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 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
576 spin_lock(&dq_list_lock);
577 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
578 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
580 if (dquot->dq_sb != sb)
582 /* Now we have active dquot so we can just increase use count */
583 atomic_inc(&dquot->dq_count);
584 spin_unlock(&dq_list_lock);
585 dqstats_inc(DQST_LOOKUPS);
589 * ->release_dquot() can be racing with us. Our reference
590 * protects us from new calls to it so just wait for any
591 * outstanding call and recheck the DQ_ACTIVE_B after that.
593 wait_on_dquot(dquot);
594 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
595 ret = fn(dquot, priv);
599 spin_lock(&dq_list_lock);
600 /* We are safe to continue now because our dquot could not
601 * be moved out of the inuse list while we hold the reference */
603 spin_unlock(&dq_list_lock);
606 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
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 mutex_lock(&dqopt->dqonoff_mutex);
621 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
622 if (type != -1 && cnt != type)
624 if (!sb_has_quota_active(sb, cnt))
626 spin_lock(&dq_list_lock);
627 dirty = &dqopt->info[cnt].dqi_dirty_list;
628 while (!list_empty(dirty)) {
629 dquot = list_first_entry(dirty, struct dquot,
631 /* Dirty and inactive can be only bad dquot... */
632 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
633 clear_dquot_dirty(dquot);
636 /* Now we have active dquot from which someone is
637 * holding reference so we can safely just increase
640 spin_unlock(&dq_list_lock);
641 dqstats_inc(DQST_LOOKUPS);
642 err = sb->dq_op->write_dquot(dquot);
646 spin_lock(&dq_list_lock);
648 spin_unlock(&dq_list_lock);
651 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
652 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
653 && info_dirty(&dqopt->info[cnt]))
654 sb->dq_op->write_info(sb, cnt);
655 dqstats_inc(DQST_SYNCS);
656 mutex_unlock(&dqopt->dqonoff_mutex);
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 mutex_lock(&dqopt->dqonoff_mutex);
687 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
688 if (type != -1 && cnt != type)
690 if (!sb_has_quota_active(sb, cnt))
692 inode_lock(dqopt->files[cnt]);
693 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
694 inode_unlock(dqopt->files[cnt]);
696 mutex_unlock(&dqopt->dqonoff_mutex);
700 EXPORT_SYMBOL(dquot_quota_sync);
703 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
705 struct list_head *head;
707 unsigned long freed = 0;
709 spin_lock(&dq_list_lock);
710 head = free_dquots.prev;
711 while (head != &free_dquots && sc->nr_to_scan) {
712 dquot = list_entry(head, struct dquot, dq_free);
713 remove_dquot_hash(dquot);
714 remove_free_dquot(dquot);
716 do_destroy_dquot(dquot);
719 head = free_dquots.prev;
721 spin_unlock(&dq_list_lock);
726 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
728 return vfs_pressure_ratio(
729 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
732 static struct shrinker dqcache_shrinker = {
733 .count_objects = dqcache_shrink_count,
734 .scan_objects = dqcache_shrink_scan,
735 .seeks = DEFAULT_SEEKS,
739 * Put reference to dquot
741 void dqput(struct dquot *dquot)
747 #ifdef CONFIG_QUOTA_DEBUG
748 if (!atomic_read(&dquot->dq_count)) {
749 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
750 quotatypes[dquot->dq_id.type],
751 from_kqid(&init_user_ns, dquot->dq_id));
755 dqstats_inc(DQST_DROPS);
757 spin_lock(&dq_list_lock);
758 if (atomic_read(&dquot->dq_count) > 1) {
759 /* We have more than one user... nothing to do */
760 atomic_dec(&dquot->dq_count);
761 /* Releasing dquot during quotaoff phase? */
762 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
763 atomic_read(&dquot->dq_count) == 1)
764 wake_up(&dquot->dq_wait_unused);
765 spin_unlock(&dq_list_lock);
768 /* Need to release dquot? */
769 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
770 spin_unlock(&dq_list_lock);
771 /* Commit dquot before releasing */
772 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
774 quota_error(dquot->dq_sb, "Can't write quota structure"
775 " (error %d). Quota may get out of sync!",
778 * We clear dirty bit anyway, so that we avoid
781 spin_lock(&dq_list_lock);
782 clear_dquot_dirty(dquot);
783 spin_unlock(&dq_list_lock);
787 /* Clear flag in case dquot was inactive (something bad happened) */
788 clear_dquot_dirty(dquot);
789 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
790 spin_unlock(&dq_list_lock);
791 dquot->dq_sb->dq_op->release_dquot(dquot);
794 atomic_dec(&dquot->dq_count);
795 #ifdef CONFIG_QUOTA_DEBUG
797 BUG_ON(!list_empty(&dquot->dq_free));
799 put_dquot_last(dquot);
800 spin_unlock(&dq_list_lock);
802 EXPORT_SYMBOL(dqput);
804 struct dquot *dquot_alloc(struct super_block *sb, int type)
806 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
808 EXPORT_SYMBOL(dquot_alloc);
810 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
814 dquot = sb->dq_op->alloc_dquot(sb, type);
818 mutex_init(&dquot->dq_lock);
819 INIT_LIST_HEAD(&dquot->dq_free);
820 INIT_LIST_HEAD(&dquot->dq_inuse);
821 INIT_HLIST_NODE(&dquot->dq_hash);
822 INIT_LIST_HEAD(&dquot->dq_dirty);
823 init_waitqueue_head(&dquot->dq_wait_unused);
825 dquot->dq_id = make_kqid_invalid(type);
826 atomic_set(&dquot->dq_count, 1);
832 * Get reference to dquot
834 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
835 * destroying our dquot by:
836 * a) checking for quota flags under dq_list_lock and
837 * b) getting a reference to dquot before we release dq_list_lock
839 struct dquot *dqget(struct super_block *sb, struct kqid qid)
841 unsigned int hashent = hashfn(sb, qid);
842 struct dquot *dquot, *empty = NULL;
844 if (!sb_has_quota_active(sb, qid.type))
845 return ERR_PTR(-ESRCH);
847 spin_lock(&dq_list_lock);
848 spin_lock(&dq_state_lock);
849 if (!sb_has_quota_active(sb, qid.type)) {
850 spin_unlock(&dq_state_lock);
851 spin_unlock(&dq_list_lock);
852 dquot = ERR_PTR(-ESRCH);
855 spin_unlock(&dq_state_lock);
857 dquot = find_dquot(hashent, sb, qid);
860 spin_unlock(&dq_list_lock);
861 empty = get_empty_dquot(sb, qid.type);
863 schedule(); /* Try to wait for a moment... */
869 /* all dquots go on the inuse_list */
871 /* hash it first so it can be found */
872 insert_dquot_hash(dquot);
873 spin_unlock(&dq_list_lock);
874 dqstats_inc(DQST_LOOKUPS);
876 if (!atomic_read(&dquot->dq_count))
877 remove_free_dquot(dquot);
878 atomic_inc(&dquot->dq_count);
879 spin_unlock(&dq_list_lock);
880 dqstats_inc(DQST_CACHE_HITS);
881 dqstats_inc(DQST_LOOKUPS);
883 /* Wait for dq_lock - after this we know that either dquot_release() is
884 * already finished or it will be canceled due to dq_count > 1 test */
885 wait_on_dquot(dquot);
886 /* Read the dquot / allocate space in quota file */
887 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
890 err = sb->dq_op->acquire_dquot(dquot);
893 dquot = ERR_PTR(err);
898 * Make sure following reads see filled structure - paired with
899 * smp_mb__before_atomic() in dquot_acquire().
