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 "../internal.h" /* ugh */
84 #include <linux/uaccess.h>
87 * There are five quota SMP locks:
88 * * dq_list_lock protects all lists with quotas and quota formats.
89 * * dquot->dq_dqb_lock protects data from dq_dqb
90 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
91 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
92 * dquot_transfer() can stabilize amount it transfers
93 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
94 * pointers in the inode
95 * * dq_state_lock protects modifications of quota state (on quotaon and
96 * quotaoff) and readers who care about latest values take it as well.
98 * The spinlock ordering is hence:
99 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
100 * dq_list_lock > dq_state_lock
102 * Note that some things (eg. sb pointer, type, id) doesn't change during
103 * the life of the dquot structure and so needn't to be protected by a lock
105 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
106 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
107 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
108 * inode and before dropping dquot references to avoid use of dquots after
109 * they are freed. dq_data_lock is used to serialize the pointer setting and
110 * clearing operations.
111 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
112 * inode is a quota file). Functions adding pointers from inode to dquots have
113 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
114 * have to do all pointer modifications before dropping dq_data_lock. This makes
115 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
116 * then drops all pointers to dquots from an inode.
118 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
119 * memory (or space for it is being allocated) on the first dqget(), when it is
120 * being written out, and when it is being released on the last dqput(). The
121 * allocation and release operations are serialized by the dq_lock and by
122 * checking the use count in dquot_release().
124 * Lock ordering (including related VFS locks) is the following:
125 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
128 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
129 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
130 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
131 EXPORT_SYMBOL(dq_data_lock);
132 DEFINE_STATIC_SRCU(dquot_srcu);
134 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
136 void __quota_error(struct super_block *sb, const char *func,
137 const char *fmt, ...)
139 if (printk_ratelimit()) {
141 struct va_format vaf;
148 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
149 sb->s_id, func, &vaf);
154 EXPORT_SYMBOL(__quota_error);
156 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
157 static char *quotatypes[] = INITQFNAMES;
159 static struct quota_format_type *quota_formats; /* List of registered formats */
160 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
162 /* SLAB cache for dquot structures */
163 static struct kmem_cache *dquot_cachep;
165 int register_quota_format(struct quota_format_type *fmt)
167 spin_lock(&dq_list_lock);
168 fmt->qf_next = quota_formats;
170 spin_unlock(&dq_list_lock);
173 EXPORT_SYMBOL(register_quota_format);
175 void unregister_quota_format(struct quota_format_type *fmt)
177 struct quota_format_type **actqf;
179 spin_lock(&dq_list_lock);
180 for (actqf = "a_formats; *actqf && *actqf != fmt;
181 actqf = &(*actqf)->qf_next)
184 *actqf = (*actqf)->qf_next;
185 spin_unlock(&dq_list_lock);
187 EXPORT_SYMBOL(unregister_quota_format);
189 static struct quota_format_type *find_quota_format(int id)
191 struct quota_format_type *actqf;
193 spin_lock(&dq_list_lock);
194 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
195 actqf = actqf->qf_next)
197 if (!actqf || !try_module_get(actqf->qf_owner)) {
200 spin_unlock(&dq_list_lock);
202 for (qm = 0; module_names[qm].qm_fmt_id &&
203 module_names[qm].qm_fmt_id != id; qm++)
205 if (!module_names[qm].qm_fmt_id ||
206 request_module(module_names[qm].qm_mod_name))
209 spin_lock(&dq_list_lock);
210 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
211 actqf = actqf->qf_next)
213 if (actqf && !try_module_get(actqf->qf_owner))
216 spin_unlock(&dq_list_lock);
220 static void put_quota_format(struct quota_format_type *fmt)
222 module_put(fmt->qf_owner);
226 * Dquot List Management:
227 * The quota code uses four lists for dquot management: the inuse_list,
228 * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
229 * structure may be on some of those lists, depending on its current state.
231 * All dquots are placed to the end of inuse_list when first created, and this
232 * list is used for invalidate operation, which must look at every dquot.
234 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
235 * and this list is searched whenever we need an available dquot. Dquots are
236 * removed from the list as soon as they are used again, and
237 * dqstats.free_dquots gives the number of dquots on the list. When
238 * dquot is invalidated it's completely released from memory.
240 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
241 * dirtied, and this list is searched when writing dirty dquots back to
242 * quota file. Note that some filesystems do dirty dquot tracking on their
243 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
245 * Dquots with a specific identity (device, type and id) are placed on
246 * one of the dquot_hash[] hash chains. The provides an efficient search
247 * mechanism to locate a specific dquot.
250 static LIST_HEAD(inuse_list);
251 static LIST_HEAD(free_dquots);
252 static unsigned int dq_hash_bits, dq_hash_mask;
253 static struct hlist_head *dquot_hash;
255 struct dqstats dqstats;
256 EXPORT_SYMBOL(dqstats);
258 static qsize_t inode_get_rsv_space(struct inode *inode);
259 static qsize_t __inode_get_rsv_space(struct inode *inode);
260 static int __dquot_initialize(struct inode *inode, int type);
262 static inline unsigned int
263 hashfn(const struct super_block *sb, struct kqid qid)
265 unsigned int id = from_kqid(&init_user_ns, qid);
269 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
270 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
274 * Following list functions expect dq_list_lock to be held
276 static inline void insert_dquot_hash(struct dquot *dquot)
278 struct hlist_head *head;
279 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
280 hlist_add_head(&dquot->dq_hash, head);
283 static inline void remove_dquot_hash(struct dquot *dquot)
285 hlist_del_init(&dquot->dq_hash);
288 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
291 struct hlist_node *node;
294 hlist_for_each (node, dquot_hash+hashent) {
295 dquot = hlist_entry(node, struct dquot, dq_hash);
296 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
302 /* Add a dquot to the tail of the free list */
303 static inline void put_dquot_last(struct dquot *dquot)
305 list_add_tail(&dquot->dq_free, &free_dquots);
306 dqstats_inc(DQST_FREE_DQUOTS);
309 static inline void remove_free_dquot(struct dquot *dquot)
311 if (list_empty(&dquot->dq_free))
313 list_del_init(&dquot->dq_free);
314 dqstats_dec(DQST_FREE_DQUOTS);
317 static inline void put_inuse(struct dquot *dquot)
319 /* We add to the back of inuse list so we don't have to restart
320 * when traversing this list and we block */
321 list_add_tail(&dquot->dq_inuse, &inuse_list);
322 dqstats_inc(DQST_ALLOC_DQUOTS);
325 static inline void remove_inuse(struct dquot *dquot)
327 dqstats_dec(DQST_ALLOC_DQUOTS);
328 list_del(&dquot->dq_inuse);
331 * End of list functions needing dq_list_lock
334 static void wait_on_dquot(struct dquot *dquot)
336 mutex_lock(&dquot->dq_lock);
337 mutex_unlock(&dquot->dq_lock);
340 static inline int dquot_dirty(struct dquot *dquot)
342 return test_bit(DQ_MOD_B, &dquot->dq_flags);
345 static inline int mark_dquot_dirty(struct dquot *dquot)
347 return dquot->dq_sb->dq_op->mark_dirty(dquot);
350 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
351 int dquot_mark_dquot_dirty(struct dquot *dquot)
355 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
358 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
359 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
361 /* If quota is dirty already, we don't have to acquire dq_list_lock */
362 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
365 spin_lock(&dq_list_lock);
366 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
367 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
368 info[dquot->dq_id.type].dqi_dirty_list);
371 spin_unlock(&dq_list_lock);
374 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
376 /* Dirtify all the dquots - this can block when journalling */
377 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
382 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
384 /* Even in case of error we have to continue */
385 ret = mark_dquot_dirty(dquot[cnt]);
392 static inline void dqput_all(struct dquot **dquot)
396 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
400 static inline int clear_dquot_dirty(struct dquot *dquot)
402 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
403 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
405 spin_lock(&dq_list_lock);
406 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
407 spin_unlock(&dq_list_lock);
410 list_del_init(&dquot->dq_dirty);
411 spin_unlock(&dq_list_lock);
415 void mark_info_dirty(struct super_block *sb, int type)
417 spin_lock(&dq_data_lock);
418 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
419 spin_unlock(&dq_data_lock);
421 EXPORT_SYMBOL(mark_info_dirty);
424 * Read dquot from disk and alloc space for it
427 int dquot_acquire(struct dquot *dquot)
429 int ret = 0, ret2 = 0;
430 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
432 mutex_lock(&dquot->dq_lock);
433 if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
434 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
438 /* Make sure flags update is visible after dquot has been filled */
439 smp_mb__before_atomic();
440 set_bit(DQ_READ_B, &dquot->dq_flags);
441 /* Instantiate dquot if needed */
442 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
443 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
444 /* Write the info if needed */
445 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
446 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
447 dquot->dq_sb, dquot->dq_id.type);
457 * Make sure flags update is visible after on-disk struct has been
458 * allocated. Paired with smp_rmb() in dqget().
