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_get(xlog_t
*log
,
108 STATIC
void xlog_ticket_put(xlog_t
*log
, xlog_ticket_t
*ticket
);
111 STATIC
void xlog_verify_dest_ptr(xlog_t
*log
, __psint_t ptr
);
112 STATIC
void xlog_verify_grant_head(xlog_t
*log
, int equals
);
113 STATIC
void xlog_verify_iclog(xlog_t
*log
, xlog_in_core_t
*iclog
,
114 int count
, boolean_t syncing
);
115 STATIC
void xlog_verify_tail_lsn(xlog_t
*log
, xlog_in_core_t
*iclog
,
118 #define xlog_verify_dest_ptr(a,b)
119 #define xlog_verify_grant_head(a,b)
120 #define xlog_verify_iclog(a,b,c,d)
121 #define xlog_verify_tail_lsn(a,b,c)
124 STATIC
int xlog_iclogs_empty(xlog_t
*log
);
126 #if defined(XFS_LOG_TRACE)
128 #define XLOG_TRACE_LOGGRANT_SIZE 2048
129 #define XLOG_TRACE_ICLOG_SIZE 256
132 xlog_trace_loggrant_alloc(xlog_t
*log
)
134 log
->l_grant_trace
= ktrace_alloc(XLOG_TRACE_LOGGRANT_SIZE
, KM_NOFS
);
138 xlog_trace_loggrant_dealloc(xlog_t
*log
)
140 ktrace_free(log
->l_grant_trace
);
144 xlog_trace_loggrant(xlog_t
*log
, xlog_ticket_t
*tic
, xfs_caddr_t string
)
148 /* ticket counts are 1 byte each */
149 cnts
= ((unsigned long)tic
->t_ocnt
) | ((unsigned long)tic
->t_cnt
) << 8;
151 ktrace_enter(log
->l_grant_trace
,
153 (void *)log
->l_reserve_headq
,
154 (void *)log
->l_write_headq
,
155 (void *)((unsigned long)log
->l_grant_reserve_cycle
),
156 (void *)((unsigned long)log
->l_grant_reserve_bytes
),
157 (void *)((unsigned long)log
->l_grant_write_cycle
),
158 (void *)((unsigned long)log
->l_grant_write_bytes
),
159 (void *)((unsigned long)log
->l_curr_cycle
),
160 (void *)((unsigned long)log
->l_curr_block
),
161 (void *)((unsigned long)CYCLE_LSN(log
->l_tail_lsn
)),
162 (void *)((unsigned long)BLOCK_LSN(log
->l_tail_lsn
)),
164 (void *)((unsigned long)tic
->t_trans_type
),
166 (void *)((unsigned long)tic
->t_curr_res
),
167 (void *)((unsigned long)tic
->t_unit_res
));
171 xlog_trace_iclog_alloc(xlog_in_core_t
*iclog
)
173 iclog
->ic_trace
= ktrace_alloc(XLOG_TRACE_ICLOG_SIZE
, KM_NOFS
);
177 xlog_trace_iclog_dealloc(xlog_in_core_t
*iclog
)
179 ktrace_free(iclog
->ic_trace
);
183 xlog_trace_iclog(xlog_in_core_t
*iclog
, uint state
)
185 ktrace_enter(iclog
->ic_trace
,
186 (void *)((unsigned long)state
),
187 (void *)((unsigned long)current_pid()),
188 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
189 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
190 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
191 (void *)NULL
, (void *)NULL
);
195 #define xlog_trace_loggrant_alloc(log)
196 #define xlog_trace_loggrant_dealloc(log)
197 #define xlog_trace_loggrant(log,tic,string)
199 #define xlog_trace_iclog_alloc(iclog)
200 #define xlog_trace_iclog_dealloc(iclog)
201 #define xlog_trace_iclog(iclog,state)
203 #endif /* XFS_LOG_TRACE */
207 xlog_ins_ticketq(struct xlog_ticket
**qp
, struct xlog_ticket
*tic
)
211 tic
->t_prev
= (*qp
)->t_prev
;
212 (*qp
)->t_prev
->t_next
= tic
;
215 tic
->t_prev
= tic
->t_next
= tic
;
219 tic
->t_flags
|= XLOG_TIC_IN_Q
;
223 xlog_del_ticketq(struct xlog_ticket
**qp
, struct xlog_ticket
*tic
)
225 if (tic
== tic
->t_next
) {
229 tic
->t_next
->t_prev
= tic
->t_prev
;
230 tic
->t_prev
->t_next
= tic
->t_next
;
233 tic
->t_next
= tic
->t_prev
= NULL
;
234 tic
->t_flags
&= ~XLOG_TIC_IN_Q
;
238 xlog_grant_sub_space(struct log
*log
, int bytes
)
240 log
->l_grant_write_bytes
-= bytes
;
241 if (log
->l_grant_write_bytes
< 0) {
242 log
->l_grant_write_bytes
+= log
->l_logsize
;
243 log
->l_grant_write_cycle
--;
246 log
->l_grant_reserve_bytes
-= bytes
;
247 if ((log
)->l_grant_reserve_bytes
< 0) {
248 log
->l_grant_reserve_bytes
+= log
->l_logsize
;
249 log
->l_grant_reserve_cycle
--;
255 xlog_grant_add_space_write(struct log
*log
, int bytes
)
257 int tmp
= log
->l_logsize
- log
->l_grant_write_bytes
;
259 log
->l_grant_write_bytes
+= bytes
;
261 log
->l_grant_write_cycle
++;
262 log
->l_grant_write_bytes
= bytes
- tmp
;
267 xlog_grant_add_space_reserve(struct log
*log
, int bytes
)
269 int tmp
= log
->l_logsize
- log
->l_grant_reserve_bytes
;
271 log
->l_grant_reserve_bytes
+= bytes
;
273 log
->l_grant_reserve_cycle
++;
274 log
->l_grant_reserve_bytes
= bytes
- tmp
;
279 xlog_grant_add_space(struct log
*log
, int bytes
)
281 xlog_grant_add_space_write(log
, bytes
);
282 xlog_grant_add_space_reserve(log
, bytes
);
286 xlog_tic_reset_res(xlog_ticket_t
*tic
)
289 tic
->t_res_arr_sum
= 0;
290 tic
->t_res_num_ophdrs
= 0;
294 xlog_tic_add_region(xlog_ticket_t
*tic
, uint len
, uint type
)
296 if (tic
->t_res_num
== XLOG_TIC_LEN_MAX
) {
297 /* add to overflow and start again */
298 tic
->t_res_o_flow
+= tic
->t_res_arr_sum
;
300 tic
->t_res_arr_sum
= 0;
303 tic
->t_res_arr
[tic
->t_res_num
].r_len
= len
;
304 tic
->t_res_arr
[tic
->t_res_num
].r_type
= type
;
305 tic
->t_res_arr_sum
+= len
;
312 * 1. currblock field gets updated at startup and after in-core logs
313 * marked as with WANT_SYNC.
317 * This routine is called when a user of a log manager ticket is done with
318 * the reservation. If the ticket was ever used, then a commit record for
319 * the associated transaction is written out as a log operation header with
320 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
321 * a given ticket. If the ticket was one with a permanent reservation, then
322 * a few operations are done differently. Permanent reservation tickets by
323 * default don't release the reservation. They just commit the current
324 * transaction with the belief that the reservation is still needed. A flag
325 * must be passed in before permanent reservations are actually released.
326 * When these type of tickets are not released, they need to be set into
327 * the inited state again. By doing this, a start record will be written
328 * out when the next write occurs.
331 xfs_log_done(xfs_mount_t
*mp
,
332 xfs_log_ticket_t xtic
,
336 xlog_t
*log
= mp
->m_log
;
337 xlog_ticket_t
*ticket
= (xfs_log_ticket_t
) xtic
;
340 if (XLOG_FORCED_SHUTDOWN(log
) ||
342 * If nothing was ever written, don't write out commit record.
343 * If we get an error, just continue and give back the log ticket.
345 (((ticket
->t_flags
& XLOG_TIC_INITED
) == 0) &&
346 (xlog_commit_record(mp
, ticket
,
347 (xlog_in_core_t
**)iclog
, &lsn
)))) {
348 lsn
= (xfs_lsn_t
) -1;
349 if (ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) {
350 flags
|= XFS_LOG_REL_PERM_RESERV
;
355 if ((ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) == 0 ||
356 (flags
& XFS_LOG_REL_PERM_RESERV
)) {
358 * Release ticket if not permanent reservation or a specific
359 * request has been made to release a permanent reservation.
361 xlog_trace_loggrant(log
, ticket
, "xfs_log_done: (non-permanent)");
362 xlog_ungrant_log_space(log
, ticket
);
363 xlog_ticket_put(log
, ticket
);
365 xlog_trace_loggrant(log
, ticket
, "xfs_log_done: (permanent)");
366 xlog_regrant_reserve_log_space(log
, ticket
);
367 /* If this ticket was a permanent reservation and we aren't
368 * trying to release it, reset the inited flags; so next time
369 * we write, a start record will be written out.
371 ticket
->t_flags
|= XLOG_TIC_INITED
;
379 * Force the in-core log to disk. If flags == XFS_LOG_SYNC,
380 * the force is done synchronously.
382 * Asynchronous forces are implemented by setting the WANT_SYNC
383 * bit in the appropriate in-core log and then returning.
385 * Synchronous forces are implemented with a signal variable. All callers
386 * to force a given lsn to disk will wait on a the sv attached to the
387 * specific in-core log. When given in-core log finally completes its
388 * write to disk, that thread will wake up all threads waiting on the
398 xlog_t
*log
= mp
->m_log
;
402 log_flushed
= &dummy
;
404 ASSERT(flags
& XFS_LOG_FORCE
);
406 XFS_STATS_INC(xs_log_force
);
408 if (log
->l_flags
& XLOG_IO_ERROR
)
409 return XFS_ERROR(EIO
);
411 return xlog_state_sync_all(log
, flags
, log_flushed
);
413 return xlog_state_sync(log
, lsn
, flags
, log_flushed
);
414 } /* _xfs_log_force */
417 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
418 * about errors or whether the log was flushed or not. This is the normal
419 * interface to use when trying to unpin items or move the log forward.
428 error
= _xfs_log_force(mp
, lsn
, flags
, NULL
);
430 xfs_fs_cmn_err(CE_WARN
, mp
, "xfs_log_force: "
431 "error %d returned.", error
);
437 * Attaches a new iclog I/O completion callback routine during
438 * transaction commit. If the log is in error state, a non-zero
439 * return code is handed back and the caller is responsible for
440 * executing the callback at an appropriate time.
443 xfs_log_notify(xfs_mount_t
*mp
, /* mount of partition */
444 void *iclog_hndl
, /* iclog to hang callback off */
445 xfs_log_callback_t
*cb
)
447 xlog_in_core_t
*iclog
= (xlog_in_core_t
*)iclog_hndl
;
450 spin_lock(&iclog
->ic_callback_lock
);
451 abortflg
= (iclog
->ic_state
& XLOG_STATE_IOERROR
);
453 ASSERT_ALWAYS((iclog
->ic_state
== XLOG_STATE_ACTIVE
) ||
454 (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
));
456 *(iclog
->ic_callback_tail
) = cb
;
457 iclog
->ic_callback_tail
= &(cb
->cb_next
);
459 spin_unlock(&iclog
->ic_callback_lock
);
461 } /* xfs_log_notify */
464 xfs_log_release_iclog(xfs_mount_t
*mp
,
467 xlog_t
*log
= mp
->m_log
;
468 xlog_in_core_t
*iclog
= (xlog_in_core_t
*)iclog_hndl
;
470 if (xlog_state_release_iclog(log
, iclog
)) {
471 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
479 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
480 * to the reservation.
481 * 2. Potentially, push buffers at tail of log to disk.
483 * Each reservation is going to reserve extra space for a log record header.
484 * When writes happen to the on-disk log, we don't subtract the length of the
485 * log record header from any reservation. By wasting space in each
486 * reservation, we prevent over allocation problems.
489 xfs_log_reserve(xfs_mount_t
*mp
,
492 xfs_log_ticket_t
*ticket
,
497 xlog_t
*log
= mp
->m_log
;
498 xlog_ticket_t
*internal_ticket
;
501 ASSERT(client
== XFS_TRANSACTION
|| client
== XFS_LOG
);
502 ASSERT((flags
& XFS_LOG_NOSLEEP
) == 0);
504 if (XLOG_FORCED_SHUTDOWN(log
))
505 return XFS_ERROR(EIO
);
507 XFS_STATS_INC(xs_try_logspace
);
509 if (*ticket
!= NULL
) {
510 ASSERT(flags
& XFS_LOG_PERM_RESERV
);
511 internal_ticket
= (xlog_ticket_t
*)*ticket
;
512 xlog_trace_loggrant(log
, internal_ticket
, "xfs_log_reserve: existing ticket (permanent trans)");
513 xlog_grant_push_ail(mp
, internal_ticket
->t_unit_res
);
514 retval
= xlog_regrant_write_log_space(log
, internal_ticket
);
516 /* may sleep if need to allocate more tickets */
517 internal_ticket
= xlog_ticket_get(log
, unit_bytes
, cnt
,
519 if (!internal_ticket
)
520 return XFS_ERROR(ENOMEM
);
521 internal_ticket
->t_trans_type
= t_type
;
522 *ticket
= internal_ticket
;
523 xlog_trace_loggrant(log
, internal_ticket
,
524 (internal_ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) ?
525 "xfs_log_reserve: create new ticket (permanent trans)" :
526 "xfs_log_reserve: create new ticket");
527 xlog_grant_push_ail(mp
,
528 (internal_ticket
->t_unit_res
*
529 internal_ticket
->t_cnt
));
530 retval
= xlog_grant_log_space(log
, internal_ticket
);
534 } /* xfs_log_reserve */
538 * Mount a log filesystem
540 * mp - ubiquitous xfs mount point structure
541 * log_target - buftarg of on-disk log device
542 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
543 * num_bblocks - Number of BBSIZE blocks in on-disk log
545 * Return error or zero.
