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 <linux/blkdev.h>
82 #include <linux/sched/mm.h>
83 #include "../internal.h" /* ugh */
85 #include <linux/uaccess.h>
88 * There are five quota SMP locks:
89 * * dq_list_lock protects all lists with quotas and quota formats.
90 * * dquot->dq_dqb_lock protects data from dq_dqb
91 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
92 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
93 * dquot_transfer() can stabilize amount it transfers
94 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
95 * pointers in the inode
96 * * dq_state_lock protects modifications of quota state (on quotaon and
97 * quotaoff) and readers who care about latest values take it as well.
99 * The spinlock ordering is hence:
100 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
101 * dq_list_lock > dq_state_lock
103 * Note that some things (eg. sb pointer, type, id) doesn't change during
104 * the life of the dquot structure and so needn't to be protected by a lock
106 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
107 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
108 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
109 * inode and before dropping dquot references to avoid use of dquots after
110 * they are freed. dq_data_lock is used to serialize the pointer setting and
111 * clearing operations.
112 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
113 * inode is a quota file). Functions adding pointers from inode to dquots have
114 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
115 * have to do all pointer modifications before dropping dq_data_lock. This makes
116 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
117 * then drops all pointers to dquots from an inode.
119 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
120 * memory (or space for it is being allocated) on the first dqget(), when it is
121 * being written out, and when it is being released on the last dqput(). The
122 * allocation and release operations are serialized by the dq_lock and by
123 * checking the use count in dquot_release().
125 * Lock ordering (including related VFS locks) is the following:
126 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
129 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
130 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
131 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
132 EXPORT_SYMBOL(dq_data_lock);
133 DEFINE_STATIC_SRCU(dquot_srcu);
135 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
137 void __quota_error(struct super_block *sb, const char *func,
138 const char *fmt, ...)
140 if (printk_ratelimit()) {
142 struct va_format vaf;
149 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
150 sb->s_id, func, &vaf);
155 EXPORT_SYMBOL(__quota_error);
157 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
158 static char *quotatypes[] = INITQFNAMES;
160 static struct quota_format_type *quota_formats; /* List of registered formats */
161 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
163 /* SLAB cache for dquot structures */
164 static struct kmem_cache *dquot_cachep;
166 int register_quota_format(struct quota_format_type *fmt)
168 spin_lock(&dq_list_lock);
169 fmt->qf_next = quota_formats;
171 spin_unlock(&dq_list_lock);
174 EXPORT_SYMBOL(register_quota_format);
176 void unregister_quota_format(struct quota_format_type *fmt)
178 struct quota_format_type **actqf;
180 spin_lock(&dq_list_lock);
181 for (actqf = "a_formats; *actqf && *actqf != fmt;
182 actqf = &(*actqf)->qf_next)
185 *actqf = (*actqf)->qf_next;
186 spin_unlock(&dq_list_lock);
188 EXPORT_SYMBOL(unregister_quota_format);
190 static struct quota_format_type *find_quota_format(int id)
192 struct quota_format_type *actqf;
194 spin_lock(&dq_list_lock);
195 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
196 actqf = actqf->qf_next)
198 if (!actqf || !try_module_get(actqf->qf_owner)) {
201 spin_unlock(&dq_list_lock);
203 for (qm = 0; module_names[qm].qm_fmt_id &&
204 module_names[qm].qm_fmt_id != id; qm++)
206 if (!module_names[qm].qm_fmt_id ||
207 request_module(module_names[qm].qm_mod_name))
210 spin_lock(&dq_list_lock);
211 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
212 actqf = actqf->qf_next)
214 if (actqf && !try_module_get(actqf->qf_owner))
217 spin_unlock(&dq_list_lock);
221 static void put_quota_format(struct quota_format_type *fmt)
223 module_put(fmt->qf_owner);
227 * Dquot List Management:
228 * The quota code uses four lists for dquot management: the inuse_list,
229 * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
230 * structure may be on some of those lists, depending on its current state.
232 * All dquots are placed to the end of inuse_list when first created, and this
233 * list is used for invalidate operation, which must look at every dquot.
235 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
236 * and this list is searched whenever we need an available dquot. Dquots are
237 * removed from the list as soon as they are used again, and
238 * dqstats.free_dquots gives the number of dquots on the list. When
239 * dquot is invalidated it's completely released from memory.
241 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
242 * dirtied, and this list is searched when writing dirty dquots back to
243 * quota file. Note that some filesystems do dirty dquot tracking on their
244 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
246 * Dquots with a specific identity (device, type and id) are placed on
247 * one of the dquot_hash[] hash chains. The provides an efficient search
248 * mechanism to locate a specific dquot.
251 static LIST_HEAD(inuse_list);
252 static LIST_HEAD(free_dquots);
253 static unsigned int dq_hash_bits, dq_hash_mask;
254 static struct hlist_head *dquot_hash;
256 struct dqstats dqstats;
257 EXPORT_SYMBOL(dqstats);
259 static qsize_t inode_get_rsv_space(struct inode *inode);
260 static qsize_t __inode_get_rsv_space(struct inode *inode);
261 static int __dquot_initialize(struct inode *inode, int type);
263 static inline unsigned int
264 hashfn(const struct super_block *sb, struct kqid qid)
266 unsigned int id = from_kqid(&init_user_ns, qid);
270 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
271 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
275 * Following list functions expect dq_list_lock to be held
277 static inline void insert_dquot_hash(struct dquot *dquot)
279 struct hlist_head *head;
280 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
281 hlist_add_head(&dquot->dq_hash, head);
284 static inline void remove_dquot_hash(struct dquot *dquot)
286 hlist_del_init(&dquot->dq_hash);
289 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
294 hlist_for_each_entry(dquot, dquot_hash+hashent, 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 unsigned int memalloc;
430 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
432 mutex_lock(&dquot->dq_lock);
433 memalloc = memalloc_nofs_save();
434 if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
435 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
439 /* Make sure flags update is visible after dquot has been filled */
440 smp_mb__before_atomic();
441 set_bit(DQ_READ_B, &dquot->dq_flags);
442 /* Instantiate dquot if needed */
443 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
444 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
445 /* Write the info if needed */
446 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
447 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
448 dquot->dq_sb, dquot->dq_id.type);
458 * Make sure flags update is visible after on-disk struct has been
459 * allocated. Paired with smp_rmb() in dqget().
461 smp_mb__before_atomic();
462 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
464 memalloc_nofs_restore(memalloc);
465 mutex_unlock(&dquot->dq_lock);
468 EXPORT_SYMBOL(dquot_acquire);
471 * Write dquot to disk
473 int dquot_commit(struct dquot *dquot)
476 unsigned int memalloc;
477 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
479 mutex_lock(&dquot->dq_lock);
480 memalloc = memalloc_nofs_save();
481 if (!clear_dquot_dirty(dquot))
483 /* Inactive dquot can be only if there was error during read/init
484 * => we have better not writing it */
485 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
486 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
490 memalloc_nofs_restore(memalloc);
491 mutex_unlock(&dquot->dq_lock);
494 EXPORT_SYMBOL(dquot_commit);
499 int dquot_release(struct dquot *dquot)
501 int ret = 0, ret2 = 0;
502 unsigned int memalloc;
503 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
505 mutex_lock(&dquot->dq_lock);
506 memalloc = memalloc_nofs_save();
507 /* Check whether we are not racing with some other dqget() */
508 if (dquot_is_busy(dquot))
510 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
511 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
513 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
514 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
515 dquot->dq_sb, dquot->dq_id.type);
520 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
522 memalloc_nofs_restore(memalloc);
523 mutex_unlock(&dquot->dq_lock);
526 EXPORT_SYMBOL(dquot_release);
528 void dquot_destroy(struct dquot *dquot)
530 kmem_cache_free(dquot_cachep, dquot);
532 EXPORT_SYMBOL(dquot_destroy);
534 static inline void do_destroy_dquot(struct dquot *dquot)
536 dquot->dq_sb->dq_op->destroy_dquot(dquot);
539 /* Invalidate all dquots on the list. Note that this function is called after
540 * quota is disabled and pointers from inodes removed so there cannot be new
541 * quota users. There can still be some users of quotas due to inodes being
542 * just deleted or pruned by prune_icache() (those are not attached to any
543 * list) or parallel quotactl call. We have to wait for such users.
