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.
17 * Revised list management to avoid races
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.
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
41 * Added dynamic quota structure allocation
44 * Rewritten quota interface. Implemented new quota format and
45 * formats registering.
51 * Added journalled quota support, fix lock inversion problems
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 five lists for dquot management: the inuse_list,
229 * releasing_dquots, free_dquots, dqi_dirty_list, and dquot_hash[] array.
230 * A single dquot structure may be on some of those lists, depending on
233 * All dquots are placed to the end of inuse_list when first created, and this
234 * list is used for invalidate operation, which must look at every dquot.
236 * When the last reference of a dquot is dropped, the dquot is added to
237 * releasing_dquots. We'll then queue work item which will call
238 * synchronize_srcu() and after that perform the final cleanup of all the
239 * dquots on the list. Each cleaned up dquot is moved to free_dquots list.
240 * Both releasing_dquots and free_dquots use the dq_free list_head in the dquot
243 * Unused and cleaned up dquots are in the free_dquots list and this list is
244 * searched whenever we need an available dquot. Dquots are removed from the
245 * list as soon as they are used again and dqstats.free_dquots gives the number
246 * of dquots on the list. When dquot is invalidated it's completely released
249 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
250 * dirtied, and this list is searched when writing dirty dquots back to
251 * quota file. Note that some filesystems do dirty dquot tracking on their
252 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
254 * Dquots with a specific identity (device, type and id) are placed on
255 * one of the dquot_hash[] hash chains. The provides an efficient search
256 * mechanism to locate a specific dquot.
259 static LIST_HEAD(inuse_list);
260 static LIST_HEAD(free_dquots);
261 static LIST_HEAD(releasing_dquots);
262 static unsigned int dq_hash_bits, dq_hash_mask;
263 static struct hlist_head *dquot_hash;
265 struct dqstats dqstats;
266 EXPORT_SYMBOL(dqstats);
268 static qsize_t inode_get_rsv_space(struct inode *inode);
269 static qsize_t __inode_get_rsv_space(struct inode *inode);
270 static int __dquot_initialize(struct inode *inode, int type);
272 static void quota_release_workfn(struct work_struct *work);
273 static DECLARE_DELAYED_WORK(quota_release_work, quota_release_workfn);
275 static inline unsigned int
276 hashfn(const struct super_block *sb, struct kqid qid)
278 unsigned int id = from_kqid(&init_user_ns, qid);
282 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
283 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
287 * Following list functions expect dq_list_lock to be held
289 static inline void insert_dquot_hash(struct dquot *dquot)
291 struct hlist_head *head;
292 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
293 hlist_add_head(&dquot->dq_hash, head);
296 static inline void remove_dquot_hash(struct dquot *dquot)
298 hlist_del_init(&dquot->dq_hash);
301 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
306 hlist_for_each_entry(dquot, dquot_hash+hashent, dq_hash)
307 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
313 /* Add a dquot to the tail of the free list */
314 static inline void put_dquot_last(struct dquot *dquot)
316 list_add_tail(&dquot->dq_free, &free_dquots);
317 dqstats_inc(DQST_FREE_DQUOTS);
320 static inline void put_releasing_dquots(struct dquot *dquot)
322 list_add_tail(&dquot->dq_free, &releasing_dquots);
323 set_bit(DQ_RELEASING_B, &dquot->dq_flags);
326 static inline void remove_free_dquot(struct dquot *dquot)
328 if (list_empty(&dquot->dq_free))
330 list_del_init(&dquot->dq_free);
331 if (!test_bit(DQ_RELEASING_B, &dquot->dq_flags))
332 dqstats_dec(DQST_FREE_DQUOTS);
334 clear_bit(DQ_RELEASING_B, &dquot->dq_flags);
337 static inline void put_inuse(struct dquot *dquot)
339 /* We add to the back of inuse list so we don't have to restart
340 * when traversing this list and we block */
341 list_add_tail(&dquot->dq_inuse, &inuse_list);
342 dqstats_inc(DQST_ALLOC_DQUOTS);
345 static inline void remove_inuse(struct dquot *dquot)
347 dqstats_dec(DQST_ALLOC_DQUOTS);
348 list_del(&dquot->dq_inuse);
351 * End of list functions needing dq_list_lock
354 static void wait_on_dquot(struct dquot *dquot)
356 mutex_lock(&dquot->dq_lock);
357 mutex_unlock(&dquot->dq_lock);
360 static inline int dquot_active(struct dquot *dquot)
362 return test_bit(DQ_ACTIVE_B, &dquot->dq_flags);
365 static inline int dquot_dirty(struct dquot *dquot)
367 return test_bit(DQ_MOD_B, &dquot->dq_flags);
370 static inline int mark_dquot_dirty(struct dquot *dquot)
372 return dquot->dq_sb->dq_op->mark_dirty(dquot);
375 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
376 int dquot_mark_dquot_dirty(struct dquot *dquot)
380 if (!dquot_active(dquot))
383 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
384 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
386 /* If quota is dirty already, we don't have to acquire dq_list_lock */
387 if (dquot_dirty(dquot))
390 spin_lock(&dq_list_lock);
391 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
392 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
393 info[dquot->dq_id.type].dqi_dirty_list);
396 spin_unlock(&dq_list_lock);
399 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
401 /* Dirtify all the dquots - this can block when journalling */
402 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
407 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
409 /* Even in case of error we have to continue */
410 ret = mark_dquot_dirty(dquot[cnt]);
417 static inline void dqput_all(struct dquot **dquot)
421 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
425 static inline int clear_dquot_dirty(struct dquot *dquot)
427 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
428 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
430 spin_lock(&dq_list_lock);
431 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
432 spin_unlock(&dq_list_lock);
435 list_del_init(&dquot->dq_dirty);
436 spin_unlock(&dq_list_lock);
440 void mark_info_dirty(struct super_block *sb, int type)
442 spin_lock(&dq_data_lock);
443 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
444 spin_unlock(&dq_data_lock);
446 EXPORT_SYMBOL(mark_info_dirty);
449 * Read dquot from disk and alloc space for it
452 int dquot_acquire(struct dquot *dquot)
454 int ret = 0, ret2 = 0;
455 unsigned int memalloc;
456 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
458 mutex_lock(&dquot->dq_lock);
459 memalloc = memalloc_nofs_save();
460 if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
461 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
465 /* Make sure flags update is visible after dquot has been filled */
466 smp_mb__before_atomic();
467 set_bit(DQ_READ_B, &dquot->dq_flags);
468 /* Instantiate dquot if needed */
469 if (!dquot_active(dquot) && !dquot->dq_off) {
470 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
471 /* Write the info if needed */
472 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
473 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
474 dquot->dq_sb, dquot->dq_id.type);
484 * Make sure flags update is visible after on-disk struct has been
485 * allocated. Paired with smp_rmb() in dqget().
487 smp_mb__before_atomic();
488 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
490 memalloc_nofs_restore(memalloc);
491 mutex_unlock(&dquot->dq_lock);
494 EXPORT_SYMBOL(dquot_acquire);
497 * Write dquot to disk
499 int dquot_commit(struct dquot *dquot)
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 if (!clear_dquot_dirty(dquot))
509 /* Inactive dquot can be only if there was error during read/init
510 * => we have better not writing it */
511 if (dquot_active(dquot))
512 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
516 memalloc_nofs_restore(memalloc);
517 mutex_unlock(&dquot->dq_lock);
520 EXPORT_SYMBOL(dquot_commit);
525 int dquot_release(struct dquot *dquot)
527 int ret = 0, ret2 = 0;
528 unsigned int memalloc;
529 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
531 mutex_lock(&dquot->dq_lock);
532 memalloc = memalloc_nofs_save();
533 /* Check whether we are not racing with some other dqget() */
534 if (dquot_is_busy(dquot))
536 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
537 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
539 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
540 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
541 dquot->dq_sb, dquot->dq_id.type);
546 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
548 memalloc_nofs_restore(memalloc);
549 mutex_unlock(&dquot->dq_lock);
552 EXPORT_SYMBOL(dquot_release);
554 void dquot_destroy(struct dquot *dquot)
556 kmem_cache_free(dquot_cachep, dquot);
558 EXPORT_SYMBOL(dquot_destroy);
560 static inline void do_destroy_dquot(struct dquot *dquot)
562 dquot->dq_sb->dq_op->destroy_dquot(dquot);
565 /* Invalidate all dquots on the list. Note that this function is called after
566 * quota is disabled and pointers from inodes removed so there cannot be new
567 * quota users. There can still be some users of quotas due to inodes being
568 * just deleted or pruned by prune_icache() (those are not attached to any
569 * list) or parallel quotactl call. We have to wait for such users.
