2 * Implementation of the diskquota system for the LINUX operating system. QUOTA
3 * is implemented using the BSD system call interface as the means of
4 * communication with the user level. This file contains the generic routines
5 * called by the different filesystems on allocation of an inode or block.
6 * These routines take care of the administration needed to have a consistent
7 * diskquota tracking system. The ideas of both user and group quotas are based
8 * on the Melbourne quota system as used on BSD derived systems. The internal
9 * implementation is based on one of the several variants of the LINUX
10 * inode-subsystem with added complexity of the diskquota system.
12 * Author: Marco van Wieringen <mvw@planets.elm.net>
14 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 * Revised list management to avoid races
17 * -- Bill Hawes, <whawes@star.net>, 9/98
19 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20 * As the consequence the locking was moved from dquot_decr_...(),
21 * dquot_incr_...() to calling functions.
22 * invalidate_dquots() now writes modified dquots.
23 * Serialized quota_off() and quota_on() for mount point.
24 * Fixed a few bugs in grow_dquots().
25 * Fixed deadlock in write_dquot() - we no longer account quotas on
27 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
28 * add_dquot_ref() restarts after blocking
29 * Added check for bogus uid and fixed check for group in quotactl.
30 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32 * Used struct list_head instead of own list struct
33 * Invalidation of referenced dquots is no longer possible
34 * Improved free_dquots list management
35 * Quota and i_blocks are now updated in one place to avoid races
36 * Warnings are now delayed so we won't block in critical section
37 * Write updated not to require dquot lock
38 * Jan Kara, <jack@suse.cz>, 9/2000
40 * Added dynamic quota structure allocation
41 * Jan Kara <jack@suse.cz> 12/2000
43 * Rewritten quota interface. Implemented new quota format and
44 * formats registering.
45 * Jan Kara, <jack@suse.cz>, 2001,2002
48 * Jan Kara, <jack@suse.cz>, 10/2002
50 * Added journalled quota support, fix lock inversion problems
51 * Jan Kara, <jack@suse.cz>, 2003,2004
53 * (C) Copyright 1994 - 1997 Marco van Wieringen
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
59 #include <linux/mount.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/cred.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/capability.h>
79 #include <linux/quotaops.h>
80 #include "../internal.h" /* ugh */
82 #include <linux/uaccess.h>
85 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
87 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
88 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
89 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
90 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
91 * modifications of quota state (on quotaon and quotaoff) and readers who care
92 * about latest values take it as well.
94 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
95 * dq_list_lock > dq_state_lock
97 * Note that some things (eg. sb pointer, type, id) doesn't change during
98 * the life of the dquot structure and so needn't to be protected by a lock
100 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
101 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
102 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
103 * inode and before dropping dquot references to avoid use of dquots after
104 * they are freed. dq_data_lock is used to serialize the pointer setting and
105 * clearing operations.
106 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
107 * inode is a quota file). Functions adding pointers from inode to dquots have
108 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
109 * have to do all pointer modifications before dropping dq_data_lock. This makes
110 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
111 * then drops all pointers to dquots from an inode.
113 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
114 * memory (or space for it is being allocated) on the first dqget(), when it is
115 * being written out, and when it is being released on the last dqput(). The
116 * allocation and release operations are serialized by the dq_lock and by
117 * checking the use count in dquot_release().
119 * Lock ordering (including related VFS locks) is the following:
120 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
123 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
124 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
125 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
126 EXPORT_SYMBOL(dq_data_lock);
127 DEFINE_STATIC_SRCU(dquot_srcu);
129 void __quota_error(struct super_block *sb, const char *func,
130 const char *fmt, ...)
132 if (printk_ratelimit()) {
134 struct va_format vaf;
141 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
142 sb->s_id, func, &vaf);
147 EXPORT_SYMBOL(__quota_error);
149 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
150 static char *quotatypes[] = INITQFNAMES;
152 static struct quota_format_type *quota_formats; /* List of registered formats */
153 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
155 /* SLAB cache for dquot structures */
156 static struct kmem_cache *dquot_cachep;
158 int register_quota_format(struct quota_format_type *fmt)
160 spin_lock(&dq_list_lock);
161 fmt->qf_next = quota_formats;
163 spin_unlock(&dq_list_lock);
166 EXPORT_SYMBOL(register_quota_format);
168 void unregister_quota_format(struct quota_format_type *fmt)
170 struct quota_format_type **actqf;
172 spin_lock(&dq_list_lock);
173 for (actqf = "a_formats; *actqf && *actqf != fmt;
174 actqf = &(*actqf)->qf_next)
177 *actqf = (*actqf)->qf_next;
178 spin_unlock(&dq_list_lock);
180 EXPORT_SYMBOL(unregister_quota_format);
182 static struct quota_format_type *find_quota_format(int id)
184 struct quota_format_type *actqf;
186 spin_lock(&dq_list_lock);
187 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
188 actqf = actqf->qf_next)
190 if (!actqf || !try_module_get(actqf->qf_owner)) {
193 spin_unlock(&dq_list_lock);
195 for (qm = 0; module_names[qm].qm_fmt_id &&
196 module_names[qm].qm_fmt_id != id; qm++)
198 if (!module_names[qm].qm_fmt_id ||
199 request_module(module_names[qm].qm_mod_name))
202 spin_lock(&dq_list_lock);
203 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
204 actqf = actqf->qf_next)
206 if (actqf && !try_module_get(actqf->qf_owner))
209 spin_unlock(&dq_list_lock);
213 static void put_quota_format(struct quota_format_type *fmt)
215 module_put(fmt->qf_owner);
219 * Dquot List Management:
220 * The quota code uses three lists for dquot management: the inuse_list,
221 * free_dquots, and dquot_hash[] array. A single dquot structure may be
222 * on all three lists, depending on its current state.
224 * All dquots are placed to the end of inuse_list when first created, and this
225 * list is used for invalidate operation, which must look at every dquot.
227 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
228 * and this list is searched whenever we need an available dquot. Dquots are
229 * removed from the list as soon as they are used again, and
230 * dqstats.free_dquots gives the number of dquots on the list. When
231 * dquot is invalidated it's completely released from memory.
233 * Dquots with a specific identity (device, type and id) are placed on
234 * one of the dquot_hash[] hash chains. The provides an efficient search
235 * mechanism to locate a specific dquot.
238 static LIST_HEAD(inuse_list);
239 static LIST_HEAD(free_dquots);
240 static unsigned int dq_hash_bits, dq_hash_mask;
241 static struct hlist_head *dquot_hash;
243 struct dqstats dqstats;
244 EXPORT_SYMBOL(dqstats);
246 static qsize_t inode_get_rsv_space(struct inode *inode);
247 static int __dquot_initialize(struct inode *inode, int type);
249 static inline unsigned int
250 hashfn(const struct super_block *sb, struct kqid qid)
252 unsigned int id = from_kqid(&init_user_ns, qid);
256 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
257 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
261 * Following list functions expect dq_list_lock to be held
263 static inline void insert_dquot_hash(struct dquot *dquot)
265 struct hlist_head *head;
266 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
267 hlist_add_head(&dquot->dq_hash, head);
270 static inline void remove_dquot_hash(struct dquot *dquot)
272 hlist_del_init(&dquot->dq_hash);
275 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
278 struct hlist_node *node;
281 hlist_for_each (node, dquot_hash+hashent) {
282 dquot = hlist_entry(node, struct dquot, dq_hash);
283 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
289 /* Add a dquot to the tail of the free list */
290 static inline void put_dquot_last(struct dquot *dquot)
292 list_add_tail(&dquot->dq_free, &free_dquots);
293 dqstats_inc(DQST_FREE_DQUOTS);
296 static inline void remove_free_dquot(struct dquot *dquot)
298 if (list_empty(&dquot->dq_free))
300 list_del_init(&dquot->dq_free);
301 dqstats_dec(DQST_FREE_DQUOTS);
304 static inline void put_inuse(struct dquot *dquot)
306 /* We add to the back of inuse list so we don't have to restart
307 * when traversing this list and we block */
308 list_add_tail(&dquot->dq_inuse, &inuse_list);
309 dqstats_inc(DQST_ALLOC_DQUOTS);
312 static inline void remove_inuse(struct dquot *dquot)
314 dqstats_dec(DQST_ALLOC_DQUOTS);
315 list_del(&dquot->dq_inuse);
318 * End of list functions needing dq_list_lock
321 static void wait_on_dquot(struct dquot *dquot)
323 mutex_lock(&dquot->dq_lock);
324 mutex_unlock(&dquot->dq_lock);
327 static inline int dquot_dirty(struct dquot *dquot)
329 return test_bit(DQ_MOD_B, &dquot->dq_flags);
332 static inline int mark_dquot_dirty(struct dquot *dquot)
334 return dquot->dq_sb->dq_op->mark_dirty(dquot);
337 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
338 int dquot_mark_dquot_dirty(struct dquot *dquot)
342 /* If quota is dirty already, we don't have to acquire dq_list_lock */
343 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
346 spin_lock(&dq_list_lock);
347 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
348 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
349 info[dquot->dq_id.type].dqi_dirty_list);
352 spin_unlock(&dq_list_lock);
355 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
357 /* Dirtify all the dquots - this can block when journalling */
358 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
363 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
365 /* Even in case of error we have to continue */
366 ret = mark_dquot_dirty(dquot[cnt]);
373 static inline void dqput_all(struct dquot **dquot)
377 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
381 /* This function needs dq_list_lock */
382 static inline int clear_dquot_dirty(struct dquot *dquot)
384 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
386 list_del_init(&dquot->dq_dirty);
390 void mark_info_dirty(struct super_block *sb, int type)
392 spin_lock(&dq_data_lock);
393 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
394 spin_unlock(&dq_data_lock);
396 EXPORT_SYMBOL(mark_info_dirty);
399 * Read dquot from disk and alloc space for it
402 int dquot_acquire(struct dquot *dquot)
404 int ret = 0, ret2 = 0;
405 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
407 mutex_lock(&dquot->dq_lock);
408 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
409 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
412 /* Make sure flags update is visible after dquot has been filled */
413 smp_mb__before_atomic();
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);
431 * Make sure flags update is visible after on-disk struct has been
432 * allocated. Paired with smp_rmb() in dqget().
