1 // SPDX-License-Identifier: GPL-2.0
3 * Implementation of the diskquota system for the LINUX operating system. QUOTA
4 * is implemented using the BSD system call interface as the means of
5 * communication with the user level. This file contains the generic routines
6 * called by the different filesystems on allocation of an inode or block.
7 * These routines take care of the administration needed to have a consistent
8 * diskquota tracking system. The ideas of both user and group quotas are based
9 * on the Melbourne quota system as used on BSD derived systems. The internal
10 * implementation is based on one of the several variants of the LINUX
11 * inode-subsystem with added complexity of the diskquota system.
13 * Author: Marco van Wieringen <mvw@planets.elm.net>
15 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
17 * Revised list management to avoid races
18 * -- Bill Hawes, <whawes@star.net>, 9/98
20 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
21 * As the consequence the locking was moved from dquot_decr_...(),
22 * dquot_incr_...() to calling functions.
23 * invalidate_dquots() now writes modified dquots.
24 * Serialized quota_off() and quota_on() for mount point.
25 * Fixed a few bugs in grow_dquots().
26 * Fixed deadlock in write_dquot() - we no longer account quotas on
28 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
29 * add_dquot_ref() restarts after blocking
30 * Added check for bogus uid and fixed check for group in quotactl.
31 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
33 * Used struct list_head instead of own list struct
34 * Invalidation of referenced dquots is no longer possible
35 * Improved free_dquots list management
36 * Quota and i_blocks are now updated in one place to avoid races
37 * Warnings are now delayed so we won't block in critical section
38 * Write updated not to require dquot lock
39 * Jan Kara, <jack@suse.cz>, 9/2000
41 * Added dynamic quota structure allocation
42 * Jan Kara <jack@suse.cz> 12/2000
44 * Rewritten quota interface. Implemented new quota format and
45 * formats registering.
46 * Jan Kara, <jack@suse.cz>, 2001,2002
49 * Jan Kara, <jack@suse.cz>, 10/2002
51 * Added journalled quota support, fix lock inversion problems
52 * Jan Kara, <jack@suse.cz>, 2003,2004
54 * (C) Copyright 1994 - 1997 Marco van Wieringen
57 #include <linux/errno.h>
58 #include <linux/kernel.h>
60 #include <linux/mount.h>
62 #include <linux/time.h>
63 #include <linux/types.h>
64 #include <linux/string.h>
65 #include <linux/fcntl.h>
66 #include <linux/stat.h>
67 #include <linux/tty.h>
68 #include <linux/file.h>
69 #include <linux/slab.h>
70 #include <linux/sysctl.h>
71 #include <linux/init.h>
72 #include <linux/module.h>
73 #include <linux/proc_fs.h>
74 #include <linux/security.h>
75 #include <linux/sched.h>
76 #include <linux/cred.h>
77 #include <linux/kmod.h>
78 #include <linux/namei.h>
79 #include <linux/capability.h>
80 #include <linux/quotaops.h>
81 #include "../internal.h" /* ugh */
83 #include <linux/uaccess.h>
86 * There are five quota SMP locks:
87 * * dq_list_lock protects all lists with quotas and quota formats.
88 * * dquot->dq_dqb_lock protects data from dq_dqb
89 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
90 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
91 * dquot_transfer() can stabilize amount it transfers
92 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
93 * pointers in the inode
94 * * dq_state_lock protects modifications of quota state (on quotaon and
95 * quotaoff) and readers who care about latest values take it as well.
97 * The spinlock ordering is hence:
98 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
99 * dq_list_lock > dq_state_lock
101 * Note that some things (eg. sb pointer, type, id) doesn't change during
102 * the life of the dquot structure and so needn't to be protected by a lock
104 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
105 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
106 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
107 * inode and before dropping dquot references to avoid use of dquots after
108 * they are freed. dq_data_lock is used to serialize the pointer setting and
109 * clearing operations.
110 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
111 * inode is a quota file). Functions adding pointers from inode to dquots have
112 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
113 * have to do all pointer modifications before dropping dq_data_lock. This makes
114 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
115 * then drops all pointers to dquots from an inode.
117 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
118 * memory (or space for it is being allocated) on the first dqget(), when it is
119 * being written out, and when it is being released on the last dqput(). The
120 * allocation and release operations are serialized by the dq_lock and by
121 * checking the use count in dquot_release().
123 * Lock ordering (including related VFS locks) is the following:
124 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
128 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
129 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
130 EXPORT_SYMBOL(dq_data_lock);
131 DEFINE_STATIC_SRCU(dquot_srcu);
133 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
135 void __quota_error(struct super_block *sb, const char *func,
136 const char *fmt, ...)
138 if (printk_ratelimit()) {
140 struct va_format vaf;
147 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
148 sb->s_id, func, &vaf);
153 EXPORT_SYMBOL(__quota_error);
155 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
156 static char *quotatypes[] = INITQFNAMES;
158 static struct quota_format_type *quota_formats; /* List of registered formats */
159 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
161 /* SLAB cache for dquot structures */
162 static struct kmem_cache *dquot_cachep;
164 int register_quota_format(struct quota_format_type *fmt)
166 spin_lock(&dq_list_lock);
167 fmt->qf_next = quota_formats;
169 spin_unlock(&dq_list_lock);
172 EXPORT_SYMBOL(register_quota_format);
174 void unregister_quota_format(struct quota_format_type *fmt)
176 struct quota_format_type **actqf;
178 spin_lock(&dq_list_lock);
179 for (actqf = "a_formats; *actqf && *actqf != fmt;
180 actqf = &(*actqf)->qf_next)
183 *actqf = (*actqf)->qf_next;
184 spin_unlock(&dq_list_lock);
186 EXPORT_SYMBOL(unregister_quota_format);
188 static struct quota_format_type *find_quota_format(int id)
190 struct quota_format_type *actqf;
192 spin_lock(&dq_list_lock);
193 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
194 actqf = actqf->qf_next)
196 if (!actqf || !try_module_get(actqf->qf_owner)) {
199 spin_unlock(&dq_list_lock);
201 for (qm = 0; module_names[qm].qm_fmt_id &&
202 module_names[qm].qm_fmt_id != id; qm++)
204 if (!module_names[qm].qm_fmt_id ||
205 request_module(module_names[qm].qm_mod_name))
208 spin_lock(&dq_list_lock);
209 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
210 actqf = actqf->qf_next)
212 if (actqf && !try_module_get(actqf->qf_owner))
215 spin_unlock(&dq_list_lock);
219 static void put_quota_format(struct quota_format_type *fmt)
221 module_put(fmt->qf_owner);
225 * Dquot List Management:
226 * The quota code uses four lists for dquot management: the inuse_list,
227 * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
228 * structure may be on some of those lists, depending on its current state.
230 * All dquots are placed to the end of inuse_list when first created, and this
231 * list is used for invalidate operation, which must look at every dquot.
233 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
234 * and this list is searched whenever we need an available dquot. Dquots are
235 * removed from the list as soon as they are used again, and
236 * dqstats.free_dquots gives the number of dquots on the list. When
237 * dquot is invalidated it's completely released from memory.
239 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
240 * dirtied, and this list is searched when writing dirty dquots back to
241 * quota file. Note that some filesystems do dirty dquot tracking on their
242 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
244 * Dquots with a specific identity (device, type and id) are placed on
245 * one of the dquot_hash[] hash chains. The provides an efficient search
246 * mechanism to locate a specific dquot.
