2 * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
38 #include "xfs_trans.h"
43 #include "xfs_alloc.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_quota.h"
46 #include "xfs_mount.h"
47 #include "xfs_alloc_btree.h"
48 #include "xfs_bmap_btree.h"
49 #include "xfs_ialloc_btree.h"
50 #include "xfs_btree.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode.h"
59 #include "xfs_rtalloc.h"
60 #include "xfs_error.h"
61 #include "xfs_itable.h"
67 #include "xfs_buf_item.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_utils.h"
74 * The global quota manager. There is only one of these for the entire
75 * system, _not_ one per file system. XQM keeps track of the overall
76 * quota functionality, including maintaining the freelist and hash
80 struct xfs_qm *xfs_Gqm;
82 kmem_zone_t *qm_dqzone;
83 kmem_zone_t *qm_dqtrxzone;
84 STATIC kmem_shaker_t xfs_qm_shaker;
86 STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
87 STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
89 STATIC void xfs_qm_freelist_init(xfs_frlist_t *);
90 STATIC void xfs_qm_freelist_destroy(xfs_frlist_t *);
91 STATIC int xfs_qm_mplist_nowait(xfs_mount_t *);
92 STATIC int xfs_qm_dqhashlock_nowait(xfs_dquot_t *);
94 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
95 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
96 STATIC int xfs_qm_shake(int, unsigned int);
99 extern mutex_t qcheck_lock;
103 #define XQM_LIST_PRINT(l, NXT, title) \
105 xfs_dquot_t *dqp; int i = 0; \
106 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
107 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
108 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
109 "bcnt = %d, icnt = %d, refs = %d", \
110 ++i, (int) INT_GET(dqp->q_core.d_id, ARCH_CONVERT), \
111 DQFLAGTO_TYPESTR(dqp), \
112 (int) INT_GET(dqp->q_core.d_bcount, ARCH_CONVERT), \
113 (int) INT_GET(dqp->q_core.d_icount, ARCH_CONVERT), \
114 (int) dqp->q_nrefs); } \
117 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
121 * Initialize the XQM structure.
122 * Note that there is not one quota manager per file system.
124 STATIC struct xfs_qm *
130 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
134 * Initialize the dquot hash tables.
136 hsize = (DQUOT_HASH_HEURISTIC < XFS_QM_NCSIZE_THRESHOLD) ?
137 XFS_QM_HASHSIZE_LOW : XFS_QM_HASHSIZE_HIGH;
138 xqm->qm_dqhashmask = hsize - 1;
140 xqm->qm_usr_dqhtable = (xfs_dqhash_t *)kmem_zalloc(hsize *
141 sizeof(xfs_dqhash_t),
143 xqm->qm_grp_dqhtable = (xfs_dqhash_t *)kmem_zalloc(hsize *
144 sizeof(xfs_dqhash_t),
146 ASSERT(xqm->qm_usr_dqhtable != NULL);
147 ASSERT(xqm->qm_grp_dqhtable != NULL);
149 for (i = 0; i < hsize; i++) {
150 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
151 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
155 * Freelist of all dquots of all file systems
157 xfs_qm_freelist_init(&(xqm->qm_dqfreelist));
160 * dquot zone. we register our own low-memory callback.
163 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
165 qm_dqzone = xqm->qm_dqzone;
167 xqm->qm_dqzone = qm_dqzone;
169 xfs_qm_shaker = kmem_shake_register(xfs_qm_shake);
172 * The t_dqinfo portion of transactions.
175 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
177 qm_dqtrxzone = xqm->qm_dqtrxzone;
179 xqm->qm_dqtrxzone = qm_dqtrxzone;
181 atomic_set(&xqm->qm_totaldquots, 0);
182 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
185 mutex_init(&qcheck_lock, MUTEX_DEFAULT, "qchk");
191 * Destroy the global quota manager when its reference count goes to zero.
200 ASSERT(xqm->qm_nrefs == 0);
201 kmem_shake_deregister(xfs_qm_shaker);
202 hsize = xqm->qm_dqhashmask + 1;
203 for (i = 0; i < hsize; i++) {
204 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
205 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
207 kmem_free(xqm->qm_usr_dqhtable, hsize * sizeof(xfs_dqhash_t));
208 kmem_free(xqm->qm_grp_dqhtable, hsize * sizeof(xfs_dqhash_t));
209 xqm->qm_usr_dqhtable = NULL;
210 xqm->qm_grp_dqhtable = NULL;
211 xqm->qm_dqhashmask = 0;
212 xfs_qm_freelist_destroy(&(xqm->qm_dqfreelist));
214 mutex_destroy(&qcheck_lock);
216 kmem_free(xqm, sizeof(xfs_qm_t));
220 * Called at mount time to let XQM know that another file system is
221 * starting quotas. This isn't crucial information as the individual mount
222 * structures are pretty independent, but it helps the XQM keep a
223 * global view of what's going on.
227 xfs_qm_hold_quotafs_ref(
228 struct xfs_mount *mp)
231 * Need to lock the xfs_Gqm structure for things like this. For example,
232 * the structure could disappear between the entry to this routine and
233 * a HOLD operation if not locked.
235 XFS_QM_LOCK(xfs_Gqm);
238 xfs_Gqm = xfs_Gqm_init();
240 * We can keep a list of all filesystems with quotas mounted for
241 * debugging and statistical purposes, but ...
242 * Just take a reference and get out.
244 XFS_QM_HOLD(xfs_Gqm);
245 XFS_QM_UNLOCK(xfs_Gqm);
252 * Release the reference that a filesystem took at mount time,
253 * so that we know when we need to destroy the entire quota manager.
257 xfs_qm_rele_quotafs_ref(
258 struct xfs_mount *mp)
260 xfs_dquot_t *dqp, *nextdqp;
263 ASSERT(xfs_Gqm->qm_nrefs > 0);
266 * Go thru the freelist and destroy all inactive dquots.
268 xfs_qm_freelist_lock(xfs_Gqm);
270 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
271 dqp != (xfs_dquot_t *)&(xfs_Gqm->qm_dqfreelist); ) {
273 nextdqp = dqp->dq_flnext;
274 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
275 ASSERT(dqp->q_mount == NULL);
276 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
277 ASSERT(dqp->HL_PREVP == NULL);
278 ASSERT(dqp->MPL_PREVP == NULL);
279 XQM_FREELIST_REMOVE(dqp);
281 xfs_qm_dqdestroy(dqp);
287 xfs_qm_freelist_unlock(xfs_Gqm);
290 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
293 XFS_QM_LOCK(xfs_Gqm);
294 XFS_QM_RELE(xfs_Gqm);
295 if (xfs_Gqm->qm_nrefs == 0) {
296 xfs_qm_destroy(xfs_Gqm);
299 XFS_QM_UNLOCK(xfs_Gqm);
303 * This is called at mount time from xfs_mountfs to initialize the quotainfo
304 * structure and start the global quotamanager (xfs_Gqm) if it hasn't done
305 * so already. Note that the superblock has not been read in yet.
308 xfs_qm_mount_quotainit(
313 * User, projects or group quotas has to be on.
315 ASSERT(flags & (XFSMNT_UQUOTA | XFSMNT_PQUOTA | XFSMNT_GQUOTA));
318 * Initialize the flags in the mount structure. From this point
319 * onwards we look at m_qflags to figure out if quotas's ON/OFF, etc.
320 * Note that we enforce nothing if accounting is off.
321 * ie. XFSMNT_*QUOTA must be ON for XFSMNT_*QUOTAENF.
322 * It isn't necessary to take the quotaoff lock to do this; this is
325 if (flags & XFSMNT_UQUOTA) {
326 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
327 if (flags & XFSMNT_UQUOTAENF)
328 mp->m_qflags |= XFS_UQUOTA_ENFD;
330 if (flags & XFSMNT_GQUOTA) {
331 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
332 if (flags & XFSMNT_GQUOTAENF)
333 mp->m_qflags |= XFS_OQUOTA_ENFD;
334 } else if (flags & XFSMNT_PQUOTA) {
335 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
336 if (flags & XFSMNT_PQUOTAENF)
337 mp->m_qflags |= XFS_OQUOTA_ENFD;
342 * Just destroy the quotainfo structure.
345 xfs_qm_unmount_quotadestroy(
349 xfs_qm_destroy_quotainfo(mp);
354 * This is called from xfs_mountfs to start quotas and initialize all
355 * necessary data structures like quotainfo. This is also responsible for
356 * running a quotacheck as necessary. We are guaranteed that the superblock
357 * is consistently read in at this point.
