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"
47 #include "xfs_buf_item.h"
48 #include "xfs_trans_priv.h"
52 * returns the number of iovecs needed to log the given dquot item.
56 xfs_qm_dquot_logitem_size(
57 xfs_dq_logitem_t
*logitem
)
60 * we need only two iovecs, one for the format, one for the real thing
66 * fills in the vector of log iovecs for the given dquot log item.
69 xfs_qm_dquot_logitem_format(
70 xfs_dq_logitem_t
*logitem
,
71 xfs_log_iovec_t
*logvec
)
74 ASSERT(logitem
->qli_dquot
);
76 logvec
->i_addr
= (xfs_caddr_t
)&logitem
->qli_format
;
77 logvec
->i_len
= sizeof(xfs_dq_logformat_t
);
78 XLOG_VEC_SET_TYPE(logvec
, XLOG_REG_TYPE_QFORMAT
);
80 logvec
->i_addr
= (xfs_caddr_t
)&logitem
->qli_dquot
->q_core
;
81 logvec
->i_len
= sizeof(xfs_disk_dquot_t
);
82 XLOG_VEC_SET_TYPE(logvec
, XLOG_REG_TYPE_DQUOT
);
84 ASSERT(2 == logitem
->qli_item
.li_desc
->lid_size
);
85 logitem
->qli_format
.qlf_size
= 2;
90 * Increment the pin count of the given dquot.
91 * This value is protected by pinlock spinlock in the xQM structure.
94 xfs_qm_dquot_logitem_pin(
95 xfs_dq_logitem_t
*logitem
)
99 dqp
= logitem
->qli_dquot
;
100 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
101 spin_lock(&(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
));
103 spin_unlock(&(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
));
107 * Decrement the pin count of the given dquot, and wake up
108 * anyone in xfs_dqwait_unpin() if the count goes to 0. The
109 * dquot must have been previously pinned with a call to xfs_dqpin().
113 xfs_qm_dquot_logitem_unpin(
114 xfs_dq_logitem_t
*logitem
,
119 dqp
= logitem
->qli_dquot
;
120 ASSERT(dqp
->q_pincount
> 0);
121 spin_lock(&(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
));
123 if (dqp
->q_pincount
== 0) {
124 sv_broadcast(&dqp
->q_pinwait
);
126 spin_unlock(&(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
));
131 xfs_qm_dquot_logitem_unpin_remove(
132 xfs_dq_logitem_t
*logitem
,
135 xfs_qm_dquot_logitem_unpin(logitem
, 0);
139 * Given the logitem, this writes the corresponding dquot entry to disk
140 * asynchronously. This is called with the dquot entry securely locked;
141 * we simply get xfs_qm_dqflush() to do the work, and unlock the dquot
145 xfs_qm_dquot_logitem_push(
146 xfs_dq_logitem_t
*logitem
)
151 dqp
= logitem
->qli_dquot
;
153 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
154 ASSERT(XFS_DQ_IS_FLUSH_LOCKED(dqp
));
157 * Since we were able to lock the dquot's flush lock and
158 * we found it on the AIL, the dquot must be dirty. This
159 * is because the dquot is removed from the AIL while still
160 * holding the flush lock in xfs_dqflush_done(). Thus, if
161 * we found it in the AIL and were able to obtain the flush
162 * lock without sleeping, then there must not have been
163 * anyone in the process of flushing the dquot.
165 error
= xfs_qm_dqflush(dqp
, XFS_QMOPT_DELWRI
);
167 xfs_fs_cmn_err(CE_WARN
, dqp
->q_mount
,
168 "xfs_qm_dquot_logitem_push: push error %d on dqp %p",
175 xfs_qm_dquot_logitem_committed(
180 * We always re-log the entire dquot when it becomes dirty,
181 * so, the latest copy _is_ the only one that matters.
188 * This is called to wait for the given dquot to be unpinned.
189 * Most of these pin/unpin routines are plagiarized from inode code.
195 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
196 if (dqp
->q_pincount
== 0) {
201 * Give the log a push so we don't wait here too long.
203 xfs_log_force(dqp
->q_mount
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
204 spin_lock(&(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
));
205 if (dqp
->q_pincount
== 0) {
206 spin_unlock(&(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
));
209 sv_wait(&(dqp
->q_pinwait
), PINOD
,
210 &(XFS_DQ_TO_QINF(dqp
)->qi_pinlock
), s
);
214 * This is called when IOP_TRYLOCK returns XFS_ITEM_PUSHBUF to indicate that
215 * the dquot is locked by us, but the flush lock isn't. So, here we are
216 * going to see if the relevant dquot buffer is incore, waiting on DELWRI.