460 smp_mb__before_atomic();
461 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
463 mutex_unlock(&dquot->dq_lock);
466 EXPORT_SYMBOL(dquot_acquire);
469 * Write dquot to disk
471 int dquot_commit(struct dquot *dquot)
474 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
476 mutex_lock(&dquot->dq_lock);
477 if (!clear_dquot_dirty(dquot))
479 /* Inactive dquot can be only if there was error during read/init
480 * => we have better not writing it */
481 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
482 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
486 mutex_unlock(&dquot->dq_lock);
489 EXPORT_SYMBOL(dquot_commit);
494 int dquot_release(struct dquot *dquot)
496 int ret = 0, ret2 = 0;
497 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
499 mutex_lock(&dquot->dq_lock);
500 /* Check whether we are not racing with some other dqget() */
501 if (dquot_is_busy(dquot))
503 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
504 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
506 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
507 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
508 dquot->dq_sb, dquot->dq_id.type);
513 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
515 mutex_unlock(&dquot->dq_lock);
518 EXPORT_SYMBOL(dquot_release);
520 void dquot_destroy(struct dquot *dquot)
522 kmem_cache_free(dquot_cachep, dquot);
524 EXPORT_SYMBOL(dquot_destroy);
526 static inline void do_destroy_dquot(struct dquot *dquot)
528 dquot->dq_sb->dq_op->destroy_dquot(dquot);
531 /* Invalidate all dquots on the list. Note that this function is called after
532 * quota is disabled and pointers from inodes removed so there cannot be new
533 * quota users. There can still be some users of quotas due to inodes being
534 * just deleted or pruned by prune_icache() (those are not attached to any
535 * list) or parallel quotactl call. We have to wait for such users.
537 static void invalidate_dquots(struct super_block *sb, int type)
539 struct dquot *dquot, *tmp;
542 spin_lock(&dq_list_lock);
543 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
544 if (dquot->dq_sb != sb)
546 if (dquot->dq_id.type != type)
548 /* Wait for dquot users */
549 if (atomic_read(&dquot->dq_count)) {
551 spin_unlock(&dq_list_lock);
553 * Once dqput() wakes us up, we know it's time to free
555 * IMPORTANT: we rely on the fact that there is always
556 * at most one process waiting for dquot to free.
557 * Otherwise dq_count would be > 1 and we would never
560 wait_event(dquot_ref_wq,
561 atomic_read(&dquot->dq_count) == 1);
563 /* At this moment dquot() need not exist (it could be
564 * reclaimed by prune_dqcache(). Hence we must
569 * Quota now has no users and it has been written on last
572 remove_dquot_hash(dquot);
573 remove_free_dquot(dquot);
575 do_destroy_dquot(dquot);
577 spin_unlock(&dq_list_lock);
580 /* Call callback for every active dquot on given filesystem */
581 int dquot_scan_active(struct super_block *sb,
582 int (*fn)(struct dquot *dquot, unsigned long priv),
585 struct dquot *dquot, *old_dquot = NULL;
588 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
590 spin_lock(&dq_list_lock);
591 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
592 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
594 if (dquot->dq_sb != sb)
596 /* Now we have active dquot so we can just increase use count */
597 atomic_inc(&dquot->dq_count);
598 spin_unlock(&dq_list_lock);
602 * ->release_dquot() can be racing with us. Our reference
603 * protects us from new calls to it so just wait for any
604 * outstanding call and recheck the DQ_ACTIVE_B after that.
606 wait_on_dquot(dquot);
607 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
608 ret = fn(dquot, priv);
612 spin_lock(&dq_list_lock);
613 /* We are safe to continue now because our dquot could not
614 * be moved out of the inuse list while we hold the reference */
616 spin_unlock(&dq_list_lock);
621 EXPORT_SYMBOL(dquot_scan_active);
623 /* Write all dquot structures to quota files */
624 int dquot_writeback_dquots(struct super_block *sb, int type)
626 struct list_head dirty;
628 struct quota_info *dqopt = sb_dqopt(sb);
632 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
634 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
635 if (type != -1 && cnt != type)
637 if (!sb_has_quota_active(sb, cnt))
639 spin_lock(&dq_list_lock);
640 /* Move list away to avoid livelock. */
641 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
642 while (!list_empty(&dirty)) {
643 dquot = list_first_entry(&dirty, struct dquot,
646 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
648 /* Now we have active dquot from which someone is
649 * holding reference so we can safely just increase
652 spin_unlock(&dq_list_lock);
653 err = sb->dq_op->write_dquot(dquot);
656 * Clear dirty bit anyway to avoid infinite
659 clear_dquot_dirty(dquot);
664 spin_lock(&dq_list_lock);
666 spin_unlock(&dq_list_lock);
669 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
670 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
671 && info_dirty(&dqopt->info[cnt]))
672 sb->dq_op->write_info(sb, cnt);
673 dqstats_inc(DQST_SYNCS);
677 EXPORT_SYMBOL(dquot_writeback_dquots);
679 /* Write all dquot structures to disk and make them visible from userspace */
680 int dquot_quota_sync(struct super_block *sb, int type)
682 struct quota_info *dqopt = sb_dqopt(sb);
686 ret = dquot_writeback_dquots(sb, type);
689 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
692 /* This is not very clever (and fast) but currently I don't know about
693 * any other simple way of getting quota data to disk and we must get
694 * them there for userspace to be visible... */
695 if (sb->s_op->sync_fs)
696 sb->s_op->sync_fs(sb, 1);
697 sync_blockdev(sb->s_bdev);
700 * Now when everything is written we can discard the pagecache so
701 * that userspace sees the changes.
