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 "../internal.h" /* ugh */
83 #include <linux/uaccess.h>
86 * There are five quota SMP locks:
87 * * dq_list_lock protects all lists with quotas and quota formats.
88 * * dquot->dq_dqb_lock protects data from dq_dqb
89 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
90 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
91 * dquot_transfer() can stabilize amount it transfers
92 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
93 * pointers in the inode
94 * * dq_state_lock protects modifications of quota state (on quotaon and
95 * quotaoff) and readers who care about latest values take it as well.
97 * The spinlock ordering is hence:
98 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
99 * dq_list_lock > dq_state_lock
101 * Note that some things (eg. sb pointer, type, id) doesn't change during
102 * the life of the dquot structure and so needn't to be protected by a lock
104 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
105 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
106 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
107 * inode and before dropping dquot references to avoid use of dquots after
108 * they are freed. dq_data_lock is used to serialize the pointer setting and
109 * clearing operations.
110 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
111 * inode is a quota file). Functions adding pointers from inode to dquots have
112 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
113 * have to do all pointer modifications before dropping dq_data_lock. This makes
114 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
115 * then drops all pointers to dquots from an inode.
117 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
118 * memory (or space for it is being allocated) on the first dqget(), when it is
119 * being written out, and when it is being released on the last dqput(). The
120 * allocation and release operations are serialized by the dq_lock and by
121 * checking the use count in dquot_release().
123 * Lock ordering (including related VFS locks) is the following:
124 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
128 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
129 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
130 EXPORT_SYMBOL(dq_data_lock);
131 DEFINE_STATIC_SRCU(dquot_srcu);
133 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
135 void __quota_error(struct super_block *sb, const char *func,
136 const char *fmt, ...)
138 if (printk_ratelimit()) {
140 struct va_format vaf;
147 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
148 sb->s_id, func, &vaf);
153 EXPORT_SYMBOL(__quota_error);
155 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
156 static char *quotatypes[] = INITQFNAMES;
158 static struct quota_format_type *quota_formats; /* List of registered formats */
159 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
161 /* SLAB cache for dquot structures */
162 static struct kmem_cache *dquot_cachep;
164 int register_quota_format(struct quota_format_type *fmt)
166 spin_lock(&dq_list_lock);
167 fmt->qf_next = quota_formats;
169 spin_unlock(&dq_list_lock);
172 EXPORT_SYMBOL(register_quota_format);
174 void unregister_quota_format(struct quota_format_type *fmt)
176 struct quota_format_type **actqf;
178 spin_lock(&dq_list_lock);
179 for (actqf = "a_formats; *actqf && *actqf != fmt;
180 actqf = &(*actqf)->qf_next)
183 *actqf = (*actqf)->qf_next;
184 spin_unlock(&dq_list_lock);
186 EXPORT_SYMBOL(unregister_quota_format);
188 static struct quota_format_type *find_quota_format(int id)
190 struct quota_format_type *actqf;
192 spin_lock(&dq_list_lock);
193 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
194 actqf = actqf->qf_next)
196 if (!actqf || !try_module_get(actqf->qf_owner)) {
199 spin_unlock(&dq_list_lock);
201 for (qm = 0; module_names[qm].qm_fmt_id &&
202 module_names[qm].qm_fmt_id != id; qm++)
204 if (!module_names[qm].qm_fmt_id ||
205 request_module(module_names[qm].qm_mod_name))
208 spin_lock(&dq_list_lock);
209 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
210 actqf = actqf->qf_next)
212 if (actqf && !try_module_get(actqf->qf_owner))
215 spin_unlock(&dq_list_lock);
219 static void put_quota_format(struct quota_format_type *fmt)
221 module_put(fmt->qf_owner);
225 * Dquot List Management:
226 * The quota code uses three lists for dquot management: the inuse_list,
227 * free_dquots, and dquot_hash[] array. A single dquot structure may be
228 * on all three lists, depending on its current state.
230 * All dquots are placed to the end of inuse_list when first created, and this
231 * list is used for invalidate operation, which must look at every dquot.
233 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
234 * and this list is searched whenever we need an available dquot. Dquots are
235 * removed from the list as soon as they are used again, and
236 * dqstats.free_dquots gives the number of dquots on the list. When
237 * dquot is invalidated it's completely released from memory.
239 * Dquots with a specific identity (device, type and id) are placed on
240 * one of the dquot_hash[] hash chains. The provides an efficient search
241 * mechanism to locate a specific dquot.
244 static LIST_HEAD(inuse_list);
245 static LIST_HEAD(free_dquots);
246 static unsigned int dq_hash_bits, dq_hash_mask;
247 static struct hlist_head *dquot_hash;
249 struct dqstats dqstats;
250 EXPORT_SYMBOL(dqstats);
252 static qsize_t inode_get_rsv_space(struct inode *inode);
253 static qsize_t __inode_get_rsv_space(struct inode *inode);
254 static int __dquot_initialize(struct inode *inode, int type);
256 static inline unsigned int
257 hashfn(const struct super_block *sb, struct kqid qid)
259 unsigned int id = from_kqid(&init_user_ns, qid);
263 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
264 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
268 * Following list functions expect dq_list_lock to be held
270 static inline void insert_dquot_hash(struct dquot *dquot)
272 struct hlist_head *head;
273 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
274 hlist_add_head(&dquot->dq_hash, head);
277 static inline void remove_dquot_hash(struct dquot *dquot)
279 hlist_del_init(&dquot->dq_hash);
282 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
285 struct hlist_node *node;
288 hlist_for_each (node, dquot_hash+hashent) {
289 dquot = hlist_entry(node, struct dquot, dq_hash);
290 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
296 /* Add a dquot to the tail of the free list */
297 static inline void put_dquot_last(struct dquot *dquot)
299 list_add_tail(&dquot->dq_free, &free_dquots);
300 dqstats_inc(DQST_FREE_DQUOTS);
303 static inline void remove_free_dquot(struct dquot *dquot)
305 if (list_empty(&dquot->dq_free))
307 list_del_init(&dquot->dq_free);
308 dqstats_dec(DQST_FREE_DQUOTS);
311 static inline void put_inuse(struct dquot *dquot)
313 /* We add to the back of inuse list so we don't have to restart
314 * when traversing this list and we block */
315 list_add_tail(&dquot->dq_inuse, &inuse_list);
316 dqstats_inc(DQST_ALLOC_DQUOTS);
319 static inline void remove_inuse(struct dquot *dquot)
321 dqstats_dec(DQST_ALLOC_DQUOTS);
322 list_del(&dquot->dq_inuse);
325 * End of list functions needing dq_list_lock
328 static void wait_on_dquot(struct dquot *dquot)
330 mutex_lock(&dquot->dq_lock);
331 mutex_unlock(&dquot->dq_lock);
334 static inline int dquot_dirty(struct dquot *dquot)
336 return test_bit(DQ_MOD_B, &dquot->dq_flags);
339 static inline int mark_dquot_dirty(struct dquot *dquot)
341 return dquot->dq_sb->dq_op->mark_dirty(dquot);
344 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
345 int dquot_mark_dquot_dirty(struct dquot *dquot)
349 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
352 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
353 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
355 /* If quota is dirty already, we don't have to acquire dq_list_lock */
356 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
359 spin_lock(&dq_list_lock);
360 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
361 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
362 info[dquot->dq_id.type].dqi_dirty_list);
365 spin_unlock(&dq_list_lock);
368 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
370 /* Dirtify all the dquots - this can block when journalling */
371 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
376 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
378 /* Even in case of error we have to continue */
379 ret = mark_dquot_dirty(dquot[cnt]);
386 static inline void dqput_all(struct dquot **dquot)
390 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
394 static inline int clear_dquot_dirty(struct dquot *dquot)
396 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
397 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
399 spin_lock(&dq_list_lock);
400 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
401 spin_unlock(&dq_list_lock);
404 list_del_init(&dquot->dq_dirty);
405 spin_unlock(&dq_list_lock);
409 void mark_info_dirty(struct super_block *sb, int type)
411 spin_lock(&dq_data_lock);
412 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
413 spin_unlock(&dq_data_lock);
415 EXPORT_SYMBOL(mark_info_dirty);
418 * Read dquot from disk and alloc space for it
421 int dquot_acquire(struct dquot *dquot)
423 int ret = 0, ret2 = 0;
424 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
426 mutex_lock(&dquot->dq_lock);
427 if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
428 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
432 /* Make sure flags update is visible after dquot has been filled */
433 smp_mb__before_atomic();
434 set_bit(DQ_READ_B, &dquot->dq_flags);
435 /* Instantiate dquot if needed */
436 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
437 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
438 /* Write the info if needed */
439 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
440 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
441 dquot->dq_sb, dquot->dq_id.type);
451 * Make sure flags update is visible after on-disk struct has been
452 * allocated. Paired with smp_rmb() in dqget().