369 * If a file system had quotas running earlier, but decided to
370 * mount without -o uquota/pquota/gquota options, revoke the
371 * quotachecked license, and bail out.
373 if (! XFS_IS_QUOTA_ON(mp) &&
374 (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT)) {
380 * If quotas on realtime volumes is not supported, we disable
381 * quotas immediately.
383 if (mp->m_sb.sb_rextents) {
385 "Cannot turn on quotas for realtime filesystem %s",
391 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
394 * Allocate the quotainfo structure inside the mount struct, and
395 * create quotainode(s), and change/rev superblock if necessary.
397 if ((error = xfs_qm_init_quotainfo(mp))) {
399 * We must turn off quotas.
401 ASSERT(mp->m_quotainfo == NULL);
406 * If any of the quotas are not consistent, do a quotacheck.
408 if (XFS_QM_NEED_QUOTACHECK(mp) &&
409 !(mfsi_flags & XFS_MFSI_NO_QUOTACHECK)) {
410 if ((error = xfs_qm_quotacheck(mp))) {
411 /* Quotacheck has failed and quotas have
414 return XFS_ERROR(error);
420 * We actually don't have to acquire the SB_LOCK at all.
421 * This can only be called from mount, and that's single threaded. XXX
424 sbf = mp->m_sb.sb_qflags;
425 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
426 XFS_SB_UNLOCK(mp, s);
428 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
429 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
431 * We could only have been turning quotas off.
432 * We aren't in very good shape actually because
433 * the incore structures are convinced that quotas are
434 * off, but the on disk superblock doesn't know that !
436 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
437 xfs_fs_cmn_err(CE_ALERT, mp,
438 "XFS mount_quotas: Superblock update failed!");
443 xfs_fs_cmn_err(CE_WARN, mp,
444 "Failed to initialize disk quotas.");
446 return XFS_ERROR(error);
450 * Called from the vfsops layer.
453 xfs_qm_unmount_quotas(
456 xfs_inode_t *uqp, *gqp;
460 * Release the dquots that root inode, et al might be holding,
461 * before we flush quotas and blow away the quotainfo structure.
463 ASSERT(mp->m_rootip);
464 xfs_qm_dqdetach(mp->m_rootip);
466 xfs_qm_dqdetach(mp->m_rbmip);
468 xfs_qm_dqdetach(mp->m_rsumip);
471 * Flush out the quota inodes.
474 if (mp->m_quotainfo) {
475 if ((uqp = mp->m_quotainfo->qi_uquotaip) != NULL) {
476 xfs_ilock(uqp, XFS_ILOCK_EXCL);
478 error = xfs_iflush(uqp, XFS_IFLUSH_SYNC);
479 xfs_iunlock(uqp, XFS_ILOCK_EXCL);
480 if (unlikely(error == EFSCORRUPTED)) {
481 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(1)",
482 XFS_ERRLEVEL_LOW, mp);
486 if ((gqp = mp->m_quotainfo->qi_gquotaip) != NULL) {
487 xfs_ilock(gqp, XFS_ILOCK_EXCL);
489 error = xfs_iflush(gqp, XFS_IFLUSH_SYNC);
490 xfs_iunlock(gqp, XFS_ILOCK_EXCL);
491 if (unlikely(error == EFSCORRUPTED)) {
492 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(2)",
493 XFS_ERRLEVEL_LOW, mp);
499 XFS_PURGE_INODE(uqp);
500 mp->m_quotainfo->qi_uquotaip = NULL;
503 XFS_PURGE_INODE(gqp);
504 mp->m_quotainfo->qi_gquotaip = NULL;
507 return XFS_ERROR(error);
511 * Flush all dquots of the given file system to disk. The dquots are
512 * _not_ purged from memory here, just their data written to disk.
524 if (mp->m_quotainfo == NULL)
528 xfs_qm_mplist_lock(mp);
529 FOREACH_DQUOT_IN_MP(dqp, mp) {
531 if (! XFS_DQ_IS_DIRTY(dqp)) {
535 xfs_dqtrace_entry(dqp, "FLUSHALL: DQDIRTY");
536 /* XXX a sentinel would be better */
537 recl = XFS_QI_MPLRECLAIMS(mp);
538 if (! xfs_qm_dqflock_nowait(dqp)) {
540 * If we can't grab the flush lock then check
541 * to see if the dquot has been flushed delayed
542 * write. If so, grab its buffer and send it
543 * out immediately. We'll be able to acquire
544 * the flush lock when the I/O completes.
546 xfs_qm_dqflock_pushbuf_wait(dqp);
549 * Let go of the mplist lock. We don't want to hold it
550 * across a disk write.
552 xfs_qm_mplist_unlock(mp);
553 error = xfs_qm_dqflush(dqp, flags);
558 xfs_qm_mplist_lock(mp);
559 if (recl != XFS_QI_MPLRECLAIMS(mp)) {
560 xfs_qm_mplist_unlock(mp);
561 /* XXX restart limit */
566 xfs_qm_mplist_unlock(mp);
571 * Release the group dquot pointers the user dquots may be
572 * carrying around as a hint. mplist is locked on entry and exit.
575 xfs_qm_detach_gdquots(
578 xfs_dquot_t *dqp, *gdqp;
582 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
583 dqp = XFS_QI_MPLNEXT(mp);
586 if ((gdqp = dqp->q_gdquot)) {
588 dqp->q_gdquot = NULL;
594 * Can't hold the mplist lock across a dqput.
595 * XXXmust convert to marker based iterations here.
597 nrecl = XFS_QI_MPLRECLAIMS(mp);
598 xfs_qm_mplist_unlock(mp);
601 xfs_qm_mplist_lock(mp);
602 if (nrecl != XFS_QI_MPLRECLAIMS(mp))
610 * Go through all the incore dquots of this file system and take them
611 * off the mplist and hashlist, if the dquot type matches the dqtype
612 * parameter. This is used when turning off quota accounting for
613 * users and/or groups, as well as when the filesystem is unmounting.
618 uint flags) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
623 xfs_dquot_t *nextdqp;
626 if (mp->m_quotainfo == NULL)
629 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
630 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
631 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
633 xfs_qm_mplist_lock(mp);
636 * In the first pass through all incore dquots of this filesystem,
637 * we release the group dquot pointers the user dquots may be
638 * carrying around as a hint. We need to do this irrespective of
639 * what's being turned off.
641 xfs_qm_detach_gdquots(mp);
645 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
647 * Try to get rid of all of the unwanted dquots. The idea is to
648 * get them off mplist and hashlist, but leave them on freelist.
650 dqp = XFS_QI_MPLNEXT(mp);
653 * It's OK to look at the type without taking dqlock here.
654 * We're holding the mplist lock here, and that's needed for
657 if ((dqp->dq_flags & dqtype) == 0) {
662 if (! xfs_qm_dqhashlock_nowait(dqp)) {
663 nrecl = XFS_QI_MPLRECLAIMS(mp);
664 xfs_qm_mplist_unlock(mp);
665 XFS_DQ_HASH_LOCK(dqp->q_hash);
666 xfs_qm_mplist_lock(mp);
669 * XXXTheoretically, we can get into a very long
670 * ping pong game here.
671 * No one can be adding dquots to the mplist at
672 * this point, but somebody might be taking things off.
674 if (nrecl != XFS_QI_MPLRECLAIMS(mp)) {
675 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
681 * Take the dquot off the mplist and hashlist. It may remain on
682 * freelist in INACTIVE state.
684 nextdqp = dqp->MPL_NEXT;
685 nmisses += xfs_qm_dqpurge(dqp, flags);
688 xfs_qm_mplist_unlock(mp);
700 * Purge the dquot cache.
701 * None of the dquots should really be busy at this point.
703 if (mp->m_quotainfo) {
704 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
718 xfs_dquot_t *udqhint, /* hint */
719 xfs_dquot_t **IO_idqpp)
724 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
727 * See if we already have it in the inode itself. IO_idqpp is
728 * &i_udquot or &i_gdquot. This made the code look weird, but
729 * made the logic a lot simpler.
731 if ((dqp = *IO_idqpp)) {
734 xfs_dqtrace_entry(dqp, "DQATTACH: found in ip");
739 * udqhint is the i_udquot field in inode, and is non-NULL only
740 * when the type arg is group/project. Its purpose is to save a
741 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
744 ASSERT(!udqhint || type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
745 if (udqhint && !dolock)
749 * No need to take dqlock to look at the id.