703 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
704 if (type != -1 && cnt != type)
706 if (!sb_has_quota_active(sb, cnt))
708 inode_lock(dqopt->files[cnt]);
709 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
710 inode_unlock(dqopt->files[cnt]);
715 EXPORT_SYMBOL(dquot_quota_sync);
718 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
721 unsigned long freed = 0;
723 spin_lock(&dq_list_lock);
724 while (!list_empty(&free_dquots) && sc->nr_to_scan) {
725 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
726 remove_dquot_hash(dquot);
727 remove_free_dquot(dquot);
729 do_destroy_dquot(dquot);
733 spin_unlock(&dq_list_lock);
738 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
740 return vfs_pressure_ratio(
741 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
744 static struct shrinker dqcache_shrinker = {
745 .count_objects = dqcache_shrink_count,
746 .scan_objects = dqcache_shrink_scan,
747 .seeks = DEFAULT_SEEKS,
751 * Put reference to dquot
753 void dqput(struct dquot *dquot)
759 #ifdef CONFIG_QUOTA_DEBUG
760 if (!atomic_read(&dquot->dq_count)) {
761 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
762 quotatypes[dquot->dq_id.type],
763 from_kqid(&init_user_ns, dquot->dq_id));
767 dqstats_inc(DQST_DROPS);
769 spin_lock(&dq_list_lock);
770 if (atomic_read(&dquot->dq_count) > 1) {
771 /* We have more than one user... nothing to do */
772 atomic_dec(&dquot->dq_count);
773 /* Releasing dquot during quotaoff phase? */
774 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
775 atomic_read(&dquot->dq_count) == 1)
776 wake_up(&dquot_ref_wq);
777 spin_unlock(&dq_list_lock);
780 /* Need to release dquot? */
781 if (dquot_dirty(dquot)) {
782 spin_unlock(&dq_list_lock);
783 /* Commit dquot before releasing */
784 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
786 quota_error(dquot->dq_sb, "Can't write quota structure"
787 " (error %d). Quota may get out of sync!",
790 * We clear dirty bit anyway, so that we avoid
793 clear_dquot_dirty(dquot);
797 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
798 spin_unlock(&dq_list_lock);
799 dquot->dq_sb->dq_op->release_dquot(dquot);
802 atomic_dec(&dquot->dq_count);
803 #ifdef CONFIG_QUOTA_DEBUG
805 BUG_ON(!list_empty(&dquot->dq_free));
807 put_dquot_last(dquot);
808 spin_unlock(&dq_list_lock);
810 EXPORT_SYMBOL(dqput);
812 struct dquot *dquot_alloc(struct super_block *sb, int type)
814 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
816 EXPORT_SYMBOL(dquot_alloc);
818 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
822 dquot = sb->dq_op->alloc_dquot(sb, type);
826 mutex_init(&dquot->dq_lock);
827 INIT_LIST_HEAD(&dquot->dq_free);
828 INIT_LIST_HEAD(&dquot->dq_inuse);
829 INIT_HLIST_NODE(&dquot->dq_hash);
830 INIT_LIST_HEAD(&dquot->dq_dirty);
832 dquot->dq_id = make_kqid_invalid(type);
833 atomic_set(&dquot->dq_count, 1);
834 spin_lock_init(&dquot->dq_dqb_lock);
840 * Get reference to dquot
842 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
843 * destroying our dquot by:
844 * a) checking for quota flags under dq_list_lock and
845 * b) getting a reference to dquot before we release dq_list_lock
847 struct dquot *dqget(struct super_block *sb, struct kqid qid)
849 unsigned int hashent = hashfn(sb, qid);
850 struct dquot *dquot, *empty = NULL;
852 if (!qid_has_mapping(sb->s_user_ns, qid))
853 return ERR_PTR(-EINVAL);
855 if (!sb_has_quota_active(sb, qid.type))
856 return ERR_PTR(-ESRCH);
858 spin_lock(&dq_list_lock);
859 spin_lock(&dq_state_lock);
860 if (!sb_has_quota_active(sb, qid.type)) {
861 spin_unlock(&dq_state_lock);
862 spin_unlock(&dq_list_lock);
863 dquot = ERR_PTR(-ESRCH);
866 spin_unlock(&dq_state_lock);
868 dquot = find_dquot(hashent, sb, qid);
871 spin_unlock(&dq_list_lock);
872 empty = get_empty_dquot(sb, qid.type);
874 schedule(); /* Try to wait for a moment... */
880 /* all dquots go on the inuse_list */
882 /* hash it first so it can be found */
883 insert_dquot_hash(dquot);
884 spin_unlock(&dq_list_lock);
885 dqstats_inc(DQST_LOOKUPS);
887 if (!atomic_read(&dquot->dq_count))
888 remove_free_dquot(dquot);
889 atomic_inc(&dquot->dq_count);
890 spin_unlock(&dq_list_lock);
891 dqstats_inc(DQST_CACHE_HITS);
892 dqstats_inc(DQST_LOOKUPS);
894 /* Wait for dq_lock - after this we know that either dquot_release() is
895 * already finished or it will be canceled due to dq_count > 1 test */
896 wait_on_dquot(dquot);
897 /* Read the dquot / allocate space in quota file */
898 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
901 err = sb->dq_op->acquire_dquot(dquot);
904 dquot = ERR_PTR(err);
909 * Make sure following reads see filled structure - paired with
910 * smp_mb__before_atomic() in dquot_acquire().
913 #ifdef CONFIG_QUOTA_DEBUG
914 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
918 do_destroy_dquot(empty);
922 EXPORT_SYMBOL(dqget);
924 static inline struct dquot **i_dquot(struct inode *inode)
926 return inode->i_sb->s_op->get_dquots(inode);
929 static int dqinit_needed(struct inode *inode, int type)
931 struct dquot * const *dquots;
934 if (IS_NOQUOTA(inode))
937 dquots = i_dquot(inode);
939 return !dquots[type];
940 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
946 /* This routine is guarded by s_umount semaphore */
947 static int add_dquot_ref(struct super_block *sb, int type)
949 struct inode *inode, *old_inode = NULL;
950 #ifdef CONFIG_QUOTA_DEBUG
955 spin_lock(&sb->s_inode_list_lock);
956 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
957 spin_lock(&inode->i_lock);
958 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
959 !atomic_read(&inode->i_writecount) ||
960 !dqinit_needed(inode, type)) {
961 spin_unlock(&inode->i_lock);
965 spin_unlock(&inode->i_lock);
966 spin_unlock(&sb->s_inode_list_lock);
968 #ifdef CONFIG_QUOTA_DEBUG
969 if (unlikely(inode_get_rsv_space(inode) > 0))
973 err = __dquot_initialize(inode, type);
980 * We hold a reference to 'inode' so it couldn't have been
981 * removed from s_inodes list while we dropped the
982 * s_inode_list_lock. We cannot iput the inode now as we can be
983 * holding the last reference and we cannot iput it under
984 * s_inode_list_lock. So we keep the reference and iput it
989 spin_lock(&sb->s_inode_list_lock);
991 spin_unlock(&sb->s_inode_list_lock);
994 #ifdef CONFIG_QUOTA_DEBUG
996 quota_error(sb, "Writes happened before quota was turned on "
997 "thus quota information is probably inconsistent. "
998 "Please run quotacheck(8)");
1005 * Remove references to dquots from inode and add dquot to list for freeing
1006 * if we have the last reference to dquot
1008 static void remove_inode_dquot_ref(struct inode *inode, int type,
1009 struct list_head *tofree_head)
1011 struct dquot **dquots = i_dquot(inode);
1012 struct dquot *dquot = dquots[type];
1017 dquots[type] = NULL;
1018 if (list_empty(&dquot->dq_free)) {
1020 * The inode still has reference to dquot so it can't be in the
1023 spin_lock(&dq_list_lock);
1024 list_add(&dquot->dq_free, tofree_head);
1025 spin_unlock(&dq_list_lock);
1028 * Dquot is already in a list to put so we won't drop the last
1036 * Free list of dquots
1037 * Dquots are removed from inodes and no new references can be got so we are
1038 * the only ones holding reference
1040 static void put_dquot_list(struct list_head *tofree_head)
1042 struct list_head *act_head;
1043 struct dquot *dquot;
1045 act_head = tofree_head->next;
1046 while (act_head != tofree_head) {
1047 dquot = list_entry(act_head, struct dquot, dq_free);
1048 act_head = act_head->next;
1049 /* Remove dquot from the list so we won't have problems... */
1050 list_del_init(&dquot->dq_free);
1055 static void remove_dquot_ref(struct super_block *sb, int type,
1056 struct list_head *tofree_head)
1058 struct inode *inode;
1059 #ifdef CONFIG_QUOTA_DEBUG
1063 spin_lock(&sb->s_inode_list_lock);
1064 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1066 * We have to scan also I_NEW inodes because they can already
1067 * have quota pointer initialized. Luckily, we need to touch
1068 * only quota pointers and these have separate locking
1071 spin_lock(&dq_data_lock);
1072 if (!IS_NOQUOTA(inode)) {
1073 #ifdef CONFIG_QUOTA_DEBUG
1074 if (unlikely(inode_get_rsv_space(inode) > 0))
1077 remove_inode_dquot_ref(inode, type, tofree_head);
1079 spin_unlock(&dq_data_lock);
1081 spin_unlock(&sb->s_inode_list_lock);
1082 #ifdef CONFIG_QUOTA_DEBUG
1084 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1085 " was disabled thus quota information is probably "
1086 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1091 /* Gather all references from inodes and drop them */
1092 static void drop_dquot_ref(struct super_block *sb, int type)
1094 LIST_HEAD(tofree_head);
1097 remove_dquot_ref(sb, type, &tofree_head);
1098 synchronize_srcu(&dquot_srcu);