550 xfs_buftarg_t
*log_target
,
551 xfs_daddr_t blk_offset
,
556 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
557 cmn_err(CE_NOTE
, "XFS mounting filesystem %s", mp
->m_fsname
);
560 "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.",
562 ASSERT(mp
->m_flags
& XFS_MOUNT_RDONLY
);
565 mp
->m_log
= xlog_alloc_log(mp
, log_target
, blk_offset
, num_bblks
);
567 cmn_err(CE_WARN
, "XFS: Log allocation failed: No memory!");
573 * Initialize the AIL now we have a log.
575 spin_lock_init(&mp
->m_ail_lock
);
576 error
= xfs_trans_ail_init(mp
);
578 cmn_err(CE_WARN
, "XFS: AIL initialisation failed: error %d", error
);
583 * skip log recovery on a norecovery mount. pretend it all
586 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
)) {
587 int readonly
= (mp
->m_flags
& XFS_MOUNT_RDONLY
);
590 mp
->m_flags
&= ~XFS_MOUNT_RDONLY
;
592 error
= xlog_recover(mp
->m_log
);
595 mp
->m_flags
|= XFS_MOUNT_RDONLY
;
597 cmn_err(CE_WARN
, "XFS: log mount/recovery failed: error %d", error
);
602 /* Normal transactions can now occur */
603 mp
->m_log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
605 /* End mounting message in xfs_log_mount_finish */
608 xfs_log_unmount_dealloc(mp
);
611 } /* xfs_log_mount */
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 * Unmount processing for the log.
640 xfs_log_unmount(xfs_mount_t
*mp
)
644 error
= xfs_log_unmount_write(mp
);
645 xfs_log_unmount_dealloc(mp
);
650 * Final log writes as part of unmount.
652 * Mark the filesystem clean as unmount happens. Note that during relocation
653 * this routine needs to be executed as part of source-bag while the
654 * deallocation must not be done until source-end.
658 * Unmount record used to have a string "Unmount filesystem--" in the
659 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
660 * We just write the magic number now since that particular field isn't
661 * currently architecture converted and "nUmount" is a bit foo.
662 * As far as I know, there weren't any dependencies on the old behaviour.
666 xfs_log_unmount_write(xfs_mount_t
*mp
)
668 xlog_t
*log
= mp
->m_log
;
669 xlog_in_core_t
*iclog
;
671 xlog_in_core_t
*first_iclog
;
673 xfs_log_iovec_t reg
[1];
674 xfs_log_ticket_t tic
= NULL
;
678 /* the data section must be 32 bit size aligned */
682 __uint32_t pad2
; /* may as well make it 64 bits */
683 } magic
= { XLOG_UNMOUNT_TYPE
, 0, 0 };
686 * Don't write out unmount record on read-only mounts.
687 * Or, if we are doing a forced umount (typically because of IO errors).
689 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
692 error
= _xfs_log_force(mp
, 0, XFS_LOG_FORCE
|XFS_LOG_SYNC
, NULL
);
693 ASSERT(error
|| !(XLOG_FORCED_SHUTDOWN(log
)));
696 first_iclog
= iclog
= log
->l_iclog
;
698 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
699 ASSERT(iclog
->ic_state
& XLOG_STATE_ACTIVE
);
700 ASSERT(iclog
->ic_offset
== 0);
702 iclog
= iclog
->ic_next
;
703 } while (iclog
!= first_iclog
);
705 if (! (XLOG_FORCED_SHUTDOWN(log
))) {
706 reg
[0].i_addr
= (void*)&magic
;
707 reg
[0].i_len
= sizeof(magic
);
708 XLOG_VEC_SET_TYPE(®
[0], XLOG_REG_TYPE_UNMOUNT
);
710 error
= xfs_log_reserve(mp
, 600, 1, &tic
,
711 XFS_LOG
, 0, XLOG_UNMOUNT_REC_TYPE
);
713 /* remove inited flag */
714 ((xlog_ticket_t
*)tic
)->t_flags
= 0;
715 error
= xlog_write(mp
, reg
, 1, tic
, &lsn
,
716 NULL
, XLOG_UNMOUNT_TRANS
);
718 * At this point, we're umounting anyway,
719 * so there's no point in transitioning log state
720 * to IOERROR. Just continue...
725 xfs_fs_cmn_err(CE_ALERT
, mp
,
726 "xfs_log_unmount: unmount record failed");
730 spin_lock(&log
->l_icloglock
);
731 iclog
= log
->l_iclog
;
732 atomic_inc(&iclog
->ic_refcnt
);
733 spin_unlock(&log
->l_icloglock
);
734 xlog_state_want_sync(log
, iclog
);
735 error
= xlog_state_release_iclog(log
, iclog
);
737 spin_lock(&log
->l_icloglock
);
738 if (!(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
739 iclog
->ic_state
== XLOG_STATE_DIRTY
)) {
740 if (!XLOG_FORCED_SHUTDOWN(log
)) {
741 sv_wait(&iclog
->ic_force_wait
, PMEM
,
742 &log
->l_icloglock
, s
);
744 spin_unlock(&log
->l_icloglock
);
747 spin_unlock(&log
->l_icloglock
);
750 xlog_trace_loggrant(log
, tic
, "unmount rec");
751 xlog_ungrant_log_space(log
, tic
);
752 xlog_ticket_put(log
, tic
);
756 * We're already in forced_shutdown mode, couldn't
757 * even attempt to write out the unmount transaction.
759 * Go through the motions of sync'ing and releasing
760 * the iclog, even though no I/O will actually happen,
761 * we need to wait for other log I/Os that may already
762 * be in progress. Do this as a separate section of
763 * code so we'll know if we ever get stuck here that
764 * we're in this odd situation of trying to unmount
765 * a file system that went into forced_shutdown as
766 * the result of an unmount..
768 spin_lock(&log
->l_icloglock
);
769 iclog
= log
->l_iclog
;
770 atomic_inc(&iclog
->ic_refcnt
);
771 spin_unlock(&log
->l_icloglock
);
773 xlog_state_want_sync(log
, iclog
);
774 error
= xlog_state_release_iclog(log
, iclog
);
776 spin_lock(&log
->l_icloglock
);
778 if ( ! ( iclog
->ic_state
== XLOG_STATE_ACTIVE
779 || iclog
->ic_state
== XLOG_STATE_DIRTY
780 || iclog
->ic_state
== XLOG_STATE_IOERROR
) ) {
782 sv_wait(&iclog
->ic_force_wait
, PMEM
,
783 &log
->l_icloglock
, s
);
785 spin_unlock(&log
->l_icloglock
);
790 } /* xfs_log_unmount_write */
793 * Deallocate log structures for unmount/relocation.
795 * We need to stop the aild from running before we destroy
796 * and deallocate the log as the aild references the log.
799 xfs_log_unmount_dealloc(xfs_mount_t
*mp
)
801 xfs_trans_ail_destroy(mp
);
802 xlog_dealloc_log(mp
->m_log
);
806 * Write region vectors to log. The write happens using the space reservation
807 * of the ticket (tic). It is not a requirement that all writes for a given
808 * transaction occur with one call to xfs_log_write().
811 xfs_log_write(xfs_mount_t
* mp
,
812 xfs_log_iovec_t reg
[],
814 xfs_log_ticket_t tic
,
815 xfs_lsn_t
*start_lsn
)
818 xlog_t
*log
= mp
->m_log
;
820 if (XLOG_FORCED_SHUTDOWN(log
))
821 return XFS_ERROR(EIO
);
823 if ((error
= xlog_write(mp
, reg
, nentries
, tic
, start_lsn
, NULL
, 0))) {
824 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
827 } /* xfs_log_write */
831 xfs_log_move_tail(xfs_mount_t
*mp
,
835 xlog_t
*log
= mp
->m_log
;
836 int need_bytes
, free_bytes
, cycle
, bytes
;
838 if (XLOG_FORCED_SHUTDOWN(log
))
842 /* needed since sync_lsn is 64 bits */
843 spin_lock(&log
->l_icloglock
);
844 tail_lsn
= log
->l_last_sync_lsn
;
845 spin_unlock(&log
->l_icloglock
);
848 spin_lock(&log
->l_grant_lock
);
850 /* Also an invalid lsn. 1 implies that we aren't passing in a valid
854 log
->l_tail_lsn
= tail_lsn
;
857 if ((tic
= log
->l_write_headq
)) {
859 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
860 panic("Recovery problem");
862 cycle
= log
->l_grant_write_cycle
;
863 bytes
= log
->l_grant_write_bytes
;
864 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
866 ASSERT(tic
->t_flags
& XLOG_TIC_PERM_RESERV
);
868 if (free_bytes
< tic
->t_unit_res
&& tail_lsn
!= 1)
871 free_bytes
-= tic
->t_unit_res
;
872 sv_signal(&tic
->t_wait
);
874 } while (tic
!= log
->l_write_headq
);
876 if ((tic
= log
->l_reserve_headq
)) {
878 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
879 panic("Recovery problem");
881 cycle
= log
->l_grant_reserve_cycle
;
882 bytes
= log
->l_grant_reserve_bytes
;
883 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
885 if (tic
->t_flags
& XLOG_TIC_PERM_RESERV
)
886 need_bytes
= tic
->t_unit_res
*tic
->t_cnt
;
888 need_bytes
= tic
->t_unit_res
;
889 if (free_bytes
< need_bytes
&& tail_lsn
!= 1)
892 free_bytes
-= need_bytes
;
893 sv_signal(&tic
->t_wait
);
895 } while (tic
!= log
->l_reserve_headq
);
897 spin_unlock(&log
->l_grant_lock
);
898 } /* xfs_log_move_tail */
901 * Determine if we have a transaction that has gone to disk
902 * that needs to be covered. Log activity needs to be idle (no AIL and
903 * nothing in the iclogs). And, we need to be in the right state indicating
904 * something has gone out.
907 xfs_log_need_covered(xfs_mount_t
*mp
)
910 xlog_t
*log
= mp
->m_log
;
912 if (!xfs_fs_writable(mp
))
915 spin_lock(&log
->l_icloglock
);
916 if (((log
->l_covered_state
== XLOG_STATE_COVER_NEED
) ||
917 (log
->l_covered_state
== XLOG_STATE_COVER_NEED2
))
918 && !xfs_trans_first_ail(mp
, &gen
)
919 && xlog_iclogs_empty(log
)) {
920 if (log
->l_covered_state
== XLOG_STATE_COVER_NEED
)
921 log
->l_covered_state
= XLOG_STATE_COVER_DONE
;
923 ASSERT(log
->l_covered_state
== XLOG_STATE_COVER_NEED2
);
924 log
->l_covered_state
= XLOG_STATE_COVER_DONE2
;
928 spin_unlock(&log
->l_icloglock
);
932 /******************************************************************************
936 ******************************************************************************
939 /* xfs_trans_tail_ail returns 0 when there is nothing in the list.
940 * The log manager must keep track of the last LR which was committed
941 * to disk. The lsn of this LR will become the new tail_lsn whenever
942 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
943 * the situation where stuff could be written into the log but nothing
944 * was ever in the AIL when asked. Eventually, we panic since the
945 * tail hits the head.
947 * We may be holding the log iclog lock upon entering this routine.
950 xlog_assign_tail_lsn(xfs_mount_t
*mp
)
953 xlog_t
*log
= mp
->m_log
;
955 tail_lsn
= xfs_trans_tail_ail(mp
);
956 spin_lock(&log
->l_grant_lock
);
958 log
->l_tail_lsn
= tail_lsn
;
960 tail_lsn
= log
->l_tail_lsn
= log
->l_last_sync_lsn
;
962 spin_unlock(&log
->l_grant_lock
);
965 } /* xlog_assign_tail_lsn */
969 * Return the space in the log between the tail and the head. The head
970 * is passed in the cycle/bytes formal parms. In the special case where
971 * the reserve head has wrapped passed the tail, this calculation is no
972 * longer valid. In this case, just return 0 which means there is no space
973 * in the log. This works for all places where this function is called
974 * with the reserve head. Of course, if the write head were to ever
975 * wrap the tail, we should blow up. Rather than catch this case here,
976 * we depend on other ASSERTions in other parts of the code. XXXmiken
978 * This code also handles the case where the reservation head is behind
979 * the tail. The details of this case are described below, but the end
980 * result is that we return the size of the log as the amount of space left.
983 xlog_space_left(xlog_t
*log
, int cycle
, int bytes
)
989 tail_bytes
= BBTOB(BLOCK_LSN(log
->l_tail_lsn
));
990 tail_cycle
= CYCLE_LSN(log
->l_tail_lsn
);
991 if ((tail_cycle
== cycle
) && (bytes
>= tail_bytes
)) {
992 free_bytes
= log
->l_logsize
- (bytes
- tail_bytes
);
993 } else if ((tail_cycle
+ 1) < cycle
) {
995 } else if (tail_cycle
< cycle
) {
996 ASSERT(tail_cycle
== (cycle
- 1));
997 free_bytes
= tail_bytes
- bytes
;
1000 * The reservation head is behind the tail.
1001 * In this case we just want to return the size of the
1002 * log as the amount of space left.
1004 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
1005 "xlog_space_left: head behind tail\n"
1006 " tail_cycle = %d, tail_bytes = %d\n"
1007 " GH cycle = %d, GH bytes = %d",
1008 tail_cycle
, tail_bytes
, cycle
, bytes
);
1010 free_bytes
= log
->l_logsize
;
1013 } /* xlog_space_left */
1017 * Log function which is called when an io completes.
1019 * The log manager needs its own routine, in order to control what
1020 * happens with the buffer after the write completes.
1023 xlog_iodone(xfs_buf_t
*bp
)
1025 xlog_in_core_t
*iclog
;
1029 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
1030 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long) 2);
1031 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1035 * Some versions of cpp barf on the recursive definition of
1036 * ic_log -> hic_fields.ic_log and expand ic_log twice when
1037 * it is passed through two macros. Workaround broken cpp.
1042 * If the _XFS_BARRIER_FAILED flag was set by a lower
1043 * layer, it means the underlying device no longer supports
1044 * barrier I/O. Warn loudly and turn off barriers.
1046 if (bp
->b_flags
& _XFS_BARRIER_FAILED
) {
1047 bp
->b_flags
&= ~_XFS_BARRIER_FAILED
;
1048 l
->l_mp
->m_flags
&= ~XFS_MOUNT_BARRIER
;
1049 xfs_fs_cmn_err(CE_WARN
, l
->l_mp
,
1050 "xlog_iodone: Barriers are no longer supported"
1051 " by device. Disabling barriers\n");
1052 xfs_buftrace("XLOG_IODONE BARRIERS OFF", bp
);
1056 * Race to shutdown the filesystem if we see an error.