454 smp_mb__before_atomic();
455 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
457 mutex_unlock(&dquot->dq_lock);
460 EXPORT_SYMBOL(dquot_acquire);
463 * Write dquot to disk
465 int dquot_commit(struct dquot *dquot)
468 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
470 mutex_lock(&dquot->dq_lock);
471 if (!clear_dquot_dirty(dquot))
473 /* Inactive dquot can be only if there was error during read/init
474 * => we have better not writing it */
475 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
476 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
480 mutex_unlock(&dquot->dq_lock);
483 EXPORT_SYMBOL(dquot_commit);
488 int dquot_release(struct dquot *dquot)
490 int ret = 0, ret2 = 0;
491 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
493 mutex_lock(&dquot->dq_lock);
494 /* Check whether we are not racing with some other dqget() */
495 if (atomic_read(&dquot->dq_count) > 1)
497 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
498 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
500 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
501 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
502 dquot->dq_sb, dquot->dq_id.type);
507 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
509 mutex_unlock(&dquot->dq_lock);
512 EXPORT_SYMBOL(dquot_release);
514 void dquot_destroy(struct dquot *dquot)
516 kmem_cache_free(dquot_cachep, dquot);
518 EXPORT_SYMBOL(dquot_destroy);
520 static inline void do_destroy_dquot(struct dquot *dquot)
522 dquot->dq_sb->dq_op->destroy_dquot(dquot);
525 /* Invalidate all dquots on the list. Note that this function is called after
526 * quota is disabled and pointers from inodes removed so there cannot be new
527 * quota users. There can still be some users of quotas due to inodes being
528 * just deleted or pruned by prune_icache() (those are not attached to any
529 * list) or parallel quotactl call. We have to wait for such users.
531 static void invalidate_dquots(struct super_block *sb, int type)
533 struct dquot *dquot, *tmp;
536 spin_lock(&dq_list_lock);
537 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
538 if (dquot->dq_sb != sb)
540 if (dquot->dq_id.type != type)
542 /* Wait for dquot users */
543 if (atomic_read(&dquot->dq_count)) {
545 spin_unlock(&dq_list_lock);
547 * Once dqput() wakes us up, we know it's time to free
549 * IMPORTANT: we rely on the fact that there is always
550 * at most one process waiting for dquot to free.
551 * Otherwise dq_count would be > 1 and we would never
554 wait_event(dquot_ref_wq,
555 atomic_read(&dquot->dq_count) == 1);
557 /* At this moment dquot() need not exist (it could be
558 * reclaimed by prune_dqcache(). Hence we must
563 * Quota now has no users and it has been written on last
566 remove_dquot_hash(dquot);
567 remove_free_dquot(dquot);
569 do_destroy_dquot(dquot);
571 spin_unlock(&dq_list_lock);
574 /* Call callback for every active dquot on given filesystem */
575 int dquot_scan_active(struct super_block *sb,
576 int (*fn)(struct dquot *dquot, unsigned long priv),
579 struct dquot *dquot, *old_dquot = NULL;
582 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
584 spin_lock(&dq_list_lock);
585 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
586 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
588 if (dquot->dq_sb != sb)
590 /* Now we have active dquot so we can just increase use count */
591 atomic_inc(&dquot->dq_count);
592 spin_unlock(&dq_list_lock);
593 dqstats_inc(DQST_LOOKUPS);
597 * ->release_dquot() can be racing with us. Our reference
598 * protects us from new calls to it so just wait for any
599 * outstanding call and recheck the DQ_ACTIVE_B after that.
601 wait_on_dquot(dquot);
602 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
603 ret = fn(dquot, priv);
607 spin_lock(&dq_list_lock);
608 /* We are safe to continue now because our dquot could not
609 * be moved out of the inuse list while we hold the reference */
611 spin_unlock(&dq_list_lock);
616 EXPORT_SYMBOL(dquot_scan_active);
618 /* Write all dquot structures to quota files */
619 int dquot_writeback_dquots(struct super_block *sb, int type)
621 struct list_head *dirty;
623 struct quota_info *dqopt = sb_dqopt(sb);
627 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
629 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
630 if (type != -1 && cnt != type)
632 if (!sb_has_quota_active(sb, cnt))
634 spin_lock(&dq_list_lock);
635 dirty = &dqopt->info[cnt].dqi_dirty_list;
636 while (!list_empty(dirty)) {
637 dquot = list_first_entry(dirty, struct dquot,
640 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
642 /* Now we have active dquot from which someone is
643 * holding reference so we can safely just increase
646 spin_unlock(&dq_list_lock);
647 dqstats_inc(DQST_LOOKUPS);
648 err = sb->dq_op->write_dquot(dquot);
651 * Clear dirty bit anyway to avoid infinite
654 clear_dquot_dirty(dquot);
659 spin_lock(&dq_list_lock);
661 spin_unlock(&dq_list_lock);
664 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
665 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
666 && info_dirty(&dqopt->info[cnt]))
667 sb->dq_op->write_info(sb, cnt);
668 dqstats_inc(DQST_SYNCS);
672 EXPORT_SYMBOL(dquot_writeback_dquots);
674 /* Write all dquot structures to disk and make them visible from userspace */
675 int dquot_quota_sync(struct super_block *sb, int type)
677 struct quota_info *dqopt = sb_dqopt(sb);
681 ret = dquot_writeback_dquots(sb, type);
684 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
687 /* This is not very clever (and fast) but currently I don't know about
688 * any other simple way of getting quota data to disk and we must get
689 * them there for userspace to be visible... */
690 if (sb->s_op->sync_fs)
691 sb->s_op->sync_fs(sb, 1);
692 sync_blockdev(sb->s_bdev);
695 * Now when everything is written we can discard the pagecache so
696 * that userspace sees the changes.
