  However,
1296  * some other stuff may be filled in too.
1297  */
1298 STATIC struct xlog *
1299 xlog_alloc_log(
1300         struct xfs_mount        *mp,
1301         struct xfs_buftarg      *log_target,
1302         xfs_daddr_t             blk_offset,
1303         int                     num_bblks)
1304 {
1305         struct xlog             *log;
1306         xlog_rec_header_t       *head;
1307         xlog_in_core_t          **iclogp;
1308         xlog_in_core_t          *iclog, *prev_iclog=NULL;
1309         xfs_buf_t               *bp;
1310         int                     i;
1311         int                     error = ENOMEM;
1312         uint                    log2_size = 0;
1313 
1314         log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL);
1315         if (!log) {
1316                 xfs_warn(mp, "Log allocation failed: No memory!");
1317                 goto out;
1318         }
1319 
1320         log->l_mp          = mp;
1321         log->l_targ        = log_target;
1322         log->l_logsize     = BBTOB(num_bblks);
1323         log->l_logBBstart  = blk_offset;
1324         log->l_logBBsize   = num_bblks;
1325         log->l_covered_state = XLOG_STATE_COVER_IDLE;
1326         log->l_flags       |= XLOG_ACTIVE_RECOVERY;
1327         INIT_DELAYED_WORK(&log->l_work, xfs_log_worker);
1328 
1329         log->l_prev_block  = -1;
1330         /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1331         xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
1332         xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
1333         log->l_curr_cycle  = 1;     /* 0 is bad since this is initial value */
1334 
1335         xlog_grant_head_init(&log->l_reserve_head);
1336         xlog_grant_head_init(&log->l_write_head);
1337 
1338         error = EFSCORRUPTED;
1339         if (xfs_sb_version_hassector(&mp->m_sb)) {
1340                 log2_size = mp->m_sb.sb_logsectlog;
1341                 if (log2_size < BBSHIFT) {
1342                         xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
1343                                 log2_size, BBSHIFT);
1344                         goto out_free_log;
1345                 }
1346 
1347                 log2_size -= BBSHIFT;
1348                 if (log2_size > mp->m_sectbb_log) {
1349                         xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
1350                                 log2_size, mp->m_sectbb_log);
1351                         goto out_free_log;
1352                 }
1353 
1354                 /* for larger sector sizes, must have v2 or external log */
1355                 if (log2_size && log->l_logBBstart > 0 &&
1356                             !xfs_sb_version_haslogv2(&mp->m_sb)) {
1357                         xfs_warn(mp,
1358                 "log sector size (0x%x) invalid for configuration.",
1359                                 log2_size);
1360                         goto out_free_log;
1361                 }
1362         }
1363         log->l_sectBBsize = 1 << log2_size;
1364 
1365         xlog_get_iclog_buffer_size(mp, log);
1366 
1367         error = ENOMEM;
1368         bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0);
1369         if (!bp)
1370                 goto out_free_log;
1371         bp->b_iodone = xlog_iodone;
1372         ASSERT(xfs_buf_islocked(bp));
1373         log->l_xbuf = bp;
1374 
1375         spin_lock_init(&log->l_icloglock);
1376         init_waitqueue_head(&log->l_flush_wait);
1377 
1378         iclogp = &log->l_iclog;
1379         /*
1380          * The amount of memory to allocate for the iclog structure is
1381          * rather funky due to the way the structure is defined.  It is
1382          * done this way so that we can use different sizes for machines
1383          * with different amounts of memory.  See the definition of
1384          * xlog_in_core_t in xfs_log_priv.h for details.
1385          */
1386         ASSERT(log->l_iclog_size >= 4096);
1387         for (i=0; i < log->l_iclog_bufs; i++) {
1388                 *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
1389                 if (!*iclogp)
1390                         goto out_free_iclog;
1391 
1392                 iclog = *iclogp;
1393                 iclog->ic_prev = prev_iclog;
1394                 prev_iclog = iclog;
1395 
1396                 bp = xfs_buf_get_uncached(mp->m_logdev_targp,
1397                                                 BTOBB(log->l_iclog_size), 0);
1398                 if (!bp)
1399                         goto out_free_iclog;
1400 
1401                 bp->b_iodone = xlog_iodone;
1402                 iclog->ic_bp = bp;
1403                 iclog->ic_data = bp->b_addr;
1404 #ifdef DEBUG
1405                 log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
1406 #endif
1407                 head = &iclog->ic_header;
1408                 memset(head, 0, sizeof(xlog_rec_header_t));
1409                 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1410                 head->h_version = cpu_to_be32(
1411                         xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1412                 head->h_size = cpu_to_be32(log->l_iclog_size);
1413                 /* new fields */
1414                 head->h_fmt = cpu_to_be32(XLOG_FMT);
1415                 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1416 
1417                 iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize;
1418                 iclog->ic_state = XLOG_STATE_ACTIVE;
1419                 iclog->ic_log = log;
1420                 atomic_set(&iclog->ic_refcnt, 0);
1421                 spin_lock_init(&iclog->ic_callback_lock);
1422                 iclog->ic_callback_tail = &(iclog->ic_callback);
1423                 iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
1424 
1425                 ASSERT(xfs_buf_islocked(iclog->ic_bp));
1426                 init_waitqueue_head(&iclog->ic_force_wait);
1427                 init_waitqueue_head(&iclog->ic_write_wait);
1428 
1429                 iclogp = &iclog->ic_next;
1430         }
1431         *iclogp = log->l_iclog;                 /* complete ring */
1432         log->l_iclog->ic_prev = prev_iclog;     /* re-write 1st prev ptr */
1433 
1434         error = xlog_cil_init(log);
1435         if (error)
1436                 goto out_free_iclog;
1437         return log;
1438 
1439 out_free_iclog:
1440         for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1441                 prev_iclog = iclog->ic_next;
1442                 if (iclog->ic_bp)
1443                         xfs_buf_free(iclog->ic_bp);
1444                 kmem_free(iclog);
1445         }
1446         spinlock_destroy(&log->l_icloglock);
1447         xfs_buf_free(log->l_xbuf);
1448 out_free_log:
1449         kmem_free(log);
1450 out:
1451         return ERR_PTR(-error);
1452 }       /* xlog_alloc_log */
1453 
1454 
1455 /*
1456  * Write out the commit record of a transaction associated with the given
1457  * ticket.  Return the lsn of the commit record.
1458  */
1459 STATIC int
1460 xlog_commit_record(
1461         struct xlog             *log,
1462         struct xlog_ticket      *ticket,
1463         struct xlog_in_core     **iclog,
1464         xfs_lsn_t               *commitlsnp)
1465 {
1466         struct xfs_mount *mp = log->l_mp;
1467         int     error;
1468         struct xfs_log_iovec reg = {
1469                 .i_addr = NULL,
1470                 .i_len = 0,
1471                 .i_type = XLOG_REG_TYPE_COMMIT,
1472         };
1473         struct xfs_log_vec vec = {
1474                 .lv_niovecs = 1,
1475                 .lv_iovecp = &reg,
1476         };
1477 
1478         ASSERT_ALWAYS(iclog);
1479         error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
1480                                         XLOG_COMMIT_TRANS);
1481         if (error)
1482                 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1483         return error;
1484 }
1485 
1486 /*
1487  * Push on the buffer cache code if we ever use more than 75% of the on-disk
1488  * log space.  This code pushes on the lsn which would supposedly free up
1489  * the 25% which we want to leave free.  We may need to adopt a policy which
1490  * pushes on an lsn which is further along in the log once we reach the high
1491  * water mark.  In this manner, we would be creating a low water mark.
1492  */
1493 STATIC void
1494 xlog_grant_push_ail(
1495         struct xlog     *log,
1496         int             need_bytes)
1497 {
1498         xfs_lsn_t       threshold_lsn = 0;
1499         xfs_lsn_t       last_sync_lsn;
1500         int             free_blocks;
1501         int             free_bytes;
1502         int             threshold_block;
1503         int             threshold_cycle;
1504         int             free_threshold;
1505 
1506         ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
1507 
1508         free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
1509         free_blocks = BTOBBT(free_bytes);
1510 
1511         /*
1512          * Set the threshold for the minimum number of free blocks in the
1513          * log to the maximum of what the caller needs, one quarter of the
1514          * log, and 256 blocks.
1515          */
1516         free_threshold = BTOBB(need_bytes);
1517         free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1518         free_threshold = MAX(free_threshold, 256);
1519         if (free_blocks >= free_threshold)
1520                 return;
1521 
1522         xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
1523                                                 &threshold_block);
1524         threshold_block += free_threshold;
1525         if (threshold_block >= log->l_logBBsize) {
1526                 threshold_block -= log->l_logBBsize;
1527                 threshold_cycle += 1;
1528         }
1529         threshold_lsn = xlog_assign_lsn(threshold_cycle,
1530                                         threshold_block);
1531         /*
1532          * Don't pass in an lsn greater than the lsn of the last
1533          * log record known to be on disk. Use a snapshot of the last sync lsn
1534          * so that it doesn't change between the compare and the set.
1535          */
1536         last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
1537         if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
1538                 threshold_lsn = last_sync_lsn;
1539 
1540         /*
1541          * Get the transaction layer to kick the dirty buffers out to
1542          * disk asynchronously. No point in trying to do this if
1543          * the filesystem is shutting down.
1544          */
1545         if (!XLOG_FORCED_SHUTDOWN(log))
1546                 xfs_ail_push(log->l_ailp, threshold_lsn);
1547 }
1548 
1549 /*
1550  * Stamp cycle number in every block
1551  */
1552 STATIC void
1553 xlog_pack_data(
1554         struct xlog             *log,
1555         struct xlog_in_core     *iclog,
1556         int                     roundoff)
1557 {
1558         int                     i, j, k;
1559         int                     size = iclog->ic_offset + roundoff;
1560         __be32                  cycle_lsn;
1561         xfs_caddr_t             dp;
1562 
1563         cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
1564 
1565         dp = iclog->ic_datap;
1566         for (i = 0; i < BTOBB(size); i++) {
1567                 if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE))
1568                         break;
1569                 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
1570                 *(__be32 *)dp = cycle_lsn;
1571                 dp += BBSIZE;
1572         }
1573 
1574         if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1575                 xlog_in_core_2_t *xhdr = iclog->ic_data;
1576 
1577                 for ( ; i < BTOBB(size); i++) {
1578                         j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1579                         k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1580                         xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
1581                         *(__be32 *)dp = cycle_lsn;
1582                         dp += BBSIZE;
1583                 }
1584 
1585                 for (i = 1; i < log->l_iclog_heads; i++)
1586                         xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
1587         }
1588 }
1589 
1590 /*
1591  * Calculate the checksum for a log buffer.
1592  *
1593  * This is a little more complicated than it should be because the various
1594  * headers and the actual data are non-contiguous.
1595  */
1596 __le32
1597 xlog_cksum(
1598         struct xlog             *log,
1599         struct xlog_rec_header  *rhead,
1600         char                    *dp,
1601         int                     size)
1602 {
1603         __uint32_t              crc;
1604 
1605         /* first generate the crc for the record header ... */
1606         crc = xfs_start_cksum((char *)rhead,
1607                               sizeof(struct xlog_rec_header),
1608                               offsetof(struct xlog_rec_header, h_crc));
1609 
1610         /* ... then for additional cycle data for v2 logs ... */
1611         if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1612                 union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead;
1613                 int             i;
1614 
1615                 for (i = 1; i < log->l_iclog_heads; i++) {
1616                         crc = crc32c(crc, &xhdr[i].hic_xheader,
1617                                      sizeof(struct xlog_rec_ext_header));
1618                 }
1619         }
1620 
1621         /* ... and finally for the payload */
1622         crc = crc32c(crc, dp, size);
1623 
1624         return xfs_end_cksum(crc);
1625 }
1626 
1627 /*
1628  * The bdstrat callback function for log bufs. This gives us a central
1629  * place to trap bufs in case we get hit by a log I/O error and need to
1630  * shutdown. Actually, in practice, even when we didn't get a log error,
1631  * we transition the iclogs to IOERROR state *after* flushing all existing
1632  * iclogs to disk. This is because we don't want anymore new transactions to be
1633  * started or completed afterwards.