434 smp_mb__before_atomic();
435 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
437 mutex_unlock(&dquot->dq_lock);
440 EXPORT_SYMBOL(dquot_acquire);
443 * Write dquot to disk
445 int dquot_commit(struct dquot *dquot)
448 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
450 mutex_lock(&dquot->dq_lock);
451 spin_lock(&dq_list_lock);
452 if (!clear_dquot_dirty(dquot)) {
453 spin_unlock(&dq_list_lock);
456 spin_unlock(&dq_list_lock);
457 /* Inactive dquot can be only if there was error during read/init
458 * => we have better not writing it */
459 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
460 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
464 mutex_unlock(&dquot->dq_lock);
467 EXPORT_SYMBOL(dquot_commit);
472 int dquot_release(struct dquot *dquot)
474 int ret = 0, ret2 = 0;
475 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
477 mutex_lock(&dquot->dq_lock);
478 /* Check whether we are not racing with some other dqget() */
479 if (atomic_read(&dquot->dq_count) > 1)
481 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
482 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
484 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
485 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
486 dquot->dq_sb, dquot->dq_id.type);
491 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
493 mutex_unlock(&dquot->dq_lock);
496 EXPORT_SYMBOL(dquot_release);
498 void dquot_destroy(struct dquot *dquot)
500 kmem_cache_free(dquot_cachep, dquot);
502 EXPORT_SYMBOL(dquot_destroy);
504 static inline void do_destroy_dquot(struct dquot *dquot)
506 dquot->dq_sb->dq_op->destroy_dquot(dquot);
509 /* Invalidate all dquots on the list. Note that this function is called after
510 * quota is disabled and pointers from inodes removed so there cannot be new
511 * quota users. There can still be some users of quotas due to inodes being
512 * just deleted or pruned by prune_icache() (those are not attached to any
513 * list) or parallel quotactl call. We have to wait for such users.
515 static void invalidate_dquots(struct super_block *sb, int type)
517 struct dquot *dquot, *tmp;
520 spin_lock(&dq_list_lock);
521 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
522 if (dquot->dq_sb != sb)
524 if (dquot->dq_id.type != type)
526 /* Wait for dquot users */
527 if (atomic_read(&dquot->dq_count)) {
531 prepare_to_wait(&dquot->dq_wait_unused, &wait,
532 TASK_UNINTERRUPTIBLE);
533 spin_unlock(&dq_list_lock);
534 /* Once dqput() wakes us up, we know it's time to free
536 * IMPORTANT: we rely on the fact that there is always
537 * at most one process waiting for dquot to free.
538 * Otherwise dq_count would be > 1 and we would never
541 if (atomic_read(&dquot->dq_count) > 1)
543 finish_wait(&dquot->dq_wait_unused, &wait);
545 /* At this moment dquot() need not exist (it could be
546 * reclaimed by prune_dqcache(). Hence we must
551 * Quota now has no users and it has been written on last
554 remove_dquot_hash(dquot);
555 remove_free_dquot(dquot);
557 do_destroy_dquot(dquot);
559 spin_unlock(&dq_list_lock);
562 /* Call callback for every active dquot on given filesystem */
563 int dquot_scan_active(struct super_block *sb,
564 int (*fn)(struct dquot *dquot, unsigned long priv),
567 struct dquot *dquot, *old_dquot = NULL;
570 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
572 spin_lock(&dq_list_lock);
573 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
574 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
576 if (dquot->dq_sb != sb)
578 /* Now we have active dquot so we can just increase use count */
579 atomic_inc(&dquot->dq_count);
580 spin_unlock(&dq_list_lock);
581 dqstats_inc(DQST_LOOKUPS);
585 * ->release_dquot() can be racing with us. Our reference
586 * protects us from new calls to it so just wait for any
587 * outstanding call and recheck the DQ_ACTIVE_B after that.
589 wait_on_dquot(dquot);
590 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
591 ret = fn(dquot, priv);
595 spin_lock(&dq_list_lock);
596 /* We are safe to continue now because our dquot could not
597 * be moved out of the inuse list while we hold the reference */
599 spin_unlock(&dq_list_lock);
604 EXPORT_SYMBOL(dquot_scan_active);
606 /* Write all dquot structures to quota files */
607 int dquot_writeback_dquots(struct super_block *sb, int type)
609 struct list_head *dirty;
611 struct quota_info *dqopt = sb_dqopt(sb);
615 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
617 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
618 if (type != -1 && cnt != type)
620 if (!sb_has_quota_active(sb, cnt))
622 spin_lock(&dq_list_lock);
623 dirty = &dqopt->info[cnt].dqi_dirty_list;
624 while (!list_empty(dirty)) {
625 dquot = list_first_entry(dirty, struct dquot,
627 /* Dirty and inactive can be only bad dquot... */
628 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
629 clear_dquot_dirty(dquot);
632 /* Now we have active dquot from which someone is
633 * holding reference so we can safely just increase
636 spin_unlock(&dq_list_lock);
637 dqstats_inc(DQST_LOOKUPS);
638 err = sb->dq_op->write_dquot(dquot);
642 spin_lock(&dq_list_lock);
644 spin_unlock(&dq_list_lock);
647 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
648 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
649 && info_dirty(&dqopt->info[cnt]))
650 sb->dq_op->write_info(sb, cnt);
651 dqstats_inc(DQST_SYNCS);
655 EXPORT_SYMBOL(dquot_writeback_dquots);
657 /* Write all dquot structures to disk and make them visible from userspace */
658 int dquot_quota_sync(struct super_block *sb, int type)
660 struct quota_info *dqopt = sb_dqopt(sb);
664 ret = dquot_writeback_dquots(sb, type);
667 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
670 /* This is not very clever (and fast) but currently I don't know about
671 * any other simple way of getting quota data to disk and we must get
672 * them there for userspace to be visible... */
673 if (sb->s_op->sync_fs)
674 sb->s_op->sync_fs(sb, 1);
675 sync_blockdev(sb->s_bdev);
678 * Now when everything is written we can discard the pagecache so
679 * that userspace sees the changes.
