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
), XBF_TRYLOCK
);
242 if (XFS_BUF_ISDELAYWRITE(bp
)) {
243 dopush
= ((qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
) &&
244 !completion_done(&dqp
->q_flush
));
245 qip
->qli_pushbuf_flag
= 0;
248 if (XFS_BUF_ISPINNED(bp
)) {
249 xfs_log_force(mp
, (xfs_lsn_t
)0,
258 error
= xfs_bawrite(mp
, bp
);
260 xfs_fs_cmn_err(CE_WARN
, mp
,
261 "xfs_qm_dquot_logitem_pushbuf: pushbuf error %d on qip %p, bp %p",
267 qip
->qli_pushbuf_flag
= 0;
274 qip
->qli_pushbuf_flag
= 0;
279 * This is called to attempt to lock the dquot associated with this
280 * dquot log item. Don't sleep on the dquot lock or the flush lock.
281 * If the flush lock is already held, indicating that the dquot has
282 * been or is in the process of being flushed, then see if we can
283 * find the dquot's buffer in the buffer cache without sleeping. If
284 * we can and it is marked delayed write, then we want to send it out.
285 * We delay doing so until the push routine, though, to avoid sleeping
286 * in any device strategy routines.
289 xfs_qm_dquot_logitem_trylock(
290 xfs_dq_logitem_t
*qip
)
295 dqp
= qip
->qli_dquot
;
296 if (atomic_read(&dqp
->q_pincount
) > 0)
297 return (XFS_ITEM_PINNED
);
299 if (! xfs_qm_dqlock_nowait(dqp
))
300 return (XFS_ITEM_LOCKED
);
302 retval
= XFS_ITEM_SUCCESS
;
303 if (!xfs_dqflock_nowait(dqp
)) {
305 * The dquot is already being flushed. It may have been
306 * flushed delayed write, however, and we don't want to
307 * get stuck waiting for that to complete. So, we want to check
308 * to see if we can lock the dquot's buffer without sleeping.
309 * If we can and it is marked for delayed write, then we
310 * hold it and send it out from the push routine. We don't
311 * want to do that now since we might sleep in the device
312 * strategy routine. We also don't want to grab the buffer lock
313 * here because we'd like not to call into the buffer cache
314 * while holding the AIL lock.
315 * Make sure to only return PUSHBUF if we set pushbuf_flag
316 * ourselves. If someone else is doing it then we don't
317 * want to go to the push routine and duplicate their efforts.
319 if (qip
->qli_pushbuf_flag
== 0) {
320 qip
->qli_pushbuf_flag
= 1;
321 ASSERT(qip
->qli_format
.qlf_blkno
== dqp
->q_blkno
);
323 qip
->qli_push_owner
= current_pid();
326 * The dquot is left locked.
328 retval
= XFS_ITEM_PUSHBUF
;
330 retval
= XFS_ITEM_FLUSHING
;
331 xfs_dqunlock_nonotify(dqp
);
335 ASSERT(qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
);
341 * Unlock the dquot associated with the log item.
342 * Clear the fields of the dquot and dquot log item that
343 * are specific to the current transaction. If the
344 * hold flags is set, do not unlock the dquot.
347 xfs_qm_dquot_logitem_unlock(
348 xfs_dq_logitem_t
*ql
)
354 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
357 * Clear the transaction pointer in the dquot
359 dqp
->q_transp
= NULL
;
362 * dquots are never 'held' from getting unlocked at the end of
363 * a transaction. Their locking and unlocking is hidden inside the
364 * transaction layer, within trans_commit. Hence, no LI_HOLD flag
372 * this needs to stamp an lsn into the dquot, I think.