249 static LIST_HEAD(inuse_list);
250 static LIST_HEAD(free_dquots);
251 static unsigned int dq_hash_bits, dq_hash_mask;
252 static struct hlist_head *dquot_hash;
254 struct dqstats dqstats;
255 EXPORT_SYMBOL(dqstats);
257 static qsize_t inode_get_rsv_space(struct inode *inode);
258 static qsize_t __inode_get_rsv_space(struct inode *inode);
259 static int __dquot_initialize(struct inode *inode, int type);
261 static inline unsigned int
262 hashfn(const struct super_block *sb, struct kqid qid)
264 unsigned int id = from_kqid(&init_user_ns, qid);
268 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
269 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
273 * Following list functions expect dq_list_lock to be held
275 static inline void insert_dquot_hash(struct dquot *dquot)
277 struct hlist_head *head;
278 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
279 hlist_add_head(&dquot->dq_hash, head);
282 static inline void remove_dquot_hash(struct dquot *dquot)
284 hlist_del_init(&dquot->dq_hash);
287 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
290 struct hlist_node *node;
293 hlist_for_each (node, dquot_hash+hashent) {
294 dquot = hlist_entry(node, struct dquot, dq_hash);
295 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
301 /* Add a dquot to the tail of the free list */
302 static inline void put_dquot_last(struct dquot *dquot)
304 list_add_tail(&dquot->dq_free, &free_dquots);
305 dqstats_inc(DQST_FREE_DQUOTS);
308 static inline void remove_free_dquot(struct dquot *dquot)
310 if (list_empty(&dquot->dq_free))
312 list_del_init(&dquot->dq_free);
313 dqstats_dec(DQST_FREE_DQUOTS);
316 static inline void put_inuse(struct dquot *dquot)
318 /* We add to the back of inuse list so we don't have to restart
319 * when traversing this list and we block */
320 list_add_tail(&dquot->dq_inuse, &inuse_list);
321 dqstats_inc(DQST_ALLOC_DQUOTS);
324 static inline void remove_inuse(struct dquot *dquot)
326 dqstats_dec(DQST_ALLOC_DQUOTS);
327 list_del(&dquot->dq_inuse);
330 * End of list functions needing dq_list_lock
333 static void wait_on_dquot(struct dquot *dquot)
335 mutex_lock(&dquot->dq_lock);
336 mutex_unlock(&dquot->dq_lock);
339 static inline int dquot_dirty(struct dquot *dquot)
341 return test_bit(DQ_MOD_B, &dquot->dq_flags);
344 static inline int mark_dquot_dirty(struct dquot *dquot)
346 return dquot->dq_sb->dq_op->mark_dirty(dquot);
349 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
350 int dquot_mark_dquot_dirty(struct dquot *dquot)
354 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
357 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
358 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
360 /* If quota is dirty already, we don't have to acquire dq_list_lock */
361 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
364 spin_lock(&dq_list_lock);
365 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
366 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
367 info[dquot->dq_id.type].dqi_dirty_list);
370 spin_unlock(&dq_list_lock);
373 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
375 /* Dirtify all the dquots - this can block when journalling */
376 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
381 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
383 /* Even in case of error we have to continue */
384 ret = mark_dquot_dirty(dquot[cnt]);
391 static inline void dqput_all(struct dquot **dquot)
395 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
399 static inline int clear_dquot_dirty(struct dquot *dquot)
401 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
402 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
404 spin_lock(&dq_list_lock);
405 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
406 spin_unlock(&dq_list_lock);
409 list_del_init(&dquot->dq_dirty);
410 spin_unlock(&dq_list_lock);
414 void mark_info_dirty(struct super_block *sb, int type)
416 spin_lock(&dq_data_lock);
417 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
418 spin_unlock(&dq_data_lock);
420 EXPORT_SYMBOL(mark_info_dirty);
423 * Read dquot from disk and alloc space for it
426 int dquot_acquire(struct dquot *dquot)
428 int ret = 0, ret2 = 0;
429 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
431 mutex_lock(&dquot->dq_lock);
432 if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
433 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
437 /* Make sure flags update is visible after dquot has been filled */
438 smp_mb__before_atomic();
439 set_bit(DQ_READ_B, &dquot->dq_flags);
440 /* Instantiate dquot if needed */
441 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
442 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
443 /* Write the info if needed */
444 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
445 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
446 dquot->dq_sb, dquot->dq_id.type);
456 * Make sure flags update is visible after on-disk struct has been
457 * allocated. Paired with smp_rmb() in dqget().
459 smp_mb__before_atomic();
460 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
462 mutex_unlock(&dquot->dq_lock);
465 EXPORT_SYMBOL(dquot_acquire);
468 * Write dquot to disk
470 int dquot_commit(struct dquot *dquot)
473 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
475 mutex_lock(&dquot->dq_lock);
476 if (!clear_dquot_dirty(dquot))
478 /* Inactive dquot can be only if there was error during read/init
479 * => we have better not writing it */
480 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
481 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
485 mutex_unlock(&dquot->dq_lock);
488 EXPORT_SYMBOL(dquot_commit);
493 int dquot_release(struct dquot *dquot)
495 int ret = 0, ret2 = 0;
496 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
498 mutex_lock(&dquot->dq_lock);
499 /* Check whether we are not racing with some other dqget() */
500 if (dquot_is_busy(dquot))
502 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
503 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
505 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
506 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
507 dquot->dq_sb, dquot->dq_id.type);
512 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
514 mutex_unlock(&dquot->dq_lock);
517 EXPORT_SYMBOL(dquot_release);
519 void dquot_destroy(struct dquot *dquot)
521 kmem_cache_free(dquot_cachep, dquot);
523 EXPORT_SYMBOL(dquot_destroy);
525 static inline void do_destroy_dquot(struct dquot *dquot)
527 dquot->dq_sb->dq_op->destroy_dquot(dquot);
530 /* Invalidate all dquots on the list. Note that this function is called after
531 * quota is disabled and pointers from inodes removed so there cannot be new
532 * quota users. There can still be some users of quotas due to inodes being
533 * just deleted or pruned by prune_icache() (those are not attached to any
534 * list) or parallel quotactl call. We have to wait for such users.
536 static void invalidate_dquots(struct super_block *sb, int type)
538 struct dquot *dquot, *tmp;
541 spin_lock(&dq_list_lock);
542 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
543 if (dquot->dq_sb != sb)
545 if (dquot->dq_id.type != type)
547 /* Wait for dquot users */
548 if (atomic_read(&dquot->dq_count)) {
550 spin_unlock(&dq_list_lock);
552 * Once dqput() wakes us up, we know it's time to free
554 * IMPORTANT: we rely on the fact that there is always
555 * at most one process waiting for dquot to free.
556 * Otherwise dq_count would be > 1 and we would never
559 wait_event(dquot_ref_wq,
560 atomic_read(&dquot->dq_count) == 1);
562 /* At this moment dquot() need not exist (it could be
563 * reclaimed by prune_dqcache(). Hence we must
568 * Quota now has no users and it has been written on last
571 remove_dquot_hash(dquot);
572 remove_free_dquot(dquot);
574 do_destroy_dquot(dquot);
576 spin_unlock(&dq_list_lock);
579 /* Call callback for every active dquot on given filesystem */
580 int dquot_scan_active(struct super_block *sb,
581 int (*fn)(struct dquot *dquot, unsigned long priv),
584 struct dquot *dquot, *old_dquot = NULL;
587 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
589 spin_lock(&dq_list_lock);
590 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
591 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
593 if (dquot->dq_sb != sb)
595 /* Now we have active dquot so we can just increase use count */
596 atomic_inc(&dquot->dq_count);
597 spin_unlock(&dq_list_lock);
601 * ->release_dquot() can be racing with us. Our reference
602 * protects us from new calls to it so just wait for any
603 * outstanding call and recheck the DQ_ACTIVE_B after that.
605 wait_on_dquot(dquot);
606 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
607 ret = fn(dquot, priv);
611 spin_lock(&dq_list_lock);
612 /* We are safe to continue now because our dquot could not
613 * be moved out of the inuse list while we hold the reference */
615 spin_unlock(&dq_list_lock);
620 EXPORT_SYMBOL(dquot_scan_active);
622 /* Write all dquot structures to quota files */
623 int dquot_writeback_dquots(struct super_block *sb, int type)
625 struct list_head dirty;
627 struct quota_info *dqopt = sb_dqopt(sb);
631 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
633 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
634 if (type != -1 && cnt != type)
636 if (!sb_has_quota_active(sb, cnt))
638 spin_lock(&dq_list_lock);
639 /* Move list away to avoid livelock. */
640 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
641 while (!list_empty(&dirty)) {
642 dquot = list_first_entry(&dirty, struct dquot,
645 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
647 /* Now we have active dquot from which someone is
648 * holding reference so we can safely just increase
651 spin_unlock(&dq_list_lock);
652 err = sb->dq_op->write_dquot(dquot);
655 * Clear dirty bit anyway to avoid infinite
658 clear_dquot_dirty(dquot);
663 spin_lock(&dq_list_lock);
665 spin_unlock(&dq_list_lock);
668 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
669 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
670 && info_dirty(&dqopt->info[cnt]))
671 sb->dq_op->write_info(sb, cnt);
672 dqstats_inc(DQST_SYNCS);
676 EXPORT_SYMBOL(dquot_writeback_dquots);
678 /* Write all dquot structures to disk and make them visible from userspace */
679 int dquot_quota_sync(struct super_block *sb, int type)
681 struct quota_info *dqopt = sb_dqopt(sb);
685 ret = dquot_writeback_dquots(sb, type);
688 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
691 /* This is not very clever (and fast) but currently I don't know about
692 * any other simple way of getting quota data to disk and we must get
693 * them there for userspace to be visible... */
694 if (sb->s_op->sync_fs)
695 sb->s_op->sync_fs(sb, 1);
696 sync_blockdev(sb->s_bdev);
699 * Now when everything is written we can discard the pagecache so
700 * that userspace sees the changes.