681 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
682 if (type != -1 && cnt != type)
684 if (!sb_has_quota_active(sb, cnt))
686 inode_lock(dqopt->files[cnt]);
687 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
688 inode_unlock(dqopt->files[cnt]);
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 (!qid_has_mapping(sb->s_user_ns, qid))
838 return ERR_PTR(-EINVAL);
840 if (!sb_has_quota_active(sb, qid.type))
841 return ERR_PTR(-ESRCH);
843 spin_lock(&dq_list_lock);
844 spin_lock(&dq_state_lock);
845 if (!sb_has_quota_active(sb, qid.type)) {
846 spin_unlock(&dq_state_lock);
847 spin_unlock(&dq_list_lock);
848 dquot = ERR_PTR(-ESRCH);
851 spin_unlock(&dq_state_lock);
853 dquot = find_dquot(hashent, sb, qid);
856 spin_unlock(&dq_list_lock);
857 empty = get_empty_dquot(sb, qid.type);
859 schedule(); /* Try to wait for a moment... */
865 /* all dquots go on the inuse_list */
867 /* hash it first so it can be found */
868 insert_dquot_hash(dquot);
869 spin_unlock(&dq_list_lock);
870 dqstats_inc(DQST_LOOKUPS);
872 if (!atomic_read(&dquot->dq_count))
873 remove_free_dquot(dquot);
874 atomic_inc(&dquot->dq_count);
875 spin_unlock(&dq_list_lock);
876 dqstats_inc(DQST_CACHE_HITS);
877 dqstats_inc(DQST_LOOKUPS);
879 /* Wait for dq_lock - after this we know that either dquot_release() is
880 * already finished or it will be canceled due to dq_count > 1 test */
881 wait_on_dquot(dquot);
882 /* Read the dquot / allocate space in quota file */
883 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
886 err = sb->dq_op->acquire_dquot(dquot);
889 dquot = ERR_PTR(err);
894 * Make sure following reads see filled structure - paired with
895 * smp_mb__before_atomic() in dquot_acquire().
898 #ifdef CONFIG_QUOTA_DEBUG
899 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
903 do_destroy_dquot(empty);
907 EXPORT_SYMBOL(dqget);
909 static inline struct dquot **i_dquot(struct inode *inode)
911 return inode->i_sb->s_op->get_dquots(inode);
914 static int dqinit_needed(struct inode *inode, int type)
916 struct dquot * const *dquots;
919 if (IS_NOQUOTA(inode))
922 dquots = i_dquot(inode);
924 return !dquots[type];
925 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
931 /* This routine is guarded by s_umount semaphore */
932 static void add_dquot_ref(struct super_block *sb, int type)
934 struct inode *inode, *old_inode = NULL;
935 #ifdef CONFIG_QUOTA_DEBUG
939 spin_lock(&sb->s_inode_list_lock);
940 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
941 spin_lock(&inode->i_lock);
942 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
943 !atomic_read(&inode->i_writecount) ||
944 !dqinit_needed(inode, type)) {
945 spin_unlock(&inode->i_lock);
949 spin_unlock(&inode->i_lock);
950 spin_unlock(&sb->s_inode_list_lock);
952 #ifdef CONFIG_QUOTA_DEBUG
953 if (unlikely(inode_get_rsv_space(inode) > 0))
957 __dquot_initialize(inode, type);
960 * We hold a reference to 'inode' so it couldn't have been
961 * removed from s_inodes list while we dropped the
962 * s_inode_list_lock. We cannot iput the inode now as we can be
963 * holding the last reference and we cannot iput it under
964 * s_inode_list_lock. So we keep the reference and iput it
968 spin_lock(&sb->s_inode_list_lock);
970 spin_unlock(&sb->s_inode_list_lock);
973 #ifdef CONFIG_QUOTA_DEBUG
975 quota_error(sb, "Writes happened before quota was turned on "
976 "thus quota information is probably inconsistent. "
977 "Please run quotacheck(8)");
983 * Remove references to dquots from inode and add dquot to list for freeing
984 * if we have the last reference to dquot
986 static void remove_inode_dquot_ref(struct inode *inode, int type,
987 struct list_head *tofree_head)
989 struct dquot **dquots = i_dquot(inode);
990 struct dquot *dquot = dquots[type];
996 if (list_empty(&dquot->dq_free)) {
998 * The inode still has reference to dquot so it can't be in the
1001 spin_lock(&dq_list_lock);
1002 list_add(&dquot->dq_free, tofree_head);
1003 spin_unlock(&dq_list_lock);
1006 * Dquot is already in a list to put so we won't drop the last
1014 * Free list of dquots
1015 * Dquots are removed from inodes and no new references can be got so we are
1016 * the only ones holding reference
1018 static void put_dquot_list(struct list_head *tofree_head)
1020 struct list_head *act_head;
1021 struct dquot *dquot;
1023 act_head = tofree_head->next;
1024 while (act_head != tofree_head) {
1025 dquot = list_entry(act_head, struct dquot, dq_free);
1026 act_head = act_head->next;
1027 /* Remove dquot from the list so we won't have problems... */
1028 list_del_init(&dquot->dq_free);
1033 static void remove_dquot_ref(struct super_block *sb, int type,
1034 struct list_head *tofree_head)
1036 struct inode *inode;
1039 spin_lock(&sb->s_inode_list_lock);
1040 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1042 * We have to scan also I_NEW inodes because they can already
1043 * have quota pointer initialized. Luckily, we need to touch
1044 * only quota pointers and these have separate locking
1047 spin_lock(&dq_data_lock);
1048 if (!IS_NOQUOTA(inode)) {
1049 if (unlikely(inode_get_rsv_space(inode) > 0))
1051 remove_inode_dquot_ref(inode, type, tofree_head);
1053 spin_unlock(&dq_data_lock);
1055 spin_unlock(&sb->s_inode_list_lock);
1056 #ifdef CONFIG_QUOTA_DEBUG
1058 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1059 " was disabled thus quota information is probably "
1060 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1065 /* Gather all references from inodes and drop them */
1066 static void drop_dquot_ref(struct super_block *sb, int type)
1068 LIST_HEAD(tofree_head);
1071 remove_dquot_ref(sb, type, &tofree_head);
1072 synchronize_srcu(&dquot_srcu);
1073 put_dquot_list(&tofree_head);
1077 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1079 dquot->dq_dqb.dqb_curinodes += number;
1082 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1084 dquot->dq_dqb.dqb_curspace += number;
1087 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1089 dquot->dq_dqb.dqb_rsvspace += number;
1093 * Claim reserved quota space
1095 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1097 if (dquot->dq_dqb.dqb_rsvspace < number) {
1099 number = dquot->dq_dqb.dqb_rsvspace;
1101 dquot->dq_dqb.dqb_curspace += number;
1102 dquot->dq_dqb.dqb_rsvspace -= number;
1105 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1107 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1108 number = dquot->dq_dqb.dqb_curspace;
1109 dquot->dq_dqb.dqb_rsvspace += number;
1110 dquot->dq_dqb.dqb_curspace -= number;
1114 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1116 if (dquot->dq_dqb.dqb_rsvspace >= number)
1117 dquot->dq_dqb.dqb_rsvspace -= number;
1120 dquot->dq_dqb.dqb_rsvspace = 0;
1124 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1126 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1127 dquot->dq_dqb.dqb_curinodes >= number)
1128 dquot->dq_dqb.dqb_curinodes -= number;
1130 dquot->dq_dqb.dqb_curinodes = 0;
1131 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1132 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1133 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1136 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1138 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1139 dquot->dq_dqb.dqb_curspace >= number)
1140 dquot->dq_dqb.dqb_curspace -= number;
1142 dquot->dq_dqb.dqb_curspace = 0;
1143 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1144 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1145 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1149 struct super_block *w_sb;
1150 struct kqid w_dq_id;
1154 static int warning_issued(struct dquot *dquot, const int warntype)
1156 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1157 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1158 ((warntype == QUOTA_NL_IHARDWARN ||
1159 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1163 return test_and_set_bit(flag, &dquot->dq_flags);
1166 #ifdef CONFIG_PRINT_QUOTA_WARNING
1167 static int flag_print_warnings = 1;
1169 static int need_print_warning(struct dquot_warn *warn)
1171 if (!flag_print_warnings)
1174 switch (warn->w_dq_id.type) {
1176 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1178 return in_group_p(warn->w_dq_id.gid);
1185 /* Print warning to user which exceeded quota */
1186 static void print_warning(struct dquot_warn *warn)
1189 struct tty_struct *tty;
1190 int warntype = warn->w_type;
1192 if (warntype == QUOTA_NL_IHARDBELOW ||
1193 warntype == QUOTA_NL_ISOFTBELOW ||
1194 warntype == QUOTA_NL_BHARDBELOW ||
1195 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1198 tty = get_current_tty();
1201 tty_write_message(tty, warn->w_sb->s_id);
1202 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1203 tty_write_message(tty, ": warning, ");
1205 tty_write_message(tty, ": write failed, ");
1206 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1208 case QUOTA_NL_IHARDWARN:
1209 msg = " file limit reached.\r\n";
1211 case QUOTA_NL_ISOFTLONGWARN:
1212 msg = " file quota exceeded too long.\r\n";
1214 case QUOTA_NL_ISOFTWARN:
1215 msg = " file quota exceeded.\r\n";
1217 case QUOTA_NL_BHARDWARN:
1218 msg = " block limit reached.\r\n";
1220 case QUOTA_NL_BSOFTLONGWARN:
1221 msg = " block quota exceeded too long.\r\n";
1223 case QUOTA_NL_BSOFTWARN:
1224 msg = " block quota exceeded.\r\n";
1227 tty_write_message(tty, msg);
1232 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1235 if (warning_issued(dquot, warntype))
1237 warn->w_type = warntype;
1238 warn->w_sb = dquot->dq_sb;
1239 warn->w_dq_id = dquot->dq_id;
1243 * Write warnings to the console and send warning messages over netlink.
