2 * Implementation of the diskquota system for the LINUX operating system. QUOTA
3 * is implemented using the BSD system call interface as the means of
4 * communication with the user level. This file contains the generic routines
5 * called by the different filesystems on allocation of an inode or block.
6 * These routines take care of the administration needed to have a consistent
7 * diskquota tracking system. The ideas of both user and group quotas are based
8 * on the Melbourne quota system as used on BSD derived systems. The internal
9 * implementation is based on one of the several variants of the LINUX
10 * inode-subsystem with added complexity of the diskquota system.
12 * Author: Marco van Wieringen <mvw@planets.elm.net>
14 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 * Revised list management to avoid races
17 * -- Bill Hawes, <whawes@star.net>, 9/98
19 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20 * As the consequence the locking was moved from dquot_decr_...(),
21 * dquot_incr_...() to calling functions.
22 * invalidate_dquots() now writes modified dquots.
23 * Serialized quota_off() and quota_on() for mount point.
24 * Fixed a few bugs in grow_dquots().
25 * Fixed deadlock in write_dquot() - we no longer account quotas on
27 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
28 * add_dquot_ref() restarts after blocking
29 * Added check for bogus uid and fixed check for group in quotactl.
30 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32 * Used struct list_head instead of own list struct
33 * Invalidation of referenced dquots is no longer possible
34 * Improved free_dquots list management
35 * Quota and i_blocks are now updated in one place to avoid races
36 * Warnings are now delayed so we won't block in critical section
37 * Write updated not to require dquot lock
38 * Jan Kara, <jack@suse.cz>, 9/2000
40 * Added dynamic quota structure allocation
41 * Jan Kara <jack@suse.cz> 12/2000
43 * Rewritten quota interface. Implemented new quota format and
44 * formats registering.
45 * Jan Kara, <jack@suse.cz>, 2001,2002
48 * Jan Kara, <jack@suse.cz>, 10/2002
50 * Added journalled quota support, fix lock inversion problems
51 * Jan Kara, <jack@suse.cz>, 2003,2004
53 * (C) Copyright 1994 - 1997 Marco van Wieringen
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
59 #include <linux/mount.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/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 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, *empty = NULL;
837 if (!sb_has_quota_active(sb, qid.type))
838 return ERR_PTR(-ESRCH);
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);
845 dquot = ERR_PTR(-ESRCH);
848 spin_unlock(&dq_state_lock);
850 dquot = find_dquot(hashent, sb, qid);
853 spin_unlock(&dq_list_lock);
854 empty = get_empty_dquot(sb, qid.type);
856 schedule(); /* Try to wait for a moment... */
862 /* all dquots go on the inuse_list */
864 /* hash it first so it can be found */
865 insert_dquot_hash(dquot);
866 spin_unlock(&dq_list_lock);
867 dqstats_inc(DQST_LOOKUPS);
869 if (!atomic_read(&dquot->dq_count))
870 remove_free_dquot(dquot);
871 atomic_inc(&dquot->dq_count);
872 spin_unlock(&dq_list_lock);
873 dqstats_inc(DQST_CACHE_HITS);
874 dqstats_inc(DQST_LOOKUPS);
876 /* Wait for dq_lock - after this we know that either dquot_release() is
877 * already finished or it will be canceled due to dq_count > 1 test */
878 wait_on_dquot(dquot);
879 /* Read the dquot / allocate space in quota file */
880 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
883 err = sb->dq_op->acquire_dquot(dquot);
886 dquot = ERR_PTR(err);
890 #ifdef CONFIG_QUOTA_DEBUG
891 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
895 do_destroy_dquot(empty);
899 EXPORT_SYMBOL(dqget);
901 static inline struct dquot **i_dquot(struct inode *inode)
903 return inode->i_sb->s_op->get_dquots(inode);
906 static int dqinit_needed(struct inode *inode, int type)
908 struct dquot * const *dquots;
911 if (IS_NOQUOTA(inode))
914 dquots = i_dquot(inode);
916 return !dquots[type];
917 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
923 /* This routine is guarded by dqonoff_mutex mutex */
924 static void add_dquot_ref(struct super_block *sb, int type)
926 struct inode *inode, *old_inode = NULL;
927 #ifdef CONFIG_QUOTA_DEBUG
931 spin_lock(&sb->s_inode_list_lock);
932 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
933 spin_lock(&inode->i_lock);
934 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
935 !atomic_read(&inode->i_writecount) ||
936 !dqinit_needed(inode, type)) {
937 spin_unlock(&inode->i_lock);
941 spin_unlock(&inode->i_lock);
942 spin_unlock(&sb->s_inode_list_lock);
944 #ifdef CONFIG_QUOTA_DEBUG
945 if (unlikely(inode_get_rsv_space(inode) > 0))
949 __dquot_initialize(inode, type);
952 * We hold a reference to 'inode' so it couldn't have been
953 * removed from s_inodes list while we dropped the
954 * s_inode_list_lock. We cannot iput the inode now as we can be
955 * holding the last reference and we cannot iput it under
956 * s_inode_list_lock. So we keep the reference and iput it
960 spin_lock(&sb->s_inode_list_lock);
962 spin_unlock(&sb->s_inode_list_lock);
965 #ifdef CONFIG_QUOTA_DEBUG
967 quota_error(sb, "Writes happened before quota was turned on "
968 "thus quota information is probably inconsistent. "
969 "Please run quotacheck(8)");
975 * Remove references to dquots from inode and add dquot to list for freeing
976 * if we have the last reference to dquot
978 static void remove_inode_dquot_ref(struct inode *inode, int type,
979 struct list_head *tofree_head)
981 struct dquot **dquots = i_dquot(inode);
982 struct dquot *dquot = dquots[type];
988 if (list_empty(&dquot->dq_free)) {
990 * The inode still has reference to dquot so it can't be in the
993 spin_lock(&dq_list_lock);
994 list_add(&dquot->dq_free, tofree_head);
995 spin_unlock(&dq_list_lock);
998 * Dquot is already in a list to put so we won't drop the last
1006 * Free list of dquots
1007 * Dquots are removed from inodes and no new references can be got so we are
1008 * the only ones holding reference
1010 static void put_dquot_list(struct list_head *tofree_head)
1012 struct list_head *act_head;
1013 struct dquot *dquot;
1015 act_head = tofree_head->next;
1016 while (act_head != tofree_head) {
1017 dquot = list_entry(act_head, struct dquot, dq_free);
1018 act_head = act_head->next;
1019 /* Remove dquot from the list so we won't have problems... */
1020 list_del_init(&dquot->dq_free);
1025 static void remove_dquot_ref(struct super_block *sb, int type,
1026 struct list_head *tofree_head)
1028 struct inode *inode;
1031 spin_lock(&sb->s_inode_list_lock);
1032 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1034 * We have to scan also I_NEW inodes because they can already
1035 * have quota pointer initialized. Luckily, we need to touch
1036 * only quota pointers and these have separate locking
1039 spin_lock(&dq_data_lock);
1040 if (!IS_NOQUOTA(inode)) {
1041 if (unlikely(inode_get_rsv_space(inode) > 0))
1043 remove_inode_dquot_ref(inode, type, tofree_head);
1045 spin_unlock(&dq_data_lock);
1047 spin_unlock(&sb->s_inode_list_lock);
1048 #ifdef CONFIG_QUOTA_DEBUG
1050 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1051 " was disabled thus quota information is probably "
1052 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1057 /* Gather all references from inodes and drop them */
1058 static void drop_dquot_ref(struct super_block *sb, int type)
1060 LIST_HEAD(tofree_head);
1063 remove_dquot_ref(sb, type, &tofree_head);
1064 synchronize_srcu(&dquot_srcu);
1065 put_dquot_list(&tofree_head);
1069 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1071 dquot->dq_dqb.dqb_curinodes += number;
1074 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1076 dquot->dq_dqb.dqb_curspace += number;
1079 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1081 dquot->dq_dqb.dqb_rsvspace += number;