1058 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp
)), l
->l_mp
,
1059 XFS_ERRTAG_IODONE_IOERR
, XFS_RANDOM_IODONE_IOERR
)) {
1060 xfs_ioerror_alert("xlog_iodone", l
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
1062 xfs_force_shutdown(l
->l_mp
, SHUTDOWN_LOG_IO_ERROR
);
1064 * This flag will be propagated to the trans-committed
1065 * callback routines to let them know that the log-commit
1068 aborted
= XFS_LI_ABORTED
;
1069 } else if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
1070 aborted
= XFS_LI_ABORTED
;
1073 /* log I/O is always issued ASYNC */
1074 ASSERT(XFS_BUF_ISASYNC(bp
));
1075 xlog_state_done_syncing(iclog
, aborted
);
1077 * do not reference the buffer (bp) here as we could race
1078 * with it being freed after writing the unmount record to the
1085 * The bdstrat callback function for log bufs. This gives us a central
1086 * place to trap bufs in case we get hit by a log I/O error and need to
1087 * shutdown. Actually, in practice, even when we didn't get a log error,
1088 * we transition the iclogs to IOERROR state *after* flushing all existing
1089 * iclogs to disk. This is because we don't want anymore new transactions to be
1090 * started or completed afterwards.
1093 xlog_bdstrat_cb(struct xfs_buf
*bp
)
1095 xlog_in_core_t
*iclog
;
1097 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
1099 if ((iclog
->ic_state
& XLOG_STATE_IOERROR
) == 0) {
1100 /* note for irix bstrat will need struct bdevsw passed
1101 * Fix the following macro if the code ever is merged
1107 xfs_buftrace("XLOG__BDSTRAT IOERROR", bp
);
1108 XFS_BUF_ERROR(bp
, EIO
);
1111 return XFS_ERROR(EIO
);
1117 * Return size of each in-core log record buffer.
1119 * All machines get 8 x 32KB buffers by default, unless tuned otherwise.
1121 * If the filesystem blocksize is too large, we may need to choose a
1122 * larger size since the directory code currently logs entire blocks.
1126 xlog_get_iclog_buffer_size(xfs_mount_t
*mp
,
1132 if (mp
->m_logbufs
<= 0)
1133 log
->l_iclog_bufs
= XLOG_MAX_ICLOGS
;
1135 log
->l_iclog_bufs
= mp
->m_logbufs
;
1138 * Buffer size passed in from mount system call.
1140 if (mp
->m_logbsize
> 0) {
1141 size
= log
->l_iclog_size
= mp
->m_logbsize
;
1142 log
->l_iclog_size_log
= 0;
1144 log
->l_iclog_size_log
++;
1148 if (xfs_sb_version_haslogv2(&mp
->m_sb
)) {
1149 /* # headers = size / 32K
1150 * one header holds cycles from 32K of data
1153 xhdrs
= mp
->m_logbsize
/ XLOG_HEADER_CYCLE_SIZE
;
1154 if (mp
->m_logbsize
% XLOG_HEADER_CYCLE_SIZE
)
1156 log
->l_iclog_hsize
= xhdrs
<< BBSHIFT
;
1157 log
->l_iclog_heads
= xhdrs
;
1159 ASSERT(mp
->m_logbsize
<= XLOG_BIG_RECORD_BSIZE
);
1160 log
->l_iclog_hsize
= BBSIZE
;
1161 log
->l_iclog_heads
= 1;
1166 /* All machines use 32KB buffers by default. */
1167 log
->l_iclog_size
= XLOG_BIG_RECORD_BSIZE
;
1168 log
->l_iclog_size_log
= XLOG_BIG_RECORD_BSHIFT
;
1170 /* the default log size is 16k or 32k which is one header sector */
1171 log
->l_iclog_hsize
= BBSIZE
;
1172 log
->l_iclog_heads
= 1;
1175 * For 16KB, we use 3 32KB buffers. For 32KB block sizes, we use
1176 * 4 32KB buffers. For 64KB block sizes, we use 8 32KB buffers.
1178 if (mp
->m_sb
.sb_blocksize
>= 16*1024) {
1179 log
->l_iclog_size
= XLOG_BIG_RECORD_BSIZE
;
1180 log
->l_iclog_size_log
= XLOG_BIG_RECORD_BSHIFT
;
1181 if (mp
->m_logbufs
<= 0) {
1182 switch (mp
->m_sb
.sb_blocksize
) {
1183 case 16*1024: /* 16 KB */
1184 log
->l_iclog_bufs
= 3;
1186 case 32*1024: /* 32 KB */
1187 log
->l_iclog_bufs
= 4;
1189 case 64*1024: /* 64 KB */
1190 log
->l_iclog_bufs
= 8;
1193 xlog_panic("XFS: Invalid blocksize");
1199 done
: /* are we being asked to make the sizes selected above visible? */
1200 if (mp
->m_logbufs
== 0)
1201 mp
->m_logbufs
= log
->l_iclog_bufs
;
1202 if (mp
->m_logbsize
== 0)
1203 mp
->m_logbsize
= log
->l_iclog_size
;
1204 } /* xlog_get_iclog_buffer_size */
1208 * This routine initializes some of the log structure for a given mount point.
1209 * Its primary purpose is to fill in enough, so recovery can occur. However,
1210 * some other stuff may be filled in too.
1213 xlog_alloc_log(xfs_mount_t
*mp
,
1214 xfs_buftarg_t
*log_target
,
1215 xfs_daddr_t blk_offset
,
1219 xlog_rec_header_t
*head
;
1220 xlog_in_core_t
**iclogp
;
1221 xlog_in_core_t
*iclog
, *prev_iclog
=NULL
;
1226 log
= kmem_zalloc(sizeof(xlog_t
), KM_MAYFAIL
);
1231 log
->l_targ
= log_target
;
1232 log
->l_logsize
= BBTOB(num_bblks
);
1233 log
->l_logBBstart
= blk_offset
;
1234 log
->l_logBBsize
= num_bblks
;
1235 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
1236 log
->l_flags
|= XLOG_ACTIVE_RECOVERY
;
1238 log
->l_prev_block
= -1;
1239 log
->l_tail_lsn
= xlog_assign_lsn(1, 0);
1240 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1241 log
->l_last_sync_lsn
= log
->l_tail_lsn
;
1242 log
->l_curr_cycle
= 1; /* 0 is bad since this is initial value */
1243 log
->l_grant_reserve_cycle
= 1;
1244 log
->l_grant_write_cycle
= 1;
1246 if (xfs_sb_version_hassector(&mp
->m_sb
)) {
1247 log
->l_sectbb_log
= mp
->m_sb
.sb_logsectlog
- BBSHIFT
;
1248 ASSERT(log
->l_sectbb_log
<= mp
->m_sectbb_log
);
1249 /* for larger sector sizes, must have v2 or external log */
1250 ASSERT(log
->l_sectbb_log
== 0 ||
1251 log
->l_logBBstart
== 0 ||
1252 xfs_sb_version_haslogv2(&mp
->m_sb
));
1253 ASSERT(mp
->m_sb
.sb_logsectlog
>= BBSHIFT
);
1255 log
->l_sectbb_mask
= (1 << log
->l_sectbb_log
) - 1;
1257 xlog_get_iclog_buffer_size(mp
, log
);
1259 bp
= xfs_buf_get_empty(log
->l_iclog_size
, mp
->m_logdev_targp
);
1262 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1263 XFS_BUF_SET_BDSTRAT_FUNC(bp
, xlog_bdstrat_cb
);
1264 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1265 ASSERT(XFS_BUF_ISBUSY(bp
));
1266 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
1269 spin_lock_init(&log
->l_icloglock
);
1270 spin_lock_init(&log
->l_grant_lock
);
1271 sv_init(&log
->l_flush_wait
, 0, "flush_wait");
1273 xlog_trace_loggrant_alloc(log
);
1274 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1275 ASSERT((XFS_BUF_SIZE(bp
) & BBMASK
) == 0);
1277 iclogp
= &log
->l_iclog
;
1279 * The amount of memory to allocate for the iclog structure is
1280 * rather funky due to the way the structure is defined. It is
1281 * done this way so that we can use different sizes for machines
1282 * with different amounts of memory. See the definition of
1283 * xlog_in_core_t in xfs_log_priv.h for details.
1285 iclogsize
= log
->l_iclog_size
;
1286 ASSERT(log
->l_iclog_size
>= 4096);
1287 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
1288 *iclogp
= kmem_zalloc(sizeof(xlog_in_core_t
), KM_MAYFAIL
);
1290 goto out_free_iclog
;
1293 iclog
->ic_prev
= prev_iclog
;
1296 bp
= xfs_buf_get_noaddr(log
->l_iclog_size
, mp
->m_logdev_targp
);
1298 goto out_free_iclog
;
1299 if (!XFS_BUF_CPSEMA(bp
))
1301 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1302 XFS_BUF_SET_BDSTRAT_FUNC(bp
, xlog_bdstrat_cb
);
1303 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1305 iclog
->hic_data
= bp
->b_addr
;
1307 log
->l_iclog_bak
[i
] = (xfs_caddr_t
)&(iclog
->ic_header
);
1309 head
= &iclog
->ic_header
;
1310 memset(head
, 0, sizeof(xlog_rec_header_t
));
1311 head
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1312 head
->h_version
= cpu_to_be32(
1313 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1314 head
->h_size
= cpu_to_be32(log
->l_iclog_size
);
1316 head
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1317 memcpy(&head
->h_fs_uuid
, &mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1319 iclog
->ic_size
= XFS_BUF_SIZE(bp
) - log
->l_iclog_hsize
;
1320 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
1321 iclog
->ic_log
= log
;
1322 atomic_set(&iclog
->ic_refcnt
, 0);
1323 spin_lock_init(&iclog
->ic_callback_lock
);
1324 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
1325 iclog
->ic_datap
= (char *)iclog
->hic_data
+ log
->l_iclog_hsize
;
1327 ASSERT(XFS_BUF_ISBUSY(iclog
->ic_bp
));
1328 ASSERT(XFS_BUF_VALUSEMA(iclog
->ic_bp
) <= 0);
1329 sv_init(&iclog
->ic_force_wait
, SV_DEFAULT
, "iclog-force");
1330 sv_init(&iclog
->ic_write_wait
, SV_DEFAULT
, "iclog-write");
1332 xlog_trace_iclog_alloc(iclog
);
1334 iclogp
= &iclog
->ic_next
;
1336 *iclogp
= log
->l_iclog
; /* complete ring */
1337 log
->l_iclog
->ic_prev
= prev_iclog
; /* re-write 1st prev ptr */
1342 for (iclog
= log
->l_iclog
; iclog
; iclog
= prev_iclog
) {
1343 prev_iclog
= iclog
->ic_next
;
1345 sv_destroy(&iclog
->ic_force_wait
);
1346 sv_destroy(&iclog
->ic_write_wait
);
1347 xfs_buf_free(iclog
->ic_bp
);
1348 xlog_trace_iclog_dealloc(iclog
);
1352 spinlock_destroy(&log
->l_icloglock
);
1353 spinlock_destroy(&log
->l_grant_lock
);
1354 xlog_trace_loggrant_dealloc(log
);
1355 xfs_buf_free(log
->l_xbuf
);
1359 } /* xlog_alloc_log */
1363 * Write out the commit record of a transaction associated with the given
1364 * ticket. Return the lsn of the commit record.
1367 xlog_commit_record(xfs_mount_t
*mp
,
1368 xlog_ticket_t
*ticket
,
1369 xlog_in_core_t
**iclog
,
1370 xfs_lsn_t
*commitlsnp
)
1373 xfs_log_iovec_t reg
[1];
1375 reg
[0].i_addr
= NULL
;
1377 XLOG_VEC_SET_TYPE(®
[0], XLOG_REG_TYPE_COMMIT
);
1379 ASSERT_ALWAYS(iclog
);
1380 if ((error
= xlog_write(mp
, reg
, 1, ticket
, commitlsnp
,
1381 iclog
, XLOG_COMMIT_TRANS
))) {
1382 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
1385 } /* xlog_commit_record */
1389 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1390 * log space. This code pushes on the lsn which would supposedly free up
1391 * the 25% which we want to leave free. We may need to adopt a policy which
1392 * pushes on an lsn which is further along in the log once we reach the high
1393 * water mark. In this manner, we would be creating a low water mark.
1396 xlog_grant_push_ail(xfs_mount_t
*mp
,
1399 xlog_t
*log
= mp
->m_log
; /* pointer to the log */
1400 xfs_lsn_t tail_lsn
; /* lsn of the log tail */
1401 xfs_lsn_t threshold_lsn
= 0; /* lsn we'd like to be at */
1402 int free_blocks
; /* free blocks left to write to */
1403 int free_bytes
; /* free bytes left to write to */
1404 int threshold_block
; /* block in lsn we'd like to be at */
1405 int threshold_cycle
; /* lsn cycle we'd like to be at */
1408 ASSERT(BTOBB(need_bytes
) < log
->l_logBBsize
);
1410 spin_lock(&log
->l_grant_lock
);
1411 free_bytes
= xlog_space_left(log
,
1412 log
->l_grant_reserve_cycle
,
1413 log
->l_grant_reserve_bytes
);
1414 tail_lsn
= log
->l_tail_lsn
;
1415 free_blocks
= BTOBBT(free_bytes
);
1418 * Set the threshold for the minimum number of free blocks in the
1419 * log to the maximum of what the caller needs, one quarter of the
1420 * log, and 256 blocks.
1422 free_threshold
= BTOBB(need_bytes
);
1423 free_threshold
= MAX(free_threshold
, (log
->l_logBBsize
>> 2));
1424 free_threshold
= MAX(free_threshold
, 256);
1425 if (free_blocks
< free_threshold
) {
1426 threshold_block
= BLOCK_LSN(tail_lsn
) + free_threshold
;
1427 threshold_cycle
= CYCLE_LSN(tail_lsn
);
1428 if (threshold_block
>= log
->l_logBBsize
) {
1429 threshold_block
-= log
->l_logBBsize
;
1430 threshold_cycle
+= 1;
1432 threshold_lsn
= xlog_assign_lsn(threshold_cycle
, threshold_block
);
1434 /* Don't pass in an lsn greater than the lsn of the last
1435 * log record known to be on disk.