1245 * Note that this function can call into tty and networking code.
1247 static void flush_warnings(struct dquot_warn *warn)
1251 for (i = 0; i < MAXQUOTAS; i++) {
1252 if (warn[i].w_type == QUOTA_NL_NOWARN)
1254 #ifdef CONFIG_PRINT_QUOTA_WARNING
1255 print_warning(&warn[i]);
1257 quota_send_warning(warn[i].w_dq_id,
1258 warn[i].w_sb->s_dev, warn[i].w_type);
1262 static int ignore_hardlimit(struct dquot *dquot)
1264 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1266 return capable(CAP_SYS_RESOURCE) &&
1267 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1268 !(info->dqi_flags & DQF_ROOT_SQUASH));
1271 /* needs dq_data_lock */
1272 static int check_idq(struct dquot *dquot, qsize_t inodes,
1273 struct dquot_warn *warn)
1275 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1277 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1278 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1281 if (dquot->dq_dqb.dqb_ihardlimit &&
1282 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1283 !ignore_hardlimit(dquot)) {
1284 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1288 if (dquot->dq_dqb.dqb_isoftlimit &&
1289 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1290 dquot->dq_dqb.dqb_itime &&
1291 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1292 !ignore_hardlimit(dquot)) {
1293 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1297 if (dquot->dq_dqb.dqb_isoftlimit &&
1298 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1299 dquot->dq_dqb.dqb_itime == 0) {
1300 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1301 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1302 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1308 /* needs dq_data_lock */
1309 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1310 struct dquot_warn *warn)
1313 struct super_block *sb = dquot->dq_sb;
1315 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1316 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1319 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1322 if (dquot->dq_dqb.dqb_bhardlimit &&
1323 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1324 !ignore_hardlimit(dquot)) {
1326 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1330 if (dquot->dq_dqb.dqb_bsoftlimit &&
1331 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1332 dquot->dq_dqb.dqb_btime &&
1333 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1334 !ignore_hardlimit(dquot)) {
1336 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1340 if (dquot->dq_dqb.dqb_bsoftlimit &&
1341 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1342 dquot->dq_dqb.dqb_btime == 0) {
1344 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1345 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1346 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1350 * We don't allow preallocation to exceed softlimit so exceeding will
1359 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1363 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1364 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1365 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1366 return QUOTA_NL_NOWARN;
1368 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1369 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1370 return QUOTA_NL_ISOFTBELOW;
1371 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1372 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1373 return QUOTA_NL_IHARDBELOW;
1374 return QUOTA_NL_NOWARN;
1377 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1379 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1380 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1381 return QUOTA_NL_NOWARN;
1383 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1384 return QUOTA_NL_BSOFTBELOW;
1385 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1386 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1387 return QUOTA_NL_BHARDBELOW;
1388 return QUOTA_NL_NOWARN;
1391 static int dquot_active(const struct inode *inode)
1393 struct super_block *sb = inode->i_sb;
1395 if (IS_NOQUOTA(inode))
1397 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1401 * Initialize quota pointers in inode
1403 * It is better to call this function outside of any transaction as it
1404 * might need a lot of space in journal for dquot structure allocation.
1406 static int __dquot_initialize(struct inode *inode, int type)
1408 int cnt, init_needed = 0;
1409 struct dquot **dquots, *got[MAXQUOTAS] = {};
1410 struct super_block *sb = inode->i_sb;
1414 if (!dquot_active(inode))
1417 dquots = i_dquot(inode);
1419 /* First get references to structures we might need. */
1420 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1424 struct dquot *dquot;
1426 if (type != -1 && cnt != type)
1429 * The i_dquot should have been initialized in most cases,
1430 * we check it without locking here to avoid unnecessary
1431 * dqget()/dqput() calls.
1436 if (!sb_has_quota_active(sb, cnt))
1443 qid = make_kqid_uid(inode->i_uid);
1446 qid = make_kqid_gid(inode->i_gid);
1449 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1452 qid = make_kqid_projid(projid);
1455 dquot = dqget(sb, qid);
1456 if (IS_ERR(dquot)) {
1457 /* We raced with somebody turning quotas off... */
1458 if (PTR_ERR(dquot) != -ESRCH) {
1459 ret = PTR_ERR(dquot);
1467 /* All required i_dquot has been initialized */
1471 spin_lock(&dq_data_lock);
1472 if (IS_NOQUOTA(inode))
1474 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1475 if (type != -1 && cnt != type)
1477 /* Avoid races with quotaoff() */
1478 if (!sb_has_quota_active(sb, cnt))
1480 /* We could race with quotaon or dqget() could have failed */
1484 dquots[cnt] = got[cnt];
1487 * Make quota reservation system happy if someone
1488 * did a write before quota was turned on
1490 rsv = inode_get_rsv_space(inode);
1492 dquot_resv_space(dquots[cnt], rsv);
1496 spin_unlock(&dq_data_lock);
1498 /* Drop unused references */
1504 int dquot_initialize(struct inode *inode)
1506 return __dquot_initialize(inode, -1);
1508 EXPORT_SYMBOL(dquot_initialize);
1510 bool dquot_initialize_needed(struct inode *inode)
1512 struct dquot **dquots;
1515 if (!dquot_active(inode))
1518 dquots = i_dquot(inode);
1519 for (i = 0; i < MAXQUOTAS; i++)
1520 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1524 EXPORT_SYMBOL(dquot_initialize_needed);
1527 * Release all quotas referenced by inode.
1529 * This function only be called on inode free or converting
1530 * a file to quota file, no other users for the i_dquot in
1531 * both cases, so we needn't call synchronize_srcu() after
1534 static void __dquot_drop(struct inode *inode)
1537 struct dquot **dquots = i_dquot(inode);
1538 struct dquot *put[MAXQUOTAS];
1540 spin_lock(&dq_data_lock);
1541 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1542 put[cnt] = dquots[cnt];
1545 spin_unlock(&dq_data_lock);
1549 void dquot_drop(struct inode *inode)
1551 struct dquot * const *dquots;
1554 if (IS_NOQUOTA(inode))
1558 * Test before calling to rule out calls from proc and such
1559 * where we are not allowed to block. Note that this is
1560 * actually reliable test even without the lock - the caller
1561 * must assure that nobody can come after the DQUOT_DROP and
1562 * add quota pointers back anyway.
1564 dquots = i_dquot(inode);
1565 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1570 if (cnt < MAXQUOTAS)
1571 __dquot_drop(inode);
1573 EXPORT_SYMBOL(dquot_drop);
1576 * inode_reserved_space is managed internally by quota, and protected by
1577 * i_lock similar to i_blocks+i_bytes.