217 * If so, we want to push it out to help us take this item off the AIL as soon
220 * We must not be holding the AIL lock at this point. Calling incore() to
221 * search the buffer cache can be a time consuming thing, and AIL lock is a
225 xfs_qm_dquot_logitem_pushbuf(
226 xfs_dq_logitem_t
*qip
)
233 dqp
= qip
->qli_dquot
;
234 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
237 * The qli_pushbuf_flag keeps others from
238 * trying to duplicate our effort.
240 ASSERT(qip
->qli_pushbuf_flag
!= 0);
241 ASSERT(qip
->qli_push_owner
== current_pid());
244 * If flushlock isn't locked anymore, chances are that the
245 * inode flush completed and the inode was taken off the AIL.
248 if (!issemalocked(&(dqp
->q_flock
)) ||
249 ((qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
) == 0)) {
250 qip
->qli_pushbuf_flag
= 0;
255 bp
= xfs_incore(mp
->m_ddev_targp
, qip
->qli_format
.qlf_blkno
,
256 XFS_QI_DQCHUNKLEN(mp
),
259 if (XFS_BUF_ISDELAYWRITE(bp
)) {
260 dopush
= ((qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
) &&
261 issemalocked(&(dqp
->q_flock
)));
262 qip
->qli_pushbuf_flag
= 0;
265 if (XFS_BUF_ISPINNED(bp
)) {
266 xfs_log_force(mp
, (xfs_lsn_t
)0,
275 error
= xfs_bawrite(mp
, bp
);
277 xfs_fs_cmn_err(CE_WARN
, mp
,
278 "xfs_qm_dquot_logitem_pushbuf: pushbuf error %d on qip %p, bp %p",
284 qip
->qli_pushbuf_flag
= 0;
291 qip
->qli_pushbuf_flag
= 0;
296 * This is called to attempt to lock the dquot associated with this
297 * dquot log item. Don't sleep on the dquot lock or the flush lock.
298 * If the flush lock is already held, indicating that the dquot has
299 * been or is in the process of being flushed, then see if we can
300 * find the dquot's buffer in the buffer cache without sleeping. If
301 * we can and it is marked delayed write, then we want to send it out.
302 * We delay doing so until the push routine, though, to avoid sleeping
303 * in any device strategy routines.
306 xfs_qm_dquot_logitem_trylock(
307 xfs_dq_logitem_t
*qip
)
312 dqp
= qip
->qli_dquot
;
313 if (dqp
->q_pincount
> 0)
314 return (XFS_ITEM_PINNED
);
316 if (! xfs_qm_dqlock_nowait(dqp
))
317 return (XFS_ITEM_LOCKED
);
319 retval
= XFS_ITEM_SUCCESS
;
320 if (! xfs_qm_dqflock_nowait(dqp
)) {
322 * The dquot is already being flushed. It may have been
323 * flushed delayed write, however, and we don't want to
324 * get stuck waiting for that to complete. So, we want to check
325 * to see if we can lock the dquot's buffer without sleeping.
326 * If we can and it is marked for delayed write, then we
327 * hold it and send it out from the push routine. We don't
328 * want to do that now since we might sleep in the device
329 * strategy routine. We also don't want to grab the buffer lock
330 * here because we'd like not to call into the buffer cache
331 * while holding the AIL lock.
332 * Make sure to only return PUSHBUF if we set pushbuf_flag
333 * ourselves. If someone else is doing it then we don't
334 * want to go to the push routine and duplicate their efforts.
336 if (qip
->qli_pushbuf_flag
== 0) {
337 qip
->qli_pushbuf_flag
= 1;
338 ASSERT(qip
->qli_format
.qlf_blkno
== dqp
->q_blkno
);
340 qip
->qli_push_owner
= current_pid();
343 * The dquot is left locked.
345 retval
= XFS_ITEM_PUSHBUF
;
347 retval
= XFS_ITEM_FLUSHING
;
348 xfs_dqunlock_nonotify(dqp
);
352 ASSERT(qip
->qli_item
.li_flags
& XFS_LI_IN_AIL
);
358 * Unlock the dquot associated with the log item.
359 * Clear the fields of the dquot and dquot log item that
360 * are specific to the current transaction. If the
361 * hold flags is set, do not unlock the dquot.