698 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
699 if (type != -1 && cnt != type)
701 if (!sb_has_quota_active(sb, cnt))
703 inode_lock(dqopt->files[cnt]);
704 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
705 inode_unlock(dqopt->files[cnt]);
710 EXPORT_SYMBOL(dquot_quota_sync);
713 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
716 unsigned long freed = 0;
718 spin_lock(&dq_list_lock);
719 while (!list_empty(&free_dquots) && sc->nr_to_scan) {
720 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
721 remove_dquot_hash(dquot);
722 remove_free_dquot(dquot);
724 do_destroy_dquot(dquot);
728 spin_unlock(&dq_list_lock);
733 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
735 return vfs_pressure_ratio(
736 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
739 static struct shrinker dqcache_shrinker = {
740 .count_objects = dqcache_shrink_count,
741 .scan_objects = dqcache_shrink_scan,
742 .seeks = DEFAULT_SEEKS,
746 * Put reference to dquot
748 void dqput(struct dquot *dquot)
754 #ifdef CONFIG_QUOTA_DEBUG
755 if (!atomic_read(&dquot->dq_count)) {
756 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
757 quotatypes[dquot->dq_id.type],
758 from_kqid(&init_user_ns, dquot->dq_id));
762 dqstats_inc(DQST_DROPS);
764 spin_lock(&dq_list_lock);
765 if (atomic_read(&dquot->dq_count) > 1) {
766 /* We have more than one user... nothing to do */
767 atomic_dec(&dquot->dq_count);
768 /* Releasing dquot during quotaoff phase? */
769 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
770 atomic_read(&dquot->dq_count) == 1)
771 wake_up(&dquot_ref_wq);
772 spin_unlock(&dq_list_lock);
775 /* Need to release dquot? */
776 if (dquot_dirty(dquot)) {
777 spin_unlock(&dq_list_lock);
778 /* Commit dquot before releasing */
779 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
781 quota_error(dquot->dq_sb, "Can't write quota structure"
782 " (error %d). Quota may get out of sync!",
785 * We clear dirty bit anyway, so that we avoid
788 clear_dquot_dirty(dquot);
792 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
793 spin_unlock(&dq_list_lock);
794 dquot->dq_sb->dq_op->release_dquot(dquot);
797 atomic_dec(&dquot->dq_count);
798 #ifdef CONFIG_QUOTA_DEBUG
800 BUG_ON(!list_empty(&dquot->dq_free));
802 put_dquot_last(dquot);
803 spin_unlock(&dq_list_lock);
805 EXPORT_SYMBOL(dqput);
807 struct dquot *dquot_alloc(struct super_block *sb, int type)
809 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
811 EXPORT_SYMBOL(dquot_alloc);
813 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
817 dquot = sb->dq_op->alloc_dquot(sb, type);
821 mutex_init(&dquot->dq_lock);
822 INIT_LIST_HEAD(&dquot->dq_free);
823 INIT_LIST_HEAD(&dquot->dq_inuse);
824 INIT_HLIST_NODE(&dquot->dq_hash);
825 INIT_LIST_HEAD(&dquot->dq_dirty);
827 dquot->dq_id = make_kqid_invalid(type);
828 atomic_set(&dquot->dq_count, 1);
829 spin_lock_init(&dquot->dq_dqb_lock);
835 * Get reference to dquot
837 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
838 * destroying our dquot by:
839 * a) checking for quota flags under dq_list_lock and
840 * b) getting a reference to dquot before we release dq_list_lock
842 struct dquot *dqget(struct super_block *sb, struct kqid qid)
844 unsigned int hashent = hashfn(sb, qid);
845 struct dquot *dquot, *empty = NULL;
847 if (!qid_has_mapping(sb->s_user_ns, qid))
848 return ERR_PTR(-EINVAL);
850 if (!sb_has_quota_active(sb, qid.type))
851 return ERR_PTR(-ESRCH);
853 spin_lock(&dq_list_lock);
854 spin_lock(&dq_state_lock);
855 if (!sb_has_quota_active(sb, qid.type)) {
856 spin_unlock(&dq_state_lock);
857 spin_unlock(&dq_list_lock);
858 dquot = ERR_PTR(-ESRCH);
861 spin_unlock(&dq_state_lock);
863 dquot = find_dquot(hashent, sb, qid);
866 spin_unlock(&dq_list_lock);
867 empty = get_empty_dquot(sb, qid.type);
869 schedule(); /* Try to wait for a moment... */
875 /* all dquots go on the inuse_list */
877 /* hash it first so it can be found */
878 insert_dquot_hash(dquot);
879 spin_unlock(&dq_list_lock);
880 dqstats_inc(DQST_LOOKUPS);
882 if (!atomic_read(&dquot->dq_count))
883 remove_free_dquot(dquot);
884 atomic_inc(&dquot->dq_count);
885 spin_unlock(&dq_list_lock);
886 dqstats_inc(DQST_CACHE_HITS);
887 dqstats_inc(DQST_LOOKUPS);
889 /* Wait for dq_lock - after this we know that either dquot_release() is
890 * already finished or it will be canceled due to dq_count > 1 test */
891 wait_on_dquot(dquot);
892 /* Read the dquot / allocate space in quota file */
893 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
896 err = sb->dq_op->acquire_dquot(dquot);
899 dquot = ERR_PTR(err);
904 * Make sure following reads see filled structure - paired with
905 * smp_mb__before_atomic() in dquot_acquire().
908 #ifdef CONFIG_QUOTA_DEBUG
909 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
913 do_destroy_dquot(empty);
917 EXPORT_SYMBOL(dqget);
919 static inline struct dquot **i_dquot(struct inode *inode)
921 return inode->i_sb->s_op->get_dquots(inode);
924 static int dqinit_needed(struct inode *inode, int type)
926 struct dquot * const *dquots;
929 if (IS_NOQUOTA(inode))
932 dquots = i_dquot(inode);
934 return !dquots[type];
935 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
941 /* This routine is guarded by s_umount semaphore */
942 static int add_dquot_ref(struct super_block *sb, int type)
944 struct inode *inode, *old_inode = NULL;
945 #ifdef CONFIG_QUOTA_DEBUG
950 spin_lock(&sb->s_inode_list_lock);
951 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
952 spin_lock(&inode->i_lock);
953 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
954 !atomic_read(&inode->i_writecount) ||
955 !dqinit_needed(inode, type)) {
956 spin_unlock(&inode->i_lock);
960 spin_unlock(&inode->i_lock);
961 spin_unlock(&sb->s_inode_list_lock);
963 #ifdef CONFIG_QUOTA_DEBUG
964 if (unlikely(inode_get_rsv_space(inode) > 0))
968 err = __dquot_initialize(inode, type);
975 * We hold a reference to 'inode' so it couldn't have been
976 * removed from s_inodes list while we dropped the
977 * s_inode_list_lock. We cannot iput the inode now as we can be
978 * holding the last reference and we cannot iput it under
979 * s_inode_list_lock. So we keep the reference and iput it
983 spin_lock(&sb->s_inode_list_lock);
985 spin_unlock(&sb->s_inode_list_lock);
988 #ifdef CONFIG_QUOTA_DEBUG
990 quota_error(sb, "Writes happened before quota was turned on "
991 "thus quota information is probably inconsistent. "
992 "Please run quotacheck(8)");
999 * Remove references to dquots from inode and add dquot to list for freeing
1000 * if we have the last reference to dquot
1002 static void remove_inode_dquot_ref(struct inode *inode, int type,
1003 struct list_head *tofree_head)
1005 struct dquot **dquots = i_dquot(inode);
1006 struct dquot *dquot = dquots[type];
1011 dquots[type] = NULL;
1012 if (list_empty(&dquot->dq_free)) {
1014 * The inode still has reference to dquot so it can't be in the
1017 spin_lock(&dq_list_lock);
1018 list_add(&dquot->dq_free, tofree_head);
1019 spin_unlock(&dq_list_lock);
1022 * Dquot is already in a list to put so we won't drop the last
1030 * Free list of dquots
1031 * Dquots are removed from inodes and no new references can be got so we are
1032 * the only ones holding reference
1034 static void put_dquot_list(struct list_head *tofree_head)
1036 struct list_head *act_head;
1037 struct dquot *dquot;
1039 act_head = tofree_head->next;
1040 while (act_head != tofree_head) {
1041 dquot = list_entry(act_head, struct dquot, dq_free);
1042 act_head = act_head->next;
1043 /* Remove dquot from the list so we won't have problems... */
1044 list_del_init(&dquot->dq_free);
1049 static void remove_dquot_ref(struct super_block *sb, int type,
1050 struct list_head *tofree_head)
1052 struct inode *inode;
1053 #ifdef CONFIG_QUOTA_DEBUG
1057 spin_lock(&sb->s_inode_list_lock);
1058 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1060 * We have to scan also I_NEW inodes because they can already
1061 * have quota pointer initialized. Luckily, we need to touch
1062 * only quota pointers and these have separate locking
1065 spin_lock(&dq_data_lock);
1066 if (!IS_NOQUOTA(inode)) {
1067 #ifdef CONFIG_QUOTA_DEBUG
1068 if (unlikely(inode_get_rsv_space(inode) > 0))
1071 remove_inode_dquot_ref(inode, type, tofree_head);
1073 spin_unlock(&dq_data_lock);
1075 spin_unlock(&sb->s_inode_list_lock);
1076 #ifdef CONFIG_QUOTA_DEBUG
1078 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1079 " was disabled thus quota information is probably "
1080 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1085 /* Gather all references from inodes and drop them */
1086 static void drop_dquot_ref(struct super_block *sb, int type)
1088 LIST_HEAD(tofree_head);
1091 remove_dquot_ref(sb, type, &tofree_head);
1092 synchronize_srcu(&dquot_srcu);
1093 put_dquot_list(&tofree_head);
1098 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1100 if (dquot->dq_dqb.dqb_rsvspace >= number)
1101 dquot->dq_dqb.dqb_rsvspace -= number;
1104 dquot->dq_dqb.dqb_rsvspace = 0;
1106 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1107 dquot->dq_dqb.dqb_bsoftlimit)
1108 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1109 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1112 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1114 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1115 dquot->dq_dqb.dqb_curinodes >= number)
1116 dquot->dq_dqb.dqb_curinodes -= number;
1118 dquot->dq_dqb.dqb_curinodes = 0;
1119 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1120 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1121 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1124 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1126 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1127 dquot->dq_dqb.dqb_curspace >= number)