1085 * Claim reserved quota space
1087 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1089 if (dquot->dq_dqb.dqb_rsvspace < number) {
1091 number = dquot->dq_dqb.dqb_rsvspace;
1093 dquot->dq_dqb.dqb_curspace += number;
1094 dquot->dq_dqb.dqb_rsvspace -= number;
1097 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1099 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1100 number = dquot->dq_dqb.dqb_curspace;
1101 dquot->dq_dqb.dqb_rsvspace += number;
1102 dquot->dq_dqb.dqb_curspace -= number;
1106 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1108 if (dquot->dq_dqb.dqb_rsvspace >= number)
1109 dquot->dq_dqb.dqb_rsvspace -= number;
1112 dquot->dq_dqb.dqb_rsvspace = 0;
1116 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1118 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1119 dquot->dq_dqb.dqb_curinodes >= number)
1120 dquot->dq_dqb.dqb_curinodes -= number;
1122 dquot->dq_dqb.dqb_curinodes = 0;
1123 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1124 dquot->dq_dqb.dqb_itime = (time_t) 0;
1125 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1128 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1130 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1131 dquot->dq_dqb.dqb_curspace >= number)
1132 dquot->dq_dqb.dqb_curspace -= number;
1134 dquot->dq_dqb.dqb_curspace = 0;
1135 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1136 dquot->dq_dqb.dqb_btime = (time_t) 0;
1137 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1141 struct super_block *w_sb;
1142 struct kqid w_dq_id;
1146 static int warning_issued(struct dquot *dquot, const int warntype)
1148 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1149 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1150 ((warntype == QUOTA_NL_IHARDWARN ||
1151 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1155 return test_and_set_bit(flag, &dquot->dq_flags);
1158 #ifdef CONFIG_PRINT_QUOTA_WARNING
1159 static int flag_print_warnings = 1;
1161 static int need_print_warning(struct dquot_warn *warn)
1163 if (!flag_print_warnings)
1166 switch (warn->w_dq_id.type) {
1168 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1170 return in_group_p(warn->w_dq_id.gid);
1177 /* Print warning to user which exceeded quota */
1178 static void print_warning(struct dquot_warn *warn)
1181 struct tty_struct *tty;
1182 int warntype = warn->w_type;
1184 if (warntype == QUOTA_NL_IHARDBELOW ||
1185 warntype == QUOTA_NL_ISOFTBELOW ||
1186 warntype == QUOTA_NL_BHARDBELOW ||
1187 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1190 tty = get_current_tty();
1193 tty_write_message(tty, warn->w_sb->s_id);
1194 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1195 tty_write_message(tty, ": warning, ");
1197 tty_write_message(tty, ": write failed, ");
1198 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1200 case QUOTA_NL_IHARDWARN:
1201 msg = " file limit reached.\r\n";
1203 case QUOTA_NL_ISOFTLONGWARN:
1204 msg = " file quota exceeded too long.\r\n";
1206 case QUOTA_NL_ISOFTWARN:
1207 msg = " file quota exceeded.\r\n";
1209 case QUOTA_NL_BHARDWARN:
1210 msg = " block limit reached.\r\n";
1212 case QUOTA_NL_BSOFTLONGWARN:
1213 msg = " block quota exceeded too long.\r\n";
1215 case QUOTA_NL_BSOFTWARN:
1216 msg = " block quota exceeded.\r\n";
1219 tty_write_message(tty, msg);
1224 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1227 if (warning_issued(dquot, warntype))
1229 warn->w_type = warntype;
1230 warn->w_sb = dquot->dq_sb;
1231 warn->w_dq_id = dquot->dq_id;
1235 * Write warnings to the console and send warning messages over netlink.
1237 * Note that this function can call into tty and networking code.
1239 static void flush_warnings(struct dquot_warn *warn)
1243 for (i = 0; i < MAXQUOTAS; i++) {
1244 if (warn[i].w_type == QUOTA_NL_NOWARN)
1246 #ifdef CONFIG_PRINT_QUOTA_WARNING
1247 print_warning(&warn[i]);
1249 quota_send_warning(warn[i].w_dq_id,
1250 warn[i].w_sb->s_dev, warn[i].w_type);
1254 static int ignore_hardlimit(struct dquot *dquot)
1256 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1258 return capable(CAP_SYS_RESOURCE) &&
1259 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1260 !(info->dqi_flags & DQF_ROOT_SQUASH));
1263 /* needs dq_data_lock */
1264 static int check_idq(struct dquot *dquot, qsize_t inodes,
1265 struct dquot_warn *warn)
1267 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1269 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1270 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1273 if (dquot->dq_dqb.dqb_ihardlimit &&
1274 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1275 !ignore_hardlimit(dquot)) {
1276 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1280 if (dquot->dq_dqb.dqb_isoftlimit &&
1281 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1282 dquot->dq_dqb.dqb_itime &&
1283 get_seconds() >= dquot->dq_dqb.dqb_itime &&
1284 !ignore_hardlimit(dquot)) {
1285 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1289 if (dquot->dq_dqb.dqb_isoftlimit &&
1290 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1291 dquot->dq_dqb.dqb_itime == 0) {
1292 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1293 dquot->dq_dqb.dqb_itime = get_seconds() +
1294 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1300 /* needs dq_data_lock */
1301 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1302 struct dquot_warn *warn)
1305 struct super_block *sb = dquot->dq_sb;
1307 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1308 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1311 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1314 if (dquot->dq_dqb.dqb_bhardlimit &&
1315 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1316 !ignore_hardlimit(dquot)) {
1318 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1322 if (dquot->dq_dqb.dqb_bsoftlimit &&
1323 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1324 dquot->dq_dqb.dqb_btime &&
1325 get_seconds() >= dquot->dq_dqb.dqb_btime &&
1326 !ignore_hardlimit(dquot)) {
1328 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1332 if (dquot->dq_dqb.dqb_bsoftlimit &&
1333 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1334 dquot->dq_dqb.dqb_btime == 0) {
1336 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1337 dquot->dq_dqb.dqb_btime = get_seconds() +
1338 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1342 * We don't allow preallocation to exceed softlimit so exceeding will
1351 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1355 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1356 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1357 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1358 return QUOTA_NL_NOWARN;
1360 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1361 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1362 return QUOTA_NL_ISOFTBELOW;
1363 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1364 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1365 return QUOTA_NL_IHARDBELOW;
1366 return QUOTA_NL_NOWARN;
1369 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1371 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1372 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1373 return QUOTA_NL_NOWARN;
1375 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1376 return QUOTA_NL_BSOFTBELOW;
1377 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1378 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1379 return QUOTA_NL_BHARDBELOW;
1380 return QUOTA_NL_NOWARN;
1383 static int dquot_active(const struct inode *inode)
1385 struct super_block *sb = inode->i_sb;
1387 if (IS_NOQUOTA(inode))
1389 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1393 * Initialize quota pointers in inode
1395 * It is better to call this function outside of any transaction as it
1396 * might need a lot of space in journal for dquot structure allocation.