1099 put_dquot_list(&tofree_head);
1104 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1106 if (dquot->dq_dqb.dqb_rsvspace >= number)
1107 dquot->dq_dqb.dqb_rsvspace -= number;
1110 dquot->dq_dqb.dqb_rsvspace = 0;
1112 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1113 dquot->dq_dqb.dqb_bsoftlimit)
1114 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1115 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1118 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1120 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1121 dquot->dq_dqb.dqb_curinodes >= number)
1122 dquot->dq_dqb.dqb_curinodes -= number;
1124 dquot->dq_dqb.dqb_curinodes = 0;
1125 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1126 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1127 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1130 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1132 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1133 dquot->dq_dqb.dqb_curspace >= number)
1134 dquot->dq_dqb.dqb_curspace -= number;
1136 dquot->dq_dqb.dqb_curspace = 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);
1144 struct super_block *w_sb;
1145 struct kqid w_dq_id;
1149 static int warning_issued(struct dquot *dquot, const int warntype)
1151 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1152 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1153 ((warntype == QUOTA_NL_IHARDWARN ||
1154 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1158 return test_and_set_bit(flag, &dquot->dq_flags);
1161 #ifdef CONFIG_PRINT_QUOTA_WARNING
1162 static int flag_print_warnings = 1;
1164 static int need_print_warning(struct dquot_warn *warn)
1166 if (!flag_print_warnings)
1169 switch (warn->w_dq_id.type) {
1171 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1173 return in_group_p(warn->w_dq_id.gid);
1180 /* Print warning to user which exceeded quota */
1181 static void print_warning(struct dquot_warn *warn)
1184 struct tty_struct *tty;
1185 int warntype = warn->w_type;
1187 if (warntype == QUOTA_NL_IHARDBELOW ||
1188 warntype == QUOTA_NL_ISOFTBELOW ||
1189 warntype == QUOTA_NL_BHARDBELOW ||
1190 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1193 tty = get_current_tty();
1196 tty_write_message(tty, warn->w_sb->s_id);
1197 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1198 tty_write_message(tty, ": warning, ");
1200 tty_write_message(tty, ": write failed, ");
1201 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1203 case QUOTA_NL_IHARDWARN:
1204 msg = " file limit reached.\r\n";
1206 case QUOTA_NL_ISOFTLONGWARN:
1207 msg = " file quota exceeded too long.\r\n";
1209 case QUOTA_NL_ISOFTWARN:
1210 msg = " file quota exceeded.\r\n";
1212 case QUOTA_NL_BHARDWARN:
1213 msg = " block limit reached.\r\n";
1215 case QUOTA_NL_BSOFTLONGWARN:
1216 msg = " block quota exceeded too long.\r\n";
1218 case QUOTA_NL_BSOFTWARN:
1219 msg = " block quota exceeded.\r\n";
1222 tty_write_message(tty, msg);
1227 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1230 if (warning_issued(dquot, warntype))
1232 warn->w_type = warntype;
1233 warn->w_sb = dquot->dq_sb;
1234 warn->w_dq_id = dquot->dq_id;
1238 * Write warnings to the console and send warning messages over netlink.
1240 * Note that this function can call into tty and networking code.
1242 static void flush_warnings(struct dquot_warn *warn)
1246 for (i = 0; i < MAXQUOTAS; i++) {
1247 if (warn[i].w_type == QUOTA_NL_NOWARN)
1249 #ifdef CONFIG_PRINT_QUOTA_WARNING
1250 print_warning(&warn[i]);
1252 quota_send_warning(warn[i].w_dq_id,
1253 warn[i].w_sb->s_dev, warn[i].w_type);
1257 static int ignore_hardlimit(struct dquot *dquot)
1259 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1261 return capable(CAP_SYS_RESOURCE) &&
1262 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1263 !(info->dqi_flags & DQF_ROOT_SQUASH));
1266 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1267 struct dquot_warn *warn)
1272 spin_lock(&dquot->dq_dqb_lock);
1273 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1274 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1275 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1278 if (dquot->dq_dqb.dqb_ihardlimit &&
1279 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1280 !ignore_hardlimit(dquot)) {
1281 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1286 if (dquot->dq_dqb.dqb_isoftlimit &&
1287 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1288 dquot->dq_dqb.dqb_itime &&
1289 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1290 !ignore_hardlimit(dquot)) {
1291 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1296 if (dquot->dq_dqb.dqb_isoftlimit &&
1297 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1298 dquot->dq_dqb.dqb_itime == 0) {
1299 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1300 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1301 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1304 dquot->dq_dqb.dqb_curinodes = newinodes;
1307 spin_unlock(&dquot->dq_dqb_lock);
1311 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1312 qsize_t rsv_space, unsigned int flags,
1313 struct dquot_warn *warn)
1316 struct super_block *sb = dquot->dq_sb;
1319 spin_lock(&dquot->dq_dqb_lock);
1320 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1321 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1324 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1325 + space + rsv_space;
1327 if (dquot->dq_dqb.dqb_bhardlimit &&
1328 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1329 !ignore_hardlimit(dquot)) {
1330 if (flags & DQUOT_SPACE_WARN)
1331 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1336 if (dquot->dq_dqb.dqb_bsoftlimit &&
1337 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1338 dquot->dq_dqb.dqb_btime &&
1339 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1340 !ignore_hardlimit(dquot)) {
1341 if (flags & DQUOT_SPACE_WARN)
1342 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1347 if (dquot->dq_dqb.dqb_bsoftlimit &&
1348 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1349 dquot->dq_dqb.dqb_btime == 0) {
1350 if (flags & DQUOT_SPACE_WARN) {
1351 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1352 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1353 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1356 * We don't allow preallocation to exceed softlimit so exceeding will
1365 * We have to be careful and go through warning generation & grace time
1366 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1369 if (flags & DQUOT_SPACE_NOFAIL)
1372 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1373 dquot->dq_dqb.dqb_curspace += space;
1375 spin_unlock(&dquot->dq_dqb_lock);
1379 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1383 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1384 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1385 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1386 return QUOTA_NL_NOWARN;
1388 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1389 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1390 return QUOTA_NL_ISOFTBELOW;
1391 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1392 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1393 return QUOTA_NL_IHARDBELOW;
1394 return QUOTA_NL_NOWARN;
1397 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1401 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1403 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1404 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1405 return QUOTA_NL_NOWARN;
1407 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1408 return QUOTA_NL_BSOFTBELOW;
1409 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1410 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1411 return QUOTA_NL_BHARDBELOW;
1412 return QUOTA_NL_NOWARN;
1415 static int dquot_active(const struct inode *inode)
1417 struct super_block *sb = inode->i_sb;
1419 if (IS_NOQUOTA(inode))
1421 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1425 * Initialize quota pointers in inode
1427 * It is better to call this function outside of any transaction as it
1428 * might need a lot of space in journal for dquot structure allocation.
1430 static int __dquot_initialize(struct inode *inode, int type)
1432 int cnt, init_needed = 0;
1433 struct dquot **dquots, *got[MAXQUOTAS] = {};
1434 struct super_block *sb = inode->i_sb;
1438 if (!dquot_active(inode))
1441 dquots = i_dquot(inode);
1443 /* First get references to structures we might need. */
1444 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1448 struct dquot *dquot;
1450 if (type != -1 && cnt != type)
1453 * The i_dquot should have been initialized in most cases,
1454 * we check it without locking here to avoid unnecessary
1455 * dqget()/dqput() calls.
