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 void __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 set_bit(DQ_READ_B, &dquot->dq_flags);
415 /* Instantiate dquot if needed */
416 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
417 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
418 /* Write the info if needed */
419 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
420 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
421 dquot->dq_sb, dquot->dq_id.type);
430 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
432 mutex_unlock(&dqopt->dqio_mutex);
433 mutex_unlock(&dquot->dq_lock);
436 EXPORT_SYMBOL(dquot_acquire);
439 * Write dquot to disk
441 int dquot_commit(struct dquot *dquot)
444 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
446 mutex_lock(&dqopt->dqio_mutex);
447 spin_lock(&dq_list_lock);
448 if (!clear_dquot_dirty(dquot)) {
449 spin_unlock(&dq_list_lock);
452 spin_unlock(&dq_list_lock);
453 /* Inactive dquot can be only if there was error during read/init
454 * => we have better not writing it */
455 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
456 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
460 mutex_unlock(&dqopt->dqio_mutex);
463 EXPORT_SYMBOL(dquot_commit);
468 int dquot_release(struct dquot *dquot)
470 int ret = 0, ret2 = 0;
471 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
473 mutex_lock(&dquot->dq_lock);
474 /* Check whether we are not racing with some other dqget() */
475 if (atomic_read(&dquot->dq_count) > 1)
477 mutex_lock(&dqopt->dqio_mutex);
478 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
479 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
481 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
482 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
483 dquot->dq_sb, dquot->dq_id.type);
488 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
489 mutex_unlock(&dqopt->dqio_mutex);
491 mutex_unlock(&dquot->dq_lock);
494 EXPORT_SYMBOL(dquot_release);
496 void dquot_destroy(struct dquot *dquot)
498 kmem_cache_free(dquot_cachep, dquot);
500 EXPORT_SYMBOL(dquot_destroy);
502 static inline void do_destroy_dquot(struct dquot *dquot)
504 dquot->dq_sb->dq_op->destroy_dquot(dquot);
507 /* Invalidate all dquots on the list. Note that this function is called after
508 * quota is disabled and pointers from inodes removed so there cannot be new
509 * quota users. There can still be some users of quotas due to inodes being
510 * just deleted or pruned by prune_icache() (those are not attached to any
511 * list) or parallel quotactl call. We have to wait for such users.
513 static void invalidate_dquots(struct super_block *sb, int type)
515 struct dquot *dquot, *tmp;
518 spin_lock(&dq_list_lock);
519 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
520 if (dquot->dq_sb != sb)
522 if (dquot->dq_id.type != type)
524 /* Wait for dquot users */
525 if (atomic_read(&dquot->dq_count)) {
529 prepare_to_wait(&dquot->dq_wait_unused, &wait,
530 TASK_UNINTERRUPTIBLE);
531 spin_unlock(&dq_list_lock);
532 /* Once dqput() wakes us up, we know it's time to free
534 * IMPORTANT: we rely on the fact that there is always
535 * at most one process waiting for dquot to free.
536 * Otherwise dq_count would be > 1 and we would never
539 if (atomic_read(&dquot->dq_count) > 1)
541 finish_wait(&dquot->dq_wait_unused, &wait);
543 /* At this moment dquot() need not exist (it could be
544 * reclaimed by prune_dqcache(). Hence we must
549 * Quota now has no users and it has been written on last
552 remove_dquot_hash(dquot);
553 remove_free_dquot(dquot);
555 do_destroy_dquot(dquot);
557 spin_unlock(&dq_list_lock);
560 /* Call callback for every active dquot on given filesystem */
561 int dquot_scan_active(struct super_block *sb,
562 int (*fn)(struct dquot *dquot, unsigned long priv),
565 struct dquot *dquot, *old_dquot = NULL;
568 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
569 spin_lock(&dq_list_lock);
570 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
571 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
573 if (dquot->dq_sb != sb)
575 /* Now we have active dquot so we can just increase use count */
576 atomic_inc(&dquot->dq_count);
577 spin_unlock(&dq_list_lock);
578 dqstats_inc(DQST_LOOKUPS);
582 * ->release_dquot() can be racing with us. Our reference
583 * protects us from new calls to it so just wait for any
584 * outstanding call and recheck the DQ_ACTIVE_B after that.
586 wait_on_dquot(dquot);
587 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
588 ret = fn(dquot, priv);
592 spin_lock(&dq_list_lock);
593 /* We are safe to continue now because our dquot could not
594 * be moved out of the inuse list while we hold the reference */
596 spin_unlock(&dq_list_lock);
599 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
602 EXPORT_SYMBOL(dquot_scan_active);
604 /* Write all dquot structures to quota files */
605 int dquot_writeback_dquots(struct super_block *sb, int type)
607 struct list_head *dirty;
609 struct quota_info *dqopt = sb_dqopt(sb);
613 mutex_lock(&dqopt->dqonoff_mutex);
614 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
615 if (type != -1 && cnt != type)
617 if (!sb_has_quota_active(sb, cnt))
619 spin_lock(&dq_list_lock);
620 dirty = &dqopt->info[cnt].dqi_dirty_list;
621 while (!list_empty(dirty)) {
622 dquot = list_first_entry(dirty, struct dquot,
624 /* Dirty and inactive can be only bad dquot... */
625 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
626 clear_dquot_dirty(dquot);
629 /* Now we have active dquot from which someone is
630 * holding reference so we can safely just increase
633 spin_unlock(&dq_list_lock);
634 dqstats_inc(DQST_LOOKUPS);
635 err = sb->dq_op->write_dquot(dquot);
639 spin_lock(&dq_list_lock);
641 spin_unlock(&dq_list_lock);
644 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
645 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
646 && info_dirty(&dqopt->info[cnt]))
647 sb->dq_op->write_info(sb, cnt);
648 dqstats_inc(DQST_SYNCS);
649 mutex_unlock(&dqopt->dqonoff_mutex);
653 EXPORT_SYMBOL(dquot_writeback_dquots);
655 /* Write all dquot structures to disk and make them visible from userspace */
656 int dquot_quota_sync(struct super_block *sb, int type)
658 struct quota_info *dqopt = sb_dqopt(sb);
662 ret = dquot_writeback_dquots(sb, type);
665 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
668 /* This is not very clever (and fast) but currently I don't know about
669 * any other simple way of getting quota data to disk and we must get
670 * them there for userspace to be visible... */
671 if (sb->s_op->sync_fs)
672 sb->s_op->sync_fs(sb, 1);
673 sync_blockdev(sb->s_bdev);
676 * Now when everything is written we can discard the pagecache so
677 * that userspace sees the changes.