545 static void invalidate_dquots(struct super_block *sb, int type)
547 struct dquot *dquot, *tmp;
550 spin_lock(&dq_list_lock);
551 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
552 if (dquot->dq_sb != sb)
554 if (dquot->dq_id.type != type)
556 /* Wait for dquot users */
557 if (atomic_read(&dquot->dq_count)) {
559 spin_unlock(&dq_list_lock);
561 * Once dqput() wakes us up, we know it's time to free
563 * IMPORTANT: we rely on the fact that there is always
564 * at most one process waiting for dquot to free.
565 * Otherwise dq_count would be > 1 and we would never
568 wait_event(dquot_ref_wq,
569 atomic_read(&dquot->dq_count) == 1);
571 /* At this moment dquot() need not exist (it could be
572 * reclaimed by prune_dqcache(). Hence we must
577 * Quota now has no users and it has been written on last
580 remove_dquot_hash(dquot);
581 remove_free_dquot(dquot);
583 do_destroy_dquot(dquot);
585 spin_unlock(&dq_list_lock);
588 /* Call callback for every active dquot on given filesystem */
589 int dquot_scan_active(struct super_block *sb,
590 int (*fn)(struct dquot *dquot, unsigned long priv),
593 struct dquot *dquot, *old_dquot = NULL;
596 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
598 spin_lock(&dq_list_lock);
599 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
600 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
602 if (dquot->dq_sb != sb)
604 /* Now we have active dquot so we can just increase use count */
605 atomic_inc(&dquot->dq_count);
606 spin_unlock(&dq_list_lock);
610 * ->release_dquot() can be racing with us. Our reference
611 * protects us from new calls to it so just wait for any
612 * outstanding call and recheck the DQ_ACTIVE_B after that.
614 wait_on_dquot(dquot);
615 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
616 ret = fn(dquot, priv);
620 spin_lock(&dq_list_lock);
621 /* We are safe to continue now because our dquot could not
622 * be moved out of the inuse list while we hold the reference */
624 spin_unlock(&dq_list_lock);
629 EXPORT_SYMBOL(dquot_scan_active);
631 /* Write all dquot structures to quota files */
632 int dquot_writeback_dquots(struct super_block *sb, int type)
634 struct list_head dirty;
636 struct quota_info *dqopt = sb_dqopt(sb);
640 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
642 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
643 if (type != -1 && cnt != type)
645 if (!sb_has_quota_active(sb, cnt))
647 spin_lock(&dq_list_lock);
648 /* Move list away to avoid livelock. */
649 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
650 while (!list_empty(&dirty)) {
651 dquot = list_first_entry(&dirty, struct dquot,
654 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
656 /* Now we have active dquot from which someone is
657 * holding reference so we can safely just increase
660 spin_unlock(&dq_list_lock);
661 err = sb->dq_op->write_dquot(dquot);
664 * Clear dirty bit anyway to avoid infinite
667 clear_dquot_dirty(dquot);
672 spin_lock(&dq_list_lock);
674 spin_unlock(&dq_list_lock);
677 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
678 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
679 && info_dirty(&dqopt->info[cnt]))
680 sb->dq_op->write_info(sb, cnt);
681 dqstats_inc(DQST_SYNCS);
685 EXPORT_SYMBOL(dquot_writeback_dquots);
687 /* Write all dquot structures to disk and make them visible from userspace */
688 int dquot_quota_sync(struct super_block *sb, int type)
690 struct quota_info *dqopt = sb_dqopt(sb);
694 ret = dquot_writeback_dquots(sb, type);
697 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
700 /* This is not very clever (and fast) but currently I don't know about
701 * any other simple way of getting quota data to disk and we must get
702 * them there for userspace to be visible... */
703 if (sb->s_op->sync_fs) {
704 ret = sb->s_op->sync_fs(sb, 1);
708 ret = sync_blockdev(sb->s_bdev);
713 * Now when everything is written we can discard the pagecache so
714 * that userspace sees the changes.
716 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
717 if (type != -1 && cnt != type)
719 if (!sb_has_quota_active(sb, cnt))
721 inode_lock(dqopt->files[cnt]);
722 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
723 inode_unlock(dqopt->files[cnt]);
728 EXPORT_SYMBOL(dquot_quota_sync);
731 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
734 unsigned long freed = 0;
736 spin_lock(&dq_list_lock);
737 while (!list_empty(&free_dquots) && sc->nr_to_scan) {
738 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
739 remove_dquot_hash(dquot);
740 remove_free_dquot(dquot);
742 do_destroy_dquot(dquot);
746 spin_unlock(&dq_list_lock);
751 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
753 return vfs_pressure_ratio(
754 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
757 static struct shrinker dqcache_shrinker = {
758 .count_objects = dqcache_shrink_count,
759 .scan_objects = dqcache_shrink_scan,
760 .seeks = DEFAULT_SEEKS,
764 * Put reference to dquot
766 void dqput(struct dquot *dquot)
772 #ifdef CONFIG_QUOTA_DEBUG
773 if (!atomic_read(&dquot->dq_count)) {
774 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
775 quotatypes[dquot->dq_id.type],
776 from_kqid(&init_user_ns, dquot->dq_id));
780 dqstats_inc(DQST_DROPS);
782 spin_lock(&dq_list_lock);
783 if (atomic_read(&dquot->dq_count) > 1) {
784 /* We have more than one user... nothing to do */
785 atomic_dec(&dquot->dq_count);
786 /* Releasing dquot during quotaoff phase? */
787 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
788 atomic_read(&dquot->dq_count) == 1)
789 wake_up(&dquot_ref_wq);
790 spin_unlock(&dq_list_lock);
793 /* Need to release dquot? */
794 if (dquot_dirty(dquot)) {
795 spin_unlock(&dq_list_lock);
796 /* Commit dquot before releasing */
797 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
799 quota_error(dquot->dq_sb, "Can't write quota structure"
800 " (error %d). Quota may get out of sync!",
803 * We clear dirty bit anyway, so that we avoid
806 clear_dquot_dirty(dquot);
810 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
811 spin_unlock(&dq_list_lock);
812 dquot->dq_sb->dq_op->release_dquot(dquot);
815 atomic_dec(&dquot->dq_count);
816 #ifdef CONFIG_QUOTA_DEBUG
818 BUG_ON(!list_empty(&dquot->dq_free));
820 put_dquot_last(dquot);
821 spin_unlock(&dq_list_lock);
823 EXPORT_SYMBOL(dqput);
825 struct dquot *dquot_alloc(struct super_block *sb, int type)
827 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
829 EXPORT_SYMBOL(dquot_alloc);
831 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
835 dquot = sb->dq_op->alloc_dquot(sb, type);
839 mutex_init(&dquot->dq_lock);
840 INIT_LIST_HEAD(&dquot->dq_free);
841 INIT_LIST_HEAD(&dquot->dq_inuse);
842 INIT_HLIST_NODE(&dquot->dq_hash);
843 INIT_LIST_HEAD(&dquot->dq_dirty);
845 dquot->dq_id = make_kqid_invalid(type);
846 atomic_set(&dquot->dq_count, 1);
847 spin_lock_init(&dquot->dq_dqb_lock);
853 * Get reference to dquot
855 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
856 * destroying our dquot by:
857 * a) checking for quota flags under dq_list_lock and
858 * b) getting a reference to dquot before we release dq_list_lock
860 struct dquot *dqget(struct super_block *sb, struct kqid qid)
862 unsigned int hashent = hashfn(sb, qid);
863 struct dquot *dquot, *empty = NULL;
865 if (!qid_has_mapping(sb->s_user_ns, qid))
866 return ERR_PTR(-EINVAL);
868 if (!sb_has_quota_active(sb, qid.type))
869 return ERR_PTR(-ESRCH);
871 spin_lock(&dq_list_lock);
872 spin_lock(&dq_state_lock);
873 if (!sb_has_quota_active(sb, qid.type)) {
874 spin_unlock(&dq_state_lock);
875 spin_unlock(&dq_list_lock);
876 dquot = ERR_PTR(-ESRCH);
879 spin_unlock(&dq_state_lock);
881 dquot = find_dquot(hashent, sb, qid);
884 spin_unlock(&dq_list_lock);
885 empty = get_empty_dquot(sb, qid.type);
887 schedule(); /* Try to wait for a moment... */
893 /* all dquots go on the inuse_list */
895 /* hash it first so it can be found */
896 insert_dquot_hash(dquot);
897 spin_unlock(&dq_list_lock);
898 dqstats_inc(DQST_LOOKUPS);
900 if (!atomic_read(&dquot->dq_count))
901 remove_free_dquot(dquot);
902 atomic_inc(&dquot->dq_count);
903 spin_unlock(&dq_list_lock);
904 dqstats_inc(DQST_CACHE_HITS);
905 dqstats_inc(DQST_LOOKUPS);
907 /* Wait for dq_lock - after this we know that either dquot_release() is
908 * already finished or it will be canceled due to dq_count > 1 test */
909 wait_on_dquot(dquot);
910 /* Read the dquot / allocate space in quota file */
911 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
914 err = sb->dq_op->acquire_dquot(dquot);
917 dquot = ERR_PTR(err);
922 * Make sure following reads see filled structure - paired with
923 * smp_mb__before_atomic() in dquot_acquire().
