2 * Copyright (c) 2000-2003 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
23 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_trans_priv.h"
51 * returns the number of iovecs needed to log the given dquot item.
55 xfs_qm_dquot_logitem_size(
56 xfs_dq_logitem_t
*logitem
)
59 * we need only two iovecs, one for the format, one for the real thing
65 * fills in the vector of log iovecs for the given dquot log item.
68 xfs_qm_dquot_logitem_format(
69 xfs_dq_logitem_t
*logitem
,
70 xfs_log_iovec_t
*logvec
)
73 ASSERT(logitem
->qli_dquot
);
75 logvec
->i_addr
= (xfs_caddr_t
)&logitem
->qli_format
;
76 logvec
->i_len
= sizeof(xfs_dq_logformat_t
);
77 XLOG_VEC_SET_TYPE(logvec
, XLOG_REG_TYPE_QFORMAT
);
79 logvec
->i_addr
= (xfs_caddr_t
)&logitem
->qli_dquot
->q_core
;
80 logvec
->i_len
= sizeof(xfs_disk_dquot_t
);
81 XLOG_VEC_SET_TYPE(logvec
, XLOG_REG_TYPE_DQUOT
);
83 ASSERT(2 == logitem
->qli_item
.li_desc
->lid_size
);
84 logitem
->qli_format
.qlf_size
= 2;
89 * Increment the pin count of the given dquot.
92 xfs_qm_dquot_logitem_pin(
93 xfs_dq_logitem_t
*logitem
)
95 xfs_dquot_t
*dqp
= logitem
->qli_dquot
;
97 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
98 atomic_inc(&dqp
->q_pincount
);
102 * Decrement the pin count of the given dquot, and wake up
103 * anyone in xfs_dqwait_unpin() if the count goes to 0. The
104 * dquot must have been previously pinned with a call to
105 * xfs_qm_dquot_logitem_pin().
109 xfs_qm_dquot_logitem_unpin(
110 xfs_dq_logitem_t
*logitem
,
113 xfs_dquot_t
*dqp
= logitem
->qli_dquot
;
115 ASSERT(atomic_read(&dqp
->q_pincount
) > 0);
116 if (atomic_dec_and_test(&dqp
->q_pincount
))
117 wake_up(&dqp
->q_pinwait
);
122 xfs_qm_dquot_logitem_unpin_remove(
123 xfs_dq_logitem_t
*logitem
,
126 xfs_qm_dquot_logitem_unpin(logitem
, 0);
130 * Given the logitem, this writes the corresponding dquot entry to disk
131 * asynchronously. This is called with the dquot entry securely locked;
132 * we simply get xfs_qm_dqflush() to do the work, and unlock the dquot
136 xfs_qm_dquot_logitem_push(
137 xfs_dq_logitem_t
*logitem
)
142 dqp
= logitem
->qli_dquot
;
144 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
145 ASSERT(!completion_done(&dqp
->q_flush
));
148 * Since we were able to lock the dquot's flush lock and
149 * we found it on the AIL, the dquot must be dirty. This
150 * is because the dquot is removed from the AIL while still
151 * holding the flush lock in xfs_dqflush_done(). Thus, if
152 * we found it in the AIL and were able to obtain the flush
153 * lock without sleeping, then there must not have been
154 * anyone in the process of flushing the dquot.
156 error
= xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
158 xfs_fs_cmn_err(CE_WARN
, dqp
->q_mount
,
159 "xfs_qm_dquot_logitem_push: push error %d on dqp %p",
166 xfs_qm_dquot_logitem_committed(
171 * We always re-log the entire dquot when it becomes dirty,
172 * so, the latest copy _is_ the only one that matters.
179 * This is called to wait for the given dquot to be unpinned.
180 * Most of these pin/unpin routines are plagiarized from inode code.
186 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
187 if (atomic_read(&dqp
->q_pincount
) == 0)
191 * Give the log a push so we don't wait here too long.
193 xfs_log_force(dqp
->q_mount
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
194 wait_event(dqp
->q_pinwait
, (atomic_read(&dqp
->q_pincount
) == 0));
198 * This is called when IOP_TRYLOCK returns XFS_ITEM_PUSHBUF to indicate that
199 * the dquot is locked by us, but the flush lock isn't. So, here we are
200 * going to see if the relevant dquot buffer is incore, waiting on DELWRI.
