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"
28 #include "xfs_alloc.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_quota.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_rtalloc.h"
44 #include "xfs_error.h"
45 #include "xfs_itable.h"
51 #include "xfs_buf_item.h"
52 #include "xfs_trans_priv.h"
56 * returns the number of iovecs needed to log the given dquot item.
60 xfs_qm_dquot_logitem_size(
61 xfs_dq_logitem_t
*logitem
)
64 * we need only two iovecs, one for the format, one for the real thing
70 * fills in the vector of log iovecs for the given dquot log item.
73 xfs_qm_dquot_logitem_format(
74 xfs_dq_logitem_t
*logitem
,
75 xfs_log_iovec_t
*logvec
)
78 ASSERT(logitem
->qli_dquot
);
80 logvec
->i_addr
= (xfs_caddr_t
)&logitem
->qli_format
;
81 logvec
->i_len
= sizeof(xfs_dq_logformat_t
);
82 XLOG_VEC_SET_TYPE(logvec
, XLOG_REG_TYPE_QFORMAT
);
84 logvec
->i_addr
= (xfs_caddr_t
)&logitem
->qli_dquot
->q_core
;
85 logvec
->i_len
= sizeof(xfs_disk_dquot_t
);
86 XLOG_VEC_SET_TYPE(logvec
, XLOG_REG_TYPE_DQUOT
);
88 ASSERT(2 == logitem
->qli_item
.li_desc
->lid_size
);
89 logitem
->qli_format
.qlf_size
= 2;
94 * Increment the pin count of the given dquot.
95 * This value is protected by pinlock spinlock in the xQM structure.
98 xfs_qm_dquot_logitem_pin(
99 xfs_dq_logitem_t
*logitem
)
104 dqp
= logitem
->qli_dquot
;
105 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
106 s
= XFS_DQ_PINLOCK(dqp
);
108 XFS_DQ_PINUNLOCK(dqp
, s
);
112 * Decrement the pin count of the given dquot, and wake up
113 * anyone in xfs_dqwait_unpin() if the count goes to 0. The
114 * dquot must have been previously pinned with a call to xfs_dqpin().
118 xfs_qm_dquot_logitem_unpin(
119 xfs_dq_logitem_t
*logitem
,
125 dqp
= logitem
->qli_dquot
;
126 ASSERT(dqp
->q_pincount
> 0);
127 s
= XFS_DQ_PINLOCK(dqp
);
129 if (dqp
->q_pincount
== 0) {
130 sv_broadcast(&dqp
->q_pinwait
);
132 XFS_DQ_PINUNLOCK(dqp
, s
);
137 xfs_qm_dquot_logitem_unpin_remove(
138 xfs_dq_logitem_t
*logitem
,
141 xfs_qm_dquot_logitem_unpin(logitem
, 0);
145 * Given the logitem, this writes the corresponding dquot entry to disk
146 * asynchronously. This is called with the dquot entry securely locked;
147 * we simply get xfs_qm_dqflush() to do the work, and unlock the dquot
151 xfs_qm_dquot_logitem_push(
152 xfs_dq_logitem_t
*logitem
)
156 dqp
= logitem
->qli_dquot
;
158 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
159 ASSERT(XFS_DQ_IS_FLUSH_LOCKED(dqp
));
162 * Since we were able to lock the dquot's flush lock and
163 * we found it on the AIL, the dquot must be dirty. This
164 * is because the dquot is removed from the AIL while still
165 * holding the flush lock in xfs_dqflush_done(). Thus, if
166 * we found it in the AIL and were able to obtain the flush
167 * lock without sleeping, then there must not have been
168 * anyone in the process of flushing the dquot.
170 xfs_qm_dqflush(dqp
, XFS_B_DELWRI
);
176 xfs_qm_dquot_logitem_committed(
181 * We always re-log the entire dquot when it becomes dirty,
182 * so, the latest copy _is_ the only one that matters.
189 * This is called to wait for the given dquot to be unpinned.
190 * Most of these pin/unpin routines are plagiarized from inode code.
198 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
199 if (dqp
->q_pincount
== 0) {
204 * Give the log a push so we don't wait here too long.
206 xfs_log_force(dqp
->q_mount
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
207 s
= XFS_DQ_PINLOCK(dqp
);
208 if (dqp
->q_pincount
== 0) {
209 XFS_DQ_PINUNLOCK(dqp
, s
);
212 sv_wait(&(dqp
->q_pinwait
), PINOD
,
213 &(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
), s
);
217 * This is called when IOP_TRYLOCK returns XFS_ITEM_PUSHBUF to indicate that
218 * the dquot is locked by us, but the flush lock isn't. So, here we are
219 * going to see if the relevant dquot buffer is incore, waiting on DELWRI.
220 * If so, we want to push it out to help us take this item off the AIL as soon
223 * We must not be holding the AIL_LOCK at this point. Calling incore() to
224 * search the buffer cache can be a time consuming thing, and AIL_LOCK is a
228 xfs_qm_dquot_logitem_pushbuf(
229 xfs_dq_logitem_t
*qip
)
236 dqp
= qip
->qli_dquot
;
237 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
240 * The qli_pushbuf_flag keeps others from
241 * trying to duplicate our effort.
243 ASSERT(qip
->qli_pushbuf_flag
!= 0);
244 ASSERT(qip
->qli_push_owner
== current_pid());
247 * If flushlock isn't locked anymore, chances are that the
248 * inode flush completed and the inode was taken off the AIL.
251 if ((valusema(&(dqp
->q_flock
)) > 0) ||
252 ((qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
) == 0)) {
253 qip
->qli_pushbuf_flag
= 0;
258 bp
= xfs_incore(mp
->m_ddev_targp
, qip
->qli_format
.qlf_blkno
,
259 XFS_QI_DQCHUNKLEN(mp
),
262 if (XFS_BUF_ISDELAYWRITE(bp
)) {
263 dopush
= ((qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
) &&
264 (valusema(&(dqp
->q_flock
)) <= 0));
265 qip
->qli_pushbuf_flag
= 0;
268 if (XFS_BUF_ISPINNED(bp
)) {
269 xfs_log_force(mp
, (xfs_lsn_t
)0,
282 qip
->qli_pushbuf_flag
= 0;
289 qip
->qli_pushbuf_flag
= 0;
294 * This is called to attempt to lock the dquot associated with this
295 * dquot log item. Don't sleep on the dquot lock or the flush lock.
296 * If the flush lock is already held, indicating that the dquot has
297 * been or is in the process of being flushed, then see if we can
298 * find the dquot's buffer in the buffer cache without sleeping. If
299 * we can and it is marked delayed write, then we want to send it out.
300 * We delay doing so until the push routine, though, to avoid sleeping
301 * in any device strategy routines.
304 xfs_qm_dquot_logitem_trylock(
305 xfs_dq_logitem_t
*qip
)
310 dqp
= qip
->qli_dquot
;
311 if (dqp
->q_pincount
> 0)
312 return (XFS_ITEM_PINNED
);
314 if (! xfs_qm_dqlock_nowait(dqp
))
315 return (XFS_ITEM_LOCKED
);
317 retval
= XFS_ITEM_SUCCESS
;
318 if (! xfs_qm_dqflock_nowait(dqp
)) {
320 * The dquot is already being flushed. It may have been
321 * flushed delayed write, however, and we don't want to
322 * get stuck waiting for that to complete. So, we want to check
323 * to see if we can lock the dquot's buffer without sleeping.
324 * If we can and it is marked for delayed write, then we
325 * hold it and send it out from the push routine. We don't
326 * want to do that now since we might sleep in the device
327 * strategy routine. We also don't want to grab the buffer lock
328 * here because we'd like not to call into the buffer cache
329 * while holding the AIL_LOCK.
330 * Make sure to only return PUSHBUF if we set pushbuf_flag
331 * ourselves. If someone else is doing it then we don't
332 * want to go to the push routine and duplicate their efforts.
334 if (qip
->qli_pushbuf_flag
== 0) {
335 qip
->qli_pushbuf_flag
= 1;
336 ASSERT(qip
->qli_format
.qlf_blkno
== dqp
->q_blkno
);
338 qip
->qli_push_owner
= current_pid();
341 * The dquot is left locked.
343 retval
= XFS_ITEM_PUSHBUF
;
345 retval
= XFS_ITEM_FLUSHING
;
346 xfs_dqunlock_nonotify(dqp
);
350 ASSERT(qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
);
356 * Unlock the dquot associated with the log item.
357 * Clear the fields of the dquot and dquot log item that
358 * are specific to the current transaction. If the
359 * hold flags is set, do not unlock the dquot.
362 xfs_qm_dquot_logitem_unlock(
363 xfs_dq_logitem_t
*ql
)
369 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
372 * Clear the transaction pointer in the dquot
374 dqp
->q_transp
= NULL
;
377 * dquots are never 'held' from getting unlocked at the end of
378 * a transaction. Their locking and unlocking is hidden inside the
379 * transaction layer, within trans_commit. Hence, no LI_HOLD flag
387 * The transaction with the dquot locked has aborted. The dquot
388 * must not be dirty within the transaction. We simply unlock just
389 * as if the transaction had been cancelled.
392 xfs_qm_dquot_logitem_abort(
393 xfs_dq_logitem_t
*ql
)
395 xfs_qm_dquot_logitem_unlock(ql
);
399 * this needs to stamp an lsn into the dquot, I think.
400 * rpc's that look at user dquot's would then have to
401 * push on the dependency recorded in the dquot
405 xfs_qm_dquot_logitem_committing(
414 * This is the ops vector for dquots
416 STATIC
struct xfs_item_ops xfs_dquot_item_ops
= {
417 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_size
,
418 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
419 xfs_qm_dquot_logitem_format
,
420 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_pin
,
421 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
422 xfs_qm_dquot_logitem_unpin
,
423 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*, xfs_trans_t
*))
424 xfs_qm_dquot_logitem_unpin_remove
,
425 .iop_trylock
= (uint(*)(xfs_log_item_t
*))
426 xfs_qm_dquot_logitem_trylock
,
427 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_unlock
,
428 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
429 xfs_qm_dquot_logitem_committed
,
430 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_push
,
431 .iop_abort
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_abort
,
432 .iop_pushbuf
= (void(*)(xfs_log_item_t
*))
433 xfs_qm_dquot_logitem_pushbuf
,
434 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
435 xfs_qm_dquot_logitem_committing
439 * Initialize the dquot log item for a newly allocated dquot.
440 * The dquot isn't locked at this point, but it isn't on any of the lists
441 * either, so we don't care.
444 xfs_qm_dquot_logitem_init(
445 struct xfs_dquot
*dqp
)
447 xfs_dq_logitem_t
*lp
;
448 lp
= &dqp
->q_logitem
;
450 lp
->qli_item
.li_type
= XFS_LI_DQUOT
;
451 lp
->qli_item
.li_ops
= &xfs_dquot_item_ops
;
452 lp
->qli_item
.li_mountp
= dqp
->q_mount
;
454 lp
->qli_format
.qlf_type
= XFS_LI_DQUOT
;
455 lp
->qli_format
.qlf_id
= be32_to_cpu(dqp
->q_core
.d_id
);
456 lp
->qli_format
.qlf_blkno
= dqp
->q_blkno
;
457 lp
->qli_format
.qlf_len
= 1;
459 * This is just the offset of this dquot within its buffer
460 * (which is currently 1 FSB and probably won't change).
461 * Hence 32 bits for this offset should be just fine.
462 * Alternatively, we can store (bufoffset / sizeof(xfs_dqblk_t))
463 * here, and recompute it at recovery time.
465 lp
->qli_format
.qlf_boffset
= (__uint32_t
)dqp
->q_bufoffset
;
468 /*------------------ QUOTAOFF LOG ITEMS -------------------*/
471 * This returns the number of iovecs needed to log the given quotaoff item.
472 * We only need 1 iovec for an quotaoff item. It just logs the
473 * quotaoff_log_format structure.
477 xfs_qm_qoff_logitem_size(xfs_qoff_logitem_t
*qf
)
483 * This is called to fill in the vector of log iovecs for the
484 * given quotaoff log item. We use only 1 iovec, and we point that
485 * at the quotaoff_log_format structure embedded in the quotaoff item.
486 * It is at this point that we assert that all of the extent
487 * slots in the quotaoff item have been filled.
490 xfs_qm_qoff_logitem_format(xfs_qoff_logitem_t
*qf
,
491 xfs_log_iovec_t
*log_vector
)
493 ASSERT(qf
->qql_format
.qf_type
== XFS_LI_QUOTAOFF
);
495 log_vector
->i_addr
= (xfs_caddr_t
)&(qf
->qql_format
);
496 log_vector
->i_len
= sizeof(xfs_qoff_logitem_t
);
497 XLOG_VEC_SET_TYPE(log_vector
, XLOG_REG_TYPE_QUOTAOFF
);
498 qf
->qql_format
.qf_size
= 1;
503 * Pinning has no meaning for an quotaoff item, so just return.
507 xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t
*qf
)
514 * Since pinning has no meaning for an quotaoff item, unpinning does
519 xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t
*qf
, int stale
)
526 xfs_qm_qoff_logitem_unpin_remove(xfs_qoff_logitem_t
*qf
, xfs_trans_t
*tp
)
532 * Quotaoff items have no locking, so just return success.
536 xfs_qm_qoff_logitem_trylock(xfs_qoff_logitem_t
*qf
)
538 return XFS_ITEM_LOCKED
;
542 * Quotaoff items have no locking or pushing, so return failure
543 * so that the caller doesn't bother with us.
547 xfs_qm_qoff_logitem_unlock(xfs_qoff_logitem_t
*qf
)
553 * The quotaoff-start-item is logged only once and cannot be moved in the log,
554 * so simply return the lsn at which it's been logged.
558 xfs_qm_qoff_logitem_committed(xfs_qoff_logitem_t
*qf
, xfs_lsn_t lsn
)
564 * The transaction of which this QUOTAOFF is a part has been aborted.
565 * Just clean up after ourselves.
566 * Shouldn't this never happen in the case of qoffend logitems? XXX
569 xfs_qm_qoff_logitem_abort(xfs_qoff_logitem_t
*qf
)
571 kmem_free(qf
, sizeof(xfs_qoff_logitem_t
));
575 * There isn't much you can do to push on an quotaoff item. It is simply
576 * stuck waiting for the log to be flushed to disk.
580 xfs_qm_qoff_logitem_push(xfs_qoff_logitem_t
*qf
)
588 xfs_qm_qoffend_logitem_committed(
589 xfs_qoff_logitem_t
*qfe
,
592 xfs_qoff_logitem_t
*qfs
;
595 qfs
= qfe
->qql_start_lip
;
596 AIL_LOCK(qfs
->qql_item
.li_mountp
,s
);
598 * Delete the qoff-start logitem from the AIL.
599 * xfs_trans_delete_ail() drops the AIL lock.
601 xfs_trans_delete_ail(qfs
->qql_item
.li_mountp
, (xfs_log_item_t
*)qfs
, s
);
602 kmem_free(qfs
, sizeof(xfs_qoff_logitem_t
));
603 kmem_free(qfe
, sizeof(xfs_qoff_logitem_t
));
604 return (xfs_lsn_t
)-1;
608 * XXX rcc - don't know quite what to do with this. I think we can
609 * just ignore it. The only time that isn't the case is if we allow
610 * the client to somehow see that quotas have been turned off in which
611 * we can't allow that to get back until the quotaoff hits the disk.
612 * So how would that happen? Also, do we need different routines for
613 * quotaoff start and quotaoff end? I suspect the answer is yes but
614 * to be sure, I need to look at the recovery code and see how quota off
615 * recovery is handled (do we roll forward or back or do something else).
616 * If we roll forwards or backwards, then we need two separate routines,
617 * one that does nothing and one that stamps in the lsn that matters
618 * (truly makes the quotaoff irrevocable). If we do something else,
619 * then maybe we don't need two.
623 xfs_qm_qoff_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
630 xfs_qm_qoffend_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
635 STATIC
struct xfs_item_ops xfs_qm_qoffend_logitem_ops
= {
636 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
637 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
638 xfs_qm_qoff_logitem_format
,
639 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
640 .iop_unpin
= (void(*)(xfs_log_item_t
* ,int))
641 xfs_qm_qoff_logitem_unpin
,
642 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
643 xfs_qm_qoff_logitem_unpin_remove
,
644 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
645 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
646 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
647 xfs_qm_qoffend_logitem_committed
,
648 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
649 .iop_abort
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_abort
,
651 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
652 xfs_qm_qoffend_logitem_committing
656 * This is the ops vector shared by all quotaoff-start log items.
658 STATIC
struct xfs_item_ops xfs_qm_qoff_logitem_ops
= {
659 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
660 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
661 xfs_qm_qoff_logitem_format
,
662 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
663 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
664 xfs_qm_qoff_logitem_unpin
,
665 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
666 xfs_qm_qoff_logitem_unpin_remove
,
667 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
668 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
669 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
670 xfs_qm_qoff_logitem_committed
,
671 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
672 .iop_abort
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_abort
,
674 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
675 xfs_qm_qoff_logitem_committing
679 * Allocate and initialize an quotaoff item of the correct quota type(s).
682 xfs_qm_qoff_logitem_init(
683 struct xfs_mount
*mp
,
684 xfs_qoff_logitem_t
*start
,
687 xfs_qoff_logitem_t
*qf
;
689 qf
= (xfs_qoff_logitem_t
*) kmem_zalloc(sizeof(xfs_qoff_logitem_t
), KM_SLEEP
);
691 qf
->qql_item
.li_type
= XFS_LI_QUOTAOFF
;
693 qf
->qql_item
.li_ops
= &xfs_qm_qoffend_logitem_ops
;
695 qf
->qql_item
.li_ops
= &xfs_qm_qoff_logitem_ops
;
696 qf
->qql_item
.li_mountp
= mp
;
697 qf
->qql_format
.qf_type
= XFS_LI_QUOTAOFF
;
698 qf
->qql_format
.qf_flags
= flags
;
699 qf
->qql_start_lip
= start
;