1634  */
1635 STATIC int
1636 xlog_bdstrat(
1637         struct xfs_buf          *bp)
1638 {
1639         struct xlog_in_core     *iclog = bp->b_fspriv;
1640 
1641         if (iclog->ic_state & XLOG_STATE_IOERROR) {
1642                 xfs_buf_ioerror(bp, EIO);
1643                 xfs_buf_stale(bp);
1644                 xfs_buf_ioend(bp, 0);
1645                 /*
1646                  * It would seem logical to return EIO here, but we rely on
1647                  * the log state machine to propagate I/O errors instead of
1648                  * doing it here.
1649                  */
1650                 return 0;
1651         }
1652 
1653         xfs_buf_iorequest(bp);
1654         return 0;
1655 }
1656 
1657 /*
1658  * Flush out the in-core log (iclog) to the on-disk log in an asynchronous 
1659  * fashion.  Previously, we should have moved the current iclog
1660  * ptr in the log to point to the next available iclog.  This allows further
1661  * write to continue while this code syncs out an iclog ready to go.
1662  * Before an in-core log can be written out, the data section must be scanned
1663  * to save away the 1st word of each BBSIZE block into the header.  We replace
1664  * it with the current cycle count.  Each BBSIZE block is tagged with the
1665  * cycle count because there in an implicit assumption that drives will
1666  * guarantee that entire 512 byte blocks get written at once.  In other words,
1667  * we can't have part of a 512 byte block written and part not written.  By
1668  * tagging each block, we will know which blocks are valid when recovering
1669  * after an unclean shutdown.
1670  *
1671  * This routine is single threaded on the iclog.  No other thread can be in
1672  * this routine with the same iclog.  Changing contents of iclog can there-
1673  * fore be done without grabbing the state machine lock.  Updating the global
1674  * log will require grabbing the lock though.
1675  *
1676  * The entire log manager uses a logical block numbering scheme.  Only
1677  * log_sync (and then only bwrite()) know about the fact that the log may
1678  * not start with block zero on a given device.  The log block start offset
1679  * is added immediately before calling bwrite().
1680  */
1681 
1682 STATIC int
1683 xlog_sync(
1684         struct xlog             *log,
1685         struct xlog_in_core     *iclog)
1686 {
1687         xfs_buf_t       *bp;
1688         int             i;
1689         uint            count;          /* byte count of bwrite */
1690         uint            count_init;     /* initial count before roundup */
1691         int             roundoff;       /* roundoff to BB or stripe */
1692         int             split = 0;      /* split write into two regions */
1693         int             error;
1694         int             v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
1695         int             size;
1696 
1697         XFS_STATS_INC(xs_log_writes);
1698         ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
1699 
1700         /* Add for LR header */
1701         count_init = log->l_iclog_hsize + iclog->ic_offset;
1702 
1703         /* Round out the log write size */
1704         if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1705                 /* we have a v2 stripe unit to use */
1706                 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1707         } else {
1708                 count = BBTOB(BTOBB(count_init));
1709         }
1710         roundoff = count - count_init;
1711         ASSERT(roundoff >= 0);
1712         ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && 
1713                 roundoff < log->l_mp->m_sb.sb_logsunit)
1714                 || 
1715                 (log->l_mp->m_sb.sb_logsunit <= 1 && 
1716                  roundoff < BBTOB(1)));
1717 
1718         /* move grant heads by roundoff in sync */
1719         xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
1720         xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
1721 
1722         /* put cycle number in every block */
1723         xlog_pack_data(log, iclog, roundoff); 
1724 
1725         /* real byte length */
1726         size = iclog->ic_offset;
1727         if (v2)
1728                 size += roundoff;
1729         iclog->ic_header.h_len = cpu_to_be32(size);
1730 
1731         bp = iclog->ic_bp;
1732         XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
1733 
1734         XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1735 
1736         /* Do we need to split this write into 2 parts? */
1737         if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
1738                 char            *dptr;
1739 
1740                 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1741                 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
1742                 iclog->ic_bwritecnt = 2;
1743 
1744                 /*
1745                  * Bump the cycle numbers at the start of each block in the
1746                  * part of the iclog that ends up in the buffer that gets
1747                  * written to the start of the log.
1748                  *
1749                  * Watch out for the header magic number case, though.
1750                  */
1751                 dptr = (char *)&iclog->ic_header + count;
1752                 for (i = 0; i < split; i += BBSIZE) {
1753                         __uint32_t cycle = be32_to_cpu(*(__be32 *)dptr);
1754                         if (++cycle == XLOG_HEADER_MAGIC_NUM)
1755                                 cycle++;
1756                         *(__be32 *)dptr = cpu_to_be32(cycle);
1757 
1758                         dptr += BBSIZE;
1759                 }
1760         } else {
1761                 iclog->ic_bwritecnt = 1;
1762         }
1763 
1764         /* calculcate the checksum */
1765         iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
1766                                             iclog->ic_datap, size);
1767 
1768         bp->b_io_length = BTOBB(count);
1769         bp->b_fspriv = iclog;
1770         XFS_BUF_ZEROFLAGS(bp);
1771         XFS_BUF_ASYNC(bp);
1772         bp->b_flags |= XBF_SYNCIO;
1773 
1774         if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) {
1775                 bp->b_flags |= XBF_FUA;
1776 
1777                 /*
1778                  * Flush the data device before flushing the log to make
1779                  * sure all meta data written back from the AIL actually made
1780                  * it to disk before stamping the new log tail LSN into the
1781                  * log buffer.  For an external log we need to issue the
1782                  * flush explicitly, and unfortunately synchronously here;
1783                  * for an internal log we can simply use the block layer
1784                  * state machine for preflushes.
1785                  */
1786                 if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp)
1787                         xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
1788                 else
1789                         bp->b_flags |= XBF_FLUSH;
1790         }
1791 
1792         ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1793         ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1794 
1795         xlog_verify_iclog(log, iclog, count, true);
1796 
1797         /* account for log which doesn't start at block #0 */
1798         XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1799         /*
1800          * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1801          * is shutting down.
1802          */
1803         XFS_BUF_WRITE(bp);
1804 
1805         error = xlog_bdstrat(bp);
1806         if (error) {
1807                 xfs_buf_ioerror_alert(bp, "xlog_sync");
1808                 return error;
1809         }
1810         if (split) {
1811                 bp = iclog->ic_log->l_xbuf;
1812                 XFS_BUF_SET_ADDR(bp, 0);             /* logical 0 */
1813                 xfs_buf_associate_memory(bp,
1814                                 (char *)&iclog->ic_header + count, split);
1815                 bp->b_fspriv = iclog;
1816                 XFS_BUF_ZEROFLAGS(bp);
1817                 XFS_BUF_ASYNC(bp);
1818                 bp->b_flags |= XBF_SYNCIO;
1819                 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
1820                         bp->b_flags |= XBF_FUA;
1821 
1822                 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1823                 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1824 
1825                 /* account for internal log which doesn't start at block #0 */
1826                 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1827                 XFS_BUF_WRITE(bp);
1828                 error = xlog_bdstrat(bp);
1829                 if (error) {
1830                         xfs_buf_ioerror_alert(bp, "xlog_sync (split)");
1831                         return error;
1832                 }
1833         }
1834         return 0;
1835 }       /* xlog_sync */
1836 
1837 /*
1838  * Deallocate a log structure
1839  */
1840 STATIC void
1841 xlog_dealloc_log(
1842         struct xlog     *log)
1843 {
1844         xlog_in_core_t  *iclog, *next_iclog;
1845         int             i;
1846 
1847         xlog_cil_destroy(log);
1848 
1849         /*
1850          * always need to ensure that the extra buffer does not point to memory
1851          * owned by another log buffer before we free it.
1852          */
1853         xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size));
1854         xfs_buf_free(log->l_xbuf);
1855 
1856         iclog = log->l_iclog;
1857         for (i=0; i<log->l_iclog_bufs; i++) {
1858                 xfs_buf_free(iclog->ic_bp);
1859                 next_iclog = iclog->ic_next;
1860                 kmem_free(iclog);
1861                 iclog = next_iclog;
1862         }
1863         spinlock_destroy(&log->l_icloglock);
1864 
1865         log->l_mp->m_log = NULL;
1866         kmem_free(log);
1867 }       /* xlog_dealloc_log */
1868 
1869 /*
1870  * Update counters atomically now that memcpy is done.
1871  */
1872 /* ARGSUSED */
1873 static inline void
1874 xlog_state_finish_copy(
1875         struct xlog             *log,
1876         struct xlog_in_core     *iclog,
1877         int                     record_cnt,
1878         int                     copy_bytes)
1879 {
1880         spin_lock(&log->l_icloglock);
1881 
1882         be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
1883         iclog->ic_offset += copy_bytes;
1884 
1885         spin_unlock(&log->l_icloglock);
1886 }       /* xlog_state_finish_copy */
1887 
1888 
1889 
1890 
1891 /*
1892  * print out info relating to regions written which consume
1893  * the reservation
1894  */
1895 void
1896 xlog_print_tic_res(
1897         struct xfs_mount        *mp,
1898         struct xlog_ticket      *ticket)
1899 {
1900         uint i;
1901         uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1902 
1903         /* match with XLOG_REG_TYPE_* in xfs_log.h */
1904         static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1905             "bformat",
1906             "bchunk",
1907             "efi_format",
1908             "efd_format",
1909             "iformat",
1910             "icore",
1911             "iext",
1912             "ibroot",
1913             "ilocal",
1914             "iattr_ext",
1915             "iattr_broot",
1916             "iattr_local",
1917             "qformat",
1918             "dquot",
1919             "quotaoff",
1920             "LR header",
1921             "unmount",
1922             "commit",
1923             "trans header"
1924         };
1925         static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
1926             "SETATTR_NOT_SIZE",
1927             "SETATTR_SIZE",
1928             "INACTIVE",
1929             "CREATE",
1930             "CREATE_TRUNC",
1931             "TRUNCATE_FILE",
1932             "REMOVE",
1933             "LINK",
1934             "RENAME",
1935             "MKDIR",
1936             "RMDIR",
1937             "SYMLINK",
1938             "SET_DMATTRS",
1939             "GROWFS",
1940             "STRAT_WRITE",
1941             "DIOSTRAT",
1942             "WRITE_SYNC",
1943             "WRITEID",
1944             "ADDAFORK",
1945             "ATTRINVAL",
1946             "ATRUNCATE",
1947             "ATTR_SET",
1948             "ATTR_RM",
1949             "ATTR_FLAG",
1950             "CLEAR_AGI_BUCKET",
1951             "QM_SBCHANGE",
1952             "DUMMY1",
1953             "DUMMY2",
1954             "QM_QUOTAOFF",
1955             "QM_DQALLOC",
1956             "QM_SETQLIM",
1957             "QM_DQCLUSTER",
1958             "QM_QINOCREATE",
1959             "QM_QUOTAOFF_END",
1960             "SB_UNIT",
1961             "FSYNC_TS",
1962             "GROWFSRT_ALLOC",
1963             "GROWFSRT_ZERO",
1964             "GROWFSRT_FREE",
1965             "SWAPEXT"
1966         };
1967 
1968         xfs_warn(mp,
1969                 "xlog_write: reservation summary:\n"
1970                 "  trans type  = %s (%u)\n"
1971                 "  unit res    = %d bytes\n"
1972                 "  current res = %d bytes\n"
1973                 "  total reg   = %u bytes (o/flow = %u bytes)\n"
1974                 "  ophdrs      = %u (ophdr space = %u bytes)\n"
1975                 "  ophdr + reg = %u bytes\n"
1976                 "  num regions = %u\n",
1977                 ((ticket->t_trans_type <= 0 ||
1978                   ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
1979                   "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
1980                 ticket->t_trans_type,
1981                 ticket->t_unit_res,
1982                 ticket->t_curr_res,
1983                 ticket->t_res_arr_sum, ticket->t_res_o_flow,
1984                 ticket->t_res_num_ophdrs, ophdr_spc,
1985                 ticket->t_res_arr_sum +
1986                 ticket->t_res_o_flow + ophdr_spc,
1987                 ticket->t_res_num);
1988 
1989         for (i = 0; i < ticket->t_res_num; i++) {
1990                 uint r_type = ticket->t_res_arr[i].r_type;
1991                 xfs_warn(mp, "region[%u]: %s - %u bytes", i,
1992                             ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
1993                             "bad-rtype" : res_type_str[r_type-1]),
1994                             ticket->t_res_arr[i].r_len);
1995         }
1996 
1997         xfs_alert_tag(mp, XFS_PTAG_LOGRES,
1998                 "xlog_write: reservation ran out. Need to up reservation");
1999         xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
2000 }
2001 
2002 /*
2003  * Calculate the potential space needed by the log vector.  Each region gets
2004  * its own xlog_op_header_t and may need to be double word aligned.