702 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
703 if (type != -1 && cnt != type)
705 if (!sb_has_quota_active(sb, cnt))
707 inode_lock(dqopt->files[cnt]);
708 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
709 inode_unlock(dqopt->files[cnt]);
714 EXPORT_SYMBOL(dquot_quota_sync);
717 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
720 unsigned long freed = 0;
722 spin_lock(&dq_list_lock);
723 while (!list_empty(&free_dquots) && sc->nr_to_scan) {
724 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
725 remove_dquot_hash(dquot);
726 remove_free_dquot(dquot);
728 do_destroy_dquot(dquot);
732 spin_unlock(&dq_list_lock);
737 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
739 return vfs_pressure_ratio(
740 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
743 static struct shrinker dqcache_shrinker = {
744 .count_objects = dqcache_shrink_count,
745 .scan_objects = dqcache_shrink_scan,
746 .seeks = DEFAULT_SEEKS,
750 * Put reference to dquot
752 void dqput(struct dquot *dquot)
758 #ifdef CONFIG_QUOTA_DEBUG
759 if (!atomic_read(&dquot->dq_count)) {
760 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
761 quotatypes[dquot->dq_id.type],
762 from_kqid(&init_user_ns, dquot->dq_id));
766 dqstats_inc(DQST_DROPS);
768 spin_lock(&dq_list_lock);
769 if (atomic_read(&dquot->dq_count) > 1) {
770 /* We have more than one user... nothing to do */
771 atomic_dec(&dquot->dq_count);
772 /* Releasing dquot during quotaoff phase? */
773 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
774 atomic_read(&dquot->dq_count) == 1)
775 wake_up(&dquot_ref_wq);
776 spin_unlock(&dq_list_lock);
779 /* Need to release dquot? */
780 if (dquot_dirty(dquot)) {
781 spin_unlock(&dq_list_lock);
782 /* Commit dquot before releasing */
783 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
785 quota_error(dquot->dq_sb, "Can't write quota structure"
786 " (error %d). Quota may get out of sync!",
789 * We clear dirty bit anyway, so that we avoid
792 clear_dquot_dirty(dquot);
796 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
797 spin_unlock(&dq_list_lock);
798 dquot->dq_sb->dq_op->release_dquot(dquot);
801 atomic_dec(&dquot->dq_count);
802 #ifdef CONFIG_QUOTA_DEBUG
804 BUG_ON(!list_empty(&dquot->dq_free));
806 put_dquot_last(dquot);
807 spin_unlock(&dq_list_lock);
809 EXPORT_SYMBOL(dqput);
811 struct dquot *dquot_alloc(struct super_block *sb, int type)
813 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
815 EXPORT_SYMBOL(dquot_alloc);
817 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
821 dquot = sb->dq_op->alloc_dquot(sb, type);
825 mutex_init(&dquot->dq_lock);
826 INIT_LIST_HEAD(&dquot->dq_free);
827 INIT_LIST_HEAD(&dquot->dq_inuse);
828 INIT_HLIST_NODE(&dquot->dq_hash);
829 INIT_LIST_HEAD(&dquot->dq_dirty);
831 dquot->dq_id = make_kqid_invalid(type);
832 atomic_set(&dquot->dq_count, 1);
833 spin_lock_init(&dquot->dq_dqb_lock);
839 * Get reference to dquot
841 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
842 * destroying our dquot by:
843 * a) checking for quota flags under dq_list_lock and
844 * b) getting a reference to dquot before we release dq_list_lock
846 struct dquot *dqget(struct super_block *sb, struct kqid qid)
848 unsigned int hashent = hashfn(sb, qid);
849 struct dquot *dquot, *empty = NULL;
851 if (!qid_has_mapping(sb->s_user_ns, qid))
852 return ERR_PTR(-EINVAL);
854 if (!sb_has_quota_active(sb, qid.type))
855 return ERR_PTR(-ESRCH);
857 spin_lock(&dq_list_lock);
858 spin_lock(&dq_state_lock);
859 if (!sb_has_quota_active(sb, qid.type)) {
860 spin_unlock(&dq_state_lock);
861 spin_unlock(&dq_list_lock);
862 dquot = ERR_PTR(-ESRCH);
865 spin_unlock(&dq_state_lock);
867 dquot = find_dquot(hashent, sb, qid);
870 spin_unlock(&dq_list_lock);
871 empty = get_empty_dquot(sb, qid.type);
873 schedule(); /* Try to wait for a moment... */
879 /* all dquots go on the inuse_list */
881 /* hash it first so it can be found */
882 insert_dquot_hash(dquot);
883 spin_unlock(&dq_list_lock);
884 dqstats_inc(DQST_LOOKUPS);
886 if (!atomic_read(&dquot->dq_count))
887 remove_free_dquot(dquot);
888 atomic_inc(&dquot->dq_count);
889 spin_unlock(&dq_list_lock);
890 dqstats_inc(DQST_CACHE_HITS);
891 dqstats_inc(DQST_LOOKUPS);
893 /* Wait for dq_lock - after this we know that either dquot_release() is
894 * already finished or it will be canceled due to dq_count > 1 test */
895 wait_on_dquot(dquot);
896 /* Read the dquot / allocate space in quota file */
897 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
900 err = sb->dq_op->acquire_dquot(dquot);
903 dquot = ERR_PTR(err);
908 * Make sure following reads see filled structure - paired with
909 * smp_mb__before_atomic() in dquot_acquire().
912 #ifdef CONFIG_QUOTA_DEBUG
913 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
917 do_destroy_dquot(empty);
921 EXPORT_SYMBOL(dqget);
923 static inline struct dquot **i_dquot(struct inode *inode)
925 return inode->i_sb->s_op->get_dquots(inode);
928 static int dqinit_needed(struct inode *inode, int type)
930 struct dquot * const *dquots;
933 if (IS_NOQUOTA(inode))
936 dquots = i_dquot(inode);
938 return !dquots[type];
939 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
945 /* This routine is guarded by s_umount semaphore */
946 static int add_dquot_ref(struct super_block *sb, int type)
948 struct inode *inode, *old_inode = NULL;
949 #ifdef CONFIG_QUOTA_DEBUG
954 spin_lock(&sb->s_inode_list_lock);
955 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
956 spin_lock(&inode->i_lock);
957 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
958 !atomic_read(&inode->i_writecount) ||
959 !dqinit_needed(inode, type)) {
960 spin_unlock(&inode->i_lock);
964 spin_unlock(&inode->i_lock);
965 spin_unlock(&sb->s_inode_list_lock);
967 #ifdef CONFIG_QUOTA_DEBUG
968 if (unlikely(inode_get_rsv_space(inode) > 0))
972 err = __dquot_initialize(inode, type);
979 * We hold a reference to 'inode' so it couldn't have been
980 * removed from s_inodes list while we dropped the
981 * s_inode_list_lock. We cannot iput the inode now as we can be
982 * holding the last reference and we cannot iput it under
983 * s_inode_list_lock. So we keep the reference and iput it
987 spin_lock(&sb->s_inode_list_lock);
989 spin_unlock(&sb->s_inode_list_lock);
992 #ifdef CONFIG_QUOTA_DEBUG
994 quota_error(sb, "Writes happened before quota was turned on "
995 "thus quota information is probably inconsistent. "
996 "Please run quotacheck(8)");
1003 * Remove references to dquots from inode and add dquot to list for freeing
1004 * if we have the last reference to dquot
1006 static void remove_inode_dquot_ref(struct inode *inode, int type,
1007 struct list_head *tofree_head)
1009 struct dquot **dquots = i_dquot(inode);
1010 struct dquot *dquot = dquots[type];
1015 dquots[type] = NULL;
1016 if (list_empty(&dquot->dq_free)) {
1018 * The inode still has reference to dquot so it can't be in the
1021 spin_lock(&dq_list_lock);
1022 list_add(&dquot->dq_free, tofree_head);
1023 spin_unlock(&dq_list_lock);
1026 * Dquot is already in a list to put so we won't drop the last
1034 * Free list of dquots
1035 * Dquots are removed from inodes and no new references can be got so we are
1036 * the only ones holding reference
1038 static void put_dquot_list(struct list_head *tofree_head)
1040 struct list_head *act_head;
1041 struct dquot *dquot;
1043 act_head = tofree_head->next;
1044 while (act_head != tofree_head) {
1045 dquot = list_entry(act_head, struct dquot, dq_free);
1046 act_head = act_head->next;
1047 /* Remove dquot from the list so we won't have problems... */
1048 list_del_init(&dquot->dq_free);
1053 static void remove_dquot_ref(struct super_block *sb, int type,
1054 struct list_head *tofree_head)
1056 struct inode *inode;
1057 #ifdef CONFIG_QUOTA_DEBUG
1061 spin_lock(&sb->s_inode_list_lock);
1062 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1064 * We have to scan also I_NEW inodes because they can already
1065 * have quota pointer initialized. Luckily, we need to touch
1066 * only quota pointers and these have separate locking
1069 spin_lock(&dq_data_lock);
1070 if (!IS_NOQUOTA(inode)) {
1071 #ifdef CONFIG_QUOTA_DEBUG
1072 if (unlikely(inode_get_rsv_space(inode) > 0))
1075 remove_inode_dquot_ref(inode, type, tofree_head);
1077 spin_unlock(&dq_data_lock);
1079 spin_unlock(&sb->s_inode_list_lock);
1080 #ifdef CONFIG_QUOTA_DEBUG
1082 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1083 " was disabled thus quota information is probably "
1084 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1089 /* Gather all references from inodes and drop them */
1090 static void drop_dquot_ref(struct super_block *sb, int type)
1092 LIST_HEAD(tofree_head);
1095 remove_dquot_ref(sb, type, &tofree_head);
1096 synchronize_srcu(&dquot_srcu);
1097 put_dquot_list(&tofree_head);
1102 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1104 if (dquot->dq_dqb.dqb_rsvspace >= number)
1105 dquot->dq_dqb.dqb_rsvspace -= number;
1108 dquot->dq_dqb.dqb_rsvspace = 0;
1110 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1111 dquot->dq_dqb.dqb_bsoftlimit)
1112 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1113 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
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 = (time64_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_rsvspace <=
1136 dquot->dq_dqb.dqb_bsoftlimit)
1137 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1138 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1142 struct super_block *w_sb;