1398 static int __dquot_initialize(struct inode *inode, int type)
1400 int cnt, init_needed = 0;
1401 struct dquot **dquots, *got[MAXQUOTAS] = {};
1402 struct super_block *sb = inode->i_sb;
1406 if (!dquot_active(inode))
1409 dquots = i_dquot(inode);
1411 /* First get references to structures we might need. */
1412 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1416 struct dquot *dquot;
1418 if (type != -1 && cnt != type)
1421 * The i_dquot should have been initialized in most cases,
1422 * we check it without locking here to avoid unnecessary
1423 * dqget()/dqput() calls.
1428 if (!sb_has_quota_active(sb, cnt))
1435 qid = make_kqid_uid(inode->i_uid);
1438 qid = make_kqid_gid(inode->i_gid);
1441 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1444 qid = make_kqid_projid(projid);
1447 dquot = dqget(sb, qid);
1448 if (IS_ERR(dquot)) {
1449 /* We raced with somebody turning quotas off... */
1450 if (PTR_ERR(dquot) != -ESRCH) {
1451 ret = PTR_ERR(dquot);
1459 /* All required i_dquot has been initialized */
1463 spin_lock(&dq_data_lock);
1464 if (IS_NOQUOTA(inode))
1466 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1467 if (type != -1 && cnt != type)
1469 /* Avoid races with quotaoff() */
1470 if (!sb_has_quota_active(sb, cnt))
1472 /* We could race with quotaon or dqget() could have failed */
1476 dquots[cnt] = got[cnt];
1479 * Make quota reservation system happy if someone
1480 * did a write before quota was turned on
1482 rsv = inode_get_rsv_space(inode);
1484 dquot_resv_space(dquots[cnt], rsv);
1488 spin_unlock(&dq_data_lock);
1490 /* Drop unused references */
1496 int dquot_initialize(struct inode *inode)
1498 return __dquot_initialize(inode, -1);
1500 EXPORT_SYMBOL(dquot_initialize);
1503 * Release all quotas referenced by inode.
1505 * This function only be called on inode free or converting
1506 * a file to quota file, no other users for the i_dquot in
1507 * both cases, so we needn't call synchronize_srcu() after
1510 static void __dquot_drop(struct inode *inode)
1513 struct dquot **dquots = i_dquot(inode);
1514 struct dquot *put[MAXQUOTAS];
1516 spin_lock(&dq_data_lock);
1517 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1518 put[cnt] = dquots[cnt];
1521 spin_unlock(&dq_data_lock);
1525 void dquot_drop(struct inode *inode)
1527 struct dquot * const *dquots;
1530 if (IS_NOQUOTA(inode))
1534 * Test before calling to rule out calls from proc and such
1535 * where we are not allowed to block. Note that this is
1536 * actually reliable test even without the lock - the caller
1537 * must assure that nobody can come after the DQUOT_DROP and
1538 * add quota pointers back anyway.
1540 dquots = i_dquot(inode);
1541 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1546 if (cnt < MAXQUOTAS)
1547 __dquot_drop(inode);
1549 EXPORT_SYMBOL(dquot_drop);
1552 * inode_reserved_space is managed internally by quota, and protected by
1553 * i_lock similar to i_blocks+i_bytes.
1555 static qsize_t *inode_reserved_space(struct inode * inode)
1557 /* Filesystem must explicitly define it's own method in order to use
1558 * quota reservation interface */
1559 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1560 return inode->i_sb->dq_op->get_reserved_space(inode);
1563 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1565 spin_lock(&inode->i_lock);
1566 *inode_reserved_space(inode) += number;
1567 spin_unlock(&inode->i_lock);
1569 EXPORT_SYMBOL(inode_add_rsv_space);
1571 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1573 spin_lock(&inode->i_lock);
1574 *inode_reserved_space(inode) -= number;
1575 __inode_add_bytes(inode, number);
1576 spin_unlock(&inode->i_lock);
1578 EXPORT_SYMBOL(inode_claim_rsv_space);
1580 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1582 spin_lock(&inode->i_lock);
1583 *inode_reserved_space(inode) += number;
1584 __inode_sub_bytes(inode, number);
1585 spin_unlock(&inode->i_lock);
1587 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1589 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1591 spin_lock(&inode->i_lock);
1592 *inode_reserved_space(inode) -= number;
1593 spin_unlock(&inode->i_lock);
1595 EXPORT_SYMBOL(inode_sub_rsv_space);
1597 static qsize_t inode_get_rsv_space(struct inode *inode)
1601 if (!inode->i_sb->dq_op->get_reserved_space)
1603 spin_lock(&inode->i_lock);
1604 ret = *inode_reserved_space(inode);
1605 spin_unlock(&inode->i_lock);
1609 static void inode_incr_space(struct inode *inode, qsize_t number,
1613 inode_add_rsv_space(inode, number);
1615 inode_add_bytes(inode, number);
1618 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1621 inode_sub_rsv_space(inode, number);
1623 inode_sub_bytes(inode, number);
1627 * This functions updates i_blocks+i_bytes fields and quota information
1628 * (together with appropriate checks).