679 mutex_lock(&dqopt->dqonoff_mutex);
680 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
681 if (type != -1 && cnt != type)
683 if (!sb_has_quota_active(sb, cnt))
685 mutex_lock(&dqopt->files[cnt]->i_mutex);
686 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
687 mutex_unlock(&dqopt->files[cnt]->i_mutex);
689 mutex_unlock(&dqopt->dqonoff_mutex);
693 EXPORT_SYMBOL(dquot_quota_sync);
696 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
698 struct list_head *head;
700 unsigned long freed = 0;
702 spin_lock(&dq_list_lock);
703 head = free_dquots.prev;
704 while (head != &free_dquots && sc->nr_to_scan) {
705 dquot = list_entry(head, struct dquot, dq_free);
706 remove_dquot_hash(dquot);
707 remove_free_dquot(dquot);
709 do_destroy_dquot(dquot);
712 head = free_dquots.prev;
714 spin_unlock(&dq_list_lock);
719 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
721 return vfs_pressure_ratio(
722 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
725 static struct shrinker dqcache_shrinker = {
726 .count_objects = dqcache_shrink_count,
727 .scan_objects = dqcache_shrink_scan,
728 .seeks = DEFAULT_SEEKS,
732 * Put reference to dquot
734 void dqput(struct dquot *dquot)
740 #ifdef CONFIG_QUOTA_DEBUG
741 if (!atomic_read(&dquot->dq_count)) {
742 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
743 quotatypes[dquot->dq_id.type],
744 from_kqid(&init_user_ns, dquot->dq_id));
748 dqstats_inc(DQST_DROPS);
750 spin_lock(&dq_list_lock);
751 if (atomic_read(&dquot->dq_count) > 1) {
752 /* We have more than one user... nothing to do */
753 atomic_dec(&dquot->dq_count);
754 /* Releasing dquot during quotaoff phase? */
755 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
756 atomic_read(&dquot->dq_count) == 1)
757 wake_up(&dquot->dq_wait_unused);
758 spin_unlock(&dq_list_lock);
761 /* Need to release dquot? */
762 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
763 spin_unlock(&dq_list_lock);
764 /* Commit dquot before releasing */
765 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
767 quota_error(dquot->dq_sb, "Can't write quota structure"
768 " (error %d). Quota may get out of sync!",
771 * We clear dirty bit anyway, so that we avoid
774 spin_lock(&dq_list_lock);
775 clear_dquot_dirty(dquot);
776 spin_unlock(&dq_list_lock);
780 /* Clear flag in case dquot was inactive (something bad happened) */
781 clear_dquot_dirty(dquot);
782 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
783 spin_unlock(&dq_list_lock);
784 dquot->dq_sb->dq_op->release_dquot(dquot);
787 atomic_dec(&dquot->dq_count);
788 #ifdef CONFIG_QUOTA_DEBUG
790 BUG_ON(!list_empty(&dquot->dq_free));
792 put_dquot_last(dquot);
793 spin_unlock(&dq_list_lock);
795 EXPORT_SYMBOL(dqput);
797 struct dquot *dquot_alloc(struct super_block *sb, int type)
799 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
801 EXPORT_SYMBOL(dquot_alloc);
803 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
807 dquot = sb->dq_op->alloc_dquot(sb, type);
811 mutex_init(&dquot->dq_lock);
812 INIT_LIST_HEAD(&dquot->dq_free);
813 INIT_LIST_HEAD(&dquot->dq_inuse);
814 INIT_HLIST_NODE(&dquot->dq_hash);
815 INIT_LIST_HEAD(&dquot->dq_dirty);
816 init_waitqueue_head(&dquot->dq_wait_unused);
818 dquot->dq_id = make_kqid_invalid(type);
819 atomic_set(&dquot->dq_count, 1);
825 * Get reference to dquot
827 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
828 * destroying our dquot by:
829 * a) checking for quota flags under dq_list_lock and
830 * b) getting a reference to dquot before we release dq_list_lock
832 struct dquot *dqget(struct super_block *sb, struct kqid qid)
834 unsigned int hashent = hashfn(sb, qid);
835 struct dquot *dquot = NULL, *empty = NULL;
837 if (!sb_has_quota_active(sb, qid.type))
840 spin_lock(&dq_list_lock);
841 spin_lock(&dq_state_lock);
842 if (!sb_has_quota_active(sb, qid.type)) {
843 spin_unlock(&dq_state_lock);
844 spin_unlock(&dq_list_lock);
847 spin_unlock(&dq_state_lock);
849 dquot = find_dquot(hashent, sb, qid);
852 spin_unlock(&dq_list_lock);
853 empty = get_empty_dquot(sb, qid.type);
855 schedule(); /* Try to wait for a moment... */
861 /* all dquots go on the inuse_list */
863 /* hash it first so it can be found */
864 insert_dquot_hash(dquot);
865 spin_unlock(&dq_list_lock);
866 dqstats_inc(DQST_LOOKUPS);
868 if (!atomic_read(&dquot->dq_count))
869 remove_free_dquot(dquot);
870 atomic_inc(&dquot->dq_count);
871 spin_unlock(&dq_list_lock);
872 dqstats_inc(DQST_CACHE_HITS);
873 dqstats_inc(DQST_LOOKUPS);
875 /* Wait for dq_lock - after this we know that either dquot_release() is
876 * already finished or it will be canceled due to dq_count > 1 test */
877 wait_on_dquot(dquot);
878 /* Read the dquot / allocate space in quota file */
879 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
880 sb->dq_op->acquire_dquot(dquot) < 0) {
885 #ifdef CONFIG_QUOTA_DEBUG
886 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
890 do_destroy_dquot(empty);
894 EXPORT_SYMBOL(dqget);
896 static inline struct dquot **i_dquot(struct inode *inode)
898 return inode->i_sb->s_op->get_dquots(inode);
901 static int dqinit_needed(struct inode *inode, int type)
903 struct dquot * const *dquots;
906 if (IS_NOQUOTA(inode))
909 dquots = i_dquot(inode);
911 return !dquots[type];
912 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
918 /* This routine is guarded by dqonoff_mutex mutex */
919 static void add_dquot_ref(struct super_block *sb, int type)
921 struct inode *inode, *old_inode = NULL;
922 #ifdef CONFIG_QUOTA_DEBUG
926 spin_lock(&inode_sb_list_lock);
927 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
928 spin_lock(&inode->i_lock);
929 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
930 !atomic_read(&inode->i_writecount) ||
931 !dqinit_needed(inode, type)) {
932 spin_unlock(&inode->i_lock);
936 spin_unlock(&inode->i_lock);
937 spin_unlock(&inode_sb_list_lock);
939 #ifdef CONFIG_QUOTA_DEBUG
940 if (unlikely(inode_get_rsv_space(inode) > 0))
944 __dquot_initialize(inode, type);
947 * We hold a reference to 'inode' so it couldn't have been
948 * removed from s_inodes list while we dropped the
949 * inode_sb_list_lock We cannot iput the inode now as we can be
950 * holding the last reference and we cannot iput it under
951 * inode_sb_list_lock. So we keep the reference and iput it
955 spin_lock(&inode_sb_list_lock);
957 spin_unlock(&inode_sb_list_lock);
960 #ifdef CONFIG_QUOTA_DEBUG
962 quota_error(sb, "Writes happened before quota was turned on "
963 "thus quota information is probably inconsistent. "
964 "Please run quotacheck(8)");
970 * Remove references to dquots from inode and add dquot to list for freeing
971 * if we have the last reference to dquot
973 static void remove_inode_dquot_ref(struct inode *inode, int type,
974 struct list_head *tofree_head)
976 struct dquot **dquots = i_dquot(inode);
977 struct dquot *dquot = dquots[type];
983 if (list_empty(&dquot->dq_free)) {
985 * The inode still has reference to dquot so it can't be in the
988 spin_lock(&dq_list_lock);
989 list_add(&dquot->dq_free, tofree_head);
990 spin_unlock(&dq_list_lock);
993 * Dquot is already in a list to put so we won't drop the last
1001 * Free list of dquots
1002 * Dquots are removed from inodes and no new references can be got so we are
1003 * the only ones holding reference
1005 static void put_dquot_list(struct list_head *tofree_head)
1007 struct list_head *act_head;
1008 struct dquot *dquot;
1010 act_head = tofree_head->next;
1011 while (act_head != tofree_head) {
1012 dquot = list_entry(act_head, struct dquot, dq_free);
1013 act_head = act_head->next;
1014 /* Remove dquot from the list so we won't have problems... */
1015 list_del_init(&dquot->dq_free);
1020 static void remove_dquot_ref(struct super_block *sb, int type,
1021 struct list_head *tofree_head)
1023 struct inode *inode;
1026 spin_lock(&inode_sb_list_lock);
1027 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1029 * We have to scan also I_NEW inodes because they can already
1030 * have quota pointer initialized. Luckily, we need to touch
1031 * only quota pointers and these have separate locking
1034 spin_lock(&dq_data_lock);
1035 if (!IS_NOQUOTA(inode)) {
1036 if (unlikely(inode_get_rsv_space(inode) > 0))
1038 remove_inode_dquot_ref(inode, type, tofree_head);
1040 spin_unlock(&dq_data_lock);
1042 spin_unlock(&inode_sb_list_lock);
1043 #ifdef CONFIG_QUOTA_DEBUG
1045 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1046 " was disabled thus quota information is probably "