1143 struct kqid w_dq_id;
1147 static int warning_issued(struct dquot *dquot, const int warntype)
1149 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1150 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1151 ((warntype == QUOTA_NL_IHARDWARN ||
1152 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1156 return test_and_set_bit(flag, &dquot->dq_flags);
1159 #ifdef CONFIG_PRINT_QUOTA_WARNING
1160 static int flag_print_warnings = 1;
1162 static int need_print_warning(struct dquot_warn *warn)
1164 if (!flag_print_warnings)
1167 switch (warn->w_dq_id.type) {
1169 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1171 return in_group_p(warn->w_dq_id.gid);
1178 /* Print warning to user which exceeded quota */
1179 static void print_warning(struct dquot_warn *warn)
1182 struct tty_struct *tty;
1183 int warntype = warn->w_type;
1185 if (warntype == QUOTA_NL_IHARDBELOW ||
1186 warntype == QUOTA_NL_ISOFTBELOW ||
1187 warntype == QUOTA_NL_BHARDBELOW ||
1188 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1191 tty = get_current_tty();
1194 tty_write_message(tty, warn->w_sb->s_id);
1195 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1196 tty_write_message(tty, ": warning, ");
1198 tty_write_message(tty, ": write failed, ");
1199 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1201 case QUOTA_NL_IHARDWARN:
1202 msg = " file limit reached.\r\n";
1204 case QUOTA_NL_ISOFTLONGWARN:
1205 msg = " file quota exceeded too long.\r\n";
1207 case QUOTA_NL_ISOFTWARN:
1208 msg = " file quota exceeded.\r\n";
1210 case QUOTA_NL_BHARDWARN:
1211 msg = " block limit reached.\r\n";
1213 case QUOTA_NL_BSOFTLONGWARN:
1214 msg = " block quota exceeded too long.\r\n";
1216 case QUOTA_NL_BSOFTWARN:
1217 msg = " block quota exceeded.\r\n";
1220 tty_write_message(tty, msg);
1225 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1228 if (warning_issued(dquot, warntype))
1230 warn->w_type = warntype;
1231 warn->w_sb = dquot->dq_sb;
1232 warn->w_dq_id = dquot->dq_id;
1236 * Write warnings to the console and send warning messages over netlink.
1238 * Note that this function can call into tty and networking code.
1240 static void flush_warnings(struct dquot_warn *warn)
1244 for (i = 0; i < MAXQUOTAS; i++) {
1245 if (warn[i].w_type == QUOTA_NL_NOWARN)
1247 #ifdef CONFIG_PRINT_QUOTA_WARNING
1248 print_warning(&warn[i]);
1250 quota_send_warning(warn[i].w_dq_id,
1251 warn[i].w_sb->s_dev, warn[i].w_type);
1255 static int ignore_hardlimit(struct dquot *dquot)
1257 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1259 return capable(CAP_SYS_RESOURCE) &&
1260 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1261 !(info->dqi_flags & DQF_ROOT_SQUASH));
1264 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1265 struct dquot_warn *warn)
1270 spin_lock(&dquot->dq_dqb_lock);
1271 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1272 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1273 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1276 if (dquot->dq_dqb.dqb_ihardlimit &&
1277 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1278 !ignore_hardlimit(dquot)) {
1279 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1284 if (dquot->dq_dqb.dqb_isoftlimit &&
1285 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1286 dquot->dq_dqb.dqb_itime &&
1287 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1288 !ignore_hardlimit(dquot)) {
1289 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1294 if (dquot->dq_dqb.dqb_isoftlimit &&
1295 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1296 dquot->dq_dqb.dqb_itime == 0) {
1297 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1298 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1299 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1302 dquot->dq_dqb.dqb_curinodes = newinodes;
1305 spin_unlock(&dquot->dq_dqb_lock);
1309 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1310 qsize_t rsv_space, unsigned int flags,
1311 struct dquot_warn *warn)
1314 struct super_block *sb = dquot->dq_sb;
1317 spin_lock(&dquot->dq_dqb_lock);
1318 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1319 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1322 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1323 + space + rsv_space;
1325 if (dquot->dq_dqb.dqb_bhardlimit &&
1326 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1327 !ignore_hardlimit(dquot)) {
1328 if (flags & DQUOT_SPACE_WARN)
1329 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1334 if (dquot->dq_dqb.dqb_bsoftlimit &&
1335 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1336 dquot->dq_dqb.dqb_btime &&
1337 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1338 !ignore_hardlimit(dquot)) {
1339 if (flags & DQUOT_SPACE_WARN)
1340 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1345 if (dquot->dq_dqb.dqb_bsoftlimit &&
1346 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1347 dquot->dq_dqb.dqb_btime == 0) {
1348 if (flags & DQUOT_SPACE_WARN) {
1349 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1350 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1351 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1354 * We don't allow preallocation to exceed softlimit so exceeding will
1363 * We have to be careful and go through warning generation & grace time
1364 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1367 if (flags & DQUOT_SPACE_NOFAIL)
1370 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1371 dquot->dq_dqb.dqb_curspace += space;
1373 spin_unlock(&dquot->dq_dqb_lock);
1377 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1381 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1382 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1383 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1384 return QUOTA_NL_NOWARN;
1386 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1387 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1388 return QUOTA_NL_ISOFTBELOW;
1389 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1390 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1391 return QUOTA_NL_IHARDBELOW;
1392 return QUOTA_NL_NOWARN;
1395 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1399 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1401 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1402 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1403 return QUOTA_NL_NOWARN;
1405 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1406 return QUOTA_NL_BSOFTBELOW;
1407 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1408 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1409 return QUOTA_NL_BHARDBELOW;
1410 return QUOTA_NL_NOWARN;
1413 static int dquot_active(const struct inode *inode)
1415 struct super_block *sb = inode->i_sb;
1417 if (IS_NOQUOTA(inode))
1419 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1423 * Initialize quota pointers in inode
1425 * It is better to call this function outside of any transaction as it
1426 * might need a lot of space in journal for dquot structure allocation.
1428 static int __dquot_initialize(struct inode *inode, int type)
1430 int cnt, init_needed = 0;
1431 struct dquot **dquots, *got[MAXQUOTAS] = {};
1432 struct super_block *sb = inode->i_sb;
1436 if (!dquot_active(inode))
1439 dquots = i_dquot(inode);
1441 /* First get references to structures we might need. */
1442 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1446 struct dquot *dquot;
1448 if (type != -1 && cnt != type)
1451 * The i_dquot should have been initialized in most cases,
1452 * we check it without locking here to avoid unnecessary
1453 * dqget()/dqput() calls.
1458 if (!sb_has_quota_active(sb, cnt))
1465 qid = make_kqid_uid(inode->i_uid);
1468 qid = make_kqid_gid(inode->i_gid);
1471 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1474 qid = make_kqid_projid(projid);
1477 dquot = dqget(sb, qid);
1478 if (IS_ERR(dquot)) {
1479 /* We raced with somebody turning quotas off... */
1480 if (PTR_ERR(dquot) != -ESRCH) {
1481 ret = PTR_ERR(dquot);
1489 /* All required i_dquot has been initialized */
1493 spin_lock(&dq_data_lock);
1494 if (IS_NOQUOTA(inode))
1496 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1497 if (type != -1 && cnt != type)
1499 /* Avoid races with quotaoff() */
1500 if (!sb_has_quota_active(sb, cnt))
1502 /* We could race with quotaon or dqget() could have failed */
1506 dquots[cnt] = got[cnt];
1509 * Make quota reservation system happy if someone
1510 * did a write before quota was turned on
1512 rsv = inode_get_rsv_space(inode);
1513 if (unlikely(rsv)) {
1514 spin_lock(&inode->i_lock);
1515 /* Get reservation again under proper lock */
1516 rsv = __inode_get_rsv_space(inode);
1517 spin_lock(&dquots[cnt]->dq_dqb_lock);
1518 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1519 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1520 spin_unlock(&inode->i_lock);
1525 spin_unlock(&dq_data_lock);
1527 /* Drop unused references */
1533 int dquot_initialize(struct inode *inode)
1535 return __dquot_initialize(inode, -1);
1537 EXPORT_SYMBOL(dquot_initialize);
1539 bool dquot_initialize_needed(struct inode *inode)
1541 struct dquot **dquots;
1544 if (!dquot_active(inode))
1547 dquots = i_dquot(inode);
1548 for (i = 0; i < MAXQUOTAS; i++)
1549 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1553 EXPORT_SYMBOL(dquot_initialize_needed);
1556 * Release all quotas referenced by inode.