571 static void invalidate_dquots(struct super_block *sb, int type)
573 struct dquot *dquot, *tmp;
576 flush_delayed_work("a_release_work);
578 spin_lock(&dq_list_lock);
579 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
580 if (dquot->dq_sb != sb)
582 if (dquot->dq_id.type != type)
584 /* Wait for dquot users */
585 if (atomic_read(&dquot->dq_count)) {
586 atomic_inc(&dquot->dq_count);
587 spin_unlock(&dq_list_lock);
589 * Once dqput() wakes us up, we know it's time to free
591 * IMPORTANT: we rely on the fact that there is always
592 * at most one process waiting for dquot to free.
593 * Otherwise dq_count would be > 1 and we would never
596 wait_event(dquot_ref_wq,
597 atomic_read(&dquot->dq_count) == 1);
599 /* At this moment dquot() need not exist (it could be
600 * reclaimed by prune_dqcache(). Hence we must
605 * The last user already dropped its reference but dquot didn't
606 * get fully cleaned up yet. Restart the scan which flushes the
607 * work cleaning up released dquots.
609 if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
610 spin_unlock(&dq_list_lock);
614 * Quota now has no users and it has been written on last
617 remove_dquot_hash(dquot);
618 remove_free_dquot(dquot);
620 do_destroy_dquot(dquot);
622 spin_unlock(&dq_list_lock);
625 /* Call callback for every active dquot on given filesystem */
626 int dquot_scan_active(struct super_block *sb,
627 int (*fn)(struct dquot *dquot, unsigned long priv),
630 struct dquot *dquot, *old_dquot = NULL;
633 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
635 spin_lock(&dq_list_lock);
636 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
637 if (!dquot_active(dquot))
639 if (dquot->dq_sb != sb)
641 /* Now we have active dquot so we can just increase use count */
642 atomic_inc(&dquot->dq_count);
643 spin_unlock(&dq_list_lock);
647 * ->release_dquot() can be racing with us. Our reference
648 * protects us from new calls to it so just wait for any
649 * outstanding call and recheck the DQ_ACTIVE_B after that.
651 wait_on_dquot(dquot);
652 if (dquot_active(dquot)) {
653 ret = fn(dquot, priv);
657 spin_lock(&dq_list_lock);
658 /* We are safe to continue now because our dquot could not
659 * be moved out of the inuse list while we hold the reference */
661 spin_unlock(&dq_list_lock);
666 EXPORT_SYMBOL(dquot_scan_active);
668 static inline int dquot_write_dquot(struct dquot *dquot)
670 int ret = dquot->dq_sb->dq_op->write_dquot(dquot);
672 quota_error(dquot->dq_sb, "Can't write quota structure "
673 "(error %d). Quota may get out of sync!", ret);
674 /* Clear dirty bit anyway to avoid infinite loop. */
675 clear_dquot_dirty(dquot);
680 /* Write all dquot structures to quota files */
681 int dquot_writeback_dquots(struct super_block *sb, int type)
683 struct list_head dirty;
685 struct quota_info *dqopt = sb_dqopt(sb);
689 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
691 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
692 if (type != -1 && cnt != type)
694 if (!sb_has_quota_active(sb, cnt))
696 spin_lock(&dq_list_lock);
697 /* Move list away to avoid livelock. */
698 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
699 while (!list_empty(&dirty)) {
700 dquot = list_first_entry(&dirty, struct dquot,
703 WARN_ON(!dquot_active(dquot));
704 /* If the dquot is releasing we should not touch it */
705 if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
706 spin_unlock(&dq_list_lock);
707 flush_delayed_work("a_release_work);
708 spin_lock(&dq_list_lock);
712 /* Now we have active dquot from which someone is
713 * holding reference so we can safely just increase
716 spin_unlock(&dq_list_lock);
717 err = dquot_write_dquot(dquot);
721 spin_lock(&dq_list_lock);
723 spin_unlock(&dq_list_lock);
726 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
727 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
728 && info_dirty(&dqopt->info[cnt]))
729 sb->dq_op->write_info(sb, cnt);
730 dqstats_inc(DQST_SYNCS);
734 EXPORT_SYMBOL(dquot_writeback_dquots);
736 /* Write all dquot structures to disk and make them visible from userspace */
737 int dquot_quota_sync(struct super_block *sb, int type)
739 struct quota_info *dqopt = sb_dqopt(sb);
743 ret = dquot_writeback_dquots(sb, type);
746 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
749 /* This is not very clever (and fast) but currently I don't know about
750 * any other simple way of getting quota data to disk and we must get
751 * them there for userspace to be visible... */
752 if (sb->s_op->sync_fs) {
753 ret = sb->s_op->sync_fs(sb, 1);
757 ret = sync_blockdev(sb->s_bdev);
762 * Now when everything is written we can discard the pagecache so
763 * that userspace sees the changes.
765 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
766 if (type != -1 && cnt != type)
768 if (!sb_has_quota_active(sb, cnt))
770 inode_lock(dqopt->files[cnt]);
771 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
772 inode_unlock(dqopt->files[cnt]);
777 EXPORT_SYMBOL(dquot_quota_sync);
780 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
783 unsigned long freed = 0;
785 spin_lock(&dq_list_lock);
786 while (!list_empty(&free_dquots) && sc->nr_to_scan) {
787 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
788 remove_dquot_hash(dquot);
789 remove_free_dquot(dquot);
791 do_destroy_dquot(dquot);
795 spin_unlock(&dq_list_lock);
800 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
802 return vfs_pressure_ratio(
803 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
807 * Safely release dquot and put reference to dquot.
809 static void quota_release_workfn(struct work_struct *work)
812 struct list_head rls_head;
814 spin_lock(&dq_list_lock);
815 /* Exchange the list head to avoid livelock. */
816 list_replace_init(&releasing_dquots, &rls_head);
817 spin_unlock(&dq_list_lock);
818 synchronize_srcu(&dquot_srcu);
821 spin_lock(&dq_list_lock);
822 while (!list_empty(&rls_head)) {
823 dquot = list_first_entry(&rls_head, struct dquot, dq_free);
824 WARN_ON_ONCE(atomic_read(&dquot->dq_count));
826 * Note that DQ_RELEASING_B protects us from racing with
827 * invalidate_dquots() calls so we are safe to work with the
828 * dquot even after we drop dq_list_lock.
830 if (dquot_dirty(dquot)) {
831 spin_unlock(&dq_list_lock);
832 /* Commit dquot before releasing */
833 dquot_write_dquot(dquot);
836 if (dquot_active(dquot)) {
837 spin_unlock(&dq_list_lock);
838 dquot->dq_sb->dq_op->release_dquot(dquot);
841 /* Dquot is inactive and clean, now move it to free list */
842 remove_free_dquot(dquot);
843 put_dquot_last(dquot);
845 spin_unlock(&dq_list_lock);
849 * Put reference to dquot
851 void dqput(struct dquot *dquot)
855 #ifdef CONFIG_QUOTA_DEBUG
856 if (!atomic_read(&dquot->dq_count)) {
857 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
858 quotatypes[dquot->dq_id.type],
859 from_kqid(&init_user_ns, dquot->dq_id));
863 dqstats_inc(DQST_DROPS);
865 spin_lock(&dq_list_lock);
866 if (atomic_read(&dquot->dq_count) > 1) {
867 /* We have more than one user... nothing to do */
868 atomic_dec(&dquot->dq_count);
869 /* Releasing dquot during quotaoff phase? */
870 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
871 atomic_read(&dquot->dq_count) == 1)
872 wake_up(&dquot_ref_wq);
873 spin_unlock(&dq_list_lock);
877 /* Need to release dquot? */
878 #ifdef CONFIG_QUOTA_DEBUG
880 BUG_ON(!list_empty(&dquot->dq_free));
882 put_releasing_dquots(dquot);
883 atomic_dec(&dquot->dq_count);
884 spin_unlock(&dq_list_lock);
885 queue_delayed_work(system_unbound_wq, "a_release_work, 1);
887 EXPORT_SYMBOL(dqput);
889 struct dquot *dquot_alloc(struct super_block *sb, int type)
891 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
893 EXPORT_SYMBOL(dquot_alloc);
895 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
899 dquot = sb->dq_op->alloc_dquot(sb, type);
903 mutex_init(&dquot->dq_lock);
904 INIT_LIST_HEAD(&dquot->dq_free);
905 INIT_LIST_HEAD(&dquot->dq_inuse);
906 INIT_HLIST_NODE(&dquot->dq_hash);
907 INIT_LIST_HEAD(&dquot->dq_dirty);
909 dquot->dq_id = make_kqid_invalid(type);
910 atomic_set(&dquot->dq_count, 1);
911 spin_lock_init(&dquot->dq_dqb_lock);
917 * Get reference to dquot
919 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
920 * destroying our dquot by:
921 * a) checking for quota flags under dq_list_lock and
922 * b) getting a reference to dquot before we release dq_list_lock
924 struct dquot *dqget(struct super_block *sb, struct kqid qid)
926 unsigned int hashent = hashfn(sb, qid);
927 struct dquot *dquot, *empty = NULL;
929 if (!qid_has_mapping(sb->s_user_ns, qid))
930 return ERR_PTR(-EINVAL);
932 if (!sb_has_quota_active(sb, qid.type))
933 return ERR_PTR(-ESRCH);
935 spin_lock(&dq_list_lock);
936 spin_lock(&dq_state_lock);
937 if (!sb_has_quota_active(sb, qid.type)) {
938 spin_unlock(&dq_state_lock);
939 spin_unlock(&dq_list_lock);
940 dquot = ERR_PTR(-ESRCH);
943 spin_unlock(&dq_state_lock);
945 dquot = find_dquot(hashent, sb, qid);
948 spin_unlock(&dq_list_lock);
949 empty = get_empty_dquot(sb, qid.type);
951 schedule(); /* Try to wait for a moment... */
957 /* all dquots go on the inuse_list */
959 /* hash it first so it can be found */
960 insert_dquot_hash(dquot);
961 spin_unlock(&dq_list_lock);
962 dqstats_inc(DQST_LOOKUPS);
964 if (!atomic_read(&dquot->dq_count))
965 remove_free_dquot(dquot);
966 atomic_inc(&dquot->dq_count);
967 spin_unlock(&dq_list_lock);
968 dqstats_inc(DQST_CACHE_HITS);
969 dqstats_inc(DQST_LOOKUPS);
971 /* Wait for dq_lock - after this we know that either dquot_release() is
972 * already finished or it will be canceled due to dq_count > 0 test */
973 wait_on_dquot(dquot);
974 /* Read the dquot / allocate space in quota file */
975 if (!dquot_active(dquot)) {
978 err = sb->dq_op->acquire_dquot(dquot);
981 dquot = ERR_PTR(err);
986 * Make sure following reads see filled structure - paired with
987 * smp_mb__before_atomic() in dquot_acquire().
