2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_log_priv.h"
32 #include "xfs_buf_item.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_log_recover.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_dir2_sf.h"
39 #include "xfs_attr_sf.h"
40 #include "xfs_dinode.h"
41 #include "xfs_inode.h"
44 kmem_zone_t
*xfs_log_ticket_zone
;
46 #define xlog_write_adv_cnt(ptr, len, off, bytes) \
51 /* Local miscellaneous function prototypes */
52 STATIC
int xlog_bdstrat_cb(struct xfs_buf
*);
53 STATIC
int xlog_commit_record(xfs_mount_t
*mp
, xlog_ticket_t
*ticket
,
54 xlog_in_core_t
**, xfs_lsn_t
*);
55 STATIC xlog_t
* xlog_alloc_log(xfs_mount_t
*mp
,
56 xfs_buftarg_t
*log_target
,
57 xfs_daddr_t blk_offset
,
59 STATIC
int xlog_space_left(xlog_t
*log
, int cycle
, int bytes
);
60 STATIC
int xlog_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
61 STATIC
void xlog_dealloc_log(xlog_t
*log
);
62 STATIC
int xlog_write(xfs_mount_t
*mp
, xfs_log_iovec_t region
[],
63 int nentries
, xfs_log_ticket_t tic
,
65 xlog_in_core_t
**commit_iclog
,
68 /* local state machine functions */
69 STATIC
void xlog_state_done_syncing(xlog_in_core_t
*iclog
, int);
70 STATIC
void xlog_state_do_callback(xlog_t
*log
,int aborted
, xlog_in_core_t
*iclog
);
71 STATIC
int xlog_state_get_iclog_space(xlog_t
*log
,
73 xlog_in_core_t
**iclog
,
74 xlog_ticket_t
*ticket
,
77 STATIC
int xlog_state_release_iclog(xlog_t
*log
,
78 xlog_in_core_t
*iclog
);
79 STATIC
void xlog_state_switch_iclogs(xlog_t
*log
,
80 xlog_in_core_t
*iclog
,
82 STATIC
int xlog_state_sync(xlog_t
*log
,
86 STATIC
int xlog_state_sync_all(xlog_t
*log
, uint flags
, int *log_flushed
);
87 STATIC
void xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
89 /* local functions to manipulate grant head */
90 STATIC
int xlog_grant_log_space(xlog_t
*log
,
92 STATIC
void xlog_grant_push_ail(xfs_mount_t
*mp
,
94 STATIC
void xlog_regrant_reserve_log_space(xlog_t
*log
,
95 xlog_ticket_t
*ticket
);
96 STATIC
int xlog_regrant_write_log_space(xlog_t
*log
,
97 xlog_ticket_t
*ticket
);
98 STATIC
void xlog_ungrant_log_space(xlog_t
*log
,
99 xlog_ticket_t
*ticket
);
102 /* local ticket functions */
103 STATIC xlog_ticket_t
*xlog_ticket_alloc(xlog_t
*log
,
110 STATIC
void xlog_verify_dest_ptr(xlog_t
*log
, __psint_t ptr
);
111 STATIC
void xlog_verify_grant_head(xlog_t
*log
, int equals
);
112 STATIC
void xlog_verify_iclog(xlog_t
*log
, xlog_in_core_t
*iclog
,
113 int count
, boolean_t syncing
);
114 STATIC
void xlog_verify_tail_lsn(xlog_t
*log
, xlog_in_core_t
*iclog
,
117 #define xlog_verify_dest_ptr(a,b)
118 #define xlog_verify_grant_head(a,b)
119 #define xlog_verify_iclog(a,b,c,d)
120 #define xlog_verify_tail_lsn(a,b,c)
123 STATIC
int xlog_iclogs_empty(xlog_t
*log
);
125 #if defined(XFS_LOG_TRACE)
127 #define XLOG_TRACE_LOGGRANT_SIZE 2048
128 #define XLOG_TRACE_ICLOG_SIZE 256
131 xlog_trace_loggrant_alloc(xlog_t
*log
)
133 log
->l_grant_trace
= ktrace_alloc(XLOG_TRACE_LOGGRANT_SIZE
, KM_NOFS
);
137 xlog_trace_loggrant_dealloc(xlog_t
*log
)
139 ktrace_free(log
->l_grant_trace
);
143 xlog_trace_loggrant(xlog_t
*log
, xlog_ticket_t
*tic
, xfs_caddr_t string
)
147 /* ticket counts are 1 byte each */
148 cnts
= ((unsigned long)tic
->t_ocnt
) | ((unsigned long)tic
->t_cnt
) << 8;
150 ktrace_enter(log
->l_grant_trace
,
152 (void *)log
->l_reserve_headq
,
153 (void *)log
->l_write_headq
,
154 (void *)((unsigned long)log
->l_grant_reserve_cycle
),
155 (void *)((unsigned long)log
->l_grant_reserve_bytes
),
156 (void *)((unsigned long)log
->l_grant_write_cycle
),
157 (void *)((unsigned long)log
->l_grant_write_bytes
),
158 (void *)((unsigned long)log
->l_curr_cycle
),
159 (void *)((unsigned long)log
->l_curr_block
),
160 (void *)((unsigned long)CYCLE_LSN(log
->l_tail_lsn
)),
161 (void *)((unsigned long)BLOCK_LSN(log
->l_tail_lsn
)),
163 (void *)((unsigned long)tic
->t_trans_type
),
165 (void *)((unsigned long)tic
->t_curr_res
),
166 (void *)((unsigned long)tic
->t_unit_res
));
170 xlog_trace_iclog_alloc(xlog_in_core_t
*iclog
)
172 iclog
->ic_trace
= ktrace_alloc(XLOG_TRACE_ICLOG_SIZE
, KM_NOFS
);
176 xlog_trace_iclog_dealloc(xlog_in_core_t
*iclog
)
178 ktrace_free(iclog
->ic_trace
);
182 xlog_trace_iclog(xlog_in_core_t
*iclog
, uint state
)
184 ktrace_enter(iclog
->ic_trace
,
185 (void *)((unsigned long)state
),
186 (void *)((unsigned long)current_pid()),
187 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
188 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
189 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
190 (void *)NULL
, (void *)NULL
);
194 #define xlog_trace_loggrant_alloc(log)
195 #define xlog_trace_loggrant_dealloc(log)
196 #define xlog_trace_loggrant(log,tic,string)
198 #define xlog_trace_iclog_alloc(iclog)
199 #define xlog_trace_iclog_dealloc(iclog)
200 #define xlog_trace_iclog(iclog,state)
202 #endif /* XFS_LOG_TRACE */
206 xlog_ins_ticketq(struct xlog_ticket
**qp
, struct xlog_ticket
*tic
)
210 tic
->t_prev
= (*qp
)->t_prev
;
211 (*qp
)->t_prev
->t_next
= tic
;
214 tic
->t_prev
= tic
->t_next
= tic
;
218 tic
->t_flags
|= XLOG_TIC_IN_Q
;
222 xlog_del_ticketq(struct xlog_ticket
**qp
, struct xlog_ticket
*tic
)
224 if (tic
== tic
->t_next
) {
228 tic
->t_next
->t_prev
= tic
->t_prev
;
229 tic
->t_prev
->t_next
= tic
->t_next
;
232 tic
->t_next
= tic
->t_prev
= NULL
;
233 tic
->t_flags
&= ~XLOG_TIC_IN_Q
;
237 xlog_grant_sub_space(struct log
*log
, int bytes
)
239 log
->l_grant_write_bytes
-= bytes
;
240 if (log
->l_grant_write_bytes
< 0) {
241 log
->l_grant_write_bytes
+= log
->l_logsize
;
242 log
->l_grant_write_cycle
--;
245 log
->l_grant_reserve_bytes
-= bytes
;
246 if ((log
)->l_grant_reserve_bytes
< 0) {
247 log
->l_grant_reserve_bytes
+= log
->l_logsize
;
248 log
->l_grant_reserve_cycle
--;
254 xlog_grant_add_space_write(struct log
*log
, int bytes
)
256 int tmp
= log
->l_logsize
- log
->l_grant_write_bytes
;
258 log
->l_grant_write_bytes
+= bytes
;
260 log
->l_grant_write_cycle
++;
261 log
->l_grant_write_bytes
= bytes
- tmp
;
266 xlog_grant_add_space_reserve(struct log
*log
, int bytes
)
268 int tmp
= log
->l_logsize
- log
->l_grant_reserve_bytes
;
270 log
->l_grant_reserve_bytes
+= bytes
;
272 log
->l_grant_reserve_cycle
++;
273 log
->l_grant_reserve_bytes
= bytes
- tmp
;
278 xlog_grant_add_space(struct log
*log
, int bytes
)
280 xlog_grant_add_space_write(log
, bytes
);
281 xlog_grant_add_space_reserve(log
, bytes
);
285 xlog_tic_reset_res(xlog_ticket_t
*tic
)
288 tic
->t_res_arr_sum
= 0;
289 tic
->t_res_num_ophdrs
= 0;
293 xlog_tic_add_region(xlog_ticket_t
*tic
, uint len
, uint type
)
295 if (tic
->t_res_num
== XLOG_TIC_LEN_MAX
) {
296 /* add to overflow and start again */
297 tic
->t_res_o_flow
+= tic
->t_res_arr_sum
;
299 tic
->t_res_arr_sum
= 0;
302 tic
->t_res_arr
[tic
->t_res_num
].r_len
= len
;
303 tic
->t_res_arr
[tic
->t_res_num
].r_type
= type
;
304 tic
->t_res_arr_sum
+= len
;
311 * 1. currblock field gets updated at startup and after in-core logs
312 * marked as with WANT_SYNC.
316 * This routine is called when a user of a log manager ticket is done with
317 * the reservation. If the ticket was ever used, then a commit record for
318 * the associated transaction is written out as a log operation header with
319 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
320 * a given ticket. If the ticket was one with a permanent reservation, then
321 * a few operations are done differently. Permanent reservation tickets by
322 * default don't release the reservation. They just commit the current
323 * transaction with the belief that the reservation is still needed. A flag
324 * must be passed in before permanent reservations are actually released.
325 * When these type of tickets are not released, they need to be set into
326 * the inited state again. By doing this, a start record will be written
327 * out when the next write occurs.
330 xfs_log_done(xfs_mount_t
*mp
,
331 xfs_log_ticket_t xtic
,
335 xlog_t
*log
= mp
->m_log
;
336 xlog_ticket_t
*ticket
= (xfs_log_ticket_t
) xtic
;
339 if (XLOG_FORCED_SHUTDOWN(log
) ||
341 * If nothing was ever written, don't write out commit record.
342 * If we get an error, just continue and give back the log ticket.
344 (((ticket
->t_flags
& XLOG_TIC_INITED
) == 0) &&
345 (xlog_commit_record(mp
, ticket
,
346 (xlog_in_core_t
**)iclog
, &lsn
)))) {
347 lsn
= (xfs_lsn_t
) -1;
348 if (ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) {
349 flags
|= XFS_LOG_REL_PERM_RESERV
;
354 if ((ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) == 0 ||
355 (flags
& XFS_LOG_REL_PERM_RESERV
)) {
357 * Release ticket if not permanent reservation or a specific
358 * request has been made to release a permanent reservation.
360 xlog_trace_loggrant(log
, ticket
, "xfs_log_done: (non-permanent)");
361 xlog_ungrant_log_space(log
, ticket
);
362 xfs_log_ticket_put(ticket
);
364 xlog_trace_loggrant(log
, ticket
, "xfs_log_done: (permanent)");
365 xlog_regrant_reserve_log_space(log
, ticket
);
366 /* If this ticket was a permanent reservation and we aren't
367 * trying to release it, reset the inited flags; so next time
368 * we write, a start record will be written out.
370 ticket
->t_flags
|= XLOG_TIC_INITED
;
378 * Force the in-core log to disk. If flags == XFS_LOG_SYNC,
379 * the force is done synchronously.
381 * Asynchronous forces are implemented by setting the WANT_SYNC
382 * bit in the appropriate in-core log and then returning.
384 * Synchronous forces are implemented with a signal variable. All callers
385 * to force a given lsn to disk will wait on a the sv attached to the
386 * specific in-core log. When given in-core log finally completes its
387 * write to disk, that thread will wake up all threads waiting on the
397 xlog_t
*log
= mp
->m_log
;
401 log_flushed
= &dummy
;
403 ASSERT(flags
& XFS_LOG_FORCE
);
405 XFS_STATS_INC(xs_log_force
);
407 if (log
->l_flags
& XLOG_IO_ERROR
)
408 return XFS_ERROR(EIO
);
410 return xlog_state_sync_all(log
, flags
, log_flushed
);
412 return xlog_state_sync(log
, lsn
, flags
, log_flushed
);
413 } /* _xfs_log_force */
416 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
417 * about errors or whether the log was flushed or not. This is the normal
418 * interface to use when trying to unpin items or move the log forward.
427 error
= _xfs_log_force(mp
, lsn
, flags
, NULL
);
429 xfs_fs_cmn_err(CE_WARN
, mp
, "xfs_log_force: "
430 "error %d returned.", error
);
436 * Attaches a new iclog I/O completion callback routine during
437 * transaction commit. If the log is in error state, a non-zero
438 * return code is handed back and the caller is responsible for
439 * executing the callback at an appropriate time.
442 xfs_log_notify(xfs_mount_t
*mp
, /* mount of partition */
443 void *iclog_hndl
, /* iclog to hang callback off */
444 xfs_log_callback_t
*cb
)
446 xlog_in_core_t
*iclog
= (xlog_in_core_t
*)iclog_hndl
;
449 spin_lock(&iclog
->ic_callback_lock
);
450 abortflg
= (iclog
->ic_state
& XLOG_STATE_IOERROR
);
452 ASSERT_ALWAYS((iclog
->ic_state
== XLOG_STATE_ACTIVE
) ||
453 (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
));
455 *(iclog
->ic_callback_tail
) = cb
;
456 iclog
->ic_callback_tail
= &(cb
->cb_next
);
458 spin_unlock(&iclog
->ic_callback_lock
);
460 } /* xfs_log_notify */
463 xfs_log_release_iclog(xfs_mount_t
*mp
,
466 xlog_t
*log
= mp
->m_log
;
467 xlog_in_core_t
*iclog
= (xlog_in_core_t
*)iclog_hndl
;
469 if (xlog_state_release_iclog(log
, iclog
)) {
470 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
478 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
479 * to the reservation.
480 * 2. Potentially, push buffers at tail of log to disk.
482 * Each reservation is going to reserve extra space for a log record header.