1630 * NOTE: We absolutely rely on the fact that caller dirties the inode
1631 * (usually helpers in quotaops.h care about this) and holds a handle for
1632 * the current transaction so that dquot write and inode write go into the
1637 * This operation can block, but only after everything is updated
1639 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1641 int cnt, ret = 0, index;
1642 struct dquot_warn warn[MAXQUOTAS];
1643 int reserve = flags & DQUOT_SPACE_RESERVE;
1644 struct dquot **dquots;
1646 if (!dquot_active(inode)) {
1647 inode_incr_space(inode, number, reserve);
1651 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1652 warn[cnt].w_type = QUOTA_NL_NOWARN;
1654 dquots = i_dquot(inode);
1655 index = srcu_read_lock(&dquot_srcu);
1656 spin_lock(&dq_data_lock);
1657 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1660 ret = check_bdq(dquots[cnt], number,
1661 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1662 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1663 spin_unlock(&dq_data_lock);
1664 goto out_flush_warn;
1667 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1671 dquot_resv_space(dquots[cnt], number);
1673 dquot_incr_space(dquots[cnt], number);
1675 inode_incr_space(inode, number, reserve);
1676 spin_unlock(&dq_data_lock);
1679 goto out_flush_warn;
1680 mark_all_dquot_dirty(dquots);
1682 srcu_read_unlock(&dquot_srcu, index);
1683 flush_warnings(warn);
1687 EXPORT_SYMBOL(__dquot_alloc_space);
1690 * This operation can block, but only after everything is updated
1692 int dquot_alloc_inode(struct inode *inode)
1694 int cnt, ret = 0, index;
1695 struct dquot_warn warn[MAXQUOTAS];
1696 struct dquot * const *dquots;
1698 if (!dquot_active(inode))
1700 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1701 warn[cnt].w_type = QUOTA_NL_NOWARN;
1703 dquots = i_dquot(inode);
1704 index = srcu_read_lock(&dquot_srcu);
1705 spin_lock(&dq_data_lock);
1706 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1709 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1714 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1717 dquot_incr_inodes(dquots[cnt], 1);
1721 spin_unlock(&dq_data_lock);
1723 mark_all_dquot_dirty(dquots);
1724 srcu_read_unlock(&dquot_srcu, index);
1725 flush_warnings(warn);
1728 EXPORT_SYMBOL(dquot_alloc_inode);
1731 * Convert in-memory reserved quotas to real consumed quotas
1733 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1735 struct dquot **dquots;
1738 if (!dquot_active(inode)) {
1739 inode_claim_rsv_space(inode, number);
1743 dquots = i_dquot(inode);
1744 index = srcu_read_lock(&dquot_srcu);
1745 spin_lock(&dq_data_lock);
1746 /* Claim reserved quotas to allocated quotas */
1747 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1749 dquot_claim_reserved_space(dquots[cnt], number);
1751 /* Update inode bytes */
1752 inode_claim_rsv_space(inode, number);
1753 spin_unlock(&dq_data_lock);
1754 mark_all_dquot_dirty(dquots);
1755 srcu_read_unlock(&dquot_srcu, index);
1758 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1761 * Convert allocated space back to in-memory reserved quotas
1763 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1765 struct dquot **dquots;
1768 if (!dquot_active(inode)) {
1769 inode_reclaim_rsv_space(inode, number);
1773 dquots = i_dquot(inode);
1774 index = srcu_read_lock(&dquot_srcu);
1775 spin_lock(&dq_data_lock);
1776 /* Claim reserved quotas to allocated quotas */
1777 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1779 dquot_reclaim_reserved_space(dquots[cnt], number);
1781 /* Update inode bytes */
1782 inode_reclaim_rsv_space(inode, number);
1783 spin_unlock(&dq_data_lock);
1784 mark_all_dquot_dirty(dquots);
1785 srcu_read_unlock(&dquot_srcu, index);
1788 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1791 * This operation can block, but only after everything is updated
1793 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1796 struct dquot_warn warn[MAXQUOTAS];
1797 struct dquot **dquots;
1798 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1800 if (!dquot_active(inode)) {
1801 inode_decr_space(inode, number, reserve);
1805 dquots = i_dquot(inode);
1806 index = srcu_read_lock(&dquot_srcu);
1807 spin_lock(&dq_data_lock);
1808 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1811 warn[cnt].w_type = QUOTA_NL_NOWARN;
1814 wtype = info_bdq_free(dquots[cnt], number);
1815 if (wtype != QUOTA_NL_NOWARN)
1816 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1818 dquot_free_reserved_space(dquots[cnt], number);
1820 dquot_decr_space(dquots[cnt], number);
1822 inode_decr_space(inode, number, reserve);
1823 spin_unlock(&dq_data_lock);
1827 mark_all_dquot_dirty(dquots);
1829 srcu_read_unlock(&dquot_srcu, index);
1830 flush_warnings(warn);
1832 EXPORT_SYMBOL(__dquot_free_space);
1835 * This operation can block, but only after everything is updated
1837 void dquot_free_inode(struct inode *inode)
1840 struct dquot_warn warn[MAXQUOTAS];
1841 struct dquot * const *dquots;
1844 if (!dquot_active(inode))
1847 dquots = i_dquot(inode);
1848 index = srcu_read_lock(&dquot_srcu);
1849 spin_lock(&dq_data_lock);
1850 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1853 warn[cnt].w_type = QUOTA_NL_NOWARN;
1856 wtype = info_idq_free(dquots[cnt], 1);
1857 if (wtype != QUOTA_NL_NOWARN)
1858 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1859 dquot_decr_inodes(dquots[cnt], 1);
1861 spin_unlock(&dq_data_lock);
1862 mark_all_dquot_dirty(dquots);
1863 srcu_read_unlock(&dquot_srcu, index);
1864 flush_warnings(warn);
1866 EXPORT_SYMBOL(dquot_free_inode);
1869 * Transfer the number of inode and blocks from one diskquota to an other.
1870 * On success, dquot references in transfer_to are consumed and references
1871 * to original dquots that need to be released are placed there. On failure,
1872 * references are kept untouched.
1874 * This operation can block, but only after everything is updated
1875 * A transaction must be started when entering this function.
1877 * We are holding reference on transfer_from & transfer_to, no need to
1878 * protect them by srcu_read_lock().
1880 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1882 qsize_t space, cur_space;
1883 qsize_t rsv_space = 0;
1884 struct dquot *transfer_from[MAXQUOTAS] = {};
1886 char is_valid[MAXQUOTAS] = {};
1887 struct dquot_warn warn_to[MAXQUOTAS];
1888 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1889 struct dquot_warn warn_from_space[MAXQUOTAS];
1891 if (IS_NOQUOTA(inode))
1893 /* Initialize the arrays */
1894 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1895 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1896 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1897 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1900 spin_lock(&dq_data_lock);
1901 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1902 spin_unlock(&dq_data_lock);
1905 cur_space = inode_get_bytes(inode);
1906 rsv_space = inode_get_rsv_space(inode);
1907 space = cur_space + rsv_space;
1908 /* Build the transfer_from list and check the limits */
1909 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1911 * Skip changes for same uid or gid or for turned off quota-type.