1558 * This function only be called on inode free or converting
1559 * a file to quota file, no other users for the i_dquot in
1560 * both cases, so we needn't call synchronize_srcu() after
1563 static void __dquot_drop(struct inode *inode)
1566 struct dquot **dquots = i_dquot(inode);
1567 struct dquot *put[MAXQUOTAS];
1569 spin_lock(&dq_data_lock);
1570 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1571 put[cnt] = dquots[cnt];
1574 spin_unlock(&dq_data_lock);
1578 void dquot_drop(struct inode *inode)
1580 struct dquot * const *dquots;
1583 if (IS_NOQUOTA(inode))
1587 * Test before calling to rule out calls from proc and such
1588 * where we are not allowed to block. Note that this is
1589 * actually reliable test even without the lock - the caller
1590 * must assure that nobody can come after the DQUOT_DROP and
1591 * add quota pointers back anyway.
1593 dquots = i_dquot(inode);
1594 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1599 if (cnt < MAXQUOTAS)
1600 __dquot_drop(inode);
1602 EXPORT_SYMBOL(dquot_drop);
1605 * inode_reserved_space is managed internally by quota, and protected by
1606 * i_lock similar to i_blocks+i_bytes.
1608 static qsize_t *inode_reserved_space(struct inode * inode)
1610 /* Filesystem must explicitly define it's own method in order to use
1611 * quota reservation interface */
1612 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1613 return inode->i_sb->dq_op->get_reserved_space(inode);
1616 static qsize_t __inode_get_rsv_space(struct inode *inode)
1618 if (!inode->i_sb->dq_op->get_reserved_space)
1620 return *inode_reserved_space(inode);
1623 static qsize_t inode_get_rsv_space(struct inode *inode)
1627 if (!inode->i_sb->dq_op->get_reserved_space)
1629 spin_lock(&inode->i_lock);
1630 ret = __inode_get_rsv_space(inode);
1631 spin_unlock(&inode->i_lock);
1636 * This functions updates i_blocks+i_bytes fields and quota information
1637 * (together with appropriate checks).
1639 * NOTE: We absolutely rely on the fact that caller dirties the inode
1640 * (usually helpers in quotaops.h care about this) and holds a handle for
1641 * the current transaction so that dquot write and inode write go into the
1646 * This operation can block, but only after everything is updated
1648 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1650 int cnt, ret = 0, index;
1651 struct dquot_warn warn[MAXQUOTAS];
1652 int reserve = flags & DQUOT_SPACE_RESERVE;
1653 struct dquot **dquots;
1655 if (!dquot_active(inode)) {
1657 spin_lock(&inode->i_lock);
1658 *inode_reserved_space(inode) += number;
1659 spin_unlock(&inode->i_lock);
1661 inode_add_bytes(inode, number);
1666 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1667 warn[cnt].w_type = QUOTA_NL_NOWARN;
1669 dquots = i_dquot(inode);
1670 index = srcu_read_lock(&dquot_srcu);
1671 spin_lock(&inode->i_lock);
1672 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1676 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1679 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1683 /* Back out changes we already did */
1684 for (cnt--; cnt >= 0; cnt--) {
1687 spin_lock(&dquots[cnt]->dq_dqb_lock);
1689 dquot_free_reserved_space(dquots[cnt],
1692 dquot_decr_space(dquots[cnt], number);
1693 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1695 spin_unlock(&inode->i_lock);
1696 goto out_flush_warn;
1700 *inode_reserved_space(inode) += number;
1702 __inode_add_bytes(inode, number);
1703 spin_unlock(&inode->i_lock);
1706 goto out_flush_warn;
1707 mark_all_dquot_dirty(dquots);
1709 srcu_read_unlock(&dquot_srcu, index);
1710 flush_warnings(warn);
1714 EXPORT_SYMBOL(__dquot_alloc_space);
1717 * This operation can block, but only after everything is updated
1719 int dquot_alloc_inode(struct inode *inode)
1721 int cnt, ret = 0, index;
1722 struct dquot_warn warn[MAXQUOTAS];
1723 struct dquot * const *dquots;
1725 if (!dquot_active(inode))
1727 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1728 warn[cnt].w_type = QUOTA_NL_NOWARN;
1730 dquots = i_dquot(inode);
1731 index = srcu_read_lock(&dquot_srcu);
1732 spin_lock(&inode->i_lock);
1733 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1736 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1738 for (cnt--; cnt >= 0; cnt--) {
1741 /* Back out changes we already did */
1742 spin_lock(&dquots[cnt]->dq_dqb_lock);
1743 dquot_decr_inodes(dquots[cnt], 1);
1744 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1751 spin_unlock(&inode->i_lock);
1753 mark_all_dquot_dirty(dquots);
1754 srcu_read_unlock(&dquot_srcu, index);
1755 flush_warnings(warn);
1758 EXPORT_SYMBOL(dquot_alloc_inode);
1761 * Convert in-memory reserved quotas to real consumed quotas
1763 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1765 struct dquot **dquots;
1768 if (!dquot_active(inode)) {
1769 spin_lock(&inode->i_lock);
1770 *inode_reserved_space(inode) -= number;
1771 __inode_add_bytes(inode, number);
1772 spin_unlock(&inode->i_lock);
1776 dquots = i_dquot(inode);
1777 index = srcu_read_lock(&dquot_srcu);
1778 spin_lock(&inode->i_lock);
1779 /* Claim reserved quotas to allocated quotas */
1780 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1782 struct dquot *dquot = dquots[cnt];
1784 spin_lock(&dquot->dq_dqb_lock);
1785 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1786 number = dquot->dq_dqb.dqb_rsvspace;
1787 dquot->dq_dqb.dqb_curspace += number;
1788 dquot->dq_dqb.dqb_rsvspace -= number;
1789 spin_unlock(&dquot->dq_dqb_lock);
1792 /* Update inode bytes */
1793 *inode_reserved_space(inode) -= number;
1794 __inode_add_bytes(inode, number);
1795 spin_unlock(&inode->i_lock);
1796 mark_all_dquot_dirty(dquots);
1797 srcu_read_unlock(&dquot_srcu, index);
1800 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1803 * Convert allocated space back to in-memory reserved quotas
1805 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1807 struct dquot **dquots;
1810 if (!dquot_active(inode)) {
1811 spin_lock(&inode->i_lock);
1812 *inode_reserved_space(inode) += number;
1813 __inode_sub_bytes(inode, number);
1814 spin_unlock(&inode->i_lock);
1818 dquots = i_dquot(inode);
1819 index = srcu_read_lock(&dquot_srcu);
1820 spin_lock(&inode->i_lock);
1821 /* Claim reserved quotas to allocated quotas */
1822 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1824 struct dquot *dquot = dquots[cnt];
1826 spin_lock(&dquot->dq_dqb_lock);
1827 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1828 number = dquot->dq_dqb.dqb_curspace;
1829 dquot->dq_dqb.dqb_rsvspace += number;
1830 dquot->dq_dqb.dqb_curspace -= number;
1831 spin_unlock(&dquot->dq_dqb_lock);
1834 /* Update inode bytes */
1835 *inode_reserved_space(inode) += number;
1836 __inode_sub_bytes(inode, number);
1837 spin_unlock(&inode->i_lock);
1838 mark_all_dquot_dirty(dquots);
1839 srcu_read_unlock(&dquot_srcu, index);
1842 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1845 * This operation can block, but only after everything is updated
1847 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1850 struct dquot_warn warn[MAXQUOTAS];
1851 struct dquot **dquots;
1852 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1854 if (!dquot_active(inode)) {
1856 spin_lock(&inode->i_lock);
1857 *inode_reserved_space(inode) -= number;
1858 spin_unlock(&inode->i_lock);
1860 inode_sub_bytes(inode, number);
1865 dquots = i_dquot(inode);
1866 index = srcu_read_lock(&dquot_srcu);
1867 spin_lock(&inode->i_lock);
1868 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1871 warn[cnt].w_type = QUOTA_NL_NOWARN;
1874 spin_lock(&dquots[cnt]->dq_dqb_lock);
1875 wtype = info_bdq_free(dquots[cnt], number);
1876 if (wtype != QUOTA_NL_NOWARN)
1877 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1879 dquot_free_reserved_space(dquots[cnt], number);
1881 dquot_decr_space(dquots[cnt], number);
1882 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1885 *inode_reserved_space(inode) -= number;
1887 __inode_sub_bytes(inode, number);
1888 spin_unlock(&inode->i_lock);
1892 mark_all_dquot_dirty(dquots);
1894 srcu_read_unlock(&dquot_srcu, index);
1895 flush_warnings(warn);
1897 EXPORT_SYMBOL(__dquot_free_space);
1900 * This operation can block, but only after everything is updated
1902 void dquot_free_inode(struct inode *inode)
1905 struct dquot_warn warn[MAXQUOTAS];
1906 struct dquot * const *dquots;
1909 if (!dquot_active(inode))
1912 dquots = i_dquot(inode);
1913 index = srcu_read_lock(&dquot_srcu);
1914 spin_lock(&inode->i_lock);
1915 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1918 warn[cnt].w_type = QUOTA_NL_NOWARN;
1921 spin_lock(&dquots[cnt]->dq_dqb_lock);
1922 wtype = info_idq_free(dquots[cnt], 1);
1923 if (wtype != QUOTA_NL_NOWARN)
1924 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1925 dquot_decr_inodes(dquots[cnt], 1);
1926 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1928 spin_unlock(&inode->i_lock);
1929 mark_all_dquot_dirty(dquots);
1930 srcu_read_unlock(&dquot_srcu, index);
1931 flush_warnings(warn);
1933 EXPORT_SYMBOL(dquot_free_inode);
1936 * Transfer the number of inode and blocks from one diskquota to an other.