926 #ifdef CONFIG_QUOTA_DEBUG
927 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
931 do_destroy_dquot(empty);
935 EXPORT_SYMBOL(dqget);
937 static inline struct dquot **i_dquot(struct inode *inode)
939 return inode->i_sb->s_op->get_dquots(inode);
942 static int dqinit_needed(struct inode *inode, int type)
944 struct dquot * const *dquots;
947 if (IS_NOQUOTA(inode))
950 dquots = i_dquot(inode);
952 return !dquots[type];
953 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
959 /* This routine is guarded by s_umount semaphore */
960 static int add_dquot_ref(struct super_block *sb, int type)
962 struct inode *inode, *old_inode = NULL;
963 #ifdef CONFIG_QUOTA_DEBUG
968 spin_lock(&sb->s_inode_list_lock);
969 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
970 spin_lock(&inode->i_lock);
971 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
972 !atomic_read(&inode->i_writecount) ||
973 !dqinit_needed(inode, type)) {
974 spin_unlock(&inode->i_lock);
978 spin_unlock(&inode->i_lock);
979 spin_unlock(&sb->s_inode_list_lock);
981 #ifdef CONFIG_QUOTA_DEBUG
982 if (unlikely(inode_get_rsv_space(inode) > 0))
986 err = __dquot_initialize(inode, type);
993 * We hold a reference to 'inode' so it couldn't have been
994 * removed from s_inodes list while we dropped the
995 * s_inode_list_lock. We cannot iput the inode now as we can be
996 * holding the last reference and we cannot iput it under
997 * s_inode_list_lock. So we keep the reference and iput it
1002 spin_lock(&sb->s_inode_list_lock);
1004 spin_unlock(&sb->s_inode_list_lock);
1007 #ifdef CONFIG_QUOTA_DEBUG
1009 quota_error(sb, "Writes happened before quota was turned on "
1010 "thus quota information is probably inconsistent. "
1011 "Please run quotacheck(8)");
1018 * Remove references to dquots from inode and add dquot to list for freeing
1019 * if we have the last reference to dquot
1021 static void remove_inode_dquot_ref(struct inode *inode, int type,
1022 struct list_head *tofree_head)
1024 struct dquot **dquots = i_dquot(inode);
1025 struct dquot *dquot = dquots[type];
1030 dquots[type] = NULL;
1031 if (list_empty(&dquot->dq_free)) {
1033 * The inode still has reference to dquot so it can't be in the
1036 spin_lock(&dq_list_lock);
1037 list_add(&dquot->dq_free, tofree_head);
1038 spin_unlock(&dq_list_lock);
1041 * Dquot is already in a list to put so we won't drop the last
1049 * Free list of dquots
1050 * Dquots are removed from inodes and no new references can be got so we are
1051 * the only ones holding reference
1053 static void put_dquot_list(struct list_head *tofree_head)
1055 struct list_head *act_head;
1056 struct dquot *dquot;
1058 act_head = tofree_head->next;
1059 while (act_head != tofree_head) {
1060 dquot = list_entry(act_head, struct dquot, dq_free);
1061 act_head = act_head->next;
1062 /* Remove dquot from the list so we won't have problems... */
1063 list_del_init(&dquot->dq_free);
1068 static void remove_dquot_ref(struct super_block *sb, int type,
1069 struct list_head *tofree_head)
1071 struct inode *inode;
1072 #ifdef CONFIG_QUOTA_DEBUG
1076 spin_lock(&sb->s_inode_list_lock);
1077 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1079 * We have to scan also I_NEW inodes because they can already
1080 * have quota pointer initialized. Luckily, we need to touch
1081 * only quota pointers and these have separate locking
1084 spin_lock(&dq_data_lock);
1085 if (!IS_NOQUOTA(inode)) {
1086 #ifdef CONFIG_QUOTA_DEBUG
1087 if (unlikely(inode_get_rsv_space(inode) > 0))
1090 remove_inode_dquot_ref(inode, type, tofree_head);
1092 spin_unlock(&dq_data_lock);
1094 spin_unlock(&sb->s_inode_list_lock);
1095 #ifdef CONFIG_QUOTA_DEBUG
1097 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1098 " was disabled thus quota information is probably "
1099 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1104 /* Gather all references from inodes and drop them */
1105 static void drop_dquot_ref(struct super_block *sb, int type)
1107 LIST_HEAD(tofree_head);
1110 remove_dquot_ref(sb, type, &tofree_head);
1111 synchronize_srcu(&dquot_srcu);
1112 put_dquot_list(&tofree_head);
1117 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1119 if (dquot->dq_dqb.dqb_rsvspace >= number)
1120 dquot->dq_dqb.dqb_rsvspace -= number;
1123 dquot->dq_dqb.dqb_rsvspace = 0;
1125 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1126 dquot->dq_dqb.dqb_bsoftlimit)
1127 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1128 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1131 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1133 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1134 dquot->dq_dqb.dqb_curinodes >= number)
1135 dquot->dq_dqb.dqb_curinodes -= number;
1137 dquot->dq_dqb.dqb_curinodes = 0;
1138 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1139 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1140 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1143 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1145 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1146 dquot->dq_dqb.dqb_curspace >= number)
1147 dquot->dq_dqb.dqb_curspace -= number;
1149 dquot->dq_dqb.dqb_curspace = 0;
1150 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1151 dquot->dq_dqb.dqb_bsoftlimit)
1152 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1153 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1157 struct super_block *w_sb;
1158 struct kqid w_dq_id;
1162 static int warning_issued(struct dquot *dquot, const int warntype)
1164 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1165 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1166 ((warntype == QUOTA_NL_IHARDWARN ||
1167 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1171 return test_and_set_bit(flag, &dquot->dq_flags);
1174 #ifdef CONFIG_PRINT_QUOTA_WARNING
1175 static int flag_print_warnings = 1;
1177 static int need_print_warning(struct dquot_warn *warn)
1179 if (!flag_print_warnings)
1182 switch (warn->w_dq_id.type) {
1184 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1186 return in_group_p(warn->w_dq_id.gid);
1193 /* Print warning to user which exceeded quota */
1194 static void print_warning(struct dquot_warn *warn)
1197 struct tty_struct *tty;
1198 int warntype = warn->w_type;
1200 if (warntype == QUOTA_NL_IHARDBELOW ||
1201 warntype == QUOTA_NL_ISOFTBELOW ||
1202 warntype == QUOTA_NL_BHARDBELOW ||
1203 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1206 tty = get_current_tty();
1209 tty_write_message(tty, warn->w_sb->s_id);
1210 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1211 tty_write_message(tty, ": warning, ");
1213 tty_write_message(tty, ": write failed, ");
1214 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1216 case QUOTA_NL_IHARDWARN:
1217 msg = " file limit reached.\r\n";
1219 case QUOTA_NL_ISOFTLONGWARN:
1220 msg = " file quota exceeded too long.\r\n";
1222 case QUOTA_NL_ISOFTWARN:
1223 msg = " file quota exceeded.\r\n";
1225 case QUOTA_NL_BHARDWARN:
1226 msg = " block limit reached.\r\n";
1228 case QUOTA_NL_BSOFTLONGWARN:
1229 msg = " block quota exceeded too long.\r\n";
1231 case QUOTA_NL_BSOFTWARN:
1232 msg = " block quota exceeded.\r\n";
1235 tty_write_message(tty, msg);
1240 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1243 if (warning_issued(dquot, warntype))
1245 warn->w_type = warntype;
1246 warn->w_sb = dquot->dq_sb;
1247 warn->w_dq_id = dquot->dq_id;
1251 * Write warnings to the console and send warning messages over netlink.