1579 static qsize_t *inode_reserved_space(struct inode * inode)
1581 /* Filesystem must explicitly define it's own method in order to use
1582 * quota reservation interface */
1583 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1584 return inode->i_sb->dq_op->get_reserved_space(inode);
1587 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1589 spin_lock(&inode->i_lock);
1590 *inode_reserved_space(inode) += number;
1591 spin_unlock(&inode->i_lock);
1593 EXPORT_SYMBOL(inode_add_rsv_space);
1595 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1597 spin_lock(&inode->i_lock);
1598 *inode_reserved_space(inode) -= number;
1599 __inode_add_bytes(inode, number);
1600 spin_unlock(&inode->i_lock);
1602 EXPORT_SYMBOL(inode_claim_rsv_space);
1604 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1606 spin_lock(&inode->i_lock);
1607 *inode_reserved_space(inode) += number;
1608 __inode_sub_bytes(inode, number);
1609 spin_unlock(&inode->i_lock);
1611 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1613 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1615 spin_lock(&inode->i_lock);
1616 *inode_reserved_space(inode) -= number;
1617 spin_unlock(&inode->i_lock);
1619 EXPORT_SYMBOL(inode_sub_rsv_space);
1621 static qsize_t inode_get_rsv_space(struct inode *inode)
1625 if (!inode->i_sb->dq_op->get_reserved_space)
1627 spin_lock(&inode->i_lock);
1628 ret = *inode_reserved_space(inode);
1629 spin_unlock(&inode->i_lock);
1633 static void inode_incr_space(struct inode *inode, qsize_t number,
1637 inode_add_rsv_space(inode, number);
1639 inode_add_bytes(inode, number);
1642 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1645 inode_sub_rsv_space(inode, number);
1647 inode_sub_bytes(inode, number);
1651 * This functions updates i_blocks+i_bytes fields and quota information
1652 * (together with appropriate checks).
1654 * NOTE: We absolutely rely on the fact that caller dirties the inode
1655 * (usually helpers in quotaops.h care about this) and holds a handle for
1656 * the current transaction so that dquot write and inode write go into the
1661 * This operation can block, but only after everything is updated
1663 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1665 int cnt, ret = 0, index;
1666 struct dquot_warn warn[MAXQUOTAS];
1667 int reserve = flags & DQUOT_SPACE_RESERVE;
1668 struct dquot **dquots;
1670 if (!dquot_active(inode)) {
1671 inode_incr_space(inode, number, reserve);
1675 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1676 warn[cnt].w_type = QUOTA_NL_NOWARN;
1678 dquots = i_dquot(inode);
1679 index = srcu_read_lock(&dquot_srcu);
1680 spin_lock(&dq_data_lock);
1681 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1684 ret = check_bdq(dquots[cnt], number,
1685 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1686 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1687 spin_unlock(&dq_data_lock);
1688 goto out_flush_warn;
1691 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1695 dquot_resv_space(dquots[cnt], number);
1697 dquot_incr_space(dquots[cnt], number);
1699 inode_incr_space(inode, number, reserve);
1700 spin_unlock(&dq_data_lock);
1703 goto out_flush_warn;
1704 mark_all_dquot_dirty(dquots);
1706 srcu_read_unlock(&dquot_srcu, index);
1707 flush_warnings(warn);
1711 EXPORT_SYMBOL(__dquot_alloc_space);
1714 * This operation can block, but only after everything is updated
1716 int dquot_alloc_inode(struct inode *inode)
1718 int cnt, ret = 0, index;
1719 struct dquot_warn warn[MAXQUOTAS];
1720 struct dquot * const *dquots;
1722 if (!dquot_active(inode))
1724 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1725 warn[cnt].w_type = QUOTA_NL_NOWARN;
1727 dquots = i_dquot(inode);
1728 index = srcu_read_lock(&dquot_srcu);
1729 spin_lock(&dq_data_lock);
1730 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1733 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1738 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1741 dquot_incr_inodes(dquots[cnt], 1);
1745 spin_unlock(&dq_data_lock);
1747 mark_all_dquot_dirty(dquots);
1748 srcu_read_unlock(&dquot_srcu, index);
1749 flush_warnings(warn);
1752 EXPORT_SYMBOL(dquot_alloc_inode);
1755 * Convert in-memory reserved quotas to real consumed quotas
1757 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1759 struct dquot **dquots;
1762 if (!dquot_active(inode)) {
1763 inode_claim_rsv_space(inode, number);
1767 dquots = i_dquot(inode);
1768 index = srcu_read_lock(&dquot_srcu);
1769 spin_lock(&dq_data_lock);
1770 /* Claim reserved quotas to allocated quotas */
1771 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1773 dquot_claim_reserved_space(dquots[cnt], number);
1775 /* Update inode bytes */
1776 inode_claim_rsv_space(inode, number);
1777 spin_unlock(&dq_data_lock);
1778 mark_all_dquot_dirty(dquots);
1779 srcu_read_unlock(&dquot_srcu, index);
1782 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1785 * Convert allocated space back to in-memory reserved quotas
1787 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1789 struct dquot **dquots;
1792 if (!dquot_active(inode)) {
1793 inode_reclaim_rsv_space(inode, number);
1797 dquots = i_dquot(inode);
1798 index = srcu_read_lock(&dquot_srcu);
1799 spin_lock(&dq_data_lock);
1800 /* Claim reserved quotas to allocated quotas */
1801 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1803 dquot_reclaim_reserved_space(dquots[cnt], number);
1805 /* Update inode bytes */
1806 inode_reclaim_rsv_space(inode, number);
1807 spin_unlock(&dq_data_lock);
1808 mark_all_dquot_dirty(dquots);
1809 srcu_read_unlock(&dquot_srcu, index);
1812 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1815 * This operation can block, but only after everything is updated
1817 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1820 struct dquot_warn warn[MAXQUOTAS];
1821 struct dquot **dquots;
1822 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1824 if (!dquot_active(inode)) {
1825 inode_decr_space(inode, number, reserve);
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_bdq_free(dquots[cnt], number);
1839 if (wtype != QUOTA_NL_NOWARN)
1840 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1842 dquot_free_reserved_space(dquots[cnt], number);
1844 dquot_decr_space(dquots[cnt], number);
1846 inode_decr_space(inode, number, reserve);
1847 spin_unlock(&dq_data_lock);
1851 mark_all_dquot_dirty(dquots);
1853 srcu_read_unlock(&dquot_srcu, index);
1854 flush_warnings(warn);
1856 EXPORT_SYMBOL(__dquot_free_space);
1859 * This operation can block, but only after everything is updated
1861 void dquot_free_inode(struct inode *inode)
1864 struct dquot_warn warn[MAXQUOTAS];
1865 struct dquot * const *dquots;
1868 if (!dquot_active(inode))
1871 dquots = i_dquot(inode);
1872 index = srcu_read_lock(&dquot_srcu);
1873 spin_lock(&dq_data_lock);
1874 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1877 warn[cnt].w_type = QUOTA_NL_NOWARN;
1880 wtype = info_idq_free(dquots[cnt], 1);
1881 if (wtype != QUOTA_NL_NOWARN)
1882 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1883 dquot_decr_inodes(dquots[cnt], 1);
1885 spin_unlock(&dq_data_lock);
1886 mark_all_dquot_dirty(dquots);
1887 srcu_read_unlock(&dquot_srcu, index);
1888 flush_warnings(warn);
1890 EXPORT_SYMBOL(dquot_free_inode);
1893 * Transfer the number of inode and blocks from one diskquota to an other.
1894 * On success, dquot references in transfer_to are consumed and references
1895 * to original dquots that need to be released are placed there. On failure,
1896 * references are kept untouched.
1898 * This operation can block, but only after everything is updated
1899 * A transaction must be started when entering this function.
1901 * We are holding reference on transfer_from & transfer_to, no need to
1902 * protect them by srcu_read_lock().
1904 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1906 qsize_t space, cur_space;
1907 qsize_t rsv_space = 0;
1908 qsize_t inode_usage = 1;
1909 struct dquot *transfer_from[MAXQUOTAS] = {};
1911 char is_valid[MAXQUOTAS] = {};
1912 struct dquot_warn warn_to[MAXQUOTAS];
1913 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1914 struct dquot_warn warn_from_space[MAXQUOTAS];
1916 if (IS_NOQUOTA(inode))
1919 if (inode->i_sb->dq_op->get_inode_usage) {
1920 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1925 /* Initialize the arrays */
1926 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1927 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1928 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1929 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1932 spin_lock(&dq_data_lock);
1933 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1934 spin_unlock(&dq_data_lock);
1937 cur_space = inode_get_bytes(inode);
1938 rsv_space = inode_get_rsv_space(inode);
1939 space = cur_space + rsv_space;
1940 /* Build the transfer_from list and check the limits */
1941 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1943 * Skip changes for same uid or gid or for turned off quota-type.