990 #ifdef CONFIG_QUOTA_DEBUG
991 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
995 do_destroy_dquot(empty);
999 EXPORT_SYMBOL(dqget);
1001 static inline struct dquot **i_dquot(struct inode *inode)
1003 return inode->i_sb->s_op->get_dquots(inode);
1006 static int dqinit_needed(struct inode *inode, int type)
1008 struct dquot * const *dquots;
1011 if (IS_NOQUOTA(inode))
1014 dquots = i_dquot(inode);
1016 return !dquots[type];
1017 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1023 /* This routine is guarded by s_umount semaphore */
1024 static int add_dquot_ref(struct super_block *sb, int type)
1026 struct inode *inode, *old_inode = NULL;
1027 #ifdef CONFIG_QUOTA_DEBUG
1032 spin_lock(&sb->s_inode_list_lock);
1033 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1034 spin_lock(&inode->i_lock);
1035 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1036 !atomic_read(&inode->i_writecount) ||
1037 !dqinit_needed(inode, type)) {
1038 spin_unlock(&inode->i_lock);
1042 spin_unlock(&inode->i_lock);
1043 spin_unlock(&sb->s_inode_list_lock);
1045 #ifdef CONFIG_QUOTA_DEBUG
1046 if (unlikely(inode_get_rsv_space(inode) > 0))
1050 err = __dquot_initialize(inode, type);
1057 * We hold a reference to 'inode' so it couldn't have been
1058 * removed from s_inodes list while we dropped the
1059 * s_inode_list_lock. We cannot iput the inode now as we can be
1060 * holding the last reference and we cannot iput it under
1061 * s_inode_list_lock. So we keep the reference and iput it
1066 spin_lock(&sb->s_inode_list_lock);
1068 spin_unlock(&sb->s_inode_list_lock);
1071 #ifdef CONFIG_QUOTA_DEBUG
1073 quota_error(sb, "Writes happened before quota was turned on "
1074 "thus quota information is probably inconsistent. "
1075 "Please run quotacheck(8)");
1081 static void remove_dquot_ref(struct super_block *sb, int type)
1083 struct inode *inode;
1084 #ifdef CONFIG_QUOTA_DEBUG
1088 spin_lock(&sb->s_inode_list_lock);
1089 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1091 * We have to scan also I_NEW inodes because they can already
1092 * have quota pointer initialized. Luckily, we need to touch
1093 * only quota pointers and these have separate locking
1096 spin_lock(&dq_data_lock);
1097 if (!IS_NOQUOTA(inode)) {
1098 struct dquot **dquots = i_dquot(inode);
1099 struct dquot *dquot = dquots[type];
1101 #ifdef CONFIG_QUOTA_DEBUG
1102 if (unlikely(inode_get_rsv_space(inode) > 0))
1105 dquots[type] = NULL;
1109 spin_unlock(&dq_data_lock);
1111 spin_unlock(&sb->s_inode_list_lock);
1112 #ifdef CONFIG_QUOTA_DEBUG
1114 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1115 " was disabled thus quota information is probably "
1116 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1121 /* Gather all references from inodes and drop them */
1122 static void drop_dquot_ref(struct super_block *sb, int type)
1125 remove_dquot_ref(sb, type);
1129 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1131 if (dquot->dq_dqb.dqb_rsvspace >= number)
1132 dquot->dq_dqb.dqb_rsvspace -= number;
1135 dquot->dq_dqb.dqb_rsvspace = 0;
1137 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1138 dquot->dq_dqb.dqb_bsoftlimit)
1139 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1140 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1143 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1145 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1146 dquot->dq_dqb.dqb_curinodes >= number)
1147 dquot->dq_dqb.dqb_curinodes -= number;
1149 dquot->dq_dqb.dqb_curinodes = 0;
1150 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1151 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1152 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1155 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1157 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1158 dquot->dq_dqb.dqb_curspace >= number)
1159 dquot->dq_dqb.dqb_curspace -= number;
1161 dquot->dq_dqb.dqb_curspace = 0;
1162 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1163 dquot->dq_dqb.dqb_bsoftlimit)
1164 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1165 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1169 struct super_block *w_sb;
1170 struct kqid w_dq_id;
1174 static int warning_issued(struct dquot *dquot, const int warntype)
1176 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1177 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1178 ((warntype == QUOTA_NL_IHARDWARN ||
1179 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1183 return test_and_set_bit(flag, &dquot->dq_flags);
1186 #ifdef CONFIG_PRINT_QUOTA_WARNING
1187 static int flag_print_warnings = 1;
1189 static int need_print_warning(struct dquot_warn *warn)
1191 if (!flag_print_warnings)
1194 switch (warn->w_dq_id.type) {
1196 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1198 return in_group_p(warn->w_dq_id.gid);
1205 /* Print warning to user which exceeded quota */
1206 static void print_warning(struct dquot_warn *warn)
1209 struct tty_struct *tty;
1210 int warntype = warn->w_type;
1212 if (warntype == QUOTA_NL_IHARDBELOW ||
1213 warntype == QUOTA_NL_ISOFTBELOW ||
1214 warntype == QUOTA_NL_BHARDBELOW ||
1215 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1218 tty = get_current_tty();
1221 tty_write_message(tty, warn->w_sb->s_id);
1222 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1223 tty_write_message(tty, ": warning, ");
1225 tty_write_message(tty, ": write failed, ");
1226 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1228 case QUOTA_NL_IHARDWARN:
1229 msg = " file limit reached.\r\n";
1231 case QUOTA_NL_ISOFTLONGWARN:
1232 msg = " file quota exceeded too long.\r\n";
1234 case QUOTA_NL_ISOFTWARN:
1235 msg = " file quota exceeded.\r\n";
1237 case QUOTA_NL_BHARDWARN:
1238 msg = " block limit reached.\r\n";
1240 case QUOTA_NL_BSOFTLONGWARN:
1241 msg = " block quota exceeded too long.\r\n";
1243 case QUOTA_NL_BSOFTWARN:
1244 msg = " block quota exceeded.\r\n";
1247 tty_write_message(tty, msg);
1252 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1255 if (warning_issued(dquot, warntype))
1257 warn->w_type = warntype;
1258 warn->w_sb = dquot->dq_sb;
1259 warn->w_dq_id = dquot->dq_id;
1263 * Write warnings to the console and send warning messages over netlink.
1265 * Note that this function can call into tty and networking code.