2005  */
2006 static int
2007 xlog_write_calc_vec_length(
2008         struct xlog_ticket      *ticket,
2009         struct xfs_log_vec      *log_vector)
2010 {
2011         struct xfs_log_vec      *lv;
2012         int                     headers = 0;
2013         int                     len = 0;
2014         int                     i;
2015 
2016         /* acct for start rec of xact */
2017         if (ticket->t_flags & XLOG_TIC_INITED)
2018                 headers++;
2019 
2020         for (lv = log_vector; lv; lv = lv->lv_next) {
2021                 /* we don't write ordered log vectors */
2022                 if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED)
2023                         continue;
2024 
2025                 headers += lv->lv_niovecs;
2026 
2027                 for (i = 0; i < lv->lv_niovecs; i++) {
2028                         struct xfs_log_iovec    *vecp = &lv->lv_iovecp[i];
2029 
2030                         len += vecp->i_len;
2031                         xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
2032                 }
2033         }
2034 
2035         ticket->t_res_num_ophdrs += headers;
2036         len += headers * sizeof(struct xlog_op_header);
2037 
2038         return len;
2039 }
2040 
2041 /*
2042  * If first write for transaction, insert start record  We can't be trying to
2043  * commit if we are inited.  We can't have any "partial_copy" if we are inited.
2044  */
2045 static int
2046 xlog_write_start_rec(
2047         struct xlog_op_header   *ophdr,
2048         struct xlog_ticket      *ticket)
2049 {
2050         if (!(ticket->t_flags & XLOG_TIC_INITED))
2051                 return 0;
2052 
2053         ophdr->oh_tid   = cpu_to_be32(ticket->t_tid);
2054         ophdr->oh_clientid = ticket->t_clientid;
2055         ophdr->oh_len = 0;
2056         ophdr->oh_flags = XLOG_START_TRANS;
2057         ophdr->oh_res2 = 0;
2058 
2059         ticket->t_flags &= ~XLOG_TIC_INITED;
2060 
2061         return sizeof(struct xlog_op_header);
2062 }
2063 
2064 static xlog_op_header_t *
2065 xlog_write_setup_ophdr(
2066         struct xlog             *log,
2067         struct xlog_op_header   *ophdr,
2068         struct xlog_ticket      *ticket,
2069         uint                    flags)
2070 {
2071         ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2072         ophdr->oh_clientid = ticket->t_clientid;
2073         ophdr->oh_res2 = 0;
2074 
2075         /* are we copying a commit or unmount record? */
2076         ophdr->oh_flags = flags;
2077 
2078         /*
2079          * We've seen logs corrupted with bad transaction client ids.  This
2080          * makes sure that XFS doesn't generate them on.  Turn this into an EIO
2081          * and shut down the filesystem.
2082          */
2083         switch (ophdr->oh_clientid)  {
2084         case XFS_TRANSACTION:
2085         case XFS_VOLUME:
2086         case XFS_LOG:
2087                 break;
2088         default:
2089                 xfs_warn(log->l_mp,
2090                         "Bad XFS transaction clientid 0x%x in ticket 0x%p",
2091                         ophdr->oh_clientid, ticket);
2092                 return NULL;
2093         }
2094 
2095         return ophdr;
2096 }
2097 
2098 /*
2099  * Set up the parameters of the region copy into the log. This has
2100  * to handle region write split across multiple log buffers - this
2101  * state is kept external to this function so that this code can
2102  * be written in an obvious, self documenting manner.
2103  */
2104 static int
2105 xlog_write_setup_copy(
2106         struct xlog_ticket      *ticket,
2107         struct xlog_op_header   *ophdr,
2108         int                     space_available,
2109         int                     space_required,
2110         int                     *copy_off,
2111         int                     *copy_len,
2112         int                     *last_was_partial_copy,
2113         int                     *bytes_consumed)
2114 {
2115         int                     still_to_copy;
2116 
2117         still_to_copy = space_required - *bytes_consumed;
2118         *copy_off = *bytes_consumed;
2119 
2120         if (still_to_copy <= space_available) {
2121                 /* write of region completes here */
2122                 *copy_len = still_to_copy;
2123                 ophdr->oh_len = cpu_to_be32(*copy_len);
2124                 if (*last_was_partial_copy)
2125                         ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
2126                 *last_was_partial_copy = 0;
2127                 *bytes_consumed = 0;
2128                 return 0;
2129         }
2130 
2131         /* partial write of region, needs extra log op header reservation */
2132         *copy_len = space_available;
2133         ophdr->oh_len = cpu_to_be32(*copy_len);
2134         ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
2135         if (*last_was_partial_copy)
2136                 ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
2137         *bytes_consumed += *copy_len;
2138         (*last_was_partial_copy)++;
2139 
2140         /* account for new log op header */
2141         ticket->t_curr_res -= sizeof(struct xlog_op_header);
2142         ticket->t_res_num_ophdrs++;
2143 
2144         return sizeof(struct xlog_op_header);
2145 }
2146 
2147 static int
2148 xlog_write_copy_finish(
2149         struct xlog             *log,
2150         struct xlog_in_core     *iclog,
2151         uint                    flags,
2152         int                     *record_cnt,
2153         int                     *data_cnt,
2154         int                     *partial_copy,
2155         int                     *partial_copy_len,
2156         int                     log_offset,
2157         struct xlog_in_core     **commit_iclog)
2158 {
2159         if (*partial_copy) {
2160                 /*
2161                  * This iclog has already been marked WANT_SYNC by
2162                  * xlog_state_get_iclog_space.
2163                  */
2164                 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
2165                 *record_cnt = 0;
2166                 *data_cnt = 0;
2167                 return xlog_state_release_iclog(log, iclog);
2168         }
2169 
2170         *partial_copy = 0;
2171         *partial_copy_len = 0;
2172 
2173         if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
2174                 /* no more space in this iclog - push it. */
2175                 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
2176                 *record_cnt = 0;
2177                 *data_cnt = 0;
2178 
2179                 spin_lock(&log->l_icloglock);
2180                 xlog_state_want_sync(log, iclog);
2181                 spin_unlock(&log->l_icloglock);
2182 
2183                 if (!commit_iclog)
2184                         return xlog_state_release_iclog(log, iclog);
2185                 ASSERT(flags & XLOG_COMMIT_TRANS);
2186                 *commit_iclog = iclog;
2187         }
2188 
2189         return 0;
2190 }
2191 
2192 /*
2193  * Write some region out to in-core log
2194  *
2195  * This will be called when writing externally provided regions or when
2196  * writing out a commit record for a given transaction.
2197  *
2198  * General algorithm:
2199  *      1. Find total length of this write.  This may include adding to the
2200  *              lengths passed in.
2201  *      2. Check whether we violate the tickets reservation.
2202  *      3. While writing to this iclog
2203  *          A. Reserve as much space in this iclog as can get
2204  *          B. If this is first write, save away start lsn
2205  *          C. While writing this region:
2206  *              1. If first write of transaction, write start record
2207  *              2. Write log operation header (header per region)
2208  *              3. Find out if we can fit entire region into this iclog
2209  *              4. Potentially, verify destination memcpy ptr
2210  *              5. Memcpy (partial) region
2211  *              6. If partial copy, release iclog; otherwise, continue
2212  *                      copying more regions into current iclog
2213  *      4. Mark want sync bit (in simulation mode)
2214  *      5. Release iclog for potential flush to on-disk log.
2215  *
2216  * ERRORS:
2217  * 1.   Panic if reservation is overrun.  This should never happen since
2218  *      reservation amounts are generated internal to the filesystem.
2219  * NOTES:
2220  * 1. Tickets are single threaded data structures.
2221  * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
2222  *      syncing routine.  When a single log_write region needs to span
2223  *      multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
2224  *      on all log operation writes which don't contain the end of the
2225  *      region.  The XLOG_END_TRANS bit is used for the in-core log
2226  *      operation which contains the end of the continued log_write region.
2227  * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
2228  *      we don't really know exactly how much space will be used.  As a result,
2229  *      we don't update ic_offset until the end when we know exactly how many
2230  *      bytes have been written out.
2231  */
2232 int
2233 xlog_write(
2234         struct xlog             *log,
2235         struct xfs_log_vec      *log_vector,
2236         struct xlog_ticket      *ticket,
2237         xfs_lsn_t               *start_lsn,
2238         struct xlog_in_core     **commit_iclog,
2239         uint                    flags)
2240 {
2241         struct xlog_in_core     *iclog = NULL;
2242         struct xfs_log_iovec    *vecp;
2243         struct xfs_log_vec      *lv;
2244         int                     len;
2245         int                     index;
2246         int                     partial_copy = 0;
2247         int                     partial_copy_len = 0;
2248         int                     contwr = 0;
2249         int                     record_cnt = 0;
2250         int                     data_cnt = 0;
2251         int                     error;
2252 
2253         *start_lsn = 0;
2254 
2255         len = xlog_write_calc_vec_length(ticket, log_vector);
2256 
2257         /*
2258          * Region headers and bytes are already accounted for.
2259          * We only need to take into account start records and
2260          * split regions in this function.
2261          */
2262         if (ticket->t_flags & XLOG_TIC_INITED)
2263                 ticket->t_curr_res -= sizeof(xlog_op_header_t);
2264 
2265         /*
2266          * Commit record headers need to be accounted for. These
2267          * come in as separate writes so are easy to detect.
2268          */
2269         if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
2270                 ticket->t_curr_res -= sizeof(xlog_op_header_t);
2271 
2272         if (ticket->t_curr_res < 0)
2273                 xlog_print_tic_res(log->l_mp, ticket);
2274 
2275         index = 0;
2276         lv = log_vector;
2277         vecp = lv->lv_iovecp;
2278         while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
2279                 void            *ptr;
2280                 int             log_offset;
2281 
2282                 error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
2283                                                    &contwr, &log_offset);
2284                 if (error)
2285                         return error;
2286 
2287                 ASSERT(log_offset <= iclog->ic_size - 1);
2288                 ptr = iclog->ic_datap + log_offset;
2289 
2290                 /* start_lsn is the first lsn written to. That's all we need. */
2291                 if (!*start_lsn)
2292                         *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
2293 
2294                 /*
2295                  * This loop writes out as many regions as can fit in the amount
2296                  * of space which was allocated by xlog_state_get_iclog_space().