201 * If so, we want to push it out to help us take this item off the AIL as soon
204 * We must not be holding the AIL lock at this point. Calling incore() to
205 * search the buffer cache can be a time consuming thing, and AIL lock is a
209 xfs_qm_dquot_logitem_pushbuf(
210 xfs_dq_logitem_t
*qip
)
217 dqp
= qip
->qli_dquot
;
218 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
221 * The qli_pushbuf_flag keeps others from
222 * trying to duplicate our effort.
224 ASSERT(qip
->qli_pushbuf_flag
!= 0);
225 ASSERT(qip
->qli_push_owner
== current_pid());
228 * If flushlock isn't locked anymore, chances are that the
229 * inode flush completed and the inode was taken off the AIL.
232 if (completion_done(&dqp
->q_flush
) ||
233 ((qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
) == 0)) {
234 qip
->qli_pushbuf_flag
= 0;
239 bp
= xfs_incore(mp
->m_ddev_targp
, qip
->qli_format
.qlf_blkno
,
240 XFS_QI_DQCHUNKLEN(mp
),
243 if (XFS_BUF_ISDELAYWRITE(bp
)) {
244 dopush
= ((qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
) &&
245 !completion_done(&dqp
->q_flush
));
246 qip
->qli_pushbuf_flag
= 0;
249 if (XFS_BUF_ISPINNED(bp
)) {
250 xfs_log_force(mp
, (xfs_lsn_t
)0,
259 error
= xfs_bawrite(mp
, bp
);
261 xfs_fs_cmn_err(CE_WARN
, mp
,
262 "xfs_qm_dquot_logitem_pushbuf: pushbuf error %d on qip %p, bp %p",
268 qip
->qli_pushbuf_flag
= 0;
275 qip
->qli_pushbuf_flag
= 0;
280 * This is called to attempt to lock the dquot associated with this
281 * dquot log item. Don't sleep on the dquot lock or the flush lock.
282 * If the flush lock is already held, indicating that the dquot has
283 * been or is in the process of being flushed, then see if we can
284 * find the dquot's buffer in the buffer cache without sleeping. If
285 * we can and it is marked delayed write, then we want to send it out.
286 * We delay doing so until the push routine, though, to avoid sleeping
287 * in any device strategy routines.
290 xfs_qm_dquot_logitem_trylock(
291 xfs_dq_logitem_t
*qip
)
296 dqp
= qip
->qli_dquot
;
297 if (atomic_read(&dqp
->q_pincount
) > 0)
298 return (XFS_ITEM_PINNED
);
300 if (! xfs_qm_dqlock_nowait(dqp
))
301 return (XFS_ITEM_LOCKED
);
303 retval
= XFS_ITEM_SUCCESS
;
304 if (!xfs_dqflock_nowait(dqp
)) {
306 * The dquot is already being flushed. It may have been
307 * flushed delayed write, however, and we don't want to
308 * get stuck waiting for that to complete. So, we want to check
309 * to see if we can lock the dquot's buffer without sleeping.
310 * If we can and it is marked for delayed write, then we
311 * hold it and send it out from the push routine. We don't
312 * want to do that now since we might sleep in the device
313 * strategy routine. We also don't want to grab the buffer lock
314 * here because we'd like not to call into the buffer cache
315 * while holding the AIL lock.
316 * Make sure to only return PUSHBUF if we set pushbuf_flag
317 * ourselves. If someone else is doing it then we don't
318 * want to go to the push routine and duplicate their efforts.
320 if (qip
->qli_pushbuf_flag
== 0) {
321 qip
->qli_pushbuf_flag
= 1;
322 ASSERT(qip
->qli_format
.qlf_blkno
== dqp
->q_blkno
);
324 qip
->qli_push_owner
= current_pid();
327 * The dquot is left locked.
329 retval
= XFS_ITEM_PUSHBUF
;
331 retval
= XFS_ITEM_FLUSHING
;
332 xfs_dqunlock_nonotify(dqp
);
336 ASSERT(qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
);
342 * Unlock the dquot associated with the log item.
343 * Clear the fields of the dquot and dquot log item that
344 * are specific to the current transaction. If the
345 * hold flags is set, do not unlock the dquot.