1253 * Note that this function can call into tty and networking code.
1255 static void flush_warnings(struct dquot_warn *warn)
1259 for (i = 0; i < MAXQUOTAS; i++) {
1260 if (warn[i].w_type == QUOTA_NL_NOWARN)
1262 #ifdef CONFIG_PRINT_QUOTA_WARNING
1263 print_warning(&warn[i]);
1265 quota_send_warning(warn[i].w_dq_id,
1266 warn[i].w_sb->s_dev, warn[i].w_type);
1270 static int ignore_hardlimit(struct dquot *dquot)
1272 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1274 return capable(CAP_SYS_RESOURCE) &&
1275 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1276 !(info->dqi_flags & DQF_ROOT_SQUASH));
1279 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1280 struct dquot_warn *warn)
1285 spin_lock(&dquot->dq_dqb_lock);
1286 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1287 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1288 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1291 if (dquot->dq_dqb.dqb_ihardlimit &&
1292 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1293 !ignore_hardlimit(dquot)) {
1294 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1299 if (dquot->dq_dqb.dqb_isoftlimit &&
1300 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1301 dquot->dq_dqb.dqb_itime &&
1302 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1303 !ignore_hardlimit(dquot)) {
1304 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1309 if (dquot->dq_dqb.dqb_isoftlimit &&
1310 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1311 dquot->dq_dqb.dqb_itime == 0) {
1312 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1313 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1314 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1317 dquot->dq_dqb.dqb_curinodes = newinodes;
1320 spin_unlock(&dquot->dq_dqb_lock);
1324 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1325 qsize_t rsv_space, unsigned int flags,
1326 struct dquot_warn *warn)
1329 struct super_block *sb = dquot->dq_sb;
1332 spin_lock(&dquot->dq_dqb_lock);
1333 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1334 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1337 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1338 + space + rsv_space;
1340 if (dquot->dq_dqb.dqb_bhardlimit &&
1341 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1342 !ignore_hardlimit(dquot)) {
1343 if (flags & DQUOT_SPACE_WARN)
1344 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1349 if (dquot->dq_dqb.dqb_bsoftlimit &&
1350 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1351 dquot->dq_dqb.dqb_btime &&
1352 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1353 !ignore_hardlimit(dquot)) {
1354 if (flags & DQUOT_SPACE_WARN)
1355 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1360 if (dquot->dq_dqb.dqb_bsoftlimit &&
1361 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1362 dquot->dq_dqb.dqb_btime == 0) {
1363 if (flags & DQUOT_SPACE_WARN) {
1364 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1365 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1366 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1369 * We don't allow preallocation to exceed softlimit so exceeding will
1378 * We have to be careful and go through warning generation & grace time
1379 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1382 if (flags & DQUOT_SPACE_NOFAIL)
1385 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1386 dquot->dq_dqb.dqb_curspace += space;
1388 spin_unlock(&dquot->dq_dqb_lock);
1392 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1396 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1397 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1398 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1399 return QUOTA_NL_NOWARN;
1401 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1402 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1403 return QUOTA_NL_ISOFTBELOW;
1404 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1405 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1406 return QUOTA_NL_IHARDBELOW;
1407 return QUOTA_NL_NOWARN;
1410 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1414 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1416 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1417 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1418 return QUOTA_NL_NOWARN;
1420 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1421 return QUOTA_NL_BSOFTBELOW;
1422 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1423 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1424 return QUOTA_NL_BHARDBELOW;
1425 return QUOTA_NL_NOWARN;
1428 static int dquot_active(const struct inode *inode)
1430 struct super_block *sb = inode->i_sb;
1432 if (IS_NOQUOTA(inode))
1434 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1438 * Initialize quota pointers in inode
1440 * It is better to call this function outside of any transaction as it
1441 * might need a lot of space in journal for dquot structure allocation.
1443 static int __dquot_initialize(struct inode *inode, int type)
1445 int cnt, init_needed = 0;
1446 struct dquot **dquots, *got[MAXQUOTAS] = {};
1447 struct super_block *sb = inode->i_sb;
1451 if (!dquot_active(inode))
1454 dquots = i_dquot(inode);
1456 /* First get references to structures we might need. */
1457 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1461 struct dquot *dquot;
1463 if (type != -1 && cnt != type)
1466 * The i_dquot should have been initialized in most cases,
1467 * we check it without locking here to avoid unnecessary
1468 * dqget()/dqput() calls.
1473 if (!sb_has_quota_active(sb, cnt))
1480 qid = make_kqid_uid(inode->i_uid);
1483 qid = make_kqid_gid(inode->i_gid);
1486 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1489 qid = make_kqid_projid(projid);
1492 dquot = dqget(sb, qid);
1493 if (IS_ERR(dquot)) {
1494 /* We raced with somebody turning quotas off... */
1495 if (PTR_ERR(dquot) != -ESRCH) {
1496 ret = PTR_ERR(dquot);
1504 /* All required i_dquot has been initialized */
1508 spin_lock(&dq_data_lock);
1509 if (IS_NOQUOTA(inode))
1511 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1512 if (type != -1 && cnt != type)
1514 /* Avoid races with quotaoff() */
1515 if (!sb_has_quota_active(sb, cnt))
1517 /* We could race with quotaon or dqget() could have failed */
1521 dquots[cnt] = got[cnt];
1524 * Make quota reservation system happy if someone
1525 * did a write before quota was turned on
1527 rsv = inode_get_rsv_space(inode);
1528 if (unlikely(rsv)) {
1529 spin_lock(&inode->i_lock);
1530 /* Get reservation again under proper lock */
1531 rsv = __inode_get_rsv_space(inode);
1532 spin_lock(&dquots[cnt]->dq_dqb_lock);
1533 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1534 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1535 spin_unlock(&inode->i_lock);
1540 spin_unlock(&dq_data_lock);
1542 /* Drop unused references */
1548 int dquot_initialize(struct inode *inode)
1550 return __dquot_initialize(inode, -1);
1552 EXPORT_SYMBOL(dquot_initialize);
1554 bool dquot_initialize_needed(struct inode *inode)
1556 struct dquot **dquots;
1559 if (!dquot_active(inode))
1562 dquots = i_dquot(inode);
1563 for (i = 0; i < MAXQUOTAS; i++)
1564 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1568 EXPORT_SYMBOL(dquot_initialize_needed);
1571 * Release all quotas referenced by inode.
1573 * This function only be called on inode free or converting
1574 * a file to quota file, no other users for the i_dquot in
1575 * both cases, so we needn't call synchronize_srcu() after
1578 static void __dquot_drop(struct inode *inode)
1581 struct dquot **dquots = i_dquot(inode);
1582 struct dquot *put[MAXQUOTAS];
1584 spin_lock(&dq_data_lock);
1585 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1586 put[cnt] = dquots[cnt];
1589 spin_unlock(&dq_data_lock);
1593 void dquot_drop(struct inode *inode)
1595 struct dquot * const *dquots;
1598 if (IS_NOQUOTA(inode))
1602 * Test before calling to rule out calls from proc and such
1603 * where we are not allowed to block. Note that this is
1604 * actually reliable test even without the lock - the caller
1605 * must assure that nobody can come after the DQUOT_DROP and
1606 * add quota pointers back anyway.
1608 dquots = i_dquot(inode);
1609 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1614 if (cnt < MAXQUOTAS)
1615 __dquot_drop(inode);
1617 EXPORT_SYMBOL(dquot_drop);
1620 * inode_reserved_space is managed internally by quota, and protected by
1621 * i_lock similar to i_blocks+i_bytes.