1437 if (XFS_LSN_CMP(threshold_lsn
, log
->l_last_sync_lsn
) > 0)
1438 threshold_lsn
= log
->l_last_sync_lsn
;
1440 spin_unlock(&log
->l_grant_lock
);
1443 * Get the transaction layer to kick the dirty buffers out to
1444 * disk asynchronously. No point in trying to do this if
1445 * the filesystem is shutting down.
1447 if (threshold_lsn
&&
1448 !XLOG_FORCED_SHUTDOWN(log
))
1449 xfs_trans_push_ail(mp
, threshold_lsn
);
1450 } /* xlog_grant_push_ail */
1454 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1455 * fashion. Previously, we should have moved the current iclog
1456 * ptr in the log to point to the next available iclog. This allows further
1457 * write to continue while this code syncs out an iclog ready to go.
1458 * Before an in-core log can be written out, the data section must be scanned
1459 * to save away the 1st word of each BBSIZE block into the header. We replace
1460 * it with the current cycle count. Each BBSIZE block is tagged with the
1461 * cycle count because there in an implicit assumption that drives will
1462 * guarantee that entire 512 byte blocks get written at once. In other words,
1463 * we can't have part of a 512 byte block written and part not written. By
1464 * tagging each block, we will know which blocks are valid when recovering
1465 * after an unclean shutdown.
1467 * This routine is single threaded on the iclog. No other thread can be in
1468 * this routine with the same iclog. Changing contents of iclog can there-
1469 * fore be done without grabbing the state machine lock. Updating the global
1470 * log will require grabbing the lock though.
1472 * The entire log manager uses a logical block numbering scheme. Only
1473 * log_sync (and then only bwrite()) know about the fact that the log may
1474 * not start with block zero on a given device. The log block start offset
1475 * is added immediately before calling bwrite().
1479 xlog_sync(xlog_t
*log
,
1480 xlog_in_core_t
*iclog
)
1482 xfs_caddr_t dptr
; /* pointer to byte sized element */
1485 uint count
; /* byte count of bwrite */
1486 uint count_init
; /* initial count before roundup */
1487 int roundoff
; /* roundoff to BB or stripe */
1488 int split
= 0; /* split write into two regions */
1490 int v2
= xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
);
1492 XFS_STATS_INC(xs_log_writes
);
1493 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
1495 /* Add for LR header */
1496 count_init
= log
->l_iclog_hsize
+ iclog
->ic_offset
;
1498 /* Round out the log write size */
1499 if (v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1) {
1500 /* we have a v2 stripe unit to use */
1501 count
= XLOG_LSUNITTOB(log
, XLOG_BTOLSUNIT(log
, count_init
));
1503 count
= BBTOB(BTOBB(count_init
));
1505 roundoff
= count
- count_init
;
1506 ASSERT(roundoff
>= 0);
1507 ASSERT((v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1 &&
1508 roundoff
< log
->l_mp
->m_sb
.sb_logsunit
)
1510 (log
->l_mp
->m_sb
.sb_logsunit
<= 1 &&
1511 roundoff
< BBTOB(1)));
1513 /* move grant heads by roundoff in sync */
1514 spin_lock(&log
->l_grant_lock
);
1515 xlog_grant_add_space(log
, roundoff
);
1516 spin_unlock(&log
->l_grant_lock
);
1518 /* put cycle number in every block */
1519 xlog_pack_data(log
, iclog
, roundoff
);
1521 /* real byte length */
1523 iclog
->ic_header
.h_len
=
1524 cpu_to_be32(iclog
->ic_offset
+ roundoff
);
1526 iclog
->ic_header
.h_len
=
1527 cpu_to_be32(iclog
->ic_offset
);
1531 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long)1);
1532 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1533 XFS_BUF_SET_ADDR(bp
, BLOCK_LSN(be64_to_cpu(iclog
->ic_header
.h_lsn
)));
1535 XFS_STATS_ADD(xs_log_blocks
, BTOBB(count
));
1537 /* Do we need to split this write into 2 parts? */
1538 if (XFS_BUF_ADDR(bp
) + BTOBB(count
) > log
->l_logBBsize
) {
1539 split
= count
- (BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
)));
1540 count
= BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
));
1541 iclog
->ic_bwritecnt
= 2; /* split into 2 writes */
1543 iclog
->ic_bwritecnt
= 1;
1545 XFS_BUF_SET_COUNT(bp
, count
);
1546 XFS_BUF_SET_FSPRIVATE(bp
, iclog
); /* save for later */
1547 XFS_BUF_ZEROFLAGS(bp
);
1551 * Do an ordered write for the log block.
1552 * Its unnecessary to flush the first split block in the log wrap case.
1554 if (!split
&& (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
))
1555 XFS_BUF_ORDERED(bp
);
1557 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1558 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1560 xlog_verify_iclog(log
, iclog
, count
, B_TRUE
);
1562 /* account for log which doesn't start at block #0 */
1563 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1565 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1570 if ((error
= XFS_bwrite(bp
))) {
1571 xfs_ioerror_alert("xlog_sync", log
->l_mp
, bp
,
1576 bp
= iclog
->ic_log
->l_xbuf
;
1577 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) ==
1579 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1580 XFS_BUF_SET_ADDR(bp
, 0); /* logical 0 */
1581 XFS_BUF_SET_PTR(bp
, (xfs_caddr_t
)((__psint_t
)&(iclog
->ic_header
)+
1582 (__psint_t
)count
), split
);
1583 XFS_BUF_SET_FSPRIVATE(bp
, iclog
);
1584 XFS_BUF_ZEROFLAGS(bp
);
1587 if (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
)
1588 XFS_BUF_ORDERED(bp
);
1589 dptr
= XFS_BUF_PTR(bp
);
1591 * Bump the cycle numbers at the start of each block
1592 * since this part of the buffer is at the start of
1593 * a new cycle. Watch out for the header magic number
1596 for (i
= 0; i
< split
; i
+= BBSIZE
) {
1597 be32_add_cpu((__be32
*)dptr
, 1);
1598 if (be32_to_cpu(*(__be32
*)dptr
) == XLOG_HEADER_MAGIC_NUM
)
1599 be32_add_cpu((__be32
*)dptr
, 1);
1603 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1604 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1606 /* account for internal log which doesn't start at block #0 */
1607 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1609 if ((error
= XFS_bwrite(bp
))) {
1610 xfs_ioerror_alert("xlog_sync (split)", log
->l_mp
,
1611 bp
, XFS_BUF_ADDR(bp
));
1620 * Deallocate a log structure
1623 xlog_dealloc_log(xlog_t
*log
)
1625 xlog_in_core_t
*iclog
, *next_iclog
;
1628 iclog
= log
->l_iclog
;
1629 for (i
=0; i
<log
->l_iclog_bufs
; i
++) {
1630 sv_destroy(&iclog
->ic_force_wait
);
1631 sv_destroy(&iclog
->ic_write_wait
);
1632 xfs_buf_free(iclog
->ic_bp
);
1633 xlog_trace_iclog_dealloc(iclog
);
1634 next_iclog
= iclog
->ic_next
;
1638 spinlock_destroy(&log
->l_icloglock
);
1639 spinlock_destroy(&log
->l_grant_lock
);
1641 xfs_buf_free(log
->l_xbuf
);
1642 xlog_trace_loggrant_dealloc(log
);
1643 log
->l_mp
->m_log
= NULL
;
1645 } /* xlog_dealloc_log */
1648 * Update counters atomically now that memcpy is done.
1652 xlog_state_finish_copy(xlog_t
*log
,
1653 xlog_in_core_t
*iclog
,
1657 spin_lock(&log
->l_icloglock
);
1659 be32_add_cpu(&iclog
->ic_header
.h_num_logops
, record_cnt
);
1660 iclog
->ic_offset
+= copy_bytes
;
1662 spin_unlock(&log
->l_icloglock
);
1663 } /* xlog_state_finish_copy */
1669 * print out info relating to regions written which consume
1673 xlog_print_tic_res(xfs_mount_t
*mp
, xlog_ticket_t
*ticket
)
1676 uint ophdr_spc
= ticket
->t_res_num_ophdrs
* (uint
)sizeof(xlog_op_header_t
);
1678 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1679 static char *res_type_str
[XLOG_REG_TYPE_MAX
] = {
1700 static char *trans_type_str
[XFS_TRANS_TYPE_MAX
] = {
1743 xfs_fs_cmn_err(CE_WARN
, mp
,
1744 "xfs_log_write: reservation summary:\n"
1745 " trans type = %s (%u)\n"
1746 " unit res = %d bytes\n"
1747 " current res = %d bytes\n"
1748 " total reg = %u bytes (o/flow = %u bytes)\n"
1749 " ophdrs = %u (ophdr space = %u bytes)\n"
1750 " ophdr + reg = %u bytes\n"
1751 " num regions = %u\n",
1752 ((ticket
->t_trans_type
<= 0 ||
1753 ticket
->t_trans_type
> XFS_TRANS_TYPE_MAX
) ?
1754 "bad-trans-type" : trans_type_str
[ticket
->t_trans_type
-1]),
1755 ticket
->t_trans_type
,
1758 ticket
->t_res_arr_sum
, ticket
->t_res_o_flow
,
1759 ticket
->t_res_num_ophdrs
, ophdr_spc
,
1760 ticket
->t_res_arr_sum
+
1761 ticket
->t_res_o_flow
+ ophdr_spc
,
1764 for (i
= 0; i
< ticket
->t_res_num
; i
++) {
1765 uint r_type
= ticket
->t_res_arr
[i
].r_type
;
1767 "region[%u]: %s - %u bytes\n",
1769 ((r_type
<= 0 || r_type
> XLOG_REG_TYPE_MAX
) ?
1770 "bad-rtype" : res_type_str
[r_type
-1]),
1771 ticket
->t_res_arr
[i
].r_len
);
1776 * Write some region out to in-core log
1778 * This will be called when writing externally provided regions or when
1779 * writing out a commit record for a given transaction.
1781 * General algorithm:
1782 * 1. Find total length of this write. This may include adding to the
1783 * lengths passed in.
1784 * 2. Check whether we violate the tickets reservation.
1785 * 3. While writing to this iclog
1786 * A. Reserve as much space in this iclog as can get
1787 * B. If this is first write, save away start lsn
1788 * C. While writing this region:
1789 * 1. If first write of transaction, write start record
1790 * 2. Write log operation header (header per region)
1791 * 3. Find out if we can fit entire region into this iclog
1792 * 4. Potentially, verify destination memcpy ptr
1793 * 5. Memcpy (partial) region
1794 * 6. If partial copy, release iclog; otherwise, continue
1795 * copying more regions into current iclog
1796 * 4. Mark want sync bit (in simulation mode)
1797 * 5. Release iclog for potential flush to on-disk log.
1800 * 1. Panic if reservation is overrun. This should never happen since
1801 * reservation amounts are generated internal to the filesystem.
1803 * 1. Tickets are single threaded data structures.
1804 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1805 * syncing routine. When a single log_write region needs to span
1806 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1807 * on all log operation writes which don't contain the end of the
1808 * region. The XLOG_END_TRANS bit is used for the in-core log
1809 * operation which contains the end of the continued log_write region.
1810 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1811 * we don't really know exactly how much space will be used. As a result,
1812 * we don't update ic_offset until the end when we know exactly how many
1813 * bytes have been written out.
1816 xlog_write(xfs_mount_t
* mp
,
1817 xfs_log_iovec_t reg
[],
1819 xfs_log_ticket_t tic
,
1820 xfs_lsn_t
*start_lsn
,
1821 xlog_in_core_t
**commit_iclog
,
1824 xlog_t
*log
= mp
->m_log
;
1825 xlog_ticket_t
*ticket
= (xlog_ticket_t
*)tic
;
1826 xlog_in_core_t
*iclog
= NULL
; /* ptr to current in-core log */
1827 xlog_op_header_t
*logop_head
; /* ptr to log operation header */
1828 __psint_t ptr
; /* copy address into data region */
1829 int len
; /* # xlog_write() bytes 2 still copy */
1830 int index
; /* region index currently copying */
1831 int log_offset
; /* offset (from 0) into data region */
1832 int start_rec_copy
; /* # bytes to copy for start record */
1833 int partial_copy
; /* did we split a region? */
1834 int partial_copy_len
;/* # bytes copied if split region */
1835 int need_copy
; /* # bytes need to memcpy this region */
1836 int copy_len
; /* # bytes actually memcpy'ing */
1837 int copy_off
; /* # bytes from entry start */
1838 int contwr
; /* continued write of in-core log? */
1840 int record_cnt
= 0, data_cnt
= 0;
1842 partial_copy_len
= partial_copy
= 0;
1844 /* Calculate potential maximum space. Each region gets its own
1845 * xlog_op_header_t and may need to be double word aligned.
1848 if (ticket
->t_flags
& XLOG_TIC_INITED
) { /* acct for start rec of xact */
1849 len
+= sizeof(xlog_op_header_t
);
1850 ticket
->t_res_num_ophdrs
++;
1853 for (index
= 0; index
< nentries
; index
++) {
1854 len
+= sizeof(xlog_op_header_t
); /* each region gets >= 1 */
1855 ticket
->t_res_num_ophdrs
++;
1856 len
+= reg
[index
].i_len
;
1857 xlog_tic_add_region(ticket
, reg
[index
].i_len
, reg
[index
].i_type
);
1859 contwr
= *start_lsn
= 0;
1861 if (ticket
->t_curr_res
< len
) {
1862 xlog_print_tic_res(mp
, ticket
);
1865 "xfs_log_write: reservation ran out. Need to up reservation");
1867 /* Customer configurable panic */
1868 xfs_cmn_err(XFS_PTAG_LOGRES
, CE_ALERT
, mp
,
1869 "xfs_log_write: reservation ran out. Need to up reservation");
1870 /* If we did not panic, shutdown the filesystem */
1871 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1874 ticket
->t_curr_res
-= len
;
1876 for (index
= 0; index
< nentries
; ) {
1877 if ((error
= xlog_state_get_iclog_space(log
, len
, &iclog
, ticket
,
1878 &contwr
, &log_offset
)))
1881 ASSERT(log_offset
<= iclog
->ic_size
- 1);
1882 ptr
= (__psint_t
) ((char *)iclog
->ic_datap
+log_offset
);
1884 /* start_lsn is the first lsn written to. That's all we need. */
1886 *start_lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
1888 /* This loop writes out as many regions as can fit in the amount
1889 * of space which was allocated by xlog_state_get_iclog_space().