1267 static void flush_warnings(struct dquot_warn *warn)
1271 for (i = 0; i < MAXQUOTAS; i++) {
1272 if (warn[i].w_type == QUOTA_NL_NOWARN)
1274 #ifdef CONFIG_PRINT_QUOTA_WARNING
1275 print_warning(&warn[i]);
1277 quota_send_warning(warn[i].w_dq_id,
1278 warn[i].w_sb->s_dev, warn[i].w_type);
1282 static int ignore_hardlimit(struct dquot *dquot)
1284 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1286 return capable(CAP_SYS_RESOURCE) &&
1287 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1288 !(info->dqi_flags & DQF_ROOT_SQUASH));
1291 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1292 struct dquot_warn *warn)
1297 spin_lock(&dquot->dq_dqb_lock);
1298 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1299 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1300 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1303 if (dquot->dq_dqb.dqb_ihardlimit &&
1304 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1305 !ignore_hardlimit(dquot)) {
1306 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1311 if (dquot->dq_dqb.dqb_isoftlimit &&
1312 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1313 dquot->dq_dqb.dqb_itime &&
1314 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1315 !ignore_hardlimit(dquot)) {
1316 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1321 if (dquot->dq_dqb.dqb_isoftlimit &&
1322 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1323 dquot->dq_dqb.dqb_itime == 0) {
1324 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1325 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1326 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1329 dquot->dq_dqb.dqb_curinodes = newinodes;
1332 spin_unlock(&dquot->dq_dqb_lock);
1336 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1337 qsize_t rsv_space, unsigned int flags,
1338 struct dquot_warn *warn)
1341 struct super_block *sb = dquot->dq_sb;
1344 spin_lock(&dquot->dq_dqb_lock);
1345 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1346 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1349 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1350 + space + rsv_space;
1352 if (dquot->dq_dqb.dqb_bhardlimit &&
1353 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1354 !ignore_hardlimit(dquot)) {
1355 if (flags & DQUOT_SPACE_WARN)
1356 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1361 if (dquot->dq_dqb.dqb_bsoftlimit &&
1362 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1363 dquot->dq_dqb.dqb_btime &&
1364 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1365 !ignore_hardlimit(dquot)) {
1366 if (flags & DQUOT_SPACE_WARN)
1367 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1372 if (dquot->dq_dqb.dqb_bsoftlimit &&
1373 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1374 dquot->dq_dqb.dqb_btime == 0) {
1375 if (flags & DQUOT_SPACE_WARN) {
1376 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1377 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1378 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1381 * We don't allow preallocation to exceed softlimit so exceeding will
1390 * We have to be careful and go through warning generation & grace time
1391 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1394 if (flags & DQUOT_SPACE_NOFAIL)
1397 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1398 dquot->dq_dqb.dqb_curspace += space;
1400 spin_unlock(&dquot->dq_dqb_lock);
1404 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1408 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1409 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1410 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1411 return QUOTA_NL_NOWARN;
1413 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1414 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1415 return QUOTA_NL_ISOFTBELOW;
1416 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1417 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1418 return QUOTA_NL_IHARDBELOW;
1419 return QUOTA_NL_NOWARN;
1422 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1426 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1428 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1429 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1430 return QUOTA_NL_NOWARN;
1432 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1433 return QUOTA_NL_BSOFTBELOW;
1434 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1435 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1436 return QUOTA_NL_BHARDBELOW;
1437 return QUOTA_NL_NOWARN;
1440 static int inode_quota_active(const struct inode *inode)
1442 struct super_block *sb = inode->i_sb;
1444 if (IS_NOQUOTA(inode))
1446 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1450 * Initialize quota pointers in inode
1452 * It is better to call this function outside of any transaction as it
1453 * might need a lot of space in journal for dquot structure allocation.
1455 static int __dquot_initialize(struct inode *inode, int type)
1457 int cnt, init_needed = 0;
1458 struct dquot **dquots, *got[MAXQUOTAS] = {};
1459 struct super_block *sb = inode->i_sb;
1463 if (!inode_quota_active(inode))
1466 dquots = i_dquot(inode);
1468 /* First get references to structures we might need. */
1469 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1473 struct dquot *dquot;
1475 if (type != -1 && cnt != type)
1478 * The i_dquot should have been initialized in most cases,
1479 * we check it without locking here to avoid unnecessary
1480 * dqget()/dqput() calls.
1485 if (!sb_has_quota_active(sb, cnt))
1492 qid = make_kqid_uid(inode->i_uid);
1495 qid = make_kqid_gid(inode->i_gid);
1498 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1501 qid = make_kqid_projid(projid);
1504 dquot = dqget(sb, qid);
1505 if (IS_ERR(dquot)) {
1506 /* We raced with somebody turning quotas off... */
1507 if (PTR_ERR(dquot) != -ESRCH) {
1508 ret = PTR_ERR(dquot);
1516 /* All required i_dquot has been initialized */
1520 spin_lock(&dq_data_lock);
1521 if (IS_NOQUOTA(inode))
1523 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1524 if (type != -1 && cnt != type)
1526 /* Avoid races with quotaoff() */
1527 if (!sb_has_quota_active(sb, cnt))
1529 /* We could race with quotaon or dqget() could have failed */
1533 dquots[cnt] = got[cnt];
1536 * Make quota reservation system happy if someone
1537 * did a write before quota was turned on
1539 rsv = inode_get_rsv_space(inode);
1540 if (unlikely(rsv)) {
1541 spin_lock(&inode->i_lock);
1542 /* Get reservation again under proper lock */
1543 rsv = __inode_get_rsv_space(inode);
1544 spin_lock(&dquots[cnt]->dq_dqb_lock);
1545 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1546 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1547 spin_unlock(&inode->i_lock);
1552 spin_unlock(&dq_data_lock);
1554 /* Drop unused references */
1560 int dquot_initialize(struct inode *inode)
1562 return __dquot_initialize(inode, -1);
1564 EXPORT_SYMBOL(dquot_initialize);
1566 bool dquot_initialize_needed(struct inode *inode)
1568 struct dquot **dquots;
1571 if (!inode_quota_active(inode))
1574 dquots = i_dquot(inode);
1575 for (i = 0; i < MAXQUOTAS; i++)
1576 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1580 EXPORT_SYMBOL(dquot_initialize_needed);
1583 * Release all quotas referenced by inode.
1585 * This function only be called on inode free or converting
1586 * a file to quota file, no other users for the i_dquot in
1587 * both cases, so we needn't call synchronize_srcu() after
1590 static void __dquot_drop(struct inode *inode)
1593 struct dquot **dquots = i_dquot(inode);
1594 struct dquot *put[MAXQUOTAS];
1596 spin_lock(&dq_data_lock);
1597 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1598 put[cnt] = dquots[cnt];
1601 spin_unlock(&dq_data_lock);
1605 void dquot_drop(struct inode *inode)
1607 struct dquot * const *dquots;
1610 if (IS_NOQUOTA(inode))
1614 * Test before calling to rule out calls from proc and such
1615 * where we are not allowed to block. Note that this is
1616 * actually reliable test even without the lock - the caller
1617 * must assure that nobody can come after the DQUOT_DROP and
1618 * add quota pointers back anyway.
1620 dquots = i_dquot(inode);
1621 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1626 if (cnt < MAXQUOTAS)
1627 __dquot_drop(inode);
1629 EXPORT_SYMBOL(dquot_drop);
1632 * inode_reserved_space is managed internally by quota, and protected by
1633 * i_lock similar to i_blocks+i_bytes.
1635 static qsize_t *inode_reserved_space(struct inode * inode)
1637 /* Filesystem must explicitly define it's own method in order to use
1638 * quota reservation interface */
1639 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1640 return inode->i_sb->dq_op->get_reserved_space(inode);
1643 static qsize_t __inode_get_rsv_space(struct inode *inode)
1645 if (!inode->i_sb->dq_op->get_reserved_space)
1647 return *inode_reserved_space(inode);
1650 static qsize_t inode_get_rsv_space(struct inode *inode)
1654 if (!inode->i_sb->dq_op->get_reserved_space)
1656 spin_lock(&inode->i_lock);
1657 ret = __inode_get_rsv_space(inode);
1658 spin_unlock(&inode->i_lock);
1663 * This functions updates i_blocks+i_bytes fields and quota information
1664 * (together with appropriate checks).