1623 static qsize_t *inode_reserved_space(struct inode * inode)
1625 /* Filesystem must explicitly define it's own method in order to use
1626 * quota reservation interface */
1627 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1628 return inode->i_sb->dq_op->get_reserved_space(inode);
1631 static qsize_t __inode_get_rsv_space(struct inode *inode)
1633 if (!inode->i_sb->dq_op->get_reserved_space)
1635 return *inode_reserved_space(inode);
1638 static qsize_t inode_get_rsv_space(struct inode *inode)
1642 if (!inode->i_sb->dq_op->get_reserved_space)
1644 spin_lock(&inode->i_lock);
1645 ret = __inode_get_rsv_space(inode);
1646 spin_unlock(&inode->i_lock);
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)) {
1672 spin_lock(&inode->i_lock);
1673 *inode_reserved_space(inode) += number;
1674 spin_unlock(&inode->i_lock);
1676 inode_add_bytes(inode, number);
1681 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1682 warn[cnt].w_type = QUOTA_NL_NOWARN;
1684 dquots = i_dquot(inode);
1685 index = srcu_read_lock(&dquot_srcu);
1686 spin_lock(&inode->i_lock);
1687 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1691 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1694 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1698 /* Back out changes we already did */
1699 for (cnt--; cnt >= 0; cnt--) {
1702 spin_lock(&dquots[cnt]->dq_dqb_lock);
1704 dquot_free_reserved_space(dquots[cnt],
1707 dquot_decr_space(dquots[cnt], number);
1708 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1710 spin_unlock(&inode->i_lock);
1711 goto out_flush_warn;
1715 *inode_reserved_space(inode) += number;
1717 __inode_add_bytes(inode, number);
1718 spin_unlock(&inode->i_lock);
1721 goto out_flush_warn;
1722 mark_all_dquot_dirty(dquots);
1724 srcu_read_unlock(&dquot_srcu, index);
1725 flush_warnings(warn);
1729 EXPORT_SYMBOL(__dquot_alloc_space);
1732 * This operation can block, but only after everything is updated
1734 int dquot_alloc_inode(struct inode *inode)
1736 int cnt, ret = 0, index;
1737 struct dquot_warn warn[MAXQUOTAS];
1738 struct dquot * const *dquots;
1740 if (!dquot_active(inode))
1742 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1743 warn[cnt].w_type = QUOTA_NL_NOWARN;
1745 dquots = i_dquot(inode);
1746 index = srcu_read_lock(&dquot_srcu);
1747 spin_lock(&inode->i_lock);
1748 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1751 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1753 for (cnt--; cnt >= 0; cnt--) {
1756 /* Back out changes we already did */
1757 spin_lock(&dquots[cnt]->dq_dqb_lock);
1758 dquot_decr_inodes(dquots[cnt], 1);
1759 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1766 spin_unlock(&inode->i_lock);
1768 mark_all_dquot_dirty(dquots);
1769 srcu_read_unlock(&dquot_srcu, index);
1770 flush_warnings(warn);
1773 EXPORT_SYMBOL(dquot_alloc_inode);
1776 * Convert in-memory reserved quotas to real consumed quotas
1778 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1780 struct dquot **dquots;
1783 if (!dquot_active(inode)) {
1784 spin_lock(&inode->i_lock);
1785 *inode_reserved_space(inode) -= number;
1786 __inode_add_bytes(inode, number);
1787 spin_unlock(&inode->i_lock);
1791 dquots = i_dquot(inode);
1792 index = srcu_read_lock(&dquot_srcu);
1793 spin_lock(&inode->i_lock);
1794 /* Claim reserved quotas to allocated quotas */
1795 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1797 struct dquot *dquot = dquots[cnt];
1799 spin_lock(&dquot->dq_dqb_lock);
1800 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1801 number = dquot->dq_dqb.dqb_rsvspace;
1802 dquot->dq_dqb.dqb_curspace += number;
1803 dquot->dq_dqb.dqb_rsvspace -= number;
1804 spin_unlock(&dquot->dq_dqb_lock);
1807 /* Update inode bytes */
1808 *inode_reserved_space(inode) -= number;
1809 __inode_add_bytes(inode, number);
1810 spin_unlock(&inode->i_lock);
1811 mark_all_dquot_dirty(dquots);
1812 srcu_read_unlock(&dquot_srcu, index);
1815 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1818 * Convert allocated space back to in-memory reserved quotas
1820 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1822 struct dquot **dquots;
1825 if (!dquot_active(inode)) {
1826 spin_lock(&inode->i_lock);
1827 *inode_reserved_space(inode) += number;
1828 __inode_sub_bytes(inode, number);
1829 spin_unlock(&inode->i_lock);
1833 dquots = i_dquot(inode);
1834 index = srcu_read_lock(&dquot_srcu);
1835 spin_lock(&inode->i_lock);
1836 /* Claim reserved quotas to allocated quotas */
1837 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1839 struct dquot *dquot = dquots[cnt];
1841 spin_lock(&dquot->dq_dqb_lock);
1842 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1843 number = dquot->dq_dqb.dqb_curspace;
1844 dquot->dq_dqb.dqb_rsvspace += number;
1845 dquot->dq_dqb.dqb_curspace -= number;
1846 spin_unlock(&dquot->dq_dqb_lock);
1849 /* Update inode bytes */
1850 *inode_reserved_space(inode) += number;
1851 __inode_sub_bytes(inode, number);
1852 spin_unlock(&inode->i_lock);
1853 mark_all_dquot_dirty(dquots);
1854 srcu_read_unlock(&dquot_srcu, index);
1857 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1860 * This operation can block, but only after everything is updated
1862 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1865 struct dquot_warn warn[MAXQUOTAS];
1866 struct dquot **dquots;
1867 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1869 if (!dquot_active(inode)) {
1871 spin_lock(&inode->i_lock);
1872 *inode_reserved_space(inode) -= number;
1873 spin_unlock(&inode->i_lock);
1875 inode_sub_bytes(inode, number);
1880 dquots = i_dquot(inode);
1881 index = srcu_read_lock(&dquot_srcu);
1882 spin_lock(&inode->i_lock);
1883 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1886 warn[cnt].w_type = QUOTA_NL_NOWARN;
1889 spin_lock(&dquots[cnt]->dq_dqb_lock);
1890 wtype = info_bdq_free(dquots[cnt], number);
1891 if (wtype != QUOTA_NL_NOWARN)
1892 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1894 dquot_free_reserved_space(dquots[cnt], number);
1896 dquot_decr_space(dquots[cnt], number);
1897 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1900 *inode_reserved_space(inode) -= number;
1902 __inode_sub_bytes(inode, number);
1903 spin_unlock(&inode->i_lock);
1907 mark_all_dquot_dirty(dquots);
1909 srcu_read_unlock(&dquot_srcu, index);
1910 flush_warnings(warn);
1912 EXPORT_SYMBOL(__dquot_free_space);
1915 * This operation can block, but only after everything is updated
1917 void dquot_free_inode(struct inode *inode)
1920 struct dquot_warn warn[MAXQUOTAS];
1921 struct dquot * const *dquots;
1924 if (!dquot_active(inode))
1927 dquots = i_dquot(inode);
1928 index = srcu_read_lock(&dquot_srcu);
1929 spin_lock(&inode->i_lock);
1930 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1933 warn[cnt].w_type = QUOTA_NL_NOWARN;
1936 spin_lock(&dquots[cnt]->dq_dqb_lock);
1937 wtype = info_idq_free(dquots[cnt], 1);
1938 if (wtype != QUOTA_NL_NOWARN)
1939 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1940 dquot_decr_inodes(dquots[cnt], 1);
1941 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1943 spin_unlock(&inode->i_lock);
1944 mark_all_dquot_dirty(dquots);
1945 srcu_read_unlock(&dquot_srcu, index);
1946 flush_warnings(warn);
1948 EXPORT_SYMBOL(dquot_free_inode);
1951 * Transfer the number of inode and blocks from one diskquota to an other.
1952 * On success, dquot references in transfer_to are consumed and references
1953 * to original dquots that need to be released are placed there. On failure,
1954 * references are kept untouched.
1956 * This operation can block, but only after everything is updated
1957 * A transaction must be started when entering this function.
1959 * We are holding reference on transfer_from & transfer_to, no need to
1960 * protect them by srcu_read_lock().
1962 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1965 qsize_t rsv_space = 0;
1966 qsize_t inode_usage = 1;
1967 struct dquot *transfer_from[MAXQUOTAS] = {};
1969 char is_valid[MAXQUOTAS] = {};
1970 struct dquot_warn warn_to[MAXQUOTAS];
1971 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1972 struct dquot_warn warn_from_space[MAXQUOTAS];
1974 if (IS_NOQUOTA(inode))
1977 if (inode->i_sb->dq_op->get_inode_usage) {
1978 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1983 /* Initialize the arrays */
1984 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1985 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1986 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1987 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1990 spin_lock(&dq_data_lock);
1991 spin_lock(&inode->i_lock);
1992 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1993 spin_unlock(&inode->i_lock);
1994 spin_unlock(&dq_data_lock);
1997 cur_space = __inode_get_bytes(inode);
1998 rsv_space = __inode_get_rsv_space(inode);
2000 * Build the transfer_from list, check limits, and update usage in
2001 * the target structures.