373 * rpc's that look at user dquot's would then have to
374 * push on the dependency recorded in the dquot
378 xfs_qm_dquot_logitem_committing(
387 * This is the ops vector for dquots
389 static struct xfs_item_ops xfs_dquot_item_ops
= {
390 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_size
,
391 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
392 xfs_qm_dquot_logitem_format
,
393 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_pin
,
394 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
395 xfs_qm_dquot_logitem_unpin
,
396 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*, xfs_trans_t
*))
397 xfs_qm_dquot_logitem_unpin_remove
,
398 .iop_trylock
= (uint(*)(xfs_log_item_t
*))
399 xfs_qm_dquot_logitem_trylock
,
400 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_unlock
,
401 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
402 xfs_qm_dquot_logitem_committed
,
403 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_push
,
404 .iop_pushbuf
= (void(*)(xfs_log_item_t
*))
405 xfs_qm_dquot_logitem_pushbuf
,
406 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
407 xfs_qm_dquot_logitem_committing
411 * Initialize the dquot log item for a newly allocated dquot.
412 * The dquot isn't locked at this point, but it isn't on any of the lists
413 * either, so we don't care.
416 xfs_qm_dquot_logitem_init(
417 struct xfs_dquot
*dqp
)
419 xfs_dq_logitem_t
*lp
;
420 lp
= &dqp
->q_logitem
;
422 lp
->qli_item
.li_type
= XFS_LI_DQUOT
;
423 lp
->qli_item
.li_ops
= &xfs_dquot_item_ops
;
424 lp
->qli_item
.li_mountp
= dqp
->q_mount
;
426 lp
->qli_format
.qlf_type
= XFS_LI_DQUOT
;
427 lp
->qli_format
.qlf_id
= be32_to_cpu(dqp
->q_core
.d_id
);
428 lp
->qli_format
.qlf_blkno
= dqp
->q_blkno
;
429 lp
->qli_format
.qlf_len
= 1;
431 * This is just the offset of this dquot within its buffer
432 * (which is currently 1 FSB and probably won't change).
433 * Hence 32 bits for this offset should be just fine.
434 * Alternatively, we can store (bufoffset / sizeof(xfs_dqblk_t))
435 * here, and recompute it at recovery time.
437 lp
->qli_format
.qlf_boffset
= (__uint32_t
)dqp
->q_bufoffset
;
440 /*------------------ QUOTAOFF LOG ITEMS -------------------*/
443 * This returns the number of iovecs needed to log the given quotaoff item.
444 * We only need 1 iovec for an quotaoff item. It just logs the
445 * quotaoff_log_format structure.
449 xfs_qm_qoff_logitem_size(xfs_qoff_logitem_t
*qf
)
455 * This is called to fill in the vector of log iovecs for the
456 * given quotaoff log item. We use only 1 iovec, and we point that
457 * at the quotaoff_log_format structure embedded in the quotaoff item.
458 * It is at this point that we assert that all of the extent
459 * slots in the quotaoff item have been filled.
462 xfs_qm_qoff_logitem_format(xfs_qoff_logitem_t
*qf
,
463 xfs_log_iovec_t
*log_vector
)
465 ASSERT(qf
->qql_format
.qf_type
== XFS_LI_QUOTAOFF
);
467 log_vector
->i_addr
= (xfs_caddr_t
)&(qf
->qql_format
);
468 log_vector
->i_len
= sizeof(xfs_qoff_logitem_t
);
469 XLOG_VEC_SET_TYPE(log_vector
, XLOG_REG_TYPE_QUOTAOFF
);
470 qf
->qql_format
.qf_size
= 1;
475 * Pinning has no meaning for an quotaoff item, so just return.
479 xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t
*qf
)
486 * Since pinning has no meaning for an quotaoff item, unpinning does
491 xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t
*qf
, int stale
)
498 xfs_qm_qoff_logitem_unpin_remove(xfs_qoff_logitem_t
*qf
, xfs_trans_t
*tp
)
504 * Quotaoff items have no locking, so just return success.
508 xfs_qm_qoff_logitem_trylock(xfs_qoff_logitem_t
*qf
)
510 return XFS_ITEM_LOCKED
;
514 * Quotaoff items have no locking or pushing, so return failure
515 * so that the caller doesn't bother with us.