902 #ifdef CONFIG_QUOTA_DEBUG
903 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
907 do_destroy_dquot(empty);
911 EXPORT_SYMBOL(dqget);
913 static inline struct dquot **i_dquot(struct inode *inode)
915 return inode->i_sb->s_op->get_dquots(inode);
918 static int dqinit_needed(struct inode *inode, int type)
920 struct dquot * const *dquots;
923 if (IS_NOQUOTA(inode))
926 dquots = i_dquot(inode);
928 return !dquots[type];
929 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
935 /* This routine is guarded by dqonoff_mutex mutex */
936 static void add_dquot_ref(struct super_block *sb, int type)
938 struct inode *inode, *old_inode = NULL;
939 #ifdef CONFIG_QUOTA_DEBUG
943 spin_lock(&sb->s_inode_list_lock);
944 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
945 spin_lock(&inode->i_lock);
946 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
947 !atomic_read(&inode->i_writecount) ||
948 !dqinit_needed(inode, type)) {
949 spin_unlock(&inode->i_lock);
953 spin_unlock(&inode->i_lock);
954 spin_unlock(&sb->s_inode_list_lock);
956 #ifdef CONFIG_QUOTA_DEBUG
957 if (unlikely(inode_get_rsv_space(inode) > 0))
961 __dquot_initialize(inode, type);
964 * We hold a reference to 'inode' so it couldn't have been
965 * removed from s_inodes list while we dropped the
966 * s_inode_list_lock. We cannot iput the inode now as we can be
967 * holding the last reference and we cannot iput it under
968 * s_inode_list_lock. So we keep the reference and iput it
972 spin_lock(&sb->s_inode_list_lock);
974 spin_unlock(&sb->s_inode_list_lock);
977 #ifdef CONFIG_QUOTA_DEBUG
979 quota_error(sb, "Writes happened before quota was turned on "
980 "thus quota information is probably inconsistent. "
981 "Please run quotacheck(8)");
987 * Remove references to dquots from inode and add dquot to list for freeing
988 * if we have the last reference to dquot
990 static void remove_inode_dquot_ref(struct inode *inode, int type,
991 struct list_head *tofree_head)
993 struct dquot **dquots = i_dquot(inode);
994 struct dquot *dquot = dquots[type];
1000 if (list_empty(&dquot->dq_free)) {
1002 * The inode still has reference to dquot so it can't be in the
1005 spin_lock(&dq_list_lock);
1006 list_add(&dquot->dq_free, tofree_head);
1007 spin_unlock(&dq_list_lock);
1010 * Dquot is already in a list to put so we won't drop the last
1018 * Free list of dquots
1019 * Dquots are removed from inodes and no new references can be got so we are
1020 * the only ones holding reference
1022 static void put_dquot_list(struct list_head *tofree_head)
1024 struct list_head *act_head;
1025 struct dquot *dquot;
1027 act_head = tofree_head->next;
1028 while (act_head != tofree_head) {
1029 dquot = list_entry(act_head, struct dquot, dq_free);
1030 act_head = act_head->next;
1031 /* Remove dquot from the list so we won't have problems... */
1032 list_del_init(&dquot->dq_free);
1037 static void remove_dquot_ref(struct super_block *sb, int type,
1038 struct list_head *tofree_head)
1040 struct inode *inode;
1043 spin_lock(&sb->s_inode_list_lock);
1044 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1046 * We have to scan also I_NEW inodes because they can already
1047 * have quota pointer initialized. Luckily, we need to touch
1048 * only quota pointers and these have separate locking
1051 spin_lock(&dq_data_lock);
1052 if (!IS_NOQUOTA(inode)) {
1053 if (unlikely(inode_get_rsv_space(inode) > 0))
1055 remove_inode_dquot_ref(inode, type, tofree_head);
1057 spin_unlock(&dq_data_lock);
1059 spin_unlock(&sb->s_inode_list_lock);
1060 #ifdef CONFIG_QUOTA_DEBUG
1062 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1063 " was disabled thus quota information is probably "
1064 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1069 /* Gather all references from inodes and drop them */
1070 static void drop_dquot_ref(struct super_block *sb, int type)
1072 LIST_HEAD(tofree_head);
1075 remove_dquot_ref(sb, type, &tofree_head);
1076 synchronize_srcu(&dquot_srcu);
1077 put_dquot_list(&tofree_head);
1081 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1083 dquot->dq_dqb.dqb_curinodes += number;
1086 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1088 dquot->dq_dqb.dqb_curspace += number;
1091 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1093 dquot->dq_dqb.dqb_rsvspace += number;
1097 * Claim reserved quota space
1099 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1101 if (dquot->dq_dqb.dqb_rsvspace < number) {
1103 number = dquot->dq_dqb.dqb_rsvspace;
1105 dquot->dq_dqb.dqb_curspace += number;
1106 dquot->dq_dqb.dqb_rsvspace -= number;
1109 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1111 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1112 number = dquot->dq_dqb.dqb_curspace;
1113 dquot->dq_dqb.dqb_rsvspace += number;
1114 dquot->dq_dqb.dqb_curspace -= number;
1118 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1120 if (dquot->dq_dqb.dqb_rsvspace >= number)
1121 dquot->dq_dqb.dqb_rsvspace -= number;
1124 dquot->dq_dqb.dqb_rsvspace = 0;
1128 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1130 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1131 dquot->dq_dqb.dqb_curinodes >= number)
1132 dquot->dq_dqb.dqb_curinodes -= number;
1134 dquot->dq_dqb.dqb_curinodes = 0;
1135 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1136 dquot->dq_dqb.dqb_itime = (time_t) 0;
1137 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1140 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1142 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1143 dquot->dq_dqb.dqb_curspace >= number)
1144 dquot->dq_dqb.dqb_curspace -= number;
1146 dquot->dq_dqb.dqb_curspace = 0;
1147 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1148 dquot->dq_dqb.dqb_btime = (time_t) 0;
1149 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1153 struct super_block *w_sb;
1154 struct kqid w_dq_id;
1158 static int warning_issued(struct dquot *dquot, const int warntype)
1160 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1161 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1162 ((warntype == QUOTA_NL_IHARDWARN ||
1163 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1167 return test_and_set_bit(flag, &dquot->dq_flags);
1170 #ifdef CONFIG_PRINT_QUOTA_WARNING
1171 static int flag_print_warnings = 1;
1173 static int need_print_warning(struct dquot_warn *warn)
1175 if (!flag_print_warnings)
1178 switch (warn->w_dq_id.type) {
1180 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1182 return in_group_p(warn->w_dq_id.gid);
1189 /* Print warning to user which exceeded quota */
1190 static void print_warning(struct dquot_warn *warn)
1193 struct tty_struct *tty;
1194 int warntype = warn->w_type;
1196 if (warntype == QUOTA_NL_IHARDBELOW ||
1197 warntype == QUOTA_NL_ISOFTBELOW ||
1198 warntype == QUOTA_NL_BHARDBELOW ||
1199 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1202 tty = get_current_tty();
1205 tty_write_message(tty, warn->w_sb->s_id);
1206 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1207 tty_write_message(tty, ": warning, ");
1209 tty_write_message(tty, ": write failed, ");
1210 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1212 case QUOTA_NL_IHARDWARN:
1213 msg = " file limit reached.\r\n";
1215 case QUOTA_NL_ISOFTLONGWARN:
1216 msg = " file quota exceeded too long.\r\n";
1218 case QUOTA_NL_ISOFTWARN:
1219 msg = " file quota exceeded.\r\n";
1221 case QUOTA_NL_BHARDWARN:
1222 msg = " block limit reached.\r\n";
1224 case QUOTA_NL_BSOFTLONGWARN:
1225 msg = " block quota exceeded too long.\r\n";
1227 case QUOTA_NL_BSOFTWARN:
1228 msg = " block quota exceeded.\r\n";
1231 tty_write_message(tty, msg);
1236 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1239 if (warning_issued(dquot, warntype))
1241 warn->w_type = warntype;
1242 warn->w_sb = dquot->dq_sb;
1243 warn->w_dq_id = dquot->dq_id;
1247 * Write warnings to the console and send warning messages over netlink.
1249 * Note that this function can call into tty and networking code.
1251 static void flush_warnings(struct dquot_warn *warn)
1255 for (i = 0; i < MAXQUOTAS; i++) {
1256 if (warn[i].w_type == QUOTA_NL_NOWARN)
1258 #ifdef CONFIG_PRINT_QUOTA_WARNING
1259 print_warning(&warn[i]);
1261 quota_send_warning(warn[i].w_dq_id,
1262 warn[i].w_sb->s_dev, warn[i].w_type);
1266 static int ignore_hardlimit(struct dquot *dquot)
1268 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1270 return capable(CAP_SYS_RESOURCE) &&
1271 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1272 !(info->dqi_flags & DQF_ROOT_SQUASH));
1275 /* needs dq_data_lock */
1276 static int check_idq(struct dquot *dquot, qsize_t inodes,
1277 struct dquot_warn *warn)
1279 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1281 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1282 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1285 if (dquot->dq_dqb.dqb_ihardlimit &&
1286 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1287 !ignore_hardlimit(dquot)) {
1288 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1292 if (dquot->dq_dqb.dqb_isoftlimit &&
1293 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1294 dquot->dq_dqb.dqb_itime &&
1295 get_seconds() >= dquot->dq_dqb.dqb_itime &&
1296 !ignore_hardlimit(dquot)) {
1297 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1301 if (dquot->dq_dqb.dqb_isoftlimit &&
1302 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1303 dquot->dq_dqb.dqb_itime == 0) {
1304 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1305 dquot->dq_dqb.dqb_itime = get_seconds() +
1306 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1312 /* needs dq_data_lock */
1313 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1314 struct dquot_warn *warn)
1317 struct super_block *sb = dquot->dq_sb;
1319 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1320 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1323 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1326 if (dquot->dq_dqb.dqb_bhardlimit &&
1327 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1328 !ignore_hardlimit(dquot)) {
1330 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1334 if (dquot->dq_dqb.dqb_bsoftlimit &&
1335 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1336 dquot->dq_dqb.dqb_btime &&
1337 get_seconds() >= dquot->dq_dqb.dqb_btime &&
1338 !ignore_hardlimit(dquot)) {
1340 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1344 if (dquot->dq_dqb.dqb_bsoftlimit &&
1345 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1346 dquot->dq_dqb.dqb_btime == 0) {
1348 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1349 dquot->dq_dqb.dqb_btime = get_seconds() +
1350 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1354 * We don't allow preallocation to exceed softlimit so exceeding will
1363 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1367 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1368 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1369 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1370 return QUOTA_NL_NOWARN;
1372 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1373 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1374 return QUOTA_NL_ISOFTBELOW;
1375 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1376 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1377 return QUOTA_NL_IHARDBELOW;
1378 return QUOTA_NL_NOWARN;
1381 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1383 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1384 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1385 return QUOTA_NL_NOWARN;
1387 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1388 return QUOTA_NL_BSOFTBELOW;
1389 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1390 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1391 return QUOTA_NL_BHARDBELOW;
1392 return QUOTA_NL_NOWARN;
1395 static int dquot_active(const struct inode *inode)
1397 struct super_block *sb = inode->i_sb;
1399 if (IS_NOQUOTA(inode))
1401 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1405 * Initialize quota pointers in inode
1407 * It is better to call this function outside of any transaction as it
1408 * might need a lot of space in journal for dquot structure allocation.
1410 static int __dquot_initialize(struct inode *inode, int type)
1412 int cnt, init_needed = 0;
1413 struct dquot **dquots, *got[MAXQUOTAS] = {};
1414 struct super_block *sb = inode->i_sb;
1418 if (!dquot_active(inode))
1421 dquots = i_dquot(inode);
1423 /* First get references to structures we might need. */
1424 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1428 struct dquot *dquot;
1430 if (type != -1 && cnt != type)
1433 * The i_dquot should have been initialized in most cases,
1434 * we check it without locking here to avoid unnecessary
1435 * dqget()/dqput() calls.
1440 if (!sb_has_quota_active(sb, cnt))
1447 qid = make_kqid_uid(inode->i_uid);
1450 qid = make_kqid_gid(inode->i_gid);
1453 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1456 qid = make_kqid_projid(projid);
1459 dquot = dqget(sb, qid);
1460 if (IS_ERR(dquot)) {
1461 /* We raced with somebody turning quotas off... */
1462 if (PTR_ERR(dquot) != -ESRCH) {
1463 ret = PTR_ERR(dquot);
1471 /* All required i_dquot has been initialized */
1475 spin_lock(&dq_data_lock);
1476 if (IS_NOQUOTA(inode))
1478 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1479 if (type != -1 && cnt != type)
1481 /* Avoid races with quotaoff() */
1482 if (!sb_has_quota_active(sb, cnt))
1484 /* We could race with quotaon or dqget() could have failed */
1488 dquots[cnt] = got[cnt];
1491 * Make quota reservation system happy if someone
1492 * did a write before quota was turned on
1494 rsv = inode_get_rsv_space(inode);
1496 dquot_resv_space(dquots[cnt], rsv);
1500 spin_unlock(&dq_data_lock);
1502 /* Drop unused references */
1508 int dquot_initialize(struct inode *inode)
1510 return __dquot_initialize(inode, -1);
1512 EXPORT_SYMBOL(dquot_initialize);
1515 * Release all quotas referenced by inode.
1517 * This function only be called on inode free or converting
1518 * a file to quota file, no other users for the i_dquot in
1519 * both cases, so we needn't call synchronize_srcu() after
1522 static void __dquot_drop(struct inode *inode)
1525 struct dquot **dquots = i_dquot(inode);
1526 struct dquot *put[MAXQUOTAS];
1528 spin_lock(&dq_data_lock);
1529 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1530 put[cnt] = dquots[cnt];
1533 spin_unlock(&dq_data_lock);
1537 void dquot_drop(struct inode *inode)
1539 struct dquot * const *dquots;
1542 if (IS_NOQUOTA(inode))
1546 * Test before calling to rule out calls from proc and such
1547 * where we are not allowed to block. Note that this is
1548 * actually reliable test even without the lock - the caller
1549 * must assure that nobody can come after the DQUOT_DROP and
1550 * add quota pointers back anyway.
1552 dquots = i_dquot(inode);
1553 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1558 if (cnt < MAXQUOTAS)
1559 __dquot_drop(inode);
1561 EXPORT_SYMBOL(dquot_drop);
1564 * inode_reserved_space is managed internally by quota, and protected by
1565 * i_lock similar to i_blocks+i_bytes.
1567 static qsize_t *inode_reserved_space(struct inode * inode)
1569 /* Filesystem must explicitly define it's own method in order to use
1570 * quota reservation interface */
1571 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1572 return inode->i_sb->dq_op->get_reserved_space(inode);
1575 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1577 spin_lock(&inode->i_lock);
1578 *inode_reserved_space(inode) += number;
1579 spin_unlock(&inode->i_lock);
1581 EXPORT_SYMBOL(inode_add_rsv_space);
1583 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1585 spin_lock(&inode->i_lock);
1586 *inode_reserved_space(inode) -= number;
1587 __inode_add_bytes(inode, number);
1588 spin_unlock(&inode->i_lock);
1590 EXPORT_SYMBOL(inode_claim_rsv_space);
1592 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1594 spin_lock(&inode->i_lock);
1595 *inode_reserved_space(inode) += number;
1596 __inode_sub_bytes(inode, number);
1597 spin_unlock(&inode->i_lock);
1599 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1601 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1603 spin_lock(&inode->i_lock);
1604 *inode_reserved_space(inode) -= number;
1605 spin_unlock(&inode->i_lock);
1607 EXPORT_SYMBOL(inode_sub_rsv_space);
1609 static qsize_t inode_get_rsv_space(struct inode *inode)
1613 if (!inode->i_sb->dq_op->get_reserved_space)
1615 spin_lock(&inode->i_lock);
1616 ret = *inode_reserved_space(inode);
1617 spin_unlock(&inode->i_lock);
1621 static void inode_incr_space(struct inode *inode, qsize_t number,
1625 inode_add_rsv_space(inode, number);
1627 inode_add_bytes(inode, number);
1630 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1633 inode_sub_rsv_space(inode, number);
1635 inode_sub_bytes(inode, number);
1639 * This functions updates i_blocks+i_bytes fields and quota information
1640 * (together with appropriate checks).
1642 * NOTE: We absolutely rely on the fact that caller dirties the inode
1643 * (usually helpers in quotaops.h care about this) and holds a handle for
1644 * the current transaction so that dquot write and inode write go into the
1649 * This operation can block, but only after everything is updated
1651 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1653 int cnt, ret = 0, index;
1654 struct dquot_warn warn[MAXQUOTAS];
1655 int reserve = flags & DQUOT_SPACE_RESERVE;
1656 struct dquot **dquots;
1658 if (!dquot_active(inode)) {
1659 inode_incr_space(inode, number, reserve);
1663 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1664 warn[cnt].w_type = QUOTA_NL_NOWARN;
1666 dquots = i_dquot(inode);
1667 index = srcu_read_lock(&dquot_srcu);
1668 spin_lock(&dq_data_lock);
1669 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1672 ret = check_bdq(dquots[cnt], number,
1673 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1674 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1675 spin_unlock(&dq_data_lock);
1676 goto out_flush_warn;
1679 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1683 dquot_resv_space(dquots[cnt], number);
1685 dquot_incr_space(dquots[cnt], number);
1687 inode_incr_space(inode, number, reserve);
1688 spin_unlock(&dq_data_lock);
1691 goto out_flush_warn;
1692 mark_all_dquot_dirty(dquots);
1694 srcu_read_unlock(&dquot_srcu, index);
1695 flush_warnings(warn);
1699 EXPORT_SYMBOL(__dquot_alloc_space);
1702 * This operation can block, but only after everything is updated
1704 int dquot_alloc_inode(struct inode *inode)
1706 int cnt, ret = 0, index;
1707 struct dquot_warn warn[MAXQUOTAS];
1708 struct dquot * const *dquots;
1710 if (!dquot_active(inode))
1712 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1713 warn[cnt].w_type = QUOTA_NL_NOWARN;
1715 dquots = i_dquot(inode);
1716 index = srcu_read_lock(&dquot_srcu);
1717 spin_lock(&dq_data_lock);
1718 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1721 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1726 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1729 dquot_incr_inodes(dquots[cnt], 1);
1733 spin_unlock(&dq_data_lock);
1735 mark_all_dquot_dirty(dquots);
1736 srcu_read_unlock(&dquot_srcu, index);
1737 flush_warnings(warn);
1740 EXPORT_SYMBOL(dquot_alloc_inode);
1743 * Convert in-memory reserved quotas to real consumed quotas
1745 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1747 struct dquot **dquots;
1750 if (!dquot_active(inode)) {
1751 inode_claim_rsv_space(inode, number);
1755 dquots = i_dquot(inode);
1756 index = srcu_read_lock(&dquot_srcu);
1757 spin_lock(&dq_data_lock);
1758 /* Claim reserved quotas to allocated quotas */
1759 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1761 dquot_claim_reserved_space(dquots[cnt], number);
1763 /* Update inode bytes */
1764 inode_claim_rsv_space(inode, number);
1765 spin_unlock(&dq_data_lock);
1766 mark_all_dquot_dirty(dquots);
1767 srcu_read_unlock(&dquot_srcu, index);
1770 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1773 * Convert allocated space back to in-memory reserved quotas
1775 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1777 struct dquot **dquots;
1780 if (!dquot_active(inode)) {
1781 inode_reclaim_rsv_space(inode, number);
1785 dquots = i_dquot(inode);
1786 index = srcu_read_lock(&dquot_srcu);
1787 spin_lock(&dq_data_lock);
1788 /* Claim reserved quotas to allocated quotas */
1789 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1791 dquot_reclaim_reserved_space(dquots[cnt], number);
1793 /* Update inode bytes */
1794 inode_reclaim_rsv_space(inode, number);
1795 spin_unlock(&dq_data_lock);
1796 mark_all_dquot_dirty(dquots);
1797 srcu_read_unlock(&dquot_srcu, index);
1800 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1803 * This operation can block, but only after everything is updated
1805 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1808 struct dquot_warn warn[MAXQUOTAS];
1809 struct dquot **dquots;
1810 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1812 if (!dquot_active(inode)) {
1813 inode_decr_space(inode, number, reserve);
1817 dquots = i_dquot(inode);
1818 index = srcu_read_lock(&dquot_srcu);
1819 spin_lock(&dq_data_lock);
1820 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1823 warn[cnt].w_type = QUOTA_NL_NOWARN;
1826 wtype = info_bdq_free(dquots[cnt], number);
1827 if (wtype != QUOTA_NL_NOWARN)
1828 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1830 dquot_free_reserved_space(dquots[cnt], number);
1832 dquot_decr_space(dquots[cnt], number);
1834 inode_decr_space(inode, number, reserve);
1835 spin_unlock(&dq_data_lock);
1839 mark_all_dquot_dirty(dquots);
1841 srcu_read_unlock(&dquot_srcu, index);
1842 flush_warnings(warn);
1844 EXPORT_SYMBOL(__dquot_free_space);
1847 * This operation can block, but only after everything is updated
1849 void dquot_free_inode(struct inode *inode)
1852 struct dquot_warn warn[MAXQUOTAS];
1853 struct dquot * const *dquots;
1856 if (!dquot_active(inode))
1859 dquots = i_dquot(inode);
1860 index = srcu_read_lock(&dquot_srcu);
1861 spin_lock(&dq_data_lock);
1862 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1865 warn[cnt].w_type = QUOTA_NL_NOWARN;
1868 wtype = info_idq_free(dquots[cnt], 1);
1869 if (wtype != QUOTA_NL_NOWARN)
1870 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1871 dquot_decr_inodes(dquots[cnt], 1);
1873 spin_unlock(&dq_data_lock);
1874 mark_all_dquot_dirty(dquots);
1875 srcu_read_unlock(&dquot_srcu, index);
1876 flush_warnings(warn);
1878 EXPORT_SYMBOL(dquot_free_inode);
1881 * Transfer the number of inode and blocks from one diskquota to an other.
1882 * On success, dquot references in transfer_to are consumed and references
1883 * to original dquots that need to be released are placed there. On failure,
1884 * references are kept untouched.
1886 * This operation can block, but only after everything is updated
1887 * A transaction must be started when entering this function.
1889 * We are holding reference on transfer_from & transfer_to, no need to
1890 * protect them by srcu_read_lock().
1892 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1894 qsize_t space, cur_space;
1895 qsize_t rsv_space = 0;
1896 struct dquot *transfer_from[MAXQUOTAS] = {};
1898 char is_valid[MAXQUOTAS] = {};
1899 struct dquot_warn warn_to[MAXQUOTAS];
1900 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1901 struct dquot_warn warn_from_space[MAXQUOTAS];
1903 if (IS_NOQUOTA(inode))
1905 /* Initialize the arrays */
1906 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1907 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1908 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1909 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1912 spin_lock(&dq_data_lock);
1913 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1914 spin_unlock(&dq_data_lock);
1917 cur_space = inode_get_bytes(inode);
1918 rsv_space = inode_get_rsv_space(inode);
1919 space = cur_space + rsv_space;
1920 /* Build the transfer_from list and check the limits */
1921 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1923 * Skip changes for same uid or gid or for turned off quota-type.
1925 if (!transfer_to[cnt])
1927 /* Avoid races with quotaoff() */
1928 if (!sb_has_quota_active(inode->i_sb, cnt))
1931 transfer_from[cnt] = i_dquot(inode)[cnt];
1932 ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1935 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1941 * Finally perform the needed transfer from transfer_from to transfer_to
1943 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1946 /* Due to IO error we might not have transfer_from[] structure */
1947 if (transfer_from[cnt]) {
1949 wtype = info_idq_free(transfer_from[cnt], 1);
1950 if (wtype != QUOTA_NL_NOWARN)
1951 prepare_warning(&warn_from_inodes[cnt],
1952 transfer_from[cnt], wtype);
1953 wtype = info_bdq_free(transfer_from[cnt], space);
1954 if (wtype != QUOTA_NL_NOWARN)
1955 prepare_warning(&warn_from_space[cnt],
1956 transfer_from[cnt], wtype);
1957 dquot_decr_inodes(transfer_from[cnt], 1);
1958 dquot_decr_space(transfer_from[cnt], cur_space);
1959 dquot_free_reserved_space(transfer_from[cnt],
1963 dquot_incr_inodes(transfer_to[cnt], 1);
1964 dquot_incr_space(transfer_to[cnt], cur_space);
1965 dquot_resv_space(transfer_to[cnt], rsv_space);
1967 i_dquot(inode)[cnt] = transfer_to[cnt];
1969 spin_unlock(&dq_data_lock);
1971 mark_all_dquot_dirty(transfer_from);
1972 mark_all_dquot_dirty(transfer_to);
1973 flush_warnings(warn_to);
1974 flush_warnings(warn_from_inodes);
1975 flush_warnings(warn_from_space);
1976 /* Pass back references to put */
1977 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1979 transfer_to[cnt] = transfer_from[cnt];
1982 spin_unlock(&dq_data_lock);
1983 flush_warnings(warn_to);
1986 EXPORT_SYMBOL(__dquot_transfer);
1988 /* Wrapper for transferring ownership of an inode for uid/gid only
1989 * Called from FSXXX_setattr()
1991 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1993 struct dquot *transfer_to[MAXQUOTAS] = {};
1994 struct dquot *dquot;
1995 struct super_block *sb = inode->i_sb;
1998 if (!dquot_active(inode))
2001 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2002 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2003 if (IS_ERR(dquot)) {
2004 if (PTR_ERR(dquot) != -ESRCH) {
2005 ret = PTR_ERR(dquot);
2010 transfer_to[USRQUOTA] = dquot;
2012 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2013 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2014 if (IS_ERR(dquot)) {
2015 if (PTR_ERR(dquot) != -ESRCH) {
2016 ret = PTR_ERR(dquot);
2021 transfer_to[GRPQUOTA] = dquot;
2023 ret = __dquot_transfer(inode, transfer_to);
2025 dqput_all(transfer_to);
2028 EXPORT_SYMBOL(dquot_transfer);
2031 * Write info of quota file to disk
2033 int dquot_commit_info(struct super_block *sb, int type)
2036 struct quota_info *dqopt = sb_dqopt(sb);
2038 mutex_lock(&dqopt->dqio_mutex);
2039 ret = dqopt->ops[type]->write_file_info(sb, type);
2040 mutex_unlock(&dqopt->dqio_mutex);
2043 EXPORT_SYMBOL(dquot_commit_info);
2045 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2047 struct quota_info *dqopt = sb_dqopt(sb);
2050 if (!dqopt->ops[qid->type]->get_next_id)
2052 mutex_lock(&dqopt->dqio_mutex);
2053 err = dqopt->ops[qid->type]->get_next_id(sb, qid);
2054 mutex_unlock(&dqopt->dqio_mutex);
2058 EXPORT_SYMBOL(dquot_get_next_id);
2061 * Definitions of diskquota operations.
2063 const struct dquot_operations dquot_operations = {
2064 .write_dquot = dquot_commit,
2065 .acquire_dquot = dquot_acquire,
2066 .release_dquot = dquot_release,
2067 .mark_dirty = dquot_mark_dquot_dirty,
2068 .write_info = dquot_commit_info,
2069 .alloc_dquot = dquot_alloc,
2070 .destroy_dquot = dquot_destroy,
2071 .get_next_id = dquot_get_next_id,
2073 EXPORT_SYMBOL(dquot_operations);
2076 * Generic helper for ->open on filesystems supporting disk quotas.
2078 int dquot_file_open(struct inode *inode, struct file *file)
2082 error = generic_file_open(inode, file);
2083 if (!error && (file->f_mode & FMODE_WRITE))
2084 dquot_initialize(inode);
2087 EXPORT_SYMBOL(dquot_file_open);
2090 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2092 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2095 struct quota_info *dqopt = sb_dqopt(sb);
2096 struct inode *toputinode[MAXQUOTAS];
2098 /* Cannot turn off usage accounting without turning off limits, or
2099 * suspend quotas and simultaneously turn quotas off. */
2100 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2101 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2102 DQUOT_USAGE_ENABLED)))
2105 /* We need to serialize quota_off() for device */
2106 mutex_lock(&dqopt->dqonoff_mutex);
2109 * Skip everything if there's nothing to do. We have to do this because
2110 * sometimes we are called when fill_super() failed and calling
2111 * sync_fs() in such cases does no good.