1460 if (!sb_has_quota_active(sb, cnt))
1467 qid = make_kqid_uid(inode->i_uid);
1470 qid = make_kqid_gid(inode->i_gid);
1473 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1476 qid = make_kqid_projid(projid);
1479 dquot = dqget(sb, qid);
1480 if (IS_ERR(dquot)) {
1481 /* We raced with somebody turning quotas off... */
1482 if (PTR_ERR(dquot) != -ESRCH) {
1483 ret = PTR_ERR(dquot);
1491 /* All required i_dquot has been initialized */
1495 spin_lock(&dq_data_lock);
1496 if (IS_NOQUOTA(inode))
1498 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1499 if (type != -1 && cnt != type)
1501 /* Avoid races with quotaoff() */
1502 if (!sb_has_quota_active(sb, cnt))
1504 /* We could race with quotaon or dqget() could have failed */
1508 dquots[cnt] = got[cnt];
1511 * Make quota reservation system happy if someone
1512 * did a write before quota was turned on
1514 rsv = inode_get_rsv_space(inode);
1515 if (unlikely(rsv)) {
1516 spin_lock(&inode->i_lock);
1517 /* Get reservation again under proper lock */
1518 rsv = __inode_get_rsv_space(inode);
1519 spin_lock(&dquots[cnt]->dq_dqb_lock);
1520 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1521 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1522 spin_unlock(&inode->i_lock);
1527 spin_unlock(&dq_data_lock);
1529 /* Drop unused references */
1535 int dquot_initialize(struct inode *inode)
1537 return __dquot_initialize(inode, -1);
1539 EXPORT_SYMBOL(dquot_initialize);
1541 bool dquot_initialize_needed(struct inode *inode)
1543 struct dquot **dquots;
1546 if (!dquot_active(inode))
1549 dquots = i_dquot(inode);
1550 for (i = 0; i < MAXQUOTAS; i++)
1551 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1555 EXPORT_SYMBOL(dquot_initialize_needed);
1558 * Release all quotas referenced by inode.
1560 * This function only be called on inode free or converting
1561 * a file to quota file, no other users for the i_dquot in
1562 * both cases, so we needn't call synchronize_srcu() after
1565 static void __dquot_drop(struct inode *inode)
1568 struct dquot **dquots = i_dquot(inode);
1569 struct dquot *put[MAXQUOTAS];
1571 spin_lock(&dq_data_lock);
1572 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1573 put[cnt] = dquots[cnt];
1576 spin_unlock(&dq_data_lock);
1580 void dquot_drop(struct inode *inode)
1582 struct dquot * const *dquots;
1585 if (IS_NOQUOTA(inode))
1589 * Test before calling to rule out calls from proc and such
1590 * where we are not allowed to block. Note that this is
1591 * actually reliable test even without the lock - the caller
1592 * must assure that nobody can come after the DQUOT_DROP and
1593 * add quota pointers back anyway.
1595 dquots = i_dquot(inode);
1596 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1601 if (cnt < MAXQUOTAS)
1602 __dquot_drop(inode);
1604 EXPORT_SYMBOL(dquot_drop);
1607 * inode_reserved_space is managed internally by quota, and protected by
1608 * i_lock similar to i_blocks+i_bytes.
1610 static qsize_t *inode_reserved_space(struct inode * inode)
1612 /* Filesystem must explicitly define it's own method in order to use
1613 * quota reservation interface */
1614 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1615 return inode->i_sb->dq_op->get_reserved_space(inode);
1618 static qsize_t __inode_get_rsv_space(struct inode *inode)
1620 if (!inode->i_sb->dq_op->get_reserved_space)
1622 return *inode_reserved_space(inode);
1625 static qsize_t inode_get_rsv_space(struct inode *inode)
1629 if (!inode->i_sb->dq_op->get_reserved_space)
1631 spin_lock(&inode->i_lock);
1632 ret = __inode_get_rsv_space(inode);
1633 spin_unlock(&inode->i_lock);
1638 * This functions updates i_blocks+i_bytes fields and quota information
1639 * (together with appropriate checks).
1641 * NOTE: We absolutely rely on the fact that caller dirties the inode
1642 * (usually helpers in quotaops.h care about this) and holds a handle for
1643 * the current transaction so that dquot write and inode write go into the
1648 * This operation can block, but only after everything is updated
1650 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1652 int cnt, ret = 0, index;
1653 struct dquot_warn warn[MAXQUOTAS];
1654 int reserve = flags & DQUOT_SPACE_RESERVE;
1655 struct dquot **dquots;
1657 if (!dquot_active(inode)) {
1659 spin_lock(&inode->i_lock);
1660 *inode_reserved_space(inode) += number;
1661 spin_unlock(&inode->i_lock);
1663 inode_add_bytes(inode, number);
1668 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1669 warn[cnt].w_type = QUOTA_NL_NOWARN;
1671 dquots = i_dquot(inode);
1672 index = srcu_read_lock(&dquot_srcu);
1673 spin_lock(&inode->i_lock);
1674 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1678 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1681 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1685 /* Back out changes we already did */
1686 for (cnt--; cnt >= 0; cnt--) {
1689 spin_lock(&dquots[cnt]->dq_dqb_lock);
1691 dquot_free_reserved_space(dquots[cnt],
1694 dquot_decr_space(dquots[cnt], number);
1695 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1697 spin_unlock(&inode->i_lock);
1698 goto out_flush_warn;
1702 *inode_reserved_space(inode) += number;
1704 __inode_add_bytes(inode, number);
1705 spin_unlock(&inode->i_lock);
1708 goto out_flush_warn;
1709 mark_all_dquot_dirty(dquots);
1711 srcu_read_unlock(&dquot_srcu, index);
1712 flush_warnings(warn);
1716 EXPORT_SYMBOL(__dquot_alloc_space);
1719 * This operation can block, but only after everything is updated
1721 int dquot_alloc_inode(struct inode *inode)
1723 int cnt, ret = 0, index;
1724 struct dquot_warn warn[MAXQUOTAS];
1725 struct dquot * const *dquots;
1727 if (!dquot_active(inode))
1729 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1730 warn[cnt].w_type = QUOTA_NL_NOWARN;
1732 dquots = i_dquot(inode);
1733 index = srcu_read_lock(&dquot_srcu);
1734 spin_lock(&inode->i_lock);
1735 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1738 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1740 for (cnt--; cnt >= 0; cnt--) {
1743 /* Back out changes we already did */
1744 spin_lock(&dquots[cnt]->dq_dqb_lock);
1745 dquot_decr_inodes(dquots[cnt], 1);
1746 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1753 spin_unlock(&inode->i_lock);
1755 mark_all_dquot_dirty(dquots);
1756 srcu_read_unlock(&dquot_srcu, index);
1757 flush_warnings(warn);
1760 EXPORT_SYMBOL(dquot_alloc_inode);
1763 * Convert in-memory reserved quotas to real consumed quotas
1765 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1767 struct dquot **dquots;
1770 if (!dquot_active(inode)) {
1771 spin_lock(&inode->i_lock);
1772 *inode_reserved_space(inode) -= number;
1773 __inode_add_bytes(inode, number);
1774 spin_unlock(&inode->i_lock);
1778 dquots = i_dquot(inode);
1779 index = srcu_read_lock(&dquot_srcu);
1780 spin_lock(&inode->i_lock);
1781 /* Claim reserved quotas to allocated quotas */
1782 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1784 struct dquot *dquot = dquots[cnt];
1786 spin_lock(&dquot->dq_dqb_lock);
1787 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1788 number = dquot->dq_dqb.dqb_rsvspace;
1789 dquot->dq_dqb.dqb_curspace += number;
1790 dquot->dq_dqb.dqb_rsvspace -= number;
1791 spin_unlock(&dquot->dq_dqb_lock);
1794 /* Update inode bytes */
1795 *inode_reserved_space(inode) -= number;
1796 __inode_add_bytes(inode, number);
1797 spin_unlock(&inode->i_lock);
1798 mark_all_dquot_dirty(dquots);
1799 srcu_read_unlock(&dquot_srcu, index);
1802 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1805 * Convert allocated space back to in-memory reserved quotas
1807 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1809 struct dquot **dquots;
1812 if (!dquot_active(inode)) {
1813 spin_lock(&inode->i_lock);
1814 *inode_reserved_space(inode) += number;
1815 __inode_sub_bytes(inode, number);
1816 spin_unlock(&inode->i_lock);
1820 dquots = i_dquot(inode);
1821 index = srcu_read_lock(&dquot_srcu);
1822 spin_lock(&inode->i_lock);
1823 /* Claim reserved quotas to allocated quotas */
1824 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1826 struct dquot *dquot = dquots[cnt];
1828 spin_lock(&dquot->dq_dqb_lock);