2297                  */
2298                 while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
2299                         struct xfs_log_iovec    *reg;
2300                         struct xlog_op_header   *ophdr;
2301                         int                     start_rec_copy;
2302                         int                     copy_len;
2303                         int                     copy_off;
2304                         bool                    ordered = false;
2305 
2306                         /* ordered log vectors have no regions to write */
2307                         if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) {
2308                                 ASSERT(lv->lv_niovecs == 0);
2309                                 ordered = true;
2310                                 goto next_lv;
2311                         }
2312 
2313                         reg = &vecp[index];
2314                         ASSERT(reg->i_len % sizeof(__int32_t) == 0);
2315                         ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
2316 
2317                         start_rec_copy = xlog_write_start_rec(ptr, ticket);
2318                         if (start_rec_copy) {
2319                                 record_cnt++;
2320                                 xlog_write_adv_cnt(&ptr, &len, &log_offset,
2321                                                    start_rec_copy);
2322                         }
2323 
2324                         ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
2325                         if (!ophdr)
2326                                 return XFS_ERROR(EIO);
2327 
2328                         xlog_write_adv_cnt(&ptr, &len, &log_offset,
2329                                            sizeof(struct xlog_op_header));
2330 
2331                         len += xlog_write_setup_copy(ticket, ophdr,
2332                                                      iclog->ic_size-log_offset,
2333                                                      reg->i_len,
2334                                                      &copy_off, &copy_len,
2335                                                      &partial_copy,
2336                                                      &partial_copy_len);
2337                         xlog_verify_dest_ptr(log, ptr);
2338 
2339                         /* copy region */
2340                         ASSERT(copy_len >= 0);
2341                         memcpy(ptr, reg->i_addr + copy_off, copy_len);
2342                         xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len);
2343 
2344                         copy_len += start_rec_copy + sizeof(xlog_op_header_t);
2345                         record_cnt++;
2346                         data_cnt += contwr ? copy_len : 0;
2347 
2348                         error = xlog_write_copy_finish(log, iclog, flags,
2349                                                        &record_cnt, &data_cnt,
2350                                                        &partial_copy,
2351                                                        &partial_copy_len,
2352                                                        log_offset,
2353                                                        commit_iclog);
2354                         if (error)
2355                                 return error;
2356 
2357                         /*
2358                          * if we had a partial copy, we need to get more iclog
2359                          * space but we don't want to increment the region
2360                          * index because there is still more is this region to
2361                          * write.
2362                          *
2363                          * If we completed writing this region, and we flushed
2364                          * the iclog (indicated by resetting of the record
2365                          * count), then we also need to get more log space. If
2366                          * this was the last record, though, we are done and
2367                          * can just return.
2368                          */
2369                         if (partial_copy)
2370                                 break;
2371 
2372                         if (++index == lv->lv_niovecs) {
2373 next_lv:
2374                                 lv = lv->lv_next;
2375                                 index = 0;
2376                                 if (lv)
2377                                         vecp = lv->lv_iovecp;
2378                         }
2379                         if (record_cnt == 0 && ordered == false) {
2380                                 if (!lv)
2381                                         return 0;
2382                                 break;
2383                         }
2384                 }
2385         }
2386 
2387         ASSERT(len == 0);
2388 
2389         xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
2390         if (!commit_iclog)
2391                 return xlog_state_release_iclog(log, iclog);
2392 
2393         ASSERT(flags & XLOG_COMMIT_TRANS);
2394         *commit_iclog = iclog;
2395         return 0;
2396 }
2397 
2398 
2399 /*****************************************************************************
2400  *
2401  *              State Machine functions
2402  *
2403  *****************************************************************************
2404  */
2405 
2406 /* Clean iclogs starting from the head.  This ordering must be
2407  * maintained, so an iclog doesn't become ACTIVE beyond one that
2408  * is SYNCING.  This is also required to maintain the notion that we use
2409  * a ordered wait queue to hold off would be writers to the log when every
2410  * iclog is trying to sync to disk.
2411  *
2412  * State Change: DIRTY -> ACTIVE
2413  */
2414 STATIC void
2415 xlog_state_clean_log(
2416         struct xlog *log)
2417 {
2418         xlog_in_core_t  *iclog;
2419         int changed = 0;
2420 
2421         iclog = log->l_iclog;
2422         do {
2423                 if (iclog->ic_state == XLOG_STATE_DIRTY) {
2424                         iclog->ic_state = XLOG_STATE_ACTIVE;
2425                         iclog->ic_offset       = 0;
2426                         ASSERT(iclog->ic_callback == NULL);
2427                         /*
2428                          * If the number of ops in this iclog indicate it just
2429                          * contains the dummy transaction, we can
2430                          * change state into IDLE (the second time around).
2431                          * Otherwise we should change the state into
2432                          * NEED a dummy.
2433                          * We don't need to cover the dummy.
2434                          */
2435                         if (!changed &&
2436                            (be32_to_cpu(iclog->ic_header.h_num_logops) ==
2437                                         XLOG_COVER_OPS)) {
2438                                 changed = 1;
2439                         } else {
2440                                 /*
2441                                  * We have two dirty iclogs so start over
2442                                  * This could also be num of ops indicates
2443                                  * this is not the dummy going out.
2444                                  */
2445                                 changed = 2;
2446                         }
2447                         iclog->ic_header.h_num_logops = 0;
2448                         memset(iclog->ic_header.h_cycle_data, 0,
2449                               sizeof(iclog->ic_header.h_cycle_data));
2450                         iclog->ic_header.h_lsn = 0;
2451                 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
2452                         /* do nothing */;
2453                 else
2454                         break;  /* stop cleaning */
2455                 iclog = iclog->ic_next;
2456         } while (iclog != log->l_iclog);
2457 
2458         /* log is locked when we are called */
2459         /*
2460          * Change state for the dummy log recording.
2461          * We usually go to NEED. But we go to NEED2 if the changed indicates
2462          * we are done writing the dummy record.
2463          * If we are done with the second dummy recored (DONE2), then
2464          * we go to IDLE.
2465          */
2466         if (changed) {
2467                 switch (log->l_covered_state) {
2468                 case XLOG_STATE_COVER_IDLE:
2469                 case XLOG_STATE_COVER_NEED:
2470                 case XLOG_STATE_COVER_NEED2:
2471                         log->l_covered_state = XLOG_STATE_COVER_NEED;
2472                         break;
2473 
2474                 case XLOG_STATE_COVER_DONE:
2475                         if (changed == 1)
2476                                 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2477                         else
2478                                 log->l_covered_state = XLOG_STATE_COVER_NEED;
2479                         break;
2480 
2481                 case XLOG_STATE_COVER_DONE2:
2482                         if (changed == 1)
2483                                 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2484                         else
2485                                 log->l_covered_state = XLOG_STATE_COVER_NEED;
2486                         break;
2487 
2488                 default:
2489                         ASSERT(0);
2490                 }
2491         }
2492 }       /* xlog_state_clean_log */
2493 
2494 STATIC xfs_lsn_t
2495 xlog_get_lowest_lsn(
2496         struct xlog     *log)
2497 {
2498         xlog_in_core_t  *lsn_log;
2499         xfs_lsn_t       lowest_lsn, lsn;
2500 
2501         lsn_log = log->l_iclog;
2502         lowest_lsn = 0;
2503         do {
2504             if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
2505                 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
2506                 if ((lsn && !lowest_lsn) ||
2507                     (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2508                         lowest_lsn = lsn;
2509                 }
2510             }
2511             lsn_log = lsn_log->ic_next;
2512         } while (lsn_log != log->l_iclog);
2513         return lowest_lsn;
2514 }
2515 
2516 
2517 STATIC void
2518 xlog_state_do_callback(
2519         struct xlog             *log,
2520         int                     aborted,
2521         struct xlog_in_core     *ciclog)
2522 {
2523         xlog_in_core_t     *iclog;
2524         xlog_in_core_t     *first_iclog;        /* used to know when we've
2525                                                  * processed all iclogs once */
2526         xfs_log_callback_t *cb, *cb_next;
2527         int                flushcnt = 0;
2528         xfs_lsn_t          lowest_lsn;
2529         int                ioerrors;    /* counter: iclogs with errors */
2530         int                loopdidcallbacks; /* flag: inner loop did callbacks*/
2531         int                funcdidcallbacks; /* flag: function did callbacks */
2532         int                repeats;     /* for issuing console warnings if
2533                                          * looping too many times */
2534         int                wake = 0;
2535 
2536         spin_lock(&log->l_icloglock);
2537         first_iclog = iclog = log->l_iclog;
2538         ioerrors = 0;
2539         funcdidcallbacks = 0;
2540         repeats = 0;
2541 
2542         do {
2543                 /*
2544                  * Scan all iclogs starting with the one pointed to by the
2545                  * log.  Reset this starting point each time the log is
2546                  * unlocked (during callbacks).
2547                  *
2548                  * Keep looping through iclogs until one full pass is made
2549                  * without running any callbacks.
2550                  */
2551                 first_iclog = log->l_iclog;
2552                 iclog = log->l_iclog;
2553                 loopdidcallbacks = 0;
2554                 repeats++;
2555 
2556                 do {
2557 
2558                         /* skip all iclogs in the ACTIVE & DIRTY states */
2559                         if (iclog->ic_state &
2560                             (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2561                                 iclog = iclog->ic_next;
2562                                 continue;
2563                         }
2564 
2565                         /*
2566                          * Between marking a filesystem SHUTDOWN and stopping
2567                          * the log, we do flush all iclogs to disk (if there
2568                          * wasn't a log I/O error). So, we do want things to
2569                          * go smoothly in case of just a SHUTDOWN  w/o a
2570                          * LOG_IO_ERROR.
2571                          */
2572                         if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2573                                 /*
2574                                  * Can only perform callbacks in order.  Since
2575                                  * this iclog is not in the DONE_SYNC/
2576                                  * DO_CALLBACK state, we skip the rest and
2577                                  * just try to clean up.  If we set our iclog
2578                                  * to DO_CALLBACK, we will not process it when
2579                                  * we retry since a previous iclog is in the
2580                                  * CALLBACK and the state cannot change since
2581                                  * we are holding the l_icloglock.
2582                                  */
2583                                 if (!(iclog->ic_state &
2584                                         (XLOG_STATE_DONE_SYNC |
2585                                                  XLOG_STATE_DO_CALLBACK))) {
2586                                         if (ciclog && (ciclog->ic_state ==
2587                                                         XLOG_STATE_DONE_SYNC)) {
2588                                                 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2589                                         }
2590                                         break;
2591                                 }
2592                                 /*
2593                                  * We now have an iclog that is in either the
2594                                  * DO_CALLBACK or DONE_SYNC states. The other
2595                                  * states (WANT_SYNC, SYNCING, or CALLBACK were
2596                                  * caught by the above if and are going to
2597                                  * clean (i.e. we aren't doing their callbacks)
2598                                  * see the above if.
2599                                  */
2600 
2601                                 /*
2602                                  * We will do one more check here to see if we
2603                                  * have chased our tail around.
2604                                  */
2605 
2606                                 lowest_lsn = xlog_get_lowest_lsn(log);
2607                                 if (lowest_lsn &&
2608                                     XFS_LSN_CMP(lowest_lsn,
2609                                                 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
2610                                         iclog = iclog->ic_next;
2611                                         continue; /* Leave this iclog for
2612                                                    * another thread */
2613                                 }
2614 
2615                                 iclog->ic_state = XLOG_STATE_CALLBACK;
2616 
2617 
2618                                 /*
2619                                  * Completion of a iclog IO does not imply that
2620                                  * a transaction has completed, as transactions
2621                                  * can be large enough to span many iclogs. We
2622                                  * cannot change the tail of the log half way
2623                                  * through a transaction as this may be the only
2624                                  * transaction in the log and moving th etail to
2625                                  * point to the middle of it will prevent
2626                                  * recovery from finding the start of the
2627                                  * transaction. Hence we should only update the
2628                                  * last_sync_lsn if this iclog contains
2629                                  * transaction completion callbacks on it.