1937 * On success, dquot references in transfer_to are consumed and references
1938 * to original dquots that need to be released are placed there. On failure,
1939 * references are kept untouched.
1941 * This operation can block, but only after everything is updated
1942 * A transaction must be started when entering this function.
1944 * We are holding reference on transfer_from & transfer_to, no need to
1945 * protect them by srcu_read_lock().
1947 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1950 qsize_t rsv_space = 0;
1951 qsize_t inode_usage = 1;
1952 struct dquot *transfer_from[MAXQUOTAS] = {};
1954 char is_valid[MAXQUOTAS] = {};
1955 struct dquot_warn warn_to[MAXQUOTAS];
1956 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1957 struct dquot_warn warn_from_space[MAXQUOTAS];
1959 if (IS_NOQUOTA(inode))
1962 if (inode->i_sb->dq_op->get_inode_usage) {
1963 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1968 /* Initialize the arrays */
1969 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1970 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1971 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1972 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1975 spin_lock(&dq_data_lock);
1976 spin_lock(&inode->i_lock);
1977 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1978 spin_unlock(&inode->i_lock);
1979 spin_unlock(&dq_data_lock);
1982 cur_space = __inode_get_bytes(inode);
1983 rsv_space = __inode_get_rsv_space(inode);
1985 * Build the transfer_from list, check limits, and update usage in
1986 * the target structures.
1988 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1990 * Skip changes for same uid or gid or for turned off quota-type.
1992 if (!transfer_to[cnt])
1994 /* Avoid races with quotaoff() */
1995 if (!sb_has_quota_active(inode->i_sb, cnt))
1998 transfer_from[cnt] = i_dquot(inode)[cnt];
1999 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2003 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2004 DQUOT_SPACE_WARN, &warn_to[cnt]);
2006 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2007 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2008 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2013 /* Decrease usage for source structures and update quota pointers */
2014 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2017 /* Due to IO error we might not have transfer_from[] structure */
2018 if (transfer_from[cnt]) {
2021 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2022 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2023 if (wtype != QUOTA_NL_NOWARN)
2024 prepare_warning(&warn_from_inodes[cnt],
2025 transfer_from[cnt], wtype);
2026 wtype = info_bdq_free(transfer_from[cnt],
2027 cur_space + rsv_space);
2028 if (wtype != QUOTA_NL_NOWARN)
2029 prepare_warning(&warn_from_space[cnt],
2030 transfer_from[cnt], wtype);
2031 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2032 dquot_decr_space(transfer_from[cnt], cur_space);
2033 dquot_free_reserved_space(transfer_from[cnt],
2035 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2037 i_dquot(inode)[cnt] = transfer_to[cnt];
2039 spin_unlock(&inode->i_lock);
2040 spin_unlock(&dq_data_lock);
2042 mark_all_dquot_dirty(transfer_from);
2043 mark_all_dquot_dirty(transfer_to);
2044 flush_warnings(warn_to);
2045 flush_warnings(warn_from_inodes);
2046 flush_warnings(warn_from_space);
2047 /* Pass back references to put */
2048 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2050 transfer_to[cnt] = transfer_from[cnt];
2053 /* Back out changes we already did */
2054 for (cnt--; cnt >= 0; cnt--) {
2057 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2058 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2059 dquot_decr_space(transfer_to[cnt], cur_space);
2060 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2061 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2063 spin_unlock(&inode->i_lock);
2064 spin_unlock(&dq_data_lock);
2065 flush_warnings(warn_to);
2068 EXPORT_SYMBOL(__dquot_transfer);
2070 /* Wrapper for transferring ownership of an inode for uid/gid only
2071 * Called from FSXXX_setattr()
2073 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2075 struct dquot *transfer_to[MAXQUOTAS] = {};
2076 struct dquot *dquot;
2077 struct super_block *sb = inode->i_sb;
2080 if (!dquot_active(inode))
2083 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2084 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2085 if (IS_ERR(dquot)) {
2086 if (PTR_ERR(dquot) != -ESRCH) {
2087 ret = PTR_ERR(dquot);
2092 transfer_to[USRQUOTA] = dquot;
2094 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2095 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2096 if (IS_ERR(dquot)) {
2097 if (PTR_ERR(dquot) != -ESRCH) {
2098 ret = PTR_ERR(dquot);
2103 transfer_to[GRPQUOTA] = dquot;
2105 ret = __dquot_transfer(inode, transfer_to);
2107 dqput_all(transfer_to);
2110 EXPORT_SYMBOL(dquot_transfer);
2113 * Write info of quota file to disk
2115 int dquot_commit_info(struct super_block *sb, int type)
2117 struct quota_info *dqopt = sb_dqopt(sb);
2119 return dqopt->ops[type]->write_file_info(sb, type);
2121 EXPORT_SYMBOL(dquot_commit_info);
2123 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2125 struct quota_info *dqopt = sb_dqopt(sb);
2127 if (!sb_has_quota_active(sb, qid->type))
2129 if (!dqopt->ops[qid->type]->get_next_id)
2131 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2133 EXPORT_SYMBOL(dquot_get_next_id);
2136 * Definitions of diskquota operations.
2138 const struct dquot_operations dquot_operations = {
2139 .write_dquot = dquot_commit,
2140 .acquire_dquot = dquot_acquire,
2141 .release_dquot = dquot_release,
2142 .mark_dirty = dquot_mark_dquot_dirty,
2143 .write_info = dquot_commit_info,
2144 .alloc_dquot = dquot_alloc,
2145 .destroy_dquot = dquot_destroy,
2146 .get_next_id = dquot_get_next_id,
2148 EXPORT_SYMBOL(dquot_operations);
2151 * Generic helper for ->open on filesystems supporting disk quotas.
2153 int dquot_file_open(struct inode *inode, struct file *file)
2157 error = generic_file_open(inode, file);
2158 if (!error && (file->f_mode & FMODE_WRITE))
2159 error = dquot_initialize(inode);
2162 EXPORT_SYMBOL(dquot_file_open);
2164 static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2166 struct quota_info *dqopt = sb_dqopt(sb);
2167 struct inode *inode = dqopt->files[type];
2171 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2173 inode->i_flags &= ~S_NOQUOTA;
2174 inode_unlock(inode);
2176 dqopt->files[type] = NULL;
2181 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2183 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2186 struct quota_info *dqopt = sb_dqopt(sb);
2188 /* s_umount should be held in exclusive mode */
2189 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2190 up_read(&sb->s_umount);
2192 /* Cannot turn off usage accounting without turning off limits, or
2193 * suspend quotas and simultaneously turn quotas off. */
2194 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2195 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2196 DQUOT_USAGE_ENABLED)))
2200 * Skip everything if there's nothing to do. We have to do this because
2201 * sometimes we are called when fill_super() failed and calling
2202 * sync_fs() in such cases does no good.
2204 if (!sb_any_quota_loaded(sb))
2207 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2208 if (type != -1 && cnt != type)
2210 if (!sb_has_quota_loaded(sb, cnt))
2213 if (flags & DQUOT_SUSPENDED) {
2214 spin_lock(&dq_state_lock);
2216 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2217 spin_unlock(&dq_state_lock);
2219 spin_lock(&dq_state_lock);
2220 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2221 /* Turning off suspended quotas? */
2222 if (!sb_has_quota_loaded(sb, cnt) &&
2223 sb_has_quota_suspended(sb, cnt)) {
2224 dqopt->flags &= ~dquot_state_flag(
2225 DQUOT_SUSPENDED, cnt);
2226 spin_unlock(&dq_state_lock);
2227 vfs_cleanup_quota_inode(sb, cnt);
2230 spin_unlock(&dq_state_lock);
2233 /* We still have to keep quota loaded? */
2234 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2237 /* Note: these are blocking operations */
2238 drop_dquot_ref(sb, cnt);
2239 invalidate_dquots(sb, cnt);
2241 * Now all dquots should be invalidated, all writes done so we
2242 * should be only users of the info. No locks needed.