1945 if (!transfer_to[cnt])
1947 /* Avoid races with quotaoff() */
1948 if (!sb_has_quota_active(inode->i_sb, cnt))
1951 transfer_from[cnt] = i_dquot(inode)[cnt];
1952 ret = check_idq(transfer_to[cnt], inode_usage, &warn_to[cnt]);
1955 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1961 * Finally perform the needed transfer from transfer_from to transfer_to
1963 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1966 /* Due to IO error we might not have transfer_from[] structure */
1967 if (transfer_from[cnt]) {
1969 wtype = info_idq_free(transfer_from[cnt], inode_usage);
1970 if (wtype != QUOTA_NL_NOWARN)
1971 prepare_warning(&warn_from_inodes[cnt],
1972 transfer_from[cnt], wtype);
1973 wtype = info_bdq_free(transfer_from[cnt], space);
1974 if (wtype != QUOTA_NL_NOWARN)
1975 prepare_warning(&warn_from_space[cnt],
1976 transfer_from[cnt], wtype);
1977 dquot_decr_inodes(transfer_from[cnt], inode_usage);
1978 dquot_decr_space(transfer_from[cnt], cur_space);
1979 dquot_free_reserved_space(transfer_from[cnt],
1983 dquot_incr_inodes(transfer_to[cnt], inode_usage);
1984 dquot_incr_space(transfer_to[cnt], cur_space);
1985 dquot_resv_space(transfer_to[cnt], rsv_space);
1987 i_dquot(inode)[cnt] = transfer_to[cnt];
1989 spin_unlock(&dq_data_lock);
1991 mark_all_dquot_dirty(transfer_from);
1992 mark_all_dquot_dirty(transfer_to);
1993 flush_warnings(warn_to);
1994 flush_warnings(warn_from_inodes);
1995 flush_warnings(warn_from_space);
1996 /* Pass back references to put */
1997 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1999 transfer_to[cnt] = transfer_from[cnt];
2002 spin_unlock(&dq_data_lock);
2003 flush_warnings(warn_to);
2006 EXPORT_SYMBOL(__dquot_transfer);
2008 /* Wrapper for transferring ownership of an inode for uid/gid only
2009 * Called from FSXXX_setattr()
2011 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2013 struct dquot *transfer_to[MAXQUOTAS] = {};
2014 struct dquot *dquot;
2015 struct super_block *sb = inode->i_sb;
2018 if (!dquot_active(inode))
2021 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2022 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2023 if (IS_ERR(dquot)) {
2024 if (PTR_ERR(dquot) != -ESRCH) {
2025 ret = PTR_ERR(dquot);
2030 transfer_to[USRQUOTA] = dquot;
2032 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2033 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2034 if (IS_ERR(dquot)) {
2035 if (PTR_ERR(dquot) != -ESRCH) {
2036 ret = PTR_ERR(dquot);
2041 transfer_to[GRPQUOTA] = dquot;
2043 ret = __dquot_transfer(inode, transfer_to);
2045 dqput_all(transfer_to);
2048 EXPORT_SYMBOL(dquot_transfer);
2051 * Write info of quota file to disk
2053 int dquot_commit_info(struct super_block *sb, int type)
2055 struct quota_info *dqopt = sb_dqopt(sb);
2057 return dqopt->ops[type]->write_file_info(sb, type);
2059 EXPORT_SYMBOL(dquot_commit_info);
2061 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2063 struct quota_info *dqopt = sb_dqopt(sb);
2065 if (!sb_has_quota_active(sb, qid->type))
2067 if (!dqopt->ops[qid->type]->get_next_id)
2069 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2071 EXPORT_SYMBOL(dquot_get_next_id);
2074 * Definitions of diskquota operations.
2076 const struct dquot_operations dquot_operations = {
2077 .write_dquot = dquot_commit,
2078 .acquire_dquot = dquot_acquire,
2079 .release_dquot = dquot_release,
2080 .mark_dirty = dquot_mark_dquot_dirty,
2081 .write_info = dquot_commit_info,
2082 .alloc_dquot = dquot_alloc,
2083 .destroy_dquot = dquot_destroy,
2084 .get_next_id = dquot_get_next_id,
2086 EXPORT_SYMBOL(dquot_operations);
2089 * Generic helper for ->open on filesystems supporting disk quotas.
2091 int dquot_file_open(struct inode *inode, struct file *file)
2095 error = generic_file_open(inode, file);
2096 if (!error && (file->f_mode & FMODE_WRITE))
2097 dquot_initialize(inode);
2100 EXPORT_SYMBOL(dquot_file_open);
2103 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2105 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2108 struct quota_info *dqopt = sb_dqopt(sb);
2109 struct inode *toputinode[MAXQUOTAS];
2111 /* s_umount should be held in exclusive mode */
2112 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2113 up_read(&sb->s_umount);
2115 /* Cannot turn off usage accounting without turning off limits, or
2116 * suspend quotas and simultaneously turn quotas off. */
2117 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2118 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2119 DQUOT_USAGE_ENABLED)))
2123 * Skip everything if there's nothing to do. We have to do this because
2124 * sometimes we are called when fill_super() failed and calling
2125 * sync_fs() in such cases does no good.
2127 if (!sb_any_quota_loaded(sb))
2130 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2131 toputinode[cnt] = NULL;
2132 if (type != -1 && cnt != type)
2134 if (!sb_has_quota_loaded(sb, cnt))
2137 if (flags & DQUOT_SUSPENDED) {
2138 spin_lock(&dq_state_lock);
2140 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2141 spin_unlock(&dq_state_lock);
2143 spin_lock(&dq_state_lock);
2144 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2145 /* Turning off suspended quotas? */
2146 if (!sb_has_quota_loaded(sb, cnt) &&
2147 sb_has_quota_suspended(sb, cnt)) {
2148 dqopt->flags &= ~dquot_state_flag(
2149 DQUOT_SUSPENDED, cnt);
2150 spin_unlock(&dq_state_lock);
2151 iput(dqopt->files[cnt]);
2152 dqopt->files[cnt] = NULL;
2155 spin_unlock(&dq_state_lock);
2158 /* We still have to keep quota loaded? */
2159 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2162 /* Note: these are blocking operations */
2163 drop_dquot_ref(sb, cnt);
2164 invalidate_dquots(sb, cnt);
2166 * Now all dquots should be invalidated, all writes done so we
2167 * should be only users of the info. No locks needed.
2169 if (info_dirty(&dqopt->info[cnt]))
2170 sb->dq_op->write_info(sb, cnt);
2171 if (dqopt->ops[cnt]->free_file_info)
2172 dqopt->ops[cnt]->free_file_info(sb, cnt);
2173 put_quota_format(dqopt->info[cnt].dqi_format);
2175 toputinode[cnt] = dqopt->files[cnt];
2176 if (!sb_has_quota_loaded(sb, cnt))
2177 dqopt->files[cnt] = NULL;
2178 dqopt->info[cnt].dqi_flags = 0;
2179 dqopt->info[cnt].dqi_igrace = 0;
2180 dqopt->info[cnt].dqi_bgrace = 0;
2181 dqopt->ops[cnt] = NULL;
2184 /* Skip syncing and setting flags if quota files are hidden */
2185 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2188 /* Sync the superblock so that buffers with quota data are written to
2189 * disk (and so userspace sees correct data afterwards). */
2190 if (sb->s_op->sync_fs)
2191 sb->s_op->sync_fs(sb, 1);
2192 sync_blockdev(sb->s_bdev);
2193 /* Now the quota files are just ordinary files and we can set the
2194 * inode flags back. Moreover we discard the pagecache so that
2195 * userspace sees the writes we did bypassing the pagecache. We
2196 * must also discard the blockdev buffers so that we see the
2197 * changes done by userspace on the next quotaon() */
2198 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2199 /* This can happen when suspending quotas on remount-ro... */
2200 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2201 inode_lock(toputinode[cnt]);
2202 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2203 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2204 inode_unlock(toputinode[cnt]);
2205 mark_inode_dirty_sync(toputinode[cnt]);
2208 invalidate_bdev(sb->s_bdev);
2210 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2211 if (toputinode[cnt]) {
2212 /* On remount RO, we keep the inode pointer so that we
2213 * can reenable quota on the subsequent remount RW. We
2214 * have to check 'flags' variable and not use sb_has_
2215 * function because another quotaon / quotaoff could
2216 * change global state before we got here. We refuse
2217 * to suspend quotas when there is pending delete on
2218 * the quota file... */
2219 if (!(flags & DQUOT_SUSPENDED))
2220 iput(toputinode[cnt]);
2221 else if (!toputinode[cnt]->i_nlink)
2226 EXPORT_SYMBOL(dquot_disable);
2228 int dquot_quota_off(struct super_block *sb, int type)
2230 return dquot_disable(sb, type,
2231 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2233 EXPORT_SYMBOL(dquot_quota_off);
2236 * Turn quotas on on a device
2240 * Helper function to turn quotas on when we already have the inode of
2241 * quota file and no quota information is loaded.