1128 dquot->dq_dqb.dqb_curspace -= number;
1130 dquot->dq_dqb.dqb_curspace = 0;
1131 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1132 dquot->dq_dqb.dqb_bsoftlimit)
1133 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1134 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1138 struct super_block *w_sb;
1139 struct kqid w_dq_id;
1143 static int warning_issued(struct dquot *dquot, const int warntype)
1145 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1146 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1147 ((warntype == QUOTA_NL_IHARDWARN ||
1148 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1152 return test_and_set_bit(flag, &dquot->dq_flags);
1155 #ifdef CONFIG_PRINT_QUOTA_WARNING
1156 static int flag_print_warnings = 1;
1158 static int need_print_warning(struct dquot_warn *warn)
1160 if (!flag_print_warnings)
1163 switch (warn->w_dq_id.type) {
1165 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1167 return in_group_p(warn->w_dq_id.gid);
1174 /* Print warning to user which exceeded quota */
1175 static void print_warning(struct dquot_warn *warn)
1178 struct tty_struct *tty;
1179 int warntype = warn->w_type;
1181 if (warntype == QUOTA_NL_IHARDBELOW ||
1182 warntype == QUOTA_NL_ISOFTBELOW ||
1183 warntype == QUOTA_NL_BHARDBELOW ||
1184 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1187 tty = get_current_tty();
1190 tty_write_message(tty, warn->w_sb->s_id);
1191 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1192 tty_write_message(tty, ": warning, ");
1194 tty_write_message(tty, ": write failed, ");
1195 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1197 case QUOTA_NL_IHARDWARN:
1198 msg = " file limit reached.\r\n";
1200 case QUOTA_NL_ISOFTLONGWARN:
1201 msg = " file quota exceeded too long.\r\n";
1203 case QUOTA_NL_ISOFTWARN:
1204 msg = " file quota exceeded.\r\n";
1206 case QUOTA_NL_BHARDWARN:
1207 msg = " block limit reached.\r\n";
1209 case QUOTA_NL_BSOFTLONGWARN:
1210 msg = " block quota exceeded too long.\r\n";
1212 case QUOTA_NL_BSOFTWARN:
1213 msg = " block quota exceeded.\r\n";
1216 tty_write_message(tty, msg);
1221 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1224 if (warning_issued(dquot, warntype))
1226 warn->w_type = warntype;
1227 warn->w_sb = dquot->dq_sb;
1228 warn->w_dq_id = dquot->dq_id;
1232 * Write warnings to the console and send warning messages over netlink.
1234 * Note that this function can call into tty and networking code.
1236 static void flush_warnings(struct dquot_warn *warn)
1240 for (i = 0; i < MAXQUOTAS; i++) {
1241 if (warn[i].w_type == QUOTA_NL_NOWARN)
1243 #ifdef CONFIG_PRINT_QUOTA_WARNING
1244 print_warning(&warn[i]);
1246 quota_send_warning(warn[i].w_dq_id,
1247 warn[i].w_sb->s_dev, warn[i].w_type);
1251 static int ignore_hardlimit(struct dquot *dquot)
1253 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1255 return capable(CAP_SYS_RESOURCE) &&
1256 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1257 !(info->dqi_flags & DQF_ROOT_SQUASH));
1260 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1261 struct dquot_warn *warn)
1266 spin_lock(&dquot->dq_dqb_lock);
1267 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1268 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1269 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1272 if (dquot->dq_dqb.dqb_ihardlimit &&
1273 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1274 !ignore_hardlimit(dquot)) {
1275 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1280 if (dquot->dq_dqb.dqb_isoftlimit &&
1281 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1282 dquot->dq_dqb.dqb_itime &&
1283 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1284 !ignore_hardlimit(dquot)) {
1285 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1290 if (dquot->dq_dqb.dqb_isoftlimit &&
1291 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1292 dquot->dq_dqb.dqb_itime == 0) {
1293 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1294 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1295 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1298 dquot->dq_dqb.dqb_curinodes = newinodes;
1301 spin_unlock(&dquot->dq_dqb_lock);
1305 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1306 qsize_t rsv_space, unsigned int flags,
1307 struct dquot_warn *warn)
1310 struct super_block *sb = dquot->dq_sb;
1313 spin_lock(&dquot->dq_dqb_lock);
1314 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1315 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1318 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1319 + space + rsv_space;
1321 if (dquot->dq_dqb.dqb_bhardlimit &&
1322 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1323 !ignore_hardlimit(dquot)) {
1324 if (flags & DQUOT_SPACE_WARN)
1325 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1330 if (dquot->dq_dqb.dqb_bsoftlimit &&
1331 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1332 dquot->dq_dqb.dqb_btime &&
1333 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1334 !ignore_hardlimit(dquot)) {
1335 if (flags & DQUOT_SPACE_WARN)
1336 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1341 if (dquot->dq_dqb.dqb_bsoftlimit &&
1342 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1343 dquot->dq_dqb.dqb_btime == 0) {
1344 if (flags & DQUOT_SPACE_WARN) {
1345 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1346 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1347 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1350 * We don't allow preallocation to exceed softlimit so exceeding will
1359 * We have to be careful and go through warning generation & grace time
1360 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1363 if (flags & DQUOT_SPACE_NOFAIL)
1366 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1367 dquot->dq_dqb.dqb_curspace += space;
1369 spin_unlock(&dquot->dq_dqb_lock);
1373 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1377 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1378 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1379 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1380 return QUOTA_NL_NOWARN;
1382 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1383 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1384 return QUOTA_NL_ISOFTBELOW;
1385 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1386 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1387 return QUOTA_NL_IHARDBELOW;
1388 return QUOTA_NL_NOWARN;
1391 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1395 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1397 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1398 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1399 return QUOTA_NL_NOWARN;
1401 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1402 return QUOTA_NL_BSOFTBELOW;
1403 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1404 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1405 return QUOTA_NL_BHARDBELOW;
1406 return QUOTA_NL_NOWARN;
1409 static int dquot_active(const struct inode *inode)
1411 struct super_block *sb = inode->i_sb;
1413 if (IS_NOQUOTA(inode))
1415 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1419 * Initialize quota pointers in inode
1421 * It is better to call this function outside of any transaction as it
1422 * might need a lot of space in journal for dquot structure allocation.
1424 static int __dquot_initialize(struct inode *inode, int type)
1426 int cnt, init_needed = 0;
1427 struct dquot **dquots, *got[MAXQUOTAS] = {};
1428 struct super_block *sb = inode->i_sb;
1432 if (!dquot_active(inode))
1435 dquots = i_dquot(inode);
1437 /* First get references to structures we might need. */
1438 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1442 struct dquot *dquot;
1444 if (type != -1 && cnt != type)
1447 * The i_dquot should have been initialized in most cases,
1448 * we check it without locking here to avoid unnecessary
1449 * dqget()/dqput() calls.
1454 if (!sb_has_quota_active(sb, cnt))
1461 qid = make_kqid_uid(inode->i_uid);
1464 qid = make_kqid_gid(inode->i_gid);
1467 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1470 qid = make_kqid_projid(projid);
1473 dquot = dqget(sb, qid);
1474 if (IS_ERR(dquot)) {
1475 /* We raced with somebody turning quotas off... */
1476 if (PTR_ERR(dquot) != -ESRCH) {
1477 ret = PTR_ERR(dquot);
1485 /* All required i_dquot has been initialized */
1489 spin_lock(&dq_data_lock);
1490 if (IS_NOQUOTA(inode))
1492 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1493 if (type != -1 && cnt != type)
1495 /* Avoid races with quotaoff() */
1496 if (!sb_has_quota_active(sb, cnt))
1498 /* We could race with quotaon or dqget() could have failed */
1502 dquots[cnt] = got[cnt];
1505 * Make quota reservation system happy if someone
1506 * did a write before quota was turned on
1508 rsv = inode_get_rsv_space(inode);
1509 if (unlikely(rsv)) {
1510 spin_lock(&inode->i_lock);
1511 /* Get reservation again under proper lock */
1512 rsv = __inode_get_rsv_space(inode);
1513 spin_lock(&dquots[cnt]->dq_dqb_lock);
1514 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1515 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1516 spin_unlock(&inode->i_lock);
1521 spin_unlock(&dq_data_lock);
1523 /* Drop unused references */
1529 int dquot_initialize(struct inode *inode)
1531 return __dquot_initialize(inode, -1);
1533 EXPORT_SYMBOL(dquot_initialize);
1535 bool dquot_initialize_needed(struct inode *inode)
1537 struct dquot **dquots;
1540 if (!dquot_active(inode))
1543 dquots = i_dquot(inode);
1544 for (i = 0; i < MAXQUOTAS; i++)
1545 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1549 EXPORT_SYMBOL(dquot_initialize_needed);
1552 * Release all quotas referenced by inode.