2244 if (info_dirty(&dqopt->info[cnt]))
2245 sb->dq_op->write_info(sb, cnt);
2246 if (dqopt->ops[cnt]->free_file_info)
2247 dqopt->ops[cnt]->free_file_info(sb, cnt);
2248 put_quota_format(dqopt->info[cnt].dqi_format);
2249 dqopt->info[cnt].dqi_flags = 0;
2250 dqopt->info[cnt].dqi_igrace = 0;
2251 dqopt->info[cnt].dqi_bgrace = 0;
2252 dqopt->ops[cnt] = NULL;
2255 /* Skip syncing and setting flags if quota files are hidden */
2256 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2259 /* Sync the superblock so that buffers with quota data are written to
2260 * disk (and so userspace sees correct data afterwards). */
2261 if (sb->s_op->sync_fs)
2262 sb->s_op->sync_fs(sb, 1);
2263 sync_blockdev(sb->s_bdev);
2264 /* Now the quota files are just ordinary files and we can set the
2265 * inode flags back. Moreover we discard the pagecache so that
2266 * userspace sees the writes we did bypassing the pagecache. We
2267 * must also discard the blockdev buffers so that we see the
2268 * changes done by userspace on the next quotaon() */
2269 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2270 if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2271 inode_lock(dqopt->files[cnt]);
2272 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2273 inode_unlock(dqopt->files[cnt]);
2276 invalidate_bdev(sb->s_bdev);
2278 /* We are done when suspending quotas */
2279 if (flags & DQUOT_SUSPENDED)
2282 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2283 if (!sb_has_quota_loaded(sb, cnt))
2284 vfs_cleanup_quota_inode(sb, cnt);
2287 EXPORT_SYMBOL(dquot_disable);
2289 int dquot_quota_off(struct super_block *sb, int type)
2291 return dquot_disable(sb, type,
2292 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2294 EXPORT_SYMBOL(dquot_quota_off);
2297 * Turn quotas on on a device
2300 static int vfs_setup_quota_inode(struct inode *inode, int type)
2302 struct super_block *sb = inode->i_sb;
2303 struct quota_info *dqopt = sb_dqopt(sb);
2305 if (!S_ISREG(inode->i_mode))
2307 if (IS_RDONLY(inode))
2309 if (sb_has_quota_loaded(sb, type))
2312 dqopt->files[type] = igrab(inode);
2313 if (!dqopt->files[type])
2315 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2316 /* We don't want quota and atime on quota files (deadlocks
2317 * possible) Also nobody should write to the file - we use
2318 * special IO operations which ignore the immutable bit. */
2320 inode->i_flags |= S_NOQUOTA;
2321 inode_unlock(inode);
2323 * When S_NOQUOTA is set, remove dquot references as no more
2324 * references can be added
2326 __dquot_drop(inode);
2331 int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2334 struct quota_format_type *fmt = find_quota_format(format_id);
2335 struct quota_info *dqopt = sb_dqopt(sb);
2338 /* Just unsuspend quotas? */
2339 BUG_ON(flags & DQUOT_SUSPENDED);
2340 /* s_umount should be held in exclusive mode */
2341 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2342 up_read(&sb->s_umount);
2346 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2347 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2351 /* Filesystems outside of init_user_ns not yet supported */
2352 if (sb->s_user_ns != &init_user_ns) {
2356 /* Usage always has to be set... */
2357 if (!(flags & DQUOT_USAGE_ENABLED)) {
2361 if (sb_has_quota_loaded(sb, type)) {
2366 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2367 /* As we bypass the pagecache we must now flush all the
2368 * dirty data and invalidate caches so that kernel sees
2369 * changes from userspace. It is not enough to just flush
2370 * the quota file since if blocksize < pagesize, invalidation
2371 * of the cache could fail because of other unrelated dirty
2373 sync_filesystem(sb);
2374 invalidate_bdev(sb->s_bdev);
2378 if (!fmt->qf_ops->check_quota_file(sb, type))
2381 dqopt->ops[type] = fmt->qf_ops;
2382 dqopt->info[type].dqi_format = fmt;
2383 dqopt->info[type].dqi_fmt_id = format_id;
2384 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2385 error = dqopt->ops[type]->read_file_info(sb, type);
2388 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2389 spin_lock(&dq_data_lock);
2390 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2391 spin_unlock(&dq_data_lock);
2393 spin_lock(&dq_state_lock);
2394 dqopt->flags |= dquot_state_flag(flags, type);
2395 spin_unlock(&dq_state_lock);
2397 error = add_dquot_ref(sb, type);
2399 dquot_disable(sb, type, flags);
2403 put_quota_format(fmt);
2407 EXPORT_SYMBOL(dquot_load_quota_sb);
2410 * More powerful function for turning on quotas on given quota inode allowing
2411 * setting of individual quota flags
2413 int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2418 err = vfs_setup_quota_inode(inode, type);
2421 err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2423 vfs_cleanup_quota_inode(inode->i_sb, type);
2426 EXPORT_SYMBOL(dquot_load_quota_inode);
2428 /* Reenable quotas on remount RW */
2429 int dquot_resume(struct super_block *sb, int type)
2431 struct quota_info *dqopt = sb_dqopt(sb);
2435 /* s_umount should be held in exclusive mode */
2436 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2437 up_read(&sb->s_umount);
2439 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2440 if (type != -1 && cnt != type)
2442 if (!sb_has_quota_suspended(sb, cnt))
2445 spin_lock(&dq_state_lock);
2446 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2447 DQUOT_LIMITS_ENABLED,
2449 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2450 spin_unlock(&dq_state_lock);
2452 flags = dquot_generic_flag(flags, cnt);
2453 ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2456 vfs_cleanup_quota_inode(sb, type);
2461 EXPORT_SYMBOL(dquot_resume);
2463 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2464 const struct path *path)
2466 int error = security_quota_on(path->dentry);
2469 /* Quota file not on the same filesystem? */
2470 if (path->dentry->d_sb != sb)
2473 error = dquot_load_quota_inode(d_inode(path->dentry), type,
2474 format_id, DQUOT_USAGE_ENABLED |
2475 DQUOT_LIMITS_ENABLED);
2478 EXPORT_SYMBOL(dquot_quota_on);
2481 * This function is used when filesystem needs to initialize quotas
2482 * during mount time.
2484 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2485 int format_id, int type)
2487 struct dentry *dentry;
2490 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2492 return PTR_ERR(dentry);
2494 if (d_really_is_negative(dentry)) {
2499 error = security_quota_on(dentry);
2501 error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2502 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2508 EXPORT_SYMBOL(dquot_quota_on_mount);
2510 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2514 struct quota_info *dqopt = sb_dqopt(sb);
2516 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2518 /* Accounting cannot be turned on while fs is mounted */
2519 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2522 for (type = 0; type < MAXQUOTAS; type++) {
2523 if (!(flags & qtype_enforce_flag(type)))
2525 /* Can't enforce without accounting */
2526 if (!sb_has_quota_usage_enabled(sb, type)) {
2530 if (sb_has_quota_limits_enabled(sb, type)) {
2534 spin_lock(&dq_state_lock);
2535 dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2536 spin_unlock(&dq_state_lock);
2540 /* Backout enforcement enablement we already did */
2541 for (type--; type >= 0; type--) {
2542 if (flags & qtype_enforce_flag(type))
2543 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2545 /* Error code translation for better compatibility with XFS */
2551 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2555 struct quota_info *dqopt = sb_dqopt(sb);
2557 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2560 * We don't support turning off accounting via quotactl. In principle
2561 * quota infrastructure can do this but filesystems don't expect
2562 * userspace to be able to do it.