1913 if (!transfer_to[cnt])
1915 /* Avoid races with quotaoff() */
1916 if (!sb_has_quota_active(inode->i_sb, cnt))
1919 transfer_from[cnt] = i_dquot(inode)[cnt];
1920 ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1923 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1929 * Finally perform the needed transfer from transfer_from to transfer_to
1931 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1934 /* Due to IO error we might not have transfer_from[] structure */
1935 if (transfer_from[cnt]) {
1937 wtype = info_idq_free(transfer_from[cnt], 1);
1938 if (wtype != QUOTA_NL_NOWARN)
1939 prepare_warning(&warn_from_inodes[cnt],
1940 transfer_from[cnt], wtype);
1941 wtype = info_bdq_free(transfer_from[cnt], space);
1942 if (wtype != QUOTA_NL_NOWARN)
1943 prepare_warning(&warn_from_space[cnt],
1944 transfer_from[cnt], wtype);
1945 dquot_decr_inodes(transfer_from[cnt], 1);
1946 dquot_decr_space(transfer_from[cnt], cur_space);
1947 dquot_free_reserved_space(transfer_from[cnt],
1951 dquot_incr_inodes(transfer_to[cnt], 1);
1952 dquot_incr_space(transfer_to[cnt], cur_space);
1953 dquot_resv_space(transfer_to[cnt], rsv_space);
1955 i_dquot(inode)[cnt] = transfer_to[cnt];
1957 spin_unlock(&dq_data_lock);
1959 mark_all_dquot_dirty(transfer_from);
1960 mark_all_dquot_dirty(transfer_to);
1961 flush_warnings(warn_to);
1962 flush_warnings(warn_from_inodes);
1963 flush_warnings(warn_from_space);
1964 /* Pass back references to put */
1965 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1967 transfer_to[cnt] = transfer_from[cnt];
1970 spin_unlock(&dq_data_lock);
1971 flush_warnings(warn_to);
1974 EXPORT_SYMBOL(__dquot_transfer);
1976 /* Wrapper for transferring ownership of an inode for uid/gid only
1977 * Called from FSXXX_setattr()
1979 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1981 struct dquot *transfer_to[MAXQUOTAS] = {};
1982 struct dquot *dquot;
1983 struct super_block *sb = inode->i_sb;
1986 if (!dquot_active(inode))
1989 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
1990 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
1991 if (IS_ERR(dquot)) {
1992 if (PTR_ERR(dquot) != -ESRCH) {
1993 ret = PTR_ERR(dquot);
1998 transfer_to[USRQUOTA] = dquot;
2000 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2001 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2002 if (IS_ERR(dquot)) {
2003 if (PTR_ERR(dquot) != -ESRCH) {
2004 ret = PTR_ERR(dquot);
2009 transfer_to[GRPQUOTA] = dquot;
2011 ret = __dquot_transfer(inode, transfer_to);
2013 dqput_all(transfer_to);
2016 EXPORT_SYMBOL(dquot_transfer);
2019 * Write info of quota file to disk
2021 int dquot_commit_info(struct super_block *sb, int type)
2024 struct quota_info *dqopt = sb_dqopt(sb);
2026 mutex_lock(&dqopt->dqio_mutex);
2027 ret = dqopt->ops[type]->write_file_info(sb, type);
2028 mutex_unlock(&dqopt->dqio_mutex);
2031 EXPORT_SYMBOL(dquot_commit_info);
2034 * Definitions of diskquota operations.
2036 const struct dquot_operations dquot_operations = {
2037 .write_dquot = dquot_commit,
2038 .acquire_dquot = dquot_acquire,
2039 .release_dquot = dquot_release,
2040 .mark_dirty = dquot_mark_dquot_dirty,
2041 .write_info = dquot_commit_info,
2042 .alloc_dquot = dquot_alloc,
2043 .destroy_dquot = dquot_destroy,
2045 EXPORT_SYMBOL(dquot_operations);
2048 * Generic helper for ->open on filesystems supporting disk quotas.
2050 int dquot_file_open(struct inode *inode, struct file *file)
2054 error = generic_file_open(inode, file);
2055 if (!error && (file->f_mode & FMODE_WRITE))
2056 dquot_initialize(inode);
2059 EXPORT_SYMBOL(dquot_file_open);
2062 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2064 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2067 struct quota_info *dqopt = sb_dqopt(sb);
2068 struct inode *toputinode[MAXQUOTAS];
2070 /* Cannot turn off usage accounting without turning off limits, or
2071 * suspend quotas and simultaneously turn quotas off. */
2072 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2073 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2074 DQUOT_USAGE_ENABLED)))
2077 /* We need to serialize quota_off() for device */
2078 mutex_lock(&dqopt->dqonoff_mutex);
2081 * Skip everything if there's nothing to do. We have to do this because
2082 * sometimes we are called when fill_super() failed and calling
2083 * sync_fs() in such cases does no good.
2085 if (!sb_any_quota_loaded(sb)) {
2086 mutex_unlock(&dqopt->dqonoff_mutex);
2089 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2090 toputinode[cnt] = NULL;
2091 if (type != -1 && cnt != type)
2093 if (!sb_has_quota_loaded(sb, cnt))
2096 if (flags & DQUOT_SUSPENDED) {
2097 spin_lock(&dq_state_lock);
2099 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2100 spin_unlock(&dq_state_lock);
2102 spin_lock(&dq_state_lock);
2103 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2104 /* Turning off suspended quotas? */
2105 if (!sb_has_quota_loaded(sb, cnt) &&
2106 sb_has_quota_suspended(sb, cnt)) {
2107 dqopt->flags &= ~dquot_state_flag(
2108 DQUOT_SUSPENDED, cnt);
2109 spin_unlock(&dq_state_lock);
2110 iput(dqopt->files[cnt]);
2111 dqopt->files[cnt] = NULL;
2114 spin_unlock(&dq_state_lock);
2117 /* We still have to keep quota loaded? */
2118 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2121 /* Note: these are blocking operations */
2122 drop_dquot_ref(sb, cnt);
2123 invalidate_dquots(sb, cnt);
2125 * Now all dquots should be invalidated, all writes done so we
2126 * should be only users of the info. No locks needed.