1554 * This function only be called on inode free or converting
1555 * a file to quota file, no other users for the i_dquot in
1556 * both cases, so we needn't call synchronize_srcu() after
1559 static void __dquot_drop(struct inode *inode)
1562 struct dquot **dquots = i_dquot(inode);
1563 struct dquot *put[MAXQUOTAS];
1565 spin_lock(&dq_data_lock);
1566 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1567 put[cnt] = dquots[cnt];
1570 spin_unlock(&dq_data_lock);
1574 void dquot_drop(struct inode *inode)
1576 struct dquot * const *dquots;
1579 if (IS_NOQUOTA(inode))
1583 * Test before calling to rule out calls from proc and such
1584 * where we are not allowed to block. Note that this is
1585 * actually reliable test even without the lock - the caller
1586 * must assure that nobody can come after the DQUOT_DROP and
1587 * add quota pointers back anyway.
1589 dquots = i_dquot(inode);
1590 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1595 if (cnt < MAXQUOTAS)
1596 __dquot_drop(inode);
1598 EXPORT_SYMBOL(dquot_drop);
1601 * inode_reserved_space is managed internally by quota, and protected by
1602 * i_lock similar to i_blocks+i_bytes.
1604 static qsize_t *inode_reserved_space(struct inode * inode)
1606 /* Filesystem must explicitly define it's own method in order to use
1607 * quota reservation interface */
1608 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1609 return inode->i_sb->dq_op->get_reserved_space(inode);
1612 static qsize_t __inode_get_rsv_space(struct inode *inode)
1614 if (!inode->i_sb->dq_op->get_reserved_space)
1616 return *inode_reserved_space(inode);
1619 static qsize_t inode_get_rsv_space(struct inode *inode)
1623 if (!inode->i_sb->dq_op->get_reserved_space)
1625 spin_lock(&inode->i_lock);
1626 ret = __inode_get_rsv_space(inode);
1627 spin_unlock(&inode->i_lock);
1632 * This functions updates i_blocks+i_bytes fields and quota information
1633 * (together with appropriate checks).
1635 * NOTE: We absolutely rely on the fact that caller dirties the inode
1636 * (usually helpers in quotaops.h care about this) and holds a handle for
1637 * the current transaction so that dquot write and inode write go into the
1642 * This operation can block, but only after everything is updated
1644 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1646 int cnt, ret = 0, index;
1647 struct dquot_warn warn[MAXQUOTAS];
1648 int reserve = flags & DQUOT_SPACE_RESERVE;
1649 struct dquot **dquots;
1651 if (!dquot_active(inode)) {
1653 spin_lock(&inode->i_lock);
1654 *inode_reserved_space(inode) += number;
1655 spin_unlock(&inode->i_lock);
1657 inode_add_bytes(inode, number);
1662 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1663 warn[cnt].w_type = QUOTA_NL_NOWARN;
1665 dquots = i_dquot(inode);
1666 index = srcu_read_lock(&dquot_srcu);
1667 spin_lock(&inode->i_lock);
1668 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1672 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1675 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1679 /* Back out changes we already did */
1680 for (cnt--; cnt >= 0; cnt--) {
1683 spin_lock(&dquots[cnt]->dq_dqb_lock);
1685 dquot_free_reserved_space(dquots[cnt],
1688 dquot_decr_space(dquots[cnt], number);
1689 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1691 spin_unlock(&inode->i_lock);
1692 goto out_flush_warn;
1696 *inode_reserved_space(inode) += number;
1698 __inode_add_bytes(inode, number);
1699 spin_unlock(&inode->i_lock);
1702 goto out_flush_warn;
1703 mark_all_dquot_dirty(dquots);
1705 srcu_read_unlock(&dquot_srcu, index);
1706 flush_warnings(warn);
1710 EXPORT_SYMBOL(__dquot_alloc_space);
1713 * This operation can block, but only after everything is updated
1715 int dquot_alloc_inode(struct inode *inode)
1717 int cnt, ret = 0, index;
1718 struct dquot_warn warn[MAXQUOTAS];
1719 struct dquot * const *dquots;
1721 if (!dquot_active(inode))
1723 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1724 warn[cnt].w_type = QUOTA_NL_NOWARN;
1726 dquots = i_dquot(inode);
1727 index = srcu_read_lock(&dquot_srcu);
1728 spin_lock(&inode->i_lock);
1729 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1732 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1734 for (cnt--; cnt >= 0; cnt--) {
1737 /* Back out changes we already did */
1738 spin_lock(&dquots[cnt]->dq_dqb_lock);
1739 dquot_decr_inodes(dquots[cnt], 1);
1740 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1747 spin_unlock(&inode->i_lock);
1749 mark_all_dquot_dirty(dquots);
1750 srcu_read_unlock(&dquot_srcu, index);
1751 flush_warnings(warn);
1754 EXPORT_SYMBOL(dquot_alloc_inode);
1757 * Convert in-memory reserved quotas to real consumed quotas
1759 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1761 struct dquot **dquots;
1764 if (!dquot_active(inode)) {
1765 spin_lock(&inode->i_lock);
1766 *inode_reserved_space(inode) -= number;
1767 __inode_add_bytes(inode, number);
1768 spin_unlock(&inode->i_lock);
1772 dquots = i_dquot(inode);
1773 index = srcu_read_lock(&dquot_srcu);
1774 spin_lock(&inode->i_lock);
1775 /* Claim reserved quotas to allocated quotas */
1776 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1778 struct dquot *dquot = dquots[cnt];
1780 spin_lock(&dquot->dq_dqb_lock);
1781 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1782 number = dquot->dq_dqb.dqb_rsvspace;
1783 dquot->dq_dqb.dqb_curspace += number;
1784 dquot->dq_dqb.dqb_rsvspace -= number;
1785 spin_unlock(&dquot->dq_dqb_lock);
1788 /* Update inode bytes */
1789 *inode_reserved_space(inode) -= number;
1790 __inode_add_bytes(inode, number);
1791 spin_unlock(&inode->i_lock);
1792 mark_all_dquot_dirty(dquots);
1793 srcu_read_unlock(&dquot_srcu, index);
1796 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1799 * Convert allocated space back to in-memory reserved quotas
1801 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1803 struct dquot **dquots;
1806 if (!dquot_active(inode)) {
1807 spin_lock(&inode->i_lock);
1808 *inode_reserved_space(inode) += number;
1809 __inode_sub_bytes(inode, number);
1810 spin_unlock(&inode->i_lock);
1814 dquots = i_dquot(inode);
1815 index = srcu_read_lock(&dquot_srcu);
1816 spin_lock(&inode->i_lock);
1817 /* Claim reserved quotas to allocated quotas */
1818 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1820 struct dquot *dquot = dquots[cnt];
1822 spin_lock(&dquot->dq_dqb_lock);
1823 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1824 number = dquot->dq_dqb.dqb_curspace;
1825 dquot->dq_dqb.dqb_rsvspace += number;
1826 dquot->dq_dqb.dqb_curspace -= number;
1827 spin_unlock(&dquot->dq_dqb_lock);
1830 /* Update inode bytes */
1831 *inode_reserved_space(inode) += number;
1832 __inode_sub_bytes(inode, number);
1833 spin_unlock(&inode->i_lock);
1834 mark_all_dquot_dirty(dquots);
1835 srcu_read_unlock(&dquot_srcu, index);
1838 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1841 * This operation can block, but only after everything is updated
1843 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1846 struct dquot_warn warn[MAXQUOTAS];
1847 struct dquot **dquots;
1848 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1850 if (!dquot_active(inode)) {
1852 spin_lock(&inode->i_lock);
1853 *inode_reserved_space(inode) -= number;
1854 spin_unlock(&inode->i_lock);
1856 inode_sub_bytes(inode, number);
1861 dquots = i_dquot(inode);
1862 index = srcu_read_lock(&dquot_srcu);
1863 spin_lock(&inode->i_lock);
1864 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1867 warn[cnt].w_type = QUOTA_NL_NOWARN;
1870 spin_lock(&dquots[cnt]->dq_dqb_lock);
1871 wtype = info_bdq_free(dquots[cnt], number);
1872 if (wtype != QUOTA_NL_NOWARN)
1873 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1875 dquot_free_reserved_space(dquots[cnt], number);
1877 dquot_decr_space(dquots[cnt], number);
1878 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1881 *inode_reserved_space(inode) -= number;
1883 __inode_sub_bytes(inode, number);
1884 spin_unlock(&inode->i_lock);
1888 mark_all_dquot_dirty(dquots);
1890 srcu_read_unlock(&dquot_srcu, index);
1891 flush_warnings(warn);
1893 EXPORT_SYMBOL(__dquot_free_space);
1896 * This operation can block, but only after everything is updated
1898 void dquot_free_inode(struct inode *inode)
1901 struct dquot_warn warn[MAXQUOTAS];
1902 struct dquot * const *dquots;
1905 if (!dquot_active(inode))
1908 dquots = i_dquot(inode);
1909 index = srcu_read_lock(&dquot_srcu);
1910 spin_lock(&inode->i_lock);
1911 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1914 warn[cnt].w_type = QUOTA_NL_NOWARN;
1917 spin_lock(&dquots[cnt]->dq_dqb_lock);
1918 wtype = info_idq_free(dquots[cnt], 1);
1919 if (wtype != QUOTA_NL_NOWARN)
1920 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1921 dquot_decr_inodes(dquots[cnt], 1);
1922 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1924 spin_unlock(&inode->i_lock);
1925 mark_all_dquot_dirty(dquots);
1926 srcu_read_unlock(&dquot_srcu, index);
1927 flush_warnings(warn);
1929 EXPORT_SYMBOL(dquot_free_inode);
1932 * Transfer the number of inode and blocks from one diskquota to an other.