2565 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2568 /* Filter out limits not enabled */
2569 for (type = 0; type < MAXQUOTAS; type++)
2570 if (!sb_has_quota_limits_enabled(sb, type))
2571 flags &= ~qtype_enforce_flag(type);
2575 for (type = 0; type < MAXQUOTAS; type++) {
2576 if (flags & qtype_enforce_flag(type)) {
2577 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2584 /* Backout enforcement disabling we already did */
2585 for (type--; type >= 0; type--) {
2586 if (flags & qtype_enforce_flag(type)) {
2587 spin_lock(&dq_state_lock);
2589 dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2590 spin_unlock(&dq_state_lock);
2596 /* Generic routine for getting common part of quota structure */
2597 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2599 struct mem_dqblk *dm = &dquot->dq_dqb;
2601 memset(di, 0, sizeof(*di));
2602 spin_lock(&dquot->dq_dqb_lock);
2603 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2604 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2605 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2606 di->d_ino_softlimit = dm->dqb_isoftlimit;
2607 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2608 di->d_ino_count = dm->dqb_curinodes;
2609 di->d_spc_timer = dm->dqb_btime;
2610 di->d_ino_timer = dm->dqb_itime;
2611 spin_unlock(&dquot->dq_dqb_lock);
2614 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2615 struct qc_dqblk *di)
2617 struct dquot *dquot;
2619 dquot = dqget(sb, qid);
2621 return PTR_ERR(dquot);
2622 do_get_dqblk(dquot, di);
2627 EXPORT_SYMBOL(dquot_get_dqblk);
2629 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2630 struct qc_dqblk *di)
2632 struct dquot *dquot;
2635 if (!sb->dq_op->get_next_id)
2637 err = sb->dq_op->get_next_id(sb, qid);
2640 dquot = dqget(sb, *qid);
2642 return PTR_ERR(dquot);
2643 do_get_dqblk(dquot, di);
2648 EXPORT_SYMBOL(dquot_get_next_dqblk);
2650 #define VFS_QC_MASK \
2651 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2652 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2653 QC_SPC_TIMER | QC_INO_TIMER)
2655 /* Generic routine for setting common part of quota structure */
2656 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2658 struct mem_dqblk *dm = &dquot->dq_dqb;
2659 int check_blim = 0, check_ilim = 0;
2660 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2662 if (di->d_fieldmask & ~VFS_QC_MASK)
2665 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2666 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2667 ((di->d_fieldmask & QC_SPC_HARD) &&
2668 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2669 ((di->d_fieldmask & QC_INO_SOFT) &&
2670 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2671 ((di->d_fieldmask & QC_INO_HARD) &&
2672 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2675 spin_lock(&dquot->dq_dqb_lock);
2676 if (di->d_fieldmask & QC_SPACE) {
2677 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2679 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2682 if (di->d_fieldmask & QC_SPC_SOFT)
2683 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2684 if (di->d_fieldmask & QC_SPC_HARD)
2685 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2686 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2688 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2691 if (di->d_fieldmask & QC_INO_COUNT) {
2692 dm->dqb_curinodes = di->d_ino_count;
2694 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2697 if (di->d_fieldmask & QC_INO_SOFT)
2698 dm->dqb_isoftlimit = di->d_ino_softlimit;
2699 if (di->d_fieldmask & QC_INO_HARD)
2700 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2701 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2703 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2706 if (di->d_fieldmask & QC_SPC_TIMER) {
2707 dm->dqb_btime = di->d_spc_timer;
2709 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2712 if (di->d_fieldmask & QC_INO_TIMER) {
2713 dm->dqb_itime = di->d_ino_timer;
2715 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2719 if (!dm->dqb_bsoftlimit ||
2720 dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2722 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2723 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2724 /* Set grace only if user hasn't provided his own... */
2725 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2728 if (!dm->dqb_isoftlimit ||
2729 dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2731 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2732 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2733 /* Set grace only if user hasn't provided his own... */
2734 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2736 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2738 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2740 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2741 spin_unlock(&dquot->dq_dqb_lock);
2742 mark_dquot_dirty(dquot);
2747 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2748 struct qc_dqblk *di)
2750 struct dquot *dquot;
2753 dquot = dqget(sb, qid);
2754 if (IS_ERR(dquot)) {
2755 rc = PTR_ERR(dquot);
2758 rc = do_set_dqblk(dquot, di);
2763 EXPORT_SYMBOL(dquot_set_dqblk);
2765 /* Generic routine for getting common part of quota file information */
2766 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2768 struct mem_dqinfo *mi;
2769 struct qc_type_state *tstate;
2770 struct quota_info *dqopt = sb_dqopt(sb);
2773 memset(state, 0, sizeof(*state));
2774 for (type = 0; type < MAXQUOTAS; type++) {
2775 if (!sb_has_quota_active(sb, type))
2777 tstate = state->s_state + type;
2778 mi = sb_dqopt(sb)->info + type;
2779 tstate->flags = QCI_ACCT_ENABLED;
2780 spin_lock(&dq_data_lock);
2781 if (mi->dqi_flags & DQF_SYS_FILE)
2782 tstate->flags |= QCI_SYSFILE;
2783 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2784 tstate->flags |= QCI_ROOT_SQUASH;
2785 if (sb_has_quota_limits_enabled(sb, type))
2786 tstate->flags |= QCI_LIMITS_ENFORCED;
2787 tstate->spc_timelimit = mi->dqi_bgrace;
2788 tstate->ino_timelimit = mi->dqi_igrace;
2789 if (dqopt->files[type]) {
2790 tstate->ino = dqopt->files[type]->i_ino;
2791 tstate->blocks = dqopt->files[type]->i_blocks;
2793 tstate->nextents = 1; /* We don't know... */
2794 spin_unlock(&dq_data_lock);
2798 EXPORT_SYMBOL(dquot_get_state);
2800 /* Generic routine for setting common part of quota file information */
2801 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2803 struct mem_dqinfo *mi;
2806 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2807 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2809 if (!sb_has_quota_active(sb, type))
2811 mi = sb_dqopt(sb)->info + type;
2812 if (ii->i_fieldmask & QC_FLAGS) {
2813 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2814 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2817 spin_lock(&dq_data_lock);
2818 if (ii->i_fieldmask & QC_SPC_TIMER)
2819 mi->dqi_bgrace = ii->i_spc_timelimit;
2820 if (ii->i_fieldmask & QC_INO_TIMER)
2821 mi->dqi_igrace = ii->i_ino_timelimit;
2822 if (ii->i_fieldmask & QC_FLAGS) {
2823 if (ii->i_flags & QCI_ROOT_SQUASH)
2824 mi->dqi_flags |= DQF_ROOT_SQUASH;
2826 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2828 spin_unlock(&dq_data_lock);
2829 mark_info_dirty(sb, type);
2830 /* Force write to disk */
2831 sb->dq_op->write_info(sb, type);
2834 EXPORT_SYMBOL(dquot_set_dqinfo);
2836 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2837 .quota_enable = dquot_quota_enable,
2838 .quota_disable = dquot_quota_disable,
2839 .quota_sync = dquot_quota_sync,
2840 .get_state = dquot_get_state,
2841 .set_info = dquot_set_dqinfo,
2842 .get_dqblk = dquot_get_dqblk,
2843 .get_nextdqblk = dquot_get_next_dqblk,
2844 .set_dqblk = dquot_set_dqblk
2846 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2848 static int do_proc_dqstats(struct ctl_table *table, int write,
2849 void __user *buffer, size_t *lenp, loff_t *ppos)
2851 unsigned int type = (unsigned long *)table->data - dqstats.stat;
2852 s64 value = percpu_counter_sum(&dqstats.counter[type]);
2854 /* Filter negative values for non-monotonic counters */
2855 if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2856 type == DQST_FREE_DQUOTS))
2859 /* Update global table */
2860 dqstats.stat[type] = value;
2861 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2864 static struct ctl_table fs_dqstats_table[] = {
2866 .procname = "lookups",
2867 .data = &dqstats.stat[DQST_LOOKUPS],
2868 .maxlen = sizeof(unsigned long),
2870 .proc_handler = do_proc_dqstats,
2873 .procname = "drops",
2874 .data = &dqstats.stat[DQST_DROPS],
2875 .maxlen = sizeof(unsigned long),
2877 .proc_handler = do_proc_dqstats,
2880 .procname = "reads",
2881 .data = &dqstats.stat[DQST_READS],
2882 .maxlen = sizeof(unsigned long),
2884 .proc_handler = do_proc_dqstats,
2887 .procname = "writes",
2888 .data = &dqstats.stat[DQST_WRITES],
2889 .maxlen = sizeof(unsigned long),
2891 .proc_handler = do_proc_dqstats,
2894 .procname = "cache_hits",
2895 .data = &dqstats.stat[DQST_CACHE_HITS],
2896 .maxlen = sizeof(unsigned long),
2898 .proc_handler = do_proc_dqstats,
2901 .procname = "allocated_dquots",
2902 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2903 .maxlen = sizeof(unsigned long),
2905 .proc_handler = do_proc_dqstats,
2908 .procname = "free_dquots",
2909 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2910 .maxlen = sizeof(unsigned long),
2912 .proc_handler = do_proc_dqstats,
2915 .procname = "syncs",
2916 .data = &dqstats.stat[DQST_SYNCS],
2917 .maxlen = sizeof(unsigned long),
2919 .proc_handler = do_proc_dqstats,
2921 #ifdef CONFIG_PRINT_QUOTA_WARNING
2923 .procname = "warnings",
2924 .data = &flag_print_warnings,
2925 .maxlen = sizeof(int),
2927 .proc_handler = proc_dointvec,
2933 static struct ctl_table fs_table[] = {
2935 .procname = "quota",
2937 .child = fs_dqstats_table,
2942 static struct ctl_table sys_table[] = {
2951 static int __init dquot_init(void)
2954 unsigned long nr_hash, order;
2956 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2958 register_sysctl_table(sys_table);
2960 dquot_cachep = kmem_cache_create("dquot",
2961 sizeof(struct dquot), sizeof(unsigned long) * 4,
2962 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2963 SLAB_MEM_SPREAD|SLAB_PANIC),
2967 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2969 panic("Cannot create dquot hash table");
2971 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2972 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2974 panic("Cannot create dquot stat counters");
2977 /* Find power-of-two hlist_heads which can fit into allocation */
2978 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2979 dq_hash_bits = ilog2(nr_hash);
2981 nr_hash = 1UL << dq_hash_bits;
2982 dq_hash_mask = nr_hash - 1;
2983 for (i = 0; i < nr_hash; i++)
2984 INIT_HLIST_HEAD(dquot_hash + i);
2986 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2987 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2989 if (register_shrinker(&dqcache_shrinker))
2990 panic("Cannot register dquot shrinker");
2994 fs_initcall(dquot_init);