2003 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2005 * Skip changes for same uid or gid or for turned off quota-type.
2007 if (!transfer_to[cnt])
2009 /* Avoid races with quotaoff() */
2010 if (!sb_has_quota_active(inode->i_sb, cnt))
2013 transfer_from[cnt] = i_dquot(inode)[cnt];
2014 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2018 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2019 DQUOT_SPACE_WARN, &warn_to[cnt]);
2021 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2022 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2023 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2028 /* Decrease usage for source structures and update quota pointers */
2029 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2032 /* Due to IO error we might not have transfer_from[] structure */
2033 if (transfer_from[cnt]) {
2036 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2037 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2038 if (wtype != QUOTA_NL_NOWARN)
2039 prepare_warning(&warn_from_inodes[cnt],
2040 transfer_from[cnt], wtype);
2041 wtype = info_bdq_free(transfer_from[cnt],
2042 cur_space + rsv_space);
2043 if (wtype != QUOTA_NL_NOWARN)
2044 prepare_warning(&warn_from_space[cnt],
2045 transfer_from[cnt], wtype);
2046 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2047 dquot_decr_space(transfer_from[cnt], cur_space);
2048 dquot_free_reserved_space(transfer_from[cnt],
2050 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2052 i_dquot(inode)[cnt] = transfer_to[cnt];
2054 spin_unlock(&inode->i_lock);
2055 spin_unlock(&dq_data_lock);
2057 mark_all_dquot_dirty(transfer_from);
2058 mark_all_dquot_dirty(transfer_to);
2059 flush_warnings(warn_to);
2060 flush_warnings(warn_from_inodes);
2061 flush_warnings(warn_from_space);
2062 /* Pass back references to put */
2063 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2065 transfer_to[cnt] = transfer_from[cnt];
2068 /* Back out changes we already did */
2069 for (cnt--; cnt >= 0; cnt--) {
2072 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2073 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2074 dquot_decr_space(transfer_to[cnt], cur_space);
2075 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2076 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2078 spin_unlock(&inode->i_lock);
2079 spin_unlock(&dq_data_lock);
2080 flush_warnings(warn_to);
2083 EXPORT_SYMBOL(__dquot_transfer);
2085 /* Wrapper for transferring ownership of an inode for uid/gid only
2086 * Called from FSXXX_setattr()
2088 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2090 struct dquot *transfer_to[MAXQUOTAS] = {};
2091 struct dquot *dquot;
2092 struct super_block *sb = inode->i_sb;
2095 if (!dquot_active(inode))
2098 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2099 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2100 if (IS_ERR(dquot)) {
2101 if (PTR_ERR(dquot) != -ESRCH) {
2102 ret = PTR_ERR(dquot);
2107 transfer_to[USRQUOTA] = dquot;
2109 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2110 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2111 if (IS_ERR(dquot)) {
2112 if (PTR_ERR(dquot) != -ESRCH) {
2113 ret = PTR_ERR(dquot);
2118 transfer_to[GRPQUOTA] = dquot;
2120 ret = __dquot_transfer(inode, transfer_to);
2122 dqput_all(transfer_to);
2125 EXPORT_SYMBOL(dquot_transfer);
2128 * Write info of quota file to disk
2130 int dquot_commit_info(struct super_block *sb, int type)
2132 struct quota_info *dqopt = sb_dqopt(sb);
2134 return dqopt->ops[type]->write_file_info(sb, type);
2136 EXPORT_SYMBOL(dquot_commit_info);
2138 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2140 struct quota_info *dqopt = sb_dqopt(sb);
2142 if (!sb_has_quota_active(sb, qid->type))
2144 if (!dqopt->ops[qid->type]->get_next_id)
2146 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2148 EXPORT_SYMBOL(dquot_get_next_id);
2151 * Definitions of diskquota operations.
2153 const struct dquot_operations dquot_operations = {
2154 .write_dquot = dquot_commit,
2155 .acquire_dquot = dquot_acquire,
2156 .release_dquot = dquot_release,
2157 .mark_dirty = dquot_mark_dquot_dirty,
2158 .write_info = dquot_commit_info,
2159 .alloc_dquot = dquot_alloc,
2160 .destroy_dquot = dquot_destroy,
2161 .get_next_id = dquot_get_next_id,
2163 EXPORT_SYMBOL(dquot_operations);
2166 * Generic helper for ->open on filesystems supporting disk quotas.
2168 int dquot_file_open(struct inode *inode, struct file *file)
2172 error = generic_file_open(inode, file);
2173 if (!error && (file->f_mode & FMODE_WRITE))
2174 error = dquot_initialize(inode);
2177 EXPORT_SYMBOL(dquot_file_open);
2179 static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2181 struct quota_info *dqopt = sb_dqopt(sb);
2182 struct inode *inode = dqopt->files[type];
2186 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2188 inode->i_flags &= ~S_NOQUOTA;
2189 inode_unlock(inode);
2191 dqopt->files[type] = NULL;
2196 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2198 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2201 struct quota_info *dqopt = sb_dqopt(sb);
2203 /* s_umount should be held in exclusive mode */
2204 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2205 up_read(&sb->s_umount);
2207 /* Cannot turn off usage accounting without turning off limits, or
2208 * suspend quotas and simultaneously turn quotas off. */
2209 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2210 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2211 DQUOT_USAGE_ENABLED)))
2215 * Skip everything if there's nothing to do. We have to do this because
2216 * sometimes we are called when fill_super() failed and calling
2217 * sync_fs() in such cases does no good.
2219 if (!sb_any_quota_loaded(sb))
2222 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2223 if (type != -1 && cnt != type)
2225 if (!sb_has_quota_loaded(sb, cnt))
2228 if (flags & DQUOT_SUSPENDED) {
2229 spin_lock(&dq_state_lock);
2231 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2232 spin_unlock(&dq_state_lock);
2234 spin_lock(&dq_state_lock);
2235 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2236 /* Turning off suspended quotas? */
2237 if (!sb_has_quota_loaded(sb, cnt) &&
2238 sb_has_quota_suspended(sb, cnt)) {
2239 dqopt->flags &= ~dquot_state_flag(
2240 DQUOT_SUSPENDED, cnt);
2241 spin_unlock(&dq_state_lock);
2242 vfs_cleanup_quota_inode(sb, cnt);
2245 spin_unlock(&dq_state_lock);
2248 /* We still have to keep quota loaded? */
2249 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2252 /* Note: these are blocking operations */
2253 drop_dquot_ref(sb, cnt);
2254 invalidate_dquots(sb, cnt);
2256 * Now all dquots should be invalidated, all writes done so we
2257 * should be only users of the info. No locks needed.