1933 * On success, dquot references in transfer_to are consumed and references
1934 * to original dquots that need to be released are placed there. On failure,
1935 * references are kept untouched.
1937 * This operation can block, but only after everything is updated
1938 * A transaction must be started when entering this function.
1940 * We are holding reference on transfer_from & transfer_to, no need to
1941 * protect them by srcu_read_lock().
1943 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1946 qsize_t rsv_space = 0;
1947 qsize_t inode_usage = 1;
1948 struct dquot *transfer_from[MAXQUOTAS] = {};
1950 char is_valid[MAXQUOTAS] = {};
1951 struct dquot_warn warn_to[MAXQUOTAS];
1952 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1953 struct dquot_warn warn_from_space[MAXQUOTAS];
1955 if (IS_NOQUOTA(inode))
1958 if (inode->i_sb->dq_op->get_inode_usage) {
1959 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1964 /* Initialize the arrays */
1965 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1966 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1967 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1968 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1971 spin_lock(&dq_data_lock);
1972 spin_lock(&inode->i_lock);
1973 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1974 spin_unlock(&inode->i_lock);
1975 spin_unlock(&dq_data_lock);
1978 cur_space = __inode_get_bytes(inode);
1979 rsv_space = __inode_get_rsv_space(inode);
1981 * Build the transfer_from list, check limits, and update usage in
1982 * the target structures.
1984 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1986 * Skip changes for same uid or gid or for turned off quota-type.
1988 if (!transfer_to[cnt])
1990 /* Avoid races with quotaoff() */
1991 if (!sb_has_quota_active(inode->i_sb, cnt))
1994 transfer_from[cnt] = i_dquot(inode)[cnt];
1995 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1999 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2000 DQUOT_SPACE_WARN, &warn_to[cnt]);
2002 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2003 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2004 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2009 /* Decrease usage for source structures and update quota pointers */
2010 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2013 /* Due to IO error we might not have transfer_from[] structure */
2014 if (transfer_from[cnt]) {
2017 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2018 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2019 if (wtype != QUOTA_NL_NOWARN)
2020 prepare_warning(&warn_from_inodes[cnt],
2021 transfer_from[cnt], wtype);
2022 wtype = info_bdq_free(transfer_from[cnt],
2023 cur_space + rsv_space);
2024 if (wtype != QUOTA_NL_NOWARN)
2025 prepare_warning(&warn_from_space[cnt],
2026 transfer_from[cnt], wtype);
2027 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2028 dquot_decr_space(transfer_from[cnt], cur_space);
2029 dquot_free_reserved_space(transfer_from[cnt],
2031 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2033 i_dquot(inode)[cnt] = transfer_to[cnt];
2035 spin_unlock(&inode->i_lock);
2036 spin_unlock(&dq_data_lock);
2038 mark_all_dquot_dirty(transfer_from);
2039 mark_all_dquot_dirty(transfer_to);
2040 flush_warnings(warn_to);
2041 flush_warnings(warn_from_inodes);
2042 flush_warnings(warn_from_space);
2043 /* Pass back references to put */
2044 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2046 transfer_to[cnt] = transfer_from[cnt];
2049 /* Back out changes we already did */
2050 for (cnt--; cnt >= 0; cnt--) {
2053 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2054 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2055 dquot_decr_space(transfer_to[cnt], cur_space);
2056 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2057 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2059 spin_unlock(&inode->i_lock);
2060 spin_unlock(&dq_data_lock);
2061 flush_warnings(warn_to);
2064 EXPORT_SYMBOL(__dquot_transfer);
2066 /* Wrapper for transferring ownership of an inode for uid/gid only
2067 * Called from FSXXX_setattr()
2069 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2071 struct dquot *transfer_to[MAXQUOTAS] = {};
2072 struct dquot *dquot;
2073 struct super_block *sb = inode->i_sb;
2076 if (!dquot_active(inode))
2079 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2080 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2081 if (IS_ERR(dquot)) {
2082 if (PTR_ERR(dquot) != -ESRCH) {
2083 ret = PTR_ERR(dquot);
2088 transfer_to[USRQUOTA] = dquot;
2090 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2091 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2092 if (IS_ERR(dquot)) {
2093 if (PTR_ERR(dquot) != -ESRCH) {
2094 ret = PTR_ERR(dquot);
2099 transfer_to[GRPQUOTA] = dquot;
2101 ret = __dquot_transfer(inode, transfer_to);
2103 dqput_all(transfer_to);
2106 EXPORT_SYMBOL(dquot_transfer);
2109 * Write info of quota file to disk
2111 int dquot_commit_info(struct super_block *sb, int type)
2113 struct quota_info *dqopt = sb_dqopt(sb);
2115 return dqopt->ops[type]->write_file_info(sb, type);
2117 EXPORT_SYMBOL(dquot_commit_info);
2119 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2121 struct quota_info *dqopt = sb_dqopt(sb);
2123 if (!sb_has_quota_active(sb, qid->type))
2125 if (!dqopt->ops[qid->type]->get_next_id)
2127 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2129 EXPORT_SYMBOL(dquot_get_next_id);
2132 * Definitions of diskquota operations.
2134 const struct dquot_operations dquot_operations = {
2135 .write_dquot = dquot_commit,
2136 .acquire_dquot = dquot_acquire,
2137 .release_dquot = dquot_release,
2138 .mark_dirty = dquot_mark_dquot_dirty,
2139 .write_info = dquot_commit_info,
2140 .alloc_dquot = dquot_alloc,
2141 .destroy_dquot = dquot_destroy,
2142 .get_next_id = dquot_get_next_id,
2144 EXPORT_SYMBOL(dquot_operations);
2147 * Generic helper for ->open on filesystems supporting disk quotas.