364 xfs_qm_dquot_logitem_unlock(
365 xfs_dq_logitem_t
*ql
)
371 ASSERT(XFS_DQ_IS_LOCKED(dqp
));
374 * Clear the transaction pointer in the dquot
376 dqp
->q_transp
= NULL
;
379 * dquots are never 'held' from getting unlocked at the end of
380 * a transaction. Their locking and unlocking is hidden inside the
381 * transaction layer, within trans_commit. Hence, no LI_HOLD flag
389 * this needs to stamp an lsn into the dquot, I think.
390 * rpc's that look at user dquot's would then have to
391 * push on the dependency recorded in the dquot
395 xfs_qm_dquot_logitem_committing(
404 * This is the ops vector for dquots
406 static struct xfs_item_ops xfs_dquot_item_ops
= {
407 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_size
,
408 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
409 xfs_qm_dquot_logitem_format
,
410 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_pin
,
411 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
412 xfs_qm_dquot_logitem_unpin
,
413 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*, xfs_trans_t
*))
414 xfs_qm_dquot_logitem_unpin_remove
,
415 .iop_trylock
= (uint(*)(xfs_log_item_t
*))
416 xfs_qm_dquot_logitem_trylock
,
417 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_unlock
,
418 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
419 xfs_qm_dquot_logitem_committed
,
420 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_dquot_logitem_push
,
421 .iop_pushbuf
= (void(*)(xfs_log_item_t
*))
422 xfs_qm_dquot_logitem_pushbuf
,
423 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
424 xfs_qm_dquot_logitem_committing
428 * Initialize the dquot log item for a newly allocated dquot.
429 * The dquot isn't locked at this point, but it isn't on any of the lists
430 * either, so we don't care.
433 xfs_qm_dquot_logitem_init(
434 struct xfs_dquot
*dqp
)
436 xfs_dq_logitem_t
*lp
;
437 lp
= &dqp
->q_logitem
;
439 lp
->qli_item
.li_type
= XFS_LI_DQUOT
;
440 lp
->qli_item
.li_ops
= &xfs_dquot_item_ops
;
441 lp
->qli_item
.li_mountp
= dqp
->q_mount
;
443 lp
->qli_format
.qlf_type
= XFS_LI_DQUOT
;
444 lp
->qli_format
.qlf_id
= be32_to_cpu(dqp
->q_core
.d_id
);
445 lp
->qli_format
.qlf_blkno
= dqp
->q_blkno
;
446 lp
->qli_format
.qlf_len
= 1;
448 * This is just the offset of this dquot within its buffer
449 * (which is currently 1 FSB and probably won't change).
450 * Hence 32 bits for this offset should be just fine.
451 * Alternatively, we can store (bufoffset / sizeof(xfs_dqblk_t))
452 * here, and recompute it at recovery time.
454 lp
->qli_format
.qlf_boffset
= (__uint32_t
)dqp
->q_bufoffset
;
457 /*------------------ QUOTAOFF LOG ITEMS -------------------*/
460 * This returns the number of iovecs needed to log the given quotaoff item.
461 * We only need 1 iovec for an quotaoff item. It just logs the
462 * quotaoff_log_format structure.
466 xfs_qm_qoff_logitem_size(xfs_qoff_logitem_t
*qf
)
472 * This is called to fill in the vector of log iovecs for the
473 * given quotaoff log item. We use only 1 iovec, and we point that
474 * at the quotaoff_log_format structure embedded in the quotaoff item.
475 * It is at this point that we assert that all of the extent
476 * slots in the quotaoff item have been filled.
479 xfs_qm_qoff_logitem_format(xfs_qoff_logitem_t
*qf
,
480 xfs_log_iovec_t
*log_vector
)
482 ASSERT(qf
->qql_format
.qf_type
== XFS_LI_QUOTAOFF
);
484 log_vector
->i_addr
= (xfs_caddr_t
)&(qf
->qql_format
);
485 log_vector
->i_len
= sizeof(xfs_qoff_logitem_t
);
486 XLOG_VEC_SET_TYPE(log_vector
, XLOG_REG_TYPE_QUOTAOFF
);
487 qf
->qql_format
.qf_size
= 1;
492 * Pinning has no meaning for an quotaoff item, so just return.
496 xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t
*qf
)
503 * Since pinning has no meaning for an quotaoff item, unpinning does
508 xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t
*qf
, int stale
)
515 xfs_qm_qoff_logitem_unpin_remove(xfs_qoff_logitem_t
*qf
, xfs_trans_t
*tp
)
521 * Quotaoff items have no locking, so just return success.
525 xfs_qm_qoff_logitem_trylock(xfs_qoff_logitem_t
*qf
)
527 return XFS_ITEM_LOCKED
;
531 * Quotaoff items have no locking or pushing, so return failure
532 * so that the caller doesn't bother with us.