2630                                  *
2631                                  * We have to do this before we drop the
2632                                  * icloglock to ensure we are the only one that
2633                                  * can update it.
2634                                  */
2635                                 ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
2636                                         be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
2637                                 if (iclog->ic_callback)
2638                                         atomic64_set(&log->l_last_sync_lsn,
2639                                                 be64_to_cpu(iclog->ic_header.h_lsn));
2640 
2641                         } else
2642                                 ioerrors++;
2643 
2644                         spin_unlock(&log->l_icloglock);
2645 
2646                         /*
2647                          * Keep processing entries in the callback list until
2648                          * we come around and it is empty.  We need to
2649                          * atomically see that the list is empty and change the
2650                          * state to DIRTY so that we don't miss any more
2651                          * callbacks being added.
2652                          */
2653                         spin_lock(&iclog->ic_callback_lock);
2654                         cb = iclog->ic_callback;
2655                         while (cb) {
2656                                 iclog->ic_callback_tail = &(iclog->ic_callback);
2657                                 iclog->ic_callback = NULL;
2658                                 spin_unlock(&iclog->ic_callback_lock);
2659 
2660                                 /* perform callbacks in the order given */
2661                                 for (; cb; cb = cb_next) {
2662                                         cb_next = cb->cb_next;
2663                                         cb->cb_func(cb->cb_arg, aborted);
2664                                 }
2665                                 spin_lock(&iclog->ic_callback_lock);
2666                                 cb = iclog->ic_callback;
2667                         }
2668 
2669                         loopdidcallbacks++;
2670                         funcdidcallbacks++;
2671 
2672                         spin_lock(&log->l_icloglock);
2673                         ASSERT(iclog->ic_callback == NULL);
2674                         spin_unlock(&iclog->ic_callback_lock);
2675                         if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2676                                 iclog->ic_state = XLOG_STATE_DIRTY;
2677 
2678                         /*
2679                          * Transition from DIRTY to ACTIVE if applicable.
2680                          * NOP if STATE_IOERROR.
2681                          */
2682                         xlog_state_clean_log(log);
2683 
2684                         /* wake up threads waiting in xfs_log_force() */
2685                         wake_up_all(&iclog->ic_force_wait);
2686 
2687                         iclog = iclog->ic_next;
2688                 } while (first_iclog != iclog);
2689 
2690                 if (repeats > 5000) {
2691                         flushcnt += repeats;
2692                         repeats = 0;
2693                         xfs_warn(log->l_mp,
2694                                 "%s: possible infinite loop (%d iterations)",
2695                                 __func__, flushcnt);
2696                 }
2697         } while (!ioerrors && loopdidcallbacks);
2698 
2699         /*
2700          * make one last gasp attempt to see if iclogs are being left in
2701          * limbo..
2702          */
2703 #ifdef DEBUG
2704         if (funcdidcallbacks) {
2705                 first_iclog = iclog = log->l_iclog;
2706                 do {
2707                         ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2708                         /*
2709                          * Terminate the loop if iclogs are found in states
2710                          * which will cause other threads to clean up iclogs.
2711                          *
2712                          * SYNCING - i/o completion will go through logs
2713                          * DONE_SYNC - interrupt thread should be waiting for
2714                          *              l_icloglock
2715                          * IOERROR - give up hope all ye who enter here
2716                          */
2717                         if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2718                             iclog->ic_state == XLOG_STATE_SYNCING ||
2719                             iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2720                             iclog->ic_state == XLOG_STATE_IOERROR )
2721                                 break;
2722                         iclog = iclog->ic_next;
2723                 } while (first_iclog != iclog);
2724         }
2725 #endif
2726 
2727         if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2728                 wake = 1;
2729         spin_unlock(&log->l_icloglock);
2730 
2731         if (wake)
2732                 wake_up_all(&log->l_flush_wait);
2733 }
2734 
2735 
2736 /*
2737  * Finish transitioning this iclog to the dirty state.
2738  *
2739  * Make sure that we completely execute this routine only when this is
2740  * the last call to the iclog.  There is a good chance that iclog flushes,
2741  * when we reach the end of the physical log, get turned into 2 separate
2742  * calls to bwrite.  Hence, one iclog flush could generate two calls to this
2743  * routine.  By using the reference count bwritecnt, we guarantee that only
2744  * the second completion goes through.
2745  *
2746  * Callbacks could take time, so they are done outside the scope of the
2747  * global state machine log lock.
2748  */
2749 STATIC void
2750 xlog_state_done_syncing(
2751         xlog_in_core_t  *iclog,
2752         int             aborted)
2753 {
2754         struct xlog        *log = iclog->ic_log;
2755 
2756         spin_lock(&log->l_icloglock);
2757 
2758         ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2759                iclog->ic_state == XLOG_STATE_IOERROR);
2760         ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
2761         ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2762 
2763 
2764         /*
2765          * If we got an error, either on the first buffer, or in the case of
2766          * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2767          * and none should ever be attempted to be written to disk
2768          * again.
2769          */
2770         if (iclog->ic_state != XLOG_STATE_IOERROR) {
2771                 if (--iclog->ic_bwritecnt == 1) {
2772                         spin_unlock(&log->l_icloglock);
2773                         return;
2774                 }
2775                 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2776         }
2777 
2778         /*
2779          * Someone could be sleeping prior to writing out the next
2780          * iclog buffer, we wake them all, one will get to do the
2781          * I/O, the others get to wait for the result.
2782          */
2783         wake_up_all(&iclog->ic_write_wait);
2784         spin_unlock(&log->l_icloglock);
2785         xlog_state_do_callback(log, aborted, iclog);    /* also cleans log */
2786 }       /* xlog_state_done_syncing */
2787 
2788 
2789 /*
2790  * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2791  * sleep.  We wait on the flush queue on the head iclog as that should be
2792  * the first iclog to complete flushing. Hence if all iclogs are syncing,
2793  * we will wait here and all new writes will sleep until a sync completes.
2794  *
2795  * The in-core logs are used in a circular fashion. They are not used
2796  * out-of-order even when an iclog past the head is free.
2797  *
2798  * return:
2799  *      * log_offset where xlog_write() can start writing into the in-core
2800  *              log's data space.
2801  *      * in-core log pointer to which xlog_write() should write.
2802  *      * boolean indicating this is a continued write to an in-core log.
2803  *              If this is the last write, then the in-core log's offset field
2804  *              needs to be incremented, depending on the amount of data which
2805  *              is copied.
2806  */
2807 STATIC int
2808 xlog_state_get_iclog_space(
2809         struct xlog             *log,
2810         int                     len,
2811         struct xlog_in_core     **iclogp,
2812         struct xlog_ticket      *ticket,
2813         int                     *continued_write,
2814         int                     *logoffsetp)
2815 {
2816         int               log_offset;
2817         xlog_rec_header_t *head;
2818         xlog_in_core_t    *iclog;
2819         int               error;
2820 
2821 restart:
2822         spin_lock(&log->l_icloglock);
2823         if (XLOG_FORCED_SHUTDOWN(log)) {
2824                 spin_unlock(&log->l_icloglock);
2825                 return XFS_ERROR(EIO);
2826         }
2827 
2828         iclog = log->l_iclog;
2829         if (iclog->ic_state != XLOG_STATE_ACTIVE) {
2830                 XFS_STATS_INC(xs_log_noiclogs);
2831 
2832                 /* Wait for log writes to have flushed */
2833                 xlog_wait(&log->l_flush_wait, &log->l_icloglock);
2834                 goto restart;
2835         }
2836 
2837         head = &iclog->ic_header;
2838 
2839         atomic_inc(&iclog->ic_refcnt);  /* prevents sync */
2840         log_offset = iclog->ic_offset;
2841 
2842         /* On the 1st write to an iclog, figure out lsn.  This works
2843          * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2844          * committing to.  If the offset is set, that's how many blocks
2845          * must be written.
2846          */
2847         if (log_offset == 0) {
2848                 ticket->t_curr_res -= log->l_iclog_hsize;
2849                 xlog_tic_add_region(ticket,
2850                                     log->l_iclog_hsize,
2851                                     XLOG_REG_TYPE_LRHEADER);
2852                 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2853                 head->h_lsn = cpu_to_be64(
2854                         xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
2855                 ASSERT(log->l_curr_block >= 0);
2856         }
2857 
2858         /* If there is enough room to write everything, then do it.  Otherwise,
2859          * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2860          * bit is on, so this will get flushed out.  Don't update ic_offset
2861          * until you know exactly how many bytes get copied.  Therefore, wait
2862          * until later to update ic_offset.
2863          *
2864          * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2865          * can fit into remaining data section.
2866          */
2867         if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2868                 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2869 
2870                 /*
2871                  * If I'm the only one writing to this iclog, sync it to disk.
2872                  * We need to do an atomic compare and decrement here to avoid
2873                  * racing with concurrent atomic_dec_and_lock() calls in
2874                  * xlog_state_release_iclog() when there is more than one
2875                  * reference to the iclog.
2876                  */
2877                 if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
2878                         /* we are the only one */
2879                         spin_unlock(&log->l_icloglock);
2880                         error = xlog_state_release_iclog(log, iclog);
2881                         if (error)
2882                                 return error;
2883                 } else {
2884                         spin_unlock(&log->l_icloglock);
2885                 }
2886                 goto restart;
2887         }
2888 
2889         /* Do we have enough room to write the full amount in the remainder
2890          * of this iclog?  Or must we continue a write on the next iclog and
2891          * mark this iclog as completely taken?  In the case where we switch
2892          * iclogs (to mark it taken), this particular iclog will release/sync
2893          * to disk in xlog_write().
2894          */
2895         if (len <= iclog->ic_size - iclog->ic_offset) {
2896                 *continued_write = 0;
2897                 iclog->ic_offset += len;
2898         } else {
2899                 *continued_write = 1;
2900                 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2901         }
2902         *iclogp = iclog;
2903 
2904         ASSERT(iclog->ic_offset <= iclog->ic_size);
2905         spin_unlock(&log->l_icloglock);
2906 
2907         *logoffsetp = log_offset;
2908         return 0;
2909 }       /* xlog_state_get_iclog_space */
2910 
2911 /* The first cnt-1 times through here we don't need to
2912  * move the grant write head because the permanent
2913  * reservation has reserved cnt times the unit amount.
2914  * Release part of current permanent unit reservation and
2915  * reset current reservation to be one units worth.  Also
2916  * move grant reservation head forward.
2917  */
2918 STATIC void
2919 xlog_regrant_reserve_log_space(
2920         struct xlog             *log,
2921         struct xlog_ticket      *ticket)
2922 {
2923         trace_xfs_log_regrant_reserve_enter(log, ticket);
2924 
2925         if (ticket->t_cnt > 0)
2926                 ticket->t_cnt--;
2927 
2928         xlog_grant_sub_space(log, &log->l_reserve_head.grant,
2929                                         ticket->t_curr_res);
2930         xlog_grant_sub_space(log, &log->l_write_head.grant,
2931                                         ticket->t_curr_res);
2932         ticket->t_curr_res = ticket->t_unit_res;
2933         xlog_tic_reset_res(ticket);
2934 
2935         trace_xfs_log_regrant_reserve_sub(log, ticket);
2936 
2937         /* just return if we still have some of the pre-reserved space */
2938         if (ticket->t_cnt > 0)
2939                 return;
2940 
2941         xlog_grant_add_space(log, &log->l_reserve_head.grant,
2942                                         ticket->t_unit_res);
2943 
2944         trace_xfs_log_regrant_reserve_exit(log, ticket);
2945 
2946         ticket->t_curr_res = ticket->t_unit_res;
2947         xlog_tic_reset_res(ticket);
2948 }       /* xlog_regrant_reserve_log_space */
2949 
2950 
2951 /*
2952  * Give back the space left from a reservation.