348 xfs_qm_dquot_logitem_unlock(
349 xfs_dq_logitem_t
*ql
)
355 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
358 * Clear the transaction pointer in the dquot
360 dqp
->q_transp
= NULL
;
363 * dquots are never 'held' from getting unlocked at the end of
364 * a transaction. Their locking and unlocking is hidden inside the
365 * transaction layer, within trans_commit. Hence, no LI_HOLD flag
373 * this needs to stamp an lsn into the dquot, I think.
374 * rpc's that look at user dquot's would then have to
375 * push on the dependency recorded in the dquot
379 xfs_qm_dquot_logitem_committing(
388 * This is the ops vector for dquots
390 static struct xfs_item_ops xfs_dquot_item_ops
= {
391 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_size
,
392 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
393 xfs_qm_dquot_logitem_format
,
394 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_pin
,
395 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
396 xfs_qm_dquot_logitem_unpin
,
397 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*, xfs_trans_t
*))
398 xfs_qm_dquot_logitem_unpin_remove
,
399 .iop_trylock
= (uint(*)(xfs_log_item_t
*))
400 xfs_qm_dquot_logitem_trylock
,
401 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_unlock
,
402 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
403 xfs_qm_dquot_logitem_committed
,
404 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_push
,
405 .iop_pushbuf
= (void(*)(xfs_log_item_t
*))
406 xfs_qm_dquot_logitem_pushbuf
,
407 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
408 xfs_qm_dquot_logitem_committing
412 * Initialize the dquot log item for a newly allocated dquot.
413 * The dquot isn't locked at this point, but it isn't on any of the lists
414 * either, so we don't care.
417 xfs_qm_dquot_logitem_init(
418 struct xfs_dquot
*dqp
)
420 xfs_dq_logitem_t
*lp
;
421 lp
= &dqp
->q_logitem
;
423 lp
->qli_item
.li_type
= XFS_LI_DQUOT
;
424 lp
->qli_item
.li_ops
= &xfs_dquot_item_ops
;
425 lp
->qli_item
.li_mountp
= dqp
->q_mount
;
427 lp
->qli_format
.qlf_type
= XFS_LI_DQUOT
;
428 lp
->qli_format
.qlf_id
= be32_to_cpu(dqp
->q_core
.d_id
);
429 lp
->qli_format
.qlf_blkno
= dqp
->q_blkno
;
430 lp
->qli_format
.qlf_len
= 1;
432 * This is just the offset of this dquot within its buffer
433 * (which is currently 1 FSB and probably won't change).
434 * Hence 32 bits for this offset should be just fine.
435 * Alternatively, we can store (bufoffset / sizeof(xfs_dqblk_t))
436 * here, and recompute it at recovery time.
438 lp
->qli_format
.qlf_boffset
= (__uint32_t
)dqp
->q_bufoffset
;
441 /*------------------ QUOTAOFF LOG ITEMS -------------------*/
444 * This returns the number of iovecs needed to log the given quotaoff item.
445 * We only need 1 iovec for an quotaoff item. It just logs the
446 * quotaoff_log_format structure.
450 xfs_qm_qoff_logitem_size(xfs_qoff_logitem_t
*qf
)
456 * This is called to fill in the vector of log iovecs for the
457 * given quotaoff log item. We use only 1 iovec, and we point that
458 * at the quotaoff_log_format structure embedded in the quotaoff item.
459 * It is at this point that we assert that all of the extent
460 * slots in the quotaoff item have been filled.
463 xfs_qm_qoff_logitem_format(xfs_qoff_logitem_t
*qf
,
464 xfs_log_iovec_t
*log_vector
)
466 ASSERT(qf
->qql_format
.qf_type
== XFS_LI_QUOTAOFF
);
468 log_vector
->i_addr
= (xfs_caddr_t
)&(qf
->qql_format
);
469 log_vector
->i_len
= sizeof(xfs_qoff_logitem_t
);
470 XLOG_VEC_SET_TYPE(log_vector
, XLOG_REG_TYPE_QUOTAOFF
);
471 qf
->qql_format
.qf_size
= 1;
476 * Pinning has no meaning for an quotaoff item, so just return.
480 xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t
*qf
)
487 * Since pinning has no meaning for an quotaoff item, unpinning does
492 xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t
*qf
, int stale
)
499 xfs_qm_qoff_logitem_unpin_remove(xfs_qoff_logitem_t
*qf
, xfs_trans_t
*tp
)
505 * Quotaoff items have no locking, so just return success.