519 xfs_qm_qoff_logitem_unlock(xfs_qoff_logitem_t
*qf
)
525 * The quotaoff-start-item is logged only once and cannot be moved in the log,
526 * so simply return the lsn at which it's been logged.
530 xfs_qm_qoff_logitem_committed(xfs_qoff_logitem_t
*qf
, xfs_lsn_t lsn
)
536 * There isn't much you can do to push on an quotaoff item. It is simply
537 * stuck waiting for the log to be flushed to disk.
541 xfs_qm_qoff_logitem_push(xfs_qoff_logitem_t
*qf
)
549 xfs_qm_qoffend_logitem_committed(
550 xfs_qoff_logitem_t
*qfe
,
553 xfs_qoff_logitem_t
*qfs
;
554 struct xfs_ail
*ailp
;
556 qfs
= qfe
->qql_start_lip
;
557 ailp
= qfs
->qql_item
.li_ailp
;
558 spin_lock(&ailp
->xa_lock
);
560 * Delete the qoff-start logitem from the AIL.
561 * xfs_trans_ail_delete() drops the AIL lock.
563 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)qfs
);
566 return (xfs_lsn_t
)-1;
570 * XXX rcc - don't know quite what to do with this. I think we can
571 * just ignore it. The only time that isn't the case is if we allow
572 * the client to somehow see that quotas have been turned off in which
573 * we can't allow that to get back until the quotaoff hits the disk.
574 * So how would that happen? Also, do we need different routines for
575 * quotaoff start and quotaoff end? I suspect the answer is yes but
576 * to be sure, I need to look at the recovery code and see how quota off
577 * recovery is handled (do we roll forward or back or do something else).
578 * If we roll forwards or backwards, then we need two separate routines,
579 * one that does nothing and one that stamps in the lsn that matters
580 * (truly makes the quotaoff irrevocable). If we do something else,
581 * then maybe we don't need two.
585 xfs_qm_qoff_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
592 xfs_qm_qoffend_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
597 static struct xfs_item_ops xfs_qm_qoffend_logitem_ops
= {
598 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
599 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
600 xfs_qm_qoff_logitem_format
,
601 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
602 .iop_unpin
= (void(*)(xfs_log_item_t
* ,int))
603 xfs_qm_qoff_logitem_unpin
,
604 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
605 xfs_qm_qoff_logitem_unpin_remove
,
606 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
607 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
608 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
609 xfs_qm_qoffend_logitem_committed
,
610 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
612 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
613 xfs_qm_qoffend_logitem_committing
617 * This is the ops vector shared by all quotaoff-start log items.
619 static struct xfs_item_ops xfs_qm_qoff_logitem_ops
= {
620 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
621 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
622 xfs_qm_qoff_logitem_format
,
623 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
624 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
625 xfs_qm_qoff_logitem_unpin
,
626 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
627 xfs_qm_qoff_logitem_unpin_remove
,
628 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
629 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
630 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
631 xfs_qm_qoff_logitem_committed
,
632 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
634 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
635 xfs_qm_qoff_logitem_committing
639 * Allocate and initialize an quotaoff item of the correct quota type(s).
642 xfs_qm_qoff_logitem_init(
643 struct xfs_mount
*mp
,
644 xfs_qoff_logitem_t
*start
,
647 xfs_qoff_logitem_t
*qf
;
649 qf
= (xfs_qoff_logitem_t
*) kmem_zalloc(sizeof(xfs_qoff_logitem_t
), KM_SLEEP
);
651 qf
->qql_item
.li_type
= XFS_LI_QUOTAOFF
;
653 qf
->qql_item
.li_ops
= &xfs_qm_qoffend_logitem_ops
;
655 qf
->qql_item
.li_ops
= &xfs_qm_qoff_logitem_ops
;
656 qf
->qql_item
.li_mountp
= mp
;
657 qf
->qql_format
.qf_type
= XFS_LI_QUOTAOFF
;
658 qf
->qql_format
.qf_flags
= flags
;
659 qf
->qql_start_lip
= start
;