1047 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1052 /* Gather all references from inodes and drop them */
1053 static void drop_dquot_ref(struct super_block *sb, int type)
1055 LIST_HEAD(tofree_head);
1058 remove_dquot_ref(sb, type, &tofree_head);
1059 synchronize_srcu(&dquot_srcu);
1060 put_dquot_list(&tofree_head);
1064 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1066 dquot->dq_dqb.dqb_curinodes += number;
1069 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1071 dquot->dq_dqb.dqb_curspace += number;
1074 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1076 dquot->dq_dqb.dqb_rsvspace += number;
1080 * Claim reserved quota space
1082 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1084 if (dquot->dq_dqb.dqb_rsvspace < number) {
1086 number = dquot->dq_dqb.dqb_rsvspace;
1088 dquot->dq_dqb.dqb_curspace += number;
1089 dquot->dq_dqb.dqb_rsvspace -= number;
1092 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1094 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1095 number = dquot->dq_dqb.dqb_curspace;
1096 dquot->dq_dqb.dqb_rsvspace += number;
1097 dquot->dq_dqb.dqb_curspace -= number;
1101 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1103 if (dquot->dq_dqb.dqb_rsvspace >= number)
1104 dquot->dq_dqb.dqb_rsvspace -= number;
1107 dquot->dq_dqb.dqb_rsvspace = 0;
1111 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1113 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1114 dquot->dq_dqb.dqb_curinodes >= number)
1115 dquot->dq_dqb.dqb_curinodes -= number;
1117 dquot->dq_dqb.dqb_curinodes = 0;
1118 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1119 dquot->dq_dqb.dqb_itime = (time_t) 0;
1120 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1123 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1125 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1126 dquot->dq_dqb.dqb_curspace >= number)
1127 dquot->dq_dqb.dqb_curspace -= number;
1129 dquot->dq_dqb.dqb_curspace = 0;
1130 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1131 dquot->dq_dqb.dqb_btime = (time_t) 0;
1132 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1136 struct super_block *w_sb;
1137 struct kqid w_dq_id;
1141 static int warning_issued(struct dquot *dquot, const int warntype)
1143 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1144 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1145 ((warntype == QUOTA_NL_IHARDWARN ||
1146 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1150 return test_and_set_bit(flag, &dquot->dq_flags);
1153 #ifdef CONFIG_PRINT_QUOTA_WARNING
1154 static int flag_print_warnings = 1;
1156 static int need_print_warning(struct dquot_warn *warn)
1158 if (!flag_print_warnings)
1161 switch (warn->w_dq_id.type) {
1163 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1165 return in_group_p(warn->w_dq_id.gid);
1172 /* Print warning to user which exceeded quota */
1173 static void print_warning(struct dquot_warn *warn)
1176 struct tty_struct *tty;
1177 int warntype = warn->w_type;
1179 if (warntype == QUOTA_NL_IHARDBELOW ||
1180 warntype == QUOTA_NL_ISOFTBELOW ||
1181 warntype == QUOTA_NL_BHARDBELOW ||
1182 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1185 tty = get_current_tty();
1188 tty_write_message(tty, warn->w_sb->s_id);
1189 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1190 tty_write_message(tty, ": warning, ");
1192 tty_write_message(tty, ": write failed, ");
1193 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1195 case QUOTA_NL_IHARDWARN:
1196 msg = " file limit reached.\r\n";
1198 case QUOTA_NL_ISOFTLONGWARN:
1199 msg = " file quota exceeded too long.\r\n";
1201 case QUOTA_NL_ISOFTWARN:
1202 msg = " file quota exceeded.\r\n";
1204 case QUOTA_NL_BHARDWARN:
1205 msg = " block limit reached.\r\n";
1207 case QUOTA_NL_BSOFTLONGWARN:
1208 msg = " block quota exceeded too long.\r\n";
1210 case QUOTA_NL_BSOFTWARN:
1211 msg = " block quota exceeded.\r\n";
1214 tty_write_message(tty, msg);
1219 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1222 if (warning_issued(dquot, warntype))
1224 warn->w_type = warntype;
1225 warn->w_sb = dquot->dq_sb;
1226 warn->w_dq_id = dquot->dq_id;
1230 * Write warnings to the console and send warning messages over netlink.
1232 * Note that this function can call into tty and networking code.
1234 static void flush_warnings(struct dquot_warn *warn)
1238 for (i = 0; i < MAXQUOTAS; i++) {
1239 if (warn[i].w_type == QUOTA_NL_NOWARN)
1241 #ifdef CONFIG_PRINT_QUOTA_WARNING
1242 print_warning(&warn[i]);
1244 quota_send_warning(warn[i].w_dq_id,
1245 warn[i].w_sb->s_dev, warn[i].w_type);
1249 static int ignore_hardlimit(struct dquot *dquot)
1251 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1253 return capable(CAP_SYS_RESOURCE) &&
1254 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1255 !(info->dqi_flags & DQF_ROOT_SQUASH));
1258 /* needs dq_data_lock */
1259 static int check_idq(struct dquot *dquot, qsize_t inodes,
1260 struct dquot_warn *warn)
1262 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1264 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1265 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1268 if (dquot->dq_dqb.dqb_ihardlimit &&
1269 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1270 !ignore_hardlimit(dquot)) {
1271 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1275 if (dquot->dq_dqb.dqb_isoftlimit &&
1276 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1277 dquot->dq_dqb.dqb_itime &&
1278 get_seconds() >= dquot->dq_dqb.dqb_itime &&
1279 !ignore_hardlimit(dquot)) {
1280 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1284 if (dquot->dq_dqb.dqb_isoftlimit &&
1285 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1286 dquot->dq_dqb.dqb_itime == 0) {
1287 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1288 dquot->dq_dqb.dqb_itime = get_seconds() +
1289 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1295 /* needs dq_data_lock */
1296 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1297 struct dquot_warn *warn)
1300 struct super_block *sb = dquot->dq_sb;
1302 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1303 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1306 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1309 if (dquot->dq_dqb.dqb_bhardlimit &&
1310 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1311 !ignore_hardlimit(dquot)) {
1313 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1317 if (dquot->dq_dqb.dqb_bsoftlimit &&
1318 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1319 dquot->dq_dqb.dqb_btime &&
1320 get_seconds() >= dquot->dq_dqb.dqb_btime &&
1321 !ignore_hardlimit(dquot)) {
1323 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1327 if (dquot->dq_dqb.dqb_bsoftlimit &&
1328 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1329 dquot->dq_dqb.dqb_btime == 0) {
1331 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1332 dquot->dq_dqb.dqb_btime = get_seconds() +
1333 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1337 * We don't allow preallocation to exceed softlimit so exceeding will
1346 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1350 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1351 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1352 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1353 return QUOTA_NL_NOWARN;
1355 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1356 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1357 return QUOTA_NL_ISOFTBELOW;
1358 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1359 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1360 return QUOTA_NL_IHARDBELOW;
1361 return QUOTA_NL_NOWARN;
1364 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1366 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1367 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1368 return QUOTA_NL_NOWARN;
1370 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1371 return QUOTA_NL_BSOFTBELOW;
1372 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1373 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1374 return QUOTA_NL_BHARDBELOW;
1375 return QUOTA_NL_NOWARN;
1378 static int dquot_active(const struct inode *inode)
1380 struct super_block *sb = inode->i_sb;
1382 if (IS_NOQUOTA(inode))
1384 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1388 * Initialize quota pointers in inode
1390 * It is better to call this function outside of any transaction as it
1391 * might need a lot of space in journal for dquot structure allocation.