509 xfs_qm_qoff_logitem_trylock(xfs_qoff_logitem_t
*qf
)
511 return XFS_ITEM_LOCKED
;
515 * Quotaoff items have no locking or pushing, so return failure
516 * so that the caller doesn't bother with us.
520 xfs_qm_qoff_logitem_unlock(xfs_qoff_logitem_t
*qf
)
526 * The quotaoff-start-item is logged only once and cannot be moved in the log,
527 * so simply return the lsn at which it's been logged.
531 xfs_qm_qoff_logitem_committed(xfs_qoff_logitem_t
*qf
, xfs_lsn_t lsn
)
537 * There isn't much you can do to push on an quotaoff item. It is simply
538 * stuck waiting for the log to be flushed to disk.
542 xfs_qm_qoff_logitem_push(xfs_qoff_logitem_t
*qf
)
550 xfs_qm_qoffend_logitem_committed(
551 xfs_qoff_logitem_t
*qfe
,
554 xfs_qoff_logitem_t
*qfs
;
555 struct xfs_ail
*ailp
;
557 qfs
= qfe
->qql_start_lip
;
558 ailp
= qfs
->qql_item
.li_ailp
;
559 spin_lock(&ailp
->xa_lock
);
561 * Delete the qoff-start logitem from the AIL.
562 * xfs_trans_ail_delete() drops the AIL lock.
564 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)qfs
);
567 return (xfs_lsn_t
)-1;
571 * XXX rcc - don't know quite what to do with this. I think we can
572 * just ignore it. The only time that isn't the case is if we allow
573 * the client to somehow see that quotas have been turned off in which
574 * we can't allow that to get back until the quotaoff hits the disk.
575 * So how would that happen? Also, do we need different routines for
576 * quotaoff start and quotaoff end? I suspect the answer is yes but
577 * to be sure, I need to look at the recovery code and see how quota off
578 * recovery is handled (do we roll forward or back or do something else).
579 * If we roll forwards or backwards, then we need two separate routines,
580 * one that does nothing and one that stamps in the lsn that matters
581 * (truly makes the quotaoff irrevocable). If we do something else,
582 * then maybe we don't need two.
586 xfs_qm_qoff_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
593 xfs_qm_qoffend_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
598 static struct xfs_item_ops xfs_qm_qoffend_logitem_ops
= {
599 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
600 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
601 xfs_qm_qoff_logitem_format
,
602 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
603 .iop_unpin
= (void(*)(xfs_log_item_t
* ,int))
604 xfs_qm_qoff_logitem_unpin
,
605 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
606 xfs_qm_qoff_logitem_unpin_remove
,
607 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
608 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
609 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
610 xfs_qm_qoffend_logitem_committed
,
611 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
613 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
614 xfs_qm_qoffend_logitem_committing
618 * This is the ops vector shared by all quotaoff-start log items.
620 static struct xfs_item_ops xfs_qm_qoff_logitem_ops
= {
621 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
622 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
623 xfs_qm_qoff_logitem_format
,
624 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
625 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
626 xfs_qm_qoff_logitem_unpin
,
627 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
628 xfs_qm_qoff_logitem_unpin_remove
,
629 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
630 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
631 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
632 xfs_qm_qoff_logitem_committed
,
633 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
635 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
636 xfs_qm_qoff_logitem_committing
640 * Allocate and initialize an quotaoff item of the correct quota type(s).
643 xfs_qm_qoff_logitem_init(
644 struct xfs_mount
*mp
,
645 xfs_qoff_logitem_t
*start
,
648 xfs_qoff_logitem_t
*qf
;
650 qf
= (xfs_qoff_logitem_t
*) kmem_zalloc(sizeof(xfs_qoff_logitem_t
), KM_SLEEP
);
652 qf
->qql_item
.li_type
= XFS_LI_QUOTAOFF
;
654 qf
->qql_item
.li_ops
= &xfs_qm_qoffend_logitem_ops
;
656 qf
->qql_item
.li_ops
= &xfs_qm_qoff_logitem_ops
;
657 qf
->qql_item
.li_mountp
= mp
;
658 qf
->qql_format
.qf_type
= XFS_LI_QUOTAOFF
;
659 qf
->qql_format
.qf_flags
= flags
;
660 qf
->qql_start_lip
= start
;