1891 while (index
< nentries
) {
1892 ASSERT(reg
[index
].i_len
% sizeof(__int32_t
) == 0);
1893 ASSERT((__psint_t
)ptr
% sizeof(__int32_t
) == 0);
1896 /* If first write for transaction, insert start record.
1897 * We can't be trying to commit if we are inited. We can't
1898 * have any "partial_copy" if we are inited.
1900 if (ticket
->t_flags
& XLOG_TIC_INITED
) {
1901 logop_head
= (xlog_op_header_t
*)ptr
;
1902 logop_head
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1903 logop_head
->oh_clientid
= ticket
->t_clientid
;
1904 logop_head
->oh_len
= 0;
1905 logop_head
->oh_flags
= XLOG_START_TRANS
;
1906 logop_head
->oh_res2
= 0;
1907 ticket
->t_flags
&= ~XLOG_TIC_INITED
; /* clear bit */
1910 start_rec_copy
= sizeof(xlog_op_header_t
);
1911 xlog_write_adv_cnt(ptr
, len
, log_offset
, start_rec_copy
);
1914 /* Copy log operation header directly into data section */
1915 logop_head
= (xlog_op_header_t
*)ptr
;
1916 logop_head
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1917 logop_head
->oh_clientid
= ticket
->t_clientid
;
1918 logop_head
->oh_res2
= 0;
1920 /* header copied directly */
1921 xlog_write_adv_cnt(ptr
, len
, log_offset
, sizeof(xlog_op_header_t
));
1923 /* are we copying a commit or unmount record? */
1924 logop_head
->oh_flags
= flags
;
1927 * We've seen logs corrupted with bad transaction client
1928 * ids. This makes sure that XFS doesn't generate them on.
1929 * Turn this into an EIO and shut down the filesystem.
1931 switch (logop_head
->oh_clientid
) {
1932 case XFS_TRANSACTION
:
1937 xfs_fs_cmn_err(CE_WARN
, mp
,
1938 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1939 logop_head
->oh_clientid
, tic
);
1940 return XFS_ERROR(EIO
);
1943 /* Partial write last time? => (partial_copy != 0)
1944 * need_copy is the amount we'd like to copy if everything could
1945 * fit in the current memcpy.
1947 need_copy
= reg
[index
].i_len
- partial_copy_len
;
1949 copy_off
= partial_copy_len
;
1950 if (need_copy
<= iclog
->ic_size
- log_offset
) { /*complete write */
1951 copy_len
= need_copy
;
1952 logop_head
->oh_len
= cpu_to_be32(copy_len
);
1954 logop_head
->oh_flags
|= (XLOG_END_TRANS
|XLOG_WAS_CONT_TRANS
);
1955 partial_copy_len
= partial_copy
= 0;
1956 } else { /* partial write */
1957 copy_len
= iclog
->ic_size
- log_offset
;
1958 logop_head
->oh_len
= cpu_to_be32(copy_len
);
1959 logop_head
->oh_flags
|= XLOG_CONTINUE_TRANS
;
1961 logop_head
->oh_flags
|= XLOG_WAS_CONT_TRANS
;
1962 partial_copy_len
+= copy_len
;
1964 len
+= sizeof(xlog_op_header_t
); /* from splitting of region */
1965 /* account for new log op header */
1966 ticket
->t_curr_res
-= sizeof(xlog_op_header_t
);
1967 ticket
->t_res_num_ophdrs
++;
1969 xlog_verify_dest_ptr(log
, ptr
);
1972 ASSERT(copy_len
>= 0);
1973 memcpy((xfs_caddr_t
)ptr
, reg
[index
].i_addr
+ copy_off
, copy_len
);
1974 xlog_write_adv_cnt(ptr
, len
, log_offset
, copy_len
);
1976 /* make copy_len total bytes copied, including headers */
1977 copy_len
+= start_rec_copy
+ sizeof(xlog_op_header_t
);
1979 data_cnt
+= contwr
? copy_len
: 0;
1980 if (partial_copy
) { /* copied partial region */
1981 /* already marked WANT_SYNC by xlog_state_get_iclog_space */
1982 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1983 record_cnt
= data_cnt
= 0;
1984 if ((error
= xlog_state_release_iclog(log
, iclog
)))
1986 break; /* don't increment index */
1987 } else { /* copied entire region */
1989 partial_copy_len
= partial_copy
= 0;
1991 if (iclog
->ic_size
- log_offset
<= sizeof(xlog_op_header_t
)) {
1992 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1993 record_cnt
= data_cnt
= 0;
1994 xlog_state_want_sync(log
, iclog
);
1996 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1997 *commit_iclog
= iclog
;
1998 } else if ((error
= xlog_state_release_iclog(log
, iclog
)))
2000 if (index
== nentries
)
2001 return 0; /* we are done */
2005 } /* if (partial_copy) */
2006 } /* while (index < nentries) */
2007 } /* for (index = 0; index < nentries; ) */
2010 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
2012 ASSERT(flags
& XLOG_COMMIT_TRANS
);
2013 *commit_iclog
= iclog
;
2016 return xlog_state_release_iclog(log
, iclog
);
2020 /*****************************************************************************
2022 * State Machine functions
2024 *****************************************************************************
2027 /* Clean iclogs starting from the head. This ordering must be
2028 * maintained, so an iclog doesn't become ACTIVE beyond one that
2029 * is SYNCING. This is also required to maintain the notion that we use
2030 * a ordered wait queue to hold off would be writers to the log when every
2031 * iclog is trying to sync to disk.
2033 * State Change: DIRTY -> ACTIVE
2036 xlog_state_clean_log(xlog_t
*log
)
2038 xlog_in_core_t
*iclog
;
2041 iclog
= log
->l_iclog
;
2043 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2044 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
2045 iclog
->ic_offset
= 0;
2046 ASSERT(iclog
->ic_callback
== NULL
);
2048 * If the number of ops in this iclog indicate it just
2049 * contains the dummy transaction, we can
2050 * change state into IDLE (the second time around).
2051 * Otherwise we should change the state into
2053 * We don't need to cover the dummy.
2056 (be32_to_cpu(iclog
->ic_header
.h_num_logops
) ==
2061 * We have two dirty iclogs so start over
2062 * This could also be num of ops indicates
2063 * this is not the dummy going out.
2067 iclog
->ic_header
.h_num_logops
= 0;
2068 memset(iclog
->ic_header
.h_cycle_data
, 0,
2069 sizeof(iclog
->ic_header
.h_cycle_data
));
2070 iclog
->ic_header
.h_lsn
= 0;
2071 } else if (iclog
->ic_state
== XLOG_STATE_ACTIVE
)
2074 break; /* stop cleaning */
2075 iclog
= iclog
->ic_next
;
2076 } while (iclog
!= log
->l_iclog
);
2078 /* log is locked when we are called */
2080 * Change state for the dummy log recording.
2081 * We usually go to NEED. But we go to NEED2 if the changed indicates
2082 * we are done writing the dummy record.
2083 * If we are done with the second dummy recored (DONE2), then
2087 switch (log
->l_covered_state
) {
2088 case XLOG_STATE_COVER_IDLE
:
2089 case XLOG_STATE_COVER_NEED
:
2090 case XLOG_STATE_COVER_NEED2
:
2091 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2094 case XLOG_STATE_COVER_DONE
:
2096 log
->l_covered_state
= XLOG_STATE_COVER_NEED2
;
2098 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2101 case XLOG_STATE_COVER_DONE2
:
2103 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
2105 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2112 } /* xlog_state_clean_log */
2115 xlog_get_lowest_lsn(
2118 xlog_in_core_t
*lsn_log
;
2119 xfs_lsn_t lowest_lsn
, lsn
;
2121 lsn_log
= log
->l_iclog
;
2124 if (!(lsn_log
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
))) {
2125 lsn
= be64_to_cpu(lsn_log
->ic_header
.h_lsn
);
2126 if ((lsn
&& !lowest_lsn
) ||
2127 (XFS_LSN_CMP(lsn
, lowest_lsn
) < 0)) {
2131 lsn_log
= lsn_log
->ic_next
;
2132 } while (lsn_log
!= log
->l_iclog
);
2138 xlog_state_do_callback(
2141 xlog_in_core_t
*ciclog
)
2143 xlog_in_core_t
*iclog
;
2144 xlog_in_core_t
*first_iclog
; /* used to know when we've
2145 * processed all iclogs once */
2146 xfs_log_callback_t
*cb
, *cb_next
;
2148 xfs_lsn_t lowest_lsn
;
2149 int ioerrors
; /* counter: iclogs with errors */
2150 int loopdidcallbacks
; /* flag: inner loop did callbacks*/
2151 int funcdidcallbacks
; /* flag: function did callbacks */
2152 int repeats
; /* for issuing console warnings if
2153 * looping too many times */
2156 spin_lock(&log
->l_icloglock
);
2157 first_iclog
= iclog
= log
->l_iclog
;
2159 funcdidcallbacks
= 0;
2164 * Scan all iclogs starting with the one pointed to by the
2165 * log. Reset this starting point each time the log is
2166 * unlocked (during callbacks).
2168 * Keep looping through iclogs until one full pass is made
2169 * without running any callbacks.
2171 first_iclog
= log
->l_iclog
;
2172 iclog
= log
->l_iclog
;
2173 loopdidcallbacks
= 0;
2178 /* skip all iclogs in the ACTIVE & DIRTY states */
2179 if (iclog
->ic_state
&
2180 (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
)) {
2181 iclog
= iclog
->ic_next
;
2186 * Between marking a filesystem SHUTDOWN and stopping
2187 * the log, we do flush all iclogs to disk (if there
2188 * wasn't a log I/O error). So, we do want things to
2189 * go smoothly in case of just a SHUTDOWN w/o a
2192 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
2194 * Can only perform callbacks in order. Since
2195 * this iclog is not in the DONE_SYNC/
2196 * DO_CALLBACK state, we skip the rest and
2197 * just try to clean up. If we set our iclog
2198 * to DO_CALLBACK, we will not process it when
2199 * we retry since a previous iclog is in the
2200 * CALLBACK and the state cannot change since
2201 * we are holding the l_icloglock.
2203 if (!(iclog
->ic_state
&
2204 (XLOG_STATE_DONE_SYNC
|
2205 XLOG_STATE_DO_CALLBACK
))) {
2206 if (ciclog
&& (ciclog
->ic_state
==
2207 XLOG_STATE_DONE_SYNC
)) {
2208 ciclog
->ic_state
= XLOG_STATE_DO_CALLBACK
;
2213 * We now have an iclog that is in either the
2214 * DO_CALLBACK or DONE_SYNC states. The other
2215 * states (WANT_SYNC, SYNCING, or CALLBACK were
2216 * caught by the above if and are going to
2217 * clean (i.e. we aren't doing their callbacks)
2222 * We will do one more check here to see if we
2223 * have chased our tail around.
2226 lowest_lsn
= xlog_get_lowest_lsn(log
);
2228 XFS_LSN_CMP(lowest_lsn
,
2229 be64_to_cpu(iclog
->ic_header
.h_lsn
)) < 0) {
2230 iclog
= iclog
->ic_next
;
2231 continue; /* Leave this iclog for
2235 iclog
->ic_state
= XLOG_STATE_CALLBACK
;
2237 spin_unlock(&log
->l_icloglock
);
2239 /* l_last_sync_lsn field protected by
2240 * l_grant_lock. Don't worry about iclog's lsn.
2241 * No one else can be here except us.
2243 spin_lock(&log
->l_grant_lock
);
2244 ASSERT(XFS_LSN_CMP(log
->l_last_sync_lsn
,
2245 be64_to_cpu(iclog
->ic_header
.h_lsn
)) <= 0);
2246 log
->l_last_sync_lsn
=
2247 be64_to_cpu(iclog
->ic_header
.h_lsn
);
2248 spin_unlock(&log
->l_grant_lock
);
2251 spin_unlock(&log
->l_icloglock
);
2256 * Keep processing entries in the callback list until
2257 * we come around and it is empty. We need to
2258 * atomically see that the list is empty and change the
2259 * state to DIRTY so that we don't miss any more
2260 * callbacks being added.
2262 spin_lock(&iclog
->ic_callback_lock
);
2263 cb
= iclog
->ic_callback
;
2265 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
2266 iclog
->ic_callback
= NULL
;
2267 spin_unlock(&iclog
->ic_callback_lock
);
2269 /* perform callbacks in the order given */
2270 for (; cb
; cb
= cb_next
) {
2271 cb_next
= cb
->cb_next
;
2272 cb
->cb_func(cb
->cb_arg
, aborted
);
2274 spin_lock(&iclog
->ic_callback_lock
);
2275 cb
= iclog
->ic_callback
;
2281 spin_lock(&log
->l_icloglock
);
2282 ASSERT(iclog
->ic_callback
== NULL
);
2283 spin_unlock(&iclog
->ic_callback_lock
);
2284 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
))
2285 iclog
->ic_state
= XLOG_STATE_DIRTY
;
2288 * Transition from DIRTY to ACTIVE if applicable.
2289 * NOP if STATE_IOERROR.
2291 xlog_state_clean_log(log
);
2293 /* wake up threads waiting in xfs_log_force() */
2294 sv_broadcast(&iclog
->ic_force_wait
);
2296 iclog
= iclog
->ic_next
;
2297 } while (first_iclog
!= iclog
);
2299 if (repeats
> 5000) {
2300 flushcnt
+= repeats
;
2302 xfs_fs_cmn_err(CE_WARN
, log
->l_mp
,
2303 "%s: possible infinite loop (%d iterations)",
2304 __func__
, flushcnt
);
2306 } while (!ioerrors
&& loopdidcallbacks
);
2309 * make one last gasp attempt to see if iclogs are being left in
2313 if (funcdidcallbacks
) {
2314 first_iclog
= iclog
= log
->l_iclog
;
2316 ASSERT(iclog
->ic_state
!= XLOG_STATE_DO_CALLBACK
);
2318 * Terminate the loop if iclogs are found in states
2319 * which will cause other threads to clean up iclogs.