1393 static void __dquot_initialize(struct inode *inode, int type)
1395 int cnt, init_needed = 0;
1396 struct dquot **dquots, *got[MAXQUOTAS];
1397 struct super_block *sb = inode->i_sb;
1400 if (!dquot_active(inode))
1403 dquots = i_dquot(inode);
1405 /* First get references to structures we might need. */
1406 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1412 if (type != -1 && cnt != type)
1415 * The i_dquot should have been initialized in most cases,
1416 * we check it without locking here to avoid unnecessary
1417 * dqget()/dqput() calls.
1422 if (!sb_has_quota_active(sb, cnt))
1429 qid = make_kqid_uid(inode->i_uid);
1432 qid = make_kqid_gid(inode->i_gid);
1435 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1438 qid = make_kqid_projid(projid);
1441 got[cnt] = dqget(sb, qid);
1444 /* All required i_dquot has been initialized */
1448 spin_lock(&dq_data_lock);
1449 if (IS_NOQUOTA(inode))
1451 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1452 if (type != -1 && cnt != type)
1454 /* Avoid races with quotaoff() */
1455 if (!sb_has_quota_active(sb, cnt))
1457 /* We could race with quotaon or dqget() could have failed */
1461 dquots[cnt] = got[cnt];
1464 * Make quota reservation system happy if someone
1465 * did a write before quota was turned on
1467 rsv = inode_get_rsv_space(inode);
1469 dquot_resv_space(dquots[cnt], rsv);
1473 spin_unlock(&dq_data_lock);
1474 /* Drop unused references */
1478 void dquot_initialize(struct inode *inode)
1480 __dquot_initialize(inode, -1);
1482 EXPORT_SYMBOL(dquot_initialize);
1485 * Release all quotas referenced by inode.
1487 * This function only be called on inode free or converting
1488 * a file to quota file, no other users for the i_dquot in
1489 * both cases, so we needn't call synchronize_srcu() after
1492 static void __dquot_drop(struct inode *inode)
1495 struct dquot **dquots = i_dquot(inode);
1496 struct dquot *put[MAXQUOTAS];
1498 spin_lock(&dq_data_lock);
1499 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1500 put[cnt] = dquots[cnt];
1503 spin_unlock(&dq_data_lock);
1507 void dquot_drop(struct inode *inode)
1509 struct dquot * const *dquots;
1512 if (IS_NOQUOTA(inode))
1516 * Test before calling to rule out calls from proc and such
1517 * where we are not allowed to block. Note that this is
1518 * actually reliable test even without the lock - the caller
1519 * must assure that nobody can come after the DQUOT_DROP and
1520 * add quota pointers back anyway.
1522 dquots = i_dquot(inode);
1523 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1528 if (cnt < MAXQUOTAS)
1529 __dquot_drop(inode);
1531 EXPORT_SYMBOL(dquot_drop);
1534 * inode_reserved_space is managed internally by quota, and protected by
1535 * i_lock similar to i_blocks+i_bytes.
1537 static qsize_t *inode_reserved_space(struct inode * inode)
1539 /* Filesystem must explicitly define it's own method in order to use
1540 * quota reservation interface */
1541 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1542 return inode->i_sb->dq_op->get_reserved_space(inode);
1545 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1547 spin_lock(&inode->i_lock);
1548 *inode_reserved_space(inode) += number;
1549 spin_unlock(&inode->i_lock);
1551 EXPORT_SYMBOL(inode_add_rsv_space);
1553 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1555 spin_lock(&inode->i_lock);
1556 *inode_reserved_space(inode) -= number;
1557 __inode_add_bytes(inode, number);
1558 spin_unlock(&inode->i_lock);
1560 EXPORT_SYMBOL(inode_claim_rsv_space);
1562 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1564 spin_lock(&inode->i_lock);
1565 *inode_reserved_space(inode) += number;
1566 __inode_sub_bytes(inode, number);
1567 spin_unlock(&inode->i_lock);
1569 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1571 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1573 spin_lock(&inode->i_lock);
1574 *inode_reserved_space(inode) -= number;
1575 spin_unlock(&inode->i_lock);
1577 EXPORT_SYMBOL(inode_sub_rsv_space);
1579 static qsize_t inode_get_rsv_space(struct inode *inode)
1583 if (!inode->i_sb->dq_op->get_reserved_space)
1585 spin_lock(&inode->i_lock);
1586 ret = *inode_reserved_space(inode);
1587 spin_unlock(&inode->i_lock);
1591 static void inode_incr_space(struct inode *inode, qsize_t number,
1595 inode_add_rsv_space(inode, number);
1597 inode_add_bytes(inode, number);
1600 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1603 inode_sub_rsv_space(inode, number);
1605 inode_sub_bytes(inode, number);
1609 * This functions updates i_blocks+i_bytes fields and quota information
1610 * (together with appropriate checks).
1612 * NOTE: We absolutely rely on the fact that caller dirties the inode
1613 * (usually helpers in quotaops.h care about this) and holds a handle for
1614 * the current transaction so that dquot write and inode write go into the
1619 * This operation can block, but only after everything is updated
1621 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1623 int cnt, ret = 0, index;
1624 struct dquot_warn warn[MAXQUOTAS];
1625 int reserve = flags & DQUOT_SPACE_RESERVE;
1626 struct dquot **dquots;
1628 if (!dquot_active(inode)) {
1629 inode_incr_space(inode, number, reserve);
1633 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1634 warn[cnt].w_type = QUOTA_NL_NOWARN;
1636 dquots = i_dquot(inode);
1637 index = srcu_read_lock(&dquot_srcu);
1638 spin_lock(&dq_data_lock);
1639 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1642 ret = check_bdq(dquots[cnt], number,
1643 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1644 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1645 spin_unlock(&dq_data_lock);
1646 goto out_flush_warn;
1649 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1653 dquot_resv_space(dquots[cnt], number);
1655 dquot_incr_space(dquots[cnt], number);
1657 inode_incr_space(inode, number, reserve);
1658 spin_unlock(&dq_data_lock);
1661 goto out_flush_warn;
1662 mark_all_dquot_dirty(dquots);
1664 srcu_read_unlock(&dquot_srcu, index);
1665 flush_warnings(warn);
1669 EXPORT_SYMBOL(__dquot_alloc_space);
1672 * This operation can block, but only after everything is updated
1674 int dquot_alloc_inode(struct inode *inode)
1676 int cnt, ret = 0, index;
1677 struct dquot_warn warn[MAXQUOTAS];
1678 struct dquot * const *dquots;
1680 if (!dquot_active(inode))
1682 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1683 warn[cnt].w_type = QUOTA_NL_NOWARN;
1685 dquots = i_dquot(inode);
1686 index = srcu_read_lock(&dquot_srcu);
1687 spin_lock(&dq_data_lock);
1688 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1691 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1696 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1699 dquot_incr_inodes(dquots[cnt], 1);
1703 spin_unlock(&dq_data_lock);
1705 mark_all_dquot_dirty(dquots);
1706 srcu_read_unlock(&dquot_srcu, index);
1707 flush_warnings(warn);
1710 EXPORT_SYMBOL(dquot_alloc_inode);
1713 * Convert in-memory reserved quotas to real consumed quotas
1715 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1717 struct dquot **dquots;
1720 if (!dquot_active(inode)) {
1721 inode_claim_rsv_space(inode, number);
1725 dquots = i_dquot(inode);
1726 index = srcu_read_lock(&dquot_srcu);
1727 spin_lock(&dq_data_lock);
1728 /* Claim reserved quotas to allocated quotas */