2259 if (info_dirty(&dqopt->info[cnt]))
2260 sb->dq_op->write_info(sb, cnt);
2261 if (dqopt->ops[cnt]->free_file_info)
2262 dqopt->ops[cnt]->free_file_info(sb, cnt);
2263 put_quota_format(dqopt->info[cnt].dqi_format);
2264 dqopt->info[cnt].dqi_flags = 0;
2265 dqopt->info[cnt].dqi_igrace = 0;
2266 dqopt->info[cnt].dqi_bgrace = 0;
2267 dqopt->ops[cnt] = NULL;
2270 /* Skip syncing and setting flags if quota files are hidden */
2271 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2274 /* Sync the superblock so that buffers with quota data are written to
2275 * disk (and so userspace sees correct data afterwards). */
2276 if (sb->s_op->sync_fs)
2277 sb->s_op->sync_fs(sb, 1);
2278 sync_blockdev(sb->s_bdev);
2279 /* Now the quota files are just ordinary files and we can set the
2280 * inode flags back. Moreover we discard the pagecache so that
2281 * userspace sees the writes we did bypassing the pagecache. We
2282 * must also discard the blockdev buffers so that we see the
2283 * changes done by userspace on the next quotaon() */
2284 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2285 if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2286 inode_lock(dqopt->files[cnt]);
2287 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2288 inode_unlock(dqopt->files[cnt]);
2291 invalidate_bdev(sb->s_bdev);
2293 /* We are done when suspending quotas */
2294 if (flags & DQUOT_SUSPENDED)
2297 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2298 if (!sb_has_quota_loaded(sb, cnt))
2299 vfs_cleanup_quota_inode(sb, cnt);
2302 EXPORT_SYMBOL(dquot_disable);
2304 int dquot_quota_off(struct super_block *sb, int type)
2306 return dquot_disable(sb, type,
2307 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2309 EXPORT_SYMBOL(dquot_quota_off);
2312 * Turn quotas on on a device
2315 static int vfs_setup_quota_inode(struct inode *inode, int type)
2317 struct super_block *sb = inode->i_sb;
2318 struct quota_info *dqopt = sb_dqopt(sb);
2320 if (!S_ISREG(inode->i_mode))
2322 if (IS_RDONLY(inode))
2324 if (sb_has_quota_loaded(sb, type))
2327 dqopt->files[type] = igrab(inode);
2328 if (!dqopt->files[type])
2330 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2331 /* We don't want quota and atime on quota files (deadlocks
2332 * possible) Also nobody should write to the file - we use
2333 * special IO operations which ignore the immutable bit. */
2335 inode->i_flags |= S_NOQUOTA;
2336 inode_unlock(inode);
2338 * When S_NOQUOTA is set, remove dquot references as no more
2339 * references can be added
2341 __dquot_drop(inode);
2346 int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2349 struct quota_format_type *fmt = find_quota_format(format_id);
2350 struct quota_info *dqopt = sb_dqopt(sb);
2353 /* Just unsuspend quotas? */
2354 BUG_ON(flags & DQUOT_SUSPENDED);
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);
2361 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2362 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2366 /* Filesystems outside of init_user_ns not yet supported */
2367 if (sb->s_user_ns != &init_user_ns) {
2371 /* Usage always has to be set... */
2372 if (!(flags & DQUOT_USAGE_ENABLED)) {
2376 if (sb_has_quota_loaded(sb, type)) {
2381 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2382 /* As we bypass the pagecache we must now flush all the
2383 * dirty data and invalidate caches so that kernel sees
2384 * changes from userspace. It is not enough to just flush
2385 * the quota file since if blocksize < pagesize, invalidation
2386 * of the cache could fail because of other unrelated dirty
2388 sync_filesystem(sb);
2389 invalidate_bdev(sb->s_bdev);
2393 if (!fmt->qf_ops->check_quota_file(sb, type))
2396 dqopt->ops[type] = fmt->qf_ops;
2397 dqopt->info[type].dqi_format = fmt;
2398 dqopt->info[type].dqi_fmt_id = format_id;
2399 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2400 error = dqopt->ops[type]->read_file_info(sb, type);
2403 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2404 spin_lock(&dq_data_lock);
2405 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2406 spin_unlock(&dq_data_lock);
2408 spin_lock(&dq_state_lock);
2409 dqopt->flags |= dquot_state_flag(flags, type);
2410 spin_unlock(&dq_state_lock);
2412 error = add_dquot_ref(sb, type);
2414 dquot_disable(sb, type, flags);
2418 put_quota_format(fmt);
2422 EXPORT_SYMBOL(dquot_load_quota_sb);
2425 * More powerful function for turning on quotas on given quota inode allowing
2426 * setting of individual quota flags
2428 int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2433 err = vfs_setup_quota_inode(inode, type);
2436 err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2438 vfs_cleanup_quota_inode(inode->i_sb, type);
2441 EXPORT_SYMBOL(dquot_load_quota_inode);
2443 /* Reenable quotas on remount RW */
2444 int dquot_resume(struct super_block *sb, int type)
2446 struct quota_info *dqopt = sb_dqopt(sb);
2450 /* s_umount should be held in exclusive mode */
2451 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2452 up_read(&sb->s_umount);
2454 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2455 if (type != -1 && cnt != type)
2457 if (!sb_has_quota_suspended(sb, cnt))
2460 spin_lock(&dq_state_lock);
2461 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2462 DQUOT_LIMITS_ENABLED,
2464 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2465 spin_unlock(&dq_state_lock);
2467 flags = dquot_generic_flag(flags, cnt);
2468 ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2471 vfs_cleanup_quota_inode(sb, cnt);
2476 EXPORT_SYMBOL(dquot_resume);
2478 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2479 const struct path *path)
2481 int error = security_quota_on(path->dentry);
2484 /* Quota file not on the same filesystem? */
2485 if (path->dentry->d_sb != sb)
2488 error = dquot_load_quota_inode(d_inode(path->dentry), type,
2489 format_id, DQUOT_USAGE_ENABLED |
2490 DQUOT_LIMITS_ENABLED);
2493 EXPORT_SYMBOL(dquot_quota_on);
2496 * This function is used when filesystem needs to initialize quotas
2497 * during mount time.
2499 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2500 int format_id, int type)
2502 struct dentry *dentry;
2505 dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2507 return PTR_ERR(dentry);
2509 error = security_quota_on(dentry);
2511 error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2512 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2517 EXPORT_SYMBOL(dquot_quota_on_mount);
2519 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2523 struct quota_info *dqopt = sb_dqopt(sb);
2525 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2527 /* Accounting cannot be turned on while fs is mounted */
2528 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2531 for (type = 0; type < MAXQUOTAS; type++) {
2532 if (!(flags & qtype_enforce_flag(type)))
2534 /* Can't enforce without accounting */
2535 if (!sb_has_quota_usage_enabled(sb, type)) {
2539 if (sb_has_quota_limits_enabled(sb, type)) {
2543 spin_lock(&dq_state_lock);
2544 dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2545 spin_unlock(&dq_state_lock);
2549 /* Backout enforcement enablement we already did */
2550 for (type--; type >= 0; type--) {
2551 if (flags & qtype_enforce_flag(type))
2552 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2554 /* Error code translation for better compatibility with XFS */
2560 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2564 struct quota_info *dqopt = sb_dqopt(sb);
2566 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2569 * We don't support turning off accounting via quotactl. In principle
2570 * quota infrastructure can do this but filesystems don't expect
2571 * userspace to be able to do it.