2149 int dquot_file_open(struct inode *inode, struct file *file)
2153 error = generic_file_open(inode, file);
2154 if (!error && (file->f_mode & FMODE_WRITE))
2155 error = dquot_initialize(inode);
2158 EXPORT_SYMBOL(dquot_file_open);
2161 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2163 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2166 struct quota_info *dqopt = sb_dqopt(sb);
2167 struct inode *toputinode[MAXQUOTAS];
2169 /* s_umount should be held in exclusive mode */
2170 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2171 up_read(&sb->s_umount);
2173 /* Cannot turn off usage accounting without turning off limits, or
2174 * suspend quotas and simultaneously turn quotas off. */
2175 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2176 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2177 DQUOT_USAGE_ENABLED)))
2181 * Skip everything if there's nothing to do. We have to do this because
2182 * sometimes we are called when fill_super() failed and calling
2183 * sync_fs() in such cases does no good.
2185 if (!sb_any_quota_loaded(sb))
2188 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2189 toputinode[cnt] = NULL;
2190 if (type != -1 && cnt != type)
2192 if (!sb_has_quota_loaded(sb, cnt))
2195 if (flags & DQUOT_SUSPENDED) {
2196 spin_lock(&dq_state_lock);
2198 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2199 spin_unlock(&dq_state_lock);
2201 spin_lock(&dq_state_lock);
2202 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2203 /* Turning off suspended quotas? */
2204 if (!sb_has_quota_loaded(sb, cnt) &&
2205 sb_has_quota_suspended(sb, cnt)) {
2206 dqopt->flags &= ~dquot_state_flag(
2207 DQUOT_SUSPENDED, cnt);
2208 spin_unlock(&dq_state_lock);
2209 iput(dqopt->files[cnt]);
2210 dqopt->files[cnt] = NULL;
2213 spin_unlock(&dq_state_lock);
2216 /* We still have to keep quota loaded? */
2217 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2220 /* Note: these are blocking operations */
2221 drop_dquot_ref(sb, cnt);
2222 invalidate_dquots(sb, cnt);
2224 * Now all dquots should be invalidated, all writes done so we
2225 * should be only users of the info. No locks needed.
2227 if (info_dirty(&dqopt->info[cnt]))
2228 sb->dq_op->write_info(sb, cnt);
2229 if (dqopt->ops[cnt]->free_file_info)
2230 dqopt->ops[cnt]->free_file_info(sb, cnt);
2231 put_quota_format(dqopt->info[cnt].dqi_format);
2233 toputinode[cnt] = dqopt->files[cnt];
2234 if (!sb_has_quota_loaded(sb, cnt))
2235 dqopt->files[cnt] = NULL;
2236 dqopt->info[cnt].dqi_flags = 0;
2237 dqopt->info[cnt].dqi_igrace = 0;
2238 dqopt->info[cnt].dqi_bgrace = 0;
2239 dqopt->ops[cnt] = NULL;
2242 /* Skip syncing and setting flags if quota files are hidden */
2243 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2246 /* Sync the superblock so that buffers with quota data are written to
2247 * disk (and so userspace sees correct data afterwards). */
2248 if (sb->s_op->sync_fs)
2249 sb->s_op->sync_fs(sb, 1);
2250 sync_blockdev(sb->s_bdev);
2251 /* Now the quota files are just ordinary files and we can set the
2252 * inode flags back. Moreover we discard the pagecache so that
2253 * userspace sees the writes we did bypassing the pagecache. We
2254 * must also discard the blockdev buffers so that we see the
2255 * changes done by userspace on the next quotaon() */
2256 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2257 /* This can happen when suspending quotas on remount-ro... */
2258 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2259 inode_lock(toputinode[cnt]);
2260 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2261 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2262 inode_unlock(toputinode[cnt]);
2263 mark_inode_dirty_sync(toputinode[cnt]);
2266 invalidate_bdev(sb->s_bdev);
2268 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2269 if (toputinode[cnt]) {
2270 /* On remount RO, we keep the inode pointer so that we
2271 * can reenable quota on the subsequent remount RW. We
2272 * have to check 'flags' variable and not use sb_has_
2273 * function because another quotaon / quotaoff could
2274 * change global state before we got here. We refuse
2275 * to suspend quotas when there is pending delete on
2276 * the quota file... */
2277 if (!(flags & DQUOT_SUSPENDED))
2278 iput(toputinode[cnt]);
2279 else if (!toputinode[cnt]->i_nlink)
2284 EXPORT_SYMBOL(dquot_disable);
2286 int dquot_quota_off(struct super_block *sb, int type)
2288 return dquot_disable(sb, type,
2289 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2291 EXPORT_SYMBOL(dquot_quota_off);
2294 * Turn quotas on on a device
2298 * Helper function to turn quotas on when we already have the inode of
2299 * quota file and no quota information is loaded.
2301 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2304 struct quota_format_type *fmt = find_quota_format(format_id);
2305 struct super_block *sb = inode->i_sb;
2306 struct quota_info *dqopt = sb_dqopt(sb);
2311 if (!S_ISREG(inode->i_mode)) {
2315 if (IS_RDONLY(inode)) {
2319 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2320 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2324 /* Filesystems outside of init_user_ns not yet supported */
2325 if (sb->s_user_ns != &init_user_ns) {
2329 /* Usage always has to be set... */
2330 if (!(flags & DQUOT_USAGE_ENABLED)) {
2334 if (sb_has_quota_loaded(sb, type)) {
2339 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2340 /* As we bypass the pagecache we must now flush all the
2341 * dirty data and invalidate caches so that kernel sees
2342 * changes from userspace. It is not enough to just flush
2343 * the quota file since if blocksize < pagesize, invalidation
2344 * of the cache could fail because of other unrelated dirty
2346 sync_filesystem(sb);
2347 invalidate_bdev(sb->s_bdev);
2350 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2351 /* We don't want quota and atime on quota files (deadlocks
2352 * possible) Also nobody should write to the file - we use
2353 * special IO operations which ignore the immutable bit. */
2355 inode->i_flags |= S_NOQUOTA;
2356 inode_unlock(inode);
2358 * When S_NOQUOTA is set, remove dquot references as no more
2359 * references can be added
2361 __dquot_drop(inode);
2365 dqopt->files[type] = igrab(inode);
2366 if (!dqopt->files[type])
2367 goto out_file_flags;
2369 if (!fmt->qf_ops->check_quota_file(sb, type))
2372 dqopt->ops[type] = fmt->qf_ops;
2373 dqopt->info[type].dqi_format = fmt;
2374 dqopt->info[type].dqi_fmt_id = format_id;
2375 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2376 error = dqopt->ops[type]->read_file_info(sb, type);
2379 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2380 spin_lock(&dq_data_lock);
2381 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2382 spin_unlock(&dq_data_lock);
2384 spin_lock(&dq_state_lock);
2385 dqopt->flags |= dquot_state_flag(flags, type);
2386 spin_unlock(&dq_state_lock);
2388 error = add_dquot_ref(sb, type);
2390 dquot_disable(sb, type, flags);
2394 dqopt->files[type] = NULL;
2398 inode->i_flags &= ~S_NOQUOTA;
2399 inode_unlock(inode);
2401 put_quota_format(fmt);
2406 /* Reenable quotas on remount RW */
2407 int dquot_resume(struct super_block *sb, int type)
2409 struct quota_info *dqopt = sb_dqopt(sb);
2410 struct inode *inode;
2414 /* s_umount should be held in exclusive mode */
2415 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2416 up_read(&sb->s_umount);
2418 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2419 if (type != -1 && cnt != type)
2421 if (!sb_has_quota_suspended(sb, cnt))
2424 inode = dqopt->files[cnt];
2425 dqopt->files[cnt] = NULL;
2426 spin_lock(&dq_state_lock);
2427 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2428 DQUOT_LIMITS_ENABLED,
2430 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2431 spin_unlock(&dq_state_lock);
2433 flags = dquot_generic_flag(flags, cnt);
2434 ret = vfs_load_quota_inode(inode, cnt,
2435 dqopt->info[cnt].dqi_fmt_id, flags);
2441 EXPORT_SYMBOL(dquot_resume);
2443 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2444 const struct path *path)
2446 int error = security_quota_on(path->dentry);
2449 /* Quota file not on the same filesystem? */
2450 if (path->dentry->d_sb != sb)
2453 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2454 format_id, DQUOT_USAGE_ENABLED |
2455 DQUOT_LIMITS_ENABLED);
2458 EXPORT_SYMBOL(dquot_quota_on);
2461 * More powerful function for turning on quotas allowing setting
2462 * of individual quota flags
2464 int dquot_enable(struct inode *inode, int type, int format_id,
2467 struct super_block *sb = inode->i_sb;
2469 /* Just unsuspend quotas? */
2470 BUG_ON(flags & DQUOT_SUSPENDED);
2471 /* s_umount should be held in exclusive mode */
2472 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2473 up_read(&sb->s_umount);
2477 /* Just updating flags needed? */
2478 if (sb_has_quota_loaded(sb, type)) {
2479 if (flags & DQUOT_USAGE_ENABLED &&
2480 sb_has_quota_usage_enabled(sb, type))
2482 if (flags & DQUOT_LIMITS_ENABLED &&
2483 sb_has_quota_limits_enabled(sb, type))
2485 spin_lock(&dq_state_lock);
2486 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2487 spin_unlock(&dq_state_lock);
2491 return vfs_load_quota_inode(inode, type, format_id, flags);
2493 EXPORT_SYMBOL(dquot_enable);
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_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2507 return PTR_ERR(dentry);
2509 if (d_really_is_negative(dentry)) {
2514 error = security_quota_on(dentry);
2516 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2517 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2523 EXPORT_SYMBOL(dquot_quota_on_mount);
2525 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2529 struct quota_info *dqopt = sb_dqopt(sb);
2531 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2533 /* Accounting cannot be turned on while fs is mounted */
2534 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2537 for (type = 0; type < MAXQUOTAS; type++) {
2538 if (!(flags & qtype_enforce_flag(type)))
2540 /* Can't enforce without accounting */
2541 if (!sb_has_quota_usage_enabled(sb, type))
2543 ret = dquot_enable(dqopt->files[type], type,
2544 dqopt->info[type].dqi_fmt_id,
2545 DQUOT_LIMITS_ENABLED);
2551 /* Backout enforcement enablement we already did */
2552 for (type--; type >= 0; type--) {
2553 if (flags & qtype_enforce_flag(type))
2554 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2556 /* Error code translation for better compatibility with XFS */
2562 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2566 struct quota_info *dqopt = sb_dqopt(sb);
2568 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2571 * We don't support turning off accounting via quotactl. In principle
2572 * quota infrastructure can do this but filesystems don't expect
2573 * userspace to be able to do it.