483 * When writes happen to the on-disk log, we don't subtract the length of the
484 * log record header from any reservation. By wasting space in each
485 * reservation, we prevent over allocation problems.
488 xfs_log_reserve(xfs_mount_t
*mp
,
491 xfs_log_ticket_t
*ticket
,
496 xlog_t
*log
= mp
->m_log
;
497 xlog_ticket_t
*internal_ticket
;
500 ASSERT(client
== XFS_TRANSACTION
|| client
== XFS_LOG
);
501 ASSERT((flags
& XFS_LOG_NOSLEEP
) == 0);
503 if (XLOG_FORCED_SHUTDOWN(log
))
504 return XFS_ERROR(EIO
);
506 XFS_STATS_INC(xs_try_logspace
);
508 if (*ticket
!= NULL
) {
509 ASSERT(flags
& XFS_LOG_PERM_RESERV
);
510 internal_ticket
= (xlog_ticket_t
*)*ticket
;
511 xlog_trace_loggrant(log
, internal_ticket
, "xfs_log_reserve: existing ticket (permanent trans)");
512 xlog_grant_push_ail(mp
, internal_ticket
->t_unit_res
);
513 retval
= xlog_regrant_write_log_space(log
, internal_ticket
);
515 /* may sleep if need to allocate more tickets */
516 internal_ticket
= xlog_ticket_alloc(log
, unit_bytes
, cnt
,
518 if (!internal_ticket
)
519 return XFS_ERROR(ENOMEM
);
520 internal_ticket
->t_trans_type
= t_type
;
521 *ticket
= internal_ticket
;
522 xlog_trace_loggrant(log
, internal_ticket
,
523 (internal_ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) ?
524 "xfs_log_reserve: create new ticket (permanent trans)" :
525 "xfs_log_reserve: create new ticket");
526 xlog_grant_push_ail(mp
,
527 (internal_ticket
->t_unit_res
*
528 internal_ticket
->t_cnt
));
529 retval
= xlog_grant_log_space(log
, internal_ticket
);
533 } /* xfs_log_reserve */
537 * Mount a log filesystem
539 * mp - ubiquitous xfs mount point structure
540 * log_target - buftarg of on-disk log device
541 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
542 * num_bblocks - Number of BBSIZE blocks in on-disk log
544 * Return error or zero.
549 xfs_buftarg_t
*log_target
,
550 xfs_daddr_t blk_offset
,
555 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
556 cmn_err(CE_NOTE
, "XFS mounting filesystem %s", mp
->m_fsname
);
559 "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.",
561 ASSERT(mp
->m_flags
& XFS_MOUNT_RDONLY
);
564 mp
->m_log
= xlog_alloc_log(mp
, log_target
, blk_offset
, num_bblks
);
565 if (IS_ERR(mp
->m_log
)) {
566 error
= -PTR_ERR(mp
->m_log
);
571 * Initialize the AIL now we have a log.
573 error
= xfs_trans_ail_init(mp
);
575 cmn_err(CE_WARN
, "XFS: AIL initialisation failed: error %d", error
);
578 mp
->m_log
->l_ailp
= mp
->m_ail
;
581 * skip log recovery on a norecovery mount. pretend it all
584 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
)) {
585 int readonly
= (mp
->m_flags
& XFS_MOUNT_RDONLY
);
588 mp
->m_flags
&= ~XFS_MOUNT_RDONLY
;
590 error
= xlog_recover(mp
->m_log
);
593 mp
->m_flags
|= XFS_MOUNT_RDONLY
;
595 cmn_err(CE_WARN
, "XFS: log mount/recovery failed: error %d", error
);
596 goto out_destroy_ail
;
600 /* Normal transactions can now occur */
601 mp
->m_log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
606 xfs_trans_ail_destroy(mp
);
608 xlog_dealloc_log(mp
->m_log
);
614 * Finish the recovery of the file system. This is separate from
615 * the xfs_log_mount() call, because it depends on the code in
616 * xfs_mountfs() to read in the root and real-time bitmap inodes
617 * between calling xfs_log_mount() and here.
619 * mp - ubiquitous xfs mount point structure
622 xfs_log_mount_finish(xfs_mount_t
*mp
)
626 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
627 error
= xlog_recover_finish(mp
->m_log
);
630 ASSERT(mp
->m_flags
& XFS_MOUNT_RDONLY
);
637 * Final log writes as part of unmount.
639 * Mark the filesystem clean as unmount happens. Note that during relocation
640 * this routine needs to be executed as part of source-bag while the
641 * deallocation must not be done until source-end.
645 * Unmount record used to have a string "Unmount filesystem--" in the
646 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
647 * We just write the magic number now since that particular field isn't
648 * currently architecture converted and "nUmount" is a bit foo.
649 * As far as I know, there weren't any dependencies on the old behaviour.
653 xfs_log_unmount_write(xfs_mount_t
*mp
)
655 xlog_t
*log
= mp
->m_log
;
656 xlog_in_core_t
*iclog
;
658 xlog_in_core_t
*first_iclog
;
660 xfs_log_iovec_t reg
[1];
661 xfs_log_ticket_t tic
= NULL
;
665 /* the data section must be 32 bit size aligned */
669 __uint32_t pad2
; /* may as well make it 64 bits */
670 } magic
= { XLOG_UNMOUNT_TYPE
, 0, 0 };
673 * Don't write out unmount record on read-only mounts.
674 * Or, if we are doing a forced umount (typically because of IO errors).
676 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
679 error
= _xfs_log_force(mp
, 0, XFS_LOG_FORCE
|XFS_LOG_SYNC
, NULL
);
680 ASSERT(error
|| !(XLOG_FORCED_SHUTDOWN(log
)));
683 first_iclog
= iclog
= log
->l_iclog
;
685 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
686 ASSERT(iclog
->ic_state
& XLOG_STATE_ACTIVE
);
687 ASSERT(iclog
->ic_offset
== 0);
689 iclog
= iclog
->ic_next
;
690 } while (iclog
!= first_iclog
);
692 if (! (XLOG_FORCED_SHUTDOWN(log
))) {
693 reg
[0].i_addr
= (void*)&magic
;
694 reg
[0].i_len
= sizeof(magic
);
695 XLOG_VEC_SET_TYPE(®
[0], XLOG_REG_TYPE_UNMOUNT
);
697 error
= xfs_log_reserve(mp
, 600, 1, &tic
,
698 XFS_LOG
, 0, XLOG_UNMOUNT_REC_TYPE
);
700 /* remove inited flag */
701 ((xlog_ticket_t
*)tic
)->t_flags
= 0;
702 error
= xlog_write(mp
, reg
, 1, tic
, &lsn
,
703 NULL
, XLOG_UNMOUNT_TRANS
);
705 * At this point, we're umounting anyway,
706 * so there's no point in transitioning log state
707 * to IOERROR. Just continue...
712 xfs_fs_cmn_err(CE_ALERT
, mp
,
713 "xfs_log_unmount: unmount record failed");
717 spin_lock(&log
->l_icloglock
);
718 iclog
= log
->l_iclog
;
719 atomic_inc(&iclog
->ic_refcnt
);
720 xlog_state_want_sync(log
, iclog
);
721 spin_unlock(&log
->l_icloglock
);
722 error
= xlog_state_release_iclog(log
, iclog
);
724 spin_lock(&log
->l_icloglock
);
725 if (!(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
726 iclog
->ic_state
== XLOG_STATE_DIRTY
)) {
727 if (!XLOG_FORCED_SHUTDOWN(log
)) {
728 sv_wait(&iclog
->ic_force_wait
, PMEM
,
729 &log
->l_icloglock
, s
);
731 spin_unlock(&log
->l_icloglock
);
734 spin_unlock(&log
->l_icloglock
);
737 xlog_trace_loggrant(log
, tic
, "unmount rec");
738 xlog_ungrant_log_space(log
, tic
);
739 xfs_log_ticket_put(tic
);
743 * We're already in forced_shutdown mode, couldn't
744 * even attempt to write out the unmount transaction.
746 * Go through the motions of sync'ing and releasing
747 * the iclog, even though no I/O will actually happen,
748 * we need to wait for other log I/Os that may already
749 * be in progress. Do this as a separate section of
750 * code so we'll know if we ever get stuck here that
751 * we're in this odd situation of trying to unmount
752 * a file system that went into forced_shutdown as
753 * the result of an unmount..
755 spin_lock(&log
->l_icloglock
);
756 iclog
= log
->l_iclog
;
757 atomic_inc(&iclog
->ic_refcnt
);
759 xlog_state_want_sync(log
, iclog
);
760 spin_unlock(&log
->l_icloglock
);
761 error
= xlog_state_release_iclog(log
, iclog
);
763 spin_lock(&log
->l_icloglock
);
765 if ( ! ( iclog
->ic_state
== XLOG_STATE_ACTIVE
766 || iclog
->ic_state
== XLOG_STATE_DIRTY
767 || iclog
->ic_state
== XLOG_STATE_IOERROR
) ) {
769 sv_wait(&iclog
->ic_force_wait
, PMEM
,
770 &log
->l_icloglock
, s
);
772 spin_unlock(&log
->l_icloglock
);
777 } /* xfs_log_unmount_write */
780 * Deallocate log structures for unmount/relocation.
782 * We need to stop the aild from running before we destroy
783 * and deallocate the log as the aild references the log.
786 xfs_log_unmount(xfs_mount_t
*mp
)
788 xfs_trans_ail_destroy(mp
);
789 xlog_dealloc_log(mp
->m_log
);
793 * Write region vectors to log. The write happens using the space reservation
794 * of the ticket (tic). It is not a requirement that all writes for a given
795 * transaction occur with one call to xfs_log_write().
798 xfs_log_write(xfs_mount_t
* mp
,
799 xfs_log_iovec_t reg
[],
801 xfs_log_ticket_t tic
,
802 xfs_lsn_t
*start_lsn
)
805 xlog_t
*log
= mp
->m_log
;
807 if (XLOG_FORCED_SHUTDOWN(log
))
808 return XFS_ERROR(EIO
);
810 if ((error
= xlog_write(mp
, reg
, nentries
, tic
, start_lsn
, NULL
, 0))) {
811 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
814 } /* xfs_log_write */
818 xfs_log_move_tail(xfs_mount_t
*mp
,
822 xlog_t
*log
= mp
->m_log
;
823 int need_bytes
, free_bytes
, cycle
, bytes
;
825 if (XLOG_FORCED_SHUTDOWN(log
))
829 /* needed since sync_lsn is 64 bits */
830 spin_lock(&log
->l_icloglock
);
831 tail_lsn
= log
->l_last_sync_lsn
;
832 spin_unlock(&log
->l_icloglock
);
835 spin_lock(&log
->l_grant_lock
);
837 /* Also an invalid lsn. 1 implies that we aren't passing in a valid
841 log
->l_tail_lsn
= tail_lsn
;
844 if ((tic
= log
->l_write_headq
)) {
846 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
847 panic("Recovery problem");
849 cycle
= log
->l_grant_write_cycle
;
850 bytes
= log
->l_grant_write_bytes
;
851 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
853 ASSERT(tic
->t_flags
& XLOG_TIC_PERM_RESERV
);
855 if (free_bytes
< tic
->t_unit_res
&& tail_lsn
!= 1)
858 free_bytes
-= tic
->t_unit_res
;
859 sv_signal(&tic
->t_wait
);
861 } while (tic
!= log
->l_write_headq
);
863 if ((tic
= log
->l_reserve_headq
)) {
865 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
866 panic("Recovery problem");
868 cycle
= log
->l_grant_reserve_cycle
;
869 bytes
= log
->l_grant_reserve_bytes
;
870 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
872 if (tic
->t_flags
& XLOG_TIC_PERM_RESERV
)
873 need_bytes
= tic
->t_unit_res
*tic
->t_cnt
;
875 need_bytes
= tic
->t_unit_res
;
876 if (free_bytes
< need_bytes
&& tail_lsn
!= 1)
879 free_bytes
-= need_bytes
;
880 sv_signal(&tic
->t_wait
);
882 } while (tic
!= log
->l_reserve_headq
);
884 spin_unlock(&log
->l_grant_lock
);
885 } /* xfs_log_move_tail */
888 * Determine if we have a transaction that has gone to disk
889 * that needs to be covered. Log activity needs to be idle (no AIL and
890 * nothing in the iclogs). And, we need to be in the right state indicating
891 * something has gone out.
894 xfs_log_need_covered(xfs_mount_t
*mp
)
897 xlog_t
*log
= mp
->m_log
;
899 if (!xfs_fs_writable(mp
))
902 spin_lock(&log
->l_icloglock
);
903 if (((log
->l_covered_state
== XLOG_STATE_COVER_NEED
) ||
904 (log
->l_covered_state
== XLOG_STATE_COVER_NEED2
))
905 && !xfs_trans_ail_tail(log
->l_ailp
)
906 && xlog_iclogs_empty(log
)) {
907 if (log
->l_covered_state
== XLOG_STATE_COVER_NEED
)
908 log
->l_covered_state
= XLOG_STATE_COVER_DONE
;
910 ASSERT(log
->l_covered_state
== XLOG_STATE_COVER_NEED2
);
911 log
->l_covered_state
= XLOG_STATE_COVER_DONE2
;
915 spin_unlock(&log
->l_icloglock
);
919 /******************************************************************************
923 ******************************************************************************
926 /* xfs_trans_tail_ail returns 0 when there is nothing in the list.
927 * The log manager must keep track of the last LR which was committed
928 * to disk. The lsn of this LR will become the new tail_lsn whenever
929 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
930 * the situation where stuff could be written into the log but nothing
931 * was ever in the AIL when asked. Eventually, we panic since the
932 * tail hits the head.
934 * We may be holding the log iclog lock upon entering this routine.
937 xlog_assign_tail_lsn(xfs_mount_t
*mp
)
940 xlog_t
*log
= mp
->m_log
;
942 tail_lsn
= xfs_trans_ail_tail(mp
->m_ail
);
943 spin_lock(&log
->l_grant_lock
);
945 log
->l_tail_lsn
= tail_lsn
;
947 tail_lsn
= log
->l_tail_lsn
= log
->l_last_sync_lsn
;
949 spin_unlock(&log
->l_grant_lock
);
952 } /* xlog_assign_tail_lsn */
956 * Return the space in the log between the tail and the head. The head
957 * is passed in the cycle/bytes formal parms. In the special case where
958 * the reserve head has wrapped passed the tail, this calculation is no
959 * longer valid. In this case, just return 0 which means there is no space
960 * in the log. This works for all places where this function is called
961 * with the reserve head. Of course, if the write head were to ever
962 * wrap the tail, we should blow up. Rather than catch this case here,
963 * we depend on other ASSERTions in other parts of the code. XXXmiken
965 * This code also handles the case where the reservation head is behind
966 * the tail. The details of this case are described below, but the end
967 * result is that we return the size of the log as the amount of space left.