1729 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1731 dquot_claim_reserved_space(dquots[cnt], number);
1733 /* Update inode bytes */
1734 inode_claim_rsv_space(inode, number);
1735 spin_unlock(&dq_data_lock);
1736 mark_all_dquot_dirty(dquots);
1737 srcu_read_unlock(&dquot_srcu, index);
1740 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1743 * Convert allocated space back to in-memory reserved quotas
1745 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1747 struct dquot **dquots;
1750 if (!dquot_active(inode)) {
1751 inode_reclaim_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_reclaim_reserved_space(dquots[cnt], number);
1763 /* Update inode bytes */
1764 inode_reclaim_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_reclaim_space_nodirty);
1773 * This operation can block, but only after everything is updated
1775 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1778 struct dquot_warn warn[MAXQUOTAS];
1779 struct dquot **dquots;
1780 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1782 if (!dquot_active(inode)) {
1783 inode_decr_space(inode, number, reserve);
1787 dquots = i_dquot(inode);
1788 index = srcu_read_lock(&dquot_srcu);
1789 spin_lock(&dq_data_lock);
1790 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1793 warn[cnt].w_type = QUOTA_NL_NOWARN;
1796 wtype = info_bdq_free(dquots[cnt], number);
1797 if (wtype != QUOTA_NL_NOWARN)
1798 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1800 dquot_free_reserved_space(dquots[cnt], number);
1802 dquot_decr_space(dquots[cnt], number);
1804 inode_decr_space(inode, number, reserve);
1805 spin_unlock(&dq_data_lock);
1809 mark_all_dquot_dirty(dquots);
1811 srcu_read_unlock(&dquot_srcu, index);
1812 flush_warnings(warn);
1814 EXPORT_SYMBOL(__dquot_free_space);
1817 * This operation can block, but only after everything is updated
1819 void dquot_free_inode(struct inode *inode)
1822 struct dquot_warn warn[MAXQUOTAS];
1823 struct dquot * const *dquots;
1826 if (!dquot_active(inode))
1829 dquots = i_dquot(inode);
1830 index = srcu_read_lock(&dquot_srcu);
1831 spin_lock(&dq_data_lock);
1832 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1835 warn[cnt].w_type = QUOTA_NL_NOWARN;
1838 wtype = info_idq_free(dquots[cnt], 1);
1839 if (wtype != QUOTA_NL_NOWARN)
1840 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1841 dquot_decr_inodes(dquots[cnt], 1);
1843 spin_unlock(&dq_data_lock);
1844 mark_all_dquot_dirty(dquots);
1845 srcu_read_unlock(&dquot_srcu, index);
1846 flush_warnings(warn);
1848 EXPORT_SYMBOL(dquot_free_inode);
1851 * Transfer the number of inode and blocks from one diskquota to an other.
1852 * On success, dquot references in transfer_to are consumed and references
1853 * to original dquots that need to be released are placed there. On failure,
1854 * references are kept untouched.
1856 * This operation can block, but only after everything is updated
1857 * A transaction must be started when entering this function.
1859 * We are holding reference on transfer_from & transfer_to, no need to
1860 * protect them by srcu_read_lock().
1862 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1864 qsize_t space, cur_space;
1865 qsize_t rsv_space = 0;
1866 struct dquot *transfer_from[MAXQUOTAS] = {};
1868 char is_valid[MAXQUOTAS] = {};
1869 struct dquot_warn warn_to[MAXQUOTAS];
1870 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1871 struct dquot_warn warn_from_space[MAXQUOTAS];
1873 if (IS_NOQUOTA(inode))
1875 /* Initialize the arrays */
1876 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1877 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1878 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1879 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1882 spin_lock(&dq_data_lock);
1883 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1884 spin_unlock(&dq_data_lock);
1887 cur_space = inode_get_bytes(inode);
1888 rsv_space = inode_get_rsv_space(inode);
1889 space = cur_space + rsv_space;
1890 /* Build the transfer_from list and check the limits */
1891 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1893 * Skip changes for same uid or gid or for turned off quota-type.
1895 if (!transfer_to[cnt])
1897 /* Avoid races with quotaoff() */
1898 if (!sb_has_quota_active(inode->i_sb, cnt))
1901 transfer_from[cnt] = i_dquot(inode)[cnt];
1902 ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1905 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1911 * Finally perform the needed transfer from transfer_from to transfer_to
1913 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1916 /* Due to IO error we might not have transfer_from[] structure */
1917 if (transfer_from[cnt]) {
1919 wtype = info_idq_free(transfer_from[cnt], 1);
1920 if (wtype != QUOTA_NL_NOWARN)
1921 prepare_warning(&warn_from_inodes[cnt],
1922 transfer_from[cnt], wtype);
1923 wtype = info_bdq_free(transfer_from[cnt], space);
1924 if (wtype != QUOTA_NL_NOWARN)
1925 prepare_warning(&warn_from_space[cnt],
1926 transfer_from[cnt], wtype);
1927 dquot_decr_inodes(transfer_from[cnt], 1);
1928 dquot_decr_space(transfer_from[cnt], cur_space);
1929 dquot_free_reserved_space(transfer_from[cnt],
1933 dquot_incr_inodes(transfer_to[cnt], 1);
1934 dquot_incr_space(transfer_to[cnt], cur_space);
1935 dquot_resv_space(transfer_to[cnt], rsv_space);
1937 i_dquot(inode)[cnt] = transfer_to[cnt];
1939 spin_unlock(&dq_data_lock);
1941 mark_all_dquot_dirty(transfer_from);
1942 mark_all_dquot_dirty(transfer_to);
1943 flush_warnings(warn_to);
1944 flush_warnings(warn_from_inodes);
1945 flush_warnings(warn_from_space);
1946 /* Pass back references to put */
1947 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1949 transfer_to[cnt] = transfer_from[cnt];
1952 spin_unlock(&dq_data_lock);
1953 flush_warnings(warn_to);
1956 EXPORT_SYMBOL(__dquot_transfer);
1958 /* Wrapper for transferring ownership of an inode for uid/gid only
1959 * Called from FSXXX_setattr()
1961 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1963 struct dquot *transfer_to[MAXQUOTAS] = {};
1964 struct super_block *sb = inode->i_sb;
1967 if (!dquot_active(inode))
1970 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid))
1971 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(iattr->ia_uid));
1972 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))
1973 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(iattr->ia_gid));
1975 ret = __dquot_transfer(inode, transfer_to);
1976 dqput_all(transfer_to);
1979 EXPORT_SYMBOL(dquot_transfer);
1982 * Write info of quota file to disk
1984 int dquot_commit_info(struct super_block *sb, int type)
1987 struct quota_info *dqopt = sb_dqopt(sb);
1989 mutex_lock(&dqopt->dqio_mutex);
1990 ret = dqopt->ops[type]->write_file_info(sb, type);
1991 mutex_unlock(&dqopt->dqio_mutex);
1994 EXPORT_SYMBOL(dquot_commit_info);
1997 * Definitions of diskquota operations.
1999 const struct dquot_operations dquot_operations = {
2000 .write_dquot = dquot_commit,
2001 .acquire_dquot = dquot_acquire,
2002 .release_dquot = dquot_release,
2003 .mark_dirty = dquot_mark_dquot_dirty,
2004 .write_info = dquot_commit_info,
2005 .alloc_dquot = dquot_alloc,
2006 .destroy_dquot = dquot_destroy,
2008 EXPORT_SYMBOL(dquot_operations);
2011 * Generic helper for ->open on filesystems supporting disk quotas.
2013 int dquot_file_open(struct inode *inode, struct file *file)
2017 error = generic_file_open(inode, file);
2018 if (!error && (file->f_mode & FMODE_WRITE))
2019 dquot_initialize(inode);
2022 EXPORT_SYMBOL(dquot_file_open);
2025 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2027 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2030 struct quota_info *dqopt = sb_dqopt(sb);
2031 struct inode *toputinode[MAXQUOTAS];
2033 /* Cannot turn off usage accounting without turning off limits, or
2034 * suspend quotas and simultaneously turn quotas off. */
2035 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2036 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2037 DQUOT_USAGE_ENABLED)))
2040 /* We need to serialize quota_off() for device */
2041 mutex_lock(&dqopt->dqonoff_mutex);
2044 * Skip everything if there's nothing to do. We have to do this because
2045 * sometimes we are called when fill_super() failed and calling
2046 * sync_fs() in such cases does no good.
2048 if (!sb_any_quota_loaded(sb)) {
2049 mutex_unlock(&dqopt->dqonoff_mutex);
2052 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2053 toputinode[cnt] = NULL;
2054 if (type != -1 && cnt != type)
2056 if (!sb_has_quota_loaded(sb, cnt))
2059 if (flags & DQUOT_SUSPENDED) {
2060 spin_lock(&dq_state_lock);
2062 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2063 spin_unlock(&dq_state_lock);
2065 spin_lock(&dq_state_lock);
2066 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2067 /* Turning off suspended quotas? */
2068 if (!sb_has_quota_loaded(sb, cnt) &&
2069 sb_has_quota_suspended(sb, cnt)) {
2070 dqopt->flags &= ~dquot_state_flag(
2071 DQUOT_SUSPENDED, cnt);
2072 spin_unlock(&dq_state_lock);
2073 iput(dqopt->files[cnt]);
2074 dqopt->files[cnt] = NULL;
2077 spin_unlock(&dq_state_lock);
2080 /* We still have to keep quota loaded? */
2081 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2084 /* Note: these are blocking operations */
2085 drop_dquot_ref(sb, cnt);
2086 invalidate_dquots(sb, cnt);
2088 * Now all dquots should be invalidated, all writes done so we
2089 * should be only users of the info. No locks needed.
2091 if (info_dirty(&dqopt->info[cnt]))
2092 sb->dq_op->write_info(sb, cnt);
2093 if (dqopt->ops[cnt]->free_file_info)
2094 dqopt->ops[cnt]->free_file_info(sb, cnt);
2095 put_quota_format(dqopt->info[cnt].dqi_format);
2097 toputinode[cnt] = dqopt->files[cnt];
2098 if (!sb_has_quota_loaded(sb, cnt))
2099 dqopt->files[cnt] = NULL;
2100 dqopt->info[cnt].dqi_flags = 0;
2101 dqopt->info[cnt].dqi_igrace = 0;
2102 dqopt->info[cnt].dqi_bgrace = 0;
2103 dqopt->ops[cnt] = NULL;
2105 mutex_unlock(&dqopt->dqonoff_mutex);
2107 /* Skip syncing and setting flags if quota files are hidden */
2108 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2111 /* Sync the superblock so that buffers with quota data are written to
2112 * disk (and so userspace sees correct data afterwards). */
2113 if (sb->s_op->sync_fs)
2114 sb->s_op->sync_fs(sb, 1);
2115 sync_blockdev(sb->s_bdev);
2116 /* Now the quota files are just ordinary files and we can set the
2117 * inode flags back. Moreover we discard the pagecache so that
2118 * userspace sees the writes we did bypassing the pagecache. We
2119 * must also discard the blockdev buffers so that we see the
2120 * changes done by userspace on the next quotaon() */
2121 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2122 if (toputinode[cnt]) {
2123 mutex_lock(&dqopt->dqonoff_mutex);
2124 /* If quota was reenabled in the meantime, we have
2126 if (!sb_has_quota_loaded(sb, cnt)) {
2127 mutex_lock(&toputinode[cnt]->i_mutex);
2128 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2129 S_NOATIME | S_NOQUOTA);
2130 truncate_inode_pages(&toputinode[cnt]->i_data,
2132 mutex_unlock(&toputinode[cnt]->i_mutex);
2133 mark_inode_dirty_sync(toputinode[cnt]);
2135 mutex_unlock(&dqopt->dqonoff_mutex);
2138 invalidate_bdev(sb->s_bdev);
2140 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2141 if (toputinode[cnt]) {
2142 /* On remount RO, we keep the inode pointer so that we
2143 * can reenable quota on the subsequent remount RW. We
2144 * have to check 'flags' variable and not use sb_has_
2145 * function because another quotaon / quotaoff could
2146 * change global state before we got here. We refuse
2147 * to suspend quotas when there is pending delete on
2148 * the quota file... */
2149 if (!(flags & DQUOT_SUSPENDED))
2150 iput(toputinode[cnt]);
2151 else if (!toputinode[cnt]->i_nlink)
2156 EXPORT_SYMBOL(dquot_disable);
2158 int dquot_quota_off(struct super_block *sb, int type)
2160 return dquot_disable(sb, type,
2161 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2163 EXPORT_SYMBOL(dquot_quota_off);
2166 * Turn quotas on on a device
2170 * Helper function to turn quotas on when we already have the inode of
2171 * quota file and no quota information is loaded.