2128 if (info_dirty(&dqopt->info[cnt]))
2129 sb->dq_op->write_info(sb, cnt);
2130 if (dqopt->ops[cnt]->free_file_info)
2131 dqopt->ops[cnt]->free_file_info(sb, cnt);
2132 put_quota_format(dqopt->info[cnt].dqi_format);
2134 toputinode[cnt] = dqopt->files[cnt];
2135 if (!sb_has_quota_loaded(sb, cnt))
2136 dqopt->files[cnt] = NULL;
2137 dqopt->info[cnt].dqi_flags = 0;
2138 dqopt->info[cnt].dqi_igrace = 0;
2139 dqopt->info[cnt].dqi_bgrace = 0;
2140 dqopt->ops[cnt] = NULL;
2142 mutex_unlock(&dqopt->dqonoff_mutex);
2144 /* Skip syncing and setting flags if quota files are hidden */
2145 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2148 /* Sync the superblock so that buffers with quota data are written to
2149 * disk (and so userspace sees correct data afterwards). */
2150 if (sb->s_op->sync_fs)
2151 sb->s_op->sync_fs(sb, 1);
2152 sync_blockdev(sb->s_bdev);
2153 /* Now the quota files are just ordinary files and we can set the
2154 * inode flags back. Moreover we discard the pagecache so that
2155 * userspace sees the writes we did bypassing the pagecache. We
2156 * must also discard the blockdev buffers so that we see the
2157 * changes done by userspace on the next quotaon() */
2158 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2159 if (toputinode[cnt]) {
2160 mutex_lock(&dqopt->dqonoff_mutex);
2161 /* If quota was reenabled in the meantime, we have
2163 if (!sb_has_quota_loaded(sb, cnt)) {
2164 mutex_lock(&toputinode[cnt]->i_mutex);
2165 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2166 S_NOATIME | S_NOQUOTA);
2167 truncate_inode_pages(&toputinode[cnt]->i_data,
2169 mutex_unlock(&toputinode[cnt]->i_mutex);
2170 mark_inode_dirty_sync(toputinode[cnt]);
2172 mutex_unlock(&dqopt->dqonoff_mutex);
2175 invalidate_bdev(sb->s_bdev);
2177 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2178 if (toputinode[cnt]) {
2179 /* On remount RO, we keep the inode pointer so that we
2180 * can reenable quota on the subsequent remount RW. We
2181 * have to check 'flags' variable and not use sb_has_
2182 * function because another quotaon / quotaoff could
2183 * change global state before we got here. We refuse
2184 * to suspend quotas when there is pending delete on
2185 * the quota file... */
2186 if (!(flags & DQUOT_SUSPENDED))
2187 iput(toputinode[cnt]);
2188 else if (!toputinode[cnt]->i_nlink)
2193 EXPORT_SYMBOL(dquot_disable);
2195 int dquot_quota_off(struct super_block *sb, int type)
2197 return dquot_disable(sb, type,
2198 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2200 EXPORT_SYMBOL(dquot_quota_off);
2203 * Turn quotas on on a device
2207 * Helper function to turn quotas on when we already have the inode of
2208 * quota file and no quota information is loaded.
2210 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2213 struct quota_format_type *fmt = find_quota_format(format_id);
2214 struct super_block *sb = inode->i_sb;
2215 struct quota_info *dqopt = sb_dqopt(sb);
2221 if (!S_ISREG(inode->i_mode)) {
2225 if (IS_RDONLY(inode)) {
2229 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2230 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2234 /* Usage always has to be set... */
2235 if (!(flags & DQUOT_USAGE_ENABLED)) {
2240 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2241 /* As we bypass the pagecache we must now flush all the
2242 * dirty data and invalidate caches so that kernel sees
2243 * changes from userspace. It is not enough to just flush
2244 * the quota file since if blocksize < pagesize, invalidation
2245 * of the cache could fail because of other unrelated dirty
2247 sync_filesystem(sb);
2248 invalidate_bdev(sb->s_bdev);
2250 mutex_lock(&dqopt->dqonoff_mutex);
2251 if (sb_has_quota_loaded(sb, type)) {
2256 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2257 /* We don't want quota and atime on quota files (deadlocks
2258 * possible) Also nobody should write to the file - we use
2259 * special IO operations which ignore the immutable bit. */
2260 mutex_lock(&inode->i_mutex);
2261 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2263 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2264 mutex_unlock(&inode->i_mutex);
2266 * When S_NOQUOTA is set, remove dquot references as no more
2267 * references can be added
2269 __dquot_drop(inode);
2273 dqopt->files[type] = igrab(inode);
2274 if (!dqopt->files[type])
2277 if (!fmt->qf_ops->check_quota_file(sb, type))
2280 dqopt->ops[type] = fmt->qf_ops;
2281 dqopt->info[type].dqi_format = fmt;
2282 dqopt->info[type].dqi_fmt_id = format_id;
2283 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2284 mutex_lock(&dqopt->dqio_mutex);
2285 error = dqopt->ops[type]->read_file_info(sb, type);
2287 mutex_unlock(&dqopt->dqio_mutex);
2290 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2291 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2292 mutex_unlock(&dqopt->dqio_mutex);
2293 spin_lock(&dq_state_lock);
2294 dqopt->flags |= dquot_state_flag(flags, type);
2295 spin_unlock(&dq_state_lock);
2297 add_dquot_ref(sb, type);
2298 mutex_unlock(&dqopt->dqonoff_mutex);
2303 dqopt->files[type] = NULL;
2306 if (oldflags != -1) {
2307 mutex_lock(&inode->i_mutex);
2308 /* Set the flags back (in the case of accidental quotaon()
2309 * on a wrong file we don't want to mess up the flags) */
2310 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2311 inode->i_flags |= oldflags;
2312 mutex_unlock(&inode->i_mutex);
2314 mutex_unlock(&dqopt->dqonoff_mutex);
2316 put_quota_format(fmt);
2321 /* Reenable quotas on remount RW */
2322 int dquot_resume(struct super_block *sb, int type)
2324 struct quota_info *dqopt = sb_dqopt(sb);
2325 struct inode *inode;
2329 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2330 if (type != -1 && cnt != type)
2333 mutex_lock(&dqopt->dqonoff_mutex);
2334 if (!sb_has_quota_suspended(sb, cnt)) {
2335 mutex_unlock(&dqopt->dqonoff_mutex);
2338 inode = dqopt->files[cnt];
2339 dqopt->files[cnt] = NULL;
2340 spin_lock(&dq_state_lock);
2341 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2342 DQUOT_LIMITS_ENABLED,
2344 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2345 spin_unlock(&dq_state_lock);
2346 mutex_unlock(&dqopt->dqonoff_mutex);
2348 flags = dquot_generic_flag(flags, cnt);
2349 ret = vfs_load_quota_inode(inode, cnt,
2350 dqopt->info[cnt].dqi_fmt_id, flags);
2356 EXPORT_SYMBOL(dquot_resume);
2358 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2361 int error = security_quota_on(path->dentry);
2364 /* Quota file not on the same filesystem? */
2365 if (path->dentry->d_sb != sb)
2368 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2369 format_id, DQUOT_USAGE_ENABLED |
2370 DQUOT_LIMITS_ENABLED);
2373 EXPORT_SYMBOL(dquot_quota_on);
2376 * More powerful function for turning on quotas allowing setting
2377 * of individual quota flags
2379 int dquot_enable(struct inode *inode, int type, int format_id,
2383 struct super_block *sb = inode->i_sb;
2384 struct quota_info *dqopt = sb_dqopt(sb);
2386 /* Just unsuspend quotas? */
2387 BUG_ON(flags & DQUOT_SUSPENDED);
2391 /* Just updating flags needed? */
2392 if (sb_has_quota_loaded(sb, type)) {
2393 mutex_lock(&dqopt->dqonoff_mutex);
2394 /* Now do a reliable test... */
2395 if (!sb_has_quota_loaded(sb, type)) {
2396 mutex_unlock(&dqopt->dqonoff_mutex);
2399 if (flags & DQUOT_USAGE_ENABLED &&
2400 sb_has_quota_usage_enabled(sb, type)) {
2404 if (flags & DQUOT_LIMITS_ENABLED &&
2405 sb_has_quota_limits_enabled(sb, type)) {
2409 spin_lock(&dq_state_lock);
2410 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2411 spin_unlock(&dq_state_lock);
2413 mutex_unlock(&dqopt->dqonoff_mutex);
2418 return vfs_load_quota_inode(inode, type, format_id, flags);
2420 EXPORT_SYMBOL(dquot_enable);
2423 * This function is used when filesystem needs to initialize quotas
2424 * during mount time.
