xfs: always use iget in bulkstat
[linux/fpc-iii.git] / fs / xfs / quota / xfs_qm.c
bloba211c21062618dd8a917c014dceed6f8d69e0248
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_bit.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_dir2.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"
38 #include "xfs_btree.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_itable.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_bmap.h"
44 #include "xfs_rw.h"
45 #include "xfs_attr.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_utils.h"
49 #include "xfs_qm.h"
52 * The global quota manager. There is only one of these for the entire
53 * system, _not_ one per file system. XQM keeps track of the overall
54 * quota functionality, including maintaining the freelist and hash
55 * tables of dquots.
57 struct mutex xfs_Gqm_lock;
58 struct xfs_qm *xfs_Gqm;
59 uint ndquot;
61 kmem_zone_t *qm_dqzone;
62 kmem_zone_t *qm_dqtrxzone;
64 static cred_t xfs_zerocr;
66 STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
67 STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
69 STATIC void xfs_qm_freelist_init(xfs_frlist_t *);
70 STATIC void xfs_qm_freelist_destroy(xfs_frlist_t *);
72 STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
73 STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
74 STATIC int xfs_qm_shake(int, gfp_t);
76 static struct shrinker xfs_qm_shaker = {
77 .shrink = xfs_qm_shake,
78 .seeks = DEFAULT_SEEKS,
81 #ifdef DEBUG
82 extern struct mutex qcheck_lock;
83 #endif
85 #ifdef QUOTADEBUG
86 #define XQM_LIST_PRINT(l, NXT, title) \
87 { \
88 xfs_dquot_t *dqp; int i = 0; \
89 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
90 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
91 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
92 "bcnt = %d, icnt = %d, refs = %d", \
93 ++i, (int) be32_to_cpu(dqp->q_core.d_id), \
94 DQFLAGTO_TYPESTR(dqp), \
95 (int) be64_to_cpu(dqp->q_core.d_bcount), \
96 (int) be64_to_cpu(dqp->q_core.d_icount), \
97 (int) dqp->q_nrefs); } \
99 #else
100 #define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
101 #endif
104 * Initialize the XQM structure.
105 * Note that there is not one quota manager per file system.
107 STATIC struct xfs_qm *
108 xfs_Gqm_init(void)
110 xfs_dqhash_t *udqhash, *gdqhash;
111 xfs_qm_t *xqm;
112 size_t hsize;
113 uint i;
116 * Initialize the dquot hash tables.
118 udqhash = kmem_zalloc_greedy(&hsize,
119 XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
120 XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t),
121 KM_SLEEP | KM_MAYFAIL | KM_LARGE);
122 gdqhash = kmem_zalloc(hsize, KM_SLEEP | KM_LARGE);
123 hsize /= sizeof(xfs_dqhash_t);
124 ndquot = hsize << 8;
126 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
127 xqm->qm_dqhashmask = hsize - 1;
128 xqm->qm_usr_dqhtable = udqhash;
129 xqm->qm_grp_dqhtable = gdqhash;
130 ASSERT(xqm->qm_usr_dqhtable != NULL);
131 ASSERT(xqm->qm_grp_dqhtable != NULL);
133 for (i = 0; i < hsize; i++) {
134 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
135 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
139 * Freelist of all dquots of all file systems
141 xfs_qm_freelist_init(&(xqm->qm_dqfreelist));
144 * dquot zone. we register our own low-memory callback.
146 if (!qm_dqzone) {
147 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
148 "xfs_dquots");
149 qm_dqzone = xqm->qm_dqzone;
150 } else
151 xqm->qm_dqzone = qm_dqzone;
153 register_shrinker(&xfs_qm_shaker);
156 * The t_dqinfo portion of transactions.
158 if (!qm_dqtrxzone) {
159 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
160 "xfs_dqtrx");
161 qm_dqtrxzone = xqm->qm_dqtrxzone;
162 } else
163 xqm->qm_dqtrxzone = qm_dqtrxzone;
165 atomic_set(&xqm->qm_totaldquots, 0);
166 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
167 xqm->qm_nrefs = 0;
168 #ifdef DEBUG
169 mutex_init(&qcheck_lock);
170 #endif
171 return xqm;
175 * Destroy the global quota manager when its reference count goes to zero.
177 STATIC void
178 xfs_qm_destroy(
179 struct xfs_qm *xqm)
181 int hsize, i;
183 ASSERT(xqm != NULL);
184 ASSERT(xqm->qm_nrefs == 0);
185 unregister_shrinker(&xfs_qm_shaker);
186 hsize = xqm->qm_dqhashmask + 1;
187 for (i = 0; i < hsize; i++) {
188 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
189 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
191 kmem_free(xqm->qm_usr_dqhtable);
192 kmem_free(xqm->qm_grp_dqhtable);
193 xqm->qm_usr_dqhtable = NULL;
194 xqm->qm_grp_dqhtable = NULL;
195 xqm->qm_dqhashmask = 0;
196 xfs_qm_freelist_destroy(&(xqm->qm_dqfreelist));
197 #ifdef DEBUG
198 mutex_destroy(&qcheck_lock);
199 #endif
200 kmem_free(xqm);
204 * Called at mount time to let XQM know that another file system is
205 * starting quotas. This isn't crucial information as the individual mount
206 * structures are pretty independent, but it helps the XQM keep a
207 * global view of what's going on.
209 /* ARGSUSED */
210 STATIC int
211 xfs_qm_hold_quotafs_ref(
212 struct xfs_mount *mp)
215 * Need to lock the xfs_Gqm structure for things like this. For example,
216 * the structure could disappear between the entry to this routine and
217 * a HOLD operation if not locked.
219 mutex_lock(&xfs_Gqm_lock);
221 if (xfs_Gqm == NULL)
222 xfs_Gqm = xfs_Gqm_init();
224 * We can keep a list of all filesystems with quotas mounted for
225 * debugging and statistical purposes, but ...
226 * Just take a reference and get out.
228 xfs_Gqm->qm_nrefs++;
229 mutex_unlock(&xfs_Gqm_lock);
231 return 0;
236 * Release the reference that a filesystem took at mount time,
237 * so that we know when we need to destroy the entire quota manager.
239 /* ARGSUSED */
240 STATIC void
241 xfs_qm_rele_quotafs_ref(
242 struct xfs_mount *mp)
244 xfs_dquot_t *dqp, *nextdqp;
246 ASSERT(xfs_Gqm);
247 ASSERT(xfs_Gqm->qm_nrefs > 0);
250 * Go thru the freelist and destroy all inactive dquots.
252 xfs_qm_freelist_lock(xfs_Gqm);
254 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
255 dqp != (xfs_dquot_t *)&(xfs_Gqm->qm_dqfreelist); ) {
256 xfs_dqlock(dqp);
257 nextdqp = dqp->dq_flnext;
258 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
259 ASSERT(dqp->q_mount == NULL);
260 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
261 ASSERT(dqp->HL_PREVP == NULL);
262 ASSERT(dqp->MPL_PREVP == NULL);
263 XQM_FREELIST_REMOVE(dqp);
264 xfs_dqunlock(dqp);
265 xfs_qm_dqdestroy(dqp);
266 } else {
267 xfs_dqunlock(dqp);
269 dqp = nextdqp;
271 xfs_qm_freelist_unlock(xfs_Gqm);
274 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
275 * be restarted.
277 mutex_lock(&xfs_Gqm_lock);
278 if (--xfs_Gqm->qm_nrefs == 0) {
279 xfs_qm_destroy(xfs_Gqm);
280 xfs_Gqm = NULL;
282 mutex_unlock(&xfs_Gqm_lock);
286 * Just destroy the quotainfo structure.
288 void
289 xfs_qm_unmount(
290 struct xfs_mount *mp)
292 if (mp->m_quotainfo) {
293 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
294 xfs_qm_destroy_quotainfo(mp);
300 * This is called from xfs_mountfs to start quotas and initialize all
301 * necessary data structures like quotainfo. This is also responsible for
302 * running a quotacheck as necessary. We are guaranteed that the superblock
303 * is consistently read in at this point.
305 * If we fail here, the mount will continue with quota turned off. We don't
306 * need to inidicate success or failure at all.
308 void
309 xfs_qm_mount_quotas(
310 xfs_mount_t *mp)
312 int error = 0;
313 uint sbf;
316 * If quotas on realtime volumes is not supported, we disable
317 * quotas immediately.
319 if (mp->m_sb.sb_rextents) {
320 cmn_err(CE_NOTE,
321 "Cannot turn on quotas for realtime filesystem %s",
322 mp->m_fsname);
323 mp->m_qflags = 0;
324 goto write_changes;
327 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
330 * Allocate the quotainfo structure inside the mount struct, and
331 * create quotainode(s), and change/rev superblock if necessary.
333 error = xfs_qm_init_quotainfo(mp);
334 if (error) {
336 * We must turn off quotas.
338 ASSERT(mp->m_quotainfo == NULL);
339 mp->m_qflags = 0;
340 goto write_changes;
343 * If any of the quotas are not consistent, do a quotacheck.
345 if (XFS_QM_NEED_QUOTACHECK(mp)) {
346 error = xfs_qm_quotacheck(mp);
347 if (error) {
348 /* Quotacheck failed and disabled quotas. */
349 return;
353 * If one type of quotas is off, then it will lose its
354 * quotachecked status, since we won't be doing accounting for
355 * that type anymore.
357 if (!XFS_IS_UQUOTA_ON(mp))
358 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
359 if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
360 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
362 write_changes:
364 * We actually don't have to acquire the m_sb_lock at all.
365 * This can only be called from mount, and that's single threaded. XXX
367 spin_lock(&mp->m_sb_lock);
368 sbf = mp->m_sb.sb_qflags;
369 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
370 spin_unlock(&mp->m_sb_lock);
372 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
373 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
375 * We could only have been turning quotas off.
376 * We aren't in very good shape actually because
377 * the incore structures are convinced that quotas are
378 * off, but the on disk superblock doesn't know that !
380 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
381 xfs_fs_cmn_err(CE_ALERT, mp,
382 "XFS mount_quotas: Superblock update failed!");
386 if (error) {
387 xfs_fs_cmn_err(CE_WARN, mp,
388 "Failed to initialize disk quotas.");
389 return;
392 #ifdef QUOTADEBUG
393 if (XFS_IS_QUOTA_ON(mp))
394 xfs_qm_internalqcheck(mp);
395 #endif
399 * Called from the vfsops layer.
401 void
402 xfs_qm_unmount_quotas(
403 xfs_mount_t *mp)
406 * Release the dquots that root inode, et al might be holding,
407 * before we flush quotas and blow away the quotainfo structure.
409 ASSERT(mp->m_rootip);
410 xfs_qm_dqdetach(mp->m_rootip);
411 if (mp->m_rbmip)
412 xfs_qm_dqdetach(mp->m_rbmip);
413 if (mp->m_rsumip)
414 xfs_qm_dqdetach(mp->m_rsumip);
417 * Release the quota inodes.
419 if (mp->m_quotainfo) {
420 if (mp->m_quotainfo->qi_uquotaip) {
421 IRELE(mp->m_quotainfo->qi_uquotaip);
422 mp->m_quotainfo->qi_uquotaip = NULL;
424 if (mp->m_quotainfo->qi_gquotaip) {
425 IRELE(mp->m_quotainfo->qi_gquotaip);
426 mp->m_quotainfo->qi_gquotaip = NULL;
432 * Flush all dquots of the given file system to disk. The dquots are
433 * _not_ purged from memory here, just their data written to disk.
435 STATIC int
436 xfs_qm_dqflush_all(
437 xfs_mount_t *mp,
438 int flags)
440 int recl;
441 xfs_dquot_t *dqp;
442 int niters;
443 int error;
445 if (mp->m_quotainfo == NULL)
446 return 0;
447 niters = 0;
448 again:
449 xfs_qm_mplist_lock(mp);
450 FOREACH_DQUOT_IN_MP(dqp, mp) {
451 xfs_dqlock(dqp);
452 if (! XFS_DQ_IS_DIRTY(dqp)) {
453 xfs_dqunlock(dqp);
454 continue;
456 xfs_dqtrace_entry(dqp, "FLUSHALL: DQDIRTY");
457 /* XXX a sentinel would be better */
458 recl = XFS_QI_MPLRECLAIMS(mp);
459 if (!xfs_dqflock_nowait(dqp)) {
461 * If we can't grab the flush lock then check
462 * to see if the dquot has been flushed delayed
463 * write. If so, grab its buffer and send it
464 * out immediately. We'll be able to acquire
465 * the flush lock when the I/O completes.
467 xfs_qm_dqflock_pushbuf_wait(dqp);
470 * Let go of the mplist lock. We don't want to hold it
471 * across a disk write.
473 xfs_qm_mplist_unlock(mp);
474 error = xfs_qm_dqflush(dqp, flags);
475 xfs_dqunlock(dqp);
476 if (error)
477 return error;
479 xfs_qm_mplist_lock(mp);
480 if (recl != XFS_QI_MPLRECLAIMS(mp)) {
481 xfs_qm_mplist_unlock(mp);
482 /* XXX restart limit */
483 goto again;
487 xfs_qm_mplist_unlock(mp);
488 /* return ! busy */
489 return 0;
492 * Release the group dquot pointers the user dquots may be
493 * carrying around as a hint. mplist is locked on entry and exit.
495 STATIC void
496 xfs_qm_detach_gdquots(
497 xfs_mount_t *mp)
499 xfs_dquot_t *dqp, *gdqp;
500 int nrecl;
502 again:
503 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
504 dqp = XFS_QI_MPLNEXT(mp);
505 while (dqp) {
506 xfs_dqlock(dqp);
507 if ((gdqp = dqp->q_gdquot)) {
508 xfs_dqlock(gdqp);
509 dqp->q_gdquot = NULL;
511 xfs_dqunlock(dqp);
513 if (gdqp) {
515 * Can't hold the mplist lock across a dqput.
516 * XXXmust convert to marker based iterations here.
518 nrecl = XFS_QI_MPLRECLAIMS(mp);
519 xfs_qm_mplist_unlock(mp);
520 xfs_qm_dqput(gdqp);
522 xfs_qm_mplist_lock(mp);
523 if (nrecl != XFS_QI_MPLRECLAIMS(mp))
524 goto again;
526 dqp = dqp->MPL_NEXT;
531 * Go through all the incore dquots of this file system and take them
532 * off the mplist and hashlist, if the dquot type matches the dqtype
533 * parameter. This is used when turning off quota accounting for
534 * users and/or groups, as well as when the filesystem is unmounting.
536 STATIC int
537 xfs_qm_dqpurge_int(
538 xfs_mount_t *mp,
539 uint flags) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
541 xfs_dquot_t *dqp;
542 uint dqtype;
543 int nrecl;
544 xfs_dquot_t *nextdqp;
545 int nmisses;
547 if (mp->m_quotainfo == NULL)
548 return 0;
550 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
551 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
552 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
554 xfs_qm_mplist_lock(mp);
557 * In the first pass through all incore dquots of this filesystem,
558 * we release the group dquot pointers the user dquots may be
559 * carrying around as a hint. We need to do this irrespective of
560 * what's being turned off.
562 xfs_qm_detach_gdquots(mp);
564 again:
565 nmisses = 0;
566 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
568 * Try to get rid of all of the unwanted dquots. The idea is to
569 * get them off mplist and hashlist, but leave them on freelist.
571 dqp = XFS_QI_MPLNEXT(mp);
572 while (dqp) {
574 * It's OK to look at the type without taking dqlock here.
575 * We're holding the mplist lock here, and that's needed for
576 * a dqreclaim.
578 if ((dqp->dq_flags & dqtype) == 0) {
579 dqp = dqp->MPL_NEXT;
580 continue;
583 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
584 nrecl = XFS_QI_MPLRECLAIMS(mp);
585 xfs_qm_mplist_unlock(mp);
586 mutex_lock(&dqp->q_hash->qh_lock);
587 xfs_qm_mplist_lock(mp);
590 * XXXTheoretically, we can get into a very long
591 * ping pong game here.
592 * No one can be adding dquots to the mplist at
593 * this point, but somebody might be taking things off.
595 if (nrecl != XFS_QI_MPLRECLAIMS(mp)) {
596 mutex_unlock(&dqp->q_hash->qh_lock);
597 goto again;
602 * Take the dquot off the mplist and hashlist. It may remain on
603 * freelist in INACTIVE state.
605 nextdqp = dqp->MPL_NEXT;
606 nmisses += xfs_qm_dqpurge(dqp);
607 dqp = nextdqp;
609 xfs_qm_mplist_unlock(mp);
610 return nmisses;
614 xfs_qm_dqpurge_all(
615 xfs_mount_t *mp,
616 uint flags)
618 int ndquots;
621 * Purge the dquot cache.
622 * None of the dquots should really be busy at this point.
624 if (mp->m_quotainfo) {
625 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
626 delay(ndquots * 10);
629 return 0;
632 STATIC int
633 xfs_qm_dqattach_one(
634 xfs_inode_t *ip,
635 xfs_dqid_t id,
636 uint type,
637 uint doalloc,
638 xfs_dquot_t *udqhint, /* hint */
639 xfs_dquot_t **IO_idqpp)
641 xfs_dquot_t *dqp;
642 int error;
644 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
645 error = 0;
648 * See if we already have it in the inode itself. IO_idqpp is
649 * &i_udquot or &i_gdquot. This made the code look weird, but
650 * made the logic a lot simpler.
652 dqp = *IO_idqpp;
653 if (dqp) {
654 xfs_dqtrace_entry(dqp, "DQATTACH: found in ip");
655 return 0;
659 * udqhint is the i_udquot field in inode, and is non-NULL only
660 * when the type arg is group/project. Its purpose is to save a
661 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
662 * the user dquot.
664 if (udqhint) {
665 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
666 xfs_dqlock(udqhint);
669 * No need to take dqlock to look at the id.
671 * The ID can't change until it gets reclaimed, and it won't
672 * be reclaimed as long as we have a ref from inode and we
673 * hold the ilock.
675 dqp = udqhint->q_gdquot;
676 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
677 xfs_dqlock(dqp);
678 XFS_DQHOLD(dqp);
679 ASSERT(*IO_idqpp == NULL);
680 *IO_idqpp = dqp;
682 xfs_dqunlock(dqp);
683 xfs_dqunlock(udqhint);
684 return 0;
688 * We can't hold a dquot lock when we call the dqget code.
689 * We'll deadlock in no time, because of (not conforming to)
690 * lock ordering - the inodelock comes before any dquot lock,
691 * and we may drop and reacquire the ilock in xfs_qm_dqget().
693 xfs_dqunlock(udqhint);
697 * Find the dquot from somewhere. This bumps the
698 * reference count of dquot and returns it locked.
699 * This can return ENOENT if dquot didn't exist on
700 * disk and we didn't ask it to allocate;
701 * ESRCH if quotas got turned off suddenly.