2173 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2176 struct quota_format_type *fmt = find_quota_format(format_id);
2177 struct super_block *sb = inode->i_sb;
2178 struct quota_info *dqopt = sb_dqopt(sb);
2184 if (!S_ISREG(inode->i_mode)) {
2188 if (IS_RDONLY(inode)) {
2192 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2193 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2197 /* Usage always has to be set... */
2198 if (!(flags & DQUOT_USAGE_ENABLED)) {
2203 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2204 /* As we bypass the pagecache we must now flush all the
2205 * dirty data and invalidate caches so that kernel sees
2206 * changes from userspace. It is not enough to just flush
2207 * the quota file since if blocksize < pagesize, invalidation
2208 * of the cache could fail because of other unrelated dirty
2210 sync_filesystem(sb);
2211 invalidate_bdev(sb->s_bdev);
2213 mutex_lock(&dqopt->dqonoff_mutex);
2214 if (sb_has_quota_loaded(sb, type)) {
2219 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2220 /* We don't want quota and atime on quota files (deadlocks
2221 * possible) Also nobody should write to the file - we use
2222 * special IO operations which ignore the immutable bit. */
2223 mutex_lock(&inode->i_mutex);
2224 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2226 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2227 mutex_unlock(&inode->i_mutex);
2229 * When S_NOQUOTA is set, remove dquot references as no more
2230 * references can be added
2232 __dquot_drop(inode);
2236 dqopt->files[type] = igrab(inode);
2237 if (!dqopt->files[type])
2240 if (!fmt->qf_ops->check_quota_file(sb, type))
2243 dqopt->ops[type] = fmt->qf_ops;
2244 dqopt->info[type].dqi_format = fmt;
2245 dqopt->info[type].dqi_fmt_id = format_id;
2246 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2247 mutex_lock(&dqopt->dqio_mutex);
2248 error = dqopt->ops[type]->read_file_info(sb, type);
2250 mutex_unlock(&dqopt->dqio_mutex);
2253 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2254 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2255 mutex_unlock(&dqopt->dqio_mutex);
2256 spin_lock(&dq_state_lock);
2257 dqopt->flags |= dquot_state_flag(flags, type);
2258 spin_unlock(&dq_state_lock);
2260 add_dquot_ref(sb, type);
2261 mutex_unlock(&dqopt->dqonoff_mutex);
2266 dqopt->files[type] = NULL;
2269 if (oldflags != -1) {
2270 mutex_lock(&inode->i_mutex);
2271 /* Set the flags back (in the case of accidental quotaon()
2272 * on a wrong file we don't want to mess up the flags) */
2273 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2274 inode->i_flags |= oldflags;
2275 mutex_unlock(&inode->i_mutex);
2277 mutex_unlock(&dqopt->dqonoff_mutex);
2279 put_quota_format(fmt);
2284 /* Reenable quotas on remount RW */
2285 int dquot_resume(struct super_block *sb, int type)
2287 struct quota_info *dqopt = sb_dqopt(sb);
2288 struct inode *inode;
2292 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2293 if (type != -1 && cnt != type)
2296 mutex_lock(&dqopt->dqonoff_mutex);
2297 if (!sb_has_quota_suspended(sb, cnt)) {
2298 mutex_unlock(&dqopt->dqonoff_mutex);
2301 inode = dqopt->files[cnt];
2302 dqopt->files[cnt] = NULL;
2303 spin_lock(&dq_state_lock);
2304 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2305 DQUOT_LIMITS_ENABLED,
2307 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2308 spin_unlock(&dq_state_lock);
2309 mutex_unlock(&dqopt->dqonoff_mutex);
2311 flags = dquot_generic_flag(flags, cnt);
2312 ret = vfs_load_quota_inode(inode, cnt,
2313 dqopt->info[cnt].dqi_fmt_id, flags);
2319 EXPORT_SYMBOL(dquot_resume);
2321 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2324 int error = security_quota_on(path->dentry);
2327 /* Quota file not on the same filesystem? */
2328 if (path->dentry->d_sb != sb)
2331 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2332 format_id, DQUOT_USAGE_ENABLED |
2333 DQUOT_LIMITS_ENABLED);
2336 EXPORT_SYMBOL(dquot_quota_on);
2339 * More powerful function for turning on quotas allowing setting
2340 * of individual quota flags
2342 int dquot_enable(struct inode *inode, int type, int format_id,
2346 struct super_block *sb = inode->i_sb;
2347 struct quota_info *dqopt = sb_dqopt(sb);
2349 /* Just unsuspend quotas? */
2350 BUG_ON(flags & DQUOT_SUSPENDED);
2354 /* Just updating flags needed? */
2355 if (sb_has_quota_loaded(sb, type)) {
2356 mutex_lock(&dqopt->dqonoff_mutex);
2357 /* Now do a reliable test... */
2358 if (!sb_has_quota_loaded(sb, type)) {
2359 mutex_unlock(&dqopt->dqonoff_mutex);
2362 if (flags & DQUOT_USAGE_ENABLED &&
2363 sb_has_quota_usage_enabled(sb, type)) {
2367 if (flags & DQUOT_LIMITS_ENABLED &&
2368 sb_has_quota_limits_enabled(sb, type)) {
2372 spin_lock(&dq_state_lock);
2373 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2374 spin_unlock(&dq_state_lock);
2376 mutex_unlock(&dqopt->dqonoff_mutex);
2381 return vfs_load_quota_inode(inode, type, format_id, flags);
2383 EXPORT_SYMBOL(dquot_enable);
2386 * This function is used when filesystem needs to initialize quotas
2387 * during mount time.
2389 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2390 int format_id, int type)
2392 struct dentry *dentry;
2395 mutex_lock(&d_inode(sb->s_root)->i_mutex);
2396 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2397 mutex_unlock(&d_inode(sb->s_root)->i_mutex);
2399 return PTR_ERR(dentry);
2401 if (d_really_is_negative(dentry)) {
2406 error = security_quota_on(dentry);
2408 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2409 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2415 EXPORT_SYMBOL(dquot_quota_on_mount);
2417 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2421 struct quota_info *dqopt = sb_dqopt(sb);
2423 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2425 /* Accounting cannot be turned on while fs is mounted */
2426 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2429 for (type = 0; type < MAXQUOTAS; type++) {
2430 if (!(flags & qtype_enforce_flag(type)))
2432 /* Can't enforce without accounting */
2433 if (!sb_has_quota_usage_enabled(sb, type))
2435 ret = dquot_enable(dqopt->files[type], type,
2436 dqopt->info[type].dqi_fmt_id,
2437 DQUOT_LIMITS_ENABLED);
2443 /* Backout enforcement enablement we already did */
2444 for (type--; type >= 0; type--) {
2445 if (flags & qtype_enforce_flag(type))
2446 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2448 /* Error code translation for better compatibility with XFS */
2454 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2458 struct quota_info *dqopt = sb_dqopt(sb);
2460 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2463 * We don't support turning off accounting via quotactl. In principle
2464 * quota infrastructure can do this but filesystems don't expect
2465 * userspace to be able to do it.
2468 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2471 /* Filter out limits not enabled */
2472 for (type = 0; type < MAXQUOTAS; type++)
2473 if (!sb_has_quota_limits_enabled(sb, type))
2474 flags &= ~qtype_enforce_flag(type);
2478 for (type = 0; type < MAXQUOTAS; type++) {
2479 if (flags & qtype_enforce_flag(type)) {
2480 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2487 /* Backout enforcement disabling we already did */
2488 for (type--; type >= 0; type--) {
2489 if (flags & qtype_enforce_flag(type))
2490 dquot_enable(dqopt->files[type], type,
2491 dqopt->info[type].dqi_fmt_id,
2492 DQUOT_LIMITS_ENABLED);
2497 /* Generic routine for getting common part of quota structure */
2498 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2500 struct mem_dqblk *dm = &dquot->dq_dqb;
2502 memset(di, 0, sizeof(*di));
2503 spin_lock(&dq_data_lock);
2504 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2505 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2506 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2507 di->d_ino_softlimit = dm->dqb_isoftlimit;
2508 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2509 di->d_ino_count = dm->dqb_curinodes;
2510 di->d_spc_timer = dm->dqb_btime;
2511 di->d_ino_timer = dm->dqb_itime;
2512 spin_unlock(&dq_data_lock);
2515 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2516 struct qc_dqblk *di)
2518 struct dquot *dquot;
2520 dquot = dqget(sb, qid);
2523 do_get_dqblk(dquot, di);
2528 EXPORT_SYMBOL(dquot_get_dqblk);
2530 #define VFS_QC_MASK \
2531 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2532 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2533 QC_SPC_TIMER | QC_INO_TIMER)
2535 /* Generic routine for setting common part of quota structure */
2536 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2538 struct mem_dqblk *dm = &dquot->dq_dqb;
2539 int check_blim = 0, check_ilim = 0;
2540 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2542 if (di->d_fieldmask & ~VFS_QC_MASK)
2545 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2546 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2547 ((di->d_fieldmask & QC_SPC_HARD) &&
2548 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2549 ((di->d_fieldmask & QC_INO_SOFT) &&
2550 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2551 ((di->d_fieldmask & QC_INO_HARD) &&
2552 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2555 spin_lock(&dq_data_lock);
2556 if (di->d_fieldmask & QC_SPACE) {
2557 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2559 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2562 if (di->d_fieldmask & QC_SPC_SOFT)
2563 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2564 if (di->d_fieldmask & QC_SPC_HARD)
2565 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2566 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2568 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2571 if (di->d_fieldmask & QC_INO_COUNT) {
2572 dm->dqb_curinodes = di->d_ino_count;
2574 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2577 if (di->d_fieldmask & QC_INO_SOFT)
2578 dm->dqb_isoftlimit = di->d_ino_softlimit;
2579 if (di->d_fieldmask & QC_INO_HARD)
2580 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2581 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2583 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2586 if (di->d_fieldmask & QC_SPC_TIMER) {
2587 dm->dqb_btime = di->d_spc_timer;
2589 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2592 if (di->d_fieldmask & QC_INO_TIMER) {
2593 dm->dqb_itime = di->d_ino_timer;
2595 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2599 if (!dm->dqb_bsoftlimit ||
2600 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2602 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2603 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2604 /* Set grace only if user hasn't provided his own... */
2605 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2608 if (!dm->dqb_isoftlimit ||
2609 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2611 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2612 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2613 /* Set grace only if user hasn't provided his own... */
2614 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2616 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2618 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2620 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2621 spin_unlock(&dq_data_lock);
2622 mark_dquot_dirty(dquot);
2627 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2628 struct qc_dqblk *di)
2630 struct dquot *dquot;
2633 dquot = dqget(sb, qid);
2638 rc = do_set_dqblk(dquot, di);
2643 EXPORT_SYMBOL(dquot_set_dqblk);
2645 /* Generic routine for getting common part of quota file information */
2646 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2648 struct mem_dqinfo *mi;
2649 struct qc_type_state *tstate;
2650 struct quota_info *dqopt = sb_dqopt(sb);
2653 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2654 memset(state, 0, sizeof(*state));
2655 for (type = 0; type < MAXQUOTAS; type++) {
2656 if (!sb_has_quota_active(sb, type))
2658 tstate = state->s_state + type;
2659 mi = sb_dqopt(sb)->info + type;
2660 tstate->flags = QCI_ACCT_ENABLED;
2661 spin_lock(&dq_data_lock);
2662 if (mi->dqi_flags & DQF_SYS_FILE)
2663 tstate->flags |= QCI_SYSFILE;
2664 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2665 tstate->flags |= QCI_ROOT_SQUASH;
2666 if (sb_has_quota_limits_enabled(sb, type))
2667 tstate->flags |= QCI_LIMITS_ENFORCED;
2668 tstate->spc_timelimit = mi->dqi_bgrace;
2669 tstate->ino_timelimit = mi->dqi_igrace;
2670 tstate->ino = dqopt->files[type]->i_ino;
2671 tstate->blocks = dqopt->files[type]->i_blocks;
2672 tstate->nextents = 1; /* We don't know... */
2673 spin_unlock(&dq_data_lock);
2675 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2678 EXPORT_SYMBOL(dquot_get_state);
2680 /* Generic routine for setting common part of quota file information */
2681 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2683 struct mem_dqinfo *mi;
2686 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2687 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2689 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2690 if (!sb_has_quota_active(sb, type)) {
2694 mi = sb_dqopt(sb)->info + type;
2695 if (ii->i_fieldmask & QC_FLAGS) {
2696 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2697 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD)) {
2702 spin_lock(&dq_data_lock);
2703 if (ii->i_fieldmask & QC_SPC_TIMER)
2704 mi->dqi_bgrace = ii->i_spc_timelimit;
2705 if (ii->i_fieldmask & QC_INO_TIMER)
2706 mi->dqi_igrace = ii->i_ino_timelimit;
2707 if (ii->i_fieldmask & QC_FLAGS) {
2708 if (ii->i_flags & QCI_ROOT_SQUASH)
2709 mi->dqi_flags |= DQF_ROOT_SQUASH;
2711 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2713 spin_unlock(&dq_data_lock);
2714 mark_info_dirty(sb, type);
2715 /* Force write to disk */
2716 sb->dq_op->write_info(sb, type);
2718 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2721 EXPORT_SYMBOL(dquot_set_dqinfo);
2723 const struct quotactl_ops dquot_quotactl_ops = {
2724 .quota_on = dquot_quota_on,
2725 .quota_off = dquot_quota_off,
2726 .quota_sync = dquot_quota_sync,
2727 .get_state = dquot_get_state,
2728 .set_info = dquot_set_dqinfo,
2729 .get_dqblk = dquot_get_dqblk,
2730 .set_dqblk = dquot_set_dqblk
2732 EXPORT_SYMBOL(dquot_quotactl_ops);
2734 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2735 .quota_enable = dquot_quota_enable,
2736 .quota_disable = dquot_quota_disable,
2737 .quota_sync = dquot_quota_sync,
2738 .get_state = dquot_get_state,
2739 .set_info = dquot_set_dqinfo,
2740 .get_dqblk = dquot_get_dqblk,
2741 .set_dqblk = dquot_set_dqblk
2743 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2745 static int do_proc_dqstats(struct ctl_table *table, int write,
2746 void __user *buffer, size_t *lenp, loff_t *ppos)
2748 unsigned int type = (int *)table->data - dqstats.stat;
2750 /* Update global table */
2751 dqstats.stat[type] =
2752 percpu_counter_sum_positive(&dqstats.counter[type]);
2753 return proc_dointvec(table, write, buffer, lenp, ppos);
2756 static struct ctl_table fs_dqstats_table[] = {
2758 .procname = "lookups",
2759 .data = &dqstats.stat[DQST_LOOKUPS],
2760 .maxlen = sizeof(int),
2762 .proc_handler = do_proc_dqstats,
2765 .procname = "drops",
2766 .data = &dqstats.stat[DQST_DROPS],
2767 .maxlen = sizeof(int),
2769 .proc_handler = do_proc_dqstats,
2772 .procname = "reads",
2773 .data = &dqstats.stat[DQST_READS],
2774 .maxlen = sizeof(int),
2776 .proc_handler = do_proc_dqstats,
2779 .procname = "writes",
2780 .data = &dqstats.stat[DQST_WRITES],
2781 .maxlen = sizeof(int),
2783 .proc_handler = do_proc_dqstats,
2786 .procname = "cache_hits",
2787 .data = &dqstats.stat[DQST_CACHE_HITS],
2788 .maxlen = sizeof(int),
2790 .proc_handler = do_proc_dqstats,
2793 .procname = "allocated_dquots",
2794 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2795 .maxlen = sizeof(int),
2797 .proc_handler = do_proc_dqstats,
2800 .procname = "free_dquots",
2801 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2802 .maxlen = sizeof(int),
2804 .proc_handler = do_proc_dqstats,
2807 .procname = "syncs",
2808 .data = &dqstats.stat[DQST_SYNCS],
2809 .maxlen = sizeof(int),
2811 .proc_handler = do_proc_dqstats,
2813 #ifdef CONFIG_PRINT_QUOTA_WARNING
2815 .procname = "warnings",
2816 .data = &flag_print_warnings,
2817 .maxlen = sizeof(int),
2819 .proc_handler = proc_dointvec,
2825 static struct ctl_table fs_table[] = {
2827 .procname = "quota",
2829 .child = fs_dqstats_table,
2834 static struct ctl_table sys_table[] = {
2843 static int __init dquot_init(void)
2846 unsigned long nr_hash, order;
2848 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2850 register_sysctl_table(sys_table);
2852 dquot_cachep = kmem_cache_create("dquot",
2853 sizeof(struct dquot), sizeof(unsigned long) * 4,
2854 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2855 SLAB_MEM_SPREAD|SLAB_PANIC),
2859 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2861 panic("Cannot create dquot hash table");
2863 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2864 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2866 panic("Cannot create dquot stat counters");
2869 /* Find power-of-two hlist_heads which can fit into allocation */
2870 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2874 } while (nr_hash >> dq_hash_bits);
2877 nr_hash = 1UL << dq_hash_bits;
2878 dq_hash_mask = nr_hash - 1;
2879 for (i = 0; i < nr_hash; i++)
2880 INIT_HLIST_HEAD(dquot_hash + i);
2882 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2883 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2885 register_shrinker(&dqcache_shrinker);
2889 module_init(dquot_init);