2243 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2246 struct quota_format_type *fmt = find_quota_format(format_id);
2247 struct super_block *sb = inode->i_sb;
2248 struct quota_info *dqopt = sb_dqopt(sb);
2253 if (!S_ISREG(inode->i_mode)) {
2257 if (IS_RDONLY(inode)) {
2261 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2262 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2266 /* Filesystems outside of init_user_ns not yet supported */
2267 if (sb->s_user_ns != &init_user_ns) {
2271 /* Usage always has to be set... */
2272 if (!(flags & DQUOT_USAGE_ENABLED)) {
2276 if (sb_has_quota_loaded(sb, type)) {
2281 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2282 /* As we bypass the pagecache we must now flush all the
2283 * dirty data and invalidate caches so that kernel sees
2284 * changes from userspace. It is not enough to just flush
2285 * the quota file since if blocksize < pagesize, invalidation
2286 * of the cache could fail because of other unrelated dirty
2288 sync_filesystem(sb);
2289 invalidate_bdev(sb->s_bdev);
2292 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2293 /* We don't want quota and atime on quota files (deadlocks
2294 * possible) Also nobody should write to the file - we use
2295 * special IO operations which ignore the immutable bit. */
2297 inode->i_flags |= S_NOQUOTA;
2298 inode_unlock(inode);
2300 * When S_NOQUOTA is set, remove dquot references as no more
2301 * references can be added
2303 __dquot_drop(inode);
2307 dqopt->files[type] = igrab(inode);
2308 if (!dqopt->files[type])
2309 goto out_file_flags;
2311 if (!fmt->qf_ops->check_quota_file(sb, type))
2314 dqopt->ops[type] = fmt->qf_ops;
2315 dqopt->info[type].dqi_format = fmt;
2316 dqopt->info[type].dqi_fmt_id = format_id;
2317 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2318 error = dqopt->ops[type]->read_file_info(sb, type);
2321 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2322 spin_lock(&dq_data_lock);
2323 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2324 spin_unlock(&dq_data_lock);
2326 spin_lock(&dq_state_lock);
2327 dqopt->flags |= dquot_state_flag(flags, type);
2328 spin_unlock(&dq_state_lock);
2330 add_dquot_ref(sb, type);
2335 dqopt->files[type] = NULL;
2339 inode->i_flags &= ~S_NOQUOTA;
2340 inode_unlock(inode);
2342 put_quota_format(fmt);
2347 /* Reenable quotas on remount RW */
2348 int dquot_resume(struct super_block *sb, int type)
2350 struct quota_info *dqopt = sb_dqopt(sb);
2351 struct inode *inode;
2355 /* s_umount should be held in exclusive mode */
2356 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2357 up_read(&sb->s_umount);
2359 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2360 if (type != -1 && cnt != type)
2362 if (!sb_has_quota_suspended(sb, cnt))
2365 inode = dqopt->files[cnt];
2366 dqopt->files[cnt] = NULL;
2367 spin_lock(&dq_state_lock);
2368 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2369 DQUOT_LIMITS_ENABLED,
2371 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2372 spin_unlock(&dq_state_lock);
2374 flags = dquot_generic_flag(flags, cnt);
2375 ret = vfs_load_quota_inode(inode, cnt,
2376 dqopt->info[cnt].dqi_fmt_id, flags);
2382 EXPORT_SYMBOL(dquot_resume);
2384 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2385 const struct path *path)
2387 int error = security_quota_on(path->dentry);
2390 /* Quota file not on the same filesystem? */
2391 if (path->dentry->d_sb != sb)
2394 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2395 format_id, DQUOT_USAGE_ENABLED |
2396 DQUOT_LIMITS_ENABLED);
2399 EXPORT_SYMBOL(dquot_quota_on);
2402 * More powerful function for turning on quotas allowing setting
2403 * of individual quota flags
2405 int dquot_enable(struct inode *inode, int type, int format_id,
2408 struct super_block *sb = inode->i_sb;
2410 /* Just unsuspend quotas? */
2411 BUG_ON(flags & DQUOT_SUSPENDED);
2412 /* s_umount should be held in exclusive mode */
2413 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2414 up_read(&sb->s_umount);
2418 /* Just updating flags needed? */
2419 if (sb_has_quota_loaded(sb, type)) {
2420 if (flags & DQUOT_USAGE_ENABLED &&
2421 sb_has_quota_usage_enabled(sb, type))
2423 if (flags & DQUOT_LIMITS_ENABLED &&
2424 sb_has_quota_limits_enabled(sb, type))
2426 spin_lock(&dq_state_lock);
2427 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2428 spin_unlock(&dq_state_lock);
2432 return vfs_load_quota_inode(inode, type, format_id, flags);
2434 EXPORT_SYMBOL(dquot_enable);
2437 * This function is used when filesystem needs to initialize quotas
2438 * during mount time.
2440 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2441 int format_id, int type)
2443 struct dentry *dentry;
2446 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2448 return PTR_ERR(dentry);
2450 if (d_really_is_negative(dentry)) {
2455 error = security_quota_on(dentry);
2457 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2458 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2464 EXPORT_SYMBOL(dquot_quota_on_mount);
2466 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2470 struct quota_info *dqopt = sb_dqopt(sb);
2472 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2474 /* Accounting cannot be turned on while fs is mounted */
2475 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2478 for (type = 0; type < MAXQUOTAS; type++) {
2479 if (!(flags & qtype_enforce_flag(type)))
2481 /* Can't enforce without accounting */
2482 if (!sb_has_quota_usage_enabled(sb, type))
2484 ret = dquot_enable(dqopt->files[type], type,
2485 dqopt->info[type].dqi_fmt_id,
2486 DQUOT_LIMITS_ENABLED);
2492 /* Backout enforcement enablement we already did */
2493 for (type--; type >= 0; type--) {
2494 if (flags & qtype_enforce_flag(type))
2495 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2497 /* Error code translation for better compatibility with XFS */
2503 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2507 struct quota_info *dqopt = sb_dqopt(sb);
2509 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2512 * We don't support turning off accounting via quotactl. In principle
2513 * quota infrastructure can do this but filesystems don't expect
2514 * userspace to be able to do it.