1829 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1830 number = dquot->dq_dqb.dqb_curspace;
1831 dquot->dq_dqb.dqb_rsvspace += number;
1832 dquot->dq_dqb.dqb_curspace -= number;
1833 spin_unlock(&dquot->dq_dqb_lock);
1836 /* Update inode bytes */
1837 *inode_reserved_space(inode) += number;
1838 __inode_sub_bytes(inode, number);
1839 spin_unlock(&inode->i_lock);
1840 mark_all_dquot_dirty(dquots);
1841 srcu_read_unlock(&dquot_srcu, index);
1844 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1847 * This operation can block, but only after everything is updated
1849 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1852 struct dquot_warn warn[MAXQUOTAS];
1853 struct dquot **dquots;
1854 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1856 if (!dquot_active(inode)) {
1858 spin_lock(&inode->i_lock);
1859 *inode_reserved_space(inode) -= number;
1860 spin_unlock(&inode->i_lock);
1862 inode_sub_bytes(inode, number);
1867 dquots = i_dquot(inode);
1868 index = srcu_read_lock(&dquot_srcu);
1869 spin_lock(&inode->i_lock);
1870 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1873 warn[cnt].w_type = QUOTA_NL_NOWARN;
1876 spin_lock(&dquots[cnt]->dq_dqb_lock);
1877 wtype = info_bdq_free(dquots[cnt], number);
1878 if (wtype != QUOTA_NL_NOWARN)
1879 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1881 dquot_free_reserved_space(dquots[cnt], number);
1883 dquot_decr_space(dquots[cnt], number);
1884 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1887 *inode_reserved_space(inode) -= number;
1889 __inode_sub_bytes(inode, number);
1890 spin_unlock(&inode->i_lock);
1894 mark_all_dquot_dirty(dquots);
1896 srcu_read_unlock(&dquot_srcu, index);
1897 flush_warnings(warn);
1899 EXPORT_SYMBOL(__dquot_free_space);
1902 * This operation can block, but only after everything is updated
1904 void dquot_free_inode(struct inode *inode)
1907 struct dquot_warn warn[MAXQUOTAS];
1908 struct dquot * const *dquots;
1911 if (!dquot_active(inode))
1914 dquots = i_dquot(inode);
1915 index = srcu_read_lock(&dquot_srcu);
1916 spin_lock(&inode->i_lock);
1917 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1920 warn[cnt].w_type = QUOTA_NL_NOWARN;
1923 spin_lock(&dquots[cnt]->dq_dqb_lock);
1924 wtype = info_idq_free(dquots[cnt], 1);
1925 if (wtype != QUOTA_NL_NOWARN)
1926 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1927 dquot_decr_inodes(dquots[cnt], 1);
1928 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1930 spin_unlock(&inode->i_lock);
1931 mark_all_dquot_dirty(dquots);
1932 srcu_read_unlock(&dquot_srcu, index);
1933 flush_warnings(warn);
1935 EXPORT_SYMBOL(dquot_free_inode);
1938 * Transfer the number of inode and blocks from one diskquota to an other.
1939 * On success, dquot references in transfer_to are consumed and references
1940 * to original dquots that need to be released are placed there. On failure,
1941 * references are kept untouched.
1943 * This operation can block, but only after everything is updated
1944 * A transaction must be started when entering this function.
1946 * We are holding reference on transfer_from & transfer_to, no need to
1947 * protect them by srcu_read_lock().
1949 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1952 qsize_t rsv_space = 0;
1953 qsize_t inode_usage = 1;
1954 struct dquot *transfer_from[MAXQUOTAS] = {};
1956 char is_valid[MAXQUOTAS] = {};
1957 struct dquot_warn warn_to[MAXQUOTAS];
1958 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1959 struct dquot_warn warn_from_space[MAXQUOTAS];
1961 if (IS_NOQUOTA(inode))
1964 if (inode->i_sb->dq_op->get_inode_usage) {
1965 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1970 /* Initialize the arrays */
1971 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1972 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1973 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1974 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1977 spin_lock(&dq_data_lock);
1978 spin_lock(&inode->i_lock);
1979 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1980 spin_unlock(&inode->i_lock);
1981 spin_unlock(&dq_data_lock);
1984 cur_space = __inode_get_bytes(inode);
1985 rsv_space = __inode_get_rsv_space(inode);
1987 * Build the transfer_from list, check limits, and update usage in
1988 * the target structures.
1990 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1992 * Skip changes for same uid or gid or for turned off quota-type.
1994 if (!transfer_to[cnt])
1996 /* Avoid races with quotaoff() */
1997 if (!sb_has_quota_active(inode->i_sb, cnt))
2000 transfer_from[cnt] = i_dquot(inode)[cnt];
2001 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2005 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2006 DQUOT_SPACE_WARN, &warn_to[cnt]);
2008 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2009 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2010 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2015 /* Decrease usage for source structures and update quota pointers */
2016 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2019 /* Due to IO error we might not have transfer_from[] structure */
2020 if (transfer_from[cnt]) {
2023 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2024 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2025 if (wtype != QUOTA_NL_NOWARN)
2026 prepare_warning(&warn_from_inodes[cnt],
2027 transfer_from[cnt], wtype);
2028 wtype = info_bdq_free(transfer_from[cnt],
2029 cur_space + rsv_space);
2030 if (wtype != QUOTA_NL_NOWARN)
2031 prepare_warning(&warn_from_space[cnt],
2032 transfer_from[cnt], wtype);
2033 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2034 dquot_decr_space(transfer_from[cnt], cur_space);
2035 dquot_free_reserved_space(transfer_from[cnt],
2037 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2039 i_dquot(inode)[cnt] = transfer_to[cnt];
2041 spin_unlock(&inode->i_lock);
2042 spin_unlock(&dq_data_lock);
2044 mark_all_dquot_dirty(transfer_from);
2045 mark_all_dquot_dirty(transfer_to);
2046 flush_warnings(warn_to);
2047 flush_warnings(warn_from_inodes);
2048 flush_warnings(warn_from_space);
2049 /* Pass back references to put */
2050 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2052 transfer_to[cnt] = transfer_from[cnt];
2055 /* Back out changes we already did */
2056 for (cnt--; cnt >= 0; cnt--) {
2059 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2060 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2061 dquot_decr_space(transfer_to[cnt], cur_space);
2062 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2063 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2065 spin_unlock(&inode->i_lock);
2066 spin_unlock(&dq_data_lock);
2067 flush_warnings(warn_to);
2070 EXPORT_SYMBOL(__dquot_transfer);
2072 /* Wrapper for transferring ownership of an inode for uid/gid only
2073 * Called from FSXXX_setattr()
2075 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2077 struct dquot *transfer_to[MAXQUOTAS] = {};
2078 struct dquot *dquot;
2079 struct super_block *sb = inode->i_sb;
2082 if (!dquot_active(inode))
2085 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2086 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2087 if (IS_ERR(dquot)) {
2088 if (PTR_ERR(dquot) != -ESRCH) {
2089 ret = PTR_ERR(dquot);
2094 transfer_to[USRQUOTA] = dquot;
2096 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2097 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2098 if (IS_ERR(dquot)) {
2099 if (PTR_ERR(dquot) != -ESRCH) {
2100 ret = PTR_ERR(dquot);
2105 transfer_to[GRPQUOTA] = dquot;
2107 ret = __dquot_transfer(inode, transfer_to);
2109 dqput_all(transfer_to);
2112 EXPORT_SYMBOL(dquot_transfer);
2115 * Write info of quota file to disk
2117 int dquot_commit_info(struct super_block *sb, int type)
2119 struct quota_info *dqopt = sb_dqopt(sb);
2121 return dqopt->ops[type]->write_file_info(sb, type);
2123 EXPORT_SYMBOL(dquot_commit_info);
2125 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2127 struct quota_info *dqopt = sb_dqopt(sb);
2129 if (!sb_has_quota_active(sb, qid->type))
2131 if (!dqopt->ops[qid->type]->get_next_id)
2133 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2135 EXPORT_SYMBOL(dquot_get_next_id);
2138 * Definitions of diskquota operations.