536 xfs_qm_qoff_logitem_unlock(xfs_qoff_logitem_t
*qf
)
542 * The quotaoff-start-item is logged only once and cannot be moved in the log,
543 * so simply return the lsn at which it's been logged.
547 xfs_qm_qoff_logitem_committed(xfs_qoff_logitem_t
*qf
, xfs_lsn_t lsn
)
553 * There isn't much you can do to push on an quotaoff item. It is simply
554 * stuck waiting for the log to be flushed to disk.
558 xfs_qm_qoff_logitem_push(xfs_qoff_logitem_t
*qf
)
566 xfs_qm_qoffend_logitem_committed(
567 xfs_qoff_logitem_t
*qfe
,
570 xfs_qoff_logitem_t
*qfs
;
572 qfs
= qfe
->qql_start_lip
;
573 spin_lock(&qfs
->qql_item
.li_mountp
->m_ail_lock
);
575 * Delete the qoff-start logitem from the AIL.
576 * xfs_trans_delete_ail() drops the AIL lock.
578 xfs_trans_delete_ail(qfs
->qql_item
.li_mountp
, (xfs_log_item_t
*)qfs
);
579 kmem_free(qfs
, sizeof(xfs_qoff_logitem_t
));
580 kmem_free(qfe
, sizeof(xfs_qoff_logitem_t
));
581 return (xfs_lsn_t
)-1;
585 * XXX rcc - don't know quite what to do with this. I think we can
586 * just ignore it. The only time that isn't the case is if we allow
587 * the client to somehow see that quotas have been turned off in which
588 * we can't allow that to get back until the quotaoff hits the disk.
589 * So how would that happen? Also, do we need different routines for
590 * quotaoff start and quotaoff end? I suspect the answer is yes but
591 * to be sure, I need to look at the recovery code and see how quota off
592 * recovery is handled (do we roll forward or back or do something else).
593 * If we roll forwards or backwards, then we need two separate routines,
594 * one that does nothing and one that stamps in the lsn that matters
595 * (truly makes the quotaoff irrevocable). If we do something else,
596 * then maybe we don't need two.
600 xfs_qm_qoff_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
607 xfs_qm_qoffend_logitem_committing(xfs_qoff_logitem_t
*qip
, xfs_lsn_t commit_lsn
)
612 static struct xfs_item_ops xfs_qm_qoffend_logitem_ops
= {
613 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
614 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
615 xfs_qm_qoff_logitem_format
,
616 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
617 .iop_unpin
= (void(*)(xfs_log_item_t
* ,int))
618 xfs_qm_qoff_logitem_unpin
,
619 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
620 xfs_qm_qoff_logitem_unpin_remove
,
621 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
622 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
623 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
624 xfs_qm_qoffend_logitem_committed
,
625 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
627 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
628 xfs_qm_qoffend_logitem_committing
632 * This is the ops vector shared by all quotaoff-start log items.
634 static struct xfs_item_ops xfs_qm_qoff_logitem_ops
= {
635 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_size
,
636 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
637 xfs_qm_qoff_logitem_format
,
638 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_pin
,
639 .iop_unpin
= (void(*)(xfs_log_item_t
*, int))
640 xfs_qm_qoff_logitem_unpin
,
641 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*,xfs_trans_t
*))
642 xfs_qm_qoff_logitem_unpin_remove
,
643 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_trylock
,
644 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_unlock
,
645 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
646 xfs_qm_qoff_logitem_committed
,
647 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_qm_qoff_logitem_push
,
649 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
650 xfs_qm_qoff_logitem_committing
654 * Allocate and initialize an quotaoff item of the correct quota type(s).
657 xfs_qm_qoff_logitem_init(
658 struct xfs_mount
*mp
,
659 xfs_qoff_logitem_t
*start
,
662 xfs_qoff_logitem_t
*qf
;
664 qf
= (xfs_qoff_logitem_t
*) kmem_zalloc(sizeof(xfs_qoff_logitem_t
), KM_SLEEP
);
666 qf
->qql_item
.li_type
= XFS_LI_QUOTAOFF
;
668 qf
->qql_item
.li_ops
= &xfs_qm_qoffend_logitem_ops
;
670 qf
->qql_item
.li_ops
= &xfs_qm_qoff_logitem_ops
;
671 qf
->qql_item
.li_mountp
= mp
;
672 qf
->qql_format
.qf_type
= XFS_LI_QUOTAOFF
;
673 qf
->qql_format
.qf_flags
= flags
;
674 qf
->qql_start_lip
= start
;