2113 if (!sb_any_quota_loaded(sb)) {
2114 mutex_unlock(&dqopt->dqonoff_mutex);
2117 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2118 toputinode[cnt] = NULL;
2119 if (type != -1 && cnt != type)
2121 if (!sb_has_quota_loaded(sb, cnt))
2124 if (flags & DQUOT_SUSPENDED) {
2125 spin_lock(&dq_state_lock);
2127 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2128 spin_unlock(&dq_state_lock);
2130 spin_lock(&dq_state_lock);
2131 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2132 /* Turning off suspended quotas? */
2133 if (!sb_has_quota_loaded(sb, cnt) &&
2134 sb_has_quota_suspended(sb, cnt)) {
2135 dqopt->flags &= ~dquot_state_flag(
2136 DQUOT_SUSPENDED, cnt);
2137 spin_unlock(&dq_state_lock);
2138 iput(dqopt->files[cnt]);
2139 dqopt->files[cnt] = NULL;
2142 spin_unlock(&dq_state_lock);
2145 /* We still have to keep quota loaded? */
2146 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2149 /* Note: these are blocking operations */
2150 drop_dquot_ref(sb, cnt);
2151 invalidate_dquots(sb, cnt);
2153 * Now all dquots should be invalidated, all writes done so we
2154 * should be only users of the info. No locks needed.
2156 if (info_dirty(&dqopt->info[cnt]))
2157 sb->dq_op->write_info(sb, cnt);
2158 if (dqopt->ops[cnt]->free_file_info)
2159 dqopt->ops[cnt]->free_file_info(sb, cnt);
2160 put_quota_format(dqopt->info[cnt].dqi_format);
2162 toputinode[cnt] = dqopt->files[cnt];
2163 if (!sb_has_quota_loaded(sb, cnt))
2164 dqopt->files[cnt] = NULL;
2165 dqopt->info[cnt].dqi_flags = 0;
2166 dqopt->info[cnt].dqi_igrace = 0;
2167 dqopt->info[cnt].dqi_bgrace = 0;
2168 dqopt->ops[cnt] = NULL;
2170 mutex_unlock(&dqopt->dqonoff_mutex);
2172 /* Skip syncing and setting flags if quota files are hidden */
2173 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2176 /* Sync the superblock so that buffers with quota data are written to
2177 * disk (and so userspace sees correct data afterwards). */
2178 if (sb->s_op->sync_fs)
2179 sb->s_op->sync_fs(sb, 1);
2180 sync_blockdev(sb->s_bdev);
2181 /* Now the quota files are just ordinary files and we can set the
2182 * inode flags back. Moreover we discard the pagecache so that
2183 * userspace sees the writes we did bypassing the pagecache. We
2184 * must also discard the blockdev buffers so that we see the
2185 * changes done by userspace on the next quotaon() */
2186 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2187 if (toputinode[cnt]) {
2188 mutex_lock(&dqopt->dqonoff_mutex);
2189 /* If quota was reenabled in the meantime, we have
2191 if (!sb_has_quota_loaded(sb, cnt)) {
2192 inode_lock(toputinode[cnt]);
2193 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2194 S_NOATIME | S_NOQUOTA);
2195 truncate_inode_pages(&toputinode[cnt]->i_data,
2197 inode_unlock(toputinode[cnt]);
2198 mark_inode_dirty_sync(toputinode[cnt]);
2200 mutex_unlock(&dqopt->dqonoff_mutex);
2203 invalidate_bdev(sb->s_bdev);
2205 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2206 if (toputinode[cnt]) {
2207 /* On remount RO, we keep the inode pointer so that we
2208 * can reenable quota on the subsequent remount RW. We
2209 * have to check 'flags' variable and not use sb_has_
2210 * function because another quotaon / quotaoff could
2211 * change global state before we got here. We refuse
2212 * to suspend quotas when there is pending delete on
2213 * the quota file... */
2214 if (!(flags & DQUOT_SUSPENDED))
2215 iput(toputinode[cnt]);
2216 else if (!toputinode[cnt]->i_nlink)
2221 EXPORT_SYMBOL(dquot_disable);
2223 int dquot_quota_off(struct super_block *sb, int type)
2225 return dquot_disable(sb, type,
2226 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2228 EXPORT_SYMBOL(dquot_quota_off);
2231 * Turn quotas on on a device
2235 * Helper function to turn quotas on when we already have the inode of
2236 * quota file and no quota information is loaded.
2238 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2241 struct quota_format_type *fmt = find_quota_format(format_id);
2242 struct super_block *sb = inode->i_sb;
2243 struct quota_info *dqopt = sb_dqopt(sb);
2249 if (!S_ISREG(inode->i_mode)) {
2253 if (IS_RDONLY(inode)) {
2257 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2258 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2262 /* Usage always has to be set... */
2263 if (!(flags & DQUOT_USAGE_ENABLED)) {
2268 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2269 /* As we bypass the pagecache we must now flush all the
2270 * dirty data and invalidate caches so that kernel sees
2271 * changes from userspace. It is not enough to just flush
2272 * the quota file since if blocksize < pagesize, invalidation
2273 * of the cache could fail because of other unrelated dirty
2275 sync_filesystem(sb);
2276 invalidate_bdev(sb->s_bdev);
2278 mutex_lock(&dqopt->dqonoff_mutex);
2279 if (sb_has_quota_loaded(sb, type)) {
2284 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2285 /* We don't want quota and atime on quota files (deadlocks
2286 * possible) Also nobody should write to the file - we use
2287 * special IO operations which ignore the immutable bit. */
2289 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2291 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2292 inode_unlock(inode);
2294 * When S_NOQUOTA is set, remove dquot references as no more
2295 * references can be added
2297 __dquot_drop(inode);
2301 dqopt->files[type] = igrab(inode);
2302 if (!dqopt->files[type])
2305 if (!fmt->qf_ops->check_quota_file(sb, type))
2308 dqopt->ops[type] = fmt->qf_ops;
2309 dqopt->info[type].dqi_format = fmt;
2310 dqopt->info[type].dqi_fmt_id = format_id;
2311 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2312 mutex_lock(&dqopt->dqio_mutex);
2313 error = dqopt->ops[type]->read_file_info(sb, type);
2315 mutex_unlock(&dqopt->dqio_mutex);
2318 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2319 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2320 mutex_unlock(&dqopt->dqio_mutex);
2321 spin_lock(&dq_state_lock);
2322 dqopt->flags |= dquot_state_flag(flags, type);
2323 spin_unlock(&dq_state_lock);
2325 add_dquot_ref(sb, type);
2326 mutex_unlock(&dqopt->dqonoff_mutex);
2331 dqopt->files[type] = NULL;
2334 if (oldflags != -1) {
2336 /* Set the flags back (in the case of accidental quotaon()
2337 * on a wrong file we don't want to mess up the flags) */
2338 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2339 inode->i_flags |= oldflags;
2340 inode_unlock(inode);
2342 mutex_unlock(&dqopt->dqonoff_mutex);
2344 put_quota_format(fmt);
2349 /* Reenable quotas on remount RW */
2350 int dquot_resume(struct super_block *sb, int type)
2352 struct quota_info *dqopt = sb_dqopt(sb);
2353 struct inode *inode;
2357 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2358 if (type != -1 && cnt != type)
2361 mutex_lock(&dqopt->dqonoff_mutex);
2362 if (!sb_has_quota_suspended(sb, cnt)) {
2363 mutex_unlock(&dqopt->dqonoff_mutex);
2366 inode = dqopt->files[cnt];
2367 dqopt->files[cnt] = NULL;
2368 spin_lock(&dq_state_lock);
2369 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2370 DQUOT_LIMITS_ENABLED,
2372 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2373 spin_unlock(&dq_state_lock);
2374 mutex_unlock(&dqopt->dqonoff_mutex);
2376 flags = dquot_generic_flag(flags, cnt);
2377 ret = vfs_load_quota_inode(inode, cnt,
2378 dqopt->info[cnt].dqi_fmt_id, flags);
2384 EXPORT_SYMBOL(dquot_resume);
2386 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2389 int error = security_quota_on(path->dentry);
2392 /* Quota file not on the same filesystem? */
2393 if (path->dentry->d_sb != sb)
2396 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2397 format_id, DQUOT_USAGE_ENABLED |
2398 DQUOT_LIMITS_ENABLED);
2401 EXPORT_SYMBOL(dquot_quota_on);
2404 * More powerful function for turning on quotas allowing setting
2405 * of individual quota flags
2407 int dquot_enable(struct inode *inode, int type, int format_id,
2411 struct super_block *sb = inode->i_sb;
2412 struct quota_info *dqopt = sb_dqopt(sb);
2414 /* Just unsuspend quotas? */
2415 BUG_ON(flags & DQUOT_SUSPENDED);
2419 /* Just updating flags needed? */
2420 if (sb_has_quota_loaded(sb, type)) {
2421 mutex_lock(&dqopt->dqonoff_mutex);
2422 /* Now do a reliable test... */
2423 if (!sb_has_quota_loaded(sb, type)) {
2424 mutex_unlock(&dqopt->dqonoff_mutex);
2427 if (flags & DQUOT_USAGE_ENABLED &&
2428 sb_has_quota_usage_enabled(sb, type)) {
2432 if (flags & DQUOT_LIMITS_ENABLED &&
2433 sb_has_quota_limits_enabled(sb, type)) {
2437 spin_lock(&dq_state_lock);
2438 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2439 spin_unlock(&dq_state_lock);
2441 mutex_unlock(&dqopt->dqonoff_mutex);
2446 return vfs_load_quota_inode(inode, type, format_id, flags);
2448 EXPORT_SYMBOL(dquot_enable);
2451 * This function is used when filesystem needs to initialize quotas
2452 * during mount time.
2454 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2455 int format_id, int type)
2457 struct dentry *dentry;
2460 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2462 return PTR_ERR(dentry);
2464 if (d_really_is_negative(dentry)) {
2469 error = security_quota_on(dentry);
2471 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2472 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2478 EXPORT_SYMBOL(dquot_quota_on_mount);
2480 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2484 struct quota_info *dqopt = sb_dqopt(sb);
2486 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2488 /* Accounting cannot be turned on while fs is mounted */
2489 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2492 for (type = 0; type < MAXQUOTAS; type++) {
2493 if (!(flags & qtype_enforce_flag(type)))
2495 /* Can't enforce without accounting */
2496 if (!sb_has_quota_usage_enabled(sb, type))
2498 ret = dquot_enable(dqopt->files[type], type,
2499 dqopt->info[type].dqi_fmt_id,
2500 DQUOT_LIMITS_ENABLED);
2506 /* Backout enforcement enablement we already did */
2507 for (type--; type >= 0; type--) {
2508 if (flags & qtype_enforce_flag(type))
2509 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2511 /* Error code translation for better compatibility with XFS */
2517 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2521 struct quota_info *dqopt = sb_dqopt(sb);
2523 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2526 * We don't support turning off accounting via quotactl. In principle
2527 * quota infrastructure can do this but filesystems don't expect
2528 * userspace to be able to do it.
2531 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2534 /* Filter out limits not enabled */
2535 for (type = 0; type < MAXQUOTAS; type++)
2536 if (!sb_has_quota_limits_enabled(sb, type))
2537 flags &= ~qtype_enforce_flag(type);
2541 for (type = 0; type < MAXQUOTAS; type++) {
2542 if (flags & qtype_enforce_flag(type)) {
2543 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2550 /* Backout enforcement disabling we already did */
2551 for (type--; type >= 0; type--) {
2552 if (flags & qtype_enforce_flag(type))
2553 dquot_enable(dqopt->files[type], type,
2554 dqopt->info[type].dqi_fmt_id,
2555 DQUOT_LIMITS_ENABLED);
2560 /* Generic routine for getting common part of quota structure */
2561 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2563 struct mem_dqblk *dm = &dquot->dq_dqb;
2565 memset(di, 0, sizeof(*di));
2566 spin_lock(&dq_data_lock);
2567 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2568 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2569 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2570 di->d_ino_softlimit = dm->dqb_isoftlimit;
2571 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2572 di->d_ino_count = dm->dqb_curinodes;
2573 di->d_spc_timer = dm->dqb_btime;
2574 di->d_ino_timer = dm->dqb_itime;
2575 spin_unlock(&dq_data_lock);
2578 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2579 struct qc_dqblk *di)
2581 struct dquot *dquot;
2583 dquot = dqget(sb, qid);
2585 return PTR_ERR(dquot);
2586 do_get_dqblk(dquot, di);
2591 EXPORT_SYMBOL(dquot_get_dqblk);
2593 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2594 struct qc_dqblk *di)
2596 struct dquot *dquot;
2599 if (!sb->dq_op->get_next_id)
2601 err = sb->dq_op->get_next_id(sb, qid);
2604 dquot = dqget(sb, *qid);
2606 return PTR_ERR(dquot);
2607 do_get_dqblk(dquot, di);
2612 EXPORT_SYMBOL(dquot_get_next_dqblk);
2614 #define VFS_QC_MASK \
2615 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2616 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2617 QC_SPC_TIMER | QC_INO_TIMER)
2619 /* Generic routine for setting common part of quota structure */
2620 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2622 struct mem_dqblk *dm = &dquot->dq_dqb;
2623 int check_blim = 0, check_ilim = 0;
2624 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2626 if (di->d_fieldmask & ~VFS_QC_MASK)
2629 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2630 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2631 ((di->d_fieldmask & QC_SPC_HARD) &&
2632 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2633 ((di->d_fieldmask & QC_INO_SOFT) &&
2634 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2635 ((di->d_fieldmask & QC_INO_HARD) &&
2636 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2639 spin_lock(&dq_data_lock);
2640 if (di->d_fieldmask & QC_SPACE) {
2641 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2643 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2646 if (di->d_fieldmask & QC_SPC_SOFT)
2647 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2648 if (di->d_fieldmask & QC_SPC_HARD)
2649 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2650 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2652 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2655 if (di->d_fieldmask & QC_INO_COUNT) {
2656 dm->dqb_curinodes = di->d_ino_count;
2658 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2661 if (di->d_fieldmask & QC_INO_SOFT)
2662 dm->dqb_isoftlimit = di->d_ino_softlimit;
2663 if (di->d_fieldmask & QC_INO_HARD)
2664 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2665 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2667 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2670 if (di->d_fieldmask & QC_SPC_TIMER) {
2671 dm->dqb_btime = di->d_spc_timer;
2673 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2676 if (di->d_fieldmask & QC_INO_TIMER) {
2677 dm->dqb_itime = di->d_ino_timer;
2679 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2683 if (!dm->dqb_bsoftlimit ||
2684 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2686 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2687 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2688 /* Set grace only if user hasn't provided his own... */
2689 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2692 if (!dm->dqb_isoftlimit ||
2693 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2695 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2696 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2697 /* Set grace only if user hasn't provided his own... */
2698 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2700 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2702 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2704 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2705 spin_unlock(&dq_data_lock);
2706 mark_dquot_dirty(dquot);
2711 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2712 struct qc_dqblk *di)
2714 struct dquot *dquot;
2717 dquot = dqget(sb, qid);
2718 if (IS_ERR(dquot)) {
2719 rc = PTR_ERR(dquot);
2722 rc = do_set_dqblk(dquot, di);
2727 EXPORT_SYMBOL(dquot_set_dqblk);
2729 /* Generic routine for getting common part of quota file information */
2730 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2732 struct mem_dqinfo *mi;
2733 struct qc_type_state *tstate;
2734 struct quota_info *dqopt = sb_dqopt(sb);
2737 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2738 memset(state, 0, sizeof(*state));
2739 for (type = 0; type < MAXQUOTAS; type++) {
2740 if (!sb_has_quota_active(sb, type))
2742 tstate = state->s_state + type;
2743 mi = sb_dqopt(sb)->info + type;
2744 tstate->flags = QCI_ACCT_ENABLED;
2745 spin_lock(&dq_data_lock);
2746 if (mi->dqi_flags & DQF_SYS_FILE)
2747 tstate->flags |= QCI_SYSFILE;
2748 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2749 tstate->flags |= QCI_ROOT_SQUASH;
2750 if (sb_has_quota_limits_enabled(sb, type))
2751 tstate->flags |= QCI_LIMITS_ENFORCED;
2752 tstate->spc_timelimit = mi->dqi_bgrace;
2753 tstate->ino_timelimit = mi->dqi_igrace;
2754 tstate->ino = dqopt->files[type]->i_ino;
2755 tstate->blocks = dqopt->files[type]->i_blocks;
2756 tstate->nextents = 1; /* We don't know... */
2757 spin_unlock(&dq_data_lock);
2759 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2762 EXPORT_SYMBOL(dquot_get_state);
2764 /* Generic routine for setting common part of quota file information */
2765 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2767 struct mem_dqinfo *mi;
2770 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2771 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2773 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2774 if (!sb_has_quota_active(sb, type)) {
2778 mi = sb_dqopt(sb)->info + type;
2779 if (ii->i_fieldmask & QC_FLAGS) {
2780 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2781 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD)) {
2786 spin_lock(&dq_data_lock);
2787 if (ii->i_fieldmask & QC_SPC_TIMER)
2788 mi->dqi_bgrace = ii->i_spc_timelimit;
2789 if (ii->i_fieldmask & QC_INO_TIMER)
2790 mi->dqi_igrace = ii->i_ino_timelimit;
2791 if (ii->i_fieldmask & QC_FLAGS) {
2792 if (ii->i_flags & QCI_ROOT_SQUASH)
2793 mi->dqi_flags |= DQF_ROOT_SQUASH;
2795 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2797 spin_unlock(&dq_data_lock);
2798 mark_info_dirty(sb, type);
2799 /* Force write to disk */
2800 sb->dq_op->write_info(sb, type);
2802 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2805 EXPORT_SYMBOL(dquot_set_dqinfo);
2807 const struct quotactl_ops dquot_quotactl_ops = {
2808 .quota_on = dquot_quota_on,
2809 .quota_off = dquot_quota_off,
2810 .quota_sync = dquot_quota_sync,
2811 .get_state = dquot_get_state,
2812 .set_info = dquot_set_dqinfo,
2813 .get_dqblk = dquot_get_dqblk,
2814 .get_nextdqblk = dquot_get_next_dqblk,
2815 .set_dqblk = dquot_set_dqblk
2817 EXPORT_SYMBOL(dquot_quotactl_ops);
2819 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2820 .quota_enable = dquot_quota_enable,
2821 .quota_disable = dquot_quota_disable,
2822 .quota_sync = dquot_quota_sync,
2823 .get_state = dquot_get_state,
2824 .set_info = dquot_set_dqinfo,
2825 .get_dqblk = dquot_get_dqblk,
2826 .get_nextdqblk = dquot_get_next_dqblk,
2827 .set_dqblk = dquot_set_dqblk
2829 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2831 static int do_proc_dqstats(struct ctl_table *table, int write,
2832 void __user *buffer, size_t *lenp, loff_t *ppos)
2834 unsigned int type = (int *)table->data - dqstats.stat;
2836 /* Update global table */
2837 dqstats.stat[type] =
2838 percpu_counter_sum_positive(&dqstats.counter[type]);
2839 return proc_dointvec(table, write, buffer, lenp, ppos);
2842 static struct ctl_table fs_dqstats_table[] = {
2844 .procname = "lookups",
2845 .data = &dqstats.stat[DQST_LOOKUPS],
2846 .maxlen = sizeof(int),
2848 .proc_handler = do_proc_dqstats,
2851 .procname = "drops",
2852 .data = &dqstats.stat[DQST_DROPS],
2853 .maxlen = sizeof(int),
2855 .proc_handler = do_proc_dqstats,
2858 .procname = "reads",
2859 .data = &dqstats.stat[DQST_READS],
2860 .maxlen = sizeof(int),
2862 .proc_handler = do_proc_dqstats,
2865 .procname = "writes",
2866 .data = &dqstats.stat[DQST_WRITES],
2867 .maxlen = sizeof(int),
2869 .proc_handler = do_proc_dqstats,
2872 .procname = "cache_hits",
2873 .data = &dqstats.stat[DQST_CACHE_HITS],
2874 .maxlen = sizeof(int),
2876 .proc_handler = do_proc_dqstats,
2879 .procname = "allocated_dquots",
2880 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2881 .maxlen = sizeof(int),
2883 .proc_handler = do_proc_dqstats,
2886 .procname = "free_dquots",
2887 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2888 .maxlen = sizeof(int),
2890 .proc_handler = do_proc_dqstats,
2893 .procname = "syncs",
2894 .data = &dqstats.stat[DQST_SYNCS],
2895 .maxlen = sizeof(int),
2897 .proc_handler = do_proc_dqstats,
2899 #ifdef CONFIG_PRINT_QUOTA_WARNING
2901 .procname = "warnings",
2902 .data = &flag_print_warnings,
2903 .maxlen = sizeof(int),
2905 .proc_handler = proc_dointvec,
2911 static struct ctl_table fs_table[] = {
2913 .procname = "quota",
2915 .child = fs_dqstats_table,
2920 static struct ctl_table sys_table[] = {
2929 static int __init dquot_init(void)
2932 unsigned long nr_hash, order;
2934 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2936 register_sysctl_table(sys_table);
2938 dquot_cachep = kmem_cache_create("dquot",
2939 sizeof(struct dquot), sizeof(unsigned long) * 4,
2940 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2941 SLAB_MEM_SPREAD|SLAB_PANIC),
2945 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2947 panic("Cannot create dquot hash table");
2949 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2950 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2952 panic("Cannot create dquot stat counters");
2955 /* Find power-of-two hlist_heads which can fit into allocation */
2956 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2960 } while (nr_hash >> dq_hash_bits);
2963 nr_hash = 1UL << dq_hash_bits;
2964 dq_hash_mask = nr_hash - 1;
2965 for (i = 0; i < nr_hash; i++)
2966 INIT_HLIST_HEAD(dquot_hash + i);
2968 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2969 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2971 register_shrinker(&dqcache_shrinker);
2975 fs_initcall(dquot_init);