750 * The ID can't change until it gets reclaimed, and it won't
751 * be reclaimed as long as we have a ref from inode and we hold
755 (dqp = udqhint->q_gdquot) &&
756 (INT_GET(dqp->q_core.d_id, ARCH_CONVERT) == id)) {
757 ASSERT(XFS_DQ_IS_LOCKED(udqhint));
760 ASSERT(*IO_idqpp == NULL);
764 xfs_dqunlock(udqhint);
769 * We can't hold a dquot lock when we call the dqget code.
770 * We'll deadlock in no time, because of (not conforming to)
771 * lock ordering - the inodelock comes before any dquot lock,
772 * and we may drop and reacquire the ilock in xfs_qm_dqget().
775 xfs_dqunlock(udqhint);
777 * Find the dquot from somewhere. This bumps the
778 * reference count of dquot and returns it locked.
779 * This can return ENOENT if dquot didn't exist on
780 * disk and we didn't ask it to allocate;
781 * ESRCH if quotas got turned off suddenly.
783 if ((error = xfs_qm_dqget(ip->i_mount, ip, id, type,
784 doalloc|XFS_QMOPT_DOWARN, &dqp))) {
785 if (udqhint && dolock)
790 xfs_dqtrace_entry(dqp, "DQATTACH: found by dqget");
792 * dqget may have dropped and re-acquired the ilock, but it guarantees
793 * that the dquot returned is the one that should go in the inode.
797 ASSERT(XFS_DQ_IS_LOCKED(dqp));
807 ASSERT(XFS_DQ_IS_LOCKED(dqp));
808 if (! xfs_qm_dqlock_nowait(udqhint)) {
817 ASSERT(XFS_DQ_IS_LOCKED(udqhint));
821 ASSERT(XFS_DQ_IS_LOCKED(dqp));
829 * Given a udquot and gdquot, attach a ptr to the group dquot in the
830 * udquot as a hint for future lookups. The idea sounds simple, but the
831 * execution isn't, because the udquot might have a group dquot attached
832 * already and getting rid of that gets us into lock ordering contraints.
833 * The process is complicated more by the fact that the dquots may or may not
834 * be locked on entry.
837 xfs_qm_dqattach_grouphint(
846 ASSERT(XFS_DQ_IS_LOCKED(udq));
847 ASSERT(XFS_DQ_IS_LOCKED(gdq));
853 if ((tmp = udq->q_gdquot)) {
860 udq->q_gdquot = NULL;
862 * We can't keep any dqlocks when calling dqrele,
863 * because the freelist lock comes before dqlocks.
869 * we took a hard reference once upon a time in dqget,
870 * so give it back when the udquot no longer points at it
871 * dqput() does the unlocking of the dquot.
879 ASSERT(XFS_DQ_IS_LOCKED(udq));
885 ASSERT(XFS_DQ_IS_LOCKED(udq));
886 ASSERT(XFS_DQ_IS_LOCKED(gdq));
888 * Somebody could have attached a gdquot here,
889 * when we dropped the uqlock. If so, just do nothing.
891 if (udq->q_gdquot == NULL) {
903 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
905 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
906 * If XFS_QMOPT_DQLOCK, the dquot(s) will be returned locked. This option pretty
907 * much made this code a complete mess, but it has been pretty useful.
908 * If XFS_QMOPT_ILOCKED, then inode sent is already locked EXCL.
909 * Inode may get unlocked and relocked in here, and the caller must deal with
917 xfs_mount_t *mp = ip->i_mount;
921 if ((! XFS_IS_QUOTA_ON(mp)) ||
922 (! XFS_NOT_DQATTACHED(mp, ip)) ||
923 (ip->i_ino == mp->m_sb.sb_uquotino) ||
924 (ip->i_ino == mp->m_sb.sb_gquotino))
927 ASSERT((flags & XFS_QMOPT_ILOCKED) == 0 ||
928 XFS_ISLOCKED_INODE_EXCL(ip));
930 if (! (flags & XFS_QMOPT_ILOCKED))
931 xfs_ilock(ip, XFS_ILOCK_EXCL);
933 if (XFS_IS_UQUOTA_ON(mp)) {
934 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
935 flags & XFS_QMOPT_DQALLOC,
936 flags & XFS_QMOPT_DQLOCK,
937 NULL, &ip->i_udquot);
942 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
943 if (XFS_IS_OQUOTA_ON(mp)) {
944 error = XFS_IS_GQUOTA_ON(mp) ?
945 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
946 flags & XFS_QMOPT_DQALLOC,
947 flags & XFS_QMOPT_DQLOCK,
948 ip->i_udquot, &ip->i_gdquot) :
949 xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
950 flags & XFS_QMOPT_DQALLOC,
951 flags & XFS_QMOPT_DQLOCK,
952 ip->i_udquot, &ip->i_gdquot);
954 * Don't worry about the udquot that we may have
955 * attached above. It'll get detached, if not already.
963 * Attach this group quota to the user quota as a hint.
964 * This WON'T, in general, result in a thrash.
967 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
968 ASSERT(ip->i_udquot);
969 ASSERT(ip->i_gdquot);
972 * We may or may not have the i_udquot locked at this point,
973 * but this check is OK since we don't depend on the i_gdquot to
974 * be accurate 100% all the time. It is just a hint, and this
975 * will succeed in general.
977 if (ip->i_udquot->q_gdquot == ip->i_gdquot)
980 * Attach i_gdquot to the gdquot hint inside the i_udquot.
982 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot,
983 flags & XFS_QMOPT_DQLOCK);
991 if (flags & XFS_QMOPT_DQLOCK)
992 ASSERT(XFS_DQ_IS_LOCKED(ip->i_udquot));
995 if (flags & XFS_QMOPT_DQLOCK)
996 ASSERT(XFS_DQ_IS_LOCKED(ip->i_gdquot));
998 if (XFS_IS_UQUOTA_ON(mp))
999 ASSERT(ip->i_udquot);
1000 if (XFS_IS_OQUOTA_ON(mp))
1001 ASSERT(ip->i_gdquot);
1005 if (! (flags & XFS_QMOPT_ILOCKED))
1006 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1010 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
1016 * Release dquots (and their references) if any.
1017 * The inode should be locked EXCL except when this's called by
1024 if (!(ip->i_udquot || ip->i_gdquot))
1027 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
1028 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
1030 xfs_dqtrace_entry_ino(ip->i_udquot, "DQDETTACH", ip);
1031 xfs_qm_dqrele(ip->i_udquot);
1032 ip->i_udquot = NULL;
1035 xfs_dqtrace_entry_ino(ip->i_gdquot, "DQDETTACH", ip);
1036 xfs_qm_dqrele(ip->i_gdquot);
1037 ip->i_gdquot = NULL;
1042 * This is called by VFS_SYNC and flags arg determines the caller,
1043 * and its motives, as done in xfs_sync.
1045 * vfs_sync: SYNC_FSDATA|SYNC_ATTR|SYNC_BDFLUSH 0x31
1046 * syscall sync: SYNC_FSDATA|SYNC_ATTR|SYNC_DELWRI 0x25
1047 * umountroot : SYNC_WAIT | SYNC_CLOSE | SYNC_ATTR | SYNC_FSDATA
1063 * We won't block unless we are asked to.
1065 nowait = (boolean_t)(flags & SYNC_BDFLUSH || (flags & SYNC_WAIT) == 0);
1068 xfs_qm_mplist_lock(mp);
1070 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
1071 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
1072 * when we have the mplist lock, we know that dquots will be consistent
1073 * as long as we have it locked.
1075 if (! XFS_IS_QUOTA_ON(mp)) {
1076 xfs_qm_mplist_unlock(mp);
1079 FOREACH_DQUOT_IN_MP(dqp, mp) {
1081 * If this is vfs_sync calling, then skip the dquots that
1082 * don't 'seem' to be dirty. ie. don't acquire dqlock.
1083 * This is very similar to what xfs_sync does with inodes.
1085 if (flags & SYNC_BDFLUSH) {
1086 if (! XFS_DQ_IS_DIRTY(dqp))
1092 * Try to acquire the dquot lock. We are NOT out of
1093 * lock order, but we just don't want to wait for this
1094 * lock, unless somebody wanted us to.
1096 if (! xfs_qm_dqlock_nowait(dqp))
1103 * Now, find out for sure if this dquot is dirty or not.
1105 if (! XFS_DQ_IS_DIRTY(dqp)) {
1110 /* XXX a sentinel would be better */
1111 recl = XFS_QI_MPLRECLAIMS(mp);
1112 if (! xfs_qm_dqflock_nowait(dqp)) {
1118 * If we can't grab the flush lock then if the caller
1119 * really wanted us to give this our best shot,
1120 * see if we can give a push to the buffer before we wait
1121 * on the flush lock. At this point, we know that
1122 * eventhough the dquot is being flushed,
1123 * it has (new) dirty data.
1125 xfs_qm_dqflock_pushbuf_wait(dqp);
1128 * Let go of the mplist lock. We don't want to hold it
1129 * across a disk write
1131 flush_flags = (nowait) ? XFS_QMOPT_DELWRI : XFS_QMOPT_SYNC;
1132 xfs_qm_mplist_unlock(mp);
1133 xfs_dqtrace_entry(dqp, "XQM_SYNC: DQFLUSH");
1134 error = xfs_qm_dqflush(dqp, flush_flags);
1136 if (error && XFS_FORCED_SHUTDOWN(mp))
1137 return(0); /* Need to prevent umount failure */
1141 xfs_qm_mplist_lock(mp);
1142 if (recl != XFS_QI_MPLRECLAIMS(mp)) {
1143 if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
1146 xfs_qm_mplist_unlock(mp);
1151 xfs_qm_mplist_unlock(mp);
1157 * This initializes all the quota information that's kept in the
1161 xfs_qm_init_quotainfo(
1164 xfs_quotainfo_t *qinf;
1168 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1171 * Tell XQM that we exist as soon as possible.
1173 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1177 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1180 * See if quotainodes are setup, and if not, allocate them,
1181 * and change the superblock accordingly.
1183 if ((error = xfs_qm_init_quotainos(mp))) {
1184 kmem_free(qinf, sizeof(xfs_quotainfo_t));
1185 mp->m_quotainfo = NULL;
1189 spinlock_init(&qinf->qi_pinlock, "xfs_qinf_pin");
1190 xfs_qm_list_init(&qinf->qi_dqlist, "mpdqlist", 0);
1191 qinf->qi_dqreclaims = 0;
1193 /* mutex used to serialize quotaoffs */
1194 mutex_init(&qinf->qi_quotaofflock, MUTEX_DEFAULT, "qoff");
1196 /* Precalc some constants */
1197 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1198 ASSERT(qinf->qi_dqchunklen);
1199 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1200 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1202 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1205 * We try to get the limits from the superuser's limits fields.
1206 * This is quite hacky, but it is standard quota practice.
1207 * We look at the USR dquot with id == 0 first, but if user quotas
1208 * are not enabled we goto the GRP dquot with id == 0.
1209 * We don't really care to keep separate default limits for user
1210 * and group quotas, at least not at this point.
1212 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1213 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
1214 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1216 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1219 xfs_disk_dquot_t *ddqp = &dqp->q_core;
1222 * The warnings and timers set the grace period given to
1223 * a user or group before he or she can not perform any
1224 * more writing. If it is zero, a default is used.
1226 qinf->qi_btimelimit =
1227 INT_GET(ddqp->d_btimer, ARCH_CONVERT) ?
1228 INT_GET(ddqp->d_btimer, ARCH_CONVERT) :
1230 qinf->qi_itimelimit =
1231 INT_GET(ddqp->d_itimer, ARCH_CONVERT) ?
1232 INT_GET(ddqp->d_itimer, ARCH_CONVERT) :
1234 qinf->qi_rtbtimelimit =
1235 INT_GET(ddqp->d_rtbtimer, ARCH_CONVERT) ?
1236 INT_GET(ddqp->d_rtbtimer, ARCH_CONVERT) :
1237 XFS_QM_RTBTIMELIMIT;
1238 qinf->qi_bwarnlimit =
1239 INT_GET(ddqp->d_bwarns, ARCH_CONVERT) ?
1240 INT_GET(ddqp->d_bwarns, ARCH_CONVERT) :
1242 qinf->qi_iwarnlimit =
1243 INT_GET(ddqp->d_iwarns, ARCH_CONVERT) ?
1244 INT_GET(ddqp->d_iwarns, ARCH_CONVERT) :
1246 qinf->qi_rtbwarnlimit =
1247 INT_GET(ddqp->d_rtbwarns, ARCH_CONVERT) ?
1248 INT_GET(ddqp->d_rtbwarns, ARCH_CONVERT) :
1249 XFS_QM_RTBWARNLIMIT;
1250 qinf->qi_bhardlimit =
1251 INT_GET(ddqp->d_blk_hardlimit, ARCH_CONVERT);
1252 qinf->qi_bsoftlimit =
1253 INT_GET(ddqp->d_blk_softlimit, ARCH_CONVERT);
1254 qinf->qi_ihardlimit =
1255 INT_GET(ddqp->d_ino_hardlimit, ARCH_CONVERT);
1256 qinf->qi_isoftlimit =
1257 INT_GET(ddqp->d_ino_softlimit, ARCH_CONVERT);
1258 qinf->qi_rtbhardlimit =
1259 INT_GET(ddqp->d_rtb_hardlimit, ARCH_CONVERT);
1260 qinf->qi_rtbsoftlimit =
1261 INT_GET(ddqp->d_rtb_softlimit, ARCH_CONVERT);
1264 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1265 * we don't want this dquot cached. We haven't done a
1266 * quotacheck yet, and quotacheck doesn't like incore dquots.
1268 xfs_qm_dqdestroy(dqp);
1270 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1271 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1272 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1273 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1274 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1275 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1283 * Gets called when unmounting a filesystem or when all quotas get
1285 * This purges the quota inodes, destroys locks and frees itself.
1288 xfs_qm_destroy_quotainfo(
1291 xfs_quotainfo_t *qi;
1293 qi = mp->m_quotainfo;
1295 ASSERT(xfs_Gqm != NULL);
1298 * Release the reference that XQM kept, so that we know
1299 * when the XQM structure should be freed. We cannot assume
1300 * that xfs_Gqm is non-null after this point.
1302 xfs_qm_rele_quotafs_ref(mp);
1304 spinlock_destroy(&qi->qi_pinlock);
1305 xfs_qm_list_destroy(&qi->qi_dqlist);
1307 if (qi->qi_uquotaip) {
1308 XFS_PURGE_INODE(qi->qi_uquotaip);
1309 qi->qi_uquotaip = NULL; /* paranoia */
1311 if (qi->qi_gquotaip) {
1312 XFS_PURGE_INODE(qi->qi_gquotaip);
1313 qi->qi_gquotaip = NULL;
1315 mutex_destroy(&qi->qi_quotaofflock);
1316 kmem_free(qi, sizeof(xfs_quotainfo_t));
1317 mp->m_quotainfo = NULL;
1322 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1331 mutex_init(&list->qh_lock, MUTEX_DEFAULT, str);
1332 list->qh_next = NULL;
1333 list->qh_version = 0;
1334 list->qh_nelems = 0;
1338 xfs_qm_list_destroy(
1341 mutex_destroy(&(list->qh_lock));
1346 * Stripped down version of dqattach. This doesn't attach, or even look at the
1347 * dquots attached to the inode. The rationale is that there won't be any
1348 * attached at the time this is called from quotacheck.
1351 xfs_qm_dqget_noattach(
1353 xfs_dquot_t **O_udqpp,
1354 xfs_dquot_t **O_gdqpp)
1358 xfs_dquot_t *udqp, *gdqp;
1360 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
1365 if (XFS_IS_UQUOTA_ON(mp)) {
1366 ASSERT(ip->i_udquot == NULL);
1368 * We want the dquot allocated if it doesn't exist.
1370 if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
1371 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
1374 * Shouldn't be able to turn off quotas here.
1376 ASSERT(error != ESRCH);
1377 ASSERT(error != ENOENT);
1383 if (XFS_IS_OQUOTA_ON(mp)) {
1384 ASSERT(ip->i_gdquot == NULL);
1387 error = XFS_IS_GQUOTA_ON(mp) ?
1388 xfs_qm_dqget(mp, ip,
1389 ip->i_d.di_gid, XFS_DQ_GROUP,
1390 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1392 xfs_qm_dqget(mp, ip,
1393 ip->i_d.di_projid, XFS_DQ_PROJ,
1394 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1398 xfs_qm_dqrele(udqp);
1399 ASSERT(error != ESRCH);
1400 ASSERT(error != ENOENT);
1405 /* Reacquire the locks in the right order */
1407 if (! xfs_qm_dqlock_nowait(udqp)) {
1419 if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
1420 if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
1426 * Create an inode and return with a reference already taken, but unlocked
1427 * This is how we create quota inodes
1442 tp = xfs_trans_alloc(mp,XFS_TRANS_QM_QINOCREATE);
1443 if ((error = xfs_trans_reserve(tp,
1444 XFS_QM_QINOCREATE_SPACE_RES(mp),
1445 XFS_CREATE_LOG_RES(mp), 0,
1446 XFS_TRANS_PERM_LOG_RES,
1447 XFS_CREATE_LOG_COUNT))) {
1448 xfs_trans_cancel(tp, 0);
1451 memset(&zerocr, 0, sizeof(zerocr));
1453 if ((error = xfs_dir_ialloc(&tp, mp->m_rootip, S_IFREG, 1, 0,
1454 &zerocr, 0, 1, ip, &committed))) {
1455 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1461 * Keep an extra reference to this quota inode. This inode is
1462 * locked exclusively and joined to the transaction already.
1464 ASSERT(XFS_ISLOCKED_INODE_EXCL(*ip));
1465 VN_HOLD(XFS_ITOV((*ip)));
1468 * Make the changes in the superblock, and log those too.
1469 * sbfields arg may contain fields other than *QUOTINO;
1470 * VERSIONNUM for example.
1472 s = XFS_SB_LOCK(mp);
1473 if (flags & XFS_QMOPT_SBVERSION) {
1474 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1475 unsigned oldv = mp->m_sb.sb_versionnum;
1477 ASSERT(!XFS_SB_VERSION_HASQUOTA(&mp->m_sb));
1478 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1479 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1480 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1481 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1483 XFS_SB_VERSION_ADDQUOTA(&mp->m_sb);
1484 mp->m_sb.sb_uquotino = NULLFSINO;
1485 mp->m_sb.sb_gquotino = NULLFSINO;
1487 /* qflags will get updated _after_ quotacheck */
1488 mp->m_sb.sb_qflags = 0;
1489 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1491 "Old superblock version %x, converting to %x.",
1492 oldv, mp->m_sb.sb_versionnum);
1495 if (flags & XFS_QMOPT_UQUOTA)
1496 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1498 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1499 XFS_SB_UNLOCK(mp, s);
1500 xfs_mod_sb(tp, sbfields);
1502 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
1504 xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
1512 xfs_qm_reset_dqcounts(
1518 xfs_disk_dquot_t *ddq;
1521 xfs_buftrace("RESET DQUOTS", bp);
1523 * Reset all counters and timers. They'll be
1524 * started afresh by xfs_qm_quotacheck.
1527 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1528 do_div(j, sizeof(xfs_dqblk_t));
1529 ASSERT(XFS_QM_DQPERBLK(mp) == j);
1531 ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1532 for (j = 0; j < XFS_QM_DQPERBLK(mp); j++) {
1534 * Do a sanity check, and if needed, repair the dqblk. Don't
1535 * output any warnings because it's perfectly possible to
1536 * find unitialized dquot blks. See comment in xfs_qm_dqcheck.
1538 (void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1540 INT_SET(ddq->d_bcount, ARCH_CONVERT, 0ULL);
1541 INT_SET(ddq->d_icount, ARCH_CONVERT, 0ULL);
1542 INT_SET(ddq->d_rtbcount, ARCH_CONVERT, 0ULL);
1543 INT_SET(ddq->d_btimer, ARCH_CONVERT, (time_t)0);
1544 INT_SET(ddq->d_itimer, ARCH_CONVERT, (time_t)0);
1545 INT_SET(ddq->d_rtbtimer, ARCH_CONVERT, (time_t)0);
1546 INT_SET(ddq->d_bwarns, ARCH_CONVERT, 0UL);
1547 INT_SET(ddq->d_iwarns, ARCH_CONVERT, 0UL);
1548 INT_SET(ddq->d_rtbwarns, ARCH_CONVERT, 0UL);
1549 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1560 xfs_filblks_t blkcnt,
1571 incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
1572 XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
1573 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1574 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1578 * Blkcnt arg can be a very big number, and might even be
1579 * larger than the log itself. So, we have to break it up into
1580 * manageable-sized transactions.
1581 * Note that we don't start a permanent transaction here; we might
1582 * not be able to get a log reservation for the whole thing up front,
1583 * and we don't really care to either, because we just discard
1584 * everything if we were to crash in the middle of this loop.
1587 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1588 XFS_FSB_TO_DADDR(mp, bno),
1589 (int)XFS_QI_DQCHUNKLEN(mp), 0, &bp);
1593 (void) xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1594 xfs_bdwrite(mp, bp);
1596 * goto the next block.
1599 firstid += XFS_QM_DQPERBLK(mp);
1605 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1606 * caller supplied function for every chunk of dquots that we find.
1614 xfs_bmbt_irec_t *map;
1615 int i, nmaps; /* number of map entries */
1616 int error; /* return value */
1617 xfs_fileoff_t lblkno;
1618 xfs_filblks_t maxlblkcnt;
1620 xfs_fsblock_t rablkno;
1621 xfs_filblks_t rablkcnt;
1625 * This looks racey, but we can't keep an inode lock across a
1626 * trans_reserve. But, this gets called during quotacheck, and that
1627 * happens only at mount time which is single threaded.
1629 if (qip->i_d.di_nblocks == 0)
1632 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1635 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1637 nmaps = XFS_DQITER_MAP_SIZE;
1639 * We aren't changing the inode itself. Just changing
1640 * some of its data. No new blocks are added here, and
1641 * the inode is never added to the transaction.
1643 xfs_ilock(qip, XFS_ILOCK_SHARED);
1644 error = xfs_bmapi(NULL, qip, lblkno,
1645 maxlblkcnt - lblkno,
1648 0, map, &nmaps, NULL);
1649 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1653 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1654 for (i = 0; i < nmaps; i++) {
1655 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1656 ASSERT(map[i].br_blockcount);
1659 lblkno += map[i].br_blockcount;
1661 if (map[i].br_startblock == HOLESTARTBLOCK)
1664 firstid = (xfs_dqid_t) map[i].br_startoff *
1665 XFS_QM_DQPERBLK(mp);
1667 * Do a read-ahead on the next extent.
1669 if ((i+1 < nmaps) &&
1670 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1671 rablkcnt = map[i+1].br_blockcount;
1672 rablkno = map[i+1].br_startblock;
1673 while (rablkcnt--) {
1674 xfs_baread(mp->m_ddev_targp,
1675 XFS_FSB_TO_DADDR(mp, rablkno),
1676 (int)XFS_QI_DQCHUNKLEN(mp));
1681 * Iterate thru all the blks in the extent and
1682 * reset the counters of all the dquots inside them.
1684 if ((error = xfs_qm_dqiter_bufs(mp,
1686 map[i].br_startblock,
1687 map[i].br_blockcount,
1695 } while (nmaps > 0);
1697 kmem_free(map, XFS_DQITER_MAP_SIZE * sizeof(*map));
1703 * Called by dqusage_adjust in doing a quotacheck.
1704 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1705 * this updates its incore copy as well as the buffer copy. This is
1706 * so that once the quotacheck is done, we can just log all the buffers,
1707 * as opposed to logging numerous updates to individual dquots.
1710 xfs_qm_quotacheck_dqadjust(
1715 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1716 xfs_dqtrace_entry(dqp, "QCHECK DQADJUST");
1718 * Adjust the inode count and the block count to reflect this inode's
1721 INT_MOD(dqp->q_core.d_icount, ARCH_CONVERT, +1);
1722 dqp->q_res_icount++;
1724 INT_MOD(dqp->q_core.d_bcount, ARCH_CONVERT, nblks);
1725 dqp->q_res_bcount += nblks;
1728 INT_MOD(dqp->q_core.d_rtbcount, ARCH_CONVERT, rtblks);
1729 dqp->q_res_rtbcount += rtblks;
1733 * Set default limits, adjust timers (since we changed usages)
1735 if (! XFS_IS_SUSER_DQUOT(dqp)) {
1736 xfs_qm_adjust_dqlimits(dqp->q_mount, &dqp->q_core);
1737 xfs_qm_adjust_dqtimers(dqp->q_mount, &dqp->q_core);
1740 dqp->dq_flags |= XFS_DQ_DIRTY;
1746 xfs_qcnt_t *O_rtblks)
1748 xfs_filblks_t rtblks; /* total rt blks */
1749 xfs_ifork_t *ifp; /* inode fork pointer */
1750 xfs_extnum_t nextents; /* number of extent entries */
1751 xfs_bmbt_rec_t *base; /* base of extent array */
1752 xfs_bmbt_rec_t *ep; /* pointer to an extent entry */
1755 ASSERT(XFS_IS_REALTIME_INODE(ip));
1756 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1757 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1758 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1762 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1763 base = &ifp->if_u1.if_extents[0];
1764 for (ep = base; ep < &base[nextents]; ep++)
1765 rtblks += xfs_bmbt_get_blockcount(ep);
1766 *O_rtblks = (xfs_qcnt_t)rtblks;
1771 * callback routine supplied to bulkstat(). Given an inumber, find its
1772 * dquots and update them to account for resources taken by that inode.
1776 xfs_qm_dqusage_adjust(
1777 xfs_mount_t *mp, /* mount point for filesystem */
1778 xfs_ino_t ino, /* inode number to get data for */
1779 void __user *buffer, /* not used */
1780 int ubsize, /* not used */
1781 void *private_data, /* not used */
1782 xfs_daddr_t bno, /* starting block of inode cluster */
1783 int *ubused, /* not used */
1784 void *dip, /* on-disk inode pointer (not used) */
1785 int *res) /* result code value */
1788 xfs_dquot_t *udqp, *gdqp;
1789 xfs_qcnt_t nblks, rtblks;
1792 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1795 * rootino must have its resources accounted for, not so with the quota
1798 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1799 *res = BULKSTAT_RV_NOTHING;
1800 return XFS_ERROR(EINVAL);
1804 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1805 * interface expects the inode to be exclusively locked because that's
1806 * the case in all other instances. It's OK that we do this because
1807 * quotacheck is done only at mount time.
1809 if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip, bno))) {
1810 *res = BULKSTAT_RV_NOTHING;
1814 if (ip->i_d.di_mode == 0) {
1815 xfs_iput_new(ip, XFS_ILOCK_EXCL);
1816 *res = BULKSTAT_RV_NOTHING;
1817 return XFS_ERROR(ENOENT);
1821 * Obtain the locked dquots. In case of an error (eg. allocation
1822 * fails for ENOSPC), we return the negative of the error number
1823 * to bulkstat, so that it can get propagated to quotacheck() and
1824 * making us disable quotas for the file system.
1826 if ((error = xfs_qm_dqget_noattach(ip, &udqp, &gdqp))) {
1827 xfs_iput(ip, XFS_ILOCK_EXCL);
1828 *res = BULKSTAT_RV_GIVEUP;
1833 if (! XFS_IS_REALTIME_INODE(ip)) {
1834 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks;
1837 * Walk thru the extent list and count the realtime blocks.
1839 if ((error = xfs_qm_get_rtblks(ip, &rtblks))) {
1840 xfs_iput(ip, XFS_ILOCK_EXCL);
1845 *res = BULKSTAT_RV_GIVEUP;
1848 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1850 ASSERT(ip->i_delayed_blks == 0);
1853 * We can't release the inode while holding its dquot locks.
1854 * The inode can go into inactive and might try to acquire the dquotlocks.
1855 * So, just unlock here and do a vn_rele at the end.
1857 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1860 * Add the (disk blocks and inode) resources occupied by this
1861 * inode to its dquots. We do this adjustment in the incore dquot,
1862 * and also copy the changes to its buffer.
1863 * We don't care about putting these changes in a transaction
1864 * envelope because if we crash in the middle of a 'quotacheck'
1865 * we have to start from the beginning anyway.
1866 * Once we're done, we'll log all the dquot bufs.
1868 * The *QUOTA_ON checks below may look pretty racey, but quotachecks
1869 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1871 if (XFS_IS_UQUOTA_ON(mp)) {
1873 xfs_qm_quotacheck_dqadjust(udqp, nblks, rtblks);
1876 if (XFS_IS_OQUOTA_ON(mp)) {
1878 xfs_qm_quotacheck_dqadjust(gdqp, nblks, rtblks);
1882 * Now release the inode. This will send it to 'inactive', and
1883 * possibly even free blocks.
1885 VN_RELE(XFS_ITOV(ip));
1890 *res = BULKSTAT_RV_DIDONE;
1895 * Walk thru all the filesystem inodes and construct a consistent view
1896 * of the disk quota world. If the quotacheck fails, disable quotas.
1902 int done, count, error;
1905 xfs_inode_t *uip, *gip;
1913 ASSERT(XFS_QI_UQIP(mp) || XFS_QI_GQIP(mp));
1914 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1917 * There should be no cached dquots. The (simplistic) quotacheck
1918 * algorithm doesn't like that.
1920 ASSERT(XFS_QI_MPLNDQUOTS(mp) == 0);
1922 cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);
1925 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1926 * their counters to zero. We need a clean slate.
1927 * We don't log our changes till later.
1929 if ((uip = XFS_QI_UQIP(mp))) {
1930 if ((error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA)))
1932 flags |= XFS_UQUOTA_CHKD;
1935 if ((gip = XFS_QI_GQIP(mp))) {
1936 if ((error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1937 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA)))
1939 flags |= XFS_OQUOTA_CHKD;
1944 * Iterate thru all the inodes in the file system,
1945 * adjusting the corresponding dquot counters in core.
1947 if ((error = xfs_bulkstat(mp, &lastino, &count,
1948 xfs_qm_dqusage_adjust, NULL,
1950 BULKSTAT_FG_IGET|BULKSTAT_FG_VFSLOCKED,
1957 * We can get this error if we couldn't do a dquot allocation inside
1958 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1959 * dirty dquots that might be cached, we just want to get rid of them
1960 * and turn quotaoff. The dquots won't be attached to any of the inodes
1961 * at this point (because we intentionally didn't in dqget_noattach).
1964 xfs_qm_dqpurge_all(mp,
1965 XFS_QMOPT_UQUOTA|XFS_QMOPT_GQUOTA|
1966 XFS_QMOPT_PQUOTA|XFS_QMOPT_QUOTAOFF);
1970 * We've made all the changes that we need to make incore.
1971 * Now flush_them down to disk buffers.
1973 xfs_qm_dqflush_all(mp, XFS_QMOPT_DELWRI);
1976 * We didn't log anything, because if we crashed, we'll have to
1977 * start the quotacheck from scratch anyway. However, we must make
1978 * sure that our dquot changes are secure before we put the
1979 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1982 XFS_bflush(mp->m_ddev_targp);
1985 * If one type of quotas is off, then it will lose its
1986 * quotachecked status, since we won't be doing accounting for
1987 * that type anymore.
1989 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1990 mp->m_qflags |= flags;
1992 XQM_LIST_PRINT(&(XFS_QI_MPL_LIST(mp)), MPL_NEXT, "++++ Mp list +++");
1996 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1997 "Disabling quotas.",
1998 mp->m_fsname, error);
2000 * We must turn off quotas.
2002 ASSERT(mp->m_quotainfo != NULL);
2003 ASSERT(xfs_Gqm != NULL);
2004 xfs_qm_destroy_quotainfo(mp);
2005 xfs_mount_reset_sbqflags(mp);
2007 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
2013 * This is called after the superblock has been read in and we're ready to
2014 * iget the quota inodes.
2017 xfs_qm_init_quotainos(
2020 xfs_inode_t *uip, *gip;
2025 ASSERT(mp->m_quotainfo);
2031 * Get the uquota and gquota inodes
2033 if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb)) {
2034 if (XFS_IS_UQUOTA_ON(mp) &&
2035 mp->m_sb.sb_uquotino != NULLFSINO) {
2036 ASSERT(mp->m_sb.sb_uquotino > 0);
2037 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
2039 return XFS_ERROR(error);
2041 if (XFS_IS_OQUOTA_ON(mp) &&
2042 mp->m_sb.sb_gquotino != NULLFSINO) {
2043 ASSERT(mp->m_sb.sb_gquotino > 0);
2044 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
2047 VN_RELE(XFS_ITOV(uip));
2048 return XFS_ERROR(error);
2052 flags |= XFS_QMOPT_SBVERSION;
2053 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
2054 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
2058 * Create the two inodes, if they don't exist already. The changes
2059 * made above will get added to a transaction and logged in one of
2060 * the qino_alloc calls below. If the device is readonly,
2061 * temporarily switch to read-write to do this.
2063 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
2064 if ((error = xfs_qm_qino_alloc(mp, &uip,
2065 sbflags | XFS_SB_UQUOTINO,
2066 flags | XFS_QMOPT_UQUOTA)))
2067 return XFS_ERROR(error);
2069 flags &= ~XFS_QMOPT_SBVERSION;
2071 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
2072 flags |= (XFS_IS_GQUOTA_ON(mp) ?
2073 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
2074 error = xfs_qm_qino_alloc(mp, &gip,
2075 sbflags | XFS_SB_GQUOTINO, flags);
2078 VN_RELE(XFS_ITOV(uip));
2080 return XFS_ERROR(error);
2084 XFS_QI_UQIP(mp) = uip;
2085 XFS_QI_GQIP(mp) = gip;
2092 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2093 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2094 * favor the lookup function ...
2095 * XXXsup merge this with qm_reclaim_one().
2098 xfs_qm_shake_freelist(
2103 xfs_dquot_t *dqp, *nextdqp;
2115 cmn_err(CE_DEBUG, "Shake free 0x%x", howmany);
2117 /* lock order is : hashchainlock, freelistlock, mplistlock */
2119 xfs_qm_freelist_lock(xfs_Gqm);
2121 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
2122 ((dqp != (xfs_dquot_t *) &xfs_Gqm->qm_dqfreelist) &&
2123 nreclaimed < howmany); ) {
2127 * We are racing with dqlookup here. Naturally we don't
2128 * want to reclaim a dquot that lookup wants.
2130 if (dqp->dq_flags & XFS_DQ_WANT) {
2132 xfs_qm_freelist_unlock(xfs_Gqm);
2133 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2134 return (nreclaimed);
2135 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
2140 * If the dquot is inactive, we are assured that it is
2141 * not on the mplist or the hashlist, and that makes our
2144 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2145 ASSERT(dqp->q_mount == NULL);
2146 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2147 ASSERT(dqp->HL_PREVP == NULL);
2148 ASSERT(dqp->MPL_PREVP == NULL);
2149 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2150 nextdqp = dqp->dq_flnext;
2154 ASSERT(dqp->MPL_PREVP);
2156 * Try to grab the flush lock. If this dquot is in the process of
2157 * getting flushed to disk, we don't want to reclaim it.
2159 if (! xfs_qm_dqflock_nowait(dqp)) {
2161 dqp = dqp->dq_flnext;
2166 * We have the flush lock so we know that this is not in the
2167 * process of being flushed. So, if this is dirty, flush it
2168 * DELWRI so that we don't get a freelist infested with
2171 if (XFS_DQ_IS_DIRTY(dqp)) {
2172 xfs_dqtrace_entry(dqp, "DQSHAKE: DQDIRTY");
2174 * We flush it delayed write, so don't bother
2175 * releasing the mplock.
2177 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2178 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2179 dqp = dqp->dq_flnext;
2183 * We're trying to get the hashlock out of order. This races
2184 * with dqlookup; so, we giveup and goto the next dquot if
2185 * we couldn't get the hashlock. This way, we won't starve
2186 * a dqlookup process that holds the hashlock that is
2187 * waiting for the freelist lock.
2189 if (! xfs_qm_dqhashlock_nowait(dqp)) {
2192 dqp = dqp->dq_flnext;
2196 * This races with dquot allocation code as well as dqflush_all
2197 * and reclaim code. So, if we failed to grab the mplist lock,
2198 * giveup everything and start over.
2202 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2203 /* XXX put a sentinel so that we can come back here */
2206 XFS_DQ_HASH_UNLOCK(hash);
2207 xfs_qm_freelist_unlock(xfs_Gqm);
2208 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2209 return (nreclaimed);
2212 xfs_dqtrace_entry(dqp, "DQSHAKE: UNLINKING");
2214 cmn_err(CE_DEBUG, "Shake 0x%p, ID 0x%x\n",
2215 dqp, INT_GET(dqp->q_core.d_id, ARCH_CONVERT));
2217 ASSERT(dqp->q_nrefs == 0);
2218 nextdqp = dqp->dq_flnext;
2219 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2220 XQM_HASHLIST_REMOVE(hash, dqp);
2222 xfs_qm_mplist_unlock(dqp->q_mount);
2223 XFS_DQ_HASH_UNLOCK(hash);
2226 XQM_FREELIST_REMOVE(dqp);
2229 XQM_STATS_INC(xqmstats.xs_qm_dqshake_reclaims);
2230 xfs_qm_dqdestroy(dqp);
2233 xfs_qm_freelist_unlock(xfs_Gqm);
2234 return (nreclaimed);
2239 * The kmem_shake interface is invoked when memory is running low.
2243 xfs_qm_shake(int nr_to_scan, unsigned int gfp_mask)
2245 int ndqused, nfree, n;
2247 if (!kmem_shake_allow(gfp_mask))
2252 nfree = xfs_Gqm->qm_dqfreelist.qh_nelems; /* free dquots */
2253 /* incore dquots in all f/s's */
2254 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2256 ASSERT(ndqused >= 0);
2258 if (nfree <= ndqused && nfree < ndquot)
2261 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2262 n = nfree - ndqused - ndquot; /* # over target */
2264 return xfs_qm_shake_freelist(MAX(nfree, n));
2269 * Just pop the least recently used dquot off the freelist and
2270 * recycle it. The returned dquot is locked.
2272 STATIC xfs_dquot_t *
2273 xfs_qm_dqreclaim_one(void)
2275 xfs_dquot_t *dqpout;
2284 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2286 xfs_qm_freelist_lock(xfs_Gqm);
2288 FOREACH_DQUOT_IN_FREELIST(dqp, &(xfs_Gqm->qm_dqfreelist)) {
2292 * We are racing with dqlookup here. Naturally we don't
2293 * want to reclaim a dquot that lookup wants. We release the
2294 * freelist lock and start over, so that lookup will grab
2295 * both the dquot and the freelistlock.
2297 if (dqp->dq_flags & XFS_DQ_WANT) {
2298 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
2299 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQWANT");
2301 xfs_qm_freelist_unlock(xfs_Gqm);
2302 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2304 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
2309 * If the dquot is inactive, we are assured that it is
2310 * not on the mplist or the hashlist, and that makes our
2313 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2314 ASSERT(dqp->q_mount == NULL);
2315 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2316 ASSERT(dqp->HL_PREVP == NULL);
2317 ASSERT(dqp->MPL_PREVP == NULL);
2318 XQM_FREELIST_REMOVE(dqp);
2321 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2325 ASSERT(dqp->q_hash);
2326 ASSERT(dqp->MPL_PREVP);
2329 * Try to grab the flush lock. If this dquot is in the process of
2330 * getting flushed to disk, we don't want to reclaim it.
2332 if (! xfs_qm_dqflock_nowait(dqp)) {
2338 * We have the flush lock so we know that this is not in the
2339 * process of being flushed. So, if this is dirty, flush it
2340 * DELWRI so that we don't get a freelist infested with
2343 if (XFS_DQ_IS_DIRTY(dqp)) {
2344 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQDIRTY");
2346 * We flush it delayed write, so don't bother
2347 * releasing the freelist lock.
2349 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2350 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2354 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2360 if (! xfs_qm_dqhashlock_nowait(dqp))
2363 ASSERT(dqp->q_nrefs == 0);
2364 xfs_dqtrace_entry(dqp, "DQRECLAIM: UNLINKING");
2365 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2366 XQM_HASHLIST_REMOVE(dqp->q_hash, dqp);
2367 XQM_FREELIST_REMOVE(dqp);
2369 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
2371 xfs_qm_mplist_unlock(dqp->q_mount);
2378 xfs_qm_freelist_unlock(xfs_Gqm);
2383 /*------------------------------------------------------------------*/
2386 * Return a new incore dquot. Depending on the number of
2387 * dquots in the system, we either allocate a new one on the kernel heap,
2388 * or reclaim a free one.
2389 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2390 * to reclaim an existing one from the freelist.
2393 xfs_qm_dqalloc_incore(
2394 xfs_dquot_t **O_dqpp)
2399 * Check against high water mark to see if we want to pop
2400 * a nincompoop dquot off the freelist.
2402 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2404 * Try to recycle a dquot from the freelist.
2406 if ((dqp = xfs_qm_dqreclaim_one())) {
2407 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2409 * Just zero the core here. The rest will get
2410 * reinitialized by caller. XXX we shouldn't even
2413 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2417 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2421 * Allocate a brand new dquot on the kernel heap and return it
2422 * to the caller to initialize.
2424 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2425 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2426 atomic_inc(&xfs_Gqm->qm_totaldquots);
2433 * Start a transaction and write the incore superblock changes to
2434 * disk. flags parameter indicates which fields have changed.
2437 xfs_qm_write_sb_changes(
2445 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2447 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2448 if ((error = xfs_trans_reserve(tp, 0,
2449 mp->m_sb.sb_sectsize + 128, 0,
2451 XFS_DEFAULT_LOG_COUNT))) {
2452 xfs_trans_cancel(tp, 0);
2456 xfs_mod_sb(tp, flags);
2457 (void) xfs_trans_commit(tp, 0, NULL);
2463 /* --------------- utility functions for vnodeops ---------------- */
2467 * Given an inode, a uid and gid (from cred_t) make sure that we have
2468 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2469 * quotas by creating this file.
2470 * This also attaches dquot(s) to the given inode after locking it,
2471 * and returns the dquots corresponding to the uid and/or gid.
2473 * in : inode (unlocked)
2474 * out : udquot, gdquot with references taken and unlocked
2484 xfs_dquot_t **O_udqpp,
2485 xfs_dquot_t **O_gdqpp)
2488 xfs_dquot_t *uq, *gq;
2491 if (!XFS_IS_QUOTA_ON(mp))
2494 lockflags = XFS_ILOCK_EXCL;
2495 xfs_ilock(ip, lockflags);
2497 if ((flags & XFS_QMOPT_INHERIT) &&
2498 XFS_INHERIT_GID(ip, XFS_MTOVFS(mp)))
2499 gid = ip->i_d.di_gid;
2502 * Attach the dquot(s) to this inode, doing a dquot allocation
2503 * if necessary. The dquot(s) will not be locked.
2505 if (XFS_NOT_DQATTACHED(mp, ip)) {
2506 if ((error = xfs_qm_dqattach(ip, XFS_QMOPT_DQALLOC |
2507 XFS_QMOPT_ILOCKED))) {
2508 xfs_iunlock(ip, lockflags);
2514 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
2515 if (ip->i_d.di_uid != uid) {
2517 * What we need is the dquot that has this uid, and
2518 * if we send the inode to dqget, the uid of the inode
2519 * takes priority over what's sent in the uid argument.
2520 * We must unlock inode here before calling dqget if
2521 * we're not sending the inode, because otherwise
2522 * we'll deadlock by doing trans_reserve while
2525 xfs_iunlock(ip, lockflags);
2526 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2531 ASSERT(error != ENOENT);
2535 * Get the ilock in the right order.
2538 lockflags = XFS_ILOCK_SHARED;
2539 xfs_ilock(ip, lockflags);
2542 * Take an extra reference, because we'll return
2545 ASSERT(ip->i_udquot);
2552 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
2553 if (ip->i_d.di_gid != gid) {
2554 xfs_iunlock(ip, lockflags);
2555 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2562 ASSERT(error != ENOENT);
2566 lockflags = XFS_ILOCK_SHARED;
2567 xfs_ilock(ip, lockflags);
2569 ASSERT(ip->i_gdquot);
2575 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2576 if (ip->i_d.di_projid != prid) {
2577 xfs_iunlock(ip, lockflags);
2578 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2585 ASSERT(error != ENOENT);
2589 lockflags = XFS_ILOCK_SHARED;
2590 xfs_ilock(ip, lockflags);
2592 ASSERT(ip->i_gdquot);
2600 xfs_dqtrace_entry_ino(uq, "DQALLOC", ip);
2602 xfs_iunlock(ip, lockflags);
2615 * Actually transfer ownership, and do dquot modifications.
2616 * These were already reserved.
2622 xfs_dquot_t **IO_olddq,
2625 xfs_dquot_t *prevdq;
2626 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2627 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2629 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
2630 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2635 ASSERT(prevdq != newdq);
2637 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2638 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
2640 /* the sparkling new dquot */
2641 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2642 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
2645 * Take an extra reference, because the inode
2646 * is going to keep this dquot pointer even
2647 * after the trans_commit.
2651 xfs_dqunlock(newdq);
2658 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2661 xfs_qm_vop_chown_reserve(
2670 uint delblks, blkflags;
2671 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2673 ASSERT(XFS_ISLOCKED_INODE(ip));
2675 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2677 delblks = ip->i_delayed_blks;
2678 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2679 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2680 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2682 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2683 ip->i_d.di_uid != (uid_t)INT_GET(udqp->q_core.d_id, ARCH_CONVERT)) {
2686 * If there are delayed allocation blocks, then we have to
2687 * unreserve those from the old dquot, and add them to the
2691 ASSERT(ip->i_udquot);
2692 unresudq = ip->i_udquot;
2695 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2696 if ((XFS_IS_GQUOTA_ON(ip->i_mount) && ip->i_d.di_gid !=
2697 INT_GET(gdqp->q_core.d_id, ARCH_CONVERT)) ||
2698 (XFS_IS_PQUOTA_ON(ip->i_mount) && ip->i_d.di_projid !=
2699 INT_GET(gdqp->q_core.d_id, ARCH_CONVERT))) {
2702 ASSERT(ip->i_gdquot);
2703 unresgdq = ip->i_gdquot;
2708 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2709 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2714 * Do the delayed blks reservations/unreservations now. Since, these
2715 * are done without the help of a transaction, if a reservation fails
2716 * its previous reservations won't be automatically undone by trans
2717 * code. So, we have to do it manually here.
2721 * Do the reservations first. Unreservation can't fail.
2723 ASSERT(delblksudq || delblksgdq);
2724 ASSERT(unresudq || unresgdq);
2725 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2726 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2729 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2730 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2738 xfs_qm_vop_rename_dqattach(
2739 xfs_inode_t **i_tab)
2747 if (! XFS_IS_QUOTA_ON(ip->i_mount))
2750 if (XFS_NOT_DQATTACHED(ip->i_mount, ip)) {
2751 error = xfs_qm_dqattach(ip, 0);
2755 for (i = 1; (i < 4 && i_tab[i]); i++) {
2757 * Watch out for duplicate entries in the table.
2759 if ((ip = i_tab[i]) != i_tab[i-1]) {
2760 if (XFS_NOT_DQATTACHED(ip->i_mount, ip)) {
2761 error = xfs_qm_dqattach(ip, 0);
2771 xfs_qm_vop_dqattach_and_dqmod_newinode(
2777 if (!XFS_IS_QUOTA_ON(tp->t_mountp))
2780 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
2781 ASSERT(XFS_IS_QUOTA_RUNNING(tp->t_mountp));
2787 ASSERT(ip->i_udquot == NULL);
2788 ip->i_udquot = udqp;
2789 ASSERT(ip->i_d.di_uid == INT_GET(udqp->q_core.d_id, ARCH_CONVERT));
2790 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2796 ASSERT(ip->i_gdquot == NULL);
2797 ip->i_gdquot = gdqp;
2798 ASSERT(ip->i_d.di_gid == INT_GET(gdqp->q_core.d_id, ARCH_CONVERT));
2799 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2803 /* ------------- list stuff -----------------*/
2805 xfs_qm_freelist_init(xfs_frlist_t *ql)
2807 ql->qh_next = ql->qh_prev = (xfs_dquot_t *) ql;
2808 mutex_init(&ql->qh_lock, MUTEX_DEFAULT, "dqf");
2814 xfs_qm_freelist_destroy(xfs_frlist_t *ql)
2816 xfs_dquot_t *dqp, *nextdqp;
2818 mutex_lock(&ql->qh_lock, PINOD);
2819 for (dqp = ql->qh_next;
2820 dqp != (xfs_dquot_t *)ql; ) {
2822 nextdqp = dqp->dq_flnext;
2824 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
2826 XQM_FREELIST_REMOVE(dqp);
2828 xfs_qm_dqdestroy(dqp);
2832 * Don't bother about unlocking.
2834 mutex_destroy(&ql->qh_lock);
2836 ASSERT(ql->qh_nelems == 0);
2840 xfs_qm_freelist_insert(xfs_frlist_t *ql, xfs_dquot_t *dq)
2842 dq->dq_flnext = ql->qh_next;
2843 dq->dq_flprev = (xfs_dquot_t *)ql;
2845 dq->dq_flnext->dq_flprev = dq;
2846 xfs_Gqm->qm_dqfreelist.qh_nelems++;
2847 xfs_Gqm->qm_dqfreelist.qh_version++;
2851 xfs_qm_freelist_unlink(xfs_dquot_t *dq)
2853 xfs_dquot_t *next = dq->dq_flnext;
2854 xfs_dquot_t *prev = dq->dq_flprev;
2856 next->dq_flprev = prev;
2857 prev->dq_flnext = next;
2858 dq->dq_flnext = dq->dq_flprev = dq;
2859 xfs_Gqm->qm_dqfreelist.qh_nelems--;
2860 xfs_Gqm->qm_dqfreelist.qh_version++;
2864 xfs_qm_freelist_append(xfs_frlist_t *ql, xfs_dquot_t *dq)
2866 xfs_qm_freelist_insert((xfs_frlist_t *)ql->qh_prev, dq);
2870 xfs_qm_dqhashlock_nowait(
2875 locked = mutex_trylock(&((dqp)->q_hash->qh_lock));
2880 xfs_qm_freelist_lock_nowait(
2885 locked = mutex_trylock(&(xqm->qm_dqfreelist.qh_lock));
2890 xfs_qm_mplist_nowait(
2895 ASSERT(mp->m_quotainfo);
2896 locked = mutex_trylock(&(XFS_QI_MPLLOCK(mp)));