2576 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2579 /* Filter out limits not enabled */
2580 for (type = 0; type < MAXQUOTAS; type++)
2581 if (!sb_has_quota_limits_enabled(sb, type))
2582 flags &= ~qtype_enforce_flag(type);
2586 for (type = 0; type < MAXQUOTAS; type++) {
2587 if (flags & qtype_enforce_flag(type)) {
2588 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2595 /* Backout enforcement disabling we already did */
2596 for (type--; type >= 0; type--) {
2597 if (flags & qtype_enforce_flag(type))
2598 dquot_enable(dqopt->files[type], type,
2599 dqopt->info[type].dqi_fmt_id,
2600 DQUOT_LIMITS_ENABLED);
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 tstate->ino = dqopt->files[type]->i_ino;
2799 tstate->blocks = dqopt->files[type]->i_blocks;
2800 tstate->nextents = 1; /* We don't know... */
2801 spin_unlock(&dq_data_lock);
2805 EXPORT_SYMBOL(dquot_get_state);
2807 /* Generic routine for setting common part of quota file information */
2808 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2810 struct mem_dqinfo *mi;
2813 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2814 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2816 if (!sb_has_quota_active(sb, type))
2818 mi = sb_dqopt(sb)->info + type;
2819 if (ii->i_fieldmask & QC_FLAGS) {
2820 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2821 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2824 spin_lock(&dq_data_lock);
2825 if (ii->i_fieldmask & QC_SPC_TIMER)
2826 mi->dqi_bgrace = ii->i_spc_timelimit;
2827 if (ii->i_fieldmask & QC_INO_TIMER)
2828 mi->dqi_igrace = ii->i_ino_timelimit;
2829 if (ii->i_fieldmask & QC_FLAGS) {
2830 if (ii->i_flags & QCI_ROOT_SQUASH)
2831 mi->dqi_flags |= DQF_ROOT_SQUASH;
2833 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2835 spin_unlock(&dq_data_lock);
2836 mark_info_dirty(sb, type);
2837 /* Force write to disk */
2838 sb->dq_op->write_info(sb, type);
2841 EXPORT_SYMBOL(dquot_set_dqinfo);
2843 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2844 .quota_enable = dquot_quota_enable,
2845 .quota_disable = dquot_quota_disable,
2846 .quota_sync = dquot_quota_sync,
2847 .get_state = dquot_get_state,
2848 .set_info = dquot_set_dqinfo,
2849 .get_dqblk = dquot_get_dqblk,
2850 .get_nextdqblk = dquot_get_next_dqblk,
2851 .set_dqblk = dquot_set_dqblk
2853 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2855 static int do_proc_dqstats(struct ctl_table *table, int write,
2856 void __user *buffer, size_t *lenp, loff_t *ppos)
2858 unsigned int type = (int *)table->data - dqstats.stat;
2860 /* Update global table */
2861 dqstats.stat[type] =
2862 percpu_counter_sum_positive(&dqstats.counter[type]);
2863 return proc_dointvec(table, write, buffer, lenp, ppos);
2866 static struct ctl_table fs_dqstats_table[] = {
2868 .procname = "lookups",
2869 .data = &dqstats.stat[DQST_LOOKUPS],
2870 .maxlen = sizeof(int),
2872 .proc_handler = do_proc_dqstats,
2875 .procname = "drops",
2876 .data = &dqstats.stat[DQST_DROPS],
2877 .maxlen = sizeof(int),
2879 .proc_handler = do_proc_dqstats,
2882 .procname = "reads",
2883 .data = &dqstats.stat[DQST_READS],
2884 .maxlen = sizeof(int),
2886 .proc_handler = do_proc_dqstats,
2889 .procname = "writes",
2890 .data = &dqstats.stat[DQST_WRITES],
2891 .maxlen = sizeof(int),
2893 .proc_handler = do_proc_dqstats,
2896 .procname = "cache_hits",
2897 .data = &dqstats.stat[DQST_CACHE_HITS],
2898 .maxlen = sizeof(int),
2900 .proc_handler = do_proc_dqstats,
2903 .procname = "allocated_dquots",
2904 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2905 .maxlen = sizeof(int),
2907 .proc_handler = do_proc_dqstats,
2910 .procname = "free_dquots",
2911 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2912 .maxlen = sizeof(int),
2914 .proc_handler = do_proc_dqstats,
2917 .procname = "syncs",
2918 .data = &dqstats.stat[DQST_SYNCS],
2919 .maxlen = sizeof(int),
2921 .proc_handler = do_proc_dqstats,
2923 #ifdef CONFIG_PRINT_QUOTA_WARNING
2925 .procname = "warnings",
2926 .data = &flag_print_warnings,
2927 .maxlen = sizeof(int),
2929 .proc_handler = proc_dointvec,
2935 static struct ctl_table fs_table[] = {
2937 .procname = "quota",
2939 .child = fs_dqstats_table,
2944 static struct ctl_table sys_table[] = {
2953 static int __init dquot_init(void)
2956 unsigned long nr_hash, order;
2958 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2960 register_sysctl_table(sys_table);
2962 dquot_cachep = kmem_cache_create("dquot",
2963 sizeof(struct dquot), sizeof(unsigned long) * 4,
2964 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2965 SLAB_MEM_SPREAD|SLAB_PANIC),
2969 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2971 panic("Cannot create dquot hash table");
2973 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2974 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2976 panic("Cannot create dquot stat counters");
2979 /* Find power-of-two hlist_heads which can fit into allocation */
2980 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2984 } while (nr_hash >> dq_hash_bits);
2987 nr_hash = 1UL << dq_hash_bits;
2988 dq_hash_mask = nr_hash - 1;
2989 for (i = 0; i < nr_hash; i++)
2990 INIT_HLIST_HEAD(dquot_hash + i);
2992 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2993 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2995 if (register_shrinker(&dqcache_shrinker))
2996 panic("Cannot register dquot shrinker");
3000 fs_initcall(dquot_init);