2321 * SYNCING - i/o completion will go through logs
2322 * DONE_SYNC - interrupt thread should be waiting for
2324 * IOERROR - give up hope all ye who enter here
2326 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
||
2327 iclog
->ic_state
== XLOG_STATE_SYNCING
||
2328 iclog
->ic_state
== XLOG_STATE_DONE_SYNC
||
2329 iclog
->ic_state
== XLOG_STATE_IOERROR
)
2331 iclog
= iclog
->ic_next
;
2332 } while (first_iclog
!= iclog
);
2336 if (log
->l_iclog
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_IOERROR
))
2338 spin_unlock(&log
->l_icloglock
);
2341 sv_broadcast(&log
->l_flush_wait
);
2346 * Finish transitioning this iclog to the dirty state.
2348 * Make sure that we completely execute this routine only when this is
2349 * the last call to the iclog. There is a good chance that iclog flushes,
2350 * when we reach the end of the physical log, get turned into 2 separate
2351 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2352 * routine. By using the reference count bwritecnt, we guarantee that only
2353 * the second completion goes through.
2355 * Callbacks could take time, so they are done outside the scope of the
2356 * global state machine log lock.
2359 xlog_state_done_syncing(
2360 xlog_in_core_t
*iclog
,
2363 xlog_t
*log
= iclog
->ic_log
;
2365 spin_lock(&log
->l_icloglock
);
2367 ASSERT(iclog
->ic_state
== XLOG_STATE_SYNCING
||
2368 iclog
->ic_state
== XLOG_STATE_IOERROR
);
2369 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
2370 ASSERT(iclog
->ic_bwritecnt
== 1 || iclog
->ic_bwritecnt
== 2);
2374 * If we got an error, either on the first buffer, or in the case of
2375 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2376 * and none should ever be attempted to be written to disk
2379 if (iclog
->ic_state
!= XLOG_STATE_IOERROR
) {
2380 if (--iclog
->ic_bwritecnt
== 1) {
2381 spin_unlock(&log
->l_icloglock
);
2384 iclog
->ic_state
= XLOG_STATE_DONE_SYNC
;
2388 * Someone could be sleeping prior to writing out the next
2389 * iclog buffer, we wake them all, one will get to do the
2390 * I/O, the others get to wait for the result.
2392 sv_broadcast(&iclog
->ic_write_wait
);
2393 spin_unlock(&log
->l_icloglock
);
2394 xlog_state_do_callback(log
, aborted
, iclog
); /* also cleans log */
2395 } /* xlog_state_done_syncing */
2399 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2400 * sleep. We wait on the flush queue on the head iclog as that should be
2401 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2402 * we will wait here and all new writes will sleep until a sync completes.
2404 * The in-core logs are used in a circular fashion. They are not used
2405 * out-of-order even when an iclog past the head is free.
2408 * * log_offset where xlog_write() can start writing into the in-core
2410 * * in-core log pointer to which xlog_write() should write.
2411 * * boolean indicating this is a continued write to an in-core log.
2412 * If this is the last write, then the in-core log's offset field
2413 * needs to be incremented, depending on the amount of data which
2417 xlog_state_get_iclog_space(xlog_t
*log
,
2419 xlog_in_core_t
**iclogp
,
2420 xlog_ticket_t
*ticket
,
2421 int *continued_write
,
2425 xlog_rec_header_t
*head
;
2426 xlog_in_core_t
*iclog
;
2430 spin_lock(&log
->l_icloglock
);
2431 if (XLOG_FORCED_SHUTDOWN(log
)) {
2432 spin_unlock(&log
->l_icloglock
);
2433 return XFS_ERROR(EIO
);
2436 iclog
= log
->l_iclog
;
2437 if (iclog
->ic_state
!= XLOG_STATE_ACTIVE
) {
2438 xlog_trace_iclog(iclog
, XLOG_TRACE_SLEEP_FLUSH
);
2439 XFS_STATS_INC(xs_log_noiclogs
);
2441 /* Wait for log writes to have flushed */
2442 sv_wait(&log
->l_flush_wait
, 0, &log
->l_icloglock
, 0);
2446 head
= &iclog
->ic_header
;
2448 atomic_inc(&iclog
->ic_refcnt
); /* prevents sync */
2449 log_offset
= iclog
->ic_offset
;
2451 /* On the 1st write to an iclog, figure out lsn. This works
2452 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2453 * committing to. If the offset is set, that's how many blocks
2456 if (log_offset
== 0) {
2457 ticket
->t_curr_res
-= log
->l_iclog_hsize
;
2458 xlog_tic_add_region(ticket
,
2460 XLOG_REG_TYPE_LRHEADER
);
2461 head
->h_cycle
= cpu_to_be32(log
->l_curr_cycle
);
2462 head
->h_lsn
= cpu_to_be64(
2463 xlog_assign_lsn(log
->l_curr_cycle
, log
->l_curr_block
));
2464 ASSERT(log
->l_curr_block
>= 0);
2467 /* If there is enough room to write everything, then do it. Otherwise,
2468 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2469 * bit is on, so this will get flushed out. Don't update ic_offset
2470 * until you know exactly how many bytes get copied. Therefore, wait
2471 * until later to update ic_offset.
2473 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2474 * can fit into remaining data section.
2476 if (iclog
->ic_size
- iclog
->ic_offset
< 2*sizeof(xlog_op_header_t
)) {
2477 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2480 * If I'm the only one writing to this iclog, sync it to disk.
2481 * We need to do an atomic compare and decrement here to avoid
2482 * racing with concurrent atomic_dec_and_lock() calls in
2483 * xlog_state_release_iclog() when there is more than one
2484 * reference to the iclog.
2486 if (!atomic_add_unless(&iclog
->ic_refcnt
, -1, 1)) {
2487 /* we are the only one */
2488 spin_unlock(&log
->l_icloglock
);
2489 error
= xlog_state_release_iclog(log
, iclog
);
2493 spin_unlock(&log
->l_icloglock
);
2498 /* Do we have enough room to write the full amount in the remainder
2499 * of this iclog? Or must we continue a write on the next iclog and
2500 * mark this iclog as completely taken? In the case where we switch
2501 * iclogs (to mark it taken), this particular iclog will release/sync
2502 * to disk in xlog_write().
2504 if (len
<= iclog
->ic_size
- iclog
->ic_offset
) {
2505 *continued_write
= 0;
2506 iclog
->ic_offset
+= len
;
2508 *continued_write
= 1;
2509 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2513 ASSERT(iclog
->ic_offset
<= iclog
->ic_size
);
2514 spin_unlock(&log
->l_icloglock
);
2516 *logoffsetp
= log_offset
;
2518 } /* xlog_state_get_iclog_space */
2521 * Atomically get the log space required for a log ticket.
2523 * Once a ticket gets put onto the reserveq, it will only return after
2524 * the needed reservation is satisfied.
2527 xlog_grant_log_space(xlog_t
*log
,
2538 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2539 panic("grant Recovery problem");
2542 /* Is there space or do we need to sleep? */
2543 spin_lock(&log
->l_grant_lock
);
2544 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: enter");
2546 /* something is already sleeping; insert new transaction at end */
2547 if (log
->l_reserve_headq
) {
2548 xlog_ins_ticketq(&log
->l_reserve_headq
, tic
);
2549 xlog_trace_loggrant(log
, tic
,
2550 "xlog_grant_log_space: sleep 1");
2552 * Gotta check this before going to sleep, while we're
2553 * holding the grant lock.
2555 if (XLOG_FORCED_SHUTDOWN(log
))
2558 XFS_STATS_INC(xs_sleep_logspace
);
2559 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2561 * If we got an error, and the filesystem is shutting down,
2562 * we'll catch it down below. So just continue...
2564 xlog_trace_loggrant(log
, tic
,
2565 "xlog_grant_log_space: wake 1");
2566 spin_lock(&log
->l_grant_lock
);
2568 if (tic
->t_flags
& XFS_LOG_PERM_RESERV
)
2569 need_bytes
= tic
->t_unit_res
*tic
->t_ocnt
;
2571 need_bytes
= tic
->t_unit_res
;
2574 if (XLOG_FORCED_SHUTDOWN(log
))
2577 free_bytes
= xlog_space_left(log
, log
->l_grant_reserve_cycle
,
2578 log
->l_grant_reserve_bytes
);
2579 if (free_bytes
< need_bytes
) {
2580 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2581 xlog_ins_ticketq(&log
->l_reserve_headq
, tic
);
2582 xlog_trace_loggrant(log
, tic
,
2583 "xlog_grant_log_space: sleep 2");
2584 XFS_STATS_INC(xs_sleep_logspace
);
2585 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2587 if (XLOG_FORCED_SHUTDOWN(log
)) {
2588 spin_lock(&log
->l_grant_lock
);
2592 xlog_trace_loggrant(log
, tic
,
2593 "xlog_grant_log_space: wake 2");
2594 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2595 spin_lock(&log
->l_grant_lock
);
2597 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2598 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2600 /* we've got enough space */
2601 xlog_grant_add_space(log
, need_bytes
);
2603 tail_lsn
= log
->l_tail_lsn
;
2605 * Check to make sure the grant write head didn't just over lap the
2606 * tail. If the cycles are the same, we can't be overlapping.
2607 * Otherwise, make sure that the cycles differ by exactly one and
2608 * check the byte count.
2610 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2611 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2612 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2615 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: exit");
2616 xlog_verify_grant_head(log
, 1);
2617 spin_unlock(&log
->l_grant_lock
);
2621 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2622 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2623 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: err_ret");
2625 * If we are failing, make sure the ticket doesn't have any
2626 * current reservations. We don't want to add this back when
2627 * the ticket/transaction gets cancelled.
2629 tic
->t_curr_res
= 0;
2630 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2631 spin_unlock(&log
->l_grant_lock
);
2632 return XFS_ERROR(EIO
);
2633 } /* xlog_grant_log_space */
2637 * Replenish the byte reservation required by moving the grant write head.
2642 xlog_regrant_write_log_space(xlog_t
*log
,
2645 int free_bytes
, need_bytes
;
2646 xlog_ticket_t
*ntic
;
2651 tic
->t_curr_res
= tic
->t_unit_res
;
2652 xlog_tic_reset_res(tic
);
2658 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2659 panic("regrant Recovery problem");
2662 spin_lock(&log
->l_grant_lock
);
2663 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: enter");
2665 if (XLOG_FORCED_SHUTDOWN(log
))
2668 /* If there are other waiters on the queue then give them a
2669 * chance at logspace before us. Wake up the first waiters,
2670 * if we do not wake up all the waiters then go to sleep waiting
2671 * for more free space, otherwise try to get some space for
2675 if ((ntic
= log
->l_write_headq
)) {
2676 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2677 log
->l_grant_write_bytes
);
2679 ASSERT(ntic
->t_flags
& XLOG_TIC_PERM_RESERV
);
2681 if (free_bytes
< ntic
->t_unit_res
)
2683 free_bytes
-= ntic
->t_unit_res
;
2684 sv_signal(&ntic
->t_wait
);
2685 ntic
= ntic
->t_next
;
2686 } while (ntic
!= log
->l_write_headq
);
2688 if (ntic
!= log
->l_write_headq
) {
2689 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2690 xlog_ins_ticketq(&log
->l_write_headq
, tic
);
2692 xlog_trace_loggrant(log
, tic
,
2693 "xlog_regrant_write_log_space: sleep 1");
2694 XFS_STATS_INC(xs_sleep_logspace
);
2695 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
,
2696 &log
->l_grant_lock
, s
);
2698 /* If we're shutting down, this tic is already
2700 if (XLOG_FORCED_SHUTDOWN(log
)) {
2701 spin_lock(&log
->l_grant_lock
);
2705 xlog_trace_loggrant(log
, tic
,
2706 "xlog_regrant_write_log_space: wake 1");
2707 xlog_grant_push_ail(log
->l_mp
, tic
->t_unit_res
);
2708 spin_lock(&log
->l_grant_lock
);
2712 need_bytes
= tic
->t_unit_res
;
2715 if (XLOG_FORCED_SHUTDOWN(log
))
2718 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2719 log
->l_grant_write_bytes
);
2720 if (free_bytes
< need_bytes
) {
2721 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2722 xlog_ins_ticketq(&log
->l_write_headq
, tic
);
2723 XFS_STATS_INC(xs_sleep_logspace
);
2724 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2726 /* If we're shutting down, this tic is already off the queue */
2727 if (XLOG_FORCED_SHUTDOWN(log
)) {
2728 spin_lock(&log
->l_grant_lock
);
2732 xlog_trace_loggrant(log
, tic
,
2733 "xlog_regrant_write_log_space: wake 2");
2734 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2735 spin_lock(&log
->l_grant_lock
);
2737 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2738 xlog_del_ticketq(&log
->l_write_headq
, tic
);
2740 /* we've got enough space */
2741 xlog_grant_add_space_write(log
, need_bytes
);
2743 tail_lsn
= log
->l_tail_lsn
;
2744 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2745 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2746 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2750 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: exit");
2751 xlog_verify_grant_head(log
, 1);
2752 spin_unlock(&log
->l_grant_lock
);
2757 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2758 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2759 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: err_ret");
2761 * If we are failing, make sure the ticket doesn't have any
2762 * current reservations. We don't want to add this back when
2763 * the ticket/transaction gets cancelled.
2765 tic
->t_curr_res
= 0;
2766 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2767 spin_unlock(&log
->l_grant_lock
);
2768 return XFS_ERROR(EIO
);
2769 } /* xlog_regrant_write_log_space */
2772 /* The first cnt-1 times through here we don't need to
2773 * move the grant write head because the permanent
2774 * reservation has reserved cnt times the unit amount.
2775 * Release part of current permanent unit reservation and
2776 * reset current reservation to be one units worth. Also
2777 * move grant reservation head forward.
2780 xlog_regrant_reserve_log_space(xlog_t
*log
,
2781 xlog_ticket_t
*ticket
)
2783 xlog_trace_loggrant(log
, ticket
,
2784 "xlog_regrant_reserve_log_space: enter");
2785 if (ticket
->t_cnt
> 0)
2788 spin_lock(&log
->l_grant_lock
);
2789 xlog_grant_sub_space(log
, ticket
->t_curr_res
);
2790 ticket
->t_curr_res
= ticket
->t_unit_res
;
2791 xlog_tic_reset_res(ticket
);
2792 xlog_trace_loggrant(log
, ticket
,
2793 "xlog_regrant_reserve_log_space: sub current res");
2794 xlog_verify_grant_head(log
, 1);
2796 /* just return if we still have some of the pre-reserved space */
2797 if (ticket
->t_cnt
> 0) {
2798 spin_unlock(&log
->l_grant_lock
);
2802 xlog_grant_add_space_reserve(log
, ticket
->t_unit_res
);
2803 xlog_trace_loggrant(log
, ticket
,
2804 "xlog_regrant_reserve_log_space: exit");
2805 xlog_verify_grant_head(log
, 0);
2806 spin_unlock(&log
->l_grant_lock
);
2807 ticket
->t_curr_res
= ticket
->t_unit_res
;
2808 xlog_tic_reset_res(ticket
);
2809 } /* xlog_regrant_reserve_log_space */
2813 * Give back the space left from a reservation.
2815 * All the information we need to make a correct determination of space left
2816 * is present. For non-permanent reservations, things are quite easy. The
2817 * count should have been decremented to zero. We only need to deal with the
2818 * space remaining in the current reservation part of the ticket. If the
2819 * ticket contains a permanent reservation, there may be left over space which
2820 * needs to be released. A count of N means that N-1 refills of the current
2821 * reservation can be done before we need to ask for more space. The first
2822 * one goes to fill up the first current reservation. Once we run out of
2823 * space, the count will stay at zero and the only space remaining will be
2824 * in the current reservation field.
2827 xlog_ungrant_log_space(xlog_t
*log
,
2828 xlog_ticket_t
*ticket
)
2830 if (ticket
->t_cnt
> 0)
2833 spin_lock(&log
->l_grant_lock
);
2834 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: enter");
2836 xlog_grant_sub_space(log
, ticket
->t_curr_res
);
2838 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: sub current");
2840 /* If this is a permanent reservation ticket, we may be able to free
2841 * up more space based on the remaining count.
2843 if (ticket
->t_cnt
> 0) {
2844 ASSERT(ticket
->t_flags
& XLOG_TIC_PERM_RESERV
);
2845 xlog_grant_sub_space(log
, ticket
->t_unit_res
*ticket
->t_cnt
);
2848 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: exit");
2849 xlog_verify_grant_head(log
, 1);
2850 spin_unlock(&log
->l_grant_lock
);
2851 xfs_log_move_tail(log
->l_mp
, 1);
2852 } /* xlog_ungrant_log_space */
2856 * Flush iclog to disk if this is the last reference to the given iclog and
2857 * the WANT_SYNC bit is set.
2859 * When this function is entered, the iclog is not necessarily in the
2860 * WANT_SYNC state. It may be sitting around waiting to get filled.
2865 xlog_state_release_iclog(
2867 xlog_in_core_t
*iclog
)
2869 int sync
= 0; /* do we sync? */
2871 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
2872 return XFS_ERROR(EIO
);
2874 ASSERT(atomic_read(&iclog
->ic_refcnt
) > 0);
2875 if (!atomic_dec_and_lock(&iclog
->ic_refcnt
, &log
->l_icloglock
))
2878 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2879 spin_unlock(&log
->l_icloglock
);
2880 return XFS_ERROR(EIO
);
2882 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2883 iclog
->ic_state
== XLOG_STATE_WANT_SYNC
);
2885 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
) {
2886 /* update tail before writing to iclog */
2887 xlog_assign_tail_lsn(log
->l_mp
);
2889 iclog
->ic_state
= XLOG_STATE_SYNCING
;
2890 iclog
->ic_header
.h_tail_lsn
= cpu_to_be64(log
->l_tail_lsn
);
2891 xlog_verify_tail_lsn(log
, iclog
, log
->l_tail_lsn
);
2892 /* cycle incremented when incrementing curr_block */
2894 spin_unlock(&log
->l_icloglock
);
2897 * We let the log lock go, so it's possible that we hit a log I/O
2898 * error or some other SHUTDOWN condition that marks the iclog
2899 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2900 * this iclog has consistent data, so we ignore IOERROR
2901 * flags after this point.
2904 return xlog_sync(log
, iclog
);
2906 } /* xlog_state_release_iclog */
2910 * This routine will mark the current iclog in the ring as WANT_SYNC
2911 * and move the current iclog pointer to the next iclog in the ring.
2912 * When this routine is called from xlog_state_get_iclog_space(), the
2913 * exact size of the iclog has not yet been determined. All we know is
2914 * that every data block. We have run out of space in this log record.
2917 xlog_state_switch_iclogs(xlog_t
*log
,
2918 xlog_in_core_t
*iclog
,
2921 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
);
2923 eventual_size
= iclog
->ic_offset
;
2924 iclog
->ic_state
= XLOG_STATE_WANT_SYNC
;
2925 iclog
->ic_header
.h_prev_block
= cpu_to_be32(log
->l_prev_block
);
2926 log
->l_prev_block
= log
->l_curr_block
;
2927 log
->l_prev_cycle
= log
->l_curr_cycle
;
2929 /* roll log?: ic_offset changed later */
2930 log
->l_curr_block
+= BTOBB(eventual_size
)+BTOBB(log
->l_iclog_hsize
);
2932 /* Round up to next log-sunit */
2933 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
2934 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
2935 __uint32_t sunit_bb
= BTOBB(log
->l_mp
->m_sb
.sb_logsunit
);
2936 log
->l_curr_block
= roundup(log
->l_curr_block
, sunit_bb
);
2939 if (log
->l_curr_block
>= log
->l_logBBsize
) {
2940 log
->l_curr_cycle
++;
2941 if (log
->l_curr_cycle
== XLOG_HEADER_MAGIC_NUM
)
2942 log
->l_curr_cycle
++;
2943 log
->l_curr_block
-= log
->l_logBBsize
;
2944 ASSERT(log
->l_curr_block
>= 0);
2946 ASSERT(iclog
== log
->l_iclog
);
2947 log
->l_iclog
= iclog
->ic_next
;
2948 } /* xlog_state_switch_iclogs */
2952 * Write out all data in the in-core log as of this exact moment in time.
2954 * Data may be written to the in-core log during this call. However,
2955 * we don't guarantee this data will be written out. A change from past
2956 * implementation means this routine will *not* write out zero length LRs.
2958 * Basically, we try and perform an intelligent scan of the in-core logs.
2959 * If we determine there is no flushable data, we just return. There is no
2960 * flushable data if:
2962 * 1. the current iclog is active and has no data; the previous iclog
2963 * is in the active or dirty state.
2964 * 2. the current iclog is drity, and the previous iclog is in the
2965 * active or dirty state.
2969 * 1. the current iclog is not in the active nor dirty state.
2970 * 2. the current iclog dirty, and the previous iclog is not in the
2971 * active nor dirty state.
2972 * 3. the current iclog is active, and there is another thread writing
2973 * to this particular iclog.
2974 * 4. a) the current iclog is active and has no other writers
2975 * b) when we return from flushing out this iclog, it is still
2976 * not in the active nor dirty state.
2979 xlog_state_sync_all(xlog_t
*log
, uint flags
, int *log_flushed
)
2981 xlog_in_core_t
*iclog
;
2984 spin_lock(&log
->l_icloglock
);
2986 iclog
= log
->l_iclog
;
2987 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2988 spin_unlock(&log
->l_icloglock
);
2989 return XFS_ERROR(EIO
);
2992 /* If the head iclog is not active nor dirty, we just attach
2993 * ourselves to the head and go to sleep.
2995 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2996 iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2998 * If the head is dirty or (active and empty), then
2999 * we need to look at the previous iclog. If the previous
3000 * iclog is active or dirty we are done. There is nothing
3001 * to sync out. Otherwise, we attach ourselves to the
3002 * previous iclog and go to sleep.
3004 if (iclog
->ic_state
== XLOG_STATE_DIRTY
||
3005 (atomic_read(&iclog
->ic_refcnt
) == 0
3006 && iclog
->ic_offset
== 0)) {
3007 iclog
= iclog
->ic_prev
;
3008 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
3009 iclog
->ic_state
== XLOG_STATE_DIRTY
)
3014 if (atomic_read(&iclog
->ic_refcnt
) == 0) {
3015 /* We are the only one with access to this
3016 * iclog. Flush it out now. There should
3017 * be a roundoff of zero to show that someone
3018 * has already taken care of the roundoff from
3019 * the previous sync.
3021 atomic_inc(&iclog
->ic_refcnt
);
3022 lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
3023 xlog_state_switch_iclogs(log
, iclog
, 0);
3024 spin_unlock(&log
->l_icloglock
);
3026 if (xlog_state_release_iclog(log
, iclog
))
3027 return XFS_ERROR(EIO
);
3029 spin_lock(&log
->l_icloglock
);
3030 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) == lsn
&&
3031 iclog
->ic_state
!= XLOG_STATE_DIRTY
)
3036 /* Someone else is writing to this iclog.
3037 * Use its call to flush out the data. However,
3038 * the other thread may not force out this LR,
3039 * so we mark it WANT_SYNC.
3041 xlog_state_switch_iclogs(log
, iclog
, 0);
3047 /* By the time we come around again, the iclog could've been filled
3048 * which would give it another lsn. If we have a new lsn, just
3049 * return because the relevant data has been flushed.
3052 if (flags
& XFS_LOG_SYNC
) {
3054 * We must check if we're shutting down here, before
3055 * we wait, while we're holding the l_icloglock.
3056 * Then we check again after waking up, in case our
3057 * sleep was disturbed by a bad news.
3059 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3060 spin_unlock(&log
->l_icloglock
);
3061 return XFS_ERROR(EIO
);
3063 XFS_STATS_INC(xs_log_force_sleep
);
3064 sv_wait(&iclog
->ic_force_wait
, PINOD
, &log
->l_icloglock
, s
);
3066 * No need to grab the log lock here since we're
3067 * only deciding whether or not to return EIO
3068 * and the memory read should be atomic.
3070 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3071 return XFS_ERROR(EIO
);
3077 spin_unlock(&log
->l_icloglock
);
3080 } /* xlog_state_sync_all */
3084 * Used by code which implements synchronous log forces.
3086 * Find in-core log with lsn.
3087 * If it is in the DIRTY state, just return.
3088 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3089 * state and go to sleep or return.
3090 * If it is in any other state, go to sleep or return.
3092 * If filesystem activity goes to zero, the iclog will get flushed only by
3096 xlog_state_sync(xlog_t
*log
,
3101 xlog_in_core_t
*iclog
;
3102 int already_slept
= 0;
3105 spin_lock(&log
->l_icloglock
);
3106 iclog
= log
->l_iclog
;
3108 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3109 spin_unlock(&log
->l_icloglock
);
3110 return XFS_ERROR(EIO
);
3114 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) != lsn
) {
3115 iclog
= iclog
->ic_next
;
3119 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
3120 spin_unlock(&log
->l_icloglock
);
3124 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3126 * We sleep here if we haven't already slept (e.g.
3127 * this is the first time we've looked at the correct
3128 * iclog buf) and the buffer before us is going to
3129 * be sync'ed. The reason for this is that if we
3130 * are doing sync transactions here, by waiting for
3131 * the previous I/O to complete, we can allow a few
3132 * more transactions into this iclog before we close
3135 * Otherwise, we mark the buffer WANT_SYNC, and bump
3136 * up the refcnt so we can release the log (which drops
3137 * the ref count). The state switch keeps new transaction
3138 * commits from using this buffer. When the current commits
3139 * finish writing into the buffer, the refcount will drop to
3140 * zero and the buffer will go out then.
3142 if (!already_slept
&&
3143 (iclog
->ic_prev
->ic_state
& (XLOG_STATE_WANT_SYNC
|
3144 XLOG_STATE_SYNCING
))) {
3145 ASSERT(!(iclog
->ic_state
& XLOG_STATE_IOERROR
));
3146 XFS_STATS_INC(xs_log_force_sleep
);
3147 sv_wait(&iclog
->ic_prev
->ic_write_wait
, PSWP
,
3148 &log
->l_icloglock
, s
);
3153 atomic_inc(&iclog
->ic_refcnt
);
3154 xlog_state_switch_iclogs(log
, iclog
, 0);
3155 spin_unlock(&log
->l_icloglock
);
3156 if (xlog_state_release_iclog(log
, iclog
))
3157 return XFS_ERROR(EIO
);
3159 spin_lock(&log
->l_icloglock
);
3163 if ((flags
& XFS_LOG_SYNC
) && /* sleep */
3164 !(iclog
->ic_state
& (XLOG_STATE_ACTIVE
| XLOG_STATE_DIRTY
))) {
3167 * Don't wait on completion if we know that we've
3168 * gotten a log write error.
3170 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3171 spin_unlock(&log
->l_icloglock
);
3172 return XFS_ERROR(EIO
);
3174 XFS_STATS_INC(xs_log_force_sleep
);
3175 sv_wait(&iclog
->ic_force_wait
, PSWP
, &log
->l_icloglock
, s
);
3177 * No need to grab the log lock here since we're
3178 * only deciding whether or not to return EIO
3179 * and the memory read should be atomic.
3181 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3182 return XFS_ERROR(EIO
);
3184 } else { /* just return */
3185 spin_unlock(&log
->l_icloglock
);
3189 } while (iclog
!= log
->l_iclog
);
3191 spin_unlock(&log
->l_icloglock
);
3193 } /* xlog_state_sync */
3197 * Called when we want to mark the current iclog as being ready to sync to
3201 xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
)
3203 spin_lock(&log
->l_icloglock
);
3205 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3206 xlog_state_switch_iclogs(log
, iclog
, 0);
3208 ASSERT(iclog
->ic_state
&
3209 (XLOG_STATE_WANT_SYNC
|XLOG_STATE_IOERROR
));
3212 spin_unlock(&log
->l_icloglock
);
3213 } /* xlog_state_want_sync */
3217 /*****************************************************************************
3221 *****************************************************************************
3225 * Free a used ticket.