970 xlog_space_left(xlog_t
*log
, int cycle
, int bytes
)
976 tail_bytes
= BBTOB(BLOCK_LSN(log
->l_tail_lsn
));
977 tail_cycle
= CYCLE_LSN(log
->l_tail_lsn
);
978 if ((tail_cycle
== cycle
) && (bytes
>= tail_bytes
)) {
979 free_bytes
= log
->l_logsize
- (bytes
- tail_bytes
);
980 } else if ((tail_cycle
+ 1) < cycle
) {
982 } else if (tail_cycle
< cycle
) {
983 ASSERT(tail_cycle
== (cycle
- 1));
984 free_bytes
= tail_bytes
- bytes
;
987 * The reservation head is behind the tail.
988 * In this case we just want to return the size of the
989 * log as the amount of space left.
991 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
992 "xlog_space_left: head behind tail\n"
993 " tail_cycle = %d, tail_bytes = %d\n"
994 " GH cycle = %d, GH bytes = %d",
995 tail_cycle
, tail_bytes
, cycle
, bytes
);
997 free_bytes
= log
->l_logsize
;
1000 } /* xlog_space_left */
1004 * Log function which is called when an io completes.
1006 * The log manager needs its own routine, in order to control what
1007 * happens with the buffer after the write completes.
1010 xlog_iodone(xfs_buf_t
*bp
)
1012 xlog_in_core_t
*iclog
;
1016 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
1017 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long) 2);
1018 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1023 * If the _XFS_BARRIER_FAILED flag was set by a lower
1024 * layer, it means the underlying device no longer supports
1025 * barrier I/O. Warn loudly and turn off barriers.
1027 if (bp
->b_flags
& _XFS_BARRIER_FAILED
) {
1028 bp
->b_flags
&= ~_XFS_BARRIER_FAILED
;
1029 l
->l_mp
->m_flags
&= ~XFS_MOUNT_BARRIER
;
1030 xfs_fs_cmn_err(CE_WARN
, l
->l_mp
,
1031 "xlog_iodone: Barriers are no longer supported"
1032 " by device. Disabling barriers\n");
1033 xfs_buftrace("XLOG_IODONE BARRIERS OFF", bp
);
1037 * Race to shutdown the filesystem if we see an error.
1039 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp
)), l
->l_mp
,
1040 XFS_ERRTAG_IODONE_IOERR
, XFS_RANDOM_IODONE_IOERR
)) {
1041 xfs_ioerror_alert("xlog_iodone", l
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
1043 xfs_force_shutdown(l
->l_mp
, SHUTDOWN_LOG_IO_ERROR
);
1045 * This flag will be propagated to the trans-committed
1046 * callback routines to let them know that the log-commit
1049 aborted
= XFS_LI_ABORTED
;
1050 } else if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
1051 aborted
= XFS_LI_ABORTED
;
1054 /* log I/O is always issued ASYNC */
1055 ASSERT(XFS_BUF_ISASYNC(bp
));
1056 xlog_state_done_syncing(iclog
, aborted
);
1058 * do not reference the buffer (bp) here as we could race
1059 * with it being freed after writing the unmount record to the
1066 * The bdstrat callback function for log bufs. This gives us a central
1067 * place to trap bufs in case we get hit by a log I/O error and need to
1068 * shutdown. Actually, in practice, even when we didn't get a log error,
1069 * we transition the iclogs to IOERROR state *after* flushing all existing
1070 * iclogs to disk. This is because we don't want anymore new transactions to be
1071 * started or completed afterwards.
1074 xlog_bdstrat_cb(struct xfs_buf
*bp
)
1076 xlog_in_core_t
*iclog
;
1078 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
1080 if ((iclog
->ic_state
& XLOG_STATE_IOERROR
) == 0) {
1081 /* note for irix bstrat will need struct bdevsw passed
1082 * Fix the following macro if the code ever is merged
1088 xfs_buftrace("XLOG__BDSTRAT IOERROR", bp
);
1089 XFS_BUF_ERROR(bp
, EIO
);
1092 return XFS_ERROR(EIO
);
1098 * Return size of each in-core log record buffer.
1100 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
1102 * If the filesystem blocksize is too large, we may need to choose a
1103 * larger size since the directory code currently logs entire blocks.
1107 xlog_get_iclog_buffer_size(xfs_mount_t
*mp
,
1113 if (mp
->m_logbufs
<= 0)
1114 log
->l_iclog_bufs
= XLOG_MAX_ICLOGS
;
1116 log
->l_iclog_bufs
= mp
->m_logbufs
;
1119 * Buffer size passed in from mount system call.
1121 if (mp
->m_logbsize
> 0) {
1122 size
= log
->l_iclog_size
= mp
->m_logbsize
;
1123 log
->l_iclog_size_log
= 0;
1125 log
->l_iclog_size_log
++;
1129 if (xfs_sb_version_haslogv2(&mp
->m_sb
)) {
1130 /* # headers = size / 32k
1131 * one header holds cycles from 32k of data
1134 xhdrs
= mp
->m_logbsize
/ XLOG_HEADER_CYCLE_SIZE
;
1135 if (mp
->m_logbsize
% XLOG_HEADER_CYCLE_SIZE
)
1137 log
->l_iclog_hsize
= xhdrs
<< BBSHIFT
;
1138 log
->l_iclog_heads
= xhdrs
;
1140 ASSERT(mp
->m_logbsize
<= XLOG_BIG_RECORD_BSIZE
);
1141 log
->l_iclog_hsize
= BBSIZE
;
1142 log
->l_iclog_heads
= 1;
1147 /* All machines use 32kB buffers by default. */
1148 log
->l_iclog_size
= XLOG_BIG_RECORD_BSIZE
;
1149 log
->l_iclog_size_log
= XLOG_BIG_RECORD_BSHIFT
;
1151 /* the default log size is 16k or 32k which is one header sector */
1152 log
->l_iclog_hsize
= BBSIZE
;
1153 log
->l_iclog_heads
= 1;
1156 /* are we being asked to make the sizes selected above visible? */
1157 if (mp
->m_logbufs
== 0)
1158 mp
->m_logbufs
= log
->l_iclog_bufs
;
1159 if (mp
->m_logbsize
== 0)
1160 mp
->m_logbsize
= log
->l_iclog_size
;
1161 } /* xlog_get_iclog_buffer_size */
1165 * This routine initializes some of the log structure for a given mount point.
1166 * Its primary purpose is to fill in enough, so recovery can occur. However,
1167 * some other stuff may be filled in too.
1170 xlog_alloc_log(xfs_mount_t
*mp
,
1171 xfs_buftarg_t
*log_target
,
1172 xfs_daddr_t blk_offset
,
1176 xlog_rec_header_t
*head
;
1177 xlog_in_core_t
**iclogp
;
1178 xlog_in_core_t
*iclog
, *prev_iclog
=NULL
;
1184 log
= kmem_zalloc(sizeof(xlog_t
), KM_MAYFAIL
);
1186 xlog_warn("XFS: Log allocation failed: No memory!");
1191 log
->l_targ
= log_target
;
1192 log
->l_logsize
= BBTOB(num_bblks
);
1193 log
->l_logBBstart
= blk_offset
;
1194 log
->l_logBBsize
= num_bblks
;
1195 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
1196 log
->l_flags
|= XLOG_ACTIVE_RECOVERY
;
1198 log
->l_prev_block
= -1;
1199 log
->l_tail_lsn
= xlog_assign_lsn(1, 0);
1200 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1201 log
->l_last_sync_lsn
= log
->l_tail_lsn
;
1202 log
->l_curr_cycle
= 1; /* 0 is bad since this is initial value */
1203 log
->l_grant_reserve_cycle
= 1;
1204 log
->l_grant_write_cycle
= 1;
1206 error
= EFSCORRUPTED
;
1207 if (xfs_sb_version_hassector(&mp
->m_sb
)) {
1208 log
->l_sectbb_log
= mp
->m_sb
.sb_logsectlog
- BBSHIFT
;
1209 if (log
->l_sectbb_log
< 0 ||
1210 log
->l_sectbb_log
> mp
->m_sectbb_log
) {
1211 xlog_warn("XFS: Log sector size (0x%x) out of range.",
1216 /* for larger sector sizes, must have v2 or external log */
1217 if (log
->l_sectbb_log
!= 0 &&
1218 (log
->l_logBBstart
!= 0 &&
1219 !xfs_sb_version_haslogv2(&mp
->m_sb
))) {
1220 xlog_warn("XFS: log sector size (0x%x) invalid "
1221 "for configuration.", log
->l_sectbb_log
);
1224 if (mp
->m_sb
.sb_logsectlog
< BBSHIFT
) {
1225 xlog_warn("XFS: Log sector log (0x%x) too small.",
1226 mp
->m_sb
.sb_logsectlog
);
1230 log
->l_sectbb_mask
= (1 << log
->l_sectbb_log
) - 1;
1232 xlog_get_iclog_buffer_size(mp
, log
);
1235 bp
= xfs_buf_get_empty(log
->l_iclog_size
, mp
->m_logdev_targp
);
1238 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1239 XFS_BUF_SET_BDSTRAT_FUNC(bp
, xlog_bdstrat_cb
);
1240 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1241 ASSERT(XFS_BUF_ISBUSY(bp
));
1242 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
1245 spin_lock_init(&log
->l_icloglock
);
1246 spin_lock_init(&log
->l_grant_lock
);
1247 sv_init(&log
->l_flush_wait
, 0, "flush_wait");
1249 xlog_trace_loggrant_alloc(log
);
1250 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1251 ASSERT((XFS_BUF_SIZE(bp
) & BBMASK
) == 0);
1253 iclogp
= &log
->l_iclog
;
1255 * The amount of memory to allocate for the iclog structure is
1256 * rather funky due to the way the structure is defined. It is
1257 * done this way so that we can use different sizes for machines
1258 * with different amounts of memory. See the definition of
1259 * xlog_in_core_t in xfs_log_priv.h for details.
1261 iclogsize
= log
->l_iclog_size
;
1262 ASSERT(log
->l_iclog_size
>= 4096);
1263 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
1264 *iclogp
= kmem_zalloc(sizeof(xlog_in_core_t
), KM_MAYFAIL
);
1266 goto out_free_iclog
;
1269 iclog
->ic_prev
= prev_iclog
;
1272 bp
= xfs_buf_get_noaddr(log
->l_iclog_size
, mp
->m_logdev_targp
);
1274 goto out_free_iclog
;
1275 if (!XFS_BUF_CPSEMA(bp
))
1277 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1278 XFS_BUF_SET_BDSTRAT_FUNC(bp
, xlog_bdstrat_cb
);
1279 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1281 iclog
->ic_data
= bp
->b_addr
;
1283 log
->l_iclog_bak
[i
] = (xfs_caddr_t
)&(iclog
->ic_header
);
1285 head
= &iclog
->ic_header
;
1286 memset(head
, 0, sizeof(xlog_rec_header_t
));
1287 head
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1288 head
->h_version
= cpu_to_be32(
1289 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1290 head
->h_size
= cpu_to_be32(log
->l_iclog_size
);
1292 head
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1293 memcpy(&head
->h_fs_uuid
, &mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1295 iclog
->ic_size
= XFS_BUF_SIZE(bp
) - log
->l_iclog_hsize
;
1296 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
1297 iclog
->ic_log
= log
;
1298 atomic_set(&iclog
->ic_refcnt
, 0);
1299 spin_lock_init(&iclog
->ic_callback_lock
);
1300 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
1301 iclog
->ic_datap
= (char *)iclog
->ic_data
+ log
->l_iclog_hsize
;
1303 ASSERT(XFS_BUF_ISBUSY(iclog
->ic_bp
));
1304 ASSERT(XFS_BUF_VALUSEMA(iclog
->ic_bp
) <= 0);
1305 sv_init(&iclog
->ic_force_wait
, SV_DEFAULT
, "iclog-force");
1306 sv_init(&iclog
->ic_write_wait
, SV_DEFAULT
, "iclog-write");
1308 xlog_trace_iclog_alloc(iclog
);
1310 iclogp
= &iclog
->ic_next
;
1312 *iclogp
= log
->l_iclog
; /* complete ring */
1313 log
->l_iclog
->ic_prev
= prev_iclog
; /* re-write 1st prev ptr */
1318 for (iclog
= log
->l_iclog
; iclog
; iclog
= prev_iclog
) {
1319 prev_iclog
= iclog
->ic_next
;
1321 sv_destroy(&iclog
->ic_force_wait
);
1322 sv_destroy(&iclog
->ic_write_wait
);
1323 xfs_buf_free(iclog
->ic_bp
);
1324 xlog_trace_iclog_dealloc(iclog
);
1328 spinlock_destroy(&log
->l_icloglock
);
1329 spinlock_destroy(&log
->l_grant_lock
);
1330 xlog_trace_loggrant_dealloc(log
);
1331 xfs_buf_free(log
->l_xbuf
);
1335 return ERR_PTR(-error
);
1336 } /* xlog_alloc_log */
1340 * Write out the commit record of a transaction associated with the given
1341 * ticket. Return the lsn of the commit record.
1344 xlog_commit_record(xfs_mount_t
*mp
,
1345 xlog_ticket_t
*ticket
,
1346 xlog_in_core_t
**iclog
,
1347 xfs_lsn_t
*commitlsnp
)
1350 xfs_log_iovec_t reg
[1];
1352 reg
[0].i_addr
= NULL
;
1354 XLOG_VEC_SET_TYPE(®
[0], XLOG_REG_TYPE_COMMIT
);
1356 ASSERT_ALWAYS(iclog
);
1357 if ((error
= xlog_write(mp
, reg
, 1, ticket
, commitlsnp
,
1358 iclog
, XLOG_COMMIT_TRANS
))) {
1359 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
1362 } /* xlog_commit_record */
1366 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1367 * log space. This code pushes on the lsn which would supposedly free up
1368 * the 25% which we want to leave free. We may need to adopt a policy which
1369 * pushes on an lsn which is further along in the log once we reach the high
1370 * water mark. In this manner, we would be creating a low water mark.