1666 * NOTE: We absolutely rely on the fact that caller dirties the inode
1667 * (usually helpers in quotaops.h care about this) and holds a handle for
1668 * the current transaction so that dquot write and inode write go into the
1673 * This operation can block, but only after everything is updated
1675 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1677 int cnt, ret = 0, index;
1678 struct dquot_warn warn[MAXQUOTAS];
1679 int reserve = flags & DQUOT_SPACE_RESERVE;
1680 struct dquot **dquots;
1682 if (!inode_quota_active(inode)) {
1684 spin_lock(&inode->i_lock);
1685 *inode_reserved_space(inode) += number;
1686 spin_unlock(&inode->i_lock);
1688 inode_add_bytes(inode, number);
1693 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1694 warn[cnt].w_type = QUOTA_NL_NOWARN;
1696 dquots = i_dquot(inode);
1697 index = srcu_read_lock(&dquot_srcu);
1698 spin_lock(&inode->i_lock);
1699 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1703 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1706 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1710 /* Back out changes we already did */
1711 for (cnt--; cnt >= 0; cnt--) {
1714 spin_lock(&dquots[cnt]->dq_dqb_lock);
1716 dquot_free_reserved_space(dquots[cnt],
1719 dquot_decr_space(dquots[cnt], number);
1720 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1722 spin_unlock(&inode->i_lock);
1723 goto out_flush_warn;
1727 *inode_reserved_space(inode) += number;
1729 __inode_add_bytes(inode, number);
1730 spin_unlock(&inode->i_lock);
1733 goto out_flush_warn;
1734 mark_all_dquot_dirty(dquots);
1736 srcu_read_unlock(&dquot_srcu, index);
1737 flush_warnings(warn);
1741 EXPORT_SYMBOL(__dquot_alloc_space);
1744 * This operation can block, but only after everything is updated
1746 int dquot_alloc_inode(struct inode *inode)
1748 int cnt, ret = 0, index;
1749 struct dquot_warn warn[MAXQUOTAS];
1750 struct dquot * const *dquots;
1752 if (!inode_quota_active(inode))
1754 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1755 warn[cnt].w_type = QUOTA_NL_NOWARN;
1757 dquots = i_dquot(inode);
1758 index = srcu_read_lock(&dquot_srcu);
1759 spin_lock(&inode->i_lock);
1760 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1763 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1765 for (cnt--; cnt >= 0; cnt--) {
1768 /* Back out changes we already did */
1769 spin_lock(&dquots[cnt]->dq_dqb_lock);
1770 dquot_decr_inodes(dquots[cnt], 1);
1771 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1778 spin_unlock(&inode->i_lock);
1780 mark_all_dquot_dirty(dquots);
1781 srcu_read_unlock(&dquot_srcu, index);
1782 flush_warnings(warn);
1785 EXPORT_SYMBOL(dquot_alloc_inode);
1788 * Convert in-memory reserved quotas to real consumed quotas
1790 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1792 struct dquot **dquots;
1795 if (!inode_quota_active(inode)) {
1796 spin_lock(&inode->i_lock);
1797 *inode_reserved_space(inode) -= number;
1798 __inode_add_bytes(inode, number);
1799 spin_unlock(&inode->i_lock);
1803 dquots = i_dquot(inode);
1804 index = srcu_read_lock(&dquot_srcu);
1805 spin_lock(&inode->i_lock);
1806 /* Claim reserved quotas to allocated quotas */
1807 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1809 struct dquot *dquot = dquots[cnt];
1811 spin_lock(&dquot->dq_dqb_lock);
1812 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1813 number = dquot->dq_dqb.dqb_rsvspace;
1814 dquot->dq_dqb.dqb_curspace += number;
1815 dquot->dq_dqb.dqb_rsvspace -= number;
1816 spin_unlock(&dquot->dq_dqb_lock);
1819 /* Update inode bytes */
1820 *inode_reserved_space(inode) -= number;
1821 __inode_add_bytes(inode, number);
1822 spin_unlock(&inode->i_lock);
1823 mark_all_dquot_dirty(dquots);
1824 srcu_read_unlock(&dquot_srcu, index);
1827 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1830 * Convert allocated space back to in-memory reserved quotas
1832 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1834 struct dquot **dquots;
1837 if (!inode_quota_active(inode)) {
1838 spin_lock(&inode->i_lock);
1839 *inode_reserved_space(inode) += number;
1840 __inode_sub_bytes(inode, number);
1841 spin_unlock(&inode->i_lock);
1845 dquots = i_dquot(inode);
1846 index = srcu_read_lock(&dquot_srcu);
1847 spin_lock(&inode->i_lock);
1848 /* Claim reserved quotas to allocated quotas */
1849 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1851 struct dquot *dquot = dquots[cnt];
1853 spin_lock(&dquot->dq_dqb_lock);
1854 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1855 number = dquot->dq_dqb.dqb_curspace;
1856 dquot->dq_dqb.dqb_rsvspace += number;
1857 dquot->dq_dqb.dqb_curspace -= number;
1858 spin_unlock(&dquot->dq_dqb_lock);
1861 /* Update inode bytes */
1862 *inode_reserved_space(inode) += number;
1863 __inode_sub_bytes(inode, number);
1864 spin_unlock(&inode->i_lock);
1865 mark_all_dquot_dirty(dquots);
1866 srcu_read_unlock(&dquot_srcu, index);
1869 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1872 * This operation can block, but only after everything is updated
1874 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1877 struct dquot_warn warn[MAXQUOTAS];
1878 struct dquot **dquots;
1879 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1881 if (!inode_quota_active(inode)) {
1883 spin_lock(&inode->i_lock);
1884 *inode_reserved_space(inode) -= number;
1885 spin_unlock(&inode->i_lock);
1887 inode_sub_bytes(inode, number);
1892 dquots = i_dquot(inode);
1893 index = srcu_read_lock(&dquot_srcu);
1894 spin_lock(&inode->i_lock);
1895 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1898 warn[cnt].w_type = QUOTA_NL_NOWARN;
1901 spin_lock(&dquots[cnt]->dq_dqb_lock);
1902 wtype = info_bdq_free(dquots[cnt], number);
1903 if (wtype != QUOTA_NL_NOWARN)
1904 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1906 dquot_free_reserved_space(dquots[cnt], number);
1908 dquot_decr_space(dquots[cnt], number);
1909 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1912 *inode_reserved_space(inode) -= number;
1914 __inode_sub_bytes(inode, number);
1915 spin_unlock(&inode->i_lock);
1919 mark_all_dquot_dirty(dquots);
1921 srcu_read_unlock(&dquot_srcu, index);
1922 flush_warnings(warn);
1924 EXPORT_SYMBOL(__dquot_free_space);
1927 * This operation can block, but only after everything is updated
1929 void dquot_free_inode(struct inode *inode)
1932 struct dquot_warn warn[MAXQUOTAS];
1933 struct dquot * const *dquots;
1936 if (!inode_quota_active(inode))
1939 dquots = i_dquot(inode);
1940 index = srcu_read_lock(&dquot_srcu);
1941 spin_lock(&inode->i_lock);
1942 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1945 warn[cnt].w_type = QUOTA_NL_NOWARN;
1948 spin_lock(&dquots[cnt]->dq_dqb_lock);
1949 wtype = info_idq_free(dquots[cnt], 1);
1950 if (wtype != QUOTA_NL_NOWARN)
1951 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1952 dquot_decr_inodes(dquots[cnt], 1);
1953 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1955 spin_unlock(&inode->i_lock);
1956 mark_all_dquot_dirty(dquots);
1957 srcu_read_unlock(&dquot_srcu, index);
1958 flush_warnings(warn);
1960 EXPORT_SYMBOL(dquot_free_inode);
1963 * Transfer the number of inode and blocks from one diskquota to an other.
1964 * On success, dquot references in transfer_to are consumed and references
1965 * to original dquots that need to be released are placed there. On failure,
1966 * references are kept untouched.
1968 * This operation can block, but only after everything is updated
1969 * A transaction must be started when entering this function.
1971 * We are holding reference on transfer_from & transfer_to, no need to
1972 * protect them by srcu_read_lock().
1974 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1977 qsize_t rsv_space = 0;
1978 qsize_t inode_usage = 1;
1979 struct dquot *transfer_from[MAXQUOTAS] = {};
1981 char is_valid[MAXQUOTAS] = {};
1982 struct dquot_warn warn_to[MAXQUOTAS];
1983 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1984 struct dquot_warn warn_from_space[MAXQUOTAS];
1986 if (IS_NOQUOTA(inode))
1989 if (inode->i_sb->dq_op->get_inode_usage) {
1990 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1995 /* Initialize the arrays */
1996 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1997 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1998 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1999 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
2002 spin_lock(&dq_data_lock);
2003 spin_lock(&inode->i_lock);
2004 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
2005 spin_unlock(&inode->i_lock);
2006 spin_unlock(&dq_data_lock);
2009 cur_space = __inode_get_bytes(inode);
2010 rsv_space = __inode_get_rsv_space(inode);
2012 * Build the transfer_from list, check limits, and update usage in
2013 * the target structures.
2015 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2017 * Skip changes for same uid or gid or for turned off quota-type.