2953  *
2954  * All the information we need to make a correct determination of space left
2955  * is present.  For non-permanent reservations, things are quite easy.  The
2956  * count should have been decremented to zero.  We only need to deal with the
2957  * space remaining in the current reservation part of the ticket.  If the
2958  * ticket contains a permanent reservation, there may be left over space which
2959  * needs to be released.  A count of N means that N-1 refills of the current
2960  * reservation can be done before we need to ask for more space.  The first
2961  * one goes to fill up the first current reservation.  Once we run out of
2962  * space, the count will stay at zero and the only space remaining will be
2963  * in the current reservation field.
2964  */
2965 STATIC void
2966 xlog_ungrant_log_space(
2967         struct xlog             *log,
2968         struct xlog_ticket      *ticket)
2969 {
2970         int     bytes;
2971 
2972         if (ticket->t_cnt > 0)
2973                 ticket->t_cnt--;
2974 
2975         trace_xfs_log_ungrant_enter(log, ticket);
2976         trace_xfs_log_ungrant_sub(log, ticket);
2977 
2978         /*
2979          * If this is a permanent reservation ticket, we may be able to free
2980          * up more space based on the remaining count.
2981          */
2982         bytes = ticket->t_curr_res;
2983         if (ticket->t_cnt > 0) {
2984                 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
2985                 bytes += ticket->t_unit_res*ticket->t_cnt;
2986         }
2987 
2988         xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes);
2989         xlog_grant_sub_space(log, &log->l_write_head.grant, bytes);
2990 
2991         trace_xfs_log_ungrant_exit(log, ticket);
2992 
2993         xfs_log_space_wake(log->l_mp);
2994 }
2995 
2996 /*
2997  * Flush iclog to disk if this is the last reference to the given iclog and
2998  * the WANT_SYNC bit is set.
2999  *
3000  * When this function is entered, the iclog is not necessarily in the
3001  * WANT_SYNC state.  It may be sitting around waiting to get filled.
3002  *
3003  *
3004  */
3005 STATIC int
3006 xlog_state_release_iclog(
3007         struct xlog             *log,
3008         struct xlog_in_core     *iclog)
3009 {
3010         int             sync = 0;       /* do we sync? */
3011 
3012         if (iclog->ic_state & XLOG_STATE_IOERROR)
3013                 return XFS_ERROR(EIO);
3014 
3015         ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
3016         if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
3017                 return 0;
3018 
3019         if (iclog->ic_state & XLOG_STATE_IOERROR) {
3020                 spin_unlock(&log->l_icloglock);
3021                 return XFS_ERROR(EIO);
3022         }
3023         ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
3024                iclog->ic_state == XLOG_STATE_WANT_SYNC);
3025 
3026         if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
3027                 /* update tail before writing to iclog */
3028                 xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
3029                 sync++;
3030                 iclog->ic_state = XLOG_STATE_SYNCING;
3031                 iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
3032                 xlog_verify_tail_lsn(log, iclog, tail_lsn);
3033                 /* cycle incremented when incrementing curr_block */
3034         }
3035         spin_unlock(&log->l_icloglock);
3036 
3037         /*
3038          * We let the log lock go, so it's possible that we hit a log I/O
3039          * error or some other SHUTDOWN condition that marks the iclog
3040          * as XLOG_STATE_IOERROR before the bwrite. However, we know that
3041          * this iclog has consistent data, so we ignore IOERROR
3042          * flags after this point.
3043          */
3044         if (sync)
3045                 return xlog_sync(log, iclog);
3046         return 0;
3047 }       /* xlog_state_release_iclog */
3048 
3049 
3050 /*
3051  * This routine will mark the current iclog in the ring as WANT_SYNC
3052  * and move the current iclog pointer to the next iclog in the ring.
3053  * When this routine is called from xlog_state_get_iclog_space(), the
3054  * exact size of the iclog has not yet been determined.  All we know is
3055  * that every data block.  We have run out of space in this log record.
3056  */
3057 STATIC void
3058 xlog_state_switch_iclogs(
3059         struct xlog             *log,
3060         struct xlog_in_core     *iclog,
3061         int                     eventual_size)
3062 {
3063         ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
3064         if (!eventual_size)
3065                 eventual_size = iclog->ic_offset;
3066         iclog->ic_state = XLOG_STATE_WANT_SYNC;
3067         iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
3068         log->l_prev_block = log->l_curr_block;
3069         log->l_prev_cycle = log->l_curr_cycle;
3070 
3071         /* roll log?: ic_offset changed later */
3072         log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
3073 
3074         /* Round up to next log-sunit */
3075         if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
3076             log->l_mp->m_sb.sb_logsunit > 1) {
3077                 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
3078                 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
3079         }
3080 
3081         if (log->l_curr_block >= log->l_logBBsize) {
3082                 log->l_curr_cycle++;
3083                 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
3084                         log->l_curr_cycle++;
3085                 log->l_curr_block -= log->l_logBBsize;
3086                 ASSERT(log->l_curr_block >= 0);
3087         }
3088         ASSERT(iclog == log->l_iclog);
3089         log->l_iclog = iclog->ic_next;
3090 }       /* xlog_state_switch_iclogs */
3091 
3092 /*
3093  * Write out all data in the in-core log as of this exact moment in time.
3094  *
3095  * Data may be written to the in-core log during this call.  However,
3096  * we don't guarantee this data will be written out.  A change from past
3097  * implementation means this routine will *not* write out zero length LRs.
3098  *
3099  * Basically, we try and perform an intelligent scan of the in-core logs.
3100  * If we determine there is no flushable data, we just return.  There is no
3101  * flushable data if:
3102  *
3103  *      1. the current iclog is active and has no data; the previous iclog
3104  *              is in the active or dirty state.
3105  *      2. the current iclog is drity, and the previous iclog is in the
3106  *              active or dirty state.
3107  *
3108  * We may sleep if:
3109  *
3110  *      1. the current iclog is not in the active nor dirty state.
3111  *      2. the current iclog dirty, and the previous iclog is not in the
3112  *              active nor dirty state.
3113  *      3. the current iclog is active, and there is another thread writing
3114  *              to this particular iclog.
3115  *      4. a) the current iclog is active and has no other writers
3116  *         b) when we return from flushing out this iclog, it is still
3117  *              not in the active nor dirty state.
3118  */
3119 int
3120 _xfs_log_force(
3121         struct xfs_mount        *mp,
3122         uint                    flags,
3123         int                     *log_flushed)
3124 {
3125         struct xlog             *log = mp->m_log;
3126         struct xlog_in_core     *iclog;
3127         xfs_lsn_t               lsn;
3128 
3129         XFS_STATS_INC(xs_log_force);
3130 
3131         xlog_cil_force(log);
3132 
3133         spin_lock(&log->l_icloglock);
3134 
3135         iclog = log->l_iclog;
3136         if (iclog->ic_state & XLOG_STATE_IOERROR) {
3137                 spin_unlock(&log->l_icloglock);
3138                 return XFS_ERROR(EIO);
3139         }
3140 
3141         /* If the head iclog is not active nor dirty, we just attach
3142          * ourselves to the head and go to sleep.
3143          */
3144         if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3145             iclog->ic_state == XLOG_STATE_DIRTY) {
3146                 /*
3147                  * If the head is dirty or (active and empty), then
3148                  * we need to look at the previous iclog.  If the previous
3149                  * iclog is active or dirty we are done.  There is nothing
3150                  * to sync out.  Otherwise, we attach ourselves to the
3151                  * previous iclog and go to sleep.
3152                  */
3153                 if (iclog->ic_state == XLOG_STATE_DIRTY ||
3154                     (atomic_read(&iclog->ic_refcnt) == 0
3155                      && iclog->ic_offset == 0)) {
3156                         iclog = iclog->ic_prev;
3157                         if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3158                             iclog->ic_state == XLOG_STATE_DIRTY)
3159                                 goto no_sleep;
3160                         else
3161                                 goto maybe_sleep;
3162                 } else {
3163                         if (atomic_read(&iclog->ic_refcnt) == 0) {
3164                                 /* We are the only one with access to this
3165                                  * iclog.  Flush it out now.  There should
3166                                  * be a roundoff of zero to show that someone
3167                                  * has already taken care of the roundoff from
3168                                  * the previous sync.
3169                                  */
3170                                 atomic_inc(&iclog->ic_refcnt);
3171                                 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
3172                                 xlog_state_switch_iclogs(log, iclog, 0);
3173                                 spin_unlock(&log->l_icloglock);
3174 
3175                                 if (xlog_state_release_iclog(log, iclog))
3176                                         return XFS_ERROR(EIO);
3177 
3178                                 if (log_flushed)
3179                                         *log_flushed = 1;
3180                                 spin_lock(&log->l_icloglock);
3181                                 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
3182                                     iclog->ic_state != XLOG_STATE_DIRTY)
3183                                         goto maybe_sleep;
3184                                 else
3185                                         goto no_sleep;
3186                         } else {
3187                                 /* Someone else is writing to this iclog.
3188                                  * Use its call to flush out the data.  However,
3189                                  * the other thread may not force out this LR,
3190                                  * so we mark it WANT_SYNC.
3191                                  */
3192                                 xlog_state_switch_iclogs(log, iclog, 0);
3193                                 goto maybe_sleep;
3194                         }
3195                 }
3196         }
3197 
3198         /* By the time we come around again, the iclog could've been filled
3199          * which would give it another lsn.  If we have a new lsn, just
3200          * return because the relevant data has been flushed.
3201          */
3202 maybe_sleep:
3203         if (flags & XFS_LOG_SYNC) {
3204                 /*
3205                  * We must check if we're shutting down here, before
3206                  * we wait, while we're holding the l_icloglock.
3207                  * Then we check again after waking up, in case our
3208                  * sleep was disturbed by a bad news.
3209                  */
3210                 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3211                         spin_unlock(&log->l_icloglock);
3212                         return XFS_ERROR(EIO);
3213                 }
3214                 XFS_STATS_INC(xs_log_force_sleep);
3215                 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
3216                 /*
3217                  * No need to grab the log lock here since we're
3218                  * only deciding whether or not to return EIO
3219                  * and the memory read should be atomic.
3220                  */
3221                 if (iclog->ic_state & XLOG_STATE_IOERROR)
3222                         return XFS_ERROR(EIO);
3223                 if (log_flushed)
3224                         *log_flushed = 1;
3225         } else {
3226 
3227 no_sleep:
3228                 spin_unlock(&log->l_icloglock);
3229         }
3230         return 0;
3231 }
3232 
3233 /*
3234  * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3235  * about errors or whether the log was flushed or not. This is the normal
3236  * interface to use when trying to unpin items or move the log forward.
3237  */
3238 void
3239 xfs_log_force(
3240         xfs_mount_t     *mp,
3241         uint            flags)
3242 {
3243         int     error;
3244 
3245         trace_xfs_log_force(mp, 0);
3246         error = _xfs_log_force(mp, flags, NULL);
3247         if (error)
3248                 xfs_warn(mp, "%s: error %d returned.", __func__, error);
3249 }
3250 
3251 /*
3252  * Force the in-core log to disk for a specific LSN.
3253  *
3254  * Find in-core log with lsn.
3255  *      If it is in the DIRTY state, just return.
3256  *      If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3257  *              state and go to sleep or return.
3258  *      If it is in any other state, go to sleep or return.
3259  *
3260  * Synchronous forces are implemented with a signal variable. All callers
3261  * to force a given lsn to disk will wait on a the sv attached to the
3262  * specific in-core log.  When given in-core log finally completes its
3263  * write to disk, that thread will wake up all threads waiting on the
3264  * sv.