1373 xlog_grant_push_ail(xfs_mount_t
*mp
,
1376 xlog_t
*log
= mp
->m_log
; /* pointer to the log */
1377 xfs_lsn_t tail_lsn
; /* lsn of the log tail */
1378 xfs_lsn_t threshold_lsn
= 0; /* lsn we'd like to be at */
1379 int free_blocks
; /* free blocks left to write to */
1380 int free_bytes
; /* free bytes left to write to */
1381 int threshold_block
; /* block in lsn we'd like to be at */
1382 int threshold_cycle
; /* lsn cycle we'd like to be at */
1385 ASSERT(BTOBB(need_bytes
) < log
->l_logBBsize
);
1387 spin_lock(&log
->l_grant_lock
);
1388 free_bytes
= xlog_space_left(log
,
1389 log
->l_grant_reserve_cycle
,
1390 log
->l_grant_reserve_bytes
);
1391 tail_lsn
= log
->l_tail_lsn
;
1392 free_blocks
= BTOBBT(free_bytes
);
1395 * Set the threshold for the minimum number of free blocks in the
1396 * log to the maximum of what the caller needs, one quarter of the
1397 * log, and 256 blocks.
1399 free_threshold
= BTOBB(need_bytes
);
1400 free_threshold
= MAX(free_threshold
, (log
->l_logBBsize
>> 2));
1401 free_threshold
= MAX(free_threshold
, 256);
1402 if (free_blocks
< free_threshold
) {
1403 threshold_block
= BLOCK_LSN(tail_lsn
) + free_threshold
;
1404 threshold_cycle
= CYCLE_LSN(tail_lsn
);
1405 if (threshold_block
>= log
->l_logBBsize
) {
1406 threshold_block
-= log
->l_logBBsize
;
1407 threshold_cycle
+= 1;
1409 threshold_lsn
= xlog_assign_lsn(threshold_cycle
, threshold_block
);
1411 /* Don't pass in an lsn greater than the lsn of the last
1412 * log record known to be on disk.
1414 if (XFS_LSN_CMP(threshold_lsn
, log
->l_last_sync_lsn
) > 0)
1415 threshold_lsn
= log
->l_last_sync_lsn
;
1417 spin_unlock(&log
->l_grant_lock
);
1420 * Get the transaction layer to kick the dirty buffers out to
1421 * disk asynchronously. No point in trying to do this if
1422 * the filesystem is shutting down.
1424 if (threshold_lsn
&&
1425 !XLOG_FORCED_SHUTDOWN(log
))
1426 xfs_trans_ail_push(log
->l_ailp
, threshold_lsn
);
1427 } /* xlog_grant_push_ail */
1431 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1432 * fashion. Previously, we should have moved the current iclog
1433 * ptr in the log to point to the next available iclog. This allows further
1434 * write to continue while this code syncs out an iclog ready to go.
1435 * Before an in-core log can be written out, the data section must be scanned
1436 * to save away the 1st word of each BBSIZE block into the header. We replace
1437 * it with the current cycle count. Each BBSIZE block is tagged with the
1438 * cycle count because there in an implicit assumption that drives will
1439 * guarantee that entire 512 byte blocks get written at once. In other words,
1440 * we can't have part of a 512 byte block written and part not written. By
1441 * tagging each block, we will know which blocks are valid when recovering
1442 * after an unclean shutdown.
1444 * This routine is single threaded on the iclog. No other thread can be in
1445 * this routine with the same iclog. Changing contents of iclog can there-
1446 * fore be done without grabbing the state machine lock. Updating the global
1447 * log will require grabbing the lock though.
1449 * The entire log manager uses a logical block numbering scheme. Only
1450 * log_sync (and then only bwrite()) know about the fact that the log may
1451 * not start with block zero on a given device. The log block start offset
1452 * is added immediately before calling bwrite().
1456 xlog_sync(xlog_t
*log
,
1457 xlog_in_core_t
*iclog
)
1459 xfs_caddr_t dptr
; /* pointer to byte sized element */
1462 uint count
; /* byte count of bwrite */
1463 uint count_init
; /* initial count before roundup */
1464 int roundoff
; /* roundoff to BB or stripe */
1465 int split
= 0; /* split write into two regions */
1467 int v2
= xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
);
1469 XFS_STATS_INC(xs_log_writes
);
1470 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
1472 /* Add for LR header */
1473 count_init
= log
->l_iclog_hsize
+ iclog
->ic_offset
;
1475 /* Round out the log write size */
1476 if (v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1) {
1477 /* we have a v2 stripe unit to use */
1478 count
= XLOG_LSUNITTOB(log
, XLOG_BTOLSUNIT(log
, count_init
));
1480 count
= BBTOB(BTOBB(count_init
));
1482 roundoff
= count
- count_init
;
1483 ASSERT(roundoff
>= 0);
1484 ASSERT((v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1 &&
1485 roundoff
< log
->l_mp
->m_sb
.sb_logsunit
)
1487 (log
->l_mp
->m_sb
.sb_logsunit
<= 1 &&
1488 roundoff
< BBTOB(1)));
1490 /* move grant heads by roundoff in sync */
1491 spin_lock(&log
->l_grant_lock
);
1492 xlog_grant_add_space(log
, roundoff
);
1493 spin_unlock(&log
->l_grant_lock
);
1495 /* put cycle number in every block */
1496 xlog_pack_data(log
, iclog
, roundoff
);
1498 /* real byte length */
1500 iclog
->ic_header
.h_len
=
1501 cpu_to_be32(iclog
->ic_offset
+ roundoff
);
1503 iclog
->ic_header
.h_len
=
1504 cpu_to_be32(iclog
->ic_offset
);
1508 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long)1);
1509 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1510 XFS_BUF_SET_ADDR(bp
, BLOCK_LSN(be64_to_cpu(iclog
->ic_header
.h_lsn
)));
1512 XFS_STATS_ADD(xs_log_blocks
, BTOBB(count
));
1514 /* Do we need to split this write into 2 parts? */
1515 if (XFS_BUF_ADDR(bp
) + BTOBB(count
) > log
->l_logBBsize
) {
1516 split
= count
- (BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
)));
1517 count
= BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
));
1518 iclog
->ic_bwritecnt
= 2; /* split into 2 writes */
1520 iclog
->ic_bwritecnt
= 1;
1522 XFS_BUF_SET_COUNT(bp
, count
);
1523 XFS_BUF_SET_FSPRIVATE(bp
, iclog
); /* save for later */
1524 XFS_BUF_ZEROFLAGS(bp
);
1528 * Do an ordered write for the log block.
1529 * Its unnecessary to flush the first split block in the log wrap case.
1531 if (!split
&& (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
))
1532 XFS_BUF_ORDERED(bp
);
1534 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1535 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1537 xlog_verify_iclog(log
, iclog
, count
, B_TRUE
);
1539 /* account for log which doesn't start at block #0 */
1540 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1542 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1547 if ((error
= XFS_bwrite(bp
))) {
1548 xfs_ioerror_alert("xlog_sync", log
->l_mp
, bp
,
1553 bp
= iclog
->ic_log
->l_xbuf
;
1554 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) ==
1556 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1557 XFS_BUF_SET_ADDR(bp
, 0); /* logical 0 */
1558 XFS_BUF_SET_PTR(bp
, (xfs_caddr_t
)((__psint_t
)&(iclog
->ic_header
)+
1559 (__psint_t
)count
), split
);
1560 XFS_BUF_SET_FSPRIVATE(bp
, iclog
);
1561 XFS_BUF_ZEROFLAGS(bp
);
1564 if (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
)
1565 XFS_BUF_ORDERED(bp
);
1566 dptr
= XFS_BUF_PTR(bp
);
1568 * Bump the cycle numbers at the start of each block
1569 * since this part of the buffer is at the start of
1570 * a new cycle. Watch out for the header magic number
1573 for (i
= 0; i
< split
; i
+= BBSIZE
) {
1574 be32_add_cpu((__be32
*)dptr
, 1);
1575 if (be32_to_cpu(*(__be32
*)dptr
) == XLOG_HEADER_MAGIC_NUM
)
1576 be32_add_cpu((__be32
*)dptr
, 1);
1580 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1581 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1583 /* account for internal log which doesn't start at block #0 */
1584 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1586 if ((error
= XFS_bwrite(bp
))) {
1587 xfs_ioerror_alert("xlog_sync (split)", log
->l_mp
,
1588 bp
, XFS_BUF_ADDR(bp
));
1597 * Deallocate a log structure
1600 xlog_dealloc_log(xlog_t
*log
)
1602 xlog_in_core_t
*iclog
, *next_iclog
;
1605 iclog
= log
->l_iclog
;
1606 for (i
=0; i
<log
->l_iclog_bufs
; i
++) {
1607 sv_destroy(&iclog
->ic_force_wait
);
1608 sv_destroy(&iclog
->ic_write_wait
);
1609 xfs_buf_free(iclog
->ic_bp
);
1610 xlog_trace_iclog_dealloc(iclog
);
1611 next_iclog
= iclog
->ic_next
;
1615 spinlock_destroy(&log
->l_icloglock
);
1616 spinlock_destroy(&log
->l_grant_lock
);
1618 xfs_buf_free(log
->l_xbuf
);
1619 xlog_trace_loggrant_dealloc(log
);
1620 log
->l_mp
->m_log
= NULL
;
1622 } /* xlog_dealloc_log */
1625 * Update counters atomically now that memcpy is done.
1629 xlog_state_finish_copy(xlog_t
*log
,
1630 xlog_in_core_t
*iclog
,
1634 spin_lock(&log
->l_icloglock
);
1636 be32_add_cpu(&iclog
->ic_header
.h_num_logops
, record_cnt
);
1637 iclog
->ic_offset
+= copy_bytes
;
1639 spin_unlock(&log
->l_icloglock
);
1640 } /* xlog_state_finish_copy */
1646 * print out info relating to regions written which consume
1650 xlog_print_tic_res(xfs_mount_t
*mp
, xlog_ticket_t
*ticket
)
1653 uint ophdr_spc
= ticket
->t_res_num_ophdrs
* (uint
)sizeof(xlog_op_header_t
);
1655 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1656 static char *res_type_str
[XLOG_REG_TYPE_MAX
] = {
1677 static char *trans_type_str
[XFS_TRANS_TYPE_MAX
] = {
1720 xfs_fs_cmn_err(CE_WARN
, mp
,
1721 "xfs_log_write: reservation summary:\n"
1722 " trans type = %s (%u)\n"
1723 " unit res = %d bytes\n"
1724 " current res = %d bytes\n"
1725 " total reg = %u bytes (o/flow = %u bytes)\n"
1726 " ophdrs = %u (ophdr space = %u bytes)\n"
1727 " ophdr + reg = %u bytes\n"
1728 " num regions = %u\n",
1729 ((ticket
->t_trans_type
<= 0 ||
1730 ticket
->t_trans_type
> XFS_TRANS_TYPE_MAX
) ?
1731 "bad-trans-type" : trans_type_str
[ticket
->t_trans_type
-1]),
1732 ticket
->t_trans_type
,
1735 ticket
->t_res_arr_sum
, ticket
->t_res_o_flow
,
1736 ticket
->t_res_num_ophdrs
, ophdr_spc
,
1737 ticket
->t_res_arr_sum
+
1738 ticket
->t_res_o_flow
+ ophdr_spc
,
1741 for (i
= 0; i
< ticket
->t_res_num
; i
++) {
1742 uint r_type
= ticket
->t_res_arr
[i
].r_type
;
1744 "region[%u]: %s - %u bytes\n",
1746 ((r_type
<= 0 || r_type
> XLOG_REG_TYPE_MAX
) ?
1747 "bad-rtype" : res_type_str
[r_type
-1]),
1748 ticket
->t_res_arr
[i
].r_len
);
1753 * Write some region out to in-core log
1755 * This will be called when writing externally provided regions or when
1756 * writing out a commit record for a given transaction.
1758 * General algorithm:
1759 * 1. Find total length of this write. This may include adding to the
1760 * lengths passed in.
1761 * 2. Check whether we violate the tickets reservation.
1762 * 3. While writing to this iclog
1763 * A. Reserve as much space in this iclog as can get
1764 * B. If this is first write, save away start lsn
1765 * C. While writing this region:
1766 * 1. If first write of transaction, write start record
1767 * 2. Write log operation header (header per region)
1768 * 3. Find out if we can fit entire region into this iclog
1769 * 4. Potentially, verify destination memcpy ptr
1770 * 5. Memcpy (partial) region
1771 * 6. If partial copy, release iclog; otherwise, continue
1772 * copying more regions into current iclog
1773 * 4. Mark want sync bit (in simulation mode)
1774 * 5. Release iclog for potential flush to on-disk log.
1777 * 1. Panic if reservation is overrun. This should never happen since
1778 * reservation amounts are generated internal to the filesystem.
1780 * 1. Tickets are single threaded data structures.
1781 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1782 * syncing routine. When a single log_write region needs to span
1783 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1784 * on all log operation writes which don't contain the end of the
1785 * region. The XLOG_END_TRANS bit is used for the in-core log
1786 * operation which contains the end of the continued log_write region.
1787 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1788 * we don't really know exactly how much space will be used. As a result,
1789 * we don't update ic_offset until the end when we know exactly how many
1790 * bytes have been written out.
1793 xlog_write(xfs_mount_t
* mp
,
1794 xfs_log_iovec_t reg
[],
1796 xfs_log_ticket_t tic
,
1797 xfs_lsn_t
*start_lsn
,
1798 xlog_in_core_t
**commit_iclog
,
1801 xlog_t
*log
= mp
->m_log
;
1802 xlog_ticket_t
*ticket
= (xlog_ticket_t
*)tic
;
1803 xlog_in_core_t
*iclog
= NULL
; /* ptr to current in-core log */
1804 xlog_op_header_t
*logop_head
; /* ptr to log operation header */
1805 __psint_t ptr
; /* copy address into data region */
1806 int len
; /* # xlog_write() bytes 2 still copy */
1807 int index
; /* region index currently copying */
1808 int log_offset
; /* offset (from 0) into data region */
1809 int start_rec_copy
; /* # bytes to copy for start record */
1810 int partial_copy
; /* did we split a region? */
1811 int partial_copy_len
;/* # bytes copied if split region */
1812 int need_copy
; /* # bytes need to memcpy this region */
1813 int copy_len
; /* # bytes actually memcpy'ing */
1814 int copy_off
; /* # bytes from entry start */
1815 int contwr
; /* continued write of in-core log? */
1817 int record_cnt
= 0, data_cnt
= 0;
1819 partial_copy_len
= partial_copy
= 0;
1821 /* Calculate potential maximum space. Each region gets its own
1822 * xlog_op_header_t and may need to be double word aligned.