2574 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2577 /* Filter out limits not enabled */
2578 for (type = 0; type < MAXQUOTAS; type++)
2579 if (!sb_has_quota_limits_enabled(sb, type))
2580 flags &= ~qtype_enforce_flag(type);
2584 for (type = 0; type < MAXQUOTAS; type++) {
2585 if (flags & qtype_enforce_flag(type)) {
2586 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2593 /* Backout enforcement disabling we already did */
2594 for (type--; type >= 0; type--) {
2595 if (flags & qtype_enforce_flag(type)) {
2596 spin_lock(&dq_state_lock);
2598 dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2599 spin_unlock(&dq_state_lock);
2605 /* Generic routine for getting common part of quota structure */
2606 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2608 struct mem_dqblk *dm = &dquot->dq_dqb;
2610 memset(di, 0, sizeof(*di));
2611 spin_lock(&dquot->dq_dqb_lock);
2612 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2613 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2614 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2615 di->d_ino_softlimit = dm->dqb_isoftlimit;
2616 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2617 di->d_ino_count = dm->dqb_curinodes;
2618 di->d_spc_timer = dm->dqb_btime;
2619 di->d_ino_timer = dm->dqb_itime;
2620 spin_unlock(&dquot->dq_dqb_lock);
2623 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2624 struct qc_dqblk *di)
2626 struct dquot *dquot;
2628 dquot = dqget(sb, qid);
2630 return PTR_ERR(dquot);
2631 do_get_dqblk(dquot, di);
2636 EXPORT_SYMBOL(dquot_get_dqblk);
2638 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2639 struct qc_dqblk *di)
2641 struct dquot *dquot;
2644 if (!sb->dq_op->get_next_id)
2646 err = sb->dq_op->get_next_id(sb, qid);
2649 dquot = dqget(sb, *qid);
2651 return PTR_ERR(dquot);
2652 do_get_dqblk(dquot, di);
2657 EXPORT_SYMBOL(dquot_get_next_dqblk);
2659 #define VFS_QC_MASK \
2660 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2661 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2662 QC_SPC_TIMER | QC_INO_TIMER)
2664 /* Generic routine for setting common part of quota structure */
2665 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2667 struct mem_dqblk *dm = &dquot->dq_dqb;
2668 int check_blim = 0, check_ilim = 0;
2669 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2671 if (di->d_fieldmask & ~VFS_QC_MASK)
2674 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2675 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2676 ((di->d_fieldmask & QC_SPC_HARD) &&
2677 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2678 ((di->d_fieldmask & QC_INO_SOFT) &&
2679 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2680 ((di->d_fieldmask & QC_INO_HARD) &&
2681 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2684 spin_lock(&dquot->dq_dqb_lock);
2685 if (di->d_fieldmask & QC_SPACE) {
2686 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2688 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2691 if (di->d_fieldmask & QC_SPC_SOFT)
2692 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2693 if (di->d_fieldmask & QC_SPC_HARD)
2694 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2695 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2697 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2700 if (di->d_fieldmask & QC_INO_COUNT) {
2701 dm->dqb_curinodes = di->d_ino_count;
2703 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2706 if (di->d_fieldmask & QC_INO_SOFT)
2707 dm->dqb_isoftlimit = di->d_ino_softlimit;
2708 if (di->d_fieldmask & QC_INO_HARD)
2709 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2710 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2712 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2715 if (di->d_fieldmask & QC_SPC_TIMER) {
2716 dm->dqb_btime = di->d_spc_timer;
2718 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2721 if (di->d_fieldmask & QC_INO_TIMER) {
2722 dm->dqb_itime = di->d_ino_timer;
2724 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2728 if (!dm->dqb_bsoftlimit ||
2729 dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2731 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2732 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2733 /* Set grace only if user hasn't provided his own... */
2734 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2737 if (!dm->dqb_isoftlimit ||
2738 dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2740 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2741 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2742 /* Set grace only if user hasn't provided his own... */
2743 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2745 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2747 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2749 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2750 spin_unlock(&dquot->dq_dqb_lock);
2751 mark_dquot_dirty(dquot);
2756 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2757 struct qc_dqblk *di)
2759 struct dquot *dquot;
2762 dquot = dqget(sb, qid);
2763 if (IS_ERR(dquot)) {
2764 rc = PTR_ERR(dquot);
2767 rc = do_set_dqblk(dquot, di);
2772 EXPORT_SYMBOL(dquot_set_dqblk);
2774 /* Generic routine for getting common part of quota file information */
2775 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2777 struct mem_dqinfo *mi;
2778 struct qc_type_state *tstate;
2779 struct quota_info *dqopt = sb_dqopt(sb);
2782 memset(state, 0, sizeof(*state));
2783 for (type = 0; type < MAXQUOTAS; type++) {
2784 if (!sb_has_quota_active(sb, type))
2786 tstate = state->s_state + type;
2787 mi = sb_dqopt(sb)->info + type;
2788 tstate->flags = QCI_ACCT_ENABLED;
2789 spin_lock(&dq_data_lock);
2790 if (mi->dqi_flags & DQF_SYS_FILE)
2791 tstate->flags |= QCI_SYSFILE;
2792 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2793 tstate->flags |= QCI_ROOT_SQUASH;
2794 if (sb_has_quota_limits_enabled(sb, type))
2795 tstate->flags |= QCI_LIMITS_ENFORCED;
2796 tstate->spc_timelimit = mi->dqi_bgrace;
2797 tstate->ino_timelimit = mi->dqi_igrace;
2798 if (dqopt->files[type]) {
2799 tstate->ino = dqopt->files[type]->i_ino;
2800 tstate->blocks = dqopt->files[type]->i_blocks;
2802 tstate->nextents = 1; /* We don't know... */
2803 spin_unlock(&dq_data_lock);
2807 EXPORT_SYMBOL(dquot_get_state);
2809 /* Generic routine for setting common part of quota file information */
2810 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2812 struct mem_dqinfo *mi;
2815 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2816 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2818 if (!sb_has_quota_active(sb, type))
2820 mi = sb_dqopt(sb)->info + type;
2821 if (ii->i_fieldmask & QC_FLAGS) {
2822 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2823 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2826 spin_lock(&dq_data_lock);
2827 if (ii->i_fieldmask & QC_SPC_TIMER)
2828 mi->dqi_bgrace = ii->i_spc_timelimit;
2829 if (ii->i_fieldmask & QC_INO_TIMER)
2830 mi->dqi_igrace = ii->i_ino_timelimit;
2831 if (ii->i_fieldmask & QC_FLAGS) {
2832 if (ii->i_flags & QCI_ROOT_SQUASH)
2833 mi->dqi_flags |= DQF_ROOT_SQUASH;
2835 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2837 spin_unlock(&dq_data_lock);
2838 mark_info_dirty(sb, type);
2839 /* Force write to disk */
2840 sb->dq_op->write_info(sb, type);
2843 EXPORT_SYMBOL(dquot_set_dqinfo);
2845 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2846 .quota_enable = dquot_quota_enable,
2847 .quota_disable = dquot_quota_disable,
2848 .quota_sync = dquot_quota_sync,
2849 .get_state = dquot_get_state,
2850 .set_info = dquot_set_dqinfo,
2851 .get_dqblk = dquot_get_dqblk,
2852 .get_nextdqblk = dquot_get_next_dqblk,
2853 .set_dqblk = dquot_set_dqblk
2855 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2857 static int do_proc_dqstats(struct ctl_table *table, int write,
2858 void *buffer, size_t *lenp, loff_t *ppos)
2860 unsigned int type = (unsigned long *)table->data - dqstats.stat;
2861 s64 value = percpu_counter_sum(&dqstats.counter[type]);
2863 /* Filter negative values for non-monotonic counters */
2864 if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2865 type == DQST_FREE_DQUOTS))
2868 /* Update global table */
2869 dqstats.stat[type] = value;
2870 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2873 static struct ctl_table fs_dqstats_table[] = {
2875 .procname = "lookups",
2876 .data = &dqstats.stat[DQST_LOOKUPS],
2877 .maxlen = sizeof(unsigned long),
2879 .proc_handler = do_proc_dqstats,
2882 .procname = "drops",
2883 .data = &dqstats.stat[DQST_DROPS],
2884 .maxlen = sizeof(unsigned long),
2886 .proc_handler = do_proc_dqstats,
2889 .procname = "reads",
2890 .data = &dqstats.stat[DQST_READS],
2891 .maxlen = sizeof(unsigned long),
2893 .proc_handler = do_proc_dqstats,
2896 .procname = "writes",
2897 .data = &dqstats.stat[DQST_WRITES],
2898 .maxlen = sizeof(unsigned long),
2900 .proc_handler = do_proc_dqstats,
2903 .procname = "cache_hits",
2904 .data = &dqstats.stat[DQST_CACHE_HITS],
2905 .maxlen = sizeof(unsigned long),
2907 .proc_handler = do_proc_dqstats,
2910 .procname = "allocated_dquots",
2911 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2912 .maxlen = sizeof(unsigned long),
2914 .proc_handler = do_proc_dqstats,
2917 .procname = "free_dquots",
2918 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2919 .maxlen = sizeof(unsigned long),
2921 .proc_handler = do_proc_dqstats,
2924 .procname = "syncs",
2925 .data = &dqstats.stat[DQST_SYNCS],
2926 .maxlen = sizeof(unsigned long),
2928 .proc_handler = do_proc_dqstats,
2930 #ifdef CONFIG_PRINT_QUOTA_WARNING
2932 .procname = "warnings",
2933 .data = &flag_print_warnings,
2934 .maxlen = sizeof(int),
2936 .proc_handler = proc_dointvec,
2942 static struct ctl_table fs_table[] = {
2944 .procname = "quota",
2946 .child = fs_dqstats_table,
2951 static struct ctl_table sys_table[] = {
2960 static int __init dquot_init(void)
2963 unsigned long nr_hash, order;
2965 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2967 register_sysctl_table(sys_table);
2969 dquot_cachep = kmem_cache_create("dquot",
2970 sizeof(struct dquot), sizeof(unsigned long) * 4,
2971 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2972 SLAB_MEM_SPREAD|SLAB_PANIC),
2976 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2978 panic("Cannot create dquot hash table");
2980 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2981 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2983 panic("Cannot create dquot stat counters");
2986 /* Find power-of-two hlist_heads which can fit into allocation */
2987 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2988 dq_hash_bits = ilog2(nr_hash);
2990 nr_hash = 1UL << dq_hash_bits;
2991 dq_hash_mask = nr_hash - 1;
2992 for (i = 0; i < nr_hash; i++)
2993 INIT_HLIST_HEAD(dquot_hash + i);
2995 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2996 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2998 if (register_shrinker(&dqcache_shrinker))
2999 panic("Cannot register dquot shrinker");
3003 fs_initcall(dquot_init);