3265  */
3266 int
3267 _xfs_log_force_lsn(
3268         struct xfs_mount        *mp,
3269         xfs_lsn_t               lsn,
3270         uint                    flags,
3271         int                     *log_flushed)
3272 {
3273         struct xlog             *log = mp->m_log;
3274         struct xlog_in_core     *iclog;
3275         int                     already_slept = 0;
3276 
3277         ASSERT(lsn != 0);
3278 
3279         XFS_STATS_INC(xs_log_force);
3280 
3281         lsn = xlog_cil_force_lsn(log, lsn);
3282         if (lsn == NULLCOMMITLSN)
3283                 return 0;
3284 
3285 try_again:
3286         spin_lock(&log->l_icloglock);
3287         iclog = log->l_iclog;
3288         if (iclog->ic_state & XLOG_STATE_IOERROR) {
3289                 spin_unlock(&log->l_icloglock);
3290                 return XFS_ERROR(EIO);
3291         }
3292 
3293         do {
3294                 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3295                         iclog = iclog->ic_next;
3296                         continue;
3297                 }
3298 
3299                 if (iclog->ic_state == XLOG_STATE_DIRTY) {
3300                         spin_unlock(&log->l_icloglock);
3301                         return 0;
3302                 }
3303 
3304                 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3305                         /*
3306                          * We sleep here if we haven't already slept (e.g.
3307                          * this is the first time we've looked at the correct
3308                          * iclog buf) and the buffer before us is going to
3309                          * be sync'ed. The reason for this is that if we
3310                          * are doing sync transactions here, by waiting for
3311                          * the previous I/O to complete, we can allow a few
3312                          * more transactions into this iclog before we close
3313                          * it down.
3314                          *
3315                          * Otherwise, we mark the buffer WANT_SYNC, and bump
3316                          * up the refcnt so we can release the log (which
3317                          * drops the ref count).  The state switch keeps new
3318                          * transaction commits from using this buffer.  When
3319                          * the current commits finish writing into the buffer,
3320                          * the refcount will drop to zero and the buffer will
3321                          * go out then.
3322                          */
3323                         if (!already_slept &&
3324                             (iclog->ic_prev->ic_state &
3325                              (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
3326                                 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3327 
3328                                 XFS_STATS_INC(xs_log_force_sleep);
3329 
3330                                 xlog_wait(&iclog->ic_prev->ic_write_wait,
3331                                                         &log->l_icloglock);
3332                                 if (log_flushed)
3333                                         *log_flushed = 1;
3334                                 already_slept = 1;
3335                                 goto try_again;
3336                         }
3337                         atomic_inc(&iclog->ic_refcnt);
3338                         xlog_state_switch_iclogs(log, iclog, 0);
3339                         spin_unlock(&log->l_icloglock);
3340                         if (xlog_state_release_iclog(log, iclog))
3341                                 return XFS_ERROR(EIO);
3342                         if (log_flushed)
3343                                 *log_flushed = 1;
3344                         spin_lock(&log->l_icloglock);
3345                 }
3346 
3347                 if ((flags & XFS_LOG_SYNC) && /* sleep */
3348                     !(iclog->ic_state &
3349                       (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3350                         /*
3351                          * Don't wait on completion if we know that we've
3352                          * gotten a log write error.
3353                          */
3354                         if (iclog->ic_state & XLOG_STATE_IOERROR) {
3355                                 spin_unlock(&log->l_icloglock);
3356                                 return XFS_ERROR(EIO);
3357                         }
3358                         XFS_STATS_INC(xs_log_force_sleep);
3359                         xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
3360                         /*
3361                          * No need to grab the log lock here since we're
3362                          * only deciding whether or not to return EIO
3363                          * and the memory read should be atomic.
3364                          */
3365                         if (iclog->ic_state & XLOG_STATE_IOERROR)
3366                                 return XFS_ERROR(EIO);
3367 
3368                         if (log_flushed)
3369                                 *log_flushed = 1;
3370                 } else {                /* just return */
3371                         spin_unlock(&log->l_icloglock);
3372                 }
3373 
3374                 return 0;
3375         } while (iclog != log->l_iclog);
3376 
3377         spin_unlock(&log->l_icloglock);
3378         return 0;
3379 }
3380 
3381 /*
3382  * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3383  * about errors or whether the log was flushed or not. This is the normal
3384  * interface to use when trying to unpin items or move the log forward.
3385  */
3386 void
3387 xfs_log_force_lsn(
3388         xfs_mount_t     *mp,
3389         xfs_lsn_t       lsn,
3390         uint            flags)
3391 {
3392         int     error;
3393 
3394         trace_xfs_log_force(mp, lsn);
3395         error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
3396         if (error)
3397                 xfs_warn(mp, "%s: error %d returned.", __func__, error);
3398 }
3399 
3400 /*
3401  * Called when we want to mark the current iclog as being ready to sync to
3402  * disk.
3403  */
3404 STATIC void
3405 xlog_state_want_sync(
3406         struct xlog             *log,
3407         struct xlog_in_core     *iclog)
3408 {
3409         assert_spin_locked(&log->l_icloglock);
3410 
3411         if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3412                 xlog_state_switch_iclogs(log, iclog, 0);
3413         } else {
3414                 ASSERT(iclog->ic_state &
3415                         (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3416         }
3417 }
3418 
3419 
3420 /*****************************************************************************
3421  *
3422  *              TICKET functions
3423  *
3424  *****************************************************************************
3425  */
3426 
3427 /*
3428  * Free a used ticket when its refcount falls to zero.
3429  */
3430 void
3431 xfs_log_ticket_put(
3432         xlog_ticket_t   *ticket)
3433 {
3434         ASSERT(atomic_read(&ticket->t_ref) > 0);
3435         if (atomic_dec_and_test(&ticket->t_ref))
3436                 kmem_zone_free(xfs_log_ticket_zone, ticket);
3437 }
3438 
3439 xlog_ticket_t *
3440 xfs_log_ticket_get(
3441         xlog_ticket_t   *ticket)
3442 {
3443         ASSERT(atomic_read(&ticket->t_ref) > 0);
3444         atomic_inc(&ticket->t_ref);
3445         return ticket;
3446 }
3447 
3448 /*
3449  * Figure out the total log space unit (in bytes) that would be
3450  * required for a log ticket.
3451  */
3452 int
3453 xfs_log_calc_unit_res(
3454         struct xfs_mount        *mp,
3455         int                     unit_bytes)
3456 {
3457         struct xlog             *log = mp->m_log;
3458         int                     iclog_space;
3459         uint                    num_headers;
3460 
3461         /*
3462          * Permanent reservations have up to 'cnt'-1 active log operations
3463          * in the log.  A unit in this case is the amount of space for one
3464          * of these log operations.  Normal reservations have a cnt of 1
3465          * and their unit amount is the total amount of space required.
3466          *
3467          * The following lines of code account for non-transaction data
3468          * which occupy space in the on-disk log.
3469          *
3470          * Normal form of a transaction is:
3471          * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3472          * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3473          *
3474          * We need to account for all the leadup data and trailer data
3475          * around the transaction data.
3476          * And then we need to account for the worst case in terms of using
3477          * more space.
3478          * The worst case will happen if:
3479          * - the placement of the transaction happens to be such that the
3480          *   roundoff is at its maximum
3481          * - the transaction data is synced before the commit record is synced
3482          *   i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3483          *   Therefore the commit record is in its own Log Record.
3484          *   This can happen as the commit record is called with its
3485          *   own region to xlog_write().
3486          *   This then means that in the worst case, roundoff can happen for
3487          *   the commit-rec as well.
3488          *   The commit-rec is smaller than padding in this scenario and so it is
3489          *   not added separately.
3490          */
3491 
3492         /* for trans header */
3493         unit_bytes += sizeof(xlog_op_header_t);
3494         unit_bytes += sizeof(xfs_trans_header_t);
3495 
3496         /* for start-rec */
3497         unit_bytes += sizeof(xlog_op_header_t);
3498 
3499         /*
3500          * for LR headers - the space for data in an iclog is the size minus
3501          * the space used for the headers. If we use the iclog size, then we
3502          * undercalculate the number of headers required.
3503          *
3504          * Furthermore - the addition of op headers for split-recs might
3505          * increase the space required enough to require more log and op
3506          * headers, so take that into account too.
3507          *
3508          * IMPORTANT: This reservation makes the assumption that if this
3509          * transaction is the first in an iclog and hence has the LR headers
3510          * accounted to it, then the remaining space in the iclog is
3511          * exclusively for this transaction.  i.e. if the transaction is larger
3512          * than the iclog, it will be the only thing in that iclog.
3513          * Fundamentally, this means we must pass the entire log vector to
3514          * xlog_write to guarantee this.
3515          */
3516         iclog_space = log->l_iclog_size - log->l_iclog_hsize;
3517         num_headers = howmany(unit_bytes, iclog_space);
3518 
3519         /* for split-recs - ophdrs added when data split over LRs */
3520         unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3521 
3522         /* add extra header reservations if we overrun */
3523         while (!num_headers ||
3524                howmany(unit_bytes, iclog_space) > num_headers) {
3525                 unit_bytes += sizeof(xlog_op_header_t);
3526                 num_headers++;
3527         }
3528         unit_bytes += log->l_iclog_hsize * num_headers;
3529 
3530         /* for commit-rec LR header - note: padding will subsume the ophdr */
3531         unit_bytes += log->l_iclog_hsize;
3532 
3533         /* for roundoff padding for transaction data and one for commit record */
3534         if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) {
3535                 /* log su roundoff */
3536                 unit_bytes += 2 * mp->m_sb.sb_logsunit;
3537         } else {
3538                 /* BB roundoff */
3539                 unit_bytes += 2 * BBSIZE;
3540         }
3541 
3542         return unit_bytes;
3543 }
3544 
3545 /*
3546  * Allocate and initialise a new log ticket.
3547  */
3548 struct xlog_ticket *
3549 xlog_ticket_alloc(
3550         struct xlog             *log,
3551         int                     unit_bytes,
3552         int                     cnt,
3553         char                    client,
3554         bool                    permanent,
3555         xfs_km_flags_t          alloc_flags)
3556 {
3557         struct xlog_ticket      *tic;
3558         int                     unit_res;
3559 
3560         tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
3561         if (!tic)
3562                 return NULL;
3563 
3564         unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes);
3565 
3566         atomic_set(&tic->t_ref, 1);
3567         tic->t_task             = current;
3568         INIT_LIST_HEAD(&tic->t_queue);
3569         tic->t_unit_res         = unit_res;
3570         tic->t_curr_res         = unit_res;
3571         tic->t_cnt              = cnt;
3572         tic->t_ocnt             = cnt;
3573         tic->t_tid              = prandom_u32();
3574         tic->t_clientid         = client;
3575         tic->t_flags            = XLOG_TIC_INITED;
3576         tic->t_trans_type       = 0;
3577         if (permanent)
3578                 tic->t_flags |= XLOG_TIC_PERM_RESERV;
3579 
3580         xlog_tic_reset_res(tic);
3581 
3582         return tic;
3583 }
3584 
3585 
3586 /******************************************************************************
3587  *
3588  *              Log debug routines
3589  *
3590  ******************************************************************************
3591  */
3592 #if defined(DEBUG)
3593 /*
3594  * Make sure that the destination ptr is within the valid data region of
3595  * one of the iclogs.  This uses backup pointers stored in a different
3596  * part of the log in case we trash the log structure.
3597  */
3598 void
3599 xlog_verify_dest_ptr(
3600         struct xlog     *log,
3601         char            *ptr)
3602 {
3603         int i;
3604         int good_ptr = 0;
3605 
3606         for (i = 0; i < log->l_iclog_bufs; i++) {
3607                 if (ptr >= log->l_iclog_bak[i] &&
3608                     ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
3609                         good_ptr++;
3610         }
3611 
3612         if (!good_ptr)
3613                 xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
3614 }
3615 
3616 /*
3617  * Check to make sure the grant write head didn't just over lap the tail.  If
3618  * the cycles are the same, we can't be overlapping.  Otherwise, make sure that
3619  * the cycles differ by exactly one and check the byte count.