1825 if (ticket
->t_flags
& XLOG_TIC_INITED
) { /* acct for start rec of xact */
1826 len
+= sizeof(xlog_op_header_t
);
1827 ticket
->t_res_num_ophdrs
++;
1830 for (index
= 0; index
< nentries
; index
++) {
1831 len
+= sizeof(xlog_op_header_t
); /* each region gets >= 1 */
1832 ticket
->t_res_num_ophdrs
++;
1833 len
+= reg
[index
].i_len
;
1834 xlog_tic_add_region(ticket
, reg
[index
].i_len
, reg
[index
].i_type
);
1836 contwr
= *start_lsn
= 0;
1838 if (ticket
->t_curr_res
< len
) {
1839 xlog_print_tic_res(mp
, ticket
);
1842 "xfs_log_write: reservation ran out. Need to up reservation");
1844 /* Customer configurable panic */
1845 xfs_cmn_err(XFS_PTAG_LOGRES
, CE_ALERT
, mp
,
1846 "xfs_log_write: reservation ran out. Need to up reservation");
1847 /* If we did not panic, shutdown the filesystem */
1848 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1851 ticket
->t_curr_res
-= len
;
1853 for (index
= 0; index
< nentries
; ) {
1854 if ((error
= xlog_state_get_iclog_space(log
, len
, &iclog
, ticket
,
1855 &contwr
, &log_offset
)))
1858 ASSERT(log_offset
<= iclog
->ic_size
- 1);
1859 ptr
= (__psint_t
) ((char *)iclog
->ic_datap
+log_offset
);
1861 /* start_lsn is the first lsn written to. That's all we need. */
1863 *start_lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
1865 /* This loop writes out as many regions as can fit in the amount
1866 * of space which was allocated by xlog_state_get_iclog_space().
1868 while (index
< nentries
) {
1869 ASSERT(reg
[index
].i_len
% sizeof(__int32_t
) == 0);
1870 ASSERT((__psint_t
)ptr
% sizeof(__int32_t
) == 0);
1873 /* If first write for transaction, insert start record.
1874 * We can't be trying to commit if we are inited. We can't
1875 * have any "partial_copy" if we are inited.
1877 if (ticket
->t_flags
& XLOG_TIC_INITED
) {
1878 logop_head
= (xlog_op_header_t
*)ptr
;
1879 logop_head
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1880 logop_head
->oh_clientid
= ticket
->t_clientid
;
1881 logop_head
->oh_len
= 0;
1882 logop_head
->oh_flags
= XLOG_START_TRANS
;
1883 logop_head
->oh_res2
= 0;
1884 ticket
->t_flags
&= ~XLOG_TIC_INITED
; /* clear bit */
1887 start_rec_copy
= sizeof(xlog_op_header_t
);
1888 xlog_write_adv_cnt(ptr
, len
, log_offset
, start_rec_copy
);
1891 /* Copy log operation header directly into data section */
1892 logop_head
= (xlog_op_header_t
*)ptr
;
1893 logop_head
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1894 logop_head
->oh_clientid
= ticket
->t_clientid
;
1895 logop_head
->oh_res2
= 0;
1897 /* header copied directly */
1898 xlog_write_adv_cnt(ptr
, len
, log_offset
, sizeof(xlog_op_header_t
));
1900 /* are we copying a commit or unmount record? */
1901 logop_head
->oh_flags
= flags
;
1904 * We've seen logs corrupted with bad transaction client
1905 * ids. This makes sure that XFS doesn't generate them on.
1906 * Turn this into an EIO and shut down the filesystem.
1908 switch (logop_head
->oh_clientid
) {
1909 case XFS_TRANSACTION
:
1914 xfs_fs_cmn_err(CE_WARN
, mp
,
1915 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1916 logop_head
->oh_clientid
, tic
);
1917 return XFS_ERROR(EIO
);
1920 /* Partial write last time? => (partial_copy != 0)
1921 * need_copy is the amount we'd like to copy if everything could
1922 * fit in the current memcpy.
1924 need_copy
= reg
[index
].i_len
- partial_copy_len
;
1926 copy_off
= partial_copy_len
;
1927 if (need_copy
<= iclog
->ic_size
- log_offset
) { /*complete write */
1928 copy_len
= need_copy
;
1929 logop_head
->oh_len
= cpu_to_be32(copy_len
);
1931 logop_head
->oh_flags
|= (XLOG_END_TRANS
|XLOG_WAS_CONT_TRANS
);
1932 partial_copy_len
= partial_copy
= 0;
1933 } else { /* partial write */
1934 copy_len
= iclog
->ic_size
- log_offset
;
1935 logop_head
->oh_len
= cpu_to_be32(copy_len
);
1936 logop_head
->oh_flags
|= XLOG_CONTINUE_TRANS
;
1938 logop_head
->oh_flags
|= XLOG_WAS_CONT_TRANS
;
1939 partial_copy_len
+= copy_len
;
1941 len
+= sizeof(xlog_op_header_t
); /* from splitting of region */
1942 /* account for new log op header */
1943 ticket
->t_curr_res
-= sizeof(xlog_op_header_t
);
1944 ticket
->t_res_num_ophdrs
++;
1946 xlog_verify_dest_ptr(log
, ptr
);
1949 ASSERT(copy_len
>= 0);
1950 memcpy((xfs_caddr_t
)ptr
, reg
[index
].i_addr
+ copy_off
, copy_len
);
1951 xlog_write_adv_cnt(ptr
, len
, log_offset
, copy_len
);
1953 /* make copy_len total bytes copied, including headers */
1954 copy_len
+= start_rec_copy
+ sizeof(xlog_op_header_t
);
1956 data_cnt
+= contwr
? copy_len
: 0;
1957 if (partial_copy
) { /* copied partial region */
1958 /* already marked WANT_SYNC by xlog_state_get_iclog_space */
1959 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1960 record_cnt
= data_cnt
= 0;
1961 if ((error
= xlog_state_release_iclog(log
, iclog
)))
1963 break; /* don't increment index */
1964 } else { /* copied entire region */
1966 partial_copy_len
= partial_copy
= 0;
1968 if (iclog
->ic_size
- log_offset
<= sizeof(xlog_op_header_t
)) {
1969 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1970 record_cnt
= data_cnt
= 0;
1971 spin_lock(&log
->l_icloglock
);
1972 xlog_state_want_sync(log
, iclog
);
1973 spin_unlock(&log
->l_icloglock
);
1975 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1976 *commit_iclog
= iclog
;
1977 } else if ((error
= xlog_state_release_iclog(log
, iclog
)))
1979 if (index
== nentries
)
1980 return 0; /* we are done */
1984 } /* if (partial_copy) */
1985 } /* while (index < nentries) */
1986 } /* for (index = 0; index < nentries; ) */
1989 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1991 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1992 *commit_iclog
= iclog
;
1995 return xlog_state_release_iclog(log
, iclog
);
1999 /*****************************************************************************
2001 * State Machine functions
2003 *****************************************************************************
2006 /* Clean iclogs starting from the head. This ordering must be
2007 * maintained, so an iclog doesn't become ACTIVE beyond one that
2008 * is SYNCING. This is also required to maintain the notion that we use
2009 * a ordered wait queue to hold off would be writers to the log when every
2010 * iclog is trying to sync to disk.
2012 * State Change: DIRTY -> ACTIVE
2015 xlog_state_clean_log(xlog_t
*log
)
2017 xlog_in_core_t
*iclog
;
2020 iclog
= log
->l_iclog
;
2022 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2023 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
2024 iclog
->ic_offset
= 0;
2025 ASSERT(iclog
->ic_callback
== NULL
);
2027 * If the number of ops in this iclog indicate it just
2028 * contains the dummy transaction, we can
2029 * change state into IDLE (the second time around).
2030 * Otherwise we should change the state into
2032 * We don't need to cover the dummy.
2035 (be32_to_cpu(iclog
->ic_header
.h_num_logops
) ==
2040 * We have two dirty iclogs so start over
2041 * This could also be num of ops indicates
2042 * this is not the dummy going out.
2046 iclog
->ic_header
.h_num_logops
= 0;
2047 memset(iclog
->ic_header
.h_cycle_data
, 0,
2048 sizeof(iclog
->ic_header
.h_cycle_data
));
2049 iclog
->ic_header
.h_lsn
= 0;
2050 } else if (iclog
->ic_state
== XLOG_STATE_ACTIVE
)
2053 break; /* stop cleaning */
2054 iclog
= iclog
->ic_next
;
2055 } while (iclog
!= log
->l_iclog
);
2057 /* log is locked when we are called */
2059 * Change state for the dummy log recording.
2060 * We usually go to NEED. But we go to NEED2 if the changed indicates
2061 * we are done writing the dummy record.
2062 * If we are done with the second dummy recored (DONE2), then
2066 switch (log
->l_covered_state
) {
2067 case XLOG_STATE_COVER_IDLE
:
2068 case XLOG_STATE_COVER_NEED
:
2069 case XLOG_STATE_COVER_NEED2
:
2070 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2073 case XLOG_STATE_COVER_DONE
:
2075 log
->l_covered_state
= XLOG_STATE_COVER_NEED2
;
2077 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2080 case XLOG_STATE_COVER_DONE2
:
2082 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
2084 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2091 } /* xlog_state_clean_log */
2094 xlog_get_lowest_lsn(
2097 xlog_in_core_t
*lsn_log
;
2098 xfs_lsn_t lowest_lsn
, lsn
;
2100 lsn_log
= log
->l_iclog
;
2103 if (!(lsn_log
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
))) {
2104 lsn
= be64_to_cpu(lsn_log
->ic_header
.h_lsn
);
2105 if ((lsn
&& !lowest_lsn
) ||
2106 (XFS_LSN_CMP(lsn
, lowest_lsn
) < 0)) {
2110 lsn_log
= lsn_log
->ic_next
;
2111 } while (lsn_log
!= log
->l_iclog
);
2117 xlog_state_do_callback(
2120 xlog_in_core_t
*ciclog
)
2122 xlog_in_core_t
*iclog
;
2123 xlog_in_core_t
*first_iclog
; /* used to know when we've
2124 * processed all iclogs once */
2125 xfs_log_callback_t
*cb
, *cb_next
;
2127 xfs_lsn_t lowest_lsn
;
2128 int ioerrors
; /* counter: iclogs with errors */
2129 int loopdidcallbacks
; /* flag: inner loop did callbacks*/
2130 int funcdidcallbacks
; /* flag: function did callbacks */
2131 int repeats
; /* for issuing console warnings if
2132 * looping too many times */
2135 spin_lock(&log
->l_icloglock
);
2136 first_iclog
= iclog
= log
->l_iclog
;
2138 funcdidcallbacks
= 0;
2143 * Scan all iclogs starting with the one pointed to by the
2144 * log. Reset this starting point each time the log is
2145 * unlocked (during callbacks).
2147 * Keep looping through iclogs until one full pass is made
2148 * without running any callbacks.
2150 first_iclog
= log
->l_iclog
;
2151 iclog
= log
->l_iclog
;
2152 loopdidcallbacks
= 0;
2157 /* skip all iclogs in the ACTIVE & DIRTY states */
2158 if (iclog
->ic_state
&
2159 (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
)) {
2160 iclog
= iclog
->ic_next
;
2165 * Between marking a filesystem SHUTDOWN and stopping
2166 * the log, we do flush all iclogs to disk (if there
2167 * wasn't a log I/O error). So, we do want things to
2168 * go smoothly in case of just a SHUTDOWN w/o a
2171 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
2173 * Can only perform callbacks in order. Since
2174 * this iclog is not in the DONE_SYNC/
2175 * DO_CALLBACK state, we skip the rest and
2176 * just try to clean up. If we set our iclog
2177 * to DO_CALLBACK, we will not process it when
2178 * we retry since a previous iclog is in the
2179 * CALLBACK and the state cannot change since
2180 * we are holding the l_icloglock.
2182 if (!(iclog
->ic_state
&
2183 (XLOG_STATE_DONE_SYNC
|
2184 XLOG_STATE_DO_CALLBACK
))) {
2185 if (ciclog
&& (ciclog
->ic_state
==
2186 XLOG_STATE_DONE_SYNC
)) {
2187 ciclog
->ic_state
= XLOG_STATE_DO_CALLBACK
;
2192 * We now have an iclog that is in either the
2193 * DO_CALLBACK or DONE_SYNC states. The other
2194 * states (WANT_SYNC, SYNCING, or CALLBACK were
2195 * caught by the above if and are going to
2196 * clean (i.e. we aren't doing their callbacks)
2201 * We will do one more check here to see if we
2202 * have chased our tail around.
2205 lowest_lsn
= xlog_get_lowest_lsn(log
);
2207 XFS_LSN_CMP(lowest_lsn
,
2208 be64_to_cpu(iclog
->ic_header
.h_lsn
)) < 0) {
2209 iclog
= iclog
->ic_next
;
2210 continue; /* Leave this iclog for
2214 iclog
->ic_state
= XLOG_STATE_CALLBACK
;
2216 spin_unlock(&log
->l_icloglock
);
2218 /* l_last_sync_lsn field protected by
2219 * l_grant_lock. Don't worry about iclog's lsn.
2220 * No one else can be here except us.
2222 spin_lock(&log
->l_grant_lock
);
2223 ASSERT(XFS_LSN_CMP(log
->l_last_sync_lsn
,
2224 be64_to_cpu(iclog
->ic_header
.h_lsn
)) <= 0);
2225 log
->l_last_sync_lsn
=
2226 be64_to_cpu(iclog
->ic_header
.h_lsn
);
2227 spin_unlock(&log
->l_grant_lock
);
2230 spin_unlock(&log
->l_icloglock
);
2235 * Keep processing entries in the callback list until
2236 * we come around and it is empty. We need to
2237 * atomically see that the list is empty and change the
2238 * state to DIRTY so that we don't miss any more
2239 * callbacks being added.
2241 spin_lock(&iclog
->ic_callback_lock
);
2242 cb
= iclog
->ic_callback
;
2244 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
2245 iclog
->ic_callback
= NULL
;
2246 spin_unlock(&iclog
->ic_callback_lock
);
2248 /* perform callbacks in the order given */
2249 for (; cb
; cb
= cb_next
) {
2250 cb_next
= cb
->cb_next
;
2251 cb
->cb_func(cb
->cb_arg
, aborted
);
2253 spin_lock(&iclog
->ic_callback_lock
);
2254 cb
= iclog
->ic_callback
;
2260 spin_lock(&log
->l_icloglock
);
2261 ASSERT(iclog
->ic_callback
== NULL
);
2262 spin_unlock(&iclog
->ic_callback_lock
);
2263 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
))
2264 iclog
->ic_state
= XLOG_STATE_DIRTY
;
2267 * Transition from DIRTY to ACTIVE if applicable.