2019 if (!transfer_to[cnt])
2021 /* Avoid races with quotaoff() */
2022 if (!sb_has_quota_active(inode->i_sb, cnt))
2025 transfer_from[cnt] = i_dquot(inode)[cnt];
2026 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2030 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2031 DQUOT_SPACE_WARN, &warn_to[cnt]);
2033 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2034 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2035 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2040 /* Decrease usage for source structures and update quota pointers */
2041 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2044 /* Due to IO error we might not have transfer_from[] structure */
2045 if (transfer_from[cnt]) {
2048 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2049 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2050 if (wtype != QUOTA_NL_NOWARN)
2051 prepare_warning(&warn_from_inodes[cnt],
2052 transfer_from[cnt], wtype);
2053 wtype = info_bdq_free(transfer_from[cnt],
2054 cur_space + rsv_space);
2055 if (wtype != QUOTA_NL_NOWARN)
2056 prepare_warning(&warn_from_space[cnt],
2057 transfer_from[cnt], wtype);
2058 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2059 dquot_decr_space(transfer_from[cnt], cur_space);
2060 dquot_free_reserved_space(transfer_from[cnt],
2062 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2064 i_dquot(inode)[cnt] = transfer_to[cnt];
2066 spin_unlock(&inode->i_lock);
2067 spin_unlock(&dq_data_lock);
2069 mark_all_dquot_dirty(transfer_from);
2070 mark_all_dquot_dirty(transfer_to);
2071 flush_warnings(warn_to);
2072 flush_warnings(warn_from_inodes);
2073 flush_warnings(warn_from_space);
2074 /* Pass back references to put */
2075 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2077 transfer_to[cnt] = transfer_from[cnt];
2080 /* Back out changes we already did */
2081 for (cnt--; cnt >= 0; cnt--) {
2084 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2085 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2086 dquot_decr_space(transfer_to[cnt], cur_space);
2087 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2088 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2090 spin_unlock(&inode->i_lock);
2091 spin_unlock(&dq_data_lock);
2092 flush_warnings(warn_to);
2095 EXPORT_SYMBOL(__dquot_transfer);
2097 /* Wrapper for transferring ownership of an inode for uid/gid only
2098 * Called from FSXXX_setattr()
2100 int dquot_transfer(struct mnt_idmap *idmap, struct inode *inode,
2101 struct iattr *iattr)
2103 struct dquot *transfer_to[MAXQUOTAS] = {};
2104 struct dquot *dquot;
2105 struct super_block *sb = inode->i_sb;
2108 if (!inode_quota_active(inode))
2111 if (i_uid_needs_update(idmap, iattr, inode)) {
2112 kuid_t kuid = from_vfsuid(idmap, i_user_ns(inode),
2115 dquot = dqget(sb, make_kqid_uid(kuid));
2116 if (IS_ERR(dquot)) {
2117 if (PTR_ERR(dquot) != -ESRCH) {
2118 ret = PTR_ERR(dquot);
2123 transfer_to[USRQUOTA] = dquot;
2125 if (i_gid_needs_update(idmap, iattr, inode)) {
2126 kgid_t kgid = from_vfsgid(idmap, i_user_ns(inode),
2129 dquot = dqget(sb, make_kqid_gid(kgid));
2130 if (IS_ERR(dquot)) {
2131 if (PTR_ERR(dquot) != -ESRCH) {
2132 ret = PTR_ERR(dquot);
2137 transfer_to[GRPQUOTA] = dquot;
2139 ret = __dquot_transfer(inode, transfer_to);
2141 dqput_all(transfer_to);
2144 EXPORT_SYMBOL(dquot_transfer);
2147 * Write info of quota file to disk
2149 int dquot_commit_info(struct super_block *sb, int type)
2151 struct quota_info *dqopt = sb_dqopt(sb);
2153 return dqopt->ops[type]->write_file_info(sb, type);
2155 EXPORT_SYMBOL(dquot_commit_info);
2157 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2159 struct quota_info *dqopt = sb_dqopt(sb);
2161 if (!sb_has_quota_active(sb, qid->type))
2163 if (!dqopt->ops[qid->type]->get_next_id)
2165 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2167 EXPORT_SYMBOL(dquot_get_next_id);
2170 * Definitions of diskquota operations.
2172 const struct dquot_operations dquot_operations = {
2173 .write_dquot = dquot_commit,
2174 .acquire_dquot = dquot_acquire,
2175 .release_dquot = dquot_release,
2176 .mark_dirty = dquot_mark_dquot_dirty,
2177 .write_info = dquot_commit_info,
2178 .alloc_dquot = dquot_alloc,
2179 .destroy_dquot = dquot_destroy,
2180 .get_next_id = dquot_get_next_id,
2182 EXPORT_SYMBOL(dquot_operations);
2185 * Generic helper for ->open on filesystems supporting disk quotas.
2187 int dquot_file_open(struct inode *inode, struct file *file)
2191 error = generic_file_open(inode, file);
2192 if (!error && (file->f_mode & FMODE_WRITE))
2193 error = dquot_initialize(inode);
2196 EXPORT_SYMBOL(dquot_file_open);
2198 static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2200 struct quota_info *dqopt = sb_dqopt(sb);
2201 struct inode *inode = dqopt->files[type];
2205 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2207 inode->i_flags &= ~S_NOQUOTA;
2208 inode_unlock(inode);
2210 dqopt->files[type] = NULL;
2215 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2217 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2220 struct quota_info *dqopt = sb_dqopt(sb);
2222 /* s_umount should be held in exclusive mode */
2223 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2224 up_read(&sb->s_umount);
2226 /* Cannot turn off usage accounting without turning off limits, or
2227 * suspend quotas and simultaneously turn quotas off. */
2228 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2229 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2230 DQUOT_USAGE_ENABLED)))
2234 * Skip everything if there's nothing to do. We have to do this because
2235 * sometimes we are called when fill_super() failed and calling
2236 * sync_fs() in such cases does no good.
2238 if (!sb_any_quota_loaded(sb))
2241 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2242 if (type != -1 && cnt != type)
2244 if (!sb_has_quota_loaded(sb, cnt))
2247 if (flags & DQUOT_SUSPENDED) {
2248 spin_lock(&dq_state_lock);
2250 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2251 spin_unlock(&dq_state_lock);
2253 spin_lock(&dq_state_lock);
2254 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2255 /* Turning off suspended quotas? */
2256 if (!sb_has_quota_loaded(sb, cnt) &&
2257 sb_has_quota_suspended(sb, cnt)) {
2258 dqopt->flags &= ~dquot_state_flag(
2259 DQUOT_SUSPENDED, cnt);
2260 spin_unlock(&dq_state_lock);
2261 vfs_cleanup_quota_inode(sb, cnt);
2264 spin_unlock(&dq_state_lock);
2267 /* We still have to keep quota loaded? */
2268 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2271 /* Note: these are blocking operations */
2272 drop_dquot_ref(sb, cnt);
2273 invalidate_dquots(sb, cnt);
2275 * Now all dquots should be invalidated, all writes done so we
2276 * should be only users of the info. No locks needed.
2278 if (info_dirty(&dqopt->info[cnt]))
2279 sb->dq_op->write_info(sb, cnt);
2280 if (dqopt->ops[cnt]->free_file_info)
2281 dqopt->ops[cnt]->free_file_info(sb, cnt);
2282 put_quota_format(dqopt->info[cnt].dqi_format);
2283 dqopt->info[cnt].dqi_flags = 0;
2284 dqopt->info[cnt].dqi_igrace = 0;
2285 dqopt->info[cnt].dqi_bgrace = 0;
2286 dqopt->ops[cnt] = NULL;
2289 /* Skip syncing and setting flags if quota files are hidden */
2290 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2293 /* Sync the superblock so that buffers with quota data are written to
2294 * disk (and so userspace sees correct data afterwards). */
2295 if (sb->s_op->sync_fs)
2296 sb->s_op->sync_fs(sb, 1);
2297 sync_blockdev(sb->s_bdev);
2298 /* Now the quota files are just ordinary files and we can set the
2299 * inode flags back. Moreover we discard the pagecache so that
2300 * userspace sees the writes we did bypassing the pagecache. We
2301 * must also discard the blockdev buffers so that we see the
2302 * changes done by userspace on the next quotaon() */
2303 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2304 if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2305 inode_lock(dqopt->files[cnt]);
2306 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2307 inode_unlock(dqopt->files[cnt]);
2310 invalidate_bdev(sb->s_bdev);
2312 /* We are done when suspending quotas */
2313 if (flags & DQUOT_SUSPENDED)
2316 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2317 if (!sb_has_quota_loaded(sb, cnt))
2318 vfs_cleanup_quota_inode(sb, cnt);
2321 EXPORT_SYMBOL(dquot_disable);
2323 int dquot_quota_off(struct super_block *sb, int type)
2325 return dquot_disable(sb, type,
2326 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2328 EXPORT_SYMBOL(dquot_quota_off);
2331 * Turn quotas on on a device
2334 static int vfs_setup_quota_inode(struct inode *inode, int type)
2336 struct super_block *sb = inode->i_sb;
2337 struct quota_info *dqopt = sb_dqopt(sb);
2339 if (is_bad_inode(inode))
2341 if (!S_ISREG(inode->i_mode))
2343 if (IS_RDONLY(inode))
2345 if (sb_has_quota_loaded(sb, type))
2349 * Quota files should never be encrypted. They should be thought of as
2350 * filesystem metadata, not user data. New-style internal quota files
2351 * cannot be encrypted by users anyway, but old-style external quota
2352 * files could potentially be incorrectly created in an encrypted
2353 * directory, hence this explicit check. Some reasons why encrypted
2354 * quota files don't work include: (1) some filesystems that support
2355 * encryption don't handle it in their quota_read and quota_write, and
2356 * (2) cleaning up encrypted quota files at unmount would need special
2357 * consideration, as quota files are cleaned up later than user files.