2140 const struct dquot_operations dquot_operations = {
2141 .write_dquot = dquot_commit,
2142 .acquire_dquot = dquot_acquire,
2143 .release_dquot = dquot_release,
2144 .mark_dirty = dquot_mark_dquot_dirty,
2145 .write_info = dquot_commit_info,
2146 .alloc_dquot = dquot_alloc,
2147 .destroy_dquot = dquot_destroy,
2148 .get_next_id = dquot_get_next_id,
2150 EXPORT_SYMBOL(dquot_operations);
2153 * Generic helper for ->open on filesystems supporting disk quotas.
2155 int dquot_file_open(struct inode *inode, struct file *file)
2159 error = generic_file_open(inode, file);
2160 if (!error && (file->f_mode & FMODE_WRITE))
2161 error = dquot_initialize(inode);
2164 EXPORT_SYMBOL(dquot_file_open);
2166 static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2168 struct quota_info *dqopt = sb_dqopt(sb);
2169 struct inode *inode = dqopt->files[type];
2173 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2175 inode->i_flags &= ~S_NOQUOTA;
2176 inode_unlock(inode);
2178 dqopt->files[type] = NULL;
2183 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2185 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2188 struct quota_info *dqopt = sb_dqopt(sb);
2190 /* s_umount should be held in exclusive mode */
2191 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2192 up_read(&sb->s_umount);
2194 /* Cannot turn off usage accounting without turning off limits, or
2195 * suspend quotas and simultaneously turn quotas off. */
2196 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2197 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2198 DQUOT_USAGE_ENABLED)))
2202 * Skip everything if there's nothing to do. We have to do this because
2203 * sometimes we are called when fill_super() failed and calling
2204 * sync_fs() in such cases does no good.
2206 if (!sb_any_quota_loaded(sb))
2209 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2210 if (type != -1 && cnt != type)
2212 if (!sb_has_quota_loaded(sb, cnt))
2215 if (flags & DQUOT_SUSPENDED) {
2216 spin_lock(&dq_state_lock);
2218 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2219 spin_unlock(&dq_state_lock);
2221 spin_lock(&dq_state_lock);
2222 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2223 /* Turning off suspended quotas? */
2224 if (!sb_has_quota_loaded(sb, cnt) &&
2225 sb_has_quota_suspended(sb, cnt)) {
2226 dqopt->flags &= ~dquot_state_flag(
2227 DQUOT_SUSPENDED, cnt);
2228 spin_unlock(&dq_state_lock);
2229 vfs_cleanup_quota_inode(sb, cnt);
2232 spin_unlock(&dq_state_lock);
2235 /* We still have to keep quota loaded? */
2236 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2239 /* Note: these are blocking operations */
2240 drop_dquot_ref(sb, cnt);
2241 invalidate_dquots(sb, cnt);
2243 * Now all dquots should be invalidated, all writes done so we
2244 * should be only users of the info. No locks needed.
2246 if (info_dirty(&dqopt->info[cnt]))
2247 sb->dq_op->write_info(sb, cnt);
2248 if (dqopt->ops[cnt]->free_file_info)
2249 dqopt->ops[cnt]->free_file_info(sb, cnt);
2250 put_quota_format(dqopt->info[cnt].dqi_format);
2251 dqopt->info[cnt].dqi_flags = 0;
2252 dqopt->info[cnt].dqi_igrace = 0;
2253 dqopt->info[cnt].dqi_bgrace = 0;
2254 dqopt->ops[cnt] = NULL;
2257 /* Skip syncing and setting flags if quota files are hidden */
2258 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2261 /* Sync the superblock so that buffers with quota data are written to
2262 * disk (and so userspace sees correct data afterwards). */
2263 if (sb->s_op->sync_fs)
2264 sb->s_op->sync_fs(sb, 1);
2265 sync_blockdev(sb->s_bdev);
2266 /* Now the quota files are just ordinary files and we can set the
2267 * inode flags back. Moreover we discard the pagecache so that
2268 * userspace sees the writes we did bypassing the pagecache. We
2269 * must also discard the blockdev buffers so that we see the
2270 * changes done by userspace on the next quotaon() */
2271 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2272 if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2273 inode_lock(dqopt->files[cnt]);
2274 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2275 inode_unlock(dqopt->files[cnt]);
2278 invalidate_bdev(sb->s_bdev);
2280 /* We are done when suspending quotas */
2281 if (flags & DQUOT_SUSPENDED)
2284 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2285 if (!sb_has_quota_loaded(sb, cnt))
2286 vfs_cleanup_quota_inode(sb, cnt);
2289 EXPORT_SYMBOL(dquot_disable);
2291 int dquot_quota_off(struct super_block *sb, int type)
2293 return dquot_disable(sb, type,
2294 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2296 EXPORT_SYMBOL(dquot_quota_off);
2299 * Turn quotas on on a device
2302 static int vfs_setup_quota_inode(struct inode *inode, int type)
2304 struct super_block *sb = inode->i_sb;
2305 struct quota_info *dqopt = sb_dqopt(sb);
2307 if (!S_ISREG(inode->i_mode))
2309 if (IS_RDONLY(inode))
2311 if (sb_has_quota_loaded(sb, type))
2314 dqopt->files[type] = igrab(inode);
2315 if (!dqopt->files[type])
2317 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2318 /* We don't want quota and atime on quota files (deadlocks
2319 * possible) Also nobody should write to the file - we use
2320 * special IO operations which ignore the immutable bit. */
2322 inode->i_flags |= S_NOQUOTA;
2323 inode_unlock(inode);
2325 * When S_NOQUOTA is set, remove dquot references as no more
2326 * references can be added
2328 __dquot_drop(inode);
2333 int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2336 struct quota_format_type *fmt = find_quota_format(format_id);
2337 struct quota_info *dqopt = sb_dqopt(sb);
2340 /* Just unsuspend quotas? */
2341 BUG_ON(flags & DQUOT_SUSPENDED);
2342 /* s_umount should be held in exclusive mode */
2343 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2344 up_read(&sb->s_umount);
2348 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2349 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2353 /* Filesystems outside of init_user_ns not yet supported */
2354 if (sb->s_user_ns != &init_user_ns) {
2358 /* Usage always has to be set... */
2359 if (!(flags & DQUOT_USAGE_ENABLED)) {
2363 if (sb_has_quota_loaded(sb, type)) {
2368 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2369 /* As we bypass the pagecache we must now flush all the
2370 * dirty data and invalidate caches so that kernel sees
2371 * changes from userspace. It is not enough to just flush
2372 * the quota file since if blocksize < pagesize, invalidation
2373 * of the cache could fail because of other unrelated dirty
2375 sync_filesystem(sb);
2376 invalidate_bdev(sb->s_bdev);
2380 if (!fmt->qf_ops->check_quota_file(sb, type))
2383 dqopt->ops[type] = fmt->qf_ops;
2384 dqopt->info[type].dqi_format = fmt;
2385 dqopt->info[type].dqi_fmt_id = format_id;
2386 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2387 error = dqopt->ops[type]->read_file_info(sb, type);
2390 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2391 spin_lock(&dq_data_lock);
2392 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2393 spin_unlock(&dq_data_lock);
2395 spin_lock(&dq_state_lock);
2396 dqopt->flags |= dquot_state_flag(flags, type);
2397 spin_unlock(&dq_state_lock);
2399 error = add_dquot_ref(sb, type);
2401 dquot_disable(sb, type, flags);
2405 put_quota_format(fmt);
2409 EXPORT_SYMBOL(dquot_load_quota_sb);
2412 * More powerful function for turning on quotas on given quota inode allowing
2413 * setting of individual quota flags
2415 int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2420 err = vfs_setup_quota_inode(inode, type);
2423 err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2425 vfs_cleanup_quota_inode(inode->i_sb, type);
2428 EXPORT_SYMBOL(dquot_load_quota_inode);
2430 /* Reenable quotas on remount RW */
2431 int dquot_resume(struct super_block *sb, int type)
2433 struct quota_info *dqopt = sb_dqopt(sb);
2437 /* s_umount should be held in exclusive mode */
2438 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2439 up_read(&sb->s_umount);
2441 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2442 if (type != -1 && cnt != type)
2444 if (!sb_has_quota_suspended(sb, cnt))
2447 spin_lock(&dq_state_lock);
2448 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2449 DQUOT_LIMITS_ENABLED,
2451 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2452 spin_unlock(&dq_state_lock);
2454 flags = dquot_generic_flag(flags, cnt);
2455 ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2458 vfs_cleanup_quota_inode(sb, type);
2463 EXPORT_SYMBOL(dquot_resume);
2465 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2466 const struct path *path)
2468 int error = security_quota_on(path->dentry);
2471 /* Quota file not on the same filesystem? */
2472 if (path->dentry->d_sb != sb)
2475 error = dquot_load_quota_inode(d_inode(path->dentry), type,
2476 format_id, DQUOT_USAGE_ENABLED |
2477 DQUOT_LIMITS_ENABLED);
2480 EXPORT_SYMBOL(dquot_quota_on);
2483 * This function is used when filesystem needs to initialize quotas
2484 * during mount time.
