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.
93 xfs_qm_dquot_logitem_pin(
94 xfs_dq_logitem_t *logitem)
96 xfs_dquot_t *dqp = logitem->qli_dquot;
98 ASSERT(XFS_DQ_IS_LOCKED(dqp));
99 atomic_inc(&dqp->q_pincount);
103 * Decrement the pin count of the given dquot, and wake up
104 * anyone in xfs_dqwait_unpin() if the count goes to 0. The
105 * dquot must have been previously pinned with a call to
106 * xfs_qm_dquot_logitem_pin().
110 xfs_qm_dquot_logitem_unpin(
111 xfs_dq_logitem_t *logitem,
114 xfs_dquot_t *dqp = logitem->qli_dquot;
116 ASSERT(atomic_read(&dqp->q_pincount) > 0);
117 if (atomic_dec_and_test(&dqp->q_pincount))
118 wake_up(&dqp->q_pinwait);
123 xfs_qm_dquot_logitem_unpin_remove(
124 xfs_dq_logitem_t *logitem,
127 xfs_qm_dquot_logitem_unpin(logitem, 0);
131 * Given the logitem, this writes the corresponding dquot entry to disk
132 * asynchronously. This is called with the dquot entry securely locked;
133 * we simply get xfs_qm_dqflush() to do the work, and unlock the dquot
137 xfs_qm_dquot_logitem_push(
138 xfs_dq_logitem_t *logitem)
143 dqp = logitem->qli_dquot;
145 ASSERT(XFS_DQ_IS_LOCKED(dqp));
146 ASSERT(!completion_done(&dqp->q_flush));
149 * Since we were able to lock the dquot's flush lock and
150 * we found it on the AIL, the dquot must be dirty. This
151 * is because the dquot is removed from the AIL while still
152 * holding the flush lock in xfs_dqflush_done(). Thus, if
153 * we found it in the AIL and were able to obtain the flush
154 * lock without sleeping, then there must not have been
155 * anyone in the process of flushing the dquot.
157 error = xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
159 xfs_fs_cmn_err(CE_WARN, dqp->q_mount,
160 "xfs_qm_dquot_logitem_push: push error %d on dqp %p",
167 xfs_qm_dquot_logitem_committed(
172 * We always re-log the entire dquot when it becomes dirty,
173 * so, the latest copy _is_ the only one that matters.
180 * This is called to wait for the given dquot to be unpinned.
181 * Most of these pin/unpin routines are plagiarized from inode code.
187 ASSERT(XFS_DQ_IS_LOCKED(dqp));
188 if (atomic_read(&dqp->q_pincount) == 0)
192 * Give the log a push so we don't wait here too long.
194 xfs_log_force(dqp->q_mount, (xfs_lsn_t)0, XFS_LOG_FORCE);
195 wait_event(dqp->q_pinwait, (atomic_read(&dqp->q_pincount) == 0));
199 * This is called when IOP_TRYLOCK returns XFS_ITEM_PUSHBUF to indicate that
200 * the dquot is locked by us, but the flush lock isn't. So, here we are
201 * going to see if the relevant dquot buffer is incore, waiting on DELWRI.
202 * If so, we want to push it out to help us take this item off the AIL as soon
205 * We must not be holding the AIL lock at this point. Calling incore() to
206 * search the buffer cache can be a time consuming thing, and AIL lock is a
210 xfs_qm_dquot_logitem_pushbuf(
211 xfs_dq_logitem_t *qip)
218 dqp = qip->qli_dquot;
219 ASSERT(XFS_DQ_IS_LOCKED(dqp));
222 * The qli_pushbuf_flag keeps others from
223 * trying to duplicate our effort.
225 ASSERT(qip->qli_pushbuf_flag != 0);
226 ASSERT(qip->qli_push_owner == current_pid());
229 * If flushlock isn't locked anymore, chances are that the
230 * inode flush completed and the inode was taken off the AIL.