2359 if (IS_ENCRYPTED(inode))
2362 dqopt->files[type] = igrab(inode);
2363 if (!dqopt->files[type])
2365 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2366 /* We don't want quota and atime on quota files (deadlocks
2367 * possible) Also nobody should write to the file - we use
2368 * special IO operations which ignore the immutable bit. */
2370 inode->i_flags |= S_NOQUOTA;
2371 inode_unlock(inode);
2373 * When S_NOQUOTA is set, remove dquot references as no more
2374 * references can be added
2376 __dquot_drop(inode);
2381 int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2384 struct quota_format_type *fmt = find_quota_format(format_id);
2385 struct quota_info *dqopt = sb_dqopt(sb);
2388 lockdep_assert_held_write(&sb->s_umount);
2390 /* Just unsuspend quotas? */
2391 BUG_ON(flags & DQUOT_SUSPENDED);
2395 if (!sb->dq_op || !sb->s_qcop ||
2396 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2400 /* Filesystems outside of init_user_ns not yet supported */
2401 if (sb->s_user_ns != &init_user_ns) {
2405 /* Usage always has to be set... */
2406 if (!(flags & DQUOT_USAGE_ENABLED)) {
2410 if (sb_has_quota_loaded(sb, type)) {
2415 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2416 /* As we bypass the pagecache we must now flush all the
2417 * dirty data and invalidate caches so that kernel sees
2418 * changes from userspace. It is not enough to just flush
2419 * the quota file since if blocksize < pagesize, invalidation
2420 * of the cache could fail because of other unrelated dirty
2422 sync_filesystem(sb);
2423 invalidate_bdev(sb->s_bdev);
2427 if (!fmt->qf_ops->check_quota_file(sb, type))
2430 dqopt->ops[type] = fmt->qf_ops;
2431 dqopt->info[type].dqi_format = fmt;
2432 dqopt->info[type].dqi_fmt_id = format_id;
2433 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2434 error = dqopt->ops[type]->read_file_info(sb, type);
2437 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2438 spin_lock(&dq_data_lock);
2439 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2440 spin_unlock(&dq_data_lock);
2442 spin_lock(&dq_state_lock);
2443 dqopt->flags |= dquot_state_flag(flags, type);
2444 spin_unlock(&dq_state_lock);
2446 error = add_dquot_ref(sb, type);
2448 dquot_disable(sb, type,
2449 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2453 put_quota_format(fmt);
2457 EXPORT_SYMBOL(dquot_load_quota_sb);
2460 * More powerful function for turning on quotas on given quota inode allowing
2461 * setting of individual quota flags
2463 int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2468 err = vfs_setup_quota_inode(inode, type);
2471 err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2473 vfs_cleanup_quota_inode(inode->i_sb, type);
2476 EXPORT_SYMBOL(dquot_load_quota_inode);
2478 /* Reenable quotas on remount RW */
2479 int dquot_resume(struct super_block *sb, int type)
2481 struct quota_info *dqopt = sb_dqopt(sb);
2485 /* s_umount should be held in exclusive mode */
2486 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2487 up_read(&sb->s_umount);
2489 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2490 if (type != -1 && cnt != type)
2492 if (!sb_has_quota_suspended(sb, cnt))
2495 spin_lock(&dq_state_lock);
2496 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2497 DQUOT_LIMITS_ENABLED,
2499 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2500 spin_unlock(&dq_state_lock);
2502 flags = dquot_generic_flag(flags, cnt);
2503 ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2506 vfs_cleanup_quota_inode(sb, cnt);
2511 EXPORT_SYMBOL(dquot_resume);
2513 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2514 const struct path *path)
2516 int error = security_quota_on(path->dentry);
2519 /* Quota file not on the same filesystem? */
2520 if (path->dentry->d_sb != sb)
2523 error = dquot_load_quota_inode(d_inode(path->dentry), type,
2524 format_id, DQUOT_USAGE_ENABLED |
2525 DQUOT_LIMITS_ENABLED);
2528 EXPORT_SYMBOL(dquot_quota_on);
2531 * This function is used when filesystem needs to initialize quotas
2532 * during mount time.
2534 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2535 int format_id, int type)
2537 struct dentry *dentry;
2540 dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2542 return PTR_ERR(dentry);
2544 error = security_quota_on(dentry);
2546 error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2547 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2552 EXPORT_SYMBOL(dquot_quota_on_mount);
2554 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2558 struct quota_info *dqopt = sb_dqopt(sb);
2560 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2562 /* Accounting cannot be turned on while fs is mounted */
2563 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2566 for (type = 0; type < MAXQUOTAS; type++) {
2567 if (!(flags & qtype_enforce_flag(type)))
2569 /* Can't enforce without accounting */
2570 if (!sb_has_quota_usage_enabled(sb, type)) {
2574 if (sb_has_quota_limits_enabled(sb, type)) {
2578 spin_lock(&dq_state_lock);
2579 dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2580 spin_unlock(&dq_state_lock);
2584 /* Backout enforcement enablement we already did */
2585 for (type--; type >= 0; type--) {
2586 if (flags & qtype_enforce_flag(type))
2587 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2589 /* Error code translation for better compatibility with XFS */
2595 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2599 struct quota_info *dqopt = sb_dqopt(sb);
2601 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2604 * We don't support turning off accounting via quotactl. In principle
2605 * quota infrastructure can do this but filesystems don't expect
2606 * userspace to be able to do it.