2426 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2427 int format_id, int type)
2429 struct dentry *dentry;
2432 mutex_lock(&d_inode(sb->s_root)->i_mutex);
2433 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2434 mutex_unlock(&d_inode(sb->s_root)->i_mutex);
2436 return PTR_ERR(dentry);
2438 if (d_really_is_negative(dentry)) {
2443 error = security_quota_on(dentry);
2445 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2446 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2452 EXPORT_SYMBOL(dquot_quota_on_mount);
2454 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2458 struct quota_info *dqopt = sb_dqopt(sb);
2460 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2462 /* Accounting cannot be turned on while fs is mounted */
2463 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2466 for (type = 0; type < MAXQUOTAS; type++) {
2467 if (!(flags & qtype_enforce_flag(type)))
2469 /* Can't enforce without accounting */
2470 if (!sb_has_quota_usage_enabled(sb, type))
2472 ret = dquot_enable(dqopt->files[type], type,
2473 dqopt->info[type].dqi_fmt_id,
2474 DQUOT_LIMITS_ENABLED);
2480 /* Backout enforcement enablement we already did */
2481 for (type--; type >= 0; type--) {
2482 if (flags & qtype_enforce_flag(type))
2483 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2485 /* Error code translation for better compatibility with XFS */
2491 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2495 struct quota_info *dqopt = sb_dqopt(sb);
2497 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2500 * We don't support turning off accounting via quotactl. In principle
2501 * quota infrastructure can do this but filesystems don't expect
2502 * userspace to be able to do it.
2505 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2508 /* Filter out limits not enabled */
2509 for (type = 0; type < MAXQUOTAS; type++)
2510 if (!sb_has_quota_limits_enabled(sb, type))
2511 flags &= ~qtype_enforce_flag(type);
2515 for (type = 0; type < MAXQUOTAS; type++) {
2516 if (flags & qtype_enforce_flag(type)) {
2517 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2524 /* Backout enforcement disabling we already did */
2525 for (type--; type >= 0; type--) {
2526 if (flags & qtype_enforce_flag(type))
2527 dquot_enable(dqopt->files[type], type,
2528 dqopt->info[type].dqi_fmt_id,
2529 DQUOT_LIMITS_ENABLED);
2534 /* Generic routine for getting common part of quota structure */
2535 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2537 struct mem_dqblk *dm = &dquot->dq_dqb;
2539 memset(di, 0, sizeof(*di));
2540 spin_lock(&dq_data_lock);
2541 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2542 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2543 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2544 di->d_ino_softlimit = dm->dqb_isoftlimit;
2545 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2546 di->d_ino_count = dm->dqb_curinodes;
2547 di->d_spc_timer = dm->dqb_btime;
2548 di->d_ino_timer = dm->dqb_itime;
2549 spin_unlock(&dq_data_lock);
2552 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2553 struct qc_dqblk *di)
2555 struct dquot *dquot;
2557 dquot = dqget(sb, qid);
2559 return PTR_ERR(dquot);
2560 do_get_dqblk(dquot, di);
2565 EXPORT_SYMBOL(dquot_get_dqblk);
2567 #define VFS_QC_MASK \
2568 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2569 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2570 QC_SPC_TIMER | QC_INO_TIMER)
2572 /* Generic routine for setting common part of quota structure */
2573 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2575 struct mem_dqblk *dm = &dquot->dq_dqb;
2576 int check_blim = 0, check_ilim = 0;
2577 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2579 if (di->d_fieldmask & ~VFS_QC_MASK)
2582 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2583 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2584 ((di->d_fieldmask & QC_SPC_HARD) &&
2585 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2586 ((di->d_fieldmask & QC_INO_SOFT) &&
2587 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2588 ((di->d_fieldmask & QC_INO_HARD) &&
2589 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2592 spin_lock(&dq_data_lock);
2593 if (di->d_fieldmask & QC_SPACE) {
2594 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2596 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2599 if (di->d_fieldmask & QC_SPC_SOFT)
2600 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2601 if (di->d_fieldmask & QC_SPC_HARD)
2602 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2603 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2605 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2608 if (di->d_fieldmask & QC_INO_COUNT) {
2609 dm->dqb_curinodes = di->d_ino_count;
2611 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2614 if (di->d_fieldmask & QC_INO_SOFT)
2615 dm->dqb_isoftlimit = di->d_ino_softlimit;
2616 if (di->d_fieldmask & QC_INO_HARD)
2617 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2618 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2620 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2623 if (di->d_fieldmask & QC_SPC_TIMER) {
2624 dm->dqb_btime = di->d_spc_timer;
2626 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2629 if (di->d_fieldmask & QC_INO_TIMER) {
2630 dm->dqb_itime = di->d_ino_timer;
2632 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2636 if (!dm->dqb_bsoftlimit ||
2637 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2639 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2640 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2641 /* Set grace only if user hasn't provided his own... */
2642 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2645 if (!dm->dqb_isoftlimit ||
2646 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2648 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2649 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2650 /* Set grace only if user hasn't provided his own... */
2651 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2653 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2655 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2657 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2658 spin_unlock(&dq_data_lock);
2659 mark_dquot_dirty(dquot);