2268 * NOP if STATE_IOERROR.
2270 xlog_state_clean_log(log
);
2272 /* wake up threads waiting in xfs_log_force() */
2273 sv_broadcast(&iclog
->ic_force_wait
);
2275 iclog
= iclog
->ic_next
;
2276 } while (first_iclog
!= iclog
);
2278 if (repeats
> 5000) {
2279 flushcnt
+= repeats
;
2281 xfs_fs_cmn_err(CE_WARN
, log
->l_mp
,
2282 "%s: possible infinite loop (%d iterations)",
2283 __func__
, flushcnt
);
2285 } while (!ioerrors
&& loopdidcallbacks
);
2288 * make one last gasp attempt to see if iclogs are being left in
2292 if (funcdidcallbacks
) {
2293 first_iclog
= iclog
= log
->l_iclog
;
2295 ASSERT(iclog
->ic_state
!= XLOG_STATE_DO_CALLBACK
);
2297 * Terminate the loop if iclogs are found in states
2298 * which will cause other threads to clean up iclogs.
2300 * SYNCING - i/o completion will go through logs
2301 * DONE_SYNC - interrupt thread should be waiting for
2303 * IOERROR - give up hope all ye who enter here
2305 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
||
2306 iclog
->ic_state
== XLOG_STATE_SYNCING
||
2307 iclog
->ic_state
== XLOG_STATE_DONE_SYNC
||
2308 iclog
->ic_state
== XLOG_STATE_IOERROR
)
2310 iclog
= iclog
->ic_next
;
2311 } while (first_iclog
!= iclog
);
2315 if (log
->l_iclog
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_IOERROR
))
2317 spin_unlock(&log
->l_icloglock
);
2320 sv_broadcast(&log
->l_flush_wait
);
2325 * Finish transitioning this iclog to the dirty state.
2327 * Make sure that we completely execute this routine only when this is
2328 * the last call to the iclog. There is a good chance that iclog flushes,
2329 * when we reach the end of the physical log, get turned into 2 separate
2330 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2331 * routine. By using the reference count bwritecnt, we guarantee that only
2332 * the second completion goes through.
2334 * Callbacks could take time, so they are done outside the scope of the
2335 * global state machine log lock.
2338 xlog_state_done_syncing(
2339 xlog_in_core_t
*iclog
,
2342 xlog_t
*log
= iclog
->ic_log
;
2344 spin_lock(&log
->l_icloglock
);
2346 ASSERT(iclog
->ic_state
== XLOG_STATE_SYNCING
||
2347 iclog
->ic_state
== XLOG_STATE_IOERROR
);
2348 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
2349 ASSERT(iclog
->ic_bwritecnt
== 1 || iclog
->ic_bwritecnt
== 2);
2353 * If we got an error, either on the first buffer, or in the case of
2354 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2355 * and none should ever be attempted to be written to disk
2358 if (iclog
->ic_state
!= XLOG_STATE_IOERROR
) {
2359 if (--iclog
->ic_bwritecnt
== 1) {
2360 spin_unlock(&log
->l_icloglock
);
2363 iclog
->ic_state
= XLOG_STATE_DONE_SYNC
;
2367 * Someone could be sleeping prior to writing out the next
2368 * iclog buffer, we wake them all, one will get to do the
2369 * I/O, the others get to wait for the result.
2371 sv_broadcast(&iclog
->ic_write_wait
);
2372 spin_unlock(&log
->l_icloglock
);
2373 xlog_state_do_callback(log
, aborted
, iclog
); /* also cleans log */
2374 } /* xlog_state_done_syncing */
2378 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2379 * sleep. We wait on the flush queue on the head iclog as that should be
2380 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2381 * we will wait here and all new writes will sleep until a sync completes.
2383 * The in-core logs are used in a circular fashion. They are not used
2384 * out-of-order even when an iclog past the head is free.
2387 * * log_offset where xlog_write() can start writing into the in-core
2389 * * in-core log pointer to which xlog_write() should write.
2390 * * boolean indicating this is a continued write to an in-core log.
2391 * If this is the last write, then the in-core log's offset field
2392 * needs to be incremented, depending on the amount of data which
2396 xlog_state_get_iclog_space(xlog_t
*log
,
2398 xlog_in_core_t
**iclogp
,
2399 xlog_ticket_t
*ticket
,
2400 int *continued_write
,
2404 xlog_rec_header_t
*head
;
2405 xlog_in_core_t
*iclog
;
2409 spin_lock(&log
->l_icloglock
);
2410 if (XLOG_FORCED_SHUTDOWN(log
)) {
2411 spin_unlock(&log
->l_icloglock
);
2412 return XFS_ERROR(EIO
);
2415 iclog
= log
->l_iclog
;
2416 if (iclog
->ic_state
!= XLOG_STATE_ACTIVE
) {
2417 xlog_trace_iclog(iclog
, XLOG_TRACE_SLEEP_FLUSH
);
2418 XFS_STATS_INC(xs_log_noiclogs
);
2420 /* Wait for log writes to have flushed */
2421 sv_wait(&log
->l_flush_wait
, 0, &log
->l_icloglock
, 0);
2425 head
= &iclog
->ic_header
;
2427 atomic_inc(&iclog
->ic_refcnt
); /* prevents sync */
2428 log_offset
= iclog
->ic_offset
;
2430 /* On the 1st write to an iclog, figure out lsn. This works
2431 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2432 * committing to. If the offset is set, that's how many blocks
2435 if (log_offset
== 0) {
2436 ticket
->t_curr_res
-= log
->l_iclog_hsize
;
2437 xlog_tic_add_region(ticket
,
2439 XLOG_REG_TYPE_LRHEADER
);
2440 head
->h_cycle
= cpu_to_be32(log
->l_curr_cycle
);
2441 head
->h_lsn
= cpu_to_be64(
2442 xlog_assign_lsn(log
->l_curr_cycle
, log
->l_curr_block
));
2443 ASSERT(log
->l_curr_block
>= 0);
2446 /* If there is enough room to write everything, then do it. Otherwise,
2447 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2448 * bit is on, so this will get flushed out. Don't update ic_offset
2449 * until you know exactly how many bytes get copied. Therefore, wait
2450 * until later to update ic_offset.
2452 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2453 * can fit into remaining data section.
2455 if (iclog
->ic_size
- iclog
->ic_offset
< 2*sizeof(xlog_op_header_t
)) {
2456 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2459 * If I'm the only one writing to this iclog, sync it to disk.
2460 * We need to do an atomic compare and decrement here to avoid
2461 * racing with concurrent atomic_dec_and_lock() calls in
2462 * xlog_state_release_iclog() when there is more than one
2463 * reference to the iclog.
2465 if (!atomic_add_unless(&iclog
->ic_refcnt
, -1, 1)) {
2466 /* we are the only one */
2467 spin_unlock(&log
->l_icloglock
);
2468 error
= xlog_state_release_iclog(log
, iclog
);
2472 spin_unlock(&log
->l_icloglock
);
2477 /* Do we have enough room to write the full amount in the remainder
2478 * of this iclog? Or must we continue a write on the next iclog and
2479 * mark this iclog as completely taken? In the case where we switch
2480 * iclogs (to mark it taken), this particular iclog will release/sync
2481 * to disk in xlog_write().
2483 if (len
<= iclog
->ic_size
- iclog
->ic_offset
) {
2484 *continued_write
= 0;
2485 iclog
->ic_offset
+= len
;
2487 *continued_write
= 1;
2488 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2492 ASSERT(iclog
->ic_offset
<= iclog
->ic_size
);
2493 spin_unlock(&log
->l_icloglock
);
2495 *logoffsetp
= log_offset
;
2497 } /* xlog_state_get_iclog_space */
2500 * Atomically get the log space required for a log ticket.
2502 * Once a ticket gets put onto the reserveq, it will only return after
2503 * the needed reservation is satisfied.
2506 xlog_grant_log_space(xlog_t
*log
,
2517 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2518 panic("grant Recovery problem");
2521 /* Is there space or do we need to sleep? */
2522 spin_lock(&log
->l_grant_lock
);
2523 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: enter");
2525 /* something is already sleeping; insert new transaction at end */
2526 if (log
->l_reserve_headq
) {
2527 xlog_ins_ticketq(&log
->l_reserve_headq
, tic
);
2528 xlog_trace_loggrant(log
, tic
,
2529 "xlog_grant_log_space: sleep 1");
2531 * Gotta check this before going to sleep, while we're
2532 * holding the grant lock.
2534 if (XLOG_FORCED_SHUTDOWN(log
))
2537 XFS_STATS_INC(xs_sleep_logspace
);
2538 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2540 * If we got an error, and the filesystem is shutting down,
2541 * we'll catch it down below. So just continue...
2543 xlog_trace_loggrant(log
, tic
,
2544 "xlog_grant_log_space: wake 1");
2545 spin_lock(&log
->l_grant_lock
);
2547 if (tic
->t_flags
& XFS_LOG_PERM_RESERV
)
2548 need_bytes
= tic
->t_unit_res
*tic
->t_ocnt
;
2550 need_bytes
= tic
->t_unit_res
;
2553 if (XLOG_FORCED_SHUTDOWN(log
))
2556 free_bytes
= xlog_space_left(log
, log
->l_grant_reserve_cycle
,
2557 log
->l_grant_reserve_bytes
);
2558 if (free_bytes
< need_bytes
) {
2559 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2560 xlog_ins_ticketq(&log
->l_reserve_headq
, tic
);
2561 xlog_trace_loggrant(log
, tic
,
2562 "xlog_grant_log_space: sleep 2");
2563 spin_unlock(&log
->l_grant_lock
);
2564 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2565 spin_lock(&log
->l_grant_lock
);
2567 XFS_STATS_INC(xs_sleep_logspace
);
2568 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2570 spin_lock(&log
->l_grant_lock
);
2571 if (XLOG_FORCED_SHUTDOWN(log
))
2574 xlog_trace_loggrant(log
, tic
,
2575 "xlog_grant_log_space: wake 2");
2577 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2578 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2580 /* we've got enough space */
2581 xlog_grant_add_space(log
, need_bytes
);
2583 tail_lsn
= log
->l_tail_lsn
;
2585 * Check to make sure the grant write head didn't just over lap the
2586 * tail. If the cycles are the same, we can't be overlapping.
2587 * Otherwise, make sure that the cycles differ by exactly one and
2588 * check the byte count.
2590 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2591 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2592 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2595 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: exit");
2596 xlog_verify_grant_head(log
, 1);
2597 spin_unlock(&log
->l_grant_lock
);
2601 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2602 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2603 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: err_ret");
2605 * If we are failing, make sure the ticket doesn't have any
2606 * current reservations. We don't want to add this back when
2607 * the ticket/transaction gets cancelled.
2609 tic
->t_curr_res
= 0;
2610 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2611 spin_unlock(&log
->l_grant_lock
);
2612 return XFS_ERROR(EIO
);
2613 } /* xlog_grant_log_space */
2617 * Replenish the byte reservation required by moving the grant write head.
2622 xlog_regrant_write_log_space(xlog_t
*log
,
2625 int free_bytes
, need_bytes
;
2626 xlog_ticket_t
*ntic
;
2631 tic
->t_curr_res
= tic
->t_unit_res
;
2632 xlog_tic_reset_res(tic
);
2638 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2639 panic("regrant Recovery problem");
2642 spin_lock(&log
->l_grant_lock
);
2643 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: enter");
2645 if (XLOG_FORCED_SHUTDOWN(log
))
2648 /* If there are other waiters on the queue then give them a
2649 * chance at logspace before us. Wake up the first waiters,
2650 * if we do not wake up all the waiters then go to sleep waiting
2651 * for more free space, otherwise try to get some space for
2654 need_bytes
= tic
->t_unit_res
;
2655 if ((ntic
= log
->l_write_headq
)) {
2656 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2657 log
->l_grant_write_bytes
);
2659 ASSERT(ntic
->t_flags
& XLOG_TIC_PERM_RESERV
);
2661 if (free_bytes
< ntic
->t_unit_res
)
2663 free_bytes
-= ntic
->t_unit_res
;
2664 sv_signal(&ntic
->t_wait
);
2665 ntic
= ntic
->t_next
;
2666 } while (ntic
!= log
->l_write_headq
);
2668 if (ntic
!= log
->l_write_headq
) {
2669 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2670 xlog_ins_ticketq(&log
->l_write_headq
, tic
);
2672 xlog_trace_loggrant(log
, tic
,
2673 "xlog_regrant_write_log_space: sleep 1");
2674 spin_unlock(&log
->l_grant_lock
);
2675 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2676 spin_lock(&log
->l_grant_lock
);
2678 XFS_STATS_INC(xs_sleep_logspace
);
2679 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
,
2680 &log
->l_grant_lock
, s
);
2682 /* If we're shutting down, this tic is already
2684 spin_lock(&log
->l_grant_lock
);
2685 if (XLOG_FORCED_SHUTDOWN(log
))
2688 xlog_trace_loggrant(log
, tic
,
2689 "xlog_regrant_write_log_space: wake 1");
2694 if (XLOG_FORCED_SHUTDOWN(log
))
2697 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2698 log
->l_grant_write_bytes
);
2699 if (free_bytes
< need_bytes
) {
2700 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2701 xlog_ins_ticketq(&log
->l_write_headq
, tic
);
2702 spin_unlock(&log
->l_grant_lock
);
2703 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2704 spin_lock(&log
->l_grant_lock
);
2706 XFS_STATS_INC(xs_sleep_logspace
);
2707 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2709 /* If we're shutting down, this tic is already off the queue */
2710 spin_lock(&log
->l_grant_lock
);
2711 if (XLOG_FORCED_SHUTDOWN(log
))
2714 xlog_trace_loggrant(log
, tic
,
2715 "xlog_regrant_write_log_space: wake 2");
2717 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2718 xlog_del_ticketq(&log
->l_write_headq
, tic
);
2720 /* we've got enough space */
2721 xlog_grant_add_space_write(log
, need_bytes
);
2723 tail_lsn
= log
->l_tail_lsn
;
2724 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2725 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2726 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2730 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: exit");
2731 xlog_verify_grant_head(log
, 1);
2732 spin_unlock(&log
->l_grant_lock
);
2737 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2738 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2739 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: err_ret");
2741 * If we are failing, make sure the ticket doesn't have any
2742 * current reservations. We don't want to add this back when
2743 * the ticket/transaction gets cancelled.