3620  *
3621  * This check is run unlocked, so can give false positives. Rather than assert
3622  * on failures, use a warn-once flag and a panic tag to allow the admin to
3623  * determine if they want to panic the machine when such an error occurs. For
3624  * debug kernels this will have the same effect as using an assert but, unlinke
3625  * an assert, it can be turned off at runtime.
3626  */
3627 STATIC void
3628 xlog_verify_grant_tail(
3629         struct xlog     *log)
3630 {
3631         int             tail_cycle, tail_blocks;
3632         int             cycle, space;
3633 
3634         xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space);
3635         xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
3636         if (tail_cycle != cycle) {
3637                 if (cycle - 1 != tail_cycle &&
3638                     !(log->l_flags & XLOG_TAIL_WARN)) {
3639                         xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3640                                 "%s: cycle - 1 != tail_cycle", __func__);
3641                         log->l_flags |= XLOG_TAIL_WARN;
3642                 }
3643 
3644                 if (space > BBTOB(tail_blocks) &&
3645                     !(log->l_flags & XLOG_TAIL_WARN)) {
3646                         xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3647                                 "%s: space > BBTOB(tail_blocks)", __func__);
3648                         log->l_flags |= XLOG_TAIL_WARN;
3649                 }
3650         }
3651 }
3652 
3653 /* check if it will fit */
3654 STATIC void
3655 xlog_verify_tail_lsn(
3656         struct xlog             *log,
3657         struct xlog_in_core     *iclog,
3658         xfs_lsn_t               tail_lsn)
3659 {
3660     int blocks;
3661 
3662     if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3663         blocks =
3664             log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3665         if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
3666                 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
3667     } else {
3668         ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3669 
3670         if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
3671                 xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
3672 
3673         blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3674         if (blocks < BTOBB(iclog->ic_offset) + 1)
3675                 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
3676     }
3677 }       /* xlog_verify_tail_lsn */
3678 
3679 /*
3680  * Perform a number of checks on the iclog before writing to disk.
3681  *
3682  * 1. Make sure the iclogs are still circular
3683  * 2. Make sure we have a good magic number
3684  * 3. Make sure we don't have magic numbers in the data
3685  * 4. Check fields of each log operation header for:
3686  *      A. Valid client identifier
3687  *      B. tid ptr value falls in valid ptr space (user space code)
3688  *      C. Length in log record header is correct according to the
3689  *              individual operation headers within record.
3690  * 5. When a bwrite will occur within 5 blocks of the front of the physical
3691  *      log, check the preceding blocks of the physical log to make sure all
3692  *      the cycle numbers agree with the current cycle number.
3693  */
3694 STATIC void
3695 xlog_verify_iclog(
3696         struct xlog             *log,
3697         struct xlog_in_core     *iclog,
3698         int                     count,
3699         bool                    syncing)
3700 {
3701         xlog_op_header_t        *ophead;
3702         xlog_in_core_t          *icptr;
3703         xlog_in_core_2_t        *xhdr;
3704         xfs_caddr_t             ptr;
3705         xfs_caddr_t             base_ptr;
3706         __psint_t               field_offset;
3707         __uint8_t               clientid;
3708         int                     len, i, j, k, op_len;
3709         int                     idx;
3710 
3711         /* check validity of iclog pointers */
3712         spin_lock(&log->l_icloglock);
3713         icptr = log->l_iclog;
3714         for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next)
3715                 ASSERT(icptr);
3716 
3717         if (icptr != log->l_iclog)
3718                 xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
3719         spin_unlock(&log->l_icloglock);
3720 
3721         /* check log magic numbers */
3722         if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
3723                 xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
3724 
3725         ptr = (xfs_caddr_t) &iclog->ic_header;
3726         for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
3727              ptr += BBSIZE) {
3728                 if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
3729                         xfs_emerg(log->l_mp, "%s: unexpected magic num",
3730                                 __func__);
3731         }
3732 
3733         /* check fields */
3734         len = be32_to_cpu(iclog->ic_header.h_num_logops);
3735         ptr = iclog->ic_datap;
3736         base_ptr = ptr;
3737         ophead = (xlog_op_header_t *)ptr;
3738         xhdr = iclog->ic_data;
3739         for (i = 0; i < len; i++) {
3740                 ophead = (xlog_op_header_t *)ptr;
3741 
3742                 /* clientid is only 1 byte */
3743                 field_offset = (__psint_t)
3744                                ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
3745                 if (!syncing || (field_offset & 0x1ff)) {
3746                         clientid = ophead->oh_clientid;
3747                 } else {
3748                         idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
3749                         if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3750                                 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3751                                 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3752                                 clientid = xlog_get_client_id(
3753                                         xhdr[j].hic_xheader.xh_cycle_data[k]);
3754                         } else {
3755                                 clientid = xlog_get_client_id(
3756                                         iclog->ic_header.h_cycle_data[idx]);
3757                         }
3758                 }
3759                 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
3760                         xfs_warn(log->l_mp,
3761                                 "%s: invalid clientid %d op 0x%p offset 0x%lx",
3762                                 __func__, clientid, ophead,
3763                                 (unsigned long)field_offset);
3764 
3765                 /* check length */
3766                 field_offset = (__psint_t)
3767                                ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
3768                 if (!syncing || (field_offset & 0x1ff)) {
3769                         op_len = be32_to_cpu(ophead->oh_len);
3770                 } else {
3771                         idx = BTOBBT((__psint_t)&ophead->oh_len -
3772                                     (__psint_t)iclog->ic_datap);
3773                         if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3774                                 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3775                                 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3776                                 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
3777                         } else {
3778                                 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
3779                         }
3780                 }
3781                 ptr += sizeof(xlog_op_header_t) + op_len;
3782         }
3783 }       /* xlog_verify_iclog */
3784 #endif
3785 
3786 /*
3787  * Mark all iclogs IOERROR. l_icloglock is held by the caller.
3788  */
3789 STATIC int
3790 xlog_state_ioerror(
3791         struct xlog     *log)
3792 {
3793         xlog_in_core_t  *iclog, *ic;
3794 
3795         iclog = log->l_iclog;
3796         if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3797                 /*
3798                  * Mark all the incore logs IOERROR.
3799                  * From now on, no log flushes will result.
3800                  */
3801                 ic = iclog;
3802                 do {
3803                         ic->ic_state = XLOG_STATE_IOERROR;
3804                         ic = ic->ic_next;
3805                 } while (ic != iclog);
3806                 return 0;
3807         }
3808         /*
3809          * Return non-zero, if state transition has already happened.
3810          */
3811         return 1;
3812 }
3813 
3814 /*
3815  * This is called from xfs_force_shutdown, when we're forcibly
3816  * shutting down the filesystem, typically because of an IO error.
3817  * Our main objectives here are to make sure that:
3818  *      a. the filesystem gets marked 'SHUTDOWN' for all interested
3819  *         parties to find out, 'atomically'.
3820  *      b. those who're sleeping on log reservations, pinned objects and
3821  *          other resources get woken up, and be told the bad news.
3822  *      c. nothing new gets queued up after (a) and (b) are done.
3823  *      d. if !logerror, flush the iclogs to disk, then seal them off
3824  *         for business.
3825  *
3826  * Note: for delayed logging the !logerror case needs to flush the regions
3827  * held in memory out to the iclogs before flushing them to disk. This needs
3828  * to be done before the log is marked as shutdown, otherwise the flush to the
3829  * iclogs will fail.
3830  */
3831 int
3832 xfs_log_force_umount(
3833         struct xfs_mount        *mp,
3834         int                     logerror)
3835 {
3836         struct xlog     *log;
3837         int             retval;
3838 
3839         log = mp->m_log;
3840 
3841         /*
3842          * If this happens during log recovery, don't worry about
3843          * locking; the log isn't open for business yet.
3844          */
3845         if (!log ||
3846             log->l_flags & XLOG_ACTIVE_RECOVERY) {
3847                 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
3848                 if (mp->m_sb_bp)
3849                         XFS_BUF_DONE(mp->m_sb_bp);
3850                 return 0;
3851         }
3852 
3853         /*
3854          * Somebody could've already done the hard work for us.
3855          * No need to get locks for this.
3856          */
3857         if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3858                 ASSERT(XLOG_FORCED_SHUTDOWN(log));
3859                 return 1;
3860         }
3861         retval = 0;
3862 
3863         /*
3864          * Flush the in memory commit item list before marking the log as
3865          * being shut down. We need to do it in this order to ensure all the
3866          * completed transactions are flushed to disk with the xfs_log_force()
3867          * call below.
3868          */
3869         if (!logerror)
3870                 xlog_cil_force(log);
3871 
3872         /*
3873          * mark the filesystem and the as in a shutdown state and wake
3874          * everybody up to tell them the bad news.
3875          */
3876         spin_lock(&log->l_icloglock);
3877         mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
3878         if (mp->m_sb_bp)
3879                 XFS_BUF_DONE(mp->m_sb_bp);
3880 
3881         /*
3882          * This flag is sort of redundant because of the mount flag, but
3883          * it's good to maintain the separation between the log and the rest
3884          * of XFS.
3885          */
3886         log->l_flags |= XLOG_IO_ERROR;
3887 
3888         /*
3889          * If we hit a log error, we want to mark all the iclogs IOERROR
3890          * while we're still holding the loglock.
3891          */
3892         if (logerror)
3893                 retval = xlog_state_ioerror(log);
3894         spin_unlock(&log->l_icloglock);
3895 
3896         /*
3897          * We don't want anybody waiting for log reservations after this. That
3898          * means we have to wake up everybody queued up on reserveq as well as
3899          * writeq.  In addition, we make sure in xlog_{re}grant_log_space that
3900          * we don't enqueue anything once the SHUTDOWN flag is set, and this
3901          * action is protected by the grant locks.
3902          */
3903         xlog_grant_head_wake_all(&log->l_reserve_head);
3904         xlog_grant_head_wake_all(&log->l_write_head);
3905 
3906         if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
3907                 ASSERT(!logerror);
3908                 /*
3909                  * Force the incore logs to disk before shutting the
3910                  * log down completely.
3911                  */
3912                 _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
3913 
3914                 spin_lock(&log->l_icloglock);
3915                 retval = xlog_state_ioerror(log);
3916                 spin_unlock(&log->l_icloglock);
3917         }
3918         /*
3919          * Wake up everybody waiting on xfs_log_force.
3920          * Callback all log item committed functions as if the
3921          * log writes were completed.
3922          */
3923         xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3924 
3925 #ifdef XFSERRORDEBUG
3926         {
3927                 xlog_in_core_t  *iclog;
3928 
3929                 spin_lock(&log->l_icloglock);
3930                 iclog = log->l_iclog;
3931                 do {
3932                         ASSERT(iclog->ic_callback == 0);
3933                         iclog = iclog->ic_next;
3934                 } while (iclog != log->l_iclog);
3935                 spin_unlock(&log->l_icloglock);
3936         }
3937 #endif
3938         /* return non-zero if log IOERROR transition had already happened */
3939         return retval;
3940 }
3941 
3942 STATIC int
3943 xlog_iclogs_empty(
3944         struct xlog     *log)
3945 {
3946         xlog_in_core_t  *iclog;
3947 
3948         iclog = log->l_iclog;
3949         do {
3950                 /* endianness does not matter here, zero is zero in
3951                  * any language.
3952                  */
3953                 if (iclog->ic_header.h_num_logops)
3954                         return 0;
3955                 iclog = iclog->ic_next;
3956         } while (iclog != log->l_iclog);
3957         return 1;
3958 }
3959 
3960 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp