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)
100 dqp = logitem->qli_dquot;
101 ASSERT(XFS_DQ_IS_LOCKED(dqp));
102 s = XFS_DQ_PINLOCK(dqp);
104 XFS_DQ_PINUNLOCK(dqp, s);
108 * Decrement the pin count of the given dquot, and wake up
109 * anyone in xfs_dqwait_unpin() if the count goes to 0. The
110 * dquot must have been previously pinned with a call to xfs_dqpin().
114 xfs_qm_dquot_logitem_unpin(
115 xfs_dq_logitem_t *logitem,
121 dqp = logitem->qli_dquot;
122 ASSERT(dqp->q_pincount > 0);
123 s = XFS_DQ_PINLOCK(dqp);
125 if (dqp->q_pincount == 0) {
126 sv_broadcast(&dqp->q_pinwait);
128 XFS_DQ_PINUNLOCK(dqp, s);
133 xfs_qm_dquot_logitem_unpin_remove(
134 xfs_dq_logitem_t *logitem,
137 xfs_qm_dquot_logitem_unpin(logitem, 0);
141 * Given the logitem, this writes the corresponding dquot entry to disk
142 * asynchronously. This is called with the dquot entry securely locked;
143 * we simply get xfs_qm_dqflush() to do the work, and unlock the dquot
147 xfs_qm_dquot_logitem_push(
148 xfs_dq_logitem_t *logitem)
152 dqp = logitem->qli_dquot;
154 ASSERT(XFS_DQ_IS_LOCKED(dqp));
155 ASSERT(XFS_DQ_IS_FLUSH_LOCKED(dqp));
158 * Since we were able to lock the dquot's flush lock and
159 * we found it on the AIL, the dquot must be dirty. This
160 * is because the dquot is removed from the AIL while still
161 * holding the flush lock in xfs_dqflush_done(). Thus, if
162 * we found it in the AIL and were able to obtain the flush
163 * lock without sleeping, then there must not have been
164 * anyone in the process of flushing the dquot.
166 xfs_qm_dqflush(dqp, XFS_B_DELWRI);
172 xfs_qm_dquot_logitem_committed(
177 * We always re-log the entire dquot when it becomes dirty,
178 * so, the latest copy _is_ the only one that matters.
185 * This is called to wait for the given dquot to be unpinned.
186 * Most of these pin/unpin routines are plagiarized from inode code.
194 ASSERT(XFS_DQ_IS_LOCKED(dqp));
195 if (dqp->q_pincount == 0) {
200 * Give the log a push so we don't wait here too long.
202 xfs_log_force(dqp->q_mount, (xfs_lsn_t)0, XFS_LOG_FORCE);
203 s = XFS_DQ_PINLOCK(dqp);
204 if (dqp->q_pincount == 0) {
205 XFS_DQ_PINUNLOCK(dqp, s);
208 sv_wait(&(dqp->q_pinwait), PINOD,
209 &(XFS_DQ_TO_QINF(dqp)->qi_pinlock), s);
213 * This is called when IOP_TRYLOCK returns XFS_ITEM_PUSHBUF to indicate that
214 * the dquot is locked by us, but the flush lock isn't. So, here we are
215 * going to see if the relevant dquot buffer is incore, waiting on DELWRI.
216 * If so, we want to push it out to help us take this item off the AIL as soon
219 * We must not be holding the AIL_LOCK at this point. Calling incore() to
220 * search the buffer cache can be a time consuming thing, and AIL_LOCK is a
224 xfs_qm_dquot_logitem_pushbuf(
225 xfs_dq_logitem_t *qip)
232 dqp = qip->qli_dquot;
233 ASSERT(XFS_DQ_IS_LOCKED(dqp));
236 * The qli_pushbuf_flag keeps others from
237 * trying to duplicate our effort.
239 ASSERT(qip->qli_pushbuf_flag != 0);
240 ASSERT(qip->qli_push_owner == current_pid());
243 * If flushlock isn't locked anymore, chances are that the
244 * inode flush completed and the inode was taken off the AIL.
247 if (!issemalocked(&(dqp->q_flock)) ||
248 ((qip->qli_item.li_flags & XFS_LI_IN_AIL) == 0)) {
249 qip->qli_pushbuf_flag = 0;
254 bp = xfs_incore(mp->m_ddev_targp, qip->qli_format.qlf_blkno,
255 XFS_QI_DQCHUNKLEN(mp),
258 if (XFS_BUF_ISDELAYWRITE(bp)) {
259 dopush = ((qip->qli_item.li_flags & XFS_LI_IN_AIL) &&
260 issemalocked(&(dqp->q_flock)));
261 qip->qli_pushbuf_flag = 0;
264 if (XFS_BUF_ISPINNED(bp)) {
265 xfs_log_force(mp, (xfs_lsn_t)0,
278 qip->qli_pushbuf_flag = 0;
285 qip->qli_pushbuf_flag = 0;
290 * This is called to attempt to lock the dquot associated with this
291 * dquot log item. Don't sleep on the dquot lock or the flush lock.
292 * If the flush lock is already held, indicating that the dquot has
293 * been or is in the process of being flushed, then see if we can
294 * find the dquot's buffer in the buffer cache without sleeping. If
295 * we can and it is marked delayed write, then we want to send it out.
296 * We delay doing so until the push routine, though, to avoid sleeping
297 * in any device strategy routines.
300 xfs_qm_dquot_logitem_trylock(
301 xfs_dq_logitem_t *qip)
306 dqp = qip->qli_dquot;
307 if (dqp->q_pincount > 0)
308 return (XFS_ITEM_PINNED);
310 if (! xfs_qm_dqlock_nowait(dqp))
311 return (XFS_ITEM_LOCKED);
313 retval = XFS_ITEM_SUCCESS;
314 if (! xfs_qm_dqflock_nowait(dqp)) {
316 * The dquot is already being flushed. It may have been
317 * flushed delayed write, however, and we don't want to
318 * get stuck waiting for that to complete. So, we want to check
319 * to see if we can lock the dquot's buffer without sleeping.
320 * If we can and it is marked for delayed write, then we
321 * hold it and send it out from the push routine. We don't
322 * want to do that now since we might sleep in the device
323 * strategy routine. We also don't want to grab the buffer lock
324 * here because we'd like not to call into the buffer cache
325 * while holding the AIL_LOCK.
326 * Make sure to only return PUSHBUF if we set pushbuf_flag
327 * ourselves. If someone else is doing it then we don't
328 * want to go to the push routine and duplicate their efforts.
330 if (qip->qli_pushbuf_flag == 0) {
331 qip->qli_pushbuf_flag = 1;
332 ASSERT(qip->qli_format.qlf_blkno == dqp->q_blkno);
334 qip->qli_push_owner = current_pid();
337 * The dquot is left locked.
339 retval = XFS_ITEM_PUSHBUF;
341 retval = XFS_ITEM_FLUSHING;
342 xfs_dqunlock_nonotify(dqp);
346 ASSERT(qip->qli_item.li_flags & XFS_LI_IN_AIL);
352 * Unlock the dquot associated with the log item.
353 * Clear the fields of the dquot and dquot log item that
354 * are specific to the current transaction. If the
355 * hold flags is set, do not unlock the dquot.
358 xfs_qm_dquot_logitem_unlock(
359 xfs_dq_logitem_t *ql)
365 ASSERT(XFS_DQ_IS_LOCKED(dqp));
368 * Clear the transaction pointer in the dquot
370 dqp->q_transp = NULL;
373 * dquots are never 'held' from getting unlocked at the end of
374 * a transaction. Their locking and unlocking is hidden inside the
375 * transaction layer, within trans_commit. Hence, no LI_HOLD flag
383 * this needs to stamp an lsn into the dquot, I think.
384 * rpc's that look at user dquot's would then have to
385 * push on the dependency recorded in the dquot
389 xfs_qm_dquot_logitem_committing(
398 * This is the ops vector for dquots
400 static struct xfs_item_ops xfs_dquot_item_ops = {
401 .iop_size = (uint(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_size,
402 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
403 xfs_qm_dquot_logitem_format,
404 .iop_pin = (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_pin,
405 .iop_unpin = (void(*)(xfs_log_item_t*, int))
406 xfs_qm_dquot_logitem_unpin,
407 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
408 xfs_qm_dquot_logitem_unpin_remove,
409 .iop_trylock = (uint(*)(xfs_log_item_t*))
410 xfs_qm_dquot_logitem_trylock,
411 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_unlock,
412 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
413 xfs_qm_dquot_logitem_committed,
414 .iop_push = (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_push,
415 .iop_pushbuf = (void(*)(xfs_log_item_t*))
416 xfs_qm_dquot_logitem_pushbuf,
417 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
418 xfs_qm_dquot_logitem_committing
422 * Initialize the dquot log item for a newly allocated dquot.
423 * The dquot isn't locked at this point, but it isn't on any of the lists
424 * either, so we don't care.
427 xfs_qm_dquot_logitem_init(
428 struct xfs_dquot *dqp)
430 xfs_dq_logitem_t *lp;
431 lp = &dqp->q_logitem;
433 lp->qli_item.li_type = XFS_LI_DQUOT;
434 lp->qli_item.li_ops = &xfs_dquot_item_ops;
435 lp->qli_item.li_mountp = dqp->q_mount;
437 lp->qli_format.qlf_type = XFS_LI_DQUOT;
438 lp->qli_format.qlf_id = be32_to_cpu(dqp->q_core.d_id);
439 lp->qli_format.qlf_blkno = dqp->q_blkno;
440 lp->qli_format.qlf_len = 1;
442 * This is just the offset of this dquot within its buffer
443 * (which is currently 1 FSB and probably won't change).
444 * Hence 32 bits for this offset should be just fine.
445 * Alternatively, we can store (bufoffset / sizeof(xfs_dqblk_t))
446 * here, and recompute it at recovery time.
448 lp->qli_format.qlf_boffset = (__uint32_t)dqp->q_bufoffset;
451 /*------------------ QUOTAOFF LOG ITEMS -------------------*/
454 * This returns the number of iovecs needed to log the given quotaoff item.
455 * We only need 1 iovec for an quotaoff item. It just logs the
456 * quotaoff_log_format structure.
460 xfs_qm_qoff_logitem_size(xfs_qoff_logitem_t *qf)
466 * This is called to fill in the vector of log iovecs for the
467 * given quotaoff log item. We use only 1 iovec, and we point that
468 * at the quotaoff_log_format structure embedded in the quotaoff item.
469 * It is at this point that we assert that all of the extent
470 * slots in the quotaoff item have been filled.
473 xfs_qm_qoff_logitem_format(xfs_qoff_logitem_t *qf,
474 xfs_log_iovec_t *log_vector)
476 ASSERT(qf->qql_format.qf_type == XFS_LI_QUOTAOFF);
478 log_vector->i_addr = (xfs_caddr_t)&(qf->qql_format);
479 log_vector->i_len = sizeof(xfs_qoff_logitem_t);
480 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_QUOTAOFF);
481 qf->qql_format.qf_size = 1;
486 * Pinning has no meaning for an quotaoff item, so just return.
490 xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t *qf)
497 * Since pinning has no meaning for an quotaoff item, unpinning does
502 xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t *qf, int stale)
509 xfs_qm_qoff_logitem_unpin_remove(xfs_qoff_logitem_t *qf, xfs_trans_t *tp)
515 * Quotaoff items have no locking, so just return success.
519 xfs_qm_qoff_logitem_trylock(xfs_qoff_logitem_t *qf)
521 return XFS_ITEM_LOCKED;
525 * Quotaoff items have no locking or pushing, so return failure
526 * so that the caller doesn't bother with us.
530 xfs_qm_qoff_logitem_unlock(xfs_qoff_logitem_t *qf)
536 * The quotaoff-start-item is logged only once and cannot be moved in the log,
537 * so simply return the lsn at which it's been logged.
541 xfs_qm_qoff_logitem_committed(xfs_qoff_logitem_t *qf, xfs_lsn_t lsn)
547 * There isn't much you can do to push on an quotaoff item. It is simply
548 * stuck waiting for the log to be flushed to disk.
552 xfs_qm_qoff_logitem_push(xfs_qoff_logitem_t *qf)
560 xfs_qm_qoffend_logitem_committed(
561 xfs_qoff_logitem_t *qfe,
564 xfs_qoff_logitem_t *qfs;
567 qfs = qfe->qql_start_lip;
568 AIL_LOCK(qfs->qql_item.li_mountp,s);
570 * Delete the qoff-start logitem from the AIL.
571 * xfs_trans_delete_ail() drops the AIL lock.
573 xfs_trans_delete_ail(qfs->qql_item.li_mountp, (xfs_log_item_t *)qfs, s);
574 kmem_free(qfs, sizeof(xfs_qoff_logitem_t));
575 kmem_free(qfe, sizeof(xfs_qoff_logitem_t));
576 return (xfs_lsn_t)-1;
580 * XXX rcc - don't know quite what to do with this. I think we can
581 * just ignore it. The only time that isn't the case is if we allow
582 * the client to somehow see that quotas have been turned off in which
583 * we can't allow that to get back until the quotaoff hits the disk.
584 * So how would that happen? Also, do we need different routines for
585 * quotaoff start and quotaoff end? I suspect the answer is yes but
586 * to be sure, I need to look at the recovery code and see how quota off
587 * recovery is handled (do we roll forward or back or do something else).
588 * If we roll forwards or backwards, then we need two separate routines,
589 * one that does nothing and one that stamps in the lsn that matters
590 * (truly makes the quotaoff irrevocable). If we do something else,
591 * then maybe we don't need two.
595 xfs_qm_qoff_logitem_committing(xfs_qoff_logitem_t *qip, xfs_lsn_t commit_lsn)
602 xfs_qm_qoffend_logitem_committing(xfs_qoff_logitem_t *qip, xfs_lsn_t commit_lsn)
607 static struct xfs_item_ops xfs_qm_qoffend_logitem_ops = {
608 .iop_size = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_size,
609 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
610 xfs_qm_qoff_logitem_format,
611 .iop_pin = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
612 .iop_unpin = (void(*)(xfs_log_item_t* ,int))
613 xfs_qm_qoff_logitem_unpin,
614 .iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
615 xfs_qm_qoff_logitem_unpin_remove,
616 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,
617 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unlock,
618 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
619 xfs_qm_qoffend_logitem_committed,
620 .iop_push = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_push,
622 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
623 xfs_qm_qoffend_logitem_committing
627 * This is the ops vector shared by all quotaoff-start log items.
629 static struct xfs_item_ops xfs_qm_qoff_logitem_ops = {
630 .iop_size = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_size,
631 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
632 xfs_qm_qoff_logitem_format,
633 .iop_pin = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
634 .iop_unpin = (void(*)(xfs_log_item_t*, int))
635 xfs_qm_qoff_logitem_unpin,
636 .iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
637 xfs_qm_qoff_logitem_unpin_remove,
638 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,
639 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unlock,
640 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
641 xfs_qm_qoff_logitem_committed,
642 .iop_push = (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_push,
644 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
645 xfs_qm_qoff_logitem_committing
649 * Allocate and initialize an quotaoff item of the correct quota type(s).
652 xfs_qm_qoff_logitem_init(
653 struct xfs_mount *mp,
654 xfs_qoff_logitem_t *start,
657 xfs_qoff_logitem_t *qf;
659 qf = (xfs_qoff_logitem_t*) kmem_zalloc(sizeof(xfs_qoff_logitem_t), KM_SLEEP);
661 qf->qql_item.li_type = XFS_LI_QUOTAOFF;
663 qf->qql_item.li_ops = &xfs_qm_qoffend_logitem_ops;
665 qf->qql_item.li_ops = &xfs_qm_qoff_logitem_ops;
666 qf->qql_item.li_mountp = mp;
667 qf->qql_format.qf_type = XFS_LI_QUOTAOFF;
668 qf->qql_format.qf_flags = flags;
669 qf->qql_start_lip = start;