2517 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2520 /* Filter out limits not enabled */
2521 for (type = 0; type < MAXQUOTAS; type++)
2522 if (!sb_has_quota_limits_enabled(sb, type))
2523 flags &= ~qtype_enforce_flag(type);
2527 for (type = 0; type < MAXQUOTAS; type++) {
2528 if (flags & qtype_enforce_flag(type)) {
2529 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2536 /* Backout enforcement disabling we already did */
2537 for (type--; type >= 0; type--) {
2538 if (flags & qtype_enforce_flag(type))
2539 dquot_enable(dqopt->files[type], type,
2540 dqopt->info[type].dqi_fmt_id,
2541 DQUOT_LIMITS_ENABLED);
2546 /* Generic routine for getting common part of quota structure */
2547 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2549 struct mem_dqblk *dm = &dquot->dq_dqb;
2551 memset(di, 0, sizeof(*di));
2552 spin_lock(&dq_data_lock);
2553 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2554 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2555 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2556 di->d_ino_softlimit = dm->dqb_isoftlimit;
2557 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2558 di->d_ino_count = dm->dqb_curinodes;
2559 di->d_spc_timer = dm->dqb_btime;
2560 di->d_ino_timer = dm->dqb_itime;
2561 spin_unlock(&dq_data_lock);
2564 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2565 struct qc_dqblk *di)
2567 struct dquot *dquot;
2569 dquot = dqget(sb, qid);
2571 return PTR_ERR(dquot);
2572 do_get_dqblk(dquot, di);
2577 EXPORT_SYMBOL(dquot_get_dqblk);
2579 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2580 struct qc_dqblk *di)
2582 struct dquot *dquot;
2585 if (!sb->dq_op->get_next_id)
2587 err = sb->dq_op->get_next_id(sb, qid);
2590 dquot = dqget(sb, *qid);
2592 return PTR_ERR(dquot);
2593 do_get_dqblk(dquot, di);
2598 EXPORT_SYMBOL(dquot_get_next_dqblk);
2600 #define VFS_QC_MASK \
2601 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2602 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2603 QC_SPC_TIMER | QC_INO_TIMER)
2605 /* Generic routine for setting common part of quota structure */
2606 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2608 struct mem_dqblk *dm = &dquot->dq_dqb;
2609 int check_blim = 0, check_ilim = 0;
2610 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2612 if (di->d_fieldmask & ~VFS_QC_MASK)
2615 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2616 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2617 ((di->d_fieldmask & QC_SPC_HARD) &&
2618 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2619 ((di->d_fieldmask & QC_INO_SOFT) &&
2620 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2621 ((di->d_fieldmask & QC_INO_HARD) &&
2622 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2625 spin_lock(&dq_data_lock);
2626 if (di->d_fieldmask & QC_SPACE) {
2627 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2629 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2632 if (di->d_fieldmask & QC_SPC_SOFT)
2633 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2634 if (di->d_fieldmask & QC_SPC_HARD)
2635 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2636 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2638 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2641 if (di->d_fieldmask & QC_INO_COUNT) {
2642 dm->dqb_curinodes = di->d_ino_count;
2644 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2647 if (di->d_fieldmask & QC_INO_SOFT)
2648 dm->dqb_isoftlimit = di->d_ino_softlimit;
2649 if (di->d_fieldmask & QC_INO_HARD)
2650 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2651 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2653 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2656 if (di->d_fieldmask & QC_SPC_TIMER) {
2657 dm->dqb_btime = di->d_spc_timer;
2659 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2662 if (di->d_fieldmask & QC_INO_TIMER) {
2663 dm->dqb_itime = di->d_ino_timer;
2665 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2669 if (!dm->dqb_bsoftlimit ||
2670 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2672 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2673 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2674 /* Set grace only if user hasn't provided his own... */
2675 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2678 if (!dm->dqb_isoftlimit ||
2679 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2681 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2682 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2683 /* Set grace only if user hasn't provided his own... */
2684 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2686 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2688 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2690 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2691 spin_unlock(&dq_data_lock);
2692 mark_dquot_dirty(dquot);
2697 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2698 struct qc_dqblk *di)
2700 struct dquot *dquot;
2703 dquot = dqget(sb, qid);
2704 if (IS_ERR(dquot)) {
2705 rc = PTR_ERR(dquot);
2708 rc = do_set_dqblk(dquot, di);
2713 EXPORT_SYMBOL(dquot_set_dqblk);
2715 /* Generic routine for getting common part of quota file information */
2716 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2718 struct mem_dqinfo *mi;
2719 struct qc_type_state *tstate;
2720 struct quota_info *dqopt = sb_dqopt(sb);
2723 memset(state, 0, sizeof(*state));
2724 for (type = 0; type < MAXQUOTAS; type++) {
2725 if (!sb_has_quota_active(sb, type))
2727 tstate = state->s_state + type;
2728 mi = sb_dqopt(sb)->info + type;
2729 tstate->flags = QCI_ACCT_ENABLED;
2730 spin_lock(&dq_data_lock);
2731 if (mi->dqi_flags & DQF_SYS_FILE)
2732 tstate->flags |= QCI_SYSFILE;
2733 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2734 tstate->flags |= QCI_ROOT_SQUASH;
2735 if (sb_has_quota_limits_enabled(sb, type))
2736 tstate->flags |= QCI_LIMITS_ENFORCED;
2737 tstate->spc_timelimit = mi->dqi_bgrace;
2738 tstate->ino_timelimit = mi->dqi_igrace;
2739 tstate->ino = dqopt->files[type]->i_ino;
2740 tstate->blocks = dqopt->files[type]->i_blocks;
2741 tstate->nextents = 1; /* We don't know... */
2742 spin_unlock(&dq_data_lock);
2746 EXPORT_SYMBOL(dquot_get_state);
2748 /* Generic routine for setting common part of quota file information */
2749 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2751 struct mem_dqinfo *mi;
2754 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2755 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2757 if (!sb_has_quota_active(sb, type))
2759 mi = sb_dqopt(sb)->info + type;
2760 if (ii->i_fieldmask & QC_FLAGS) {
2761 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2762 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2765 spin_lock(&dq_data_lock);
2766 if (ii->i_fieldmask & QC_SPC_TIMER)
2767 mi->dqi_bgrace = ii->i_spc_timelimit;
2768 if (ii->i_fieldmask & QC_INO_TIMER)
2769 mi->dqi_igrace = ii->i_ino_timelimit;
2770 if (ii->i_fieldmask & QC_FLAGS) {
2771 if (ii->i_flags & QCI_ROOT_SQUASH)
2772 mi->dqi_flags |= DQF_ROOT_SQUASH;
2774 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2776 spin_unlock(&dq_data_lock);
2777 mark_info_dirty(sb, type);
2778 /* Force write to disk */
2779 sb->dq_op->write_info(sb, type);
2782 EXPORT_SYMBOL(dquot_set_dqinfo);
2784 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2785 .quota_enable = dquot_quota_enable,
2786 .quota_disable = dquot_quota_disable,
2787 .quota_sync = dquot_quota_sync,
2788 .get_state = dquot_get_state,
2789 .set_info = dquot_set_dqinfo,
2790 .get_dqblk = dquot_get_dqblk,
2791 .get_nextdqblk = dquot_get_next_dqblk,
2792 .set_dqblk = dquot_set_dqblk
2794 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2796 static int do_proc_dqstats(struct ctl_table *table, int write,
2797 void __user *buffer, size_t *lenp, loff_t *ppos)
2799 unsigned int type = (int *)table->data - dqstats.stat;
2801 /* Update global table */
2802 dqstats.stat[type] =
2803 percpu_counter_sum_positive(&dqstats.counter[type]);
2804 return proc_dointvec(table, write, buffer, lenp, ppos);
2807 static struct ctl_table fs_dqstats_table[] = {
2809 .procname = "lookups",
2810 .data = &dqstats.stat[DQST_LOOKUPS],
2811 .maxlen = sizeof(int),
2813 .proc_handler = do_proc_dqstats,
2816 .procname = "drops",
2817 .data = &dqstats.stat[DQST_DROPS],
2818 .maxlen = sizeof(int),
2820 .proc_handler = do_proc_dqstats,
2823 .procname = "reads",
2824 .data = &dqstats.stat[DQST_READS],
2825 .maxlen = sizeof(int),
2827 .proc_handler = do_proc_dqstats,
2830 .procname = "writes",
2831 .data = &dqstats.stat[DQST_WRITES],
2832 .maxlen = sizeof(int),
2834 .proc_handler = do_proc_dqstats,
2837 .procname = "cache_hits",
2838 .data = &dqstats.stat[DQST_CACHE_HITS],
2839 .maxlen = sizeof(int),
2841 .proc_handler = do_proc_dqstats,
2844 .procname = "allocated_dquots",
2845 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2846 .maxlen = sizeof(int),
2848 .proc_handler = do_proc_dqstats,
2851 .procname = "free_dquots",
2852 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2853 .maxlen = sizeof(int),
2855 .proc_handler = do_proc_dqstats,
2858 .procname = "syncs",
2859 .data = &dqstats.stat[DQST_SYNCS],
2860 .maxlen = sizeof(int),
2862 .proc_handler = do_proc_dqstats,
2864 #ifdef CONFIG_PRINT_QUOTA_WARNING
2866 .procname = "warnings",
2867 .data = &flag_print_warnings,
2868 .maxlen = sizeof(int),
2870 .proc_handler = proc_dointvec,
2876 static struct ctl_table fs_table[] = {
2878 .procname = "quota",
2880 .child = fs_dqstats_table,
2885 static struct ctl_table sys_table[] = {
2894 static int __init dquot_init(void)
2897 unsigned long nr_hash, order;
2899 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2901 register_sysctl_table(sys_table);
2903 dquot_cachep = kmem_cache_create("dquot",
2904 sizeof(struct dquot), sizeof(unsigned long) * 4,
2905 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2906 SLAB_MEM_SPREAD|SLAB_PANIC),
2910 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2912 panic("Cannot create dquot hash table");
2914 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2915 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2917 panic("Cannot create dquot stat counters");
2920 /* Find power-of-two hlist_heads which can fit into allocation */
2921 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2925 } while (nr_hash >> dq_hash_bits);
2928 nr_hash = 1UL << dq_hash_bits;
2929 dq_hash_mask = nr_hash - 1;
2930 for (i = 0; i < nr_hash; i++)
2931 INIT_HLIST_HEAD(dquot_hash + i);
2933 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2934 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2936 register_shrinker(&dqcache_shrinker);
2940 fs_initcall(dquot_init);