2745 tic
->t_curr_res
= 0;
2746 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2747 spin_unlock(&log
->l_grant_lock
);
2748 return XFS_ERROR(EIO
);
2749 } /* xlog_regrant_write_log_space */
2752 /* The first cnt-1 times through here we don't need to
2753 * move the grant write head because the permanent
2754 * reservation has reserved cnt times the unit amount.
2755 * Release part of current permanent unit reservation and
2756 * reset current reservation to be one units worth. Also
2757 * move grant reservation head forward.
2760 xlog_regrant_reserve_log_space(xlog_t
*log
,
2761 xlog_ticket_t
*ticket
)
2763 xlog_trace_loggrant(log
, ticket
,
2764 "xlog_regrant_reserve_log_space: enter");
2765 if (ticket
->t_cnt
> 0)
2768 spin_lock(&log
->l_grant_lock
);
2769 xlog_grant_sub_space(log
, ticket
->t_curr_res
);
2770 ticket
->t_curr_res
= ticket
->t_unit_res
;
2771 xlog_tic_reset_res(ticket
);
2772 xlog_trace_loggrant(log
, ticket
,
2773 "xlog_regrant_reserve_log_space: sub current res");
2774 xlog_verify_grant_head(log
, 1);
2776 /* just return if we still have some of the pre-reserved space */
2777 if (ticket
->t_cnt
> 0) {
2778 spin_unlock(&log
->l_grant_lock
);
2782 xlog_grant_add_space_reserve(log
, ticket
->t_unit_res
);
2783 xlog_trace_loggrant(log
, ticket
,
2784 "xlog_regrant_reserve_log_space: exit");
2785 xlog_verify_grant_head(log
, 0);
2786 spin_unlock(&log
->l_grant_lock
);
2787 ticket
->t_curr_res
= ticket
->t_unit_res
;
2788 xlog_tic_reset_res(ticket
);
2789 } /* xlog_regrant_reserve_log_space */
2793 * Give back the space left from a reservation.
2795 * All the information we need to make a correct determination of space left
2796 * is present. For non-permanent reservations, things are quite easy. The
2797 * count should have been decremented to zero. We only need to deal with the
2798 * space remaining in the current reservation part of the ticket. If the
2799 * ticket contains a permanent reservation, there may be left over space which
2800 * needs to be released. A count of N means that N-1 refills of the current
2801 * reservation can be done before we need to ask for more space. The first
2802 * one goes to fill up the first current reservation. Once we run out of
2803 * space, the count will stay at zero and the only space remaining will be
2804 * in the current reservation field.
2807 xlog_ungrant_log_space(xlog_t
*log
,
2808 xlog_ticket_t
*ticket
)
2810 if (ticket
->t_cnt
> 0)
2813 spin_lock(&log
->l_grant_lock
);
2814 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: enter");
2816 xlog_grant_sub_space(log
, ticket
->t_curr_res
);
2818 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: sub current");
2820 /* If this is a permanent reservation ticket, we may be able to free
2821 * up more space based on the remaining count.
2823 if (ticket
->t_cnt
> 0) {
2824 ASSERT(ticket
->t_flags
& XLOG_TIC_PERM_RESERV
);
2825 xlog_grant_sub_space(log
, ticket
->t_unit_res
*ticket
->t_cnt
);
2828 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: exit");
2829 xlog_verify_grant_head(log
, 1);
2830 spin_unlock(&log
->l_grant_lock
);
2831 xfs_log_move_tail(log
->l_mp
, 1);
2832 } /* xlog_ungrant_log_space */
2836 * Flush iclog to disk if this is the last reference to the given iclog and
2837 * the WANT_SYNC bit is set.
2839 * When this function is entered, the iclog is not necessarily in the
2840 * WANT_SYNC state. It may be sitting around waiting to get filled.
2845 xlog_state_release_iclog(
2847 xlog_in_core_t
*iclog
)
2849 int sync
= 0; /* do we sync? */
2851 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
2852 return XFS_ERROR(EIO
);
2854 ASSERT(atomic_read(&iclog
->ic_refcnt
) > 0);
2855 if (!atomic_dec_and_lock(&iclog
->ic_refcnt
, &log
->l_icloglock
))
2858 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2859 spin_unlock(&log
->l_icloglock
);
2860 return XFS_ERROR(EIO
);
2862 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2863 iclog
->ic_state
== XLOG_STATE_WANT_SYNC
);
2865 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
) {
2866 /* update tail before writing to iclog */
2867 xlog_assign_tail_lsn(log
->l_mp
);
2869 iclog
->ic_state
= XLOG_STATE_SYNCING
;
2870 iclog
->ic_header
.h_tail_lsn
= cpu_to_be64(log
->l_tail_lsn
);
2871 xlog_verify_tail_lsn(log
, iclog
, log
->l_tail_lsn
);
2872 /* cycle incremented when incrementing curr_block */
2874 spin_unlock(&log
->l_icloglock
);
2877 * We let the log lock go, so it's possible that we hit a log I/O
2878 * error or some other SHUTDOWN condition that marks the iclog
2879 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2880 * this iclog has consistent data, so we ignore IOERROR
2881 * flags after this point.
2884 return xlog_sync(log
, iclog
);
2886 } /* xlog_state_release_iclog */
2890 * This routine will mark the current iclog in the ring as WANT_SYNC
2891 * and move the current iclog pointer to the next iclog in the ring.
2892 * When this routine is called from xlog_state_get_iclog_space(), the
2893 * exact size of the iclog has not yet been determined. All we know is
2894 * that every data block. We have run out of space in this log record.
2897 xlog_state_switch_iclogs(xlog_t
*log
,
2898 xlog_in_core_t
*iclog
,
2901 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
);
2903 eventual_size
= iclog
->ic_offset
;
2904 iclog
->ic_state
= XLOG_STATE_WANT_SYNC
;
2905 iclog
->ic_header
.h_prev_block
= cpu_to_be32(log
->l_prev_block
);
2906 log
->l_prev_block
= log
->l_curr_block
;
2907 log
->l_prev_cycle
= log
->l_curr_cycle
;
2909 /* roll log?: ic_offset changed later */
2910 log
->l_curr_block
+= BTOBB(eventual_size
)+BTOBB(log
->l_iclog_hsize
);
2912 /* Round up to next log-sunit */
2913 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
2914 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
2915 __uint32_t sunit_bb
= BTOBB(log
->l_mp
->m_sb
.sb_logsunit
);
2916 log
->l_curr_block
= roundup(log
->l_curr_block
, sunit_bb
);
2919 if (log
->l_curr_block
>= log
->l_logBBsize
) {
2920 log
->l_curr_cycle
++;
2921 if (log
->l_curr_cycle
== XLOG_HEADER_MAGIC_NUM
)
2922 log
->l_curr_cycle
++;
2923 log
->l_curr_block
-= log
->l_logBBsize
;
2924 ASSERT(log
->l_curr_block
>= 0);
2926 ASSERT(iclog
== log
->l_iclog
);
2927 log
->l_iclog
= iclog
->ic_next
;
2928 } /* xlog_state_switch_iclogs */
2932 * Write out all data in the in-core log as of this exact moment in time.
2934 * Data may be written to the in-core log during this call. However,
2935 * we don't guarantee this data will be written out. A change from past
2936 * implementation means this routine will *not* write out zero length LRs.
2938 * Basically, we try and perform an intelligent scan of the in-core logs.
2939 * If we determine there is no flushable data, we just return. There is no
2940 * flushable data if:
2942 * 1. the current iclog is active and has no data; the previous iclog
2943 * is in the active or dirty state.
2944 * 2. the current iclog is drity, and the previous iclog is in the
2945 * active or dirty state.
2949 * 1. the current iclog is not in the active nor dirty state.
2950 * 2. the current iclog dirty, and the previous iclog is not in the
2951 * active nor dirty state.
2952 * 3. the current iclog is active, and there is another thread writing
2953 * to this particular iclog.
2954 * 4. a) the current iclog is active and has no other writers
2955 * b) when we return from flushing out this iclog, it is still
2956 * not in the active nor dirty state.
2959 xlog_state_sync_all(xlog_t
*log
, uint flags
, int *log_flushed
)
2961 xlog_in_core_t
*iclog
;
2964 spin_lock(&log
->l_icloglock
);
2966 iclog
= log
->l_iclog
;
2967 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2968 spin_unlock(&log
->l_icloglock
);
2969 return XFS_ERROR(EIO
);
2972 /* If the head iclog is not active nor dirty, we just attach
2973 * ourselves to the head and go to sleep.
2975 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2976 iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2978 * If the head is dirty or (active and empty), then
2979 * we need to look at the previous iclog. If the previous
2980 * iclog is active or dirty we are done. There is nothing
2981 * to sync out. Otherwise, we attach ourselves to the
2982 * previous iclog and go to sleep.
2984 if (iclog
->ic_state
== XLOG_STATE_DIRTY
||
2985 (atomic_read(&iclog
->ic_refcnt
) == 0
2986 && iclog
->ic_offset
== 0)) {
2987 iclog
= iclog
->ic_prev
;
2988 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2989 iclog
->ic_state
== XLOG_STATE_DIRTY
)
2994 if (atomic_read(&iclog
->ic_refcnt
) == 0) {
2995 /* We are the only one with access to this
2996 * iclog. Flush it out now. There should
2997 * be a roundoff of zero to show that someone
2998 * has already taken care of the roundoff from
2999 * the previous sync.
3001 atomic_inc(&iclog
->ic_refcnt
);
3002 lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
3003 xlog_state_switch_iclogs(log
, iclog
, 0);
3004 spin_unlock(&log
->l_icloglock
);
3006 if (xlog_state_release_iclog(log
, iclog
))
3007 return XFS_ERROR(EIO
);
3009 spin_lock(&log
->l_icloglock
);
3010 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) == lsn
&&
3011 iclog
->ic_state
!= XLOG_STATE_DIRTY
)
3016 /* Someone else is writing to this iclog.
3017 * Use its call to flush out the data. However,
3018 * the other thread may not force out this LR,
3019 * so we mark it WANT_SYNC.
3021 xlog_state_switch_iclogs(log
, iclog
, 0);
3027 /* By the time we come around again, the iclog could've been filled
3028 * which would give it another lsn. If we have a new lsn, just
3029 * return because the relevant data has been flushed.
3032 if (flags
& XFS_LOG_SYNC
) {
3034 * We must check if we're shutting down here, before
3035 * we wait, while we're holding the l_icloglock.
3036 * Then we check again after waking up, in case our
3037 * sleep was disturbed by a bad news.
3039 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3040 spin_unlock(&log
->l_icloglock
);
3041 return XFS_ERROR(EIO
);
3043 XFS_STATS_INC(xs_log_force_sleep
);
3044 sv_wait(&iclog
->ic_force_wait
, PINOD
, &log
->l_icloglock
, s
);
3046 * No need to grab the log lock here since we're
3047 * only deciding whether or not to return EIO
3048 * and the memory read should be atomic.
3050 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3051 return XFS_ERROR(EIO
);
3057 spin_unlock(&log
->l_icloglock
);
3060 } /* xlog_state_sync_all */
3064 * Used by code which implements synchronous log forces.
3066 * Find in-core log with lsn.
3067 * If it is in the DIRTY state, just return.
3068 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3069 * state and go to sleep or return.
3070 * If it is in any other state, go to sleep or return.
3072 * If filesystem activity goes to zero, the iclog will get flushed only by
3076 xlog_state_sync(xlog_t
*log
,
3081 xlog_in_core_t
*iclog
;
3082 int already_slept
= 0;
3085 spin_lock(&log
->l_icloglock
);
3086 iclog
= log
->l_iclog
;
3088 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3089 spin_unlock(&log
->l_icloglock
);
3090 return XFS_ERROR(EIO
);
3094 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) != lsn
) {
3095 iclog
= iclog
->ic_next
;
3099 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
3100 spin_unlock(&log
->l_icloglock
);
3104 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3106 * We sleep here if we haven't already slept (e.g.
3107 * this is the first time we've looked at the correct
3108 * iclog buf) and the buffer before us is going to
3109 * be sync'ed. The reason for this is that if we
3110 * are doing sync transactions here, by waiting for
3111 * the previous I/O to complete, we can allow a few
3112 * more transactions into this iclog before we close
3115 * Otherwise, we mark the buffer WANT_SYNC, and bump
3116 * up the refcnt so we can release the log (which drops
3117 * the ref count). The state switch keeps new transaction
3118 * commits from using this buffer. When the current commits
3119 * finish writing into the buffer, the refcount will drop to
3120 * zero and the buffer will go out then.
3122 if (!already_slept
&&
3123 (iclog
->ic_prev
->ic_state
& (XLOG_STATE_WANT_SYNC
|
3124 XLOG_STATE_SYNCING
))) {
3125 ASSERT(!(iclog
->ic_state
& XLOG_STATE_IOERROR
));
3126 XFS_STATS_INC(xs_log_force_sleep
);
3127 sv_wait(&iclog
->ic_prev
->ic_write_wait
, PSWP
,
3128 &log
->l_icloglock
, s
);
3133 atomic_inc(&iclog
->ic_refcnt
);
3134 xlog_state_switch_iclogs(log
, iclog
, 0);
3135 spin_unlock(&log
->l_icloglock
);
3136 if (xlog_state_release_iclog(log
, iclog
))
3137 return XFS_ERROR(EIO
);
3139 spin_lock(&log
->l_icloglock
);
3143 if ((flags
& XFS_LOG_SYNC
) && /* sleep */
3144 !(iclog
->ic_state
& (XLOG_STATE_ACTIVE
| XLOG_STATE_DIRTY
))) {
3147 * Don't wait on completion if we know that we've
3148 * gotten a log write error.
3150 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3151 spin_unlock(&log
->l_icloglock
);
3152 return XFS_ERROR(EIO
);
3154 XFS_STATS_INC(xs_log_force_sleep
);
3155 sv_wait(&iclog
->ic_force_wait
, PSWP
, &log
->l_icloglock
, s
);
3157 * No need to grab the log lock here since we're
3158 * only deciding whether or not to return EIO
3159 * and the memory read should be atomic.
3161 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3162 return XFS_ERROR(EIO
);
3164 } else { /* just return */
3165 spin_unlock(&log
->l_icloglock
);
3169 } while (iclog
!= log
->l_iclog
);
3171 spin_unlock(&log
->l_icloglock
);
3173 } /* xlog_state_sync */
3177 * Called when we want to mark the current iclog as being ready to sync to
3181 xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
)
3183 ASSERT(spin_is_locked(&log
->l_icloglock
));
3185 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3186 xlog_state_switch_iclogs(log
, iclog
, 0);
3188 ASSERT(iclog
->ic_state
&
3189 (XLOG_STATE_WANT_SYNC
|XLOG_STATE_IOERROR
));
3194 /*****************************************************************************
3198 *****************************************************************************
3202 * Free a used ticket when its refcount falls to zero.