2609 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2612 /* Filter out limits not enabled */
2613 for (type = 0; type < MAXQUOTAS; type++)
2614 if (!sb_has_quota_limits_enabled(sb, type))
2615 flags &= ~qtype_enforce_flag(type);
2619 for (type = 0; type < MAXQUOTAS; type++) {
2620 if (flags & qtype_enforce_flag(type)) {
2621 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2628 /* Backout enforcement disabling we already did */
2629 for (type--; type >= 0; type--) {
2630 if (flags & qtype_enforce_flag(type)) {
2631 spin_lock(&dq_state_lock);
2633 dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2634 spin_unlock(&dq_state_lock);
2640 /* Generic routine for getting common part of quota structure */
2641 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2643 struct mem_dqblk *dm = &dquot->dq_dqb;
2645 memset(di, 0, sizeof(*di));
2646 spin_lock(&dquot->dq_dqb_lock);
2647 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2648 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2649 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2650 di->d_ino_softlimit = dm->dqb_isoftlimit;
2651 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2652 di->d_ino_count = dm->dqb_curinodes;
2653 di->d_spc_timer = dm->dqb_btime;
2654 di->d_ino_timer = dm->dqb_itime;
2655 spin_unlock(&dquot->dq_dqb_lock);
2658 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2659 struct qc_dqblk *di)
2661 struct dquot *dquot;
2663 dquot = dqget(sb, qid);
2665 return PTR_ERR(dquot);
2666 do_get_dqblk(dquot, di);
2671 EXPORT_SYMBOL(dquot_get_dqblk);
2673 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2674 struct qc_dqblk *di)
2676 struct dquot *dquot;
2679 if (!sb->dq_op->get_next_id)
2681 err = sb->dq_op->get_next_id(sb, qid);
2684 dquot = dqget(sb, *qid);
2686 return PTR_ERR(dquot);
2687 do_get_dqblk(dquot, di);
2692 EXPORT_SYMBOL(dquot_get_next_dqblk);
2694 #define VFS_QC_MASK \
2695 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2696 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2697 QC_SPC_TIMER | QC_INO_TIMER)
2699 /* Generic routine for setting common part of quota structure */
2700 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2702 struct mem_dqblk *dm = &dquot->dq_dqb;
2703 int check_blim = 0, check_ilim = 0;
2704 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2706 if (di->d_fieldmask & ~VFS_QC_MASK)
2709 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2710 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2711 ((di->d_fieldmask & QC_SPC_HARD) &&
2712 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2713 ((di->d_fieldmask & QC_INO_SOFT) &&
2714 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2715 ((di->d_fieldmask & QC_INO_HARD) &&
2716 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2719 spin_lock(&dquot->dq_dqb_lock);
2720 if (di->d_fieldmask & QC_SPACE) {
2721 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2723 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2726 if (di->d_fieldmask & QC_SPC_SOFT)
2727 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2728 if (di->d_fieldmask & QC_SPC_HARD)
2729 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2730 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2732 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2735 if (di->d_fieldmask & QC_INO_COUNT) {
2736 dm->dqb_curinodes = di->d_ino_count;
2738 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2741 if (di->d_fieldmask & QC_INO_SOFT)
2742 dm->dqb_isoftlimit = di->d_ino_softlimit;
2743 if (di->d_fieldmask & QC_INO_HARD)
2744 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2745 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2747 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2750 if (di->d_fieldmask & QC_SPC_TIMER) {
2751 dm->dqb_btime = di->d_spc_timer;
2753 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2756 if (di->d_fieldmask & QC_INO_TIMER) {
2757 dm->dqb_itime = di->d_ino_timer;
2759 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2763 if (!dm->dqb_bsoftlimit ||
2764 dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2766 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2767 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2768 /* Set grace only if user hasn't provided his own... */
2769 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2772 if (!dm->dqb_isoftlimit ||
2773 dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2775 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2776 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2777 /* Set grace only if user hasn't provided his own... */
2778 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2780 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2782 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2784 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2785 spin_unlock(&dquot->dq_dqb_lock);
2786 mark_dquot_dirty(dquot);
2791 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2792 struct qc_dqblk *di)
2794 struct dquot *dquot;
2797 dquot = dqget(sb, qid);
2798 if (IS_ERR(dquot)) {
2799 rc = PTR_ERR(dquot);
2802 rc = do_set_dqblk(dquot, di);
2807 EXPORT_SYMBOL(dquot_set_dqblk);
2809 /* Generic routine for getting common part of quota file information */
2810 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2812 struct mem_dqinfo *mi;
2813 struct qc_type_state *tstate;
2814 struct quota_info *dqopt = sb_dqopt(sb);
2817 memset(state, 0, sizeof(*state));
2818 for (type = 0; type < MAXQUOTAS; type++) {
2819 if (!sb_has_quota_active(sb, type))
2821 tstate = state->s_state + type;
2822 mi = sb_dqopt(sb)->info + type;
2823 tstate->flags = QCI_ACCT_ENABLED;
2824 spin_lock(&dq_data_lock);
2825 if (mi->dqi_flags & DQF_SYS_FILE)
2826 tstate->flags |= QCI_SYSFILE;
2827 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2828 tstate->flags |= QCI_ROOT_SQUASH;
2829 if (sb_has_quota_limits_enabled(sb, type))
2830 tstate->flags |= QCI_LIMITS_ENFORCED;
2831 tstate->spc_timelimit = mi->dqi_bgrace;
2832 tstate->ino_timelimit = mi->dqi_igrace;
2833 if (dqopt->files[type]) {
2834 tstate->ino = dqopt->files[type]->i_ino;
2835 tstate->blocks = dqopt->files[type]->i_blocks;
2837 tstate->nextents = 1; /* We don't know... */
2838 spin_unlock(&dq_data_lock);
2842 EXPORT_SYMBOL(dquot_get_state);
2844 /* Generic routine for setting common part of quota file information */
2845 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2847 struct mem_dqinfo *mi;
2849 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2850 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2852 if (!sb_has_quota_active(sb, type))
2854 mi = sb_dqopt(sb)->info + type;
2855 if (ii->i_fieldmask & QC_FLAGS) {
2856 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2857 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2860 spin_lock(&dq_data_lock);
2861 if (ii->i_fieldmask & QC_SPC_TIMER)
2862 mi->dqi_bgrace = ii->i_spc_timelimit;
2863 if (ii->i_fieldmask & QC_INO_TIMER)
2864 mi->dqi_igrace = ii->i_ino_timelimit;
2865 if (ii->i_fieldmask & QC_FLAGS) {
2866 if (ii->i_flags & QCI_ROOT_SQUASH)
2867 mi->dqi_flags |= DQF_ROOT_SQUASH;
2869 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2871 spin_unlock(&dq_data_lock);
2872 mark_info_dirty(sb, type);
2873 /* Force write to disk */
2874 return sb->dq_op->write_info(sb, type);
2876 EXPORT_SYMBOL(dquot_set_dqinfo);
2878 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2879 .quota_enable = dquot_quota_enable,
2880 .quota_disable = dquot_quota_disable,
2881 .quota_sync = dquot_quota_sync,
2882 .get_state = dquot_get_state,
2883 .set_info = dquot_set_dqinfo,
2884 .get_dqblk = dquot_get_dqblk,
2885 .get_nextdqblk = dquot_get_next_dqblk,
2886 .set_dqblk = dquot_set_dqblk
2888 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2890 static int do_proc_dqstats(struct ctl_table *table, int write,
2891 void *buffer, size_t *lenp, loff_t *ppos)
2893 unsigned int type = (unsigned long *)table->data - dqstats.stat;
2894 s64 value = percpu_counter_sum(&dqstats.counter[type]);
2896 /* Filter negative values for non-monotonic counters */
2897 if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2898 type == DQST_FREE_DQUOTS))
2901 /* Update global table */
2902 dqstats.stat[type] = value;
2903 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2906 static struct ctl_table fs_dqstats_table[] = {
2908 .procname = "lookups",
2909 .data = &dqstats.stat[DQST_LOOKUPS],
2910 .maxlen = sizeof(unsigned long),
2912 .proc_handler = do_proc_dqstats,
2915 .procname = "drops",
2916 .data = &dqstats.stat[DQST_DROPS],
2917 .maxlen = sizeof(unsigned long),
2919 .proc_handler = do_proc_dqstats,
2922 .procname = "reads",
2923 .data = &dqstats.stat[DQST_READS],
2924 .maxlen = sizeof(unsigned long),
2926 .proc_handler = do_proc_dqstats,
2929 .procname = "writes",
2930 .data = &dqstats.stat[DQST_WRITES],
2931 .maxlen = sizeof(unsigned long),
2933 .proc_handler = do_proc_dqstats,
2936 .procname = "cache_hits",
2937 .data = &dqstats.stat[DQST_CACHE_HITS],
2938 .maxlen = sizeof(unsigned long),
2940 .proc_handler = do_proc_dqstats,
2943 .procname = "allocated_dquots",
2944 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2945 .maxlen = sizeof(unsigned long),
2947 .proc_handler = do_proc_dqstats,
2950 .procname = "free_dquots",
2951 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2952 .maxlen = sizeof(unsigned long),
2954 .proc_handler = do_proc_dqstats,
2957 .procname = "syncs",
2958 .data = &dqstats.stat[DQST_SYNCS],
2959 .maxlen = sizeof(unsigned long),
2961 .proc_handler = do_proc_dqstats,
2963 #ifdef CONFIG_PRINT_QUOTA_WARNING
2965 .procname = "warnings",
2966 .data = &flag_print_warnings,
2967 .maxlen = sizeof(int),
2969 .proc_handler = proc_dointvec,
2975 static int __init dquot_init(void)
2978 unsigned long nr_hash, order;
2979 struct shrinker *dqcache_shrinker;
2981 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2983 register_sysctl_init("fs/quota", fs_dqstats_table);
2985 dquot_cachep = kmem_cache_create("dquot",
2986 sizeof(struct dquot), sizeof(unsigned long) * 4,
2987 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2988 SLAB_MEM_SPREAD|SLAB_PANIC),
2992 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2994 panic("Cannot create dquot hash table");
2996 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2997 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2999 panic("Cannot create dquot stat counters");
3002 /* Find power-of-two hlist_heads which can fit into allocation */
3003 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
3004 dq_hash_bits = ilog2(nr_hash);
3006 nr_hash = 1UL << dq_hash_bits;
3007 dq_hash_mask = nr_hash - 1;
3008 for (i = 0; i < nr_hash; i++)
3009 INIT_HLIST_HEAD(dquot_hash + i);
3011 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
3012 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
3014 dqcache_shrinker = shrinker_alloc(0, "dquota-cache");
3015 if (!dqcache_shrinker)
3016 panic("Cannot allocate dquot shrinker");
3018 dqcache_shrinker->count_objects = dqcache_shrink_count;
3019 dqcache_shrinker->scan_objects = dqcache_shrink_scan;
3021 shrinker_register(dqcache_shrinker);
3025 fs_initcall(dquot_init);