2664 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2665 struct qc_dqblk *di)
2667 struct dquot *dquot;
2670 dquot = dqget(sb, qid);
2671 if (IS_ERR(dquot)) {
2672 rc = PTR_ERR(dquot);
2675 rc = do_set_dqblk(dquot, di);
2680 EXPORT_SYMBOL(dquot_set_dqblk);
2682 /* Generic routine for getting common part of quota file information */
2683 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2685 struct mem_dqinfo *mi;
2686 struct qc_type_state *tstate;
2687 struct quota_info *dqopt = sb_dqopt(sb);
2690 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2691 memset(state, 0, sizeof(*state));
2692 for (type = 0; type < MAXQUOTAS; type++) {
2693 if (!sb_has_quota_active(sb, type))
2695 tstate = state->s_state + type;
2696 mi = sb_dqopt(sb)->info + type;
2697 tstate->flags = QCI_ACCT_ENABLED;
2698 spin_lock(&dq_data_lock);
2699 if (mi->dqi_flags & DQF_SYS_FILE)
2700 tstate->flags |= QCI_SYSFILE;
2701 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2702 tstate->flags |= QCI_ROOT_SQUASH;
2703 if (sb_has_quota_limits_enabled(sb, type))
2704 tstate->flags |= QCI_LIMITS_ENFORCED;
2705 tstate->spc_timelimit = mi->dqi_bgrace;
2706 tstate->ino_timelimit = mi->dqi_igrace;
2707 tstate->ino = dqopt->files[type]->i_ino;
2708 tstate->blocks = dqopt->files[type]->i_blocks;
2709 tstate->nextents = 1; /* We don't know... */
2710 spin_unlock(&dq_data_lock);
2712 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2715 EXPORT_SYMBOL(dquot_get_state);
2717 /* Generic routine for setting common part of quota file information */
2718 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2720 struct mem_dqinfo *mi;
2723 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2724 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2726 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2727 if (!sb_has_quota_active(sb, type)) {
2731 mi = sb_dqopt(sb)->info + type;
2732 if (ii->i_fieldmask & QC_FLAGS) {
2733 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2734 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD)) {
2739 spin_lock(&dq_data_lock);
2740 if (ii->i_fieldmask & QC_SPC_TIMER)
2741 mi->dqi_bgrace = ii->i_spc_timelimit;
2742 if (ii->i_fieldmask & QC_INO_TIMER)
2743 mi->dqi_igrace = ii->i_ino_timelimit;
2744 if (ii->i_fieldmask & QC_FLAGS) {
2745 if (ii->i_flags & QCI_ROOT_SQUASH)
2746 mi->dqi_flags |= DQF_ROOT_SQUASH;
2748 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2750 spin_unlock(&dq_data_lock);
2751 mark_info_dirty(sb, type);
2752 /* Force write to disk */
2753 sb->dq_op->write_info(sb, type);
2755 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2758 EXPORT_SYMBOL(dquot_set_dqinfo);
2760 const struct quotactl_ops dquot_quotactl_ops = {
2761 .quota_on = dquot_quota_on,
2762 .quota_off = dquot_quota_off,
2763 .quota_sync = dquot_quota_sync,
2764 .get_state = dquot_get_state,
2765 .set_info = dquot_set_dqinfo,
2766 .get_dqblk = dquot_get_dqblk,
2767 .set_dqblk = dquot_set_dqblk
2769 EXPORT_SYMBOL(dquot_quotactl_ops);
2771 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2772 .quota_enable = dquot_quota_enable,
2773 .quota_disable = dquot_quota_disable,
2774 .quota_sync = dquot_quota_sync,
2775 .get_state = dquot_get_state,
2776 .set_info = dquot_set_dqinfo,
2777 .get_dqblk = dquot_get_dqblk,
2778 .set_dqblk = dquot_set_dqblk
2780 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2782 static int do_proc_dqstats(struct ctl_table *table, int write,
2783 void __user *buffer, size_t *lenp, loff_t *ppos)
2785 unsigned int type = (int *)table->data - dqstats.stat;
2787 /* Update global table */
2788 dqstats.stat[type] =
2789 percpu_counter_sum_positive(&dqstats.counter[type]);
2790 return proc_dointvec(table, write, buffer, lenp, ppos);
2793 static struct ctl_table fs_dqstats_table[] = {
2795 .procname = "lookups",
2796 .data = &dqstats.stat[DQST_LOOKUPS],
2797 .maxlen = sizeof(int),
2799 .proc_handler = do_proc_dqstats,
2802 .procname = "drops",
2803 .data = &dqstats.stat[DQST_DROPS],
2804 .maxlen = sizeof(int),
2806 .proc_handler = do_proc_dqstats,
2809 .procname = "reads",
2810 .data = &dqstats.stat[DQST_READS],
2811 .maxlen = sizeof(int),
2813 .proc_handler = do_proc_dqstats,
2816 .procname = "writes",
2817 .data = &dqstats.stat[DQST_WRITES],
2818 .maxlen = sizeof(int),
2820 .proc_handler = do_proc_dqstats,
2823 .procname = "cache_hits",
2824 .data = &dqstats.stat[DQST_CACHE_HITS],
2825 .maxlen = sizeof(int),
2827 .proc_handler = do_proc_dqstats,
2830 .procname = "allocated_dquots",
2831 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2832 .maxlen = sizeof(int),
2834 .proc_handler = do_proc_dqstats,
2837 .procname = "free_dquots",
2838 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2839 .maxlen = sizeof(int),
2841 .proc_handler = do_proc_dqstats,
2844 .procname = "syncs",
2845 .data = &dqstats.stat[DQST_SYNCS],
2846 .maxlen = sizeof(int),
2848 .proc_handler = do_proc_dqstats,
2850 #ifdef CONFIG_PRINT_QUOTA_WARNING
2852 .procname = "warnings",
2853 .data = &flag_print_warnings,
2854 .maxlen = sizeof(int),
2856 .proc_handler = proc_dointvec,
2862 static struct ctl_table fs_table[] = {
2864 .procname = "quota",
2866 .child = fs_dqstats_table,
2871 static struct ctl_table sys_table[] = {
2880 static int __init dquot_init(void)
2883 unsigned long nr_hash, order;
2885 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2887 register_sysctl_table(sys_table);
2889 dquot_cachep = kmem_cache_create("dquot",
2890 sizeof(struct dquot), sizeof(unsigned long) * 4,
2891 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2892 SLAB_MEM_SPREAD|SLAB_PANIC),
2896 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2898 panic("Cannot create dquot hash table");
2900 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2901 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2903 panic("Cannot create dquot stat counters");
2906 /* Find power-of-two hlist_heads which can fit into allocation */
2907 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2911 } while (nr_hash >> dq_hash_bits);
2914 nr_hash = 1UL << dq_hash_bits;
2915 dq_hash_mask = nr_hash - 1;
2916 for (i = 0; i < nr_hash; i++)
2917 INIT_HLIST_HEAD(dquot_hash + i);
2919 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2920 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2922 register_shrinker(&dqcache_shrinker);
2926 module_init(dquot_init);