2486 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2487 int format_id, int type)
2489 struct dentry *dentry;
2492 dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2494 return PTR_ERR(dentry);
2496 error = security_quota_on(dentry);
2498 error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2499 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2504 EXPORT_SYMBOL(dquot_quota_on_mount);
2506 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2510 struct quota_info *dqopt = sb_dqopt(sb);
2512 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2514 /* Accounting cannot be turned on while fs is mounted */
2515 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2518 for (type = 0; type < MAXQUOTAS; type++) {
2519 if (!(flags & qtype_enforce_flag(type)))
2521 /* Can't enforce without accounting */
2522 if (!sb_has_quota_usage_enabled(sb, type)) {
2526 if (sb_has_quota_limits_enabled(sb, type)) {
2530 spin_lock(&dq_state_lock);
2531 dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2532 spin_unlock(&dq_state_lock);
2536 /* Backout enforcement enablement we already did */
2537 for (type--; type >= 0; type--) {
2538 if (flags & qtype_enforce_flag(type))
2539 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2541 /* Error code translation for better compatibility with XFS */
2547 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2551 struct quota_info *dqopt = sb_dqopt(sb);
2553 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2556 * We don't support turning off accounting via quotactl. In principle
2557 * quota infrastructure can do this but filesystems don't expect
2558 * userspace to be able to do it.
2561 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2564 /* Filter out limits not enabled */
2565 for (type = 0; type < MAXQUOTAS; type++)
2566 if (!sb_has_quota_limits_enabled(sb, type))
2567 flags &= ~qtype_enforce_flag(type);
2571 for (type = 0; type < MAXQUOTAS; type++) {
2572 if (flags & qtype_enforce_flag(type)) {
2573 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2580 /* Backout enforcement disabling we already did */
2581 for (type--; type >= 0; type--) {
2582 if (flags & qtype_enforce_flag(type)) {
2583 spin_lock(&dq_state_lock);
2585 dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2586 spin_unlock(&dq_state_lock);
2592 /* Generic routine for getting common part of quota structure */
2593 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2595 struct mem_dqblk *dm = &dquot->dq_dqb;
2597 memset(di, 0, sizeof(*di));
2598 spin_lock(&dquot->dq_dqb_lock);
2599 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2600 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2601 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2602 di->d_ino_softlimit = dm->dqb_isoftlimit;
2603 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2604 di->d_ino_count = dm->dqb_curinodes;
2605 di->d_spc_timer = dm->dqb_btime;
2606 di->d_ino_timer = dm->dqb_itime;
2607 spin_unlock(&dquot->dq_dqb_lock);
2610 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2611 struct qc_dqblk *di)
2613 struct dquot *dquot;
2615 dquot = dqget(sb, qid);
2617 return PTR_ERR(dquot);
2618 do_get_dqblk(dquot, di);
2623 EXPORT_SYMBOL(dquot_get_dqblk);
2625 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2626 struct qc_dqblk *di)
2628 struct dquot *dquot;
2631 if (!sb->dq_op->get_next_id)
2633 err = sb->dq_op->get_next_id(sb, qid);
2636 dquot = dqget(sb, *qid);
2638 return PTR_ERR(dquot);
2639 do_get_dqblk(dquot, di);
2644 EXPORT_SYMBOL(dquot_get_next_dqblk);
2646 #define VFS_QC_MASK \
2647 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2648 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2649 QC_SPC_TIMER | QC_INO_TIMER)
2651 /* Generic routine for setting common part of quota structure */
2652 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2654 struct mem_dqblk *dm = &dquot->dq_dqb;
2655 int check_blim = 0, check_ilim = 0;
2656 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2658 if (di->d_fieldmask & ~VFS_QC_MASK)
2661 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2662 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2663 ((di->d_fieldmask & QC_SPC_HARD) &&
2664 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2665 ((di->d_fieldmask & QC_INO_SOFT) &&
2666 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2667 ((di->d_fieldmask & QC_INO_HARD) &&
2668 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2671 spin_lock(&dquot->dq_dqb_lock);
2672 if (di->d_fieldmask & QC_SPACE) {
2673 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2675 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2678 if (di->d_fieldmask & QC_SPC_SOFT)
2679 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2680 if (di->d_fieldmask & QC_SPC_HARD)
2681 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2682 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2684 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2687 if (di->d_fieldmask & QC_INO_COUNT) {
2688 dm->dqb_curinodes = di->d_ino_count;
2690 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2693 if (di->d_fieldmask & QC_INO_SOFT)
2694 dm->dqb_isoftlimit = di->d_ino_softlimit;
2695 if (di->d_fieldmask & QC_INO_HARD)
2696 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2697 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2699 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2702 if (di->d_fieldmask & QC_SPC_TIMER) {
2703 dm->dqb_btime = di->d_spc_timer;
2705 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2708 if (di->d_fieldmask & QC_INO_TIMER) {
2709 dm->dqb_itime = di->d_ino_timer;
2711 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2715 if (!dm->dqb_bsoftlimit ||
2716 dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2718 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2719 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2720 /* Set grace only if user hasn't provided his own... */
2721 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2724 if (!dm->dqb_isoftlimit ||
2725 dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2727 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2728 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2729 /* Set grace only if user hasn't provided his own... */
2730 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2732 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2734 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2736 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2737 spin_unlock(&dquot->dq_dqb_lock);
2738 mark_dquot_dirty(dquot);
2743 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2744 struct qc_dqblk *di)
2746 struct dquot *dquot;
2749 dquot = dqget(sb, qid);
2750 if (IS_ERR(dquot)) {
2751 rc = PTR_ERR(dquot);
2754 rc = do_set_dqblk(dquot, di);
2759 EXPORT_SYMBOL(dquot_set_dqblk);
2761 /* Generic routine for getting common part of quota file information */
2762 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2764 struct mem_dqinfo *mi;
2765 struct qc_type_state *tstate;
2766 struct quota_info *dqopt = sb_dqopt(sb);
2769 memset(state, 0, sizeof(*state));
2770 for (type = 0; type < MAXQUOTAS; type++) {
2771 if (!sb_has_quota_active(sb, type))
2773 tstate = state->s_state + type;
2774 mi = sb_dqopt(sb)->info + type;
2775 tstate->flags = QCI_ACCT_ENABLED;
2776 spin_lock(&dq_data_lock);
2777 if (mi->dqi_flags & DQF_SYS_FILE)
2778 tstate->flags |= QCI_SYSFILE;
2779 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2780 tstate->flags |= QCI_ROOT_SQUASH;
2781 if (sb_has_quota_limits_enabled(sb, type))
2782 tstate->flags |= QCI_LIMITS_ENFORCED;
2783 tstate->spc_timelimit = mi->dqi_bgrace;
2784 tstate->ino_timelimit = mi->dqi_igrace;
2785 if (dqopt->files[type]) {
2786 tstate->ino = dqopt->files[type]->i_ino;
2787 tstate->blocks = dqopt->files[type]->i_blocks;
2789 tstate->nextents = 1; /* We don't know... */
2790 spin_unlock(&dq_data_lock);
2794 EXPORT_SYMBOL(dquot_get_state);
2796 /* