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.
16 * Revised list management to avoid races
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.
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
40 * Added dynamic quota structure allocation
43 * Rewritten quota interface. Implemented new quota format and
44 * formats registering.
50 * Added journalled quota support, fix lock inversion problems
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/kmod.h>
76 #include <linux/namei.h>
77 #include <linux/capability.h>
78 #include <linux/quotaops.h>
79 #include "../internal.h" /* ugh */
81 #include <linux/uaccess.h>
84 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
86 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
87 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
88 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
89 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
90 * modifications of quota state (on quotaon and quotaoff) and readers who care
91 * about latest values take it as well.
93 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
94 * dq_list_lock > dq_state_lock
96 * Note that some things (eg. sb pointer, type, id) doesn't change during
97 * the life of the dquot structure and so needn't to be protected by a lock
99 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
100 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
101 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
102 * inode and before dropping dquot references to avoid use of dquots after
103 * they are freed. dq_data_lock is used to serialize the pointer setting and
104 * clearing operations.
105 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
106 * inode is a quota file). Functions adding pointers from inode to dquots have
107 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
108 * have to do all pointer modifications before dropping dq_data_lock. This makes
109 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
110 * then drops all pointers to dquots from an inode.
112 * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
113 * from inodes (dquot_alloc_space() and such don't check the dq_lock).
114 * Currently dquot is locked only when it is being read to memory (or space for
115 * it is being allocated) on the first dqget() and when it is being released on
116 * the last dqput(). The allocation and release oparations are serialized by
117 * the dq_lock and by checking the use count in dquot_release(). Write
118 * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
119 * spinlock to internal buffers before writing.
121 * Lock ordering (including related VFS locks) is the following:
122 * dqonoff_mutex > i_mutex > journal_lock > dquot->dq_lock > dqio_mutex
123 * dqonoff_mutex > i_mutex comes from dquot_quota_sync, dquot_enable, etc.
126 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
128 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
129 EXPORT_SYMBOL(dq_data_lock);
130 DEFINE_STATIC_SRCU(dquot_srcu);
132 void __quota_error(struct super_block *sb, const char *func,
133 const char *fmt, ...)
135 if (printk_ratelimit()) {
137 struct va_format vaf;
144 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
145 sb->s_id, func, &vaf);
150 EXPORT_SYMBOL(__quota_error);
152 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
153 static char *quotatypes[] = INITQFNAMES;
155 static struct quota_format_type *quota_formats; /* List of registered formats */
156 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
158 /* SLAB cache for dquot structures */
159 static struct kmem_cache *dquot_cachep;
161 int register_quota_format(struct quota_format_type *fmt)
163 spin_lock(&dq_list_lock);
164 fmt->qf_next = quota_formats;
166 spin_unlock(&dq_list_lock);
169 EXPORT_SYMBOL(register_quota_format);
171 void unregister_quota_format(struct quota_format_type *fmt)
173 struct quota_format_type **actqf;
175 spin_lock(&dq_list_lock);
176 for (actqf = "a_formats; *actqf && *actqf != fmt;
177 actqf = &(*actqf)->qf_next)
180 *actqf = (*actqf)->qf_next;
181 spin_unlock(&dq_list_lock);
183 EXPORT_SYMBOL(unregister_quota_format);
185 static struct quota_format_type *find_quota_format(int id)
187 struct quota_format_type *actqf;
189 spin_lock(&dq_list_lock);
190 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
191 actqf = actqf->qf_next)
193 if (!actqf || !try_module_get(actqf->qf_owner)) {
196 spin_unlock(&dq_list_lock);
198 for (qm = 0; module_names[qm].qm_fmt_id &&
199 module_names[qm].qm_fmt_id != id; qm++)
201 if (!module_names[qm].qm_fmt_id ||
202 request_module(module_names[qm].qm_mod_name))
205 spin_lock(&dq_list_lock);
206 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
207 actqf = actqf->qf_next)
209 if (actqf && !try_module_get(actqf->qf_owner))
212 spin_unlock(&dq_list_lock);
216 static void put_quota_format(struct quota_format_type *fmt)
218 module_put(fmt->qf_owner);
222 * Dquot List Management:
223 * The quota code uses three lists for dquot management: the inuse_list,
224 * free_dquots, and dquot_hash[] array. A single dquot structure may be
225 * on all three lists, depending on its current state.
227 * All dquots are placed to the end of inuse_list when first created, and this
228 * list is used for invalidate operation, which must look at every dquot.
230 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
231 * and this list is searched whenever we need an available dquot. Dquots are
232 * removed from the list as soon as they are used again, and
233 * dqstats.free_dquots gives the number of dquots on the list. When
234 * dquot is invalidated it's completely released from memory.
236 * Dquots with a specific identity (device, type and id) are placed on
237 * one of the dquot_hash[] hash chains. The provides an efficient search
238 * mechanism to locate a specific dquot.
241 static LIST_HEAD(inuse_list);
242 static LIST_HEAD(free_dquots);
243 static unsigned int dq_hash_bits, dq_hash_mask;
244 static struct hlist_head *dquot_hash;
246 struct dqstats dqstats;
247 EXPORT_SYMBOL(dqstats);
249 static qsize_t inode_get_rsv_space(struct inode *inode);
250 static int __dquot_initialize(struct inode *inode, int type);
252 static inline unsigned int
253 hashfn(const struct super_block *sb, struct kqid qid)
255 unsigned int id = from_kqid(&init_user_ns, qid);
259 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
260 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
264 * Following list functions expect dq_list_lock to be held
266 static inline void insert_dquot_hash(struct dquot *dquot)
268 struct hlist_head *head;
269 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
270 hlist_add_head(&dquot->dq_hash, head);
273 static inline void remove_dquot_hash(struct dquot *dquot)
275 hlist_del_init(&dquot->dq_hash);
278 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
281 struct hlist_node *node;
284 hlist_for_each (node, dquot_hash+hashent) {
285 dquot = hlist_entry(node, struct dquot, dq_hash);
286 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
292 /* Add a dquot to the tail of the free list */
293 static inline void put_dquot_last(struct dquot *dquot)
295 list_add_tail(&dquot->dq_free, &free_dquots);
296 dqstats_inc(DQST_FREE_DQUOTS);
299 static inline void remove_free_dquot(struct dquot *dquot)
301 if (list_empty(&dquot->dq_free))
303 list_del_init(&dquot->dq_free);
304 dqstats_dec(DQST_FREE_DQUOTS);
307 static inline void put_inuse(struct dquot *dquot)
309 /* We add to the back of inuse list so we don't have to restart
310 * when traversing this list and we block */
311 list_add_tail(&dquot->dq_inuse, &inuse_list);
312 dqstats_inc(DQST_ALLOC_DQUOTS);
315 static inline void remove_inuse(struct dquot *dquot)
317 dqstats_dec(DQST_ALLOC_DQUOTS);
318 list_del(&dquot->dq_inuse);
321 * End of list functions needing dq_list_lock
324 static void wait_on_dquot(struct dquot *dquot)
326 mutex_lock(&dquot->dq_lock);
327 mutex_unlock(&dquot->dq_lock);
330 static inline int dquot_dirty(struct dquot *dquot)
332 return test_bit(DQ_MOD_B, &dquot->dq_flags);
335 static inline int mark_dquot_dirty(struct dquot *dquot)
337 return dquot->dq_sb->dq_op->mark_dirty(dquot);
340 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
341 int dquot_mark_dquot_dirty(struct dquot *dquot)
345 /* If quota is dirty already, we don't have to acquire dq_list_lock */
346 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
349 spin_lock(&dq_list_lock);
350 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
351 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
352 info[dquot->dq_id.type].dqi_dirty_list);
355 spin_unlock(&dq_list_lock);
358 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
360 /* Dirtify all the dquots - this can block when journalling */
361 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
366 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
368 /* Even in case of error we have to continue */
369 ret = mark_dquot_dirty(dquot[cnt]);
376 static inline void dqput_all(struct dquot **dquot)
380 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
384 /* This function needs dq_list_lock */
385 static inline int clear_dquot_dirty(struct dquot *dquot)
387 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
389 list_del_init(&dquot->dq_dirty);
393 void mark_info_dirty(struct super_block *sb, int type)
395 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
397 EXPORT_SYMBOL(mark_info_dirty);
400 * Read dquot from disk and alloc space for it
403 int dquot_acquire(struct dquot *dquot)
405 int ret = 0, ret2 = 0;
406 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
408 mutex_lock(&dquot->dq_lock);
409 mutex_lock(&dqopt->dqio_mutex);
410 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
411 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
414 /* Make sure flags update is visible after dquot has been filled */
415 smp_mb__before_atomic();
416 set_bit(DQ_READ_B, &dquot->dq_flags);
417 /* Instantiate dquot if needed */
418 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
419 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
420 /* Write the info if needed */
421 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
422 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
423 dquot->dq_sb, dquot->dq_id.type);
433 * Make sure flags update is visible after on-disk struct has been
434 * allocated. Paired with smp_rmb() in dqget().
436 smp_mb__before_atomic();
437 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
439 mutex_unlock(&dqopt->dqio_mutex);
440 mutex_unlock(&dquot->dq_lock);
443 EXPORT_SYMBOL(dquot_acquire);
446 * Write dquot to disk
448 int dquot_commit(struct dquot *dquot)
451 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
453 mutex_lock(&dqopt->dqio_mutex);
454 spin_lock(&dq_list_lock);
455 if (!clear_dquot_dirty(dquot)) {
456 spin_unlock(&dq_list_lock);
459 spin_unlock(&dq_list_lock);
460 /* Inactive dquot can be only if there was error during read/init
461 * => we have better not writing it */
462 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
463 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
467 mutex_unlock(&dqopt->dqio_mutex);
470 EXPORT_SYMBOL(dquot_commit);
475 int dquot_release(struct dquot *dquot)
477 int ret = 0, ret2 = 0;
478 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
480 mutex_lock(&dquot->dq_lock);
481 /* Check whether we are not racing with some other dqget() */
482 if (atomic_read(&dquot->dq_count) > 1)
484 mutex_lock(&dqopt->dqio_mutex);
485 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
486 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
488 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
489 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
490 dquot->dq_sb, dquot->dq_id.type);
495 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
496 mutex_unlock(&dqopt->dqio_mutex);
498 mutex_unlock(&dquot->dq_lock);
501 EXPORT_SYMBOL(dquot_release);
503 void dquot_destroy(struct dquot *dquot)
505 kmem_cache_free(dquot_cachep, dquot);
507 EXPORT_SYMBOL(dquot_destroy);
509 static inline void do_destroy_dquot(struct dquot *dquot)
511 dquot->dq_sb->dq_op->destroy_dquot(dquot);
514 /* Invalidate all dquots on the list. Note that this function is called after
515 * quota is disabled and pointers from inodes removed so there cannot be new
516 * quota users. There can still be some users of quotas due to inodes being
517 * just deleted or pruned by prune_icache() (those are not attached to any
518 * list) or parallel quotactl call. We have to wait for such users.
520 static void invalidate_dquots(struct super_block *sb, int type)
522 struct dquot *dquot, *tmp;
525 spin_lock(&dq_list_lock);
526 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
527 if (dquot->dq_sb != sb)
529 if (dquot->dq_id.type != type)
531 /* Wait for dquot users */
532 if (atomic_read(&dquot->dq_count)) {
536 prepare_to_wait(&dquot->dq_wait_unused, &wait,
537 TASK_UNINTERRUPTIBLE);
538 spin_unlock(&dq_list_lock);
539 /* Once dqput() wakes us up, we know it's time to free
541 * IMPORTANT: we rely on the fact that there is always
542 * at most one process waiting for dquot to free.
543 * Otherwise dq_count would be > 1 and we would never
546 if (atomic_read(&dquot->dq_count) > 1)
548 finish_wait(&dquot->dq_wait_unused, &wait);
550 /* At this moment dquot() need not exist (it could be
551 * reclaimed by prune_dqcache(). Hence we must
556 * Quota now has no users and it has been written on last
559 remove_dquot_hash(dquot);
560 remove_free_dquot(dquot);
562 do_destroy_dquot(dquot);
564 spin_unlock(&dq_list_lock);
567 /* Call callback for every active dquot on given filesystem */
568 int dquot_scan_active(struct super_block *sb,
569 int (*fn)(struct dquot *dquot, unsigned long priv),
572 struct dquot *dquot, *old_dquot = NULL;
575 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
577 spin_lock(&dq_list_lock);
578 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
579 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
581 if (dquot->dq_sb != sb)
583 /* Now we have active dquot so we can just increase use count */
584 atomic_inc(&dquot->dq_count);
585 spin_unlock(&dq_list_lock);
586 dqstats_inc(DQST_LOOKUPS);
590 * ->release_dquot() can be racing with us. Our reference
591 * protects us from new calls to it so just wait for any
592 * outstanding call and recheck the DQ_ACTIVE_B after that.
594 wait_on_dquot(dquot);
595 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
596 ret = fn(dquot, priv);
600 spin_lock(&dq_list_lock);
601 /* We are safe to continue now because our dquot could not
602 * be moved out of the inuse list while we hold the reference */
604 spin_unlock(&dq_list_lock);
609 EXPORT_SYMBOL(dquot_scan_active);
611 /* Write all dquot structures to quota files */
612 int dquot_writeback_dquots(struct super_block *sb, int type)
614 struct list_head *dirty;
616 struct quota_info *dqopt = sb_dqopt(sb);
620 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
622 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
623 if (type != -1 && cnt != type)
625 if (!sb_has_quota_active(sb, cnt))
627 spin_lock(&dq_list_lock);
628 dirty = &dqopt->info[cnt].dqi_dirty_list;
629 while (!list_empty(dirty)) {
630 dquot = list_first_entry(dirty, struct dquot,
632 /* Dirty and inactive can be only bad dquot... */
633 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
634 clear_dquot_dirty(dquot);
637 /* Now we have active dquot from which someone is
638 * holding reference so we can safely just increase
641 spin_unlock(&dq_list_lock);
642 dqstats_inc(DQST_LOOKUPS);
643 err = sb->dq_op->write_dquot(dquot);
647 spin_lock(&dq_list_lock);
649 spin_unlock(&dq_list_lock);
652 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
653 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
654 && info_dirty(&dqopt->info[cnt]))
655 sb->dq_op->write_info(sb, cnt);
656 dqstats_inc(DQST_SYNCS);
660 EXPORT_SYMBOL(dquot_writeback_dquots);
662 /* Write all dquot structures to disk and make them visible from userspace */
663 int dquot_quota_sync(struct super_block *sb, int type)
665 struct quota_info *dqopt = sb_dqopt(sb);
669 ret = dquot_writeback_dquots(sb, type);
672 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
675 /* This is not very clever (and fast) but currently I don't know about
676 * any other simple way of getting quota data to disk and we must get
677 * them there for userspace to be visible... */
678 if (sb->s_op->sync_fs)
679 sb->s_op->sync_fs(sb, 1);
680 sync_blockdev(sb->s_bdev);
683 * Now when everything is written we can discard the pagecache so
684 * that userspace sees the changes.
686 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
687 if (type != -1 && cnt != type)
689 if (!sb_has_quota_active(sb, cnt))
691 inode_lock(dqopt->files[cnt]);
692 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
693 inode_unlock(dqopt->files[cnt]);
698 EXPORT_SYMBOL(dquot_quota_sync);
701 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
703 struct list_head *head;
705 unsigned long freed = 0;
707 spin_lock(&dq_list_lock);
708 head = free_dquots.prev;
709 while (head != &free_dquots && sc->nr_to_scan) {
710 dquot = list_entry(head, struct dquot, dq_free);
711 remove_dquot_hash(dquot);
712 remove_free_dquot(dquot);
714 do_destroy_dquot(dquot);
717 head = free_dquots.prev;
719 spin_unlock(&dq_list_lock);
724 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
726 return vfs_pressure_ratio(
727 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
730 static struct shrinker dqcache_shrinker = {
731 .count_objects = dqcache_shrink_count,
732 .scan_objects = dqcache_shrink_scan,
733 .seeks = DEFAULT_SEEKS,
737 * Put reference to dquot
739 void dqput(struct dquot *dquot)
745 #ifdef CONFIG_QUOTA_DEBUG
746 if (!atomic_read(&dquot->dq_count)) {
747 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
748 quotatypes[dquot->dq_id.type],
749 from_kqid(&init_user_ns, dquot->dq_id));
753 dqstats_inc(DQST_DROPS);
755 spin_lock(&dq_list_lock);
756 if (atomic_read(&dquot->dq_count) > 1) {
757 /* We have more than one user... nothing to do */
758 atomic_dec(&dquot->dq_count);
759 /* Releasing dquot during quotaoff phase? */
760 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
761 atomic_read(&dquot->dq_count) == 1)
762 wake_up(&dquot->dq_wait_unused);
763 spin_unlock(&dq_list_lock);
766 /* Need to release dquot? */
767 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
768 spin_unlock(&dq_list_lock);
769 /* Commit dquot before releasing */
770 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
772 quota_error(dquot->dq_sb, "Can't write quota structure"
773 " (error %d). Quota may get out of sync!",
776 * We clear dirty bit anyway, so that we avoid
779 spin_lock(&dq_list_lock);
780 clear_dquot_dirty(dquot);
781 spin_unlock(&dq_list_lock);
785 /* Clear flag in case dquot was inactive (something bad happened) */
786 clear_dquot_dirty(dquot);
787 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
788 spin_unlock(&dq_list_lock);
789 dquot->dq_sb->dq_op->release_dquot(dquot);
792 atomic_dec(&dquot->dq_count);
793 #ifdef CONFIG_QUOTA_DEBUG
795 BUG_ON(!list_empty(&dquot->dq_free));
797 put_dquot_last(dquot);
798 spin_unlock(&dq_list_lock);
800 EXPORT_SYMBOL(dqput);
802 struct dquot *dquot_alloc(struct super_block *sb, int type)
804 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
806 EXPORT_SYMBOL(dquot_alloc);
808 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
812 dquot = sb->dq_op->alloc_dquot(sb, type);
816 mutex_init(&dquot->dq_lock);
817 INIT_LIST_HEAD(&dquot->dq_free);
818 INIT_LIST_HEAD(&dquot->dq_inuse);
819 INIT_HLIST_NODE(&dquot->dq_hash);
820 INIT_LIST_HEAD(&dquot->dq_dirty);
821 init_waitqueue_head(&dquot->dq_wait_unused);
823 dquot->dq_id = make_kqid_invalid(type);
824 atomic_set(&dquot->dq_count, 1);
830 * Get reference to dquot
832 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
833 * destroying our dquot by:
834 * a) checking for quota flags under dq_list_lock and
835 * b) getting a reference to dquot before we release dq_list_lock
837 struct dquot *dqget(struct super_block *sb, struct kqid qid)
839 unsigned int hashent = hashfn(sb, qid);
840 struct dquot *dquot, *empty = NULL;
842 if (!qid_has_mapping(sb->s_user_ns, qid))
843 return ERR_PTR(-EINVAL);
845 if (!sb_has_quota_active(sb, qid.type))
846 return ERR_PTR(-ESRCH);
848 spin_lock(&dq_list_lock);
849 spin_lock(&dq_state_lock);
850 if (!sb_has_quota_active(sb, qid.type)) {
851 spin_unlock(&dq_state_lock);
852 spin_unlock(&dq_list_lock);
853 dquot = ERR_PTR(-ESRCH);
856 spin_unlock(&dq_state_lock);
858 dquot = find_dquot(hashent, sb, qid);
861 spin_unlock(&dq_list_lock);
862 empty = get_empty_dquot(sb, qid.type);
864 schedule(); /* Try to wait for a moment... */
870 /* all dquots go on the inuse_list */
872 /* hash it first so it can be found */
873 insert_dquot_hash(dquot);
874 spin_unlock(&dq_list_lock);
875 dqstats_inc(DQST_LOOKUPS);
877 if (!atomic_read(&dquot->dq_count))
878 remove_free_dquot(dquot);
879 atomic_inc(&dquot->dq_count);
880 spin_unlock(&dq_list_lock);
881 dqstats_inc(DQST_CACHE_HITS);
882 dqstats_inc(DQST_LOOKUPS);
884 /* Wait for dq_lock - after this we know that either dquot_release() is
885 * already finished or it will be canceled due to dq_count > 1 test */
886 wait_on_dquot(dquot);
887 /* Read the dquot / allocate space in quota file */
888 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
891 err = sb->dq_op->acquire_dquot(dquot);
894 dquot = ERR_PTR(err);
899 * Make sure following reads see filled structure - paired with
900 * smp_mb__before_atomic() in dquot_acquire().
903 #ifdef CONFIG_QUOTA_DEBUG
904 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
908 do_destroy_dquot(empty);
912 EXPORT_SYMBOL(dqget);
914 static inline struct dquot **i_dquot(struct inode *inode)
916 return inode->i_sb->s_op->get_dquots(inode);
919 static int dqinit_needed(struct inode *inode, int type)
921 struct dquot * const *dquots;
924 if (IS_NOQUOTA(inode))
927 dquots = i_dquot(inode);
929 return !dquots[type];
930 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
936 /* This routine is guarded by dqonoff_mutex mutex */
937 static void add_dquot_ref(struct super_block *sb, int type)
939 struct inode *inode, *old_inode = NULL;
940 #ifdef CONFIG_QUOTA_DEBUG
944 spin_lock(&sb->s_inode_list_lock);
945 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
946 spin_lock(&inode->i_lock);
947 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
948 !atomic_read(&inode->i_writecount) ||
949 !dqinit_needed(inode, type)) {
950 spin_unlock(&inode->i_lock);
954 spin_unlock(&inode->i_lock);
955 spin_unlock(&sb->s_inode_list_lock);
957 #ifdef CONFIG_QUOTA_DEBUG
958 if (unlikely(inode_get_rsv_space(inode) > 0))
962 __dquot_initialize(inode, type);
965 * We hold a reference to 'inode' so it couldn't have been
966 * removed from s_inodes list while we dropped the
967 * s_inode_list_lock. We cannot iput the inode now as we can be
968 * holding the last reference and we cannot iput it under
969 * s_inode_list_lock. So we keep the reference and iput it
973 spin_lock(&sb->s_inode_list_lock);
975 spin_unlock(&sb->s_inode_list_lock);
978 #ifdef CONFIG_QUOTA_DEBUG
980 quota_error(sb, "Writes happened before quota was turned on "
981 "thus quota information is probably inconsistent. "
982 "Please run quotacheck(8)");
988 * Remove references to dquots from inode and add dquot to list for freeing
989 * if we have the last reference to dquot
991 static void remove_inode_dquot_ref(struct inode *inode, int type,
992 struct list_head *tofree_head)
994 struct dquot **dquots = i_dquot(inode);
995 struct dquot *dquot = dquots[type];
1000 dquots[type] = NULL;
1001 if (list_empty(&dquot->dq_free)) {
1003 * The inode still has reference to dquot so it can't be in the
1006 spin_lock(&dq_list_lock);
1007 list_add(&dquot->dq_free, tofree_head);
1008 spin_unlock(&dq_list_lock);
1011 * Dquot is already in a list to put so we won't drop the last
1019 * Free list of dquots
1020 * Dquots are removed from inodes and no new references can be got so we are
1021 * the only ones holding reference
1023 static void put_dquot_list(struct list_head *tofree_head)
1025 struct list_head *act_head;
1026 struct dquot *dquot;
1028 act_head = tofree_head->next;
1029 while (act_head != tofree_head) {
1030 dquot = list_entry(act_head, struct dquot, dq_free);
1031 act_head = act_head->next;
1032 /* Remove dquot from the list so we won't have problems... */
1033 list_del_init(&dquot->dq_free);
1038 static void remove_dquot_ref(struct super_block *sb, int type,
1039 struct list_head *tofree_head)
1041 struct inode *inode;
1044 spin_lock(&sb->s_inode_list_lock);
1045 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1047 * We have to scan also I_NEW inodes because they can already
1048 * have quota pointer initialized. Luckily, we need to touch
1049 * only quota pointers and these have separate locking
1052 spin_lock(&dq_data_lock);
1053 if (!IS_NOQUOTA(inode)) {
1054 if (unlikely(inode_get_rsv_space(inode) > 0))
1056 remove_inode_dquot_ref(inode, type, tofree_head);
1058 spin_unlock(&dq_data_lock);
1060 spin_unlock(&sb->s_inode_list_lock);
1061 #ifdef CONFIG_QUOTA_DEBUG
1063 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1064 " was disabled thus quota information is probably "
1065 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1070 /* Gather all references from inodes and drop them */
1071 static void drop_dquot_ref(struct super_block *sb, int type)
1073 LIST_HEAD(tofree_head);
1076 remove_dquot_ref(sb, type, &tofree_head);
1077 synchronize_srcu(&dquot_srcu);
1078 put_dquot_list(&tofree_head);
1082 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1084 dquot->dq_dqb.dqb_curinodes += number;
1087 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1089 dquot->dq_dqb.dqb_curspace += number;
1092 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1094 dquot->dq_dqb.dqb_rsvspace += number;
1098 * Claim reserved quota space
1100 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1102 if (dquot->dq_dqb.dqb_rsvspace < number) {
1104 number = dquot->dq_dqb.dqb_rsvspace;
1106 dquot->dq_dqb.dqb_curspace += number;
1107 dquot->dq_dqb.dqb_rsvspace -= number;
1110 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1112 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1113 number = dquot->dq_dqb.dqb_curspace;
1114 dquot->dq_dqb.dqb_rsvspace += number;
1115 dquot->dq_dqb.dqb_curspace -= number;
1119 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1121 if (dquot->dq_dqb.dqb_rsvspace >= number)
1122 dquot->dq_dqb.dqb_rsvspace -= number;
1125 dquot->dq_dqb.dqb_rsvspace = 0;
1129 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1131 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1132 dquot->dq_dqb.dqb_curinodes >= number)
1133 dquot->dq_dqb.dqb_curinodes -= number;
1135 dquot->dq_dqb.dqb_curinodes = 0;
1136 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1137 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1138 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1141 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1143 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1144 dquot->dq_dqb.dqb_curspace >= number)
1145 dquot->dq_dqb.dqb_curspace -= number;
1147 dquot->dq_dqb.dqb_curspace = 0;
1148 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1149 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1150 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1154 struct super_block *w_sb;
1155 struct kqid w_dq_id;
1159 static int warning_issued(struct dquot *dquot, const int warntype)
1161 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1162 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1163 ((warntype == QUOTA_NL_IHARDWARN ||
1164 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1168 return test_and_set_bit(flag, &dquot->dq_flags);
1171 #ifdef CONFIG_PRINT_QUOTA_WARNING
1172 static int flag_print_warnings = 1;
1174 static int need_print_warning(struct dquot_warn *warn)
1176 if (!flag_print_warnings)
1179 switch (warn->w_dq_id.type) {
1181 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1183 return in_group_p(warn->w_dq_id.gid);
1190 /* Print warning to user which exceeded quota */
1191 static void print_warning(struct dquot_warn *warn)
1194 struct tty_struct *tty;
1195 int warntype = warn->w_type;
1197 if (warntype == QUOTA_NL_IHARDBELOW ||
1198 warntype == QUOTA_NL_ISOFTBELOW ||
1199 warntype == QUOTA_NL_BHARDBELOW ||
1200 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1203 tty = get_current_tty();
1206 tty_write_message(tty, warn->w_sb->s_id);
1207 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1208 tty_write_message(tty, ": warning, ");
1210 tty_write_message(tty, ": write failed, ");
1211 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1213 case QUOTA_NL_IHARDWARN:
1214 msg = " file limit reached.\r\n";
1216 case QUOTA_NL_ISOFTLONGWARN:
1217 msg = " file quota exceeded too long.\r\n";
1219 case QUOTA_NL_ISOFTWARN:
1220 msg = " file quota exceeded.\r\n";
1222 case QUOTA_NL_BHARDWARN:
1223 msg = " block limit reached.\r\n";
1225 case QUOTA_NL_BSOFTLONGWARN:
1226 msg = " block quota exceeded too long.\r\n";
1228 case QUOTA_NL_BSOFTWARN:
1229 msg = " block quota exceeded.\r\n";
1232 tty_write_message(tty, msg);
1237 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1240 if (warning_issued(dquot, warntype))
1242 warn->w_type = warntype;
1243 warn->w_sb = dquot->dq_sb;
1244 warn->w_dq_id = dquot->dq_id;
1248 * Write warnings to the console and send warning messages over netlink.
1250 * Note that this function can call into tty and networking code.
1252 static void flush_warnings(struct dquot_warn *warn)
1256 for (i = 0; i < MAXQUOTAS; i++) {
1257 if (warn[i].w_type == QUOTA_NL_NOWARN)
1259 #ifdef CONFIG_PRINT_QUOTA_WARNING
1260 print_warning(&warn[i]);
1262 quota_send_warning(warn[i].w_dq_id,
1263 warn[i].w_sb->s_dev, warn[i].w_type);
1267 static int ignore_hardlimit(struct dquot *dquot)
1269 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1271 return capable(CAP_SYS_RESOURCE) &&
1272 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1273 !(info->dqi_flags & DQF_ROOT_SQUASH));
1276 /* needs dq_data_lock */
1277 static int check_idq(struct dquot *dquot, qsize_t inodes,
1278 struct dquot_warn *warn)
1280 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1282 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1283 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1286 if (dquot->dq_dqb.dqb_ihardlimit &&
1287 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1288 !ignore_hardlimit(dquot)) {
1289 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1293 if (dquot->dq_dqb.dqb_isoftlimit &&
1294 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1295 dquot->dq_dqb.dqb_itime &&
1296 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1297 !ignore_hardlimit(dquot)) {
1298 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1302 if (dquot->dq_dqb.dqb_isoftlimit &&
1303 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1304 dquot->dq_dqb.dqb_itime == 0) {
1305 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1306 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1307 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1313 /* needs dq_data_lock */
1314 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1315 struct dquot_warn *warn)
1318 struct super_block *sb = dquot->dq_sb;
1320 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1321 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1324 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1327 if (dquot->dq_dqb.dqb_bhardlimit &&
1328 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1329 !ignore_hardlimit(dquot)) {
1331 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1335 if (dquot->dq_dqb.dqb_bsoftlimit &&
1336 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1337 dquot->dq_dqb.dqb_btime &&
1338 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1339 !ignore_hardlimit(dquot)) {
1341 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1345 if (dquot->dq_dqb.dqb_bsoftlimit &&
1346 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1347 dquot->dq_dqb.dqb_btime == 0) {
1349 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1350 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1351 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1355 * We don't allow preallocation to exceed softlimit so exceeding will
1364 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1368 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1369 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1370 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1371 return QUOTA_NL_NOWARN;
1373 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1374 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1375 return QUOTA_NL_ISOFTBELOW;
1376 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1377 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1378 return QUOTA_NL_IHARDBELOW;
1379 return QUOTA_NL_NOWARN;
1382 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1384 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1385 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1386 return QUOTA_NL_NOWARN;
1388 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1389 return QUOTA_NL_BSOFTBELOW;
1390 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1391 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1392 return QUOTA_NL_BHARDBELOW;
1393 return QUOTA_NL_NOWARN;
1396 static int dquot_active(const struct inode *inode)
1398 struct super_block *sb = inode->i_sb;
1400 if (IS_NOQUOTA(inode))
1402 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1406 * Initialize quota pointers in inode
1408 * It is better to call this function outside of any transaction as it
1409 * might need a lot of space in journal for dquot structure allocation.
1411 static int __dquot_initialize(struct inode *inode, int type)
1413 int cnt, init_needed = 0;
1414 struct dquot **dquots, *got[MAXQUOTAS] = {};
1415 struct super_block *sb = inode->i_sb;
1419 if (!dquot_active(inode))
1422 dquots = i_dquot(inode);
1424 /* First get references to structures we might need. */
1425 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1429 struct dquot *dquot;
1431 if (type != -1 && cnt != type)
1434 * The i_dquot should have been initialized in most cases,
1435 * we check it without locking here to avoid unnecessary
1436 * dqget()/dqput() calls.
1441 if (!sb_has_quota_active(sb, cnt))
1448 qid = make_kqid_uid(inode->i_uid);
1451 qid = make_kqid_gid(inode->i_gid);
1454 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1457 qid = make_kqid_projid(projid);
1460 dquot = dqget(sb, qid);
1461 if (IS_ERR(dquot)) {
1462 /* We raced with somebody turning quotas off... */
1463 if (PTR_ERR(dquot) != -ESRCH) {
1464 ret = PTR_ERR(dquot);
1472 /* All required i_dquot has been initialized */
1476 spin_lock(&dq_data_lock);
1477 if (IS_NOQUOTA(inode))
1479 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1480 if (type != -1 && cnt != type)
1482 /* Avoid races with quotaoff() */
1483 if (!sb_has_quota_active(sb, cnt))
1485 /* We could race with quotaon or dqget() could have failed */
1489 dquots[cnt] = got[cnt];
1492 * Make quota reservation system happy if someone
1493 * did a write before quota was turned on
1495 rsv = inode_get_rsv_space(inode);
1497 dquot_resv_space(dquots[cnt], rsv);
1501 spin_unlock(&dq_data_lock);
1503 /* Drop unused references */
1509 int dquot_initialize(struct inode *inode)
1511 return __dquot_initialize(inode, -1);
1513 EXPORT_SYMBOL(dquot_initialize);
1516 * Release all quotas referenced by inode.
1518 * This function only be called on inode free or converting
1519 * a file to quota file, no other users for the i_dquot in
1520 * both cases, so we needn't call synchronize_srcu() after
1523 static void __dquot_drop(struct inode *inode)
1526 struct dquot **dquots = i_dquot(inode);
1527 struct dquot *put[MAXQUOTAS];
1529 spin_lock(&dq_data_lock);
1530 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1531 put[cnt] = dquots[cnt];
1534 spin_unlock(&dq_data_lock);
1538 void dquot_drop(struct inode *inode)
1540 struct dquot * const *dquots;
1543 if (IS_NOQUOTA(inode))
1547 * Test before calling to rule out calls from proc and such
1548 * where we are not allowed to block. Note that this is
1549 * actually reliable test even without the lock - the caller
1550 * must assure that nobody can come after the DQUOT_DROP and
1551 * add quota pointers back anyway.
1553 dquots = i_dquot(inode);
1554 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1559 if (cnt < MAXQUOTAS)
1560 __dquot_drop(inode);
1562 EXPORT_SYMBOL(dquot_drop);
1565 * inode_reserved_space is managed internally by quota, and protected by
1566 * i_lock similar to i_blocks+i_bytes.
1568 static qsize_t *inode_reserved_space(struct inode * inode)
1570 /* Filesystem must explicitly define it's own method in order to use
1571 * quota reservation interface */
1572 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1573 return inode->i_sb->dq_op->get_reserved_space(inode);
1576 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1578 spin_lock(&inode->i_lock);
1579 *inode_reserved_space(inode) += number;
1580 spin_unlock(&inode->i_lock);
1582 EXPORT_SYMBOL(inode_add_rsv_space);
1584 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1586 spin_lock(&inode->i_lock);
1587 *inode_reserved_space(inode) -= number;
1588 __inode_add_bytes(inode, number);
1589 spin_unlock(&inode->i_lock);
1591 EXPORT_SYMBOL(inode_claim_rsv_space);
1593 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1595 spin_lock(&inode->i_lock);
1596 *inode_reserved_space(inode) += number;
1597 __inode_sub_bytes(inode, number);
1598 spin_unlock(&inode->i_lock);
1600 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1602 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1604 spin_lock(&inode->i_lock);
1605 *inode_reserved_space(inode) -= number;
1606 spin_unlock(&inode->i_lock);
1608 EXPORT_SYMBOL(inode_sub_rsv_space);
1610 static qsize_t inode_get_rsv_space(struct inode *inode)
1614 if (!inode->i_sb->dq_op->get_reserved_space)
1616 spin_lock(&inode->i_lock);
1617 ret = *inode_reserved_space(inode);
1618 spin_unlock(&inode->i_lock);
1622 static void inode_incr_space(struct inode *inode, qsize_t number,
1626 inode_add_rsv_space(inode, number);
1628 inode_add_bytes(inode, number);
1631 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1634 inode_sub_rsv_space(inode, number);
1636 inode_sub_bytes(inode, number);
1640 * This functions updates i_blocks+i_bytes fields and quota information
1641 * (together with appropriate checks).
1643 * NOTE: We absolutely rely on the fact that caller dirties the inode
1644 * (usually helpers in quotaops.h care about this) and holds a handle for
1645 * the current transaction so that dquot write and inode write go into the
1650 * This operation can block, but only after everything is updated
1652 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1654 int cnt, ret = 0, index;
1655 struct dquot_warn warn[MAXQUOTAS];
1656 int reserve = flags & DQUOT_SPACE_RESERVE;
1657 struct dquot **dquots;
1659 if (!dquot_active(inode)) {
1660 inode_incr_space(inode, number, reserve);
1664 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1665 warn[cnt].w_type = QUOTA_NL_NOWARN;
1667 dquots = i_dquot(inode);
1668 index = srcu_read_lock(&dquot_srcu);
1669 spin_lock(&dq_data_lock);
1670 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1673 ret = check_bdq(dquots[cnt], number,
1674 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1675 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1676 spin_unlock(&dq_data_lock);
1677 goto out_flush_warn;
1680 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1684 dquot_resv_space(dquots[cnt], number);
1686 dquot_incr_space(dquots[cnt], number);
1688 inode_incr_space(inode, number, reserve);
1689 spin_unlock(&dq_data_lock);
1692 goto out_flush_warn;
1693 mark_all_dquot_dirty(dquots);
1695 srcu_read_unlock(&dquot_srcu, index);
1696 flush_warnings(warn);
1700 EXPORT_SYMBOL(__dquot_alloc_space);
1703 * This operation can block, but only after everything is updated
1705 int dquot_alloc_inode(struct inode *inode)
1707 int cnt, ret = 0, index;
1708 struct dquot_warn warn[MAXQUOTAS];
1709 struct dquot * const *dquots;
1711 if (!dquot_active(inode))
1713 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1714 warn[cnt].w_type = QUOTA_NL_NOWARN;
1716 dquots = i_dquot(inode);
1717 index = srcu_read_lock(&dquot_srcu);
1718 spin_lock(&dq_data_lock);
1719 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1722 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1727 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1730 dquot_incr_inodes(dquots[cnt], 1);
1734 spin_unlock(&dq_data_lock);
1736 mark_all_dquot_dirty(dquots);
1737 srcu_read_unlock(&dquot_srcu, index);
1738 flush_warnings(warn);
1741 EXPORT_SYMBOL(dquot_alloc_inode);
1744 * Convert in-memory reserved quotas to real consumed quotas
1746 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1748 struct dquot **dquots;
1751 if (!dquot_active(inode)) {
1752 inode_claim_rsv_space(inode, number);
1756 dquots = i_dquot(inode);
1757 index = srcu_read_lock(&dquot_srcu);
1758 spin_lock(&dq_data_lock);
1759 /* Claim reserved quotas to allocated quotas */
1760 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1762 dquot_claim_reserved_space(dquots[cnt], number);
1764 /* Update inode bytes */
1765 inode_claim_rsv_space(inode, number);
1766 spin_unlock(&dq_data_lock);
1767 mark_all_dquot_dirty(dquots);
1768 srcu_read_unlock(&dquot_srcu, index);
1771 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1774 * Convert allocated space back to in-memory reserved quotas
1776 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1778 struct dquot **dquots;
1781 if (!dquot_active(inode)) {
1782 inode_reclaim_rsv_space(inode, number);
1786 dquots = i_dquot(inode);
1787 index = srcu_read_lock(&dquot_srcu);
1788 spin_lock(&dq_data_lock);
1789 /* Claim reserved quotas to allocated quotas */
1790 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1792 dquot_reclaim_reserved_space(dquots[cnt], number);
1794 /* Update inode bytes */
1795 inode_reclaim_rsv_space(inode, number);
1796 spin_unlock(&dq_data_lock);
1797 mark_all_dquot_dirty(dquots);
1798 srcu_read_unlock(&dquot_srcu, index);
1801 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1804 * This operation can block, but only after everything is updated
1806 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1809 struct dquot_warn warn[MAXQUOTAS];
1810 struct dquot **dquots;
1811 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1813 if (!dquot_active(inode)) {
1814 inode_decr_space(inode, number, reserve);
1818 dquots = i_dquot(inode);
1819 index = srcu_read_lock(&dquot_srcu);
1820 spin_lock(&dq_data_lock);
1821 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1824 warn[cnt].w_type = QUOTA_NL_NOWARN;
1827 wtype = info_bdq_free(dquots[cnt], number);
1828 if (wtype != QUOTA_NL_NOWARN)
1829 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1831 dquot_free_reserved_space(dquots[cnt], number);
1833 dquot_decr_space(dquots[cnt], number);
1835 inode_decr_space(inode, number, reserve);
1836 spin_unlock(&dq_data_lock);
1840 mark_all_dquot_dirty(dquots);
1842 srcu_read_unlock(&dquot_srcu, index);
1843 flush_warnings(warn);
1845 EXPORT_SYMBOL(__dquot_free_space);
1848 * This operation can block, but only after everything is updated
1850 void dquot_free_inode(struct inode *inode)
1853 struct dquot_warn warn[MAXQUOTAS];
1854 struct dquot * const *dquots;
1857 if (!dquot_active(inode))
1860 dquots = i_dquot(inode);
1861 index = srcu_read_lock(&dquot_srcu);
1862 spin_lock(&dq_data_lock);
1863 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1866 warn[cnt].w_type = QUOTA_NL_NOWARN;
1869 wtype = info_idq_free(dquots[cnt], 1);
1870 if (wtype != QUOTA_NL_NOWARN)
1871 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1872 dquot_decr_inodes(dquots[cnt], 1);
1874 spin_unlock(&dq_data_lock);
1875 mark_all_dquot_dirty(dquots);
1876 srcu_read_unlock(&dquot_srcu, index);
1877 flush_warnings(warn);
1879 EXPORT_SYMBOL(dquot_free_inode);
1882 * Transfer the number of inode and blocks from one diskquota to an other.
1883 * On success, dquot references in transfer_to are consumed and references
1884 * to original dquots that need to be released are placed there. On failure,
1885 * references are kept untouched.
1887 * This operation can block, but only after everything is updated
1888 * A transaction must be started when entering this function.
1890 * We are holding reference on transfer_from & transfer_to, no need to
1891 * protect them by srcu_read_lock().
1893 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1895 qsize_t space, cur_space;
1896 qsize_t rsv_space = 0;
1897 struct dquot *transfer_from[MAXQUOTAS] = {};
1899 char is_valid[MAXQUOTAS] = {};
1900 struct dquot_warn warn_to[MAXQUOTAS];
1901 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1902 struct dquot_warn warn_from_space[MAXQUOTAS];
1904 if (IS_NOQUOTA(inode))
1906 /* Initialize the arrays */
1907 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1908 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1909 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1910 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1913 spin_lock(&dq_data_lock);
1914 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1915 spin_unlock(&dq_data_lock);
1918 cur_space = inode_get_bytes(inode);
1919 rsv_space = inode_get_rsv_space(inode);
1920 space = cur_space + rsv_space;
1921 /* Build the transfer_from list and check the limits */
1922 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1924 * Skip changes for same uid or gid or for turned off quota-type.
1926 if (!transfer_to[cnt])
1928 /* Avoid races with quotaoff() */
1929 if (!sb_has_quota_active(inode->i_sb, cnt))
1932 transfer_from[cnt] = i_dquot(inode)[cnt];
1933 ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1936 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1942 * Finally perform the needed transfer from transfer_from to transfer_to
1944 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1947 /* Due to IO error we might not have transfer_from[] structure */
1948 if (transfer_from[cnt]) {
1950 wtype = info_idq_free(transfer_from[cnt], 1);
1951 if (wtype != QUOTA_NL_NOWARN)
1952 prepare_warning(&warn_from_inodes[cnt],
1953 transfer_from[cnt], wtype);
1954 wtype = info_bdq_free(transfer_from[cnt], space);
1955 if (wtype != QUOTA_NL_NOWARN)
1956 prepare_warning(&warn_from_space[cnt],
1957 transfer_from[cnt], wtype);
1958 dquot_decr_inodes(transfer_from[cnt], 1);
1959 dquot_decr_space(transfer_from[cnt], cur_space);
1960 dquot_free_reserved_space(transfer_from[cnt],
1964 dquot_incr_inodes(transfer_to[cnt], 1);
1965 dquot_incr_space(transfer_to[cnt], cur_space);
1966 dquot_resv_space(transfer_to[cnt], rsv_space);
1968 i_dquot(inode)[cnt] = transfer_to[cnt];
1970 spin_unlock(&dq_data_lock);
1972 mark_all_dquot_dirty(transfer_from);
1973 mark_all_dquot_dirty(transfer_to);
1974 flush_warnings(warn_to);
1975 flush_warnings(warn_from_inodes);
1976 flush_warnings(warn_from_space);
1977 /* Pass back references to put */
1978 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1980 transfer_to[cnt] = transfer_from[cnt];
1983 spin_unlock(&dq_data_lock);
1984 flush_warnings(warn_to);
1987 EXPORT_SYMBOL(__dquot_transfer);
1989 /* Wrapper for transferring ownership of an inode for uid/gid only
1990 * Called from FSXXX_setattr()
1992 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1994 struct dquot *transfer_to[MAXQUOTAS] = {};
1995 struct dquot *dquot;
1996 struct super_block *sb = inode->i_sb;
1999 if (!dquot_active(inode))
2002 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2003 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2004 if (IS_ERR(dquot)) {
2005 if (PTR_ERR(dquot) != -ESRCH) {
2006 ret = PTR_ERR(dquot);
2011 transfer_to[USRQUOTA] = dquot;
2013 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2014 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2015 if (IS_ERR(dquot)) {
2016 if (PTR_ERR(dquot) != -ESRCH) {
2017 ret = PTR_ERR(dquot);
2022 transfer_to[GRPQUOTA] = dquot;
2024 ret = __dquot_transfer(inode, transfer_to);
2026 dqput_all(transfer_to);
2029 EXPORT_SYMBOL(dquot_transfer);
2032 * Write info of quota file to disk
2034 int dquot_commit_info(struct super_block *sb, int type)
2037 struct quota_info *dqopt = sb_dqopt(sb);
2039 mutex_lock(&dqopt->dqio_mutex);
2040 ret = dqopt->ops[type]->write_file_info(sb, type);
2041 mutex_unlock(&dqopt->dqio_mutex);
2044 EXPORT_SYMBOL(dquot_commit_info);
2046 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2048 struct quota_info *dqopt = sb_dqopt(sb);
2051 if (!sb_has_quota_active(sb, qid->type))
2053 if (!dqopt->ops[qid->type]->get_next_id)
2055 mutex_lock(&dqopt->dqio_mutex);
2056 err = dqopt->ops[qid->type]->get_next_id(sb, qid);
2057 mutex_unlock(&dqopt->dqio_mutex);
2060 EXPORT_SYMBOL(dquot_get_next_id);
2063 * Definitions of diskquota operations.
2065 const struct dquot_operations dquot_operations = {
2066 .write_dquot = dquot_commit,
2067 .acquire_dquot = dquot_acquire,
2068 .release_dquot = dquot_release,
2069 .mark_dirty = dquot_mark_dquot_dirty,
2070 .write_info = dquot_commit_info,
2071 .alloc_dquot = dquot_alloc,
2072 .destroy_dquot = dquot_destroy,
2073 .get_next_id = dquot_get_next_id,
2075 EXPORT_SYMBOL(dquot_operations);
2078 * Generic helper for ->open on filesystems supporting disk quotas.
2080 int dquot_file_open(struct inode *inode, struct file *file)
2084 error = generic_file_open(inode, file);
2085 if (!error && (file->f_mode & FMODE_WRITE))
2086 dquot_initialize(inode);
2089 EXPORT_SYMBOL(dquot_file_open);
2092 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2094 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2097 struct quota_info *dqopt = sb_dqopt(sb);
2098 struct inode *toputinode[MAXQUOTAS];
2100 /* s_umount should be held in exclusive mode */
2101 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2102 up_read(&sb->s_umount);
2104 /* Cannot turn off usage accounting without turning off limits, or
2105 * suspend quotas and simultaneously turn quotas off. */
2106 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2107 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2108 DQUOT_USAGE_ENABLED)))
2111 /* We need to serialize quota_off() for device */
2112 mutex_lock(&dqopt->dqonoff_mutex);
2115 * Skip everything if there's nothing to do. We have to do this because
2116 * sometimes we are called when fill_super() failed and calling
2117 * sync_fs() in such cases does no good.
2119 if (!sb_any_quota_loaded(sb)) {
2120 mutex_unlock(&dqopt->dqonoff_mutex);
2123 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2124 toputinode[cnt] = NULL;
2125 if (type != -1 && cnt != type)
2127 if (!sb_has_quota_loaded(sb, cnt))
2130 if (flags & DQUOT_SUSPENDED) {
2131 spin_lock(&dq_state_lock);
2133 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2134 spin_unlock(&dq_state_lock);
2136 spin_lock(&dq_state_lock);
2137 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2138 /* Turning off suspended quotas? */
2139 if (!sb_has_quota_loaded(sb, cnt) &&
2140 sb_has_quota_suspended(sb, cnt)) {
2141 dqopt->flags &= ~dquot_state_flag(
2142 DQUOT_SUSPENDED, cnt);
2143 spin_unlock(&dq_state_lock);
2144 iput(dqopt->files[cnt]);
2145 dqopt->files[cnt] = NULL;
2148 spin_unlock(&dq_state_lock);
2151 /* We still have to keep quota loaded? */
2152 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2155 /* Note: these are blocking operations */
2156 drop_dquot_ref(sb, cnt);
2157 invalidate_dquots(sb, cnt);
2159 * Now all dquots should be invalidated, all writes done so we
2160 * should be only users of the info. No locks needed.
2162 if (info_dirty(&dqopt->info[cnt]))
2163 sb->dq_op->write_info(sb, cnt);
2164 if (dqopt->ops[cnt]->free_file_info)
2165 dqopt->ops[cnt]->free_file_info(sb, cnt);
2166 put_quota_format(dqopt->info[cnt].dqi_format);
2168 toputinode[cnt] = dqopt->files[cnt];
2169 if (!sb_has_quota_loaded(sb, cnt))
2170 dqopt->files[cnt] = NULL;
2171 dqopt->info[cnt].dqi_flags = 0;
2172 dqopt->info[cnt].dqi_igrace = 0;
2173 dqopt->info[cnt].dqi_bgrace = 0;
2174 dqopt->ops[cnt] = NULL;
2176 mutex_unlock(&dqopt->dqonoff_mutex);
2178 /* Skip syncing and setting flags if quota files are hidden */
2179 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2182 /* Sync the superblock so that buffers with quota data are written to
2183 * disk (and so userspace sees correct data afterwards). */
2184 if (sb->s_op->sync_fs)
2185 sb->s_op->sync_fs(sb, 1);
2186 sync_blockdev(sb->s_bdev);
2187 /* Now the quota files are just ordinary files and we can set the
2188 * inode flags back. Moreover we discard the pagecache so that
2189 * userspace sees the writes we did bypassing the pagecache. We
2190 * must also discard the blockdev buffers so that we see the
2191 * changes done by userspace on the next quotaon() */
2192 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2193 if (toputinode[cnt]) {
2194 mutex_lock(&dqopt->dqonoff_mutex);
2195 /* If quota was reenabled in the meantime, we have
2197 if (!sb_has_quota_loaded(sb, cnt)) {
2198 inode_lock(toputinode[cnt]);
2199 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2200 S_NOATIME | S_NOQUOTA);
2201 truncate_inode_pages(&toputinode[cnt]->i_data,
2203 inode_unlock(toputinode[cnt]);
2204 mark_inode_dirty_sync(toputinode[cnt]);
2206 mutex_unlock(&dqopt->dqonoff_mutex);
2209 invalidate_bdev(sb->s_bdev);
2211 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2212 if (toputinode[cnt]) {
2213 /* On remount RO, we keep the inode pointer so that we
2214 * can reenable quota on the subsequent remount RW. We
2215 * have to check 'flags' variable and not use sb_has_
2216 * function because another quotaon / quotaoff could
2217 * change global state before we got here. We refuse
2218 * to suspend quotas when there is pending delete on
2219 * the quota file... */
2220 if (!(flags & DQUOT_SUSPENDED))
2221 iput(toputinode[cnt]);
2222 else if (!toputinode[cnt]->i_nlink)
2227 EXPORT_SYMBOL(dquot_disable);
2229 int dquot_quota_off(struct super_block *sb, int type)
2231 return dquot_disable(sb, type,
2232 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2234 EXPORT_SYMBOL(dquot_quota_off);
2237 * Turn quotas on on a device
2241 * Helper function to turn quotas on when we already have the inode of
2242 * quota file and no quota information is loaded.
2244 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2247 struct quota_format_type *fmt = find_quota_format(format_id);
2248 struct super_block *sb = inode->i_sb;
2249 struct quota_info *dqopt = sb_dqopt(sb);
2255 if (!S_ISREG(inode->i_mode)) {
2259 if (IS_RDONLY(inode)) {
2263 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2264 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2268 /* Filesystems outside of init_user_ns not yet supported */
2269 if (sb->s_user_ns != &init_user_ns) {
2273 /* Usage always has to be set... */
2274 if (!(flags & DQUOT_USAGE_ENABLED)) {
2279 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2280 /* As we bypass the pagecache we must now flush all the
2281 * dirty data and invalidate caches so that kernel sees
2282 * changes from userspace. It is not enough to just flush
2283 * the quota file since if blocksize < pagesize, invalidation
2284 * of the cache could fail because of other unrelated dirty
2286 sync_filesystem(sb);
2287 invalidate_bdev(sb->s_bdev);
2289 mutex_lock(&dqopt->dqonoff_mutex);
2290 if (sb_has_quota_loaded(sb, type)) {
2295 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2296 /* We don't want quota and atime on quota files (deadlocks
2297 * possible) Also nobody should write to the file - we use
2298 * special IO operations which ignore the immutable bit. */
2300 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2302 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2303 inode_unlock(inode);
2305 * When S_NOQUOTA is set, remove dquot references as no more
2306 * references can be added
2308 __dquot_drop(inode);
2312 dqopt->files[type] = igrab(inode);
2313 if (!dqopt->files[type])
2316 if (!fmt->qf_ops->check_quota_file(sb, type))
2319 dqopt->ops[type] = fmt->qf_ops;
2320 dqopt->info[type].dqi_format = fmt;
2321 dqopt->info[type].dqi_fmt_id = format_id;
2322 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2323 mutex_lock(&dqopt->dqio_mutex);
2324 error = dqopt->ops[type]->read_file_info(sb, type);
2326 mutex_unlock(&dqopt->dqio_mutex);
2329 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2330 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2331 mutex_unlock(&dqopt->dqio_mutex);
2332 spin_lock(&dq_state_lock);
2333 dqopt->flags |= dquot_state_flag(flags, type);
2334 spin_unlock(&dq_state_lock);
2336 add_dquot_ref(sb, type);
2337 mutex_unlock(&dqopt->dqonoff_mutex);
2342 dqopt->files[type] = NULL;
2345 if (oldflags != -1) {
2347 /* Set the flags back (in the case of accidental quotaon()
2348 * on a wrong file we don't want to mess up the flags) */
2349 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2350 inode->i_flags |= oldflags;
2351 inode_unlock(inode);
2353 mutex_unlock(&dqopt->dqonoff_mutex);
2355 put_quota_format(fmt);
2360 /* Reenable quotas on remount RW */
2361 int dquot_resume(struct super_block *sb, int type)
2363 struct quota_info *dqopt = sb_dqopt(sb);
2364 struct inode *inode;
2368 /* s_umount should be held in exclusive mode */
2369 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2370 up_read(&sb->s_umount);
2372 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2373 if (type != -1 && cnt != type)
2376 mutex_lock(&dqopt->dqonoff_mutex);
2377 if (!sb_has_quota_suspended(sb, cnt)) {
2378 mutex_unlock(&dqopt->dqonoff_mutex);
2381 inode = dqopt->files[cnt];
2382 dqopt->files[cnt] = NULL;
2383 spin_lock(&dq_state_lock);
2384 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2385 DQUOT_LIMITS_ENABLED,
2387 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2388 spin_unlock(&dq_state_lock);
2389 mutex_unlock(&dqopt->dqonoff_mutex);
2391 flags = dquot_generic_flag(flags, cnt);
2392 ret = vfs_load_quota_inode(inode, cnt,
2393 dqopt->info[cnt].dqi_fmt_id, flags);
2399 EXPORT_SYMBOL(dquot_resume);
2401 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2404 int error = security_quota_on(path->dentry);
2407 /* Quota file not on the same filesystem? */
2408 if (path->dentry->d_sb != sb)
2411 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2412 format_id, DQUOT_USAGE_ENABLED |
2413 DQUOT_LIMITS_ENABLED);
2416 EXPORT_SYMBOL(dquot_quota_on);
2419 * More powerful function for turning on quotas allowing setting
2420 * of individual quota flags
2422 int dquot_enable(struct inode *inode, int type, int format_id,
2426 struct super_block *sb = inode->i_sb;
2427 struct quota_info *dqopt = sb_dqopt(sb);
2429 /* Just unsuspend quotas? */
2430 BUG_ON(flags & DQUOT_SUSPENDED);
2431 /* s_umount should be held in exclusive mode */
2432 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2433 up_read(&sb->s_umount);
2437 /* Just updating flags needed? */
2438 if (sb_has_quota_loaded(sb, type)) {
2439 mutex_lock(&dqopt->dqonoff_mutex);
2440 /* Now do a reliable test... */
2441 if (!sb_has_quota_loaded(sb, type)) {
2442 mutex_unlock(&dqopt->dqonoff_mutex);
2445 if (flags & DQUOT_USAGE_ENABLED &&
2446 sb_has_quota_usage_enabled(sb, type)) {
2450 if (flags & DQUOT_LIMITS_ENABLED &&
2451 sb_has_quota_limits_enabled(sb, type)) {
2455 spin_lock(&dq_state_lock);
2456 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2457 spin_unlock(&dq_state_lock);
2459 mutex_unlock(&dqopt->dqonoff_mutex);
2464 return vfs_load_quota_inode(inode, type, format_id, flags);
2466 EXPORT_SYMBOL(dquot_enable);
2469 * This function is used when filesystem needs to initialize quotas
2470 * during mount time.
2472 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2473 int format_id, int type)
2475 struct dentry *dentry;
2478 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2480 return PTR_ERR(dentry);
2482 if (d_really_is_negative(dentry)) {
2487 error = security_quota_on(dentry);
2489 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2490 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2496 EXPORT_SYMBOL(dquot_quota_on_mount);
2498 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2502 struct quota_info *dqopt = sb_dqopt(sb);
2504 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2506 /* Accounting cannot be turned on while fs is mounted */
2507 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2510 for (type = 0; type < MAXQUOTAS; type++) {
2511 if (!(flags & qtype_enforce_flag(type)))
2513 /* Can't enforce without accounting */
2514 if (!sb_has_quota_usage_enabled(sb, type))
2516 ret = dquot_enable(dqopt->files[type], type,
2517 dqopt->info[type].dqi_fmt_id,
2518 DQUOT_LIMITS_ENABLED);
2524 /* Backout enforcement enablement we already did */
2525 for (type--; type >= 0; type--) {
2526 if (flags & qtype_enforce_flag(type))
2527 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2529 /* Error code translation for better compatibility with XFS */
2535 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2539 struct quota_info *dqopt = sb_dqopt(sb);
2541 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2544 * We don't support turning off accounting via quotactl. In principle
2545 * quota infrastructure can do this but filesystems don't expect
2546 * userspace to be able to do it.
2549 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2552 /* Filter out limits not enabled */
2553 for (type = 0; type < MAXQUOTAS; type++)
2554 if (!sb_has_quota_limits_enabled(sb, type))
2555 flags &= ~qtype_enforce_flag(type);
2559 for (type = 0; type < MAXQUOTAS; type++) {
2560 if (flags & qtype_enforce_flag(type)) {
2561 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2568 /* Backout enforcement disabling we already did */
2569 for (type--; type >= 0; type--) {
2570 if (flags & qtype_enforce_flag(type))
2571 dquot_enable(dqopt->files[type], type,
2572 dqopt->info[type].dqi_fmt_id,
2573 DQUOT_LIMITS_ENABLED);
2578 /* Generic routine for getting common part of quota structure */
2579 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2581 struct mem_dqblk *dm = &dquot->dq_dqb;
2583 memset(di, 0, sizeof(*di));
2584 spin_lock(&dq_data_lock);
2585 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2586 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2587 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2588 di->d_ino_softlimit = dm->dqb_isoftlimit;
2589 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2590 di->d_ino_count = dm->dqb_curinodes;
2591 di->d_spc_timer = dm->dqb_btime;
2592 di->d_ino_timer = dm->dqb_itime;
2593 spin_unlock(&dq_data_lock);
2596 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2597 struct qc_dqblk *di)
2599 struct dquot *dquot;
2601 dquot = dqget(sb, qid);
2603 return PTR_ERR(dquot);
2604 do_get_dqblk(dquot, di);
2609 EXPORT_SYMBOL(dquot_get_dqblk);
2611 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2612 struct qc_dqblk *di)
2614 struct dquot *dquot;
2617 if (!sb->dq_op->get_next_id)
2619 err = sb->dq_op->get_next_id(sb, qid);
2622 dquot = dqget(sb, *qid);
2624 return PTR_ERR(dquot);
2625 do_get_dqblk(dquot, di);
2630 EXPORT_SYMBOL(dquot_get_next_dqblk);
2632 #define VFS_QC_MASK \
2633 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2634 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2635 QC_SPC_TIMER | QC_INO_TIMER)
2637 /* Generic routine for setting common part of quota structure */
2638 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2640 struct mem_dqblk *dm = &dquot->dq_dqb;
2641 int check_blim = 0, check_ilim = 0;
2642 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2644 if (di->d_fieldmask & ~VFS_QC_MASK)
2647 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2648 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2649 ((di->d_fieldmask & QC_SPC_HARD) &&
2650 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2651 ((di->d_fieldmask & QC_INO_SOFT) &&
2652 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2653 ((di->d_fieldmask & QC_INO_HARD) &&
2654 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2657 spin_lock(&dq_data_lock);
2658 if (di->d_fieldmask & QC_SPACE) {
2659 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2661 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2664 if (di->d_fieldmask & QC_SPC_SOFT)
2665 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2666 if (di->d_fieldmask & QC_SPC_HARD)
2667 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2668 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2670 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2673 if (di->d_fieldmask & QC_INO_COUNT) {
2674 dm->dqb_curinodes = di->d_ino_count;
2676 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2679 if (di->d_fieldmask & QC_INO_SOFT)
2680 dm->dqb_isoftlimit = di->d_ino_softlimit;
2681 if (di->d_fieldmask & QC_INO_HARD)
2682 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2683 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2685 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2688 if (di->d_fieldmask & QC_SPC_TIMER) {
2689 dm->dqb_btime = di->d_spc_timer;
2691 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2694 if (di->d_fieldmask & QC_INO_TIMER) {
2695 dm->dqb_itime = di->d_ino_timer;
2697 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2701 if (!dm->dqb_bsoftlimit ||
2702 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2704 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2705 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2706 /* Set grace only if user hasn't provided his own... */
2707 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2710 if (!dm->dqb_isoftlimit ||
2711 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2713 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2714 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2715 /* Set grace only if user hasn't provided his own... */
2716 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2718 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2720 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2722 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2723 spin_unlock(&dq_data_lock);
2724 mark_dquot_dirty(dquot);
2729 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2730 struct qc_dqblk *di)
2732 struct dquot *dquot;
2735 dquot = dqget(sb, qid);
2736 if (IS_ERR(dquot)) {
2737 rc = PTR_ERR(dquot);
2740 rc = do_set_dqblk(dquot, di);
2745 EXPORT_SYMBOL(dquot_set_dqblk);
2747 /* Generic routine for getting common part of quota file information */
2748 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2750 struct mem_dqinfo *mi;
2751 struct qc_type_state *tstate;
2752 struct quota_info *dqopt = sb_dqopt(sb);
2755 memset(state, 0, sizeof(*state));
2756 for (type = 0; type < MAXQUOTAS; type++) {
2757 if (!sb_has_quota_active(sb, type))
2759 tstate = state->s_state + type;
2760 mi = sb_dqopt(sb)->info + type;
2761 tstate->flags = QCI_ACCT_ENABLED;
2762 spin_lock(&dq_data_lock);
2763 if (mi->dqi_flags & DQF_SYS_FILE)
2764 tstate->flags |= QCI_SYSFILE;
2765 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2766 tstate->flags |= QCI_ROOT_SQUASH;
2767 if (sb_has_quota_limits_enabled(sb, type))
2768 tstate->flags |= QCI_LIMITS_ENFORCED;
2769 tstate->spc_timelimit = mi->dqi_bgrace;
2770 tstate->ino_timelimit = mi->dqi_igrace;
2771 tstate->ino = dqopt->files[type]->i_ino;
2772 tstate->blocks = dqopt->files[type]->i_blocks;
2773 tstate->nextents = 1; /* We don't know... */
2774 spin_unlock(&dq_data_lock);
2778 EXPORT_SYMBOL(dquot_get_state);
2780 /* Generic routine for setting common part of quota file information */
2781 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2783 struct mem_dqinfo *mi;
2786 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2787 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2789 if (!sb_has_quota_active(sb, type))
2791 mi = sb_dqopt(sb)->info + type;
2792 if (ii->i_fieldmask & QC_FLAGS) {
2793 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2794 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2797 spin_lock(&dq_data_lock);
2798 if (ii->i_fieldmask & QC_SPC_TIMER)
2799 mi->dqi_bgrace = ii->i_spc_timelimit;
2800 if (ii->i_fieldmask & QC_INO_TIMER)
2801 mi->dqi_igrace = ii->i_ino_timelimit;
2802 if (ii->i_fieldmask & QC_FLAGS) {
2803 if (ii->i_flags & QCI_ROOT_SQUASH)
2804 mi->dqi_flags |= DQF_ROOT_SQUASH;
2806 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2808 spin_unlock(&dq_data_lock);
2809 mark_info_dirty(sb, type);
2810 /* Force write to disk */
2811 sb->dq_op->write_info(sb, type);
2814 EXPORT_SYMBOL(dquot_set_dqinfo);
2816 const struct quotactl_ops dquot_quotactl_ops = {
2817 .quota_on = dquot_quota_on,
2818 .quota_off = dquot_quota_off,
2819 .quota_sync = dquot_quota_sync,
2820 .get_state = dquot_get_state,
2821 .set_info = dquot_set_dqinfo,
2822 .get_dqblk = dquot_get_dqblk,
2823 .get_nextdqblk = dquot_get_next_dqblk,
2824 .set_dqblk = dquot_set_dqblk
2826 EXPORT_SYMBOL(dquot_quotactl_ops);
2828 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2829 .quota_enable = dquot_quota_enable,
2830 .quota_disable = dquot_quota_disable,
2831 .quota_sync = dquot_quota_sync,
2832 .get_state = dquot_get_state,
2833 .set_info = dquot_set_dqinfo,
2834 .get_dqblk = dquot_get_dqblk,
2835 .get_nextdqblk = dquot_get_next_dqblk,
2836 .set_dqblk = dquot_set_dqblk
2838 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2840 static int do_proc_dqstats(struct ctl_table *table, int write,
2841 void __user *buffer, size_t *lenp, loff_t *ppos)
2843 unsigned int type = (int *)table->data - dqstats.stat;
2845 /* Update global table */
2846 dqstats.stat[type] =
2847 percpu_counter_sum_positive(&dqstats.counter[type]);
2848 return proc_dointvec(table, write, buffer, lenp, ppos);
2851 static struct ctl_table fs_dqstats_table[] = {
2853 .procname = "lookups",
2854 .data = &dqstats.stat[DQST_LOOKUPS],
2855 .maxlen = sizeof(int),
2857 .proc_handler = do_proc_dqstats,
2860 .procname = "drops",
2861 .data = &dqstats.stat[DQST_DROPS],
2862 .maxlen = sizeof(int),
2864 .proc_handler = do_proc_dqstats,
2867 .procname = "reads",
2868 .data = &dqstats.stat[DQST_READS],
2869 .maxlen = sizeof(int),
2871 .proc_handler = do_proc_dqstats,
2874 .procname = "writes",
2875 .data = &dqstats.stat[DQST_WRITES],
2876 .maxlen = sizeof(int),
2878 .proc_handler = do_proc_dqstats,
2881 .procname = "cache_hits",
2882 .data = &dqstats.stat[DQST_CACHE_HITS],
2883 .maxlen = sizeof(int),
2885 .proc_handler = do_proc_dqstats,
2888 .procname = "allocated_dquots",
2889 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2890 .maxlen = sizeof(int),
2892 .proc_handler = do_proc_dqstats,
2895 .procname = "free_dquots",
2896 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2897 .maxlen = sizeof(int),
2899 .proc_handler = do_proc_dqstats,
2902 .procname = "syncs",
2903 .data = &dqstats.stat[DQST_SYNCS],
2904 .maxlen = sizeof(int),
2906 .proc_handler = do_proc_dqstats,
2908 #ifdef CONFIG_PRINT_QUOTA_WARNING
2910 .procname = "warnings",
2911 .data = &flag_print_warnings,
2912 .maxlen = sizeof(int),
2914 .proc_handler = proc_dointvec,
2920 static struct ctl_table fs_table[] = {
2922 .procname = "quota",
2924 .child = fs_dqstats_table,
2929 static struct ctl_table sys_table[] = {
2938 static int __init dquot_init(void)
2941 unsigned long nr_hash, order;
2943 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2945 register_sysctl_table(sys_table);
2947 dquot_cachep = kmem_cache_create("dquot",
2948 sizeof(struct dquot), sizeof(unsigned long) * 4,
2949 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2950 SLAB_MEM_SPREAD|SLAB_PANIC),
2954 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2956 panic("Cannot create dquot hash table");
2958 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2959 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2961 panic("Cannot create dquot stat counters");
2964 /* Find power-of-two hlist_heads which can fit into allocation */
2965 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2969 } while (nr_hash >> dq_hash_bits);
2972 nr_hash = 1UL << dq_hash_bits;
2973 dq_hash_mask = nr_hash - 1;
2974 for (i = 0; i < nr_hash; i++)
2975 INIT_HLIST_HEAD(dquot_hash + i);
2977 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2978 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2980 register_shrinker(&dqcache_shrinker);
2984 fs_initcall(dquot_init);