3206 xlog_ticket_t
*ticket
)
3208 ASSERT(atomic_read(&ticket
->t_ref
) > 0);
3209 if (atomic_dec_and_test(&ticket
->t_ref
)) {
3210 sv_destroy(&ticket
->t_wait
);
3211 kmem_zone_free(xfs_log_ticket_zone
, ticket
);
3217 xlog_ticket_t
*ticket
)
3219 ASSERT(atomic_read(&ticket
->t_ref
) > 0);
3220 atomic_inc(&ticket
->t_ref
);
3225 * Allocate and initialise a new log ticket.
3227 STATIC xlog_ticket_t
*
3228 xlog_ticket_alloc(xlog_t
*log
,
3237 tic
= kmem_zone_zalloc(xfs_log_ticket_zone
, KM_SLEEP
|KM_MAYFAIL
);
3242 * Permanent reservations have up to 'cnt'-1 active log operations
3243 * in the log. A unit in this case is the amount of space for one
3244 * of these log operations. Normal reservations have a cnt of 1
3245 * and their unit amount is the total amount of space required.
3247 * The following lines of code account for non-transaction data
3248 * which occupy space in the on-disk log.
3250 * Normal form of a transaction is:
3251 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3252 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3254 * We need to account for all the leadup data and trailer data
3255 * around the transaction data.
3256 * And then we need to account for the worst case in terms of using
3258 * The worst case will happen if:
3259 * - the placement of the transaction happens to be such that the
3260 * roundoff is at its maximum
3261 * - the transaction data is synced before the commit record is synced
3262 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3263 * Therefore the commit record is in its own Log Record.
3264 * This can happen as the commit record is called with its
3265 * own region to xlog_write().
3266 * This then means that in the worst case, roundoff can happen for
3267 * the commit-rec as well.
3268 * The commit-rec is smaller than padding in this scenario and so it is
3269 * not added separately.
3272 /* for trans header */
3273 unit_bytes
+= sizeof(xlog_op_header_t
);
3274 unit_bytes
+= sizeof(xfs_trans_header_t
);
3277 unit_bytes
+= sizeof(xlog_op_header_t
);
3279 /* for LR headers */
3280 num_headers
= ((unit_bytes
+ log
->l_iclog_size
-1) >> log
->l_iclog_size_log
);
3281 unit_bytes
+= log
->l_iclog_hsize
* num_headers
;
3283 /* for commit-rec LR header - note: padding will subsume the ophdr */
3284 unit_bytes
+= log
->l_iclog_hsize
;
3286 /* for split-recs - ophdrs added when data split over LRs */
3287 unit_bytes
+= sizeof(xlog_op_header_t
) * num_headers
;
3289 /* for roundoff padding for transaction data and one for commit record */
3290 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
3291 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
3292 /* log su roundoff */
3293 unit_bytes
+= 2*log
->l_mp
->m_sb
.sb_logsunit
;
3296 unit_bytes
+= 2*BBSIZE
;
3299 atomic_set(&tic
->t_ref
, 1);
3300 tic
->t_unit_res
= unit_bytes
;
3301 tic
->t_curr_res
= unit_bytes
;
3304 tic
->t_tid
= (xlog_tid_t
)((__psint_t
)tic
& 0xffffffff);
3305 tic
->t_clientid
= client
;
3306 tic
->t_flags
= XLOG_TIC_INITED
;
3307 tic
->t_trans_type
= 0;
3308 if (xflags
& XFS_LOG_PERM_RESERV
)
3309 tic
->t_flags
|= XLOG_TIC_PERM_RESERV
;
3310 sv_init(&(tic
->t_wait
), SV_DEFAULT
, "logtick");
3312 xlog_tic_reset_res(tic
);
3318 /******************************************************************************
3320 * Log debug routines
3322 ******************************************************************************
3326 * Make sure that the destination ptr is within the valid data region of
3327 * one of the iclogs. This uses backup pointers stored in a different
3328 * part of the log in case we trash the log structure.
3331 xlog_verify_dest_ptr(xlog_t
*log
,
3337 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3338 if (ptr
>= (__psint_t
)log
->l_iclog_bak
[i
] &&
3339 ptr
<= (__psint_t
)log
->l_iclog_bak
[i
]+log
->l_iclog_size
)
3343 xlog_panic("xlog_verify_dest_ptr: invalid ptr");
3344 } /* xlog_verify_dest_ptr */
3347 xlog_verify_grant_head(xlog_t
*log
, int equals
)
3349 if (log
->l_grant_reserve_cycle
== log
->l_grant_write_cycle
) {
3351 ASSERT(log
->l_grant_reserve_bytes
>= log
->l_grant_write_bytes
);
3353 ASSERT(log
->l_grant_reserve_bytes
> log
->l_grant_write_bytes
);
3355 ASSERT(log
->l_grant_reserve_cycle
-1 == log
->l_grant_write_cycle
);
3356 ASSERT(log
->l_grant_write_bytes
>= log
->l_grant_reserve_bytes
);
3358 } /* xlog_verify_grant_head */
3360 /* check if it will fit */
3362 xlog_verify_tail_lsn(xlog_t
*log
,
3363 xlog_in_core_t
*iclog
,
3368 if (CYCLE_LSN(tail_lsn
) == log
->l_prev_cycle
) {
3370 log
->l_logBBsize
- (log
->l_prev_block
- BLOCK_LSN(tail_lsn
));
3371 if (blocks
< BTOBB(iclog
->ic_offset
)+BTOBB(log
->l_iclog_hsize
))
3372 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3374 ASSERT(CYCLE_LSN(tail_lsn
)+1 == log
->l_prev_cycle
);
3376 if (BLOCK_LSN(tail_lsn
) == log
->l_prev_block
)
3377 xlog_panic("xlog_verify_tail_lsn: tail wrapped");
3379 blocks
= BLOCK_LSN(tail_lsn
) - log
->l_prev_block
;
3380 if (blocks
< BTOBB(iclog
->ic_offset
) + 1)
3381 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3383 } /* xlog_verify_tail_lsn */
3386 * Perform a number of checks on the iclog before writing to disk.
3388 * 1. Make sure the iclogs are still circular
3389 * 2. Make sure we have a good magic number
3390 * 3. Make sure we don't have magic numbers in the data
3391 * 4. Check fields of each log operation header for:
3392 * A. Valid client identifier
3393 * B. tid ptr value falls in valid ptr space (user space code)
3394 * C. Length in log record header is correct according to the
3395 * individual operation headers within record.
3396 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3397 * log, check the preceding blocks of the physical log to make sure all
3398 * the cycle numbers agree with the current cycle number.
3401 xlog_verify_iclog(xlog_t
*log
,
3402 xlog_in_core_t
*iclog
,
3406 xlog_op_header_t
*ophead
;
3407 xlog_in_core_t
*icptr
;
3408 xlog_in_core_2_t
*xhdr
;
3410 xfs_caddr_t base_ptr
;
3411 __psint_t field_offset
;
3413 int len
, i
, j
, k
, op_len
;
3416 /* check validity of iclog pointers */
3417 spin_lock(&log
->l_icloglock
);
3418 icptr
= log
->l_iclog
;
3419 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3421 xlog_panic("xlog_verify_iclog: invalid ptr");
3422 icptr
= icptr
->ic_next
;
3424 if (icptr
!= log
->l_iclog
)
3425 xlog_panic("xlog_verify_iclog: corrupt iclog ring");
3426 spin_unlock(&log
->l_icloglock
);
3428 /* check log magic numbers */
3429 if (be32_to_cpu(iclog
->ic_header
.h_magicno
) != XLOG_HEADER_MAGIC_NUM
)
3430 xlog_panic("xlog_verify_iclog: invalid magic num");
3432 ptr
= (xfs_caddr_t
) &iclog
->ic_header
;
3433 for (ptr
+= BBSIZE
; ptr
< ((xfs_caddr_t
)&iclog
->ic_header
) + count
;
3435 if (be32_to_cpu(*(__be32
*)ptr
) == XLOG_HEADER_MAGIC_NUM
)
3436 xlog_panic("xlog_verify_iclog: unexpected magic num");
3440 len
= be32_to_cpu(iclog
->ic_header
.h_num_logops
);
3441 ptr
= iclog
->ic_datap
;
3443 ophead
= (xlog_op_header_t
*)ptr
;
3444 xhdr
= iclog
->ic_data
;
3445 for (i
= 0; i
< len
; i
++) {
3446 ophead
= (xlog_op_header_t
*)ptr
;
3448 /* clientid is only 1 byte */
3449 field_offset
= (__psint_t
)
3450 ((xfs_caddr_t
)&(ophead
->oh_clientid
) - base_ptr
);
3451 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3452 clientid
= ophead
->oh_clientid
;
3454 idx
= BTOBBT((xfs_caddr_t
)&(ophead
->oh_clientid
) - iclog
->ic_datap
);
3455 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3456 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3457 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3458 clientid
= xlog_get_client_id(
3459 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3461 clientid
= xlog_get_client_id(
3462 iclog
->ic_header
.h_cycle_data
[idx
]);
3465 if (clientid
!= XFS_TRANSACTION
&& clientid
!= XFS_LOG
)
3466 cmn_err(CE_WARN
, "xlog_verify_iclog: "
3467 "invalid clientid %d op 0x%p offset 0x%lx",
3468 clientid
, ophead
, (unsigned long)field_offset
);
3471 field_offset
= (__psint_t
)
3472 ((xfs_caddr_t
)&(ophead
->oh_len
) - base_ptr
);
3473 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3474 op_len
= be32_to_cpu(ophead
->oh_len
);
3476 idx
= BTOBBT((__psint_t
)&ophead
->oh_len
-
3477 (__psint_t
)iclog
->ic_datap
);
3478 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3479 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3480 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3481 op_len
= be32_to_cpu(xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3483 op_len
= be32_to_cpu(iclog
->ic_header
.h_cycle_data
[idx
]);
3486 ptr
+= sizeof(xlog_op_header_t
) + op_len
;
3488 } /* xlog_verify_iclog */
3492 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
3498 xlog_in_core_t
*iclog
, *ic
;
3500 iclog
= log
->l_iclog
;
3501 if (! (iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3503 * Mark all the incore logs IOERROR.
3504 * From now on, no log flushes will result.
3508 ic
->ic_state
= XLOG_STATE_IOERROR
;
3510 } while (ic
!= iclog
);
3514 * Return non-zero, if state transition has already happened.
3520 * This is called from xfs_force_shutdown, when we're forcibly
3521 * shutting down the filesystem, typically because of an IO error.
3522 * Our main objectives here are to make sure that:
3523 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3524 * parties to find out, 'atomically'.
3525 * b. those who're sleeping on log reservations, pinned objects and
3526 * other resources get woken up, and be told the bad news.
3527 * c. nothing new gets queued up after (a) and (b) are done.
3528 * d. if !logerror, flush the iclogs to disk, then seal them off
3532 xfs_log_force_umount(
3533 struct xfs_mount
*mp
,
3544 * If this happens during log recovery, don't worry about
3545 * locking; the log isn't open for business yet.
3548 log
->l_flags
& XLOG_ACTIVE_RECOVERY
) {
3549 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3551 XFS_BUF_DONE(mp
->m_sb_bp
);
3556 * Somebody could've already done the hard work for us.
3557 * No need to get locks for this.
3559 if (logerror
&& log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3560 ASSERT(XLOG_FORCED_SHUTDOWN(log
));
3565 * We must hold both the GRANT lock and the LOG lock,
3566 * before we mark the filesystem SHUTDOWN and wake
3567 * everybody up to tell the bad news.
3569 spin_lock(&log
->l_icloglock
);
3570 spin_lock(&log
->l_grant_lock
);
3571 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3573 XFS_BUF_DONE(mp
->m_sb_bp
);
3576 * This flag is sort of redundant because of the mount flag, but
3577 * it's good to maintain the separation between the log and the rest
3580 log
->l_flags
|= XLOG_IO_ERROR
;
3583 * If we hit a log error, we want to mark all the iclogs IOERROR
3584 * while we're still holding the loglock.
3587 retval
= xlog_state_ioerror(log
);
3588 spin_unlock(&log
->l_icloglock
);
3591 * We don't want anybody waiting for log reservations
3592 * after this. That means we have to wake up everybody
3593 * queued up on reserve_headq as well as write_headq.
3594 * In addition, we make sure in xlog_{re}grant_log_space
3595 * that we don't enqueue anything once the SHUTDOWN flag
3596 * is set, and this action is protected by the GRANTLOCK.
3598 if ((tic
= log
->l_reserve_headq
)) {
3600 sv_signal(&tic
->t_wait
);
3602 } while (tic
!= log
->l_reserve_headq
);
3605 if ((tic
= log
->l_write_headq
)) {
3607 sv_signal(&tic
->t_wait
);
3609 } while (tic
!= log
->l_write_headq
);
3611 spin_unlock(&log
->l_grant_lock
);
3613 if (! (log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3616 * Force the incore logs to disk before shutting the
3617 * log down completely.
3619 xlog_state_sync_all(log
, XFS_LOG_FORCE
|XFS_LOG_SYNC
, &dummy
);
3620 spin_lock(&log
->l_icloglock
);
3621 retval
= xlog_state_ioerror(log
);
3622 spin_unlock(&log
->l_icloglock
);
3625 * Wake up everybody waiting on xfs_log_force.
3626 * Callback all log item committed functions as if the
3627 * log writes were completed.
3629 xlog_state_do_callback(log
, XFS_LI_ABORTED
, NULL
);
3631 #ifdef XFSERRORDEBUG
3633 xlog_in_core_t
*iclog
;
3635 spin_lock(&log
->l_icloglock
);
3636 iclog
= log
->l_iclog
;
3638 ASSERT(iclog
->ic_callback
== 0);
3639 iclog
= iclog
->ic_next
;
3640 } while (iclog
!= log
->l_iclog
);
3641 spin_unlock(&log
->l_icloglock
);
3644 /* return non-zero if log IOERROR transition had already happened */
3649 xlog_iclogs_empty(xlog_t
*log
)
3651 xlog_in_core_t
*iclog
;
3653 iclog
= log
->l_iclog
;
3655 /* endianness does not matter here, zero is zero in
3658 if (iclog
->ic_header
.h_num_logops
)
3660 iclog
= iclog
->ic_next
;
3661 } while (iclog
!= log
->l_iclog
);
This page took 0.103807 seconds and 5 git commands to generate.