703 error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
704 if (error)
705 return error;
707 xfs_dqtrace_entry(dqp, "DQATTACH: found by dqget");
710 * dqget may have dropped and re-acquired the ilock, but it guarantees
711 * that the dquot returned is the one that should go in the inode.
713 *IO_idqpp = dqp;
714 xfs_dqunlock(dqp);
715 return 0;
720 * Given a udquot and gdquot, attach a ptr to the group dquot in the
721 * udquot as a hint for future lookups. The idea sounds simple, but the
722 * execution isn't, because the udquot might have a group dquot attached
723 * already and getting rid of that gets us into lock ordering constraints.
724 * The process is complicated more by the fact that the dquots may or may not
725 * be locked on entry.
727 STATIC void
728 xfs_qm_dqattach_grouphint(
729 xfs_dquot_t *udq,
730 xfs_dquot_t *gdq)
732 xfs_dquot_t *tmp;
734 xfs_dqlock(udq);
736 if ((tmp = udq->q_gdquot)) {
737 if (tmp == gdq) {
738 xfs_dqunlock(udq);
739 return;
742 udq->q_gdquot = NULL;
744 * We can't keep any dqlocks when calling dqrele,
745 * because the freelist lock comes before dqlocks.
747 xfs_dqunlock(udq);
749 * we took a hard reference once upon a time in dqget,
750 * so give it back when the udquot no longer points at it
751 * dqput() does the unlocking of the dquot.
753 xfs_qm_dqrele(tmp);
755 xfs_dqlock(udq);
756 xfs_dqlock(gdq);
758 } else {
759 ASSERT(XFS_DQ_IS_LOCKED(udq));
760 xfs_dqlock(gdq);
763 ASSERT(XFS_DQ_IS_LOCKED(udq));
764 ASSERT(XFS_DQ_IS_LOCKED(gdq));
766 * Somebody could have attached a gdquot here,
767 * when we dropped the uqlock. If so, just do nothing.
769 if (udq->q_gdquot == NULL) {
770 XFS_DQHOLD(gdq);
771 udq->q_gdquot = gdq;
774 xfs_dqunlock(gdq);
775 xfs_dqunlock(udq);
780 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
781 * into account.
782 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
783 * Inode may get unlocked and relocked in here, and the caller must deal with
784 * the consequences.
787 xfs_qm_dqattach_locked(
788 xfs_inode_t *ip,
789 uint flags)
791 xfs_mount_t *mp = ip->i_mount;
792 uint nquotas = 0;
793 int error = 0;
795 if (!XFS_IS_QUOTA_RUNNING(mp) ||
796 !XFS_IS_QUOTA_ON(mp) ||
797 !XFS_NOT_DQATTACHED(mp, ip) ||
798 ip->i_ino == mp->m_sb.sb_uquotino ||
799 ip->i_ino == mp->m_sb.sb_gquotino)
800 return 0;
802 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
804 if (XFS_IS_UQUOTA_ON(mp)) {
805 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
806 flags & XFS_QMOPT_DQALLOC,
807 NULL, &ip->i_udquot);
808 if (error)
809 goto done;
810 nquotas++;
813 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
814 if (XFS_IS_OQUOTA_ON(mp)) {
815 error = XFS_IS_GQUOTA_ON(mp) ?
816 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
817 flags & XFS_QMOPT_DQALLOC,
818 ip->i_udquot, &ip->i_gdquot) :
819 xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
820 flags & XFS_QMOPT_DQALLOC,
821 ip->i_udquot, &ip->i_gdquot);
823 * Don't worry about the udquot that we may have
824 * attached above. It'll get detached, if not already.
826 if (error)
827 goto done;
828 nquotas++;
832 * Attach this group quota to the user quota as a hint.
833 * This WON'T, in general, result in a thrash.
835 if (nquotas == 2) {
836 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
837 ASSERT(ip->i_udquot);
838 ASSERT(ip->i_gdquot);
841 * We may or may not have the i_udquot locked at this point,
842 * but this check is OK since we don't depend on the i_gdquot to
843 * be accurate 100% all the time. It is just a hint, and this
844 * will succeed in general.
846 if (ip->i_udquot->q_gdquot == ip->i_gdquot)
847 goto done;
849 * Attach i_gdquot to the gdquot hint inside the i_udquot.
851 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
854 done:
855 #ifdef QUOTADEBUG
856 if (! error) {
857 if (XFS_IS_UQUOTA_ON(mp))
858 ASSERT(ip->i_udquot);
859 if (XFS_IS_OQUOTA_ON(mp))
860 ASSERT(ip->i_gdquot);
862 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
863 #endif
864 return error;
868 xfs_qm_dqattach(
869 struct xfs_inode *ip,
870 uint flags)
872 int error;
874 xfs_ilock(ip, XFS_ILOCK_EXCL);
875 error = xfs_qm_dqattach_locked(ip, flags);
876 xfs_iunlock(ip, XFS_ILOCK_EXCL);
878 return error;
882 * Release dquots (and their references) if any.
883 * The inode should be locked EXCL except when this's called by
884 * xfs_ireclaim.
886 void
887 xfs_qm_dqdetach(
888 xfs_inode_t *ip)
890 if (!(ip->i_udquot || ip->i_gdquot))
891 return;
893 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
894 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
895 if (ip->i_udquot) {
896 xfs_dqtrace_entry_ino(ip->i_udquot, "DQDETTACH", ip);
897 xfs_qm_dqrele(ip->i_udquot);
898 ip->i_udquot = NULL;
900 if (ip->i_gdquot) {
901 xfs_dqtrace_entry_ino(ip->i_gdquot, "DQDETTACH", ip);
902 xfs_qm_dqrele(ip->i_gdquot);
903 ip->i_gdquot = NULL;
908 xfs_qm_sync(
909 xfs_mount_t *mp,
910 int flags)
912 int recl, restarts;
913 xfs_dquot_t *dqp;
914 uint flush_flags;
915 int error;
917 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
918 return 0;
920 flush_flags = (flags & SYNC_WAIT) ? XFS_QMOPT_SYNC : XFS_QMOPT_DELWRI;
921 restarts = 0;
923 again:
924 xfs_qm_mplist_lock(mp);
926 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
927 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
928 * when we have the mplist lock, we know that dquots will be consistent
929 * as long as we have it locked.
931 if (! XFS_IS_QUOTA_ON(mp)) {
932 xfs_qm_mplist_unlock(mp);
933 return 0;
935 FOREACH_DQUOT_IN_MP(dqp, mp) {
937 * If this is vfs_sync calling, then skip the dquots that
938 * don't 'seem' to be dirty. ie. don't acquire dqlock.
939 * This is very similar to what xfs_sync does with inodes.
941 if (flags & SYNC_TRYLOCK) {
942 if (!XFS_DQ_IS_DIRTY(dqp))
943 continue;
944 if (!xfs_qm_dqlock_nowait(dqp))
945 continue;
946 } else {
947 xfs_dqlock(dqp);
951 * Now, find out for sure if this dquot is dirty or not.
953 if (! XFS_DQ_IS_DIRTY(dqp)) {
954 xfs_dqunlock(dqp);
955 continue;
958 /* XXX a sentinel would be better */
959 recl = XFS_QI_MPLRECLAIMS(mp);
960 if (!xfs_dqflock_nowait(dqp)) {
961 if (flags & SYNC_TRYLOCK) {
962 xfs_dqunlock(dqp);
963 continue;
966 * If we can't grab the flush lock then if the caller
967 * really wanted us to give this our best shot, so
968 * see if we can give a push to the buffer before we wait
969 * on the flush lock. At this point, we know that
970 * even though the dquot is being flushed,
971 * it has (new) dirty data.
973 xfs_qm_dqflock_pushbuf_wait(dqp);
976 * Let go of the mplist lock. We don't want to hold it
977 * across a disk write
979 xfs_qm_mplist_unlock(mp);
980 xfs_dqtrace_entry(dqp, "XQM_SYNC: DQFLUSH");
981 error = xfs_qm_dqflush(dqp, flush_flags);
982 xfs_dqunlock(dqp);
983 if (error && XFS_FORCED_SHUTDOWN(mp))
984 return 0; /* Need to prevent umount failure */
985 else if (error)
986 return error;
988 xfs_qm_mplist_lock(mp);
989 if (recl != XFS_QI_MPLRECLAIMS(mp)) {
990 if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
991 break;
993 xfs_qm_mplist_unlock(mp);
994 goto again;
998 xfs_qm_mplist_unlock(mp);
999 return 0;
1003 * The hash chains and the mplist use the same xfs_dqhash structure as
1004 * their list head, but we can take the mplist qh_lock and one of the
1005 * hash qh_locks at the same time without any problem as they aren't
1006 * related.
1008 static struct lock_class_key xfs_quota_mplist_class;
1011 * This initializes all the quota information that's kept in the
1012 * mount structure
1014 STATIC int
1015 xfs_qm_init_quotainfo(
1016 xfs_mount_t *mp)
1018 xfs_quotainfo_t *qinf;
1019 int error;
1020 xfs_dquot_t *dqp;
1022 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1025 * Tell XQM that we exist as soon as possible.
1027 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1028 return error;
1031 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1034 * See if quotainodes are setup, and if not, allocate them,
1035 * and change the superblock accordingly.
1037 if ((error = xfs_qm_init_quotainos(mp))) {
1038 kmem_free(qinf);
1039 mp->m_quotainfo = NULL;
1040 return error;
1043 xfs_qm_list_init(&qinf->qi_dqlist, "mpdqlist", 0);
1044 lockdep_set_class(&qinf->qi_dqlist.qh_lock, &xfs_quota_mplist_class);
1046 qinf->qi_dqreclaims = 0;
1048 /* mutex used to serialize quotaoffs */
1049 mutex_init(&qinf->qi_quotaofflock);
1051 /* Precalc some constants */
1052 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1053 ASSERT(qinf->qi_dqchunklen);
1054 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1055 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1057 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1060 * We try to get the limits from the superuser's limits fields.
1061 * This is quite hacky, but it is standard quota practice.
1062 * We look at the USR dquot with id == 0 first, but if user quotas
1063 * are not enabled we goto the GRP dquot with id == 0.
1064 * We don't really care to keep separate default limits for user
1065 * and group quotas, at least not at this point.
1067 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1068 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
1069 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1070 XFS_DQ_PROJ),
1071 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1072 &dqp);
1073 if (! error) {
1074 xfs_disk_dquot_t *ddqp = &dqp->q_core;
1077 * The warnings and timers set the grace period given to
1078 * a user or group before he or she can not perform any
1079 * more writing. If it is zero, a default is used.
1081 qinf->qi_btimelimit = ddqp->d_btimer ?
1082 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
1083 qinf->qi_itimelimit = ddqp->d_itimer ?
1084 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
1085 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
1086 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
1087 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
1088 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
1089 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
1090 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
1091 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
1092 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
1093 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
1094 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
1095 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
1096 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
1097 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
1098 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1101 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1102 * we don't want this dquot cached. We haven't done a
1103 * quotacheck yet, and quotacheck doesn't like incore dquots.
1105 xfs_qm_dqdestroy(dqp);
1106 } else {
1107 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1108 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1109 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1110 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1111 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1112 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1115 return 0;
1120 * Gets called when unmounting a filesystem or when all quotas get
1121 * turned off.
1122 * This purges the quota inodes, destroys locks and frees itself.
1124 void
1125 xfs_qm_destroy_quotainfo(
1126 xfs_mount_t *mp)
1128 xfs_quotainfo_t *qi;
1130 qi = mp->m_quotainfo;
1131 ASSERT(qi != NULL);
1132 ASSERT(xfs_Gqm != NULL);
1135 * Release the reference that XQM kept, so that we know
1136 * when the XQM structure should be freed. We cannot assume
1137 * that xfs_Gqm is non-null after this point.
1139 xfs_qm_rele_quotafs_ref(mp);
1141 xfs_qm_list_destroy(&qi->qi_dqlist);
1143 if (qi->qi_uquotaip) {
1144 IRELE(qi->qi_uquotaip);
1145 qi->qi_uquotaip = NULL; /* paranoia */
1147 if (qi->qi_gquotaip) {
1148 IRELE(qi->qi_gquotaip);
1149 qi->qi_gquotaip = NULL;
1151 mutex_destroy(&qi->qi_quotaofflock);
1152 kmem_free(qi);
1153 mp->m_quotainfo = NULL;
1158 /* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1160 /* ARGSUSED */
1161 STATIC void
1162 xfs_qm_list_init(
1163 xfs_dqlist_t *list,
1164 char *str,
1165 int n)
1167 mutex_init(&list->qh_lock);
1168 list->qh_next = NULL;
1169 list->qh_version = 0;
1170 list->qh_nelems = 0;
1173 STATIC void
1174 xfs_qm_list_destroy(
1175 xfs_dqlist_t *list)
1177 mutex_destroy(&(list->qh_lock));
1182 * Stripped down version of dqattach. This doesn't attach, or even look at the
1183 * dquots attached to the inode. The rationale is that there won't be any
1184 * attached at the time this is called from quotacheck.
1186 STATIC int
1187 xfs_qm_dqget_noattach(
1188 xfs_inode_t *ip,
1189 xfs_dquot_t **O_udqpp,
1190 xfs_dquot_t **O_gdqpp)
1192 int error;
1193 xfs_mount_t *mp;
1194 xfs_dquot_t *udqp, *gdqp;
1196 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1197 mp = ip->i_mount;
1198 udqp = NULL;
1199 gdqp = NULL;
1201 if (XFS_IS_UQUOTA_ON(mp)) {
1202 ASSERT(ip->i_udquot == NULL);
1204 * We want the dquot allocated if it doesn't exist.
1206 if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
1207 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
1208 &udqp))) {
1210 * Shouldn't be able to turn off quotas here.
1212 ASSERT(error != ESRCH);
1213 ASSERT(error != ENOENT);
1214 return error;
1216 ASSERT(udqp);
1219 if (XFS_IS_OQUOTA_ON(mp)) {
1220 ASSERT(ip->i_gdquot == NULL);
1221 if (udqp)
1222 xfs_dqunlock(udqp);
1223 error = XFS_IS_GQUOTA_ON(mp) ?
1224 xfs_qm_dqget(mp, ip,
1225 ip->i_d.di_gid, XFS_DQ_GROUP,
1226 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1227 &gdqp) :
1228 xfs_qm_dqget(mp, ip,
1229 ip->i_d.di_projid, XFS_DQ_PROJ,
1230 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1231 &gdqp);
1232 if (error) {
1233 if (udqp)
1234 xfs_qm_dqrele(udqp);
1235 ASSERT(error != ESRCH);
1236 ASSERT(error != ENOENT);
1237 return error;
1239 ASSERT(gdqp);
1241 /* Reacquire the locks in the right order */
1242 if (udqp) {
1243 if (! xfs_qm_dqlock_nowait(udqp)) {
1244 xfs_dqunlock(gdqp);
1245 xfs_dqlock(udqp);
1246 xfs_dqlock(gdqp);
1251 *O_udqpp = udqp;
1252 *O_gdqpp = gdqp;
1254 #ifdef QUOTADEBUG
1255 if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
1256 if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
1257 #endif
1258 return 0;
1262 * Create an inode and return with a reference already taken, but unlocked
1263 * This is how we create quota inodes
1265 STATIC int
1266 xfs_qm_qino_alloc(
1267 xfs_mount_t *mp,
1268 xfs_inode_t **ip,
1269 __int64_t sbfields,
1270 uint flags)
1272 xfs_trans_t *tp;
1273 int error;
1274 int committed;
1276 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1277 if ((error = xfs_trans_reserve(tp,
1278 XFS_QM_QINOCREATE_SPACE_RES(mp),
1279 XFS_CREATE_LOG_RES(mp), 0,
1280 XFS_TRANS_PERM_LOG_RES,
1281 XFS_CREATE_LOG_COUNT))) {
1282 xfs_trans_cancel(tp, 0);
1283 return error;
1286 if ((error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0,
1287 &xfs_zerocr, 0, 1, ip, &committed))) {
1288 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1289 XFS_TRANS_ABORT);
1290 return error;
1294 * Keep an extra reference to this quota inode. This inode is
1295 * locked exclusively and joined to the transaction already.
1297 ASSERT(xfs_isilocked(*ip, XFS_ILOCK_EXCL));
1298 IHOLD(*ip);
1301 * Make the changes in the superblock, and log those too.
1302 * sbfields arg may contain fields other than *QUOTINO;
1303 * VERSIONNUM for example.
1305 spin_lock(&mp->m_sb_lock);
1306 if (flags & XFS_QMOPT_SBVERSION) {
1307 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1308 unsigned oldv = mp->m_sb.sb_versionnum;
1309 #endif
1310 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1311 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1312 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1313 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1314 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1316 xfs_sb_version_addquota(&mp->m_sb);
1317 mp->m_sb.sb_uquotino = NULLFSINO;
1318 mp->m_sb.sb_gquotino = NULLFSINO;
1320 /* qflags will get updated _after_ quotacheck */
1321 mp->m_sb.sb_qflags = 0;
1322 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1323 cmn_err(CE_NOTE,
1324 "Old superblock version %x, converting to %x.",
1325 oldv, mp->m_sb.sb_versionnum);
1326 #endif
1328 if (flags & XFS_QMOPT_UQUOTA)
1329 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1330 else
1331 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1332 spin_unlock(&mp->m_sb_lock);
1333 xfs_mod_sb(tp, sbfields);
1335 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1336 xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
1337 return error;
1339 return 0;
1343 STATIC void
1344 xfs_qm_reset_dqcounts(
1345 xfs_mount_t *mp,
1346 xfs_buf_t *bp,
1347 xfs_dqid_t id,
1348 uint type)
1350 xfs_disk_dquot_t *ddq;
1351 int j;
1353 xfs_buftrace("RESET DQUOTS", bp);
1355 * Reset all counters and timers. They'll be
1356 * started afresh by xfs_qm_quotacheck.
1358 #ifdef DEBUG
1359 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1360 do_div(j, sizeof(xfs_dqblk_t));
1361 ASSERT(XFS_QM_DQPERBLK(mp) == j);
1362 #endif
1363 ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1364 for (j = 0; j < XFS_QM_DQPERBLK(mp); j++) {
1366 * Do a sanity check, and if needed, repair the dqblk. Don't
1367 * output any warnings because it's perfectly possible to
1368 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1370 (void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1371 "xfs_quotacheck");
1372 ddq->d_bcount = 0;
1373 ddq->d_icount = 0;
1374 ddq->d_rtbcount = 0;
1375 ddq->d_btimer = 0;
1376 ddq->d_itimer = 0;
1377 ddq->d_rtbtimer = 0;
1378 ddq->d_bwarns = 0;
1379 ddq->d_iwarns = 0;
1380 ddq->d_rtbwarns = 0;
1381 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1385 STATIC int
1386 xfs_qm_dqiter_bufs(
1387 xfs_mount_t *mp,
1388 xfs_dqid_t firstid,
1389 xfs_fsblock_t bno,
1390 xfs_filblks_t blkcnt,
1391 uint flags)
1393 xfs_buf_t *bp;
1394 int error;
1395 int notcommitted;
1396 int incr;
1397 int type;
1399 ASSERT(blkcnt > 0);
1400 notcommitted = 0;
1401 incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
1402 XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
1403 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1404 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1405 error = 0;
1408 * Blkcnt arg can be a very big number, and might even be
1409 * larger than the log itself. So, we have to break it up into
1410 * manageable-sized transactions.
1411 * Note that we don't start a permanent transaction here; we might
1412 * not be able to get a log reservation for the whole thing up front,
1413 * and we don't really care to either, because we just discard
1414 * everything if we were to crash in the middle of this loop.
1416 while (blkcnt--) {
1417 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1418 XFS_FSB_TO_DADDR(mp, bno),
1419 (int)XFS_QI_DQCHUNKLEN(mp), 0, &bp);
1420 if (error)
1421 break;
1423 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1424 xfs_bdwrite(mp, bp);
1426 * goto the next block.
1428 bno++;
1429 firstid += XFS_QM_DQPERBLK(mp);
1431 return error;
1435 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1436 * caller supplied function for every chunk of dquots that we find.
1438 STATIC int
1439 xfs_qm_dqiterate(
1440 xfs_mount_t *mp,
1441 xfs_inode_t *qip,
1442 uint flags)
1444 xfs_bmbt_irec_t *map;
1445 int i, nmaps; /* number of map entries */
1446 int error; /* return value */
1447 xfs_fileoff_t lblkno;
1448 xfs_filblks_t maxlblkcnt;
1449 xfs_dqid_t firstid;
1450 xfs_fsblock_t rablkno;
1451 xfs_filblks_t rablkcnt;
1453 error = 0;
1455 * This looks racy, but we can't keep an inode lock across a
1456 * trans_reserve. But, this gets called during quotacheck, and that
1457 * happens only at mount time which is single threaded.
1459 if (qip->i_d.di_nblocks == 0)
1460 return 0;
1462 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1464 lblkno = 0;
1465 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1466 do {
1467 nmaps = XFS_DQITER_MAP_SIZE;
1469 * We aren't changing the inode itself. Just changing
1470 * some of its data. No new blocks are added here, and
1471 * the inode is never added to the transaction.
1473 xfs_ilock(qip, XFS_ILOCK_SHARED);
1474 error = xfs_bmapi(NULL, qip, lblkno,
1475 maxlblkcnt - lblkno,
1476 XFS_BMAPI_METADATA,
1477 NULL,
1478 0, map, &nmaps, NULL, NULL);
1479 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1480 if (error)
1481 break;
1483 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1484 for (i = 0; i < nmaps; i++) {
1485 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1486 ASSERT(map[i].br_blockcount);
1489 lblkno += map[i].br_blockcount;
1491 if (map[i].br_startblock == HOLESTARTBLOCK)
1492 continue;
1494 firstid = (xfs_dqid_t) map[i].br_startoff *
1495 XFS_QM_DQPERBLK(mp);
1497 * Do a read-ahead on the next extent.
1499 if ((i+1 < nmaps) &&
1500 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1501 rablkcnt = map[i+1].br_blockcount;
1502 rablkno = map[i+1].br_startblock;
1503 while (rablkcnt--) {
1504 xfs_baread(mp->m_ddev_targp,
1505 XFS_FSB_TO_DADDR(mp, rablkno),
1506 (int)XFS_QI_DQCHUNKLEN(mp));
1507 rablkno++;
1511 * Iterate thru all the blks in the extent and
1512 * reset the counters of all the dquots inside them.
1514 if ((error = xfs_qm_dqiter_bufs(mp,
1515 firstid,
1516 map[i].br_startblock,
1517 map[i].br_blockcount,
1518 flags))) {
1519 break;
1523 if (error)
1524 break;
1525 } while (nmaps > 0);
1527 kmem_free(map);
1529 return error;
1533 * Called by dqusage_adjust in doing a quotacheck.
1534 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1535 * this updates its incore copy as well as the buffer copy. This is
1536 * so that once the quotacheck is done, we can just log all the buffers,
1537 * as opposed to logging numerous updates to individual dquots.
1539 STATIC void
1540 xfs_qm_quotacheck_dqadjust(
1541 xfs_dquot_t *dqp,
1542 xfs_qcnt_t nblks,
1543 xfs_qcnt_t rtblks)
1545 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1546 xfs_dqtrace_entry(dqp, "QCHECK DQADJUST");
1548 * Adjust the inode count and the block count to reflect this inode's
1549 * resource usage.
1551 be64_add_cpu(&dqp->q_core.d_icount, 1);
1552 dqp->q_res_icount++;
1553 if (nblks) {
1554 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1555 dqp->q_res_bcount += nblks;
1557 if (rtblks) {
1558 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1559 dqp->q_res_rtbcount += rtblks;
1563 * Set default limits, adjust timers (since we changed usages)
1565 if (! XFS_IS_SUSER_DQUOT(dqp)) {
1566 xfs_qm_adjust_dqlimits(dqp->q_mount, &dqp->q_core);
1567 xfs_qm_adjust_dqtimers(dqp->q_mount, &dqp->q_core);
1570 dqp->dq_flags |= XFS_DQ_DIRTY;
1573 STATIC int
1574 xfs_qm_get_rtblks(
1575 xfs_inode_t *ip,
1576 xfs_qcnt_t *O_rtblks)
1578 xfs_filblks_t rtblks; /* total rt blks */
1579 xfs_extnum_t idx; /* extent record index */
1580 xfs_ifork_t *ifp; /* inode fork pointer */
1581 xfs_extnum_t nextents; /* number of extent entries */
1582 int error;
1584 ASSERT(XFS_IS_REALTIME_INODE(ip));
1585 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1586 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1587 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1588 return error;
1590 rtblks = 0;
1591 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1592 for (idx = 0; idx < nextents; idx++)
1593 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1594 *O_rtblks = (xfs_qcnt_t)rtblks;
1595 return 0;
1599 * callback routine supplied to bulkstat(). Given an inumber, find its
1600 * dquots and update them to account for resources taken by that inode.
1602 /* ARGSUSED */
1603 STATIC int
1604 xfs_qm_dqusage_adjust(
1605 xfs_mount_t *mp, /* mount point for filesystem */
1606 xfs_ino_t ino, /* inode number to get data for */
1607 void __user *buffer, /* not used */
1608 int ubsize, /* not used */
1609 xfs_daddr_t bno, /* starting block of inode cluster */
1610 int *ubused, /* not used */
1611 int *res) /* result code value */
1613 xfs_inode_t *ip;
1614 xfs_dquot_t *udqp, *gdqp;
1615 xfs_qcnt_t nblks, rtblks;
1616 int error;
1618 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1621 * rootino must have its resources accounted for, not so with the quota
1622 * inodes.
1624 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1625 *res = BULKSTAT_RV_NOTHING;
1626 return XFS_ERROR(EINVAL);
1630 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1631 * interface expects the inode to be exclusively locked because that's
1632 * the case in all other instances. It's OK that we do this because
1633 * quotacheck is done only at mount time.
1635 if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip, bno))) {
1636 *res = BULKSTAT_RV_NOTHING;
1637 return error;
1641 * Obtain the locked dquots. In case of an error (eg. allocation
1642 * fails for ENOSPC), we return the negative of the error number
1643 * to bulkstat, so that it can get propagated to quotacheck() and
1644 * making us disable quotas for the file system.
1646 if ((error = xfs_qm_dqget_noattach(ip, &udqp, &gdqp))) {
1647 xfs_iput(ip, XFS_ILOCK_EXCL);
1648 *res = BULKSTAT_RV_GIVEUP;
1649 return error;
1652 rtblks = 0;
1653 if (! XFS_IS_REALTIME_INODE(ip)) {
1654 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks;
1655 } else {
1657 * Walk thru the extent list and count the realtime blocks.
1659 if ((error = xfs_qm_get_rtblks(ip, &rtblks))) {
1660 xfs_iput(ip, XFS_ILOCK_EXCL);
1661 if (udqp)
1662 xfs_qm_dqput(udqp);
1663 if (gdqp)
1664 xfs_qm_dqput(gdqp);
1665 *res = BULKSTAT_RV_GIVEUP;
1666 return error;
1668 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1670 ASSERT(ip->i_delayed_blks == 0);
1673 * We can't release the inode while holding its dquot locks.
1674 * The inode can go into inactive and might try to acquire the dquotlocks.
1675 * So, just unlock here and do a vn_rele at the end.
1677 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1680 * Add the (disk blocks and inode) resources occupied by this
1681 * inode to its dquots. We do this adjustment in the incore dquot,
1682 * and also copy the changes to its buffer.
1683 * We don't care about putting these changes in a transaction
1684 * envelope because if we crash in the middle of a 'quotacheck'
1685 * we have to start from the beginning anyway.
1686 * Once we're done, we'll log all the dquot bufs.
1688 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1689 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1691 if (XFS_IS_UQUOTA_ON(mp)) {
1692 ASSERT(udqp);
1693 xfs_qm_quotacheck_dqadjust(udqp, nblks, rtblks);
1694 xfs_qm_dqput(udqp);
1696 if (XFS_IS_OQUOTA_ON(mp)) {
1697 ASSERT(gdqp);
1698 xfs_qm_quotacheck_dqadjust(gdqp, nblks, rtblks);
1699 xfs_qm_dqput(gdqp);
1702 * Now release the inode. This will send it to 'inactive', and
1703 * possibly even free blocks.
1705 IRELE(ip);
1708 * Goto next inode.
1710 *res = BULKSTAT_RV_DIDONE;
1711 return 0;
1715 * Walk thru all the filesystem inodes and construct a consistent view
1716 * of the disk quota world. If the quotacheck fails, disable quotas.
1719 xfs_qm_quotacheck(
1720 xfs_mount_t *mp)
1722 int done, count, error;
1723 xfs_ino_t lastino;
1724 size_t structsz;
1725 xfs_inode_t *uip, *gip;
1726 uint flags;
1728 count = INT_MAX;
1729 structsz = 1;
1730 lastino = 0;
1731 flags = 0;
1733 ASSERT(XFS_QI_UQIP(mp) || XFS_QI_GQIP(mp));
1734 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1737 * There should be no cached dquots. The (simplistic) quotacheck
1738 * algorithm doesn't like that.
1740 ASSERT(XFS_QI_MPLNDQUOTS(mp) == 0);
1742 cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);
1745 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1746 * their counters to zero. We need a clean slate.
1747 * We don't log our changes till later.
1749 if ((uip = XFS_QI_UQIP(mp))) {
1750 if ((error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA)))
1751 goto error_return;
1752 flags |= XFS_UQUOTA_CHKD;
1755 if ((gip = XFS_QI_GQIP(mp))) {
1756 if ((error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1757 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA)))
1758 goto error_return;
1759 flags |= XFS_OQUOTA_CHKD;
1762 do {
1764 * Iterate thru all the inodes in the file system,
1765 * adjusting the corresponding dquot counters in core.
1767 error = xfs_bulkstat(mp, &lastino, &count,
1768 xfs_qm_dqusage_adjust,
1769 structsz, NULL, &done);
1770 if (error)
1771 break;
1773 } while (!done);
1776 * We've made all the changes that we need to make incore.
1777 * Flush them down to disk buffers if everything was updated
1778 * successfully.
1780 if (!error)
1781 error = xfs_qm_dqflush_all(mp, XFS_QMOPT_DELWRI);
1784 * We can get this error if we couldn't do a dquot allocation inside
1785 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1786 * dirty dquots that might be cached, we just want to get rid of them
1787 * and turn quotaoff. The dquots won't be attached to any of the inodes
1788 * at this point (because we intentionally didn't in dqget_noattach).
1790 if (error) {
1791 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_QUOTAOFF);
1792 goto error_return;
1796 * We didn't log anything, because if we crashed, we'll have to
1797 * start the quotacheck from scratch anyway. However, we must make
1798 * sure that our dquot changes are secure before we put the
1799 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1800 * flush.
1802 XFS_bflush(mp->m_ddev_targp);
1805 * If one type of quotas is off, then it will lose its
1806 * quotachecked status, since we won't be doing accounting for
1807 * that type anymore.
1809 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1810 mp->m_qflags |= flags;
1812 XQM_LIST_PRINT(&(XFS_QI_MPL_LIST(mp)), MPL_NEXT, "++++ Mp list +++");
1814 error_return:
1815 if (error) {
1816 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1817 "Disabling quotas.",
1818 mp->m_fsname, error);
1820 * We must turn off quotas.
1822 ASSERT(mp->m_quotainfo != NULL);
1823 ASSERT(xfs_Gqm != NULL);
1824 xfs_qm_destroy_quotainfo(mp);
1825 if (xfs_mount_reset_sbqflags(mp)) {
1826 cmn_err(CE_WARN, "XFS quotacheck %s: "
1827 "Failed to reset quota flags.", mp->m_fsname);
1829 } else {
1830 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
1832 return (error);
1836 * This is called after the superblock has been read in and we're ready to
1837 * iget the quota inodes.
1839 STATIC int
1840 xfs_qm_init_quotainos(
1841 xfs_mount_t *mp)
1843 xfs_inode_t *uip, *gip;
1844 int error;
1845 __int64_t sbflags;
1846 uint flags;
1848 ASSERT(mp->m_quotainfo);
1849 uip = gip = NULL;
1850 sbflags = 0;
1851 flags = 0;
1854 * Get the uquota and gquota inodes
1856 if (xfs_sb_version_hasquota(&mp->m_sb)) {
1857 if (XFS_IS_UQUOTA_ON(mp) &&
1858 mp->m_sb.sb_uquotino != NULLFSINO) {
1859 ASSERT(mp->m_sb.sb_uquotino > 0);
1860 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1861 0, 0, &uip, 0)))
1862 return XFS_ERROR(error);
1864 if (XFS_IS_OQUOTA_ON(mp) &&
1865 mp->m_sb.sb_gquotino != NULLFSINO) {
1866 ASSERT(mp->m_sb.sb_gquotino > 0);
1867 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1868 0, 0, &gip, 0))) {
1869 if (uip)
1870 IRELE(uip);
1871 return XFS_ERROR(error);
1874 } else {
1875 flags |= XFS_QMOPT_SBVERSION;
1876 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1877 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1881 * Create the two inodes, if they don't exist already. The changes
1882 * made above will get added to a transaction and logged in one of
1883 * the qino_alloc calls below. If the device is readonly,
1884 * temporarily switch to read-write to do this.
1886 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1887 if ((error = xfs_qm_qino_alloc(mp, &uip,
1888 sbflags | XFS_SB_UQUOTINO,
1889 flags | XFS_QMOPT_UQUOTA)))
1890 return XFS_ERROR(error);
1892 flags &= ~XFS_QMOPT_SBVERSION;
1894 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1895 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1896 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1897 error = xfs_qm_qino_alloc(mp, &gip,
1898 sbflags | XFS_SB_GQUOTINO, flags);
1899 if (error) {
1900 if (uip)
1901 IRELE(uip);
1903 return XFS_ERROR(error);
1907 XFS_QI_UQIP(mp) = uip;
1908 XFS_QI_GQIP(mp) = gip;
1910 return 0;
1915 * Traverse the freelist of dquots and attempt to reclaim a maximum of
1916 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
1917 * favor the lookup function ...
1918 * XXXsup merge this with qm_reclaim_one().
1920 STATIC int
1921 xfs_qm_shake_freelist(
1922 int howmany)
1924 int nreclaimed;
1925 xfs_dqhash_t *hash;
1926 xfs_dquot_t *dqp, *nextdqp;
1927 int restarts;
1928 int nflushes;
1930 if (howmany <= 0)
1931 return 0;
1933 nreclaimed = 0;
1934 restarts = 0;
1935 nflushes = 0;
1937 #ifdef QUOTADEBUG
1938 cmn_err(CE_DEBUG, "Shake free 0x%x", howmany);
1939 #endif
1940 /* lock order is : hashchainlock, freelistlock, mplistlock */
1941 tryagain:
1942 xfs_qm_freelist_lock(xfs_Gqm);
1944 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
1945 ((dqp != (xfs_dquot_t *) &xfs_Gqm->qm_dqfreelist) &&
1946 nreclaimed < howmany); ) {
1947 xfs_dqlock(dqp);
1950 * We are racing with dqlookup here. Naturally we don't
1951 * want to reclaim a dquot that lookup wants.
1953 if (dqp->dq_flags & XFS_DQ_WANT) {
1954 xfs_dqunlock(dqp);
1955 xfs_qm_freelist_unlock(xfs_Gqm);
1956 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1957 return nreclaimed;
1958 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
1959 goto tryagain;
1963 * If the dquot is inactive, we are assured that it is
1964 * not on the mplist or the hashlist, and that makes our
1965 * life easier.
1967 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
1968 ASSERT(dqp->q_mount == NULL);
1969 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
1970 ASSERT(dqp->HL_PREVP == NULL);
1971 ASSERT(dqp->MPL_PREVP == NULL);
1972 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
1973 nextdqp = dqp->dq_flnext;
1974 goto off_freelist;
1977 ASSERT(dqp->MPL_PREVP);
1979 * Try to grab the flush lock. If this dquot is in the process of
1980 * getting flushed to disk, we don't want to reclaim it.
1982 if (!xfs_dqflock_nowait(dqp)) {
1983 xfs_dqunlock(dqp);
1984 dqp = dqp->dq_flnext;
1985 continue;
1989 * We have the flush lock so we know that this is not in the
1990 * process of being flushed. So, if this is dirty, flush it
1991 * DELWRI so that we don't get a freelist infested with
1992 * dirty dquots.
1994 if (XFS_DQ_IS_DIRTY(dqp)) {
1995 int error;
1996 xfs_dqtrace_entry(dqp, "DQSHAKE: DQDIRTY");
1998 * We flush it delayed write, so don't bother
1999 * releasing the mplock.
2001 error = xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2002 if (error) {
2003 xfs_fs_cmn_err(CE_WARN, dqp->q_mount,
2004 "xfs_qm_dqflush_all: dquot %p flush failed", dqp);
2006 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2007 dqp = dqp->dq_flnext;
2008 continue;
2011 * We're trying to get the hashlock out of order. This races
2012 * with dqlookup; so, we giveup and goto the next dquot if
2013 * we couldn't get the hashlock. This way, we won't starve
2014 * a dqlookup process that holds the hashlock that is
2015 * waiting for the freelist lock.
2017 if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
2018 xfs_dqfunlock(dqp);
2019 xfs_dqunlock(dqp);
2020 dqp = dqp->dq_flnext;
2021 continue;
2024 * This races with dquot allocation code as well as dqflush_all
2025 * and reclaim code. So, if we failed to grab the mplist lock,
2026 * giveup everything and start over.
2028 hash = dqp->q_hash;
2029 ASSERT(hash);
2030 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2031 /* XXX put a sentinel so that we can come back here */
2032 xfs_dqfunlock(dqp);
2033 xfs_dqunlock(dqp);
2034 mutex_unlock(&hash->qh_lock);
2035 xfs_qm_freelist_unlock(xfs_Gqm);
2036 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2037 return nreclaimed;
2038 goto tryagain;
2040 xfs_dqtrace_entry(dqp, "DQSHAKE: UNLINKING");
2041 #ifdef QUOTADEBUG
2042 cmn_err(CE_DEBUG, "Shake 0x%p, ID 0x%x\n",
2043 dqp, be32_to_cpu(dqp->q_core.d_id));
2044 #endif
2045 ASSERT(dqp->q_nrefs == 0);
2046 nextdqp = dqp->dq_flnext;
2047 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2048 XQM_HASHLIST_REMOVE(hash, dqp);
2049 xfs_dqfunlock(dqp);
2050 xfs_qm_mplist_unlock(dqp->q_mount);
2051 mutex_unlock(&hash->qh_lock);
2053 off_freelist:
2054 XQM_FREELIST_REMOVE(dqp);
2055 xfs_dqunlock(dqp);
2056 nreclaimed++;
2057 XQM_STATS_INC(xqmstats.xs_qm_dqshake_reclaims);
2058 xfs_qm_dqdestroy(dqp);
2059 dqp = nextdqp;
2061 xfs_qm_freelist_unlock(xfs_Gqm);
2062 return nreclaimed;
2067 * The kmem_shake interface is invoked when memory is running low.
2069 /* ARGSUSED */
2070 STATIC int
2071 xfs_qm_shake(int nr_to_scan, gfp_t gfp_mask)
2073 int ndqused, nfree, n;
2075 if (!kmem_shake_allow(gfp_mask))
2076 return 0;
2077 if (!xfs_Gqm)
2078 return 0;
2080 nfree = xfs_Gqm->qm_dqfreelist.qh_nelems; /* free dquots */
2081 /* incore dquots in all f/s's */
2082 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2084 ASSERT(ndqused >= 0);
2086 if (nfree <= ndqused && nfree < ndquot)
2087 return 0;
2089 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2090 n = nfree - ndqused - ndquot; /* # over target */
2092 return xfs_qm_shake_freelist(MAX(nfree, n));
2097 * Just pop the least recently used dquot off the freelist and
2098 * recycle it. The returned dquot is locked.
2100 STATIC xfs_dquot_t *
2101 xfs_qm_dqreclaim_one(void)
2103 xfs_dquot_t *dqpout;
2104 xfs_dquot_t *dqp;
2105 int restarts;
2106 int nflushes;
2108 restarts = 0;
2109 dqpout = NULL;
2110 nflushes = 0;
2112 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2113 startagain:
2114 xfs_qm_freelist_lock(xfs_Gqm);
2116 FOREACH_DQUOT_IN_FREELIST(dqp, &(xfs_Gqm->qm_dqfreelist)) {
2117 xfs_dqlock(dqp);
2120 * We are racing with dqlookup here. Naturally we don't
2121 * want to reclaim a dquot that lookup wants. We release the
2122 * freelist lock and start over, so that lookup will grab
2123 * both the dquot and the freelistlock.
2125 if (dqp->dq_flags & XFS_DQ_WANT) {
2126 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
2127 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQWANT");
2128 xfs_dqunlock(dqp);
2129 xfs_qm_freelist_unlock(xfs_Gqm);
2130 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
2131 return NULL;
2132 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
2133 goto startagain;
2137 * If the dquot is inactive, we are assured that it is
2138 * not on the mplist or the hashlist, and that makes our
2139 * life easier.
2141 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2142 ASSERT(dqp->q_mount == NULL);
2143 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2144 ASSERT(dqp->HL_PREVP == NULL);
2145 ASSERT(dqp->MPL_PREVP == NULL);
2146 XQM_FREELIST_REMOVE(dqp);
2147 xfs_dqunlock(dqp);
2148 dqpout = dqp;
2149 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2150 break;
2153 ASSERT(dqp->q_hash);
2154 ASSERT(dqp->MPL_PREVP);
2157 * Try to grab the flush lock. If this dquot is in the process of
2158 * getting flushed to disk, we don't want to reclaim it.
2160 if (!xfs_dqflock_nowait(dqp)) {
2161 xfs_dqunlock(dqp);
2162 continue;
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
2169 * dirty dquots.
2171 if (XFS_DQ_IS_DIRTY(dqp)) {
2172 int error;
2173 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQDIRTY");
2175 * We flush it delayed write, so don't bother
2176 * releasing the freelist lock.
2178 error = xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2179 if (error) {
2180 xfs_fs_cmn_err(CE_WARN, dqp->q_mount,
2181 "xfs_qm_dqreclaim: dquot %p flush failed", dqp);
2183 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2184 continue;
2187 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2188 xfs_dqfunlock(dqp);
2189 xfs_dqunlock(dqp);
2190 continue;
2193 if (!mutex_trylock(&dqp->q_hash->qh_lock))
2194 goto mplistunlock;
2196 ASSERT(dqp->q_nrefs == 0);
2197 xfs_dqtrace_entry(dqp, "DQRECLAIM: UNLINKING");
2198 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2199 XQM_HASHLIST_REMOVE(dqp->q_hash, dqp);
2200 XQM_FREELIST_REMOVE(dqp);
2201 dqpout = dqp;
2202 mutex_unlock(&dqp->q_hash->qh_lock);
2203 mplistunlock:
2204 xfs_qm_mplist_unlock(dqp->q_mount);
2205 xfs_dqfunlock(dqp);
2206 xfs_dqunlock(dqp);
2207 if (dqpout)
2208 break;
2211 xfs_qm_freelist_unlock(xfs_Gqm);
2212 return dqpout;
2216 /*------------------------------------------------------------------*/
2219 * Return a new incore dquot. Depending on the number of
2220 * dquots in the system, we either allocate a new one on the kernel heap,
2221 * or reclaim a free one.
2222 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2223 * to reclaim an existing one from the freelist.
2225 boolean_t
2226 xfs_qm_dqalloc_incore(
2227 xfs_dquot_t **O_dqpp)
2229 xfs_dquot_t *dqp;
2232 * Check against high water mark to see if we want to pop
2233 * a nincompoop dquot off the freelist.
2235 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2237 * Try to recycle a dquot from the freelist.
2239 if ((dqp = xfs_qm_dqreclaim_one())) {
2240 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2242 * Just zero the core here. The rest will get
2243 * reinitialized by caller. XXX we shouldn't even
2244 * do this zero ...
2246 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2247 *O_dqpp = dqp;
2248 return B_FALSE;
2250 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2254 * Allocate a brand new dquot on the kernel heap and return it
2255 * to the caller to initialize.
2257 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2258 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2259 atomic_inc(&xfs_Gqm->qm_totaldquots);
2261 return B_TRUE;
2266 * Start a transaction and write the incore superblock changes to
2267 * disk. flags parameter indicates which fields have changed.
2270 xfs_qm_write_sb_changes(
2271 xfs_mount_t *mp,
2272 __int64_t flags)
2274 xfs_trans_t *tp;
2275 int error;
2277 #ifdef QUOTADEBUG
2278 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2279 #endif
2280 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2281 if ((error = xfs_trans_reserve(tp, 0,
2282 mp->m_sb.sb_sectsize + 128, 0,
2284 XFS_DEFAULT_LOG_COUNT))) {
2285 xfs_trans_cancel(tp, 0);
2286 return error;
2289 xfs_mod_sb(tp, flags);
2290 error = xfs_trans_commit(tp, 0);
2292 return error;
2296 /* --------------- utility functions for vnodeops ---------------- */
2300 * Given an inode, a uid and gid (from cred_t) make sure that we have
2301 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2302 * quotas by creating this file.
2303 * This also attaches dquot(s) to the given inode after locking it,
2304 * and returns the dquots corresponding to the uid and/or gid.
2306 * in : inode (unlocked)
2307 * out : udquot, gdquot with references taken and unlocked
2310 xfs_qm_vop_dqalloc(
2311 struct xfs_inode *ip,
2312 uid_t uid,
2313 gid_t gid,
2314 prid_t prid,
2315 uint flags,
2316 struct xfs_dquot **O_udqpp,
2317 struct xfs_dquot **O_gdqpp)
2319 struct xfs_mount *mp = ip->i_mount;
2320 struct xfs_dquot *uq, *gq;
2321 int error;
2322 uint lockflags;
2324 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2325 return 0;
2327 lockflags = XFS_ILOCK_EXCL;
2328 xfs_ilock(ip, lockflags);
2330 if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
2331 gid = ip->i_d.di_gid;
2334 * Attach the dquot(s) to this inode, doing a dquot allocation
2335 * if necessary. The dquot(s) will not be locked.
2337 if (XFS_NOT_DQATTACHED(mp, ip)) {
2338 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
2339 if (error) {
2340 xfs_iunlock(ip, lockflags);
2341 return error;
2345 uq = gq = NULL;
2346 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
2347 if (ip->i_d.di_uid != uid) {
2349 * What we need is the dquot that has this uid, and
2350 * if we send the inode to dqget, the uid of the inode
2351 * takes priority over what's sent in the uid argument.
2352 * We must unlock inode here before calling dqget if
2353 * we're not sending the inode, because otherwise
2354 * we'll deadlock by doing trans_reserve while
2355 * holding ilock.
2357 xfs_iunlock(ip, lockflags);
2358 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2359 XFS_DQ_USER,
2360 XFS_QMOPT_DQALLOC |
2361 XFS_QMOPT_DOWARN,
2362 &uq))) {
2363 ASSERT(error != ENOENT);
2364 return error;
2367 * Get the ilock in the right order.
2369 xfs_dqunlock(uq);
2370 lockflags = XFS_ILOCK_SHARED;
2371 xfs_ilock(ip, lockflags);
2372 } else {
2374 * Take an extra reference, because we'll return
2375 * this to caller
2377 ASSERT(ip->i_udquot);
2378 uq = ip->i_udquot;
2379 xfs_dqlock(uq);
2380 XFS_DQHOLD(uq);
2381 xfs_dqunlock(uq);
2384 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
2385 if (ip->i_d.di_gid != gid) {
2386 xfs_iunlock(ip, lockflags);
2387 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2388 XFS_DQ_GROUP,
2389 XFS_QMOPT_DQALLOC |
2390 XFS_QMOPT_DOWARN,
2391 &gq))) {
2392 if (uq)
2393 xfs_qm_dqrele(uq);
2394 ASSERT(error != ENOENT);
2395 return error;
2397 xfs_dqunlock(gq);
2398 lockflags = XFS_ILOCK_SHARED;
2399 xfs_ilock(ip, lockflags);
2400 } else {
2401 ASSERT(ip->i_gdquot);
2402 gq = ip->i_gdquot;
2403 xfs_dqlock(gq);
2404 XFS_DQHOLD(gq);
2405 xfs_dqunlock(gq);
2407 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2408 if (ip->i_d.di_projid != prid) {
2409 xfs_iunlock(ip, lockflags);
2410 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2411 XFS_DQ_PROJ,
2412 XFS_QMOPT_DQALLOC |
2413 XFS_QMOPT_DOWARN,
2414 &gq))) {
2415 if (uq)
2416 xfs_qm_dqrele(uq);
2417 ASSERT(error != ENOENT);
2418 return (error);
2420 xfs_dqunlock(gq);
2421 lockflags = XFS_ILOCK_SHARED;
2422 xfs_ilock(ip, lockflags);
2423 } else {
2424 ASSERT(ip->i_gdquot);
2425 gq = ip->i_gdquot;
2426 xfs_dqlock(gq);
2427 XFS_DQHOLD(gq);
2428 xfs_dqunlock(gq);
2431 if (uq)
2432 xfs_dqtrace_entry_ino(uq, "DQALLOC", ip);
2434 xfs_iunlock(ip, lockflags);
2435 if (O_udqpp)
2436 *O_udqpp = uq;
2437 else if (uq)
2438 xfs_qm_dqrele(uq);
2439 if (O_gdqpp)
2440 *O_gdqpp = gq;
2441 else if (gq)
2442 xfs_qm_dqrele(gq);
2443 return 0;
2447 * Actually transfer ownership, and do dquot modifications.
2448 * These were already reserved.
2450 xfs_dquot_t *
2451 xfs_qm_vop_chown(
2452 xfs_trans_t *tp,
2453 xfs_inode_t *ip,
2454 xfs_dquot_t **IO_olddq,
2455 xfs_dquot_t *newdq)
2457 xfs_dquot_t *prevdq;
2458 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2459 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2462 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2463 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2465 /* old dquot */
2466 prevdq = *IO_olddq;
2467 ASSERT(prevdq);
2468 ASSERT(prevdq != newdq);
2470 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2471 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
2473 /* the sparkling new dquot */
2474 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2475 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
2478 * Take an extra reference, because the inode
2479 * is going to keep this dquot pointer even
2480 * after the trans_commit.
2482 xfs_dqlock(newdq);
2483 XFS_DQHOLD(newdq);
2484 xfs_dqunlock(newdq);
2485 *IO_olddq = newdq;
2487 return prevdq;
2491 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2494 xfs_qm_vop_chown_reserve(
2495 xfs_trans_t *tp,
2496 xfs_inode_t *ip,
2497 xfs_dquot_t *udqp,
2498 xfs_dquot_t *gdqp,
2499 uint flags)
2501 xfs_mount_t *mp = ip->i_mount;
2502 uint delblks, blkflags, prjflags = 0;
2503 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2504 int error;
2507 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2508 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2510 delblks = ip->i_delayed_blks;
2511 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2512 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2513 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2515 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2516 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
2517 delblksudq = udqp;
2519 * If there are delayed allocation blocks, then we have to
2520 * unreserve those from the old dquot, and add them to the
2521 * new dquot.
2523 if (delblks) {
2524 ASSERT(ip->i_udquot);
2525 unresudq = ip->i_udquot;
2528 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2529 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2530 ip->i_d.di_projid != be32_to_cpu(gdqp->q_core.d_id))
2531 prjflags = XFS_QMOPT_ENOSPC;
2533 if (prjflags ||
2534 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2535 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
2536 delblksgdq = gdqp;
2537 if (delblks) {
2538 ASSERT(ip->i_gdquot);
2539 unresgdq = ip->i_gdquot;
2544 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2545 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2546 flags | blkflags | prjflags)))
2547 return (error);
2550 * Do the delayed blks reservations/unreservations now. Since, these
2551 * are done without the help of a transaction, if a reservation fails
2552 * its previous reservations won't be automatically undone by trans
2553 * code. So, we have to do it manually here.
2555 if (delblks) {
2557 * Do the reservations first. Unreservation can't fail.
2559 ASSERT(delblksudq || delblksgdq);
2560 ASSERT(unresudq || unresgdq);
2561 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2562 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2563 flags | blkflags | prjflags)))
2564 return (error);
2565 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2566 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2567 blkflags);
2570 return (0);
2574 xfs_qm_vop_rename_dqattach(
2575 struct xfs_inode **i_tab)
2577 struct xfs_mount *mp = i_tab[0]->i_mount;
2578 int i;
2580 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2581 return 0;
2583 for (i = 0; (i < 4 && i_tab[i]); i++) {
2584 struct xfs_inode *ip = i_tab[i];
2585 int error;
2588 * Watch out for duplicate entries in the table.
2590 if (i == 0 || ip != i_tab[i-1]) {
2591 if (XFS_NOT_DQATTACHED(mp, ip)) {
2592 error = xfs_qm_dqattach(ip, 0);
2593 if (error)
2594 return error;
2598 return 0;
2601 void
2602 xfs_qm_vop_create_dqattach(
2603 struct xfs_trans *tp,
2604 struct xfs_inode *ip,
2605 struct xfs_dquot *udqp,
2606 struct xfs_dquot *gdqp)
2608 struct xfs_mount *mp = tp->t_mountp;
2610 if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2611 return;
2613 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2614 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2616 if (udqp) {
2617 xfs_dqlock(udqp);
2618 XFS_DQHOLD(udqp);
2619 xfs_dqunlock(udqp);
2620 ASSERT(ip->i_udquot == NULL);
2621 ip->i_udquot = udqp;
2622 ASSERT(XFS_IS_UQUOTA_ON(mp));
2623 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2624 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2626 if (gdqp) {
2627 xfs_dqlock(gdqp);
2628 XFS_DQHOLD(gdqp);
2629 xfs_dqunlock(gdqp);
2630 ASSERT(ip->i_gdquot == NULL);
2631 ip->i_gdquot = gdqp;
2632 ASSERT(XFS_IS_OQUOTA_ON(mp));
2633 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2634 ip->i_d.di_gid : ip->i_d.di_projid) ==
2635 be32_to_cpu(gdqp->q_core.d_id));
2636 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2640 /* ------------- list stuff -----------------*/
2641 STATIC void
2642 xfs_qm_freelist_init(xfs_frlist_t *ql)
2644 ql->qh_next = ql->qh_prev = (xfs_dquot_t *) ql;
2645 mutex_init(&ql->qh_lock);
2646 ql->qh_version = 0;
2647 ql->qh_nelems = 0;
2650 STATIC void
2651 xfs_qm_freelist_destroy(xfs_frlist_t *ql)
2653 xfs_dquot_t *dqp, *nextdqp;
2655 mutex_lock(&ql->qh_lock);
2656 for (dqp = ql->qh_next;
2657 dqp != (xfs_dquot_t *)ql; ) {
2658 xfs_dqlock(dqp);
2659 nextdqp = dqp->dq_flnext;
2660 #ifdef QUOTADEBUG
2661 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
2662 #endif
2663 XQM_FREELIST_REMOVE(dqp);
2664 xfs_dqunlock(dqp);
2665 xfs_qm_dqdestroy(dqp);
2666 dqp = nextdqp;
2668 mutex_unlock(&ql->qh_lock);
2669 mutex_destroy(&ql->qh_lock);
2671 ASSERT(ql->qh_nelems == 0);
2674 STATIC void
2675 xfs_qm_freelist_insert(xfs_frlist_t *ql, xfs_dquot_t *dq)
2677 dq->dq_flnext = ql->qh_next;
2678 dq->dq_flprev = (xfs_dquot_t *)ql;
2679 ql->qh_next = dq;
2680 dq->dq_flnext->dq_flprev = dq;
2681 xfs_Gqm->qm_dqfreelist.qh_nelems++;
2682 xfs_Gqm->qm_dqfreelist.qh_version++;
2685 void
2686 xfs_qm_freelist_unlink(xfs_dquot_t *dq)
2688 xfs_dquot_t *next = dq->dq_flnext;
2689 xfs_dquot_t *prev = dq->dq_flprev;
2691 next->dq_flprev = prev;
2692 prev->dq_flnext = next;
2693 dq->dq_flnext = dq->dq_flprev = dq;
2694 xfs_Gqm->qm_dqfreelist.qh_nelems--;
2695 xfs_Gqm->qm_dqfreelist.qh_version++;
2698 void
2699 xfs_qm_freelist_append(xfs_frlist_t *ql, xfs_dquot_t *dq)
2701 xfs_qm_freelist_insert((xfs_frlist_t *)ql->qh_prev, dq);