3228 xlog_ticket_put(xlog_t
*log
,
3229 xlog_ticket_t
*ticket
)
3231 sv_destroy(&ticket
->t_wait
);
3232 kmem_zone_free(xfs_log_ticket_zone
, ticket
);
3233 } /* xlog_ticket_put */
3237 * Allocate and initialise a new log ticket.
3239 STATIC xlog_ticket_t
*
3240 xlog_ticket_get(xlog_t
*log
,
3249 tic
= kmem_zone_zalloc(xfs_log_ticket_zone
, KM_SLEEP
|KM_MAYFAIL
);
3254 * Permanent reservations have up to 'cnt'-1 active log operations
3255 * in the log. A unit in this case is the amount of space for one
3256 * of these log operations. Normal reservations have a cnt of 1
3257 * and their unit amount is the total amount of space required.
3259 * The following lines of code account for non-transaction data
3260 * which occupy space in the on-disk log.
3262 * Normal form of a transaction is:
3263 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3264 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3266 * We need to account for all the leadup data and trailer data
3267 * around the transaction data.
3268 * And then we need to account for the worst case in terms of using
3270 * The worst case will happen if:
3271 * - the placement of the transaction happens to be such that the
3272 * roundoff is at its maximum
3273 * - the transaction data is synced before the commit record is synced
3274 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3275 * Therefore the commit record is in its own Log Record.
3276 * This can happen as the commit record is called with its
3277 * own region to xlog_write().
3278 * This then means that in the worst case, roundoff can happen for
3279 * the commit-rec as well.
3280 * The commit-rec is smaller than padding in this scenario and so it is
3281 * not added separately.
3284 /* for trans header */
3285 unit_bytes
+= sizeof(xlog_op_header_t
);
3286 unit_bytes
+= sizeof(xfs_trans_header_t
);
3289 unit_bytes
+= sizeof(xlog_op_header_t
);
3291 /* for LR headers */
3292 num_headers
= ((unit_bytes
+ log
->l_iclog_size
-1) >> log
->l_iclog_size_log
);
3293 unit_bytes
+= log
->l_iclog_hsize
* num_headers
;
3295 /* for commit-rec LR header - note: padding will subsume the ophdr */
3296 unit_bytes
+= log
->l_iclog_hsize
;
3298 /* for split-recs - ophdrs added when data split over LRs */
3299 unit_bytes
+= sizeof(xlog_op_header_t
) * num_headers
;
3301 /* for roundoff padding for transaction data and one for commit record */
3302 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
3303 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
3304 /* log su roundoff */
3305 unit_bytes
+= 2*log
->l_mp
->m_sb
.sb_logsunit
;
3308 unit_bytes
+= 2*BBSIZE
;
3311 tic
->t_unit_res
= unit_bytes
;
3312 tic
->t_curr_res
= unit_bytes
;
3315 tic
->t_tid
= (xlog_tid_t
)((__psint_t
)tic
& 0xffffffff);
3316 tic
->t_clientid
= client
;
3317 tic
->t_flags
= XLOG_TIC_INITED
;
3318 tic
->t_trans_type
= 0;
3319 if (xflags
& XFS_LOG_PERM_RESERV
)
3320 tic
->t_flags
|= XLOG_TIC_PERM_RESERV
;
3321 sv_init(&(tic
->t_wait
), SV_DEFAULT
, "logtick");
3323 xlog_tic_reset_res(tic
);
3326 } /* xlog_ticket_get */
3329 /******************************************************************************
3331 * Log debug routines
3333 ******************************************************************************
3337 * Make sure that the destination ptr is within the valid data region of
3338 * one of the iclogs. This uses backup pointers stored in a different
3339 * part of the log in case we trash the log structure.
3342 xlog_verify_dest_ptr(xlog_t
*log
,
3348 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3349 if (ptr
>= (__psint_t
)log
->l_iclog_bak
[i
] &&
3350 ptr
<= (__psint_t
)log
->l_iclog_bak
[i
]+log
->l_iclog_size
)
3354 xlog_panic("xlog_verify_dest_ptr: invalid ptr");
3355 } /* xlog_verify_dest_ptr */
3358 xlog_verify_grant_head(xlog_t
*log
, int equals
)
3360 if (log
->l_grant_reserve_cycle
== log
->l_grant_write_cycle
) {
3362 ASSERT(log
->l_grant_reserve_bytes
>= log
->l_grant_write_bytes
);
3364 ASSERT(log
->l_grant_reserve_bytes
> log
->l_grant_write_bytes
);
3366 ASSERT(log
->l_grant_reserve_cycle
-1 == log
->l_grant_write_cycle
);
3367 ASSERT(log
->l_grant_write_bytes
>= log
->l_grant_reserve_bytes
);
3369 } /* xlog_verify_grant_head */
3371 /* check if it will fit */
3373 xlog_verify_tail_lsn(xlog_t
*log
,
3374 xlog_in_core_t
*iclog
,
3379 if (CYCLE_LSN(tail_lsn
) == log
->l_prev_cycle
) {
3381 log
->l_logBBsize
- (log
->l_prev_block
- BLOCK_LSN(tail_lsn
));
3382 if (blocks
< BTOBB(iclog
->ic_offset
)+BTOBB(log
->l_iclog_hsize
))
3383 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3385 ASSERT(CYCLE_LSN(tail_lsn
)+1 == log
->l_prev_cycle
);
3387 if (BLOCK_LSN(tail_lsn
) == log
->l_prev_block
)
3388 xlog_panic("xlog_verify_tail_lsn: tail wrapped");
3390 blocks
= BLOCK_LSN(tail_lsn
) - log
->l_prev_block
;
3391 if (blocks
< BTOBB(iclog
->ic_offset
) + 1)
3392 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3394 } /* xlog_verify_tail_lsn */
3397 * Perform a number of checks on the iclog before writing to disk.
3399 * 1. Make sure the iclogs are still circular
3400 * 2. Make sure we have a good magic number
3401 * 3. Make sure we don't have magic numbers in the data
3402 * 4. Check fields of each log operation header for:
3403 * A. Valid client identifier
3404 * B. tid ptr value falls in valid ptr space (user space code)
3405 * C. Length in log record header is correct according to the
3406 * individual operation headers within record.
3407 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3408 * log, check the preceding blocks of the physical log to make sure all
3409 * the cycle numbers agree with the current cycle number.
3412 xlog_verify_iclog(xlog_t
*log
,
3413 xlog_in_core_t
*iclog
,
3417 xlog_op_header_t
*ophead
;
3418 xlog_in_core_t
*icptr
;
3419 xlog_in_core_2_t
*xhdr
;
3421 xfs_caddr_t base_ptr
;
3422 __psint_t field_offset
;
3424 int len
, i
, j
, k
, op_len
;
3427 /* check validity of iclog pointers */
3428 spin_lock(&log
->l_icloglock
);
3429 icptr
= log
->l_iclog
;
3430 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3432 xlog_panic("xlog_verify_iclog: invalid ptr");
3433 icptr
= icptr
->ic_next
;
3435 if (icptr
!= log
->l_iclog
)
3436 xlog_panic("xlog_verify_iclog: corrupt iclog ring");
3437 spin_unlock(&log
->l_icloglock
);
3439 /* check log magic numbers */
3440 if (be32_to_cpu(iclog
->ic_header
.h_magicno
) != XLOG_HEADER_MAGIC_NUM
)
3441 xlog_panic("xlog_verify_iclog: invalid magic num");
3443 ptr
= (xfs_caddr_t
) &iclog
->ic_header
;
3444 for (ptr
+= BBSIZE
; ptr
< ((xfs_caddr_t
)&iclog
->ic_header
) + count
;
3446 if (be32_to_cpu(*(__be32
*)ptr
) == XLOG_HEADER_MAGIC_NUM
)
3447 xlog_panic("xlog_verify_iclog: unexpected magic num");
3451 len
= be32_to_cpu(iclog
->ic_header
.h_num_logops
);
3452 ptr
= iclog
->ic_datap
;
3454 ophead
= (xlog_op_header_t
*)ptr
;
3455 xhdr
= (xlog_in_core_2_t
*)&iclog
->ic_header
;
3456 for (i
= 0; i
< len
; i
++) {
3457 ophead
= (xlog_op_header_t
*)ptr
;
3459 /* clientid is only 1 byte */
3460 field_offset
= (__psint_t
)
3461 ((xfs_caddr_t
)&(ophead
->oh_clientid
) - base_ptr
);
3462 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3463 clientid
= ophead
->oh_clientid
;
3465 idx
= BTOBBT((xfs_caddr_t
)&(ophead
->oh_clientid
) - iclog
->ic_datap
);
3466 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3467 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3468 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3469 clientid
= xlog_get_client_id(
3470 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3472 clientid
= xlog_get_client_id(
3473 iclog
->ic_header
.h_cycle_data
[idx
]);
3476 if (clientid
!= XFS_TRANSACTION
&& clientid
!= XFS_LOG
)
3477 cmn_err(CE_WARN
, "xlog_verify_iclog: "
3478 "invalid clientid %d op 0x%p offset 0x%lx",
3479 clientid
, ophead
, (unsigned long)field_offset
);
3482 field_offset
= (__psint_t
)
3483 ((xfs_caddr_t
)&(ophead
->oh_len
) - base_ptr
);
3484 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3485 op_len
= be32_to_cpu(ophead
->oh_len
);
3487 idx
= BTOBBT((__psint_t
)&ophead
->oh_len
-
3488 (__psint_t
)iclog
->ic_datap
);
3489 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3490 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3491 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3492 op_len
= be32_to_cpu(xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3494 op_len
= be32_to_cpu(iclog
->ic_header
.h_cycle_data
[idx
]);
3497 ptr
+= sizeof(xlog_op_header_t
) + op_len
;
3499 } /* xlog_verify_iclog */
3503 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
3509 xlog_in_core_t
*iclog
, *ic
;
3511 iclog
= log
->l_iclog
;
3512 if (! (iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3514 * Mark all the incore logs IOERROR.
3515 * From now on, no log flushes will result.
3519 ic
->ic_state
= XLOG_STATE_IOERROR
;
3521 } while (ic
!= iclog
);
3525 * Return non-zero, if state transition has already happened.
3531 * This is called from xfs_force_shutdown, when we're forcibly
3532 * shutting down the filesystem, typically because of an IO error.
3533 * Our main objectives here are to make sure that:
3534 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3535 * parties to find out, 'atomically'.
3536 * b. those who're sleeping on log reservations, pinned objects and
3537 * other resources get woken up, and be told the bad news.
3538 * c. nothing new gets queued up after (a) and (b) are done.
3539 * d. if !logerror, flush the iclogs to disk, then seal them off
3543 xfs_log_force_umount(
3544 struct xfs_mount
*mp
,
3555 * If this happens during log recovery, don't worry about
3556 * locking; the log isn't open for business yet.
3559 log
->l_flags
& XLOG_ACTIVE_RECOVERY
) {
3560 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3561 XFS_BUF_DONE(mp
->m_sb_bp
);
3566 * Somebody could've already done the hard work for us.
3567 * No need to get locks for this.
3569 if (logerror
&& log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3570 ASSERT(XLOG_FORCED_SHUTDOWN(log
));
3575 * We must hold both the GRANT lock and the LOG lock,
3576 * before we mark the filesystem SHUTDOWN and wake
3577 * everybody up to tell the bad news.
3579 spin_lock(&log
->l_icloglock
);
3580 spin_lock(&log
->l_grant_lock
);
3581 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3582 XFS_BUF_DONE(mp
->m_sb_bp
);
3584 * This flag is sort of redundant because of the mount flag, but
3585 * it's good to maintain the separation between the log and the rest
3588 log
->l_flags
|= XLOG_IO_ERROR
;
3591 * If we hit a log error, we want to mark all the iclogs IOERROR
3592 * while we're still holding the loglock.
3595 retval
= xlog_state_ioerror(log
);
3596 spin_unlock(&log
->l_icloglock
);
3599 * We don't want anybody waiting for log reservations
3600 * after this. That means we have to wake up everybody
3601 * queued up on reserve_headq as well as write_headq.
3602 * In addition, we make sure in xlog_{re}grant_log_space
3603 * that we don't enqueue anything once the SHUTDOWN flag
3604 * is set, and this action is protected by the GRANTLOCK.
3606 if ((tic
= log
->l_reserve_headq
)) {
3608 sv_signal(&tic
->t_wait
);
3610 } while (tic
!= log
->l_reserve_headq
);
3613 if ((tic
= log
->l_write_headq
)) {
3615 sv_signal(&tic
->t_wait
);
3617 } while (tic
!= log
->l_write_headq
);
3619 spin_unlock(&log
->l_grant_lock
);
3621 if (! (log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3624 * Force the incore logs to disk before shutting the
3625 * log down completely.
3627 xlog_state_sync_all(log
, XFS_LOG_FORCE
|XFS_LOG_SYNC
, &dummy
);
3628 spin_lock(&log
->l_icloglock
);
3629 retval
= xlog_state_ioerror(log
);
3630 spin_unlock(&log
->l_icloglock
);
3633 * Wake up everybody waiting on xfs_log_force.
3634 * Callback all log item committed functions as if the
3635 * log writes were completed.
3637 xlog_state_do_callback(log
, XFS_LI_ABORTED
, NULL
);
3639 #ifdef XFSERRORDEBUG
3641 xlog_in_core_t
*iclog
;
3643 spin_lock(&log
->l_icloglock
);
3644 iclog
= log
->l_iclog
;
3646 ASSERT(iclog
->ic_callback
== 0);
3647 iclog
= iclog
->ic_next
;
3648 } while (iclog
!= log
->l_iclog
);
3649 spin_unlock(&log
->l_icloglock
);
3652 /* return non-zero if log IOERROR transition had already happened */
3657 xlog_iclogs_empty(xlog_t
*log
)
3659 xlog_in_core_t
*iclog
;
3661 iclog
= log
->l_iclog
;
3663 /* endianness does not matter here, zero is zero in
3666 if (iclog
->ic_header
.h_num_logops
)
3668 iclog
= iclog
->ic_next
;
3669 } while (iclog
!= log
->l_iclog
);