233 if (completion_done(&dqp->q_flush) ||
234 ((qip->qli_item.li_flags & XFS_LI_IN_AIL) == 0)) {
235 qip->qli_pushbuf_flag = 0;
240 bp = xfs_incore(mp->m_ddev_targp, qip->qli_format.qlf_blkno,
241 XFS_QI_DQCHUNKLEN(mp),
244 if (XFS_BUF_ISDELAYWRITE(bp)) {
245 dopush = ((qip->qli_item.li_flags & XFS_LI_IN_AIL) &&
246 !completion_done(&dqp->q_flush));
247 qip->qli_pushbuf_flag = 0;
250 if (XFS_BUF_ISPINNED(bp)) {
251 xfs_log_force(mp, (xfs_lsn_t)0,
260 error = xfs_bawrite(mp, bp);
262 xfs_fs_cmn_err(CE_WARN, mp,
263 "xfs_qm_dquot_logitem_pushbuf: pushbuf error %d on qip %p, bp %p",
269 qip->qli_pushbuf_flag = 0;
276 qip->qli_pushbuf_flag = 0;
281 * This is called to attempt to lock the dquot associated with this
282 * dquot log item. Don't sleep on the dquot lock or the flush lock.
283 * If the flush lock is already held, indicating that the dquot has
284 * been or is in the process of being flushed, then see if we can
285 * find the dquot's buffer in the buffer cache without sleeping. If
286 * we can and it is marked delayed write, then we want to send it out.
287 * We delay doing so until the push routine, though, to avoid sleeping
288 * in any device strategy routines.
291 xfs_qm_dquot_logitem_trylock(
292 xfs_dq_logitem_t *qip)
297 dqp = qip->qli_dquot;
298 if (atomic_read(&dqp->q_pincount) > 0)
299 return (XFS_ITEM_PINNED);
301 if (! xfs_qm_dqlock_nowait(dqp))
302 return (XFS_ITEM_LOCKED);
304 retval = XFS_ITEM_SUCCESS;
305 if (!xfs_dqflock_nowait(dqp)) {
307 * The dquot is already being flushed. It may have been
308 * flushed delayed write, however, and we don't want to
309 * get stuck waiting for that to complete. So, we want to check
310 * to see if we can lock the dquot's buffer without sleeping.
311 * If we can and it is marked for delayed write, then we
312 * hold it and send it out from the push routine. We don't
313 * want to do that now since we might sleep in the device
314 * strategy routine. We also don't want to grab the buffer lock
315 * here because we'd like not to call into the buffer cache
316 * while holding the AIL lock.
317 * Make sure to only return PUSHBUF if we set pushbuf_flag
318 * ourselves. If someone else is doing it then we don't
319 * want to go to the push routine and duplicate their efforts.
321 if (qip->qli_pushbuf_flag == 0) {
322 qip->qli_pushbuf_flag = 1;
323 ASSERT(qip->qli_format.qlf_blkno == dqp->q_blkno);
325 qip->qli_push_owner = current_pid();
328 * The dquot is left locked.
330 retval = XFS_ITEM_PUSHBUF;
332 retval = XFS_ITEM_FLUSHING;
333 xfs_dqunlock_nonotify(dqp);
337 ASSERT(qip->qli_item.li_flags & XFS_LI_IN_AIL);
343 * Unlock the dquot associated with the log item.
344 * Clear the fields of the dquot and dquot log item that
345 * are specific to the current transaction. If the
346 * hold flags is set, do not unlock the dquot.
349 xfs_qm_dquot_logitem_unlock(
350 xfs_dq_logitem_t *ql)
356 ASSERT(XFS_DQ_IS_LOCKED(dqp));
359 * Clear the transaction pointer in the dquot
361 dqp->q_transp = NULL;
364 * dquots are never 'held' from getting unlocked at the end of
365 * a transaction. Their locking and unlocking is hidden inside the
366 * transaction layer, within trans_commit. Hence, no LI_HOLD flag
374 * this needs to stamp an lsn into the dquot, I think.
375 * rpc's that look at user dquot's would then have to
376 * push on the dependency recorded in the dquot
380 xfs_qm_dquot_logitem_committing(
389 * This is the ops vector for dquots
391 static struct xfs_item_ops xfs_dquot_item_ops = {
392 .iop_size = (uint(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_size,
393 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
394 xfs_qm_dquot_logitem_format,
395 .iop_pin = (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_pin,
396 .iop_unpin = (void(*)(xfs_log_item_t*, int))
397 xfs_qm_dquot_logitem_unpin,
398 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
399 xfs_qm_dquot_logitem_unpin_remove,
400 .iop_trylock = (uint(*)(xfs_log_item_t*))
401 xfs_qm_dquot_logitem_trylock,
402 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_unlock,
403 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
404 xfs_qm_dquot_logitem_committed,
405 .iop_push = (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_push,
406 .iop_pushbuf = (void(*)(xfs_log_item_t*))
407 xfs_qm_dquot_logitem_pushbuf,
408 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
409 xfs_qm_dquot_logitem_committing
413 * Initialize the dquot log item for a newly allocated dquot.
414 * The dquot isn't locked at this point, but it isn't on any of the lists
415 * either, so we don't care.
418 xfs_qm_dquot_logitem_init(
419 struct xfs_dquot *dqp)
421 xfs_dq_logitem_t *lp;
422 lp = &dqp->q_logitem;
424 lp->qli_item.li_type = XFS_LI_DQUOT;
425 lp->qli_item.li_ops = &xfs_dquot_item_ops;
426 lp->qli_item.li_mountp = dqp->q_mount;
428 lp->qli_format.qlf_type = XFS_LI_DQUOT;
429 lp->qli_format.qlf_id = be32_to_cpu(dqp->q_core.d_id);
430 lp->qli_format.qlf_blkno = dqp->q_blkno;
431 lp->qli_format.qlf_len = 1;
433 * This is just the offset of this dquot within its buffer
434 * (which is currently 1 FSB and probably won't change).
435 * Hence 32 bits for this offset should be just fine.
436 * Alternatively, we can store (bufoffset / sizeof(xfs_dqblk_t))
437 * here, and recompute it at recovery time.
439 lp->qli_format.qlf_boffset = (__uint32_t)dqp->q_bufoffset;
442 /*------------------ QUOTAOFF LOG ITEMS -------------------*/
445 * This returns the number of iovecs needed to log the given quotaoff item.
446 * We only need 1 iovec for an quotaoff item. It just logs the
447 * quotaoff_log_format structure.
451 xfs_qm_qoff_logitem_size(xfs_qoff_logitem_t *qf)
457 * This is called to fill in the vector of log iovecs for the
458 * given quotaoff log item. We use only 1 iovec, and we point that
459 * at the quotaoff_log_format structure embedded in the quotaoff item.
460 * It is at this point that we assert that all of the extent
461 * slots in the quotaoff item have been filled.
464 xfs_qm_qoff_logitem_format(xfs_qoff_logitem_t *qf,
465 xfs_log_iovec_t *log_vector)
467 ASSERT(qf->qql_format.qf_type == XFS_LI_QUOTAOFF);
469 log_vector->i_addr = (xfs_caddr_t)&(qf->qql_format);
470 log_vector->i_len = sizeof(xfs_qoff_logitem_t);
471 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_QUOTAOFF);
472 qf->qql_format.qf_size = 1;
477 * Pinning has no meaning for an quotaoff item, so just return.
481 xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t *qf)
488 * Since pinning has no meaning for an quotaoff item, unpinning does
493 xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t *qf, int stale)
500 xfs_qm_qoff_logitem_unpin_remove(xfs_qoff_logitem_t *qf, xfs_trans_t *tp)
506 * Quotaoff items have no locking, so just return success.
510 xfs_qm_qoff_logitem_trylock(xfs_qoff_logitem_t *qf)
512 return XFS_ITEM_LOCKED;
516 * Quotaoff items have no locking or pushing, so return failure
517 * so that the caller doesn't bother with us.
521 xfs_qm_qoff_logitem_unlock(xfs_qoff_logitem_t *qf)
527 * The quotaoff-start-item is logged only once and cannot be moved in the log,
528 * so simply return the lsn at which it's been logged.
532 xfs_qm_qoff_logitem_committed(xfs_qoff_logitem_t *qf, xfs_lsn_t lsn)
538 * There isn't much you can do to push on an quotaoff item. It is simply
539 * stuck waiting for the log to be flushed to disk.
543 xfs_qm_qoff_logitem_push(xfs_qoff_logitem_t *qf)
551 xfs_qm_qoffend_logitem_committed(
552 xfs_qoff_logitem_t *qfe,
555 xfs_qoff_logitem_t *qfs;
556 struct xfs_ail *ailp;
558 qfs = qfe->qql_start_lip;
559 ailp = qfs->qql_item.li_ailp;
560 spin_lock(&ailp->xa_lock);
562 * Delete the qoff-start logitem from the AIL.
563 * xfs_trans_ail_delete() drops the AIL lock.
565 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)qfs);
568 return (xfs_lsn_t)-1;
572 * XXX rcc - don't know quite what to do with this. I think we can
573 * just ignore it. The only time that isn't the case is if we allow
574 * the client to somehow see that quotas have been turned off in which
575 * we can't allow that to get back until the quotaoff hits the disk.
576 * So how would that happen? Also, do we need different routines for
577 * quotaoff start and quotaoff end? I suspect the answer is yes but
578 * to be sure, I need to look at the recovery code and see how quota off
579 * recovery is handled (do we roll forward or back or do something else).
580 * If we roll forwards or backwards, then we need two separate routines,
581 * one that does nothing and one that stamps in the lsn that matters
582 * (truly makes the quotaoff irrevocable). If we do something else,
583 * then maybe we don't need two.
587 xfs_qm_qoff_logitem_committing(xfs_qoff_logitem_t *qip, xfs_lsn_t commit_lsn)
594 xfs_qm_qoffend_logitem_committing(xfs_qoff_logitem_t *qip, xfs_lsn_t commit_lsn)
599 static struct xfs_item_ops xfs_qm_qoffend_logitem_ops = {
600 .iop_size = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_size,
601 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
602 xfs_qm_qoff_logitem_format,
603 .iop_pin = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
604 .iop_unpin = (void(*)(xfs_log_item_t* ,int))
605 xfs_qm_qoff_logitem_unpin,
606 .iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
607 xfs_qm_qoff_logitem_unpin_remove,
608 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,
609 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unlock,
610 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
611 xfs_qm_qoffend_logitem_committed,
612 .iop_push = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_push,
614 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
615 xfs_qm_qoffend_logitem_committing
619 * This is the ops vector shared by all quotaoff-start log items.
621 static struct xfs_item_ops xfs_qm_qoff_logitem_ops = {
622 .iop_size = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_size,
623 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
624 xfs_qm_qoff_logitem_format,
625 .iop_pin = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
626 .iop_unpin = (void(*)(xfs_log_item_t*, int))
627 xfs_qm_qoff_logitem_unpin,
628 .iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
629 xfs_qm_qoff_logitem_unpin_remove,
630 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,
631 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unlock,
632 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
633 xfs_qm_qoff_logitem_committed,
634 .iop_push = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_push,
636 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
637 xfs_qm_qoff_logitem_committing
641 * Allocate and initialize an quotaoff item of the correct quota type(s).
644 xfs_qm_qoff_logitem_init(
645 struct xfs_mount *mp,
646 xfs_qoff_logitem_t *start,
649 xfs_qoff_logitem_t *qf;
651 qf = (xfs_qoff_logitem_t*) kmem_zalloc(sizeof(xfs_qoff_logitem_t), KM_SLEEP);
653 qf->qql_item.li_type = XFS_LI_QUOTAOFF;
655 qf->qql_item.li_ops = &xfs_qm_qoffend_logitem_ops;
657 qf->qql_item.li_ops = &xfs_qm_qoff_logitem_ops;
658 qf->qql_item.li_mountp = mp;
659 qf->qql_format.qf_type = XFS_LI_QUOTAOFF;
660 qf->qql_format.qf_flags = flags;
661 qf->qql_start_lip = start;