2 * Implementation of the diskquota system for the LINUX operating system. QUOTA
3 * is implemented using the BSD system call interface as the means of
4 * communication with the user level. This file contains the generic routines
5 * called by the different filesystems on allocation of an inode or block.
6 * These routines take care of the administration needed to have a consistent
7 * diskquota tracking system. The ideas of both user and group quotas are based
8 * on the Melbourne quota system as used on BSD derived systems. The internal
9 * implementation is based on one of the several variants of the LINUX
10 * inode-subsystem with added complexity of the diskquota system.
12 * Author: Marco van Wieringen <mvw@planets.elm.net>
14 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 * Revised list management to avoid races
17 * -- Bill Hawes, <whawes@star.net>, 9/98
19 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20 * As the consequence the locking was moved from dquot_decr_...(),
21 * dquot_incr_...() to calling functions.
22 * invalidate_dquots() now writes modified dquots.
23 * Serialized quota_off() and quota_on() for mount point.
24 * Fixed a few bugs in grow_dquots().
25 * Fixed deadlock in write_dquot() - we no longer account quotas on
27 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
28 * add_dquot_ref() restarts after blocking
29 * Added check for bogus uid and fixed check for group in quotactl.
30 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32 * Used struct list_head instead of own list struct
33 * Invalidation of referenced dquots is no longer possible
34 * Improved free_dquots list management
35 * Quota and i_blocks are now updated in one place to avoid races
36 * Warnings are now delayed so we won't block in critical section
37 * Write updated not to require dquot lock
38 * Jan Kara, <jack@suse.cz>, 9/2000
40 * Added dynamic quota structure allocation
41 * Jan Kara <jack@suse.cz> 12/2000
43 * Rewritten quota interface. Implemented new quota format and
44 * formats registering.
45 * Jan Kara, <jack@suse.cz>, 2001,2002
48 * Jan Kara, <jack@suse.cz>, 10/2002
50 * Added journalled quota support, fix lock inversion problems
51 * Jan Kara, <jack@suse.cz>, 2003,2004
53 * (C) Copyright 1994 - 1997 Marco van Wieringen
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
59 #include <linux/mount.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/cred.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/capability.h>
79 #include <linux/quotaops.h>
80 #include "../internal.h" /* ugh */
82 #include <linux/uaccess.h>
85 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
87 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
88 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
89 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
90 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
91 * modifications of quota state (on quotaon and quotaoff) and readers who care
92 * about latest values take it as well.
94 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
95 * dq_list_lock > dq_state_lock
97 * Note that some things (eg. sb pointer, type, id) doesn't change during
98 * the life of the dquot structure and so needn't to be protected by a lock
100 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
101 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
102 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
103 * inode and before dropping dquot references to avoid use of dquots after
104 * they are freed. dq_data_lock is used to serialize the pointer setting and
105 * clearing operations.
106 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
107 * inode is a quota file). Functions adding pointers from inode to dquots have
108 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
109 * have to do all pointer modifications before dropping dq_data_lock. This makes
110 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
111 * then drops all pointers to dquots from an inode.
113 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
114 * memory (or space for it is being allocated) on the first dqget(), when it is
115 * being written out, and when it is being released on the last dqput(). The
116 * allocation and release operations are serialized by the dq_lock and by
117 * checking the use count in dquot_release().
119 * Lock ordering (including related VFS locks) is the following:
120 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
123 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
124 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
125 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
126 EXPORT_SYMBOL(dq_data_lock);
127 DEFINE_STATIC_SRCU(dquot_srcu);
129 void __quota_error(struct super_block *sb, const char *func,
130 const char *fmt, ...)
132 if (printk_ratelimit()) {
134 struct va_format vaf;
141 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
142 sb->s_id, func, &vaf);
147 EXPORT_SYMBOL(__quota_error);
149 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
150 static char *quotatypes[] = INITQFNAMES;
152 static struct quota_format_type *quota_formats; /* List of registered formats */
153 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
155 /* SLAB cache for dquot structures */
156 static struct kmem_cache *dquot_cachep;
158 int register_quota_format(struct quota_format_type *fmt)
160 spin_lock(&dq_list_lock);
161 fmt->qf_next = quota_formats;
163 spin_unlock(&dq_list_lock);
166 EXPORT_SYMBOL(register_quota_format);
168 void unregister_quota_format(struct quota_format_type *fmt)
170 struct quota_format_type **actqf;
172 spin_lock(&dq_list_lock);
173 for (actqf = "a_formats; *actqf && *actqf != fmt;
174 actqf = &(*actqf)->qf_next)
177 *actqf = (*actqf)->qf_next;
178 spin_unlock(&dq_list_lock);
180 EXPORT_SYMBOL(unregister_quota_format);
182 static struct quota_format_type *find_quota_format(int id)
184 struct quota_format_type *actqf;
186 spin_lock(&dq_list_lock);
187 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
188 actqf = actqf->qf_next)
190 if (!actqf || !try_module_get(actqf->qf_owner)) {
193 spin_unlock(&dq_list_lock);
195 for (qm = 0; module_names[qm].qm_fmt_id &&
196 module_names[qm].qm_fmt_id != id; qm++)
198 if (!module_names[qm].qm_fmt_id ||
199 request_module(module_names[qm].qm_mod_name))
202 spin_lock(&dq_list_lock);
203 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
204 actqf = actqf->qf_next)
206 if (actqf && !try_module_get(actqf->qf_owner))
209 spin_unlock(&dq_list_lock);
213 static void put_quota_format(struct quota_format_type *fmt)
215 module_put(fmt->qf_owner);
219 * Dquot List Management:
220 * The quota code uses three lists for dquot management: the inuse_list,
221 * free_dquots, and dquot_hash[] array. A single dquot structure may be
222 * on all three lists, depending on its current state.
224 * All dquots are placed to the end of inuse_list when first created, and this
225 * list is used for invalidate operation, which must look at every dquot.
227 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
228 * and this list is searched whenever we need an available dquot. Dquots are
229 * removed from the list as soon as they are used again, and
230 * dqstats.free_dquots gives the number of dquots on the list. When
231 * dquot is invalidated it's completely released from memory.
233 * Dquots with a specific identity (device, type and id) are placed on
234 * one of the dquot_hash[] hash chains. The provides an efficient search
235 * mechanism to locate a specific dquot.
238 static LIST_HEAD(inuse_list);
239 static LIST_HEAD(free_dquots);
240 static unsigned int dq_hash_bits, dq_hash_mask;
241 static struct hlist_head *dquot_hash;
243 struct dqstats dqstats;
244 EXPORT_SYMBOL(dqstats);
246 static qsize_t inode_get_rsv_space(struct inode *inode);
247 static int __dquot_initialize(struct inode *inode, int type);
249 static inline unsigned int
250 hashfn(const struct super_block *sb, struct kqid qid)
252 unsigned int id = from_kqid(&init_user_ns, qid);
256 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
257 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
261 * Following list functions expect dq_list_lock to be held
263 static inline void insert_dquot_hash(struct dquot *dquot)
265 struct hlist_head *head;
266 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
267 hlist_add_head(&dquot->dq_hash, head);
270 static inline void remove_dquot_hash(struct dquot *dquot)
272 hlist_del_init(&dquot->dq_hash);
275 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
278 struct hlist_node *node;
281 hlist_for_each (node, dquot_hash+hashent) {
282 dquot = hlist_entry(node, struct dquot, dq_hash);
283 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
289 /* Add a dquot to the tail of the free list */
290 static inline void put_dquot_last(struct dquot *dquot)
292 list_add_tail(&dquot->dq_free, &free_dquots);
293 dqstats_inc(DQST_FREE_DQUOTS);
296 static inline void remove_free_dquot(struct dquot *dquot)
298 if (list_empty(&dquot->dq_free))
300 list_del_init(&dquot->dq_free);
301 dqstats_dec(DQST_FREE_DQUOTS);
304 static inline void put_inuse(struct dquot *dquot)
306 /* We add to the back of inuse list so we don't have to restart
307 * when traversing this list and we block */
308 list_add_tail(&dquot->dq_inuse, &inuse_list);
309 dqstats_inc(DQST_ALLOC_DQUOTS);
312 static inline void remove_inuse(struct dquot *dquot)
314 dqstats_dec(DQST_ALLOC_DQUOTS);
315 list_del(&dquot->dq_inuse);
318 * End of list functions needing dq_list_lock
321 static void wait_on_dquot(struct dquot *dquot)
323 mutex_lock(&dquot->dq_lock);
324 mutex_unlock(&dquot->dq_lock);
327 static inline int dquot_dirty(struct dquot *dquot)
329 return test_bit(DQ_MOD_B, &dquot->dq_flags);
332 static inline int mark_dquot_dirty(struct dquot *dquot)
334 return dquot->dq_sb->dq_op->mark_dirty(dquot);
337 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
338 int dquot_mark_dquot_dirty(struct dquot *dquot)
342 /* If quota is dirty already, we don't have to acquire dq_list_lock */
343 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
346 spin_lock(&dq_list_lock);
347 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
348 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
349 info[dquot->dq_id.type].dqi_dirty_list);
352 spin_unlock(&dq_list_lock);
355 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
357 /* Dirtify all the dquots - this can block when journalling */
358 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
363 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
365 /* Even in case of error we have to continue */
366 ret = mark_dquot_dirty(dquot[cnt]);
373 static inline void dqput_all(struct dquot **dquot)
377 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
381 /* This function needs dq_list_lock */
382 static inline int clear_dquot_dirty(struct dquot *dquot)
384 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
386 list_del_init(&dquot->dq_dirty);
390 void mark_info_dirty(struct super_block *sb, int type)
392 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
394 EXPORT_SYMBOL(mark_info_dirty);
397 * Read dquot from disk and alloc space for it
400 int dquot_acquire(struct dquot *dquot)
402 int ret = 0, ret2 = 0;
403 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
405 mutex_lock(&dquot->dq_lock);
406 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
407 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
410 /* Make sure flags update is visible after dquot has been filled */
411 smp_mb__before_atomic();
412 set_bit(DQ_READ_B, &dquot->dq_flags);
413 /* Instantiate dquot if needed */
414 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
415 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
416 /* Write the info if needed */
417 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
418 down_write(&dqopt->dqio_sem);
419 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
420 dquot->dq_sb, dquot->dq_id.type);
421 up_write(&dqopt->dqio_sem);
431 * Make sure flags update is visible after on-disk struct has been
432 * allocated. Paired with smp_rmb() in dqget().
434 smp_mb__before_atomic();
435 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
437 mutex_unlock(&dquot->dq_lock);
440 EXPORT_SYMBOL(dquot_acquire);
443 * Write dquot to disk
445 int dquot_commit(struct dquot *dquot)
448 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
450 mutex_lock(&dquot->dq_lock);
451 spin_lock(&dq_list_lock);
452 if (!clear_dquot_dirty(dquot)) {
453 spin_unlock(&dq_list_lock);
456 spin_unlock(&dq_list_lock);
457 /* Inactive dquot can be only if there was error during read/init
458 * => we have better not writing it */
459 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
460 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
464 mutex_unlock(&dquot->dq_lock);
467 EXPORT_SYMBOL(dquot_commit);
472 int dquot_release(struct dquot *dquot)
474 int ret = 0, ret2 = 0;
475 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
477 mutex_lock(&dquot->dq_lock);
478 /* Check whether we are not racing with some other dqget() */
479 if (atomic_read(&dquot->dq_count) > 1)
481 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
482 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
484 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
485 down_write(&dqopt->dqio_sem);
486 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
487 dquot->dq_sb, dquot->dq_id.type);
488 up_write(&dqopt->dqio_sem);
493 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
495 mutex_unlock(&dquot->dq_lock);
498 EXPORT_SYMBOL(dquot_release);
500 void dquot_destroy(struct dquot *dquot)
502 kmem_cache_free(dquot_cachep, dquot);
504 EXPORT_SYMBOL(dquot_destroy);
506 static inline void do_destroy_dquot(struct dquot *dquot)
508 dquot->dq_sb->dq_op->destroy_dquot(dquot);
511 /* Invalidate all dquots on the list. Note that this function is called after
512 * quota is disabled and pointers from inodes removed so there cannot be new
513 * quota users. There can still be some users of quotas due to inodes being
514 * just deleted or pruned by prune_icache() (those are not attached to any
515 * list) or parallel quotactl call. We have to wait for such users.
517 static void invalidate_dquots(struct super_block *sb, int type)
519 struct dquot *dquot, *tmp;
522 spin_lock(&dq_list_lock);
523 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
524 if (dquot->dq_sb != sb)
526 if (dquot->dq_id.type != type)
528 /* Wait for dquot users */
529 if (atomic_read(&dquot->dq_count)) {
533 prepare_to_wait(&dquot->dq_wait_unused, &wait,
534 TASK_UNINTERRUPTIBLE);
535 spin_unlock(&dq_list_lock);
536 /* Once dqput() wakes us up, we know it's time to free
538 * IMPORTANT: we rely on the fact that there is always
539 * at most one process waiting for dquot to free.
540 * Otherwise dq_count would be > 1 and we would never
543 if (atomic_read(&dquot->dq_count) > 1)
545 finish_wait(&dquot->dq_wait_unused, &wait);
547 /* At this moment dquot() need not exist (it could be
548 * reclaimed by prune_dqcache(). Hence we must
553 * Quota now has no users and it has been written on last
556 remove_dquot_hash(dquot);
557 remove_free_dquot(dquot);
559 do_destroy_dquot(dquot);
561 spin_unlock(&dq_list_lock);
564 /* Call callback for every active dquot on given filesystem */
565 int dquot_scan_active(struct super_block *sb,
566 int (*fn)(struct dquot *dquot, unsigned long priv),
569 struct dquot *dquot, *old_dquot = NULL;
572 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
574 spin_lock(&dq_list_lock);
575 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
576 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
578 if (dquot->dq_sb != sb)
580 /* Now we have active dquot so we can just increase use count */
581 atomic_inc(&dquot->dq_count);
582 spin_unlock(&dq_list_lock);
583 dqstats_inc(DQST_LOOKUPS);
587 * ->release_dquot() can be racing with us. Our reference
588 * protects us from new calls to it so just wait for any
589 * outstanding call and recheck the DQ_ACTIVE_B after that.
591 wait_on_dquot(dquot);
592 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
593 ret = fn(dquot, priv);
597 spin_lock(&dq_list_lock);
598 /* We are safe to continue now because our dquot could not
599 * be moved out of the inuse list while we hold the reference */
601 spin_unlock(&dq_list_lock);
606 EXPORT_SYMBOL(dquot_scan_active);
608 /* Write all dquot structures to quota files */
609 int dquot_writeback_dquots(struct super_block *sb, int type)
611 struct list_head *dirty;
613 struct quota_info *dqopt = sb_dqopt(sb);
617 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
619 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
620 if (type != -1 && cnt != type)
622 if (!sb_has_quota_active(sb, cnt))
624 spin_lock(&dq_list_lock);
625 dirty = &dqopt->info[cnt].dqi_dirty_list;
626 while (!list_empty(dirty)) {
627 dquot = list_first_entry(dirty, struct dquot,
629 /* Dirty and inactive can be only bad dquot... */
630 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
631 clear_dquot_dirty(dquot);
634 /* Now we have active dquot from which someone is
635 * holding reference so we can safely just increase
638 spin_unlock(&dq_list_lock);
639 dqstats_inc(DQST_LOOKUPS);
640 err = sb->dq_op->write_dquot(dquot);
644 spin_lock(&dq_list_lock);
646 spin_unlock(&dq_list_lock);
649 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
650 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
651 && info_dirty(&dqopt->info[cnt]))
652 sb->dq_op->write_info(sb, cnt);
653 dqstats_inc(DQST_SYNCS);
657 EXPORT_SYMBOL(dquot_writeback_dquots);
659 /* Write all dquot structures to disk and make them visible from userspace */
660 int dquot_quota_sync(struct super_block *sb, int type)
662 struct quota_info *dqopt = sb_dqopt(sb);
666 ret = dquot_writeback_dquots(sb, type);
669 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
672 /* This is not very clever (and fast) but currently I don't know about
673 * any other simple way of getting quota data to disk and we must get
674 * them there for userspace to be visible... */
675 if (sb->s_op->sync_fs)
676 sb->s_op->sync_fs(sb, 1);
677 sync_blockdev(sb->s_bdev);
680 * Now when everything is written we can discard the pagecache so
681 * that userspace sees the changes.
683 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
684 if (type != -1 && cnt != type)
686 if (!sb_has_quota_active(sb, cnt))
688 inode_lock(dqopt->files[cnt]);
689 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
690 inode_unlock(dqopt->files[cnt]);
695 EXPORT_SYMBOL(dquot_quota_sync);
698 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
700 struct list_head *head;
702 unsigned long freed = 0;
704 spin_lock(&dq_list_lock);
705 head = free_dquots.prev;
706 while (head != &free_dquots && sc->nr_to_scan) {
707 dquot = list_entry(head, struct dquot, dq_free);
708 remove_dquot_hash(dquot);
709 remove_free_dquot(dquot);
711 do_destroy_dquot(dquot);
714 head = free_dquots.prev;
716 spin_unlock(&dq_list_lock);
721 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
723 return vfs_pressure_ratio(
724 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
727 static struct shrinker dqcache_shrinker = {
728 .count_objects = dqcache_shrink_count,
729 .scan_objects = dqcache_shrink_scan,
730 .seeks = DEFAULT_SEEKS,
734 * Put reference to dquot
736 void dqput(struct dquot *dquot)
742 #ifdef CONFIG_QUOTA_DEBUG
743 if (!atomic_read(&dquot->dq_count)) {
744 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
745 quotatypes[dquot->dq_id.type],
746 from_kqid(&init_user_ns, dquot->dq_id));
750 dqstats_inc(DQST_DROPS);
752 spin_lock(&dq_list_lock);
753 if (atomic_read(&dquot->dq_count) > 1) {
754 /* We have more than one user... nothing to do */
755 atomic_dec(&dquot->dq_count);
756 /* Releasing dquot during quotaoff phase? */
757 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
758 atomic_read(&dquot->dq_count) == 1)
759 wake_up(&dquot->dq_wait_unused);
760 spin_unlock(&dq_list_lock);
763 /* Need to release dquot? */
764 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
765 spin_unlock(&dq_list_lock);
766 /* Commit dquot before releasing */
767 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
769 quota_error(dquot->dq_sb, "Can't write quota structure"
770 " (error %d). Quota may get out of sync!",
773 * We clear dirty bit anyway, so that we avoid
776 spin_lock(&dq_list_lock);
777 clear_dquot_dirty(dquot);
778 spin_unlock(&dq_list_lock);
782 /* Clear flag in case dquot was inactive (something bad happened) */
783 clear_dquot_dirty(dquot);
784 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
785 spin_unlock(&dq_list_lock);
786 dquot->dq_sb->dq_op->release_dquot(dquot);
789 atomic_dec(&dquot->dq_count);
790 #ifdef CONFIG_QUOTA_DEBUG
792 BUG_ON(!list_empty(&dquot->dq_free));
794 put_dquot_last(dquot);
795 spin_unlock(&dq_list_lock);
797 EXPORT_SYMBOL(dqput);
799 struct dquot *dquot_alloc(struct super_block *sb, int type)
801 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
803 EXPORT_SYMBOL(dquot_alloc);
805 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
809 dquot = sb->dq_op->alloc_dquot(sb, type);
813 mutex_init(&dquot->dq_lock);
814 INIT_LIST_HEAD(&dquot->dq_free);
815 INIT_LIST_HEAD(&dquot->dq_inuse);
816 INIT_HLIST_NODE(&dquot->dq_hash);
817 INIT_LIST_HEAD(&dquot->dq_dirty);
818 init_waitqueue_head(&dquot->dq_wait_unused);
820 dquot->dq_id = make_kqid_invalid(type);
821 atomic_set(&dquot->dq_count, 1);
827 * Get reference to dquot
829 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
830 * destroying our dquot by:
831 * a) checking for quota flags under dq_list_lock and
832 * b) getting a reference to dquot before we release dq_list_lock
834 struct dquot *dqget(struct super_block *sb, struct kqid qid)
836 unsigned int hashent = hashfn(sb, qid);
837 struct dquot *dquot, *empty = NULL;
839 if (!qid_has_mapping(sb->s_user_ns, qid))
840 return ERR_PTR(-EINVAL);
842 if (!sb_has_quota_active(sb, qid.type))
843 return ERR_PTR(-ESRCH);
845 spin_lock(&dq_list_lock);
846 spin_lock(&dq_state_lock);
847 if (!sb_has_quota_active(sb, qid.type)) {
848 spin_unlock(&dq_state_lock);
849 spin_unlock(&dq_list_lock);
850 dquot = ERR_PTR(-ESRCH);
853 spin_unlock(&dq_state_lock);
855 dquot = find_dquot(hashent, sb, qid);
858 spin_unlock(&dq_list_lock);
859 empty = get_empty_dquot(sb, qid.type);
861 schedule(); /* Try to wait for a moment... */
867 /* all dquots go on the inuse_list */
869 /* hash it first so it can be found */
870 insert_dquot_hash(dquot);
871 spin_unlock(&dq_list_lock);
872 dqstats_inc(DQST_LOOKUPS);
874 if (!atomic_read(&dquot->dq_count))
875 remove_free_dquot(dquot);
876 atomic_inc(&dquot->dq_count);
877 spin_unlock(&dq_list_lock);
878 dqstats_inc(DQST_CACHE_HITS);
879 dqstats_inc(DQST_LOOKUPS);
881 /* Wait for dq_lock - after this we know that either dquot_release() is
882 * already finished or it will be canceled due to dq_count > 1 test */
883 wait_on_dquot(dquot);
884 /* Read the dquot / allocate space in quota file */
885 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
888 err = sb->dq_op->acquire_dquot(dquot);
891 dquot = ERR_PTR(err);
896 * Make sure following reads see filled structure - paired with
897 * smp_mb__before_atomic() in dquot_acquire().
900 #ifdef CONFIG_QUOTA_DEBUG
901 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
905 do_destroy_dquot(empty);
909 EXPORT_SYMBOL(dqget);
911 static inline struct dquot **i_dquot(struct inode *inode)
913 return inode->i_sb->s_op->get_dquots(inode);
916 static int dqinit_needed(struct inode *inode, int type)
918 struct dquot * const *dquots;
921 if (IS_NOQUOTA(inode))
924 dquots = i_dquot(inode);
926 return !dquots[type];
927 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
933 /* This routine is guarded by s_umount semaphore */
934 static void add_dquot_ref(struct super_block *sb, int type)
936 struct inode *inode, *old_inode = NULL;
937 #ifdef CONFIG_QUOTA_DEBUG
941 spin_lock(&sb->s_inode_list_lock);
942 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
943 spin_lock(&inode->i_lock);
944 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
945 !atomic_read(&inode->i_writecount) ||
946 !dqinit_needed(inode, type)) {
947 spin_unlock(&inode->i_lock);
951 spin_unlock(&inode->i_lock);
952 spin_unlock(&sb->s_inode_list_lock);
954 #ifdef CONFIG_QUOTA_DEBUG
955 if (unlikely(inode_get_rsv_space(inode) > 0))
959 __dquot_initialize(inode, type);
962 * We hold a reference to 'inode' so it couldn't have been
963 * removed from s_inodes list while we dropped the
964 * s_inode_list_lock. We cannot iput the inode now as we can be
965 * holding the last reference and we cannot iput it under
966 * s_inode_list_lock. So we keep the reference and iput it
970 spin_lock(&sb->s_inode_list_lock);
972 spin_unlock(&sb->s_inode_list_lock);
975 #ifdef CONFIG_QUOTA_DEBUG
977 quota_error(sb, "Writes happened before quota was turned on "
978 "thus quota information is probably inconsistent. "
979 "Please run quotacheck(8)");
985 * Remove references to dquots from inode and add dquot to list for freeing
986 * if we have the last reference to dquot
988 static void remove_inode_dquot_ref(struct inode *inode, int type,
989 struct list_head *tofree_head)
991 struct dquot **dquots = i_dquot(inode);
992 struct dquot *dquot = dquots[type];
998 if (list_empty(&dquot->dq_free)) {
1000 * The inode still has reference to dquot so it can't be in the
1003 spin_lock(&dq_list_lock);
1004 list_add(&dquot->dq_free, tofree_head);
1005 spin_unlock(&dq_list_lock);
1008 * Dquot is already in a list to put so we won't drop the last
1016 * Free list of dquots
1017 * Dquots are removed from inodes and no new references can be got so we are
1018 * the only ones holding reference
1020 static void put_dquot_list(struct list_head *tofree_head)
1022 struct list_head *act_head;
1023 struct dquot *dquot;
1025 act_head = tofree_head->next;
1026 while (act_head != tofree_head) {
1027 dquot = list_entry(act_head, struct dquot, dq_free);
1028 act_head = act_head->next;
1029 /* Remove dquot from the list so we won't have problems... */
1030 list_del_init(&dquot->dq_free);
1035 static void remove_dquot_ref(struct super_block *sb, int type,
1036 struct list_head *tofree_head)
1038 struct inode *inode;
1041 spin_lock(&sb->s_inode_list_lock);
1042 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1044 * We have to scan also I_NEW inodes because they can already
1045 * have quota pointer initialized. Luckily, we need to touch
1046 * only quota pointers and these have separate locking
1049 spin_lock(&dq_data_lock);
1050 if (!IS_NOQUOTA(inode)) {
1051 if (unlikely(inode_get_rsv_space(inode) > 0))
1053 remove_inode_dquot_ref(inode, type, tofree_head);
1055 spin_unlock(&dq_data_lock);
1057 spin_unlock(&sb->s_inode_list_lock);
1058 #ifdef CONFIG_QUOTA_DEBUG
1060 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1061 " was disabled thus quota information is probably "
1062 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1067 /* Gather all references from inodes and drop them */
1068 static void drop_dquot_ref(struct super_block *sb, int type)
1070 LIST_HEAD(tofree_head);
1073 remove_dquot_ref(sb, type, &tofree_head);
1074 synchronize_srcu(&dquot_srcu);
1075 put_dquot_list(&tofree_head);
1079 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1081 dquot->dq_dqb.dqb_curinodes += number;
1084 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1086 dquot->dq_dqb.dqb_curspace += number;
1089 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1091 dquot->dq_dqb.dqb_rsvspace += number;
1095 * Claim reserved quota space
1097 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1099 if (dquot->dq_dqb.dqb_rsvspace < number) {
1101 number = dquot->dq_dqb.dqb_rsvspace;
1103 dquot->dq_dqb.dqb_curspace += number;
1104 dquot->dq_dqb.dqb_rsvspace -= number;
1107 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1109 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1110 number = dquot->dq_dqb.dqb_curspace;
1111 dquot->dq_dqb.dqb_rsvspace += number;
1112 dquot->dq_dqb.dqb_curspace -= number;
1116 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1118 if (dquot->dq_dqb.dqb_rsvspace >= number)
1119 dquot->dq_dqb.dqb_rsvspace -= number;
1122 dquot->dq_dqb.dqb_rsvspace = 0;
1126 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1128 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1129 dquot->dq_dqb.dqb_curinodes >= number)
1130 dquot->dq_dqb.dqb_curinodes -= number;
1132 dquot->dq_dqb.dqb_curinodes = 0;
1133 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1134 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1135 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1138 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1140 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1141 dquot->dq_dqb.dqb_curspace >= number)
1142 dquot->dq_dqb.dqb_curspace -= number;
1144 dquot->dq_dqb.dqb_curspace = 0;
1145 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1146 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1147 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1151 struct super_block *w_sb;
1152 struct kqid w_dq_id;
1156 static int warning_issued(struct dquot *dquot, const int warntype)
1158 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1159 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1160 ((warntype == QUOTA_NL_IHARDWARN ||
1161 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1165 return test_and_set_bit(flag, &dquot->dq_flags);
1168 #ifdef CONFIG_PRINT_QUOTA_WARNING
1169 static int flag_print_warnings = 1;
1171 static int need_print_warning(struct dquot_warn *warn)
1173 if (!flag_print_warnings)
1176 switch (warn->w_dq_id.type) {
1178 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1180 return in_group_p(warn->w_dq_id.gid);
1187 /* Print warning to user which exceeded quota */
1188 static void print_warning(struct dquot_warn *warn)
1191 struct tty_struct *tty;
1192 int warntype = warn->w_type;
1194 if (warntype == QUOTA_NL_IHARDBELOW ||
1195 warntype == QUOTA_NL_ISOFTBELOW ||
1196 warntype == QUOTA_NL_BHARDBELOW ||
1197 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1200 tty = get_current_tty();
1203 tty_write_message(tty, warn->w_sb->s_id);
1204 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1205 tty_write_message(tty, ": warning, ");
1207 tty_write_message(tty, ": write failed, ");
1208 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1210 case QUOTA_NL_IHARDWARN:
1211 msg = " file limit reached.\r\n";
1213 case QUOTA_NL_ISOFTLONGWARN:
1214 msg = " file quota exceeded too long.\r\n";
1216 case QUOTA_NL_ISOFTWARN:
1217 msg = " file quota exceeded.\r\n";
1219 case QUOTA_NL_BHARDWARN:
1220 msg = " block limit reached.\r\n";
1222 case QUOTA_NL_BSOFTLONGWARN:
1223 msg = " block quota exceeded too long.\r\n";
1225 case QUOTA_NL_BSOFTWARN:
1226 msg = " block quota exceeded.\r\n";
1229 tty_write_message(tty, msg);
1234 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1237 if (warning_issued(dquot, warntype))
1239 warn->w_type = warntype;
1240 warn->w_sb = dquot->dq_sb;
1241 warn->w_dq_id = dquot->dq_id;
1245 * Write warnings to the console and send warning messages over netlink.
1247 * Note that this function can call into tty and networking code.
1249 static void flush_warnings(struct dquot_warn *warn)
1253 for (i = 0; i < MAXQUOTAS; i++) {
1254 if (warn[i].w_type == QUOTA_NL_NOWARN)
1256 #ifdef CONFIG_PRINT_QUOTA_WARNING
1257 print_warning(&warn[i]);
1259 quota_send_warning(warn[i].w_dq_id,
1260 warn[i].w_sb->s_dev, warn[i].w_type);
1264 static int ignore_hardlimit(struct dquot *dquot)
1266 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1268 return capable(CAP_SYS_RESOURCE) &&
1269 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1270 !(info->dqi_flags & DQF_ROOT_SQUASH));
1273 /* needs dq_data_lock */
1274 static int check_idq(struct dquot *dquot, qsize_t inodes,
1275 struct dquot_warn *warn)
1277 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1279 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1280 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1283 if (dquot->dq_dqb.dqb_ihardlimit &&
1284 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1285 !ignore_hardlimit(dquot)) {
1286 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1290 if (dquot->dq_dqb.dqb_isoftlimit &&
1291 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1292 dquot->dq_dqb.dqb_itime &&
1293 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1294 !ignore_hardlimit(dquot)) {
1295 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1299 if (dquot->dq_dqb.dqb_isoftlimit &&
1300 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1301 dquot->dq_dqb.dqb_itime == 0) {
1302 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1303 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1304 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1310 /* needs dq_data_lock */
1311 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1312 struct dquot_warn *warn)
1315 struct super_block *sb = dquot->dq_sb;
1317 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1318 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1321 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1324 if (dquot->dq_dqb.dqb_bhardlimit &&
1325 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1326 !ignore_hardlimit(dquot)) {
1328 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1332 if (dquot->dq_dqb.dqb_bsoftlimit &&
1333 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1334 dquot->dq_dqb.dqb_btime &&
1335 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1336 !ignore_hardlimit(dquot)) {
1338 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1342 if (dquot->dq_dqb.dqb_bsoftlimit &&
1343 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1344 dquot->dq_dqb.dqb_btime == 0) {
1346 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1347 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1348 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1352 * We don't allow preallocation to exceed softlimit so exceeding will
1361 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1365 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1366 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1367 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1368 return QUOTA_NL_NOWARN;
1370 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1371 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1372 return QUOTA_NL_ISOFTBELOW;
1373 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1374 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1375 return QUOTA_NL_IHARDBELOW;
1376 return QUOTA_NL_NOWARN;
1379 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1381 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1382 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1383 return QUOTA_NL_NOWARN;
1385 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1386 return QUOTA_NL_BSOFTBELOW;
1387 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1388 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1389 return QUOTA_NL_BHARDBELOW;
1390 return QUOTA_NL_NOWARN;
1393 static int dquot_active(const struct inode *inode)
1395 struct super_block *sb = inode->i_sb;
1397 if (IS_NOQUOTA(inode))
1399 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1403 * Initialize quota pointers in inode
1405 * It is better to call this function outside of any transaction as it
1406 * might need a lot of space in journal for dquot structure allocation.
1408 static int __dquot_initialize(struct inode *inode, int type)
1410 int cnt, init_needed = 0;
1411 struct dquot **dquots, *got[MAXQUOTAS] = {};
1412 struct super_block *sb = inode->i_sb;
1416 if (!dquot_active(inode))
1419 dquots = i_dquot(inode);
1421 /* First get references to structures we might need. */
1422 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1426 struct dquot *dquot;
1428 if (type != -1 && cnt != type)
1431 * The i_dquot should have been initialized in most cases,
1432 * we check it without locking here to avoid unnecessary
1433 * dqget()/dqput() calls.
1438 if (!sb_has_quota_active(sb, cnt))
1445 qid = make_kqid_uid(inode->i_uid);
1448 qid = make_kqid_gid(inode->i_gid);
1451 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1454 qid = make_kqid_projid(projid);
1457 dquot = dqget(sb, qid);
1458 if (IS_ERR(dquot)) {
1459 /* We raced with somebody turning quotas off... */
1460 if (PTR_ERR(dquot) != -ESRCH) {
1461 ret = PTR_ERR(dquot);
1469 /* All required i_dquot has been initialized */
1473 spin_lock(&dq_data_lock);
1474 if (IS_NOQUOTA(inode))
1476 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1477 if (type != -1 && cnt != type)
1479 /* Avoid races with quotaoff() */
1480 if (!sb_has_quota_active(sb, cnt))
1482 /* We could race with quotaon or dqget() could have failed */
1486 dquots[cnt] = got[cnt];
1489 * Make quota reservation system happy if someone
1490 * did a write before quota was turned on
1492 rsv = inode_get_rsv_space(inode);
1494 dquot_resv_space(dquots[cnt], rsv);
1498 spin_unlock(&dq_data_lock);
1500 /* Drop unused references */
1506 int dquot_initialize(struct inode *inode)
1508 return __dquot_initialize(inode, -1);
1510 EXPORT_SYMBOL(dquot_initialize);
1512 bool dquot_initialize_needed(struct inode *inode)
1514 struct dquot **dquots;
1517 if (!dquot_active(inode))
1520 dquots = i_dquot(inode);
1521 for (i = 0; i < MAXQUOTAS; i++)
1522 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1526 EXPORT_SYMBOL(dquot_initialize_needed);
1529 * Release all quotas referenced by inode.
1531 * This function only be called on inode free or converting
1532 * a file to quota file, no other users for the i_dquot in
1533 * both cases, so we needn't call synchronize_srcu() after
1536 static void __dquot_drop(struct inode *inode)
1539 struct dquot **dquots = i_dquot(inode);
1540 struct dquot *put[MAXQUOTAS];
1542 spin_lock(&dq_data_lock);
1543 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1544 put[cnt] = dquots[cnt];
1547 spin_unlock(&dq_data_lock);
1551 void dquot_drop(struct inode *inode)
1553 struct dquot * const *dquots;
1556 if (IS_NOQUOTA(inode))
1560 * Test before calling to rule out calls from proc and such
1561 * where we are not allowed to block. Note that this is
1562 * actually reliable test even without the lock - the caller
1563 * must assure that nobody can come after the DQUOT_DROP and
1564 * add quota pointers back anyway.
1566 dquots = i_dquot(inode);
1567 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1572 if (cnt < MAXQUOTAS)
1573 __dquot_drop(inode);
1575 EXPORT_SYMBOL(dquot_drop);
1578 * inode_reserved_space is managed internally by quota, and protected by
1579 * i_lock similar to i_blocks+i_bytes.
1581 static qsize_t *inode_reserved_space(struct inode * inode)
1583 /* Filesystem must explicitly define it's own method in order to use
1584 * quota reservation interface */
1585 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1586 return inode->i_sb->dq_op->get_reserved_space(inode);
1589 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1591 spin_lock(&inode->i_lock);
1592 *inode_reserved_space(inode) += number;
1593 spin_unlock(&inode->i_lock);
1595 EXPORT_SYMBOL(inode_add_rsv_space);
1597 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1599 spin_lock(&inode->i_lock);
1600 *inode_reserved_space(inode) -= number;
1601 __inode_add_bytes(inode, number);
1602 spin_unlock(&inode->i_lock);
1604 EXPORT_SYMBOL(inode_claim_rsv_space);
1606 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1608 spin_lock(&inode->i_lock);
1609 *inode_reserved_space(inode) += number;
1610 __inode_sub_bytes(inode, number);
1611 spin_unlock(&inode->i_lock);
1613 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1615 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1617 spin_lock(&inode->i_lock);
1618 *inode_reserved_space(inode) -= number;
1619 spin_unlock(&inode->i_lock);
1621 EXPORT_SYMBOL(inode_sub_rsv_space);
1623 static qsize_t inode_get_rsv_space(struct inode *inode)
1627 if (!inode->i_sb->dq_op->get_reserved_space)
1629 spin_lock(&inode->i_lock);
1630 ret = *inode_reserved_space(inode);
1631 spin_unlock(&inode->i_lock);
1635 static void inode_incr_space(struct inode *inode, qsize_t number,
1639 inode_add_rsv_space(inode, number);
1641 inode_add_bytes(inode, number);
1644 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1647 inode_sub_rsv_space(inode, number);
1649 inode_sub_bytes(inode, number);
1653 * This functions updates i_blocks+i_bytes fields and quota information
1654 * (together with appropriate checks).
1656 * NOTE: We absolutely rely on the fact that caller dirties the inode
1657 * (usually helpers in quotaops.h care about this) and holds a handle for
1658 * the current transaction so that dquot write and inode write go into the
1663 * This operation can block, but only after everything is updated
1665 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1667 int cnt, ret = 0, index;
1668 struct dquot_warn warn[MAXQUOTAS];
1669 int reserve = flags & DQUOT_SPACE_RESERVE;
1670 struct dquot **dquots;
1672 if (!dquot_active(inode)) {
1673 inode_incr_space(inode, number, reserve);
1677 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1678 warn[cnt].w_type = QUOTA_NL_NOWARN;
1680 dquots = i_dquot(inode);
1681 index = srcu_read_lock(&dquot_srcu);
1682 spin_lock(&dq_data_lock);
1683 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1686 ret = check_bdq(dquots[cnt], number,
1687 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1688 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1689 spin_unlock(&dq_data_lock);
1690 goto out_flush_warn;
1693 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1697 dquot_resv_space(dquots[cnt], number);
1699 dquot_incr_space(dquots[cnt], number);
1701 inode_incr_space(inode, number, reserve);
1702 spin_unlock(&dq_data_lock);
1705 goto out_flush_warn;
1706 mark_all_dquot_dirty(dquots);
1708 srcu_read_unlock(&dquot_srcu, index);
1709 flush_warnings(warn);
1713 EXPORT_SYMBOL(__dquot_alloc_space);
1716 * This operation can block, but only after everything is updated
1718 int dquot_alloc_inode(struct inode *inode)
1720 int cnt, ret = 0, index;
1721 struct dquot_warn warn[MAXQUOTAS];
1722 struct dquot * const *dquots;
1724 if (!dquot_active(inode))
1726 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1727 warn[cnt].w_type = QUOTA_NL_NOWARN;
1729 dquots = i_dquot(inode);
1730 index = srcu_read_lock(&dquot_srcu);
1731 spin_lock(&dq_data_lock);
1732 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1735 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1740 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1743 dquot_incr_inodes(dquots[cnt], 1);
1747 spin_unlock(&dq_data_lock);
1749 mark_all_dquot_dirty(dquots);
1750 srcu_read_unlock(&dquot_srcu, index);
1751 flush_warnings(warn);
1754 EXPORT_SYMBOL(dquot_alloc_inode);
1757 * Convert in-memory reserved quotas to real consumed quotas
1759 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1761 struct dquot **dquots;
1764 if (!dquot_active(inode)) {
1765 inode_claim_rsv_space(inode, number);
1769 dquots = i_dquot(inode);
1770 index = srcu_read_lock(&dquot_srcu);
1771 spin_lock(&dq_data_lock);
1772 /* Claim reserved quotas to allocated quotas */
1773 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1775 dquot_claim_reserved_space(dquots[cnt], number);
1777 /* Update inode bytes */
1778 inode_claim_rsv_space(inode, number);
1779 spin_unlock(&dq_data_lock);
1780 mark_all_dquot_dirty(dquots);
1781 srcu_read_unlock(&dquot_srcu, index);
1784 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1787 * Convert allocated space back to in-memory reserved quotas
1789 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1791 struct dquot **dquots;
1794 if (!dquot_active(inode)) {
1795 inode_reclaim_rsv_space(inode, number);
1799 dquots = i_dquot(inode);
1800 index = srcu_read_lock(&dquot_srcu);
1801 spin_lock(&dq_data_lock);
1802 /* Claim reserved quotas to allocated quotas */
1803 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1805 dquot_reclaim_reserved_space(dquots[cnt], number);
1807 /* Update inode bytes */
1808 inode_reclaim_rsv_space(inode, number);
1809 spin_unlock(&dq_data_lock);
1810 mark_all_dquot_dirty(dquots);
1811 srcu_read_unlock(&dquot_srcu, index);
1814 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1817 * This operation can block, but only after everything is updated
1819 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1822 struct dquot_warn warn[MAXQUOTAS];
1823 struct dquot **dquots;
1824 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1826 if (!dquot_active(inode)) {
1827 inode_decr_space(inode, number, reserve);
1831 dquots = i_dquot(inode);
1832 index = srcu_read_lock(&dquot_srcu);
1833 spin_lock(&dq_data_lock);
1834 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1837 warn[cnt].w_type = QUOTA_NL_NOWARN;
1840 wtype = info_bdq_free(dquots[cnt], number);
1841 if (wtype != QUOTA_NL_NOWARN)
1842 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1844 dquot_free_reserved_space(dquots[cnt], number);
1846 dquot_decr_space(dquots[cnt], number);
1848 inode_decr_space(inode, number, reserve);
1849 spin_unlock(&dq_data_lock);
1853 mark_all_dquot_dirty(dquots);
1855 srcu_read_unlock(&dquot_srcu, index);
1856 flush_warnings(warn);
1858 EXPORT_SYMBOL(__dquot_free_space);
1861 * This operation can block, but only after everything is updated
1863 void dquot_free_inode(struct inode *inode)
1866 struct dquot_warn warn[MAXQUOTAS];
1867 struct dquot * const *dquots;
1870 if (!dquot_active(inode))
1873 dquots = i_dquot(inode);
1874 index = srcu_read_lock(&dquot_srcu);
1875 spin_lock(&dq_data_lock);
1876 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1879 warn[cnt].w_type = QUOTA_NL_NOWARN;
1882 wtype = info_idq_free(dquots[cnt], 1);
1883 if (wtype != QUOTA_NL_NOWARN)
1884 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1885 dquot_decr_inodes(dquots[cnt], 1);
1887 spin_unlock(&dq_data_lock);
1888 mark_all_dquot_dirty(dquots);
1889 srcu_read_unlock(&dquot_srcu, index);
1890 flush_warnings(warn);
1892 EXPORT_SYMBOL(dquot_free_inode);
1895 * Transfer the number of inode and blocks from one diskquota to an other.
1896 * On success, dquot references in transfer_to are consumed and references
1897 * to original dquots that need to be released are placed there. On failure,
1898 * references are kept untouched.
1900 * This operation can block, but only after everything is updated
1901 * A transaction must be started when entering this function.
1903 * We are holding reference on transfer_from & transfer_to, no need to
1904 * protect them by srcu_read_lock().
1906 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1908 qsize_t space, cur_space;
1909 qsize_t rsv_space = 0;
1910 qsize_t inode_usage = 1;
1911 struct dquot *transfer_from[MAXQUOTAS] = {};
1913 char is_valid[MAXQUOTAS] = {};
1914 struct dquot_warn warn_to[MAXQUOTAS];
1915 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1916 struct dquot_warn warn_from_space[MAXQUOTAS];
1918 if (IS_NOQUOTA(inode))
1921 if (inode->i_sb->dq_op->get_inode_usage) {
1922 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1927 /* Initialize the arrays */
1928 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1929 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1930 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1931 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1934 spin_lock(&dq_data_lock);
1935 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1936 spin_unlock(&dq_data_lock);
1939 cur_space = inode_get_bytes(inode);
1940 rsv_space = inode_get_rsv_space(inode);
1941 space = cur_space + rsv_space;
1942 /* Build the transfer_from list and check the limits */
1943 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1945 * Skip changes for same uid or gid or for turned off quota-type.
1947 if (!transfer_to[cnt])
1949 /* Avoid races with quotaoff() */
1950 if (!sb_has_quota_active(inode->i_sb, cnt))
1953 transfer_from[cnt] = i_dquot(inode)[cnt];
1954 ret = check_idq(transfer_to[cnt], inode_usage, &warn_to[cnt]);
1957 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1963 * Finally perform the needed transfer from transfer_from to transfer_to
1965 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1968 /* Due to IO error we might not have transfer_from[] structure */
1969 if (transfer_from[cnt]) {
1971 wtype = info_idq_free(transfer_from[cnt], inode_usage);
1972 if (wtype != QUOTA_NL_NOWARN)
1973 prepare_warning(&warn_from_inodes[cnt],
1974 transfer_from[cnt], wtype);
1975 wtype = info_bdq_free(transfer_from[cnt], space);
1976 if (wtype != QUOTA_NL_NOWARN)
1977 prepare_warning(&warn_from_space[cnt],
1978 transfer_from[cnt], wtype);
1979 dquot_decr_inodes(transfer_from[cnt], inode_usage);
1980 dquot_decr_space(transfer_from[cnt], cur_space);
1981 dquot_free_reserved_space(transfer_from[cnt],
1985 dquot_incr_inodes(transfer_to[cnt], inode_usage);
1986 dquot_incr_space(transfer_to[cnt], cur_space);
1987 dquot_resv_space(transfer_to[cnt], rsv_space);
1989 i_dquot(inode)[cnt] = transfer_to[cnt];
1991 spin_unlock(&dq_data_lock);
1993 mark_all_dquot_dirty(transfer_from);
1994 mark_all_dquot_dirty(transfer_to);
1995 flush_warnings(warn_to);
1996 flush_warnings(warn_from_inodes);
1997 flush_warnings(warn_from_space);
1998 /* Pass back references to put */
1999 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2001 transfer_to[cnt] = transfer_from[cnt];
2004 spin_unlock(&dq_data_lock);
2005 flush_warnings(warn_to);
2008 EXPORT_SYMBOL(__dquot_transfer);
2010 /* Wrapper for transferring ownership of an inode for uid/gid only
2011 * Called from FSXXX_setattr()
2013 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2015 struct dquot *transfer_to[MAXQUOTAS] = {};
2016 struct dquot *dquot;
2017 struct super_block *sb = inode->i_sb;
2020 if (!dquot_active(inode))
2023 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2024 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2025 if (IS_ERR(dquot)) {
2026 if (PTR_ERR(dquot) != -ESRCH) {
2027 ret = PTR_ERR(dquot);
2032 transfer_to[USRQUOTA] = dquot;
2034 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2035 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2036 if (IS_ERR(dquot)) {
2037 if (PTR_ERR(dquot) != -ESRCH) {
2038 ret = PTR_ERR(dquot);
2043 transfer_to[GRPQUOTA] = dquot;
2045 ret = __dquot_transfer(inode, transfer_to);
2047 dqput_all(transfer_to);
2050 EXPORT_SYMBOL(dquot_transfer);
2053 * Write info of quota file to disk
2055 int dquot_commit_info(struct super_block *sb, int type)
2058 struct quota_info *dqopt = sb_dqopt(sb);
2060 down_write(&dqopt->dqio_sem);
2061 ret = dqopt->ops[type]->write_file_info(sb, type);
2062 up_write(&dqopt->dqio_sem);
2065 EXPORT_SYMBOL(dquot_commit_info);
2067 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2069 struct quota_info *dqopt = sb_dqopt(sb);
2071 if (!sb_has_quota_active(sb, qid->type))
2073 if (!dqopt->ops[qid->type]->get_next_id)
2075 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2077 EXPORT_SYMBOL(dquot_get_next_id);
2080 * Definitions of diskquota operations.
2082 const struct dquot_operations dquot_operations = {
2083 .write_dquot = dquot_commit,
2084 .acquire_dquot = dquot_acquire,
2085 .release_dquot = dquot_release,
2086 .mark_dirty = dquot_mark_dquot_dirty,
2087 .write_info = dquot_commit_info,
2088 .alloc_dquot = dquot_alloc,
2089 .destroy_dquot = dquot_destroy,
2090 .get_next_id = dquot_get_next_id,
2092 EXPORT_SYMBOL(dquot_operations);
2095 * Generic helper for ->open on filesystems supporting disk quotas.
2097 int dquot_file_open(struct inode *inode, struct file *file)
2101 error = generic_file_open(inode, file);
2102 if (!error && (file->f_mode & FMODE_WRITE))
2103 dquot_initialize(inode);
2106 EXPORT_SYMBOL(dquot_file_open);
2109 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2111 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2114 struct quota_info *dqopt = sb_dqopt(sb);
2115 struct inode *toputinode[MAXQUOTAS];
2117 /* s_umount should be held in exclusive mode */
2118 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2119 up_read(&sb->s_umount);
2121 /* Cannot turn off usage accounting without turning off limits, or
2122 * suspend quotas and simultaneously turn quotas off. */
2123 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2124 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2125 DQUOT_USAGE_ENABLED)))
2129 * Skip everything if there's nothing to do. We have to do this because
2130 * sometimes we are called when fill_super() failed and calling
2131 * sync_fs() in such cases does no good.
2133 if (!sb_any_quota_loaded(sb))
2136 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2137 toputinode[cnt] = NULL;
2138 if (type != -1 && cnt != type)
2140 if (!sb_has_quota_loaded(sb, cnt))
2143 if (flags & DQUOT_SUSPENDED) {
2144 spin_lock(&dq_state_lock);
2146 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2147 spin_unlock(&dq_state_lock);
2149 spin_lock(&dq_state_lock);
2150 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2151 /* Turning off suspended quotas? */
2152 if (!sb_has_quota_loaded(sb, cnt) &&
2153 sb_has_quota_suspended(sb, cnt)) {
2154 dqopt->flags &= ~dquot_state_flag(
2155 DQUOT_SUSPENDED, cnt);
2156 spin_unlock(&dq_state_lock);
2157 iput(dqopt->files[cnt]);
2158 dqopt->files[cnt] = NULL;
2161 spin_unlock(&dq_state_lock);
2164 /* We still have to keep quota loaded? */
2165 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2168 /* Note: these are blocking operations */
2169 drop_dquot_ref(sb, cnt);
2170 invalidate_dquots(sb, cnt);
2172 * Now all dquots should be invalidated, all writes done so we
2173 * should be only users of the info. No locks needed.
2175 if (info_dirty(&dqopt->info[cnt]))
2176 sb->dq_op->write_info(sb, cnt);
2177 if (dqopt->ops[cnt]->free_file_info)
2178 dqopt->ops[cnt]->free_file_info(sb, cnt);
2179 put_quota_format(dqopt->info[cnt].dqi_format);
2181 toputinode[cnt] = dqopt->files[cnt];
2182 if (!sb_has_quota_loaded(sb, cnt))
2183 dqopt->files[cnt] = NULL;
2184 dqopt->info[cnt].dqi_flags = 0;
2185 dqopt->info[cnt].dqi_igrace = 0;
2186 dqopt->info[cnt].dqi_bgrace = 0;
2187 dqopt->ops[cnt] = NULL;
2190 /* Skip syncing and setting flags if quota files are hidden */
2191 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2194 /* Sync the superblock so that buffers with quota data are written to
2195 * disk (and so userspace sees correct data afterwards). */
2196 if (sb->s_op->sync_fs)
2197 sb->s_op->sync_fs(sb, 1);
2198 sync_blockdev(sb->s_bdev);
2199 /* Now the quota files are just ordinary files and we can set the
2200 * inode flags back. Moreover we discard the pagecache so that
2201 * userspace sees the writes we did bypassing the pagecache. We
2202 * must also discard the blockdev buffers so that we see the
2203 * changes done by userspace on the next quotaon() */
2204 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2205 /* This can happen when suspending quotas on remount-ro... */
2206 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2207 inode_lock(toputinode[cnt]);
2208 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2209 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2210 inode_unlock(toputinode[cnt]);
2211 mark_inode_dirty_sync(toputinode[cnt]);
2214 invalidate_bdev(sb->s_bdev);
2216 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2217 if (toputinode[cnt]) {
2218 /* On remount RO, we keep the inode pointer so that we
2219 * can reenable quota on the subsequent remount RW. We
2220 * have to check 'flags' variable and not use sb_has_
2221 * function because another quotaon / quotaoff could
2222 * change global state before we got here. We refuse
2223 * to suspend quotas when there is pending delete on
2224 * the quota file... */
2225 if (!(flags & DQUOT_SUSPENDED))
2226 iput(toputinode[cnt]);
2227 else if (!toputinode[cnt]->i_nlink)
2232 EXPORT_SYMBOL(dquot_disable);
2234 int dquot_quota_off(struct super_block *sb, int type)
2236 return dquot_disable(sb, type,
2237 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2239 EXPORT_SYMBOL(dquot_quota_off);
2242 * Turn quotas on on a device
2246 * Helper function to turn quotas on when we already have the inode of
2247 * quota file and no quota information is loaded.
2249 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2252 struct quota_format_type *fmt = find_quota_format(format_id);
2253 struct super_block *sb = inode->i_sb;
2254 struct quota_info *dqopt = sb_dqopt(sb);
2259 if (!S_ISREG(inode->i_mode)) {
2263 if (IS_RDONLY(inode)) {
2267 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2268 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2272 /* Filesystems outside of init_user_ns not yet supported */
2273 if (sb->s_user_ns != &init_user_ns) {
2277 /* Usage always has to be set... */
2278 if (!(flags & DQUOT_USAGE_ENABLED)) {
2282 if (sb_has_quota_loaded(sb, type)) {
2287 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2288 /* As we bypass the pagecache we must now flush all the
2289 * dirty data and invalidate caches so that kernel sees
2290 * changes from userspace. It is not enough to just flush
2291 * the quota file since if blocksize < pagesize, invalidation
2292 * of the cache could fail because of other unrelated dirty
2294 sync_filesystem(sb);
2295 invalidate_bdev(sb->s_bdev);
2298 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2299 /* We don't want quota and atime on quota files (deadlocks
2300 * possible) Also nobody should write to the file - we use
2301 * special IO operations which ignore the immutable bit. */
2303 inode->i_flags |= S_NOQUOTA;
2304 inode_unlock(inode);
2306 * When S_NOQUOTA is set, remove dquot references as no more
2307 * references can be added
2309 __dquot_drop(inode);
2313 dqopt->files[type] = igrab(inode);
2314 if (!dqopt->files[type])
2315 goto out_file_flags;
2317 if (!fmt->qf_ops->check_quota_file(sb, type))
2320 dqopt->ops[type] = fmt->qf_ops;
2321 dqopt->info[type].dqi_format = fmt;
2322 dqopt->info[type].dqi_fmt_id = format_id;
2323 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2324 down_read(&dqopt->dqio_sem);
2325 error = dqopt->ops[type]->read_file_info(sb, type);
2327 up_read(&dqopt->dqio_sem);
2330 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2331 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2332 up_read(&dqopt->dqio_sem);
2333 spin_lock(&dq_state_lock);
2334 dqopt->flags |= dquot_state_flag(flags, type);
2335 spin_unlock(&dq_state_lock);
2337 add_dquot_ref(sb, type);
2342 dqopt->files[type] = NULL;
2346 inode->i_flags &= ~S_NOQUOTA;
2347 inode_unlock(inode);
2349 put_quota_format(fmt);
2354 /* Reenable quotas on remount RW */
2355 int dquot_resume(struct super_block *sb, int type)
2357 struct quota_info *dqopt = sb_dqopt(sb);
2358 struct inode *inode;
2362 /* s_umount should be held in exclusive mode */
2363 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2364 up_read(&sb->s_umount);
2366 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2367 if (type != -1 && cnt != type)
2369 if (!sb_has_quota_suspended(sb, cnt))
2372 inode = dqopt->files[cnt];
2373 dqopt->files[cnt] = NULL;
2374 spin_lock(&dq_state_lock);
2375 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2376 DQUOT_LIMITS_ENABLED,
2378 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2379 spin_unlock(&dq_state_lock);
2381 flags = dquot_generic_flag(flags, cnt);
2382 ret = vfs_load_quota_inode(inode, cnt,
2383 dqopt->info[cnt].dqi_fmt_id, flags);
2389 EXPORT_SYMBOL(dquot_resume);
2391 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2392 const struct path *path)
2394 int error = security_quota_on(path->dentry);
2397 /* Quota file not on the same filesystem? */
2398 if (path->dentry->d_sb != sb)
2401 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2402 format_id, DQUOT_USAGE_ENABLED |
2403 DQUOT_LIMITS_ENABLED);
2406 EXPORT_SYMBOL(dquot_quota_on);
2409 * More powerful function for turning on quotas allowing setting
2410 * of individual quota flags
2412 int dquot_enable(struct inode *inode, int type, int format_id,
2415 struct super_block *sb = inode->i_sb;
2417 /* Just unsuspend quotas? */
2418 BUG_ON(flags & DQUOT_SUSPENDED);
2419 /* s_umount should be held in exclusive mode */
2420 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2421 up_read(&sb->s_umount);
2425 /* Just updating flags needed? */
2426 if (sb_has_quota_loaded(sb, type)) {
2427 if (flags & DQUOT_USAGE_ENABLED &&
2428 sb_has_quota_usage_enabled(sb, type))
2430 if (flags & DQUOT_LIMITS_ENABLED &&
2431 sb_has_quota_limits_enabled(sb, type))
2433 spin_lock(&dq_state_lock);
2434 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2435 spin_unlock(&dq_state_lock);
2439 return vfs_load_quota_inode(inode, type, format_id, flags);
2441 EXPORT_SYMBOL(dquot_enable);
2444 * This function is used when filesystem needs to initialize quotas
2445 * during mount time.
2447 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2448 int format_id, int type)
2450 struct dentry *dentry;
2453 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2455 return PTR_ERR(dentry);
2457 if (d_really_is_negative(dentry)) {
2462 error = security_quota_on(dentry);
2464 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2465 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2471 EXPORT_SYMBOL(dquot_quota_on_mount);
2473 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2477 struct quota_info *dqopt = sb_dqopt(sb);
2479 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2481 /* Accounting cannot be turned on while fs is mounted */
2482 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2485 for (type = 0; type < MAXQUOTAS; type++) {
2486 if (!(flags & qtype_enforce_flag(type)))
2488 /* Can't enforce without accounting */
2489 if (!sb_has_quota_usage_enabled(sb, type))
2491 ret = dquot_enable(dqopt->files[type], type,
2492 dqopt->info[type].dqi_fmt_id,
2493 DQUOT_LIMITS_ENABLED);
2499 /* Backout enforcement enablement we already did */
2500 for (type--; type >= 0; type--) {
2501 if (flags & qtype_enforce_flag(type))
2502 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2504 /* Error code translation for better compatibility with XFS */
2510 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2514 struct quota_info *dqopt = sb_dqopt(sb);
2516 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2519 * We don't support turning off accounting via quotactl. In principle
2520 * quota infrastructure can do this but filesystems don't expect
2521 * userspace to be able to do it.
2524 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2527 /* Filter out limits not enabled */
2528 for (type = 0; type < MAXQUOTAS; type++)
2529 if (!sb_has_quota_limits_enabled(sb, type))
2530 flags &= ~qtype_enforce_flag(type);
2534 for (type = 0; type < MAXQUOTAS; type++) {
2535 if (flags & qtype_enforce_flag(type)) {
2536 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2543 /* Backout enforcement disabling we already did */
2544 for (type--; type >= 0; type--) {
2545 if (flags & qtype_enforce_flag(type))
2546 dquot_enable(dqopt->files[type], type,
2547 dqopt->info[type].dqi_fmt_id,
2548 DQUOT_LIMITS_ENABLED);
2553 /* Generic routine for getting common part of quota structure */
2554 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2556 struct mem_dqblk *dm = &dquot->dq_dqb;
2558 memset(di, 0, sizeof(*di));
2559 spin_lock(&dq_data_lock);
2560 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2561 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2562 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2563 di->d_ino_softlimit = dm->dqb_isoftlimit;
2564 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2565 di->d_ino_count = dm->dqb_curinodes;
2566 di->d_spc_timer = dm->dqb_btime;
2567 di->d_ino_timer = dm->dqb_itime;
2568 spin_unlock(&dq_data_lock);
2571 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2572 struct qc_dqblk *di)
2574 struct dquot *dquot;
2576 dquot = dqget(sb, qid);
2578 return PTR_ERR(dquot);
2579 do_get_dqblk(dquot, di);
2584 EXPORT_SYMBOL(dquot_get_dqblk);
2586 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2587 struct qc_dqblk *di)
2589 struct dquot *dquot;
2592 if (!sb->dq_op->get_next_id)
2594 err = sb->dq_op->get_next_id(sb, qid);
2597 dquot = dqget(sb, *qid);
2599 return PTR_ERR(dquot);
2600 do_get_dqblk(dquot, di);
2605 EXPORT_SYMBOL(dquot_get_next_dqblk);
2607 #define VFS_QC_MASK \
2608 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2609 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2610 QC_SPC_TIMER | QC_INO_TIMER)
2612 /* Generic routine for setting common part of quota structure */
2613 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2615 struct mem_dqblk *dm = &dquot->dq_dqb;
2616 int check_blim = 0, check_ilim = 0;
2617 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2619 if (di->d_fieldmask & ~VFS_QC_MASK)
2622 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2623 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2624 ((di->d_fieldmask & QC_SPC_HARD) &&
2625 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2626 ((di->d_fieldmask & QC_INO_SOFT) &&
2627 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2628 ((di->d_fieldmask & QC_INO_HARD) &&
2629 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2632 spin_lock(&dq_data_lock);
2633 if (di->d_fieldmask & QC_SPACE) {
2634 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2636 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2639 if (di->d_fieldmask & QC_SPC_SOFT)
2640 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2641 if (di->d_fieldmask & QC_SPC_HARD)
2642 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2643 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2645 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2648 if (di->d_fieldmask & QC_INO_COUNT) {
2649 dm->dqb_curinodes = di->d_ino_count;
2651 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2654 if (di->d_fieldmask & QC_INO_SOFT)
2655 dm->dqb_isoftlimit = di->d_ino_softlimit;
2656 if (di->d_fieldmask & QC_INO_HARD)
2657 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2658 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2660 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2663 if (di->d_fieldmask & QC_SPC_TIMER) {
2664 dm->dqb_btime = di->d_spc_timer;
2666 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2669 if (di->d_fieldmask & QC_INO_TIMER) {
2670 dm->dqb_itime = di->d_ino_timer;
2672 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2676 if (!dm->dqb_bsoftlimit ||
2677 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2679 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2680 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2681 /* Set grace only if user hasn't provided his own... */
2682 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2685 if (!dm->dqb_isoftlimit ||
2686 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2688 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2689 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2690 /* Set grace only if user hasn't provided his own... */
2691 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2693 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2695 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2697 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2698 spin_unlock(&dq_data_lock);
2699 mark_dquot_dirty(dquot);
2704 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2705 struct qc_dqblk *di)
2707 struct dquot *dquot;
2710 dquot = dqget(sb, qid);
2711 if (IS_ERR(dquot)) {
2712 rc = PTR_ERR(dquot);
2715 rc = do_set_dqblk(dquot, di);
2720 EXPORT_SYMBOL(dquot_set_dqblk);
2722 /* Generic routine for getting common part of quota file information */
2723 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2725 struct mem_dqinfo *mi;
2726 struct qc_type_state *tstate;
2727 struct quota_info *dqopt = sb_dqopt(sb);
2730 memset(state, 0, sizeof(*state));
2731 for (type = 0; type < MAXQUOTAS; type++) {
2732 if (!sb_has_quota_active(sb, type))
2734 tstate = state->s_state + type;
2735 mi = sb_dqopt(sb)->info + type;
2736 tstate->flags = QCI_ACCT_ENABLED;
2737 spin_lock(&dq_data_lock);
2738 if (mi->dqi_flags & DQF_SYS_FILE)
2739 tstate->flags |= QCI_SYSFILE;
2740 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2741 tstate->flags |= QCI_ROOT_SQUASH;
2742 if (sb_has_quota_limits_enabled(sb, type))
2743 tstate->flags |= QCI_LIMITS_ENFORCED;
2744 tstate->spc_timelimit = mi->dqi_bgrace;
2745 tstate->ino_timelimit = mi->dqi_igrace;
2746 tstate->ino = dqopt->files[type]->i_ino;
2747 tstate->blocks = dqopt->files[type]->i_blocks;
2748 tstate->nextents = 1; /* We don't know... */
2749 spin_unlock(&dq_data_lock);
2753 EXPORT_SYMBOL(dquot_get_state);
2755 /* Generic routine for setting common part of quota file information */
2756 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2758 struct mem_dqinfo *mi;
2761 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2762 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2764 if (!sb_has_quota_active(sb, type))
2766 mi = sb_dqopt(sb)->info + type;
2767 if (ii->i_fieldmask & QC_FLAGS) {
2768 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2769 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2772 spin_lock(&dq_data_lock);
2773 if (ii->i_fieldmask & QC_SPC_TIMER)
2774 mi->dqi_bgrace = ii->i_spc_timelimit;
2775 if (ii->i_fieldmask & QC_INO_TIMER)
2776 mi->dqi_igrace = ii->i_ino_timelimit;
2777 if (ii->i_fieldmask & QC_FLAGS) {
2778 if (ii->i_flags & QCI_ROOT_SQUASH)
2779 mi->dqi_flags |= DQF_ROOT_SQUASH;
2781 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2783 spin_unlock(&dq_data_lock);
2784 mark_info_dirty(sb, type);
2785 /* Force write to disk */
2786 sb->dq_op->write_info(sb, type);
2789 EXPORT_SYMBOL(dquot_set_dqinfo);
2791 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2792 .quota_enable = dquot_quota_enable,
2793 .quota_disable = dquot_quota_disable,
2794 .quota_sync = dquot_quota_sync,
2795 .get_state = dquot_get_state,
2796 .set_info = dquot_set_dqinfo,
2797 .get_dqblk = dquot_get_dqblk,
2798 .get_nextdqblk = dquot_get_next_dqblk,
2799 .set_dqblk = dquot_set_dqblk
2801 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2803 static int do_proc_dqstats(struct ctl_table *table, int write,
2804 void __user *buffer, size_t *lenp, loff_t *ppos)
2806 unsigned int type = (int *)table->data - dqstats.stat;
2808 /* Update global table */
2809 dqstats.stat[type] =
2810 percpu_counter_sum_positive(&dqstats.counter[type]);
2811 return proc_dointvec(table, write, buffer, lenp, ppos);
2814 static struct ctl_table fs_dqstats_table[] = {
2816 .procname = "lookups",
2817 .data = &dqstats.stat[DQST_LOOKUPS],
2818 .maxlen = sizeof(int),
2820 .proc_handler = do_proc_dqstats,
2823 .procname = "drops",
2824 .data = &dqstats.stat[DQST_DROPS],
2825 .maxlen = sizeof(int),
2827 .proc_handler = do_proc_dqstats,
2830 .procname = "reads",
2831 .data = &dqstats.stat[DQST_READS],
2832 .maxlen = sizeof(int),
2834 .proc_handler = do_proc_dqstats,
2837 .procname = "writes",
2838 .data = &dqstats.stat[DQST_WRITES],
2839 .maxlen = sizeof(int),
2841 .proc_handler = do_proc_dqstats,
2844 .procname = "cache_hits",
2845 .data = &dqstats.stat[DQST_CACHE_HITS],
2846 .maxlen = sizeof(int),
2848 .proc_handler = do_proc_dqstats,
2851 .procname = "allocated_dquots",
2852 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2853 .maxlen = sizeof(int),
2855 .proc_handler = do_proc_dqstats,
2858 .procname = "free_dquots",
2859 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2860 .maxlen = sizeof(int),
2862 .proc_handler = do_proc_dqstats,
2865 .procname = "syncs",
2866 .data = &dqstats.stat[DQST_SYNCS],
2867 .maxlen = sizeof(int),
2869 .proc_handler = do_proc_dqstats,
2871 #ifdef CONFIG_PRINT_QUOTA_WARNING
2873 .procname = "warnings",
2874 .data = &flag_print_warnings,
2875 .maxlen = sizeof(int),
2877 .proc_handler = proc_dointvec,
2883 static struct ctl_table fs_table[] = {
2885 .procname = "quota",
2887 .child = fs_dqstats_table,
2892 static struct ctl_table sys_table[] = {
2901 static int __init dquot_init(void)
2904 unsigned long nr_hash, order;
2906 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2908 register_sysctl_table(sys_table);
2910 dquot_cachep = kmem_cache_create("dquot",
2911 sizeof(struct dquot), sizeof(unsigned long) * 4,
2912 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2913 SLAB_MEM_SPREAD|SLAB_PANIC),
2917 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2919 panic("Cannot create dquot hash table");
2921 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2922 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2924 panic("Cannot create dquot stat counters");
2927 /* Find power-of-two hlist_heads which can fit into allocation */
2928 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2932 } while (nr_hash >> dq_hash_bits);
2935 nr_hash = 1UL << dq_hash_bits;
2936 dq_hash_mask = nr_hash - 1;
2937 for (i = 0; i < nr_hash; i++)
2938 INIT_HLIST_HEAD(dquot_hash + i);
2940 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2941 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2943 register_shrinker(&dqcache_shrinker);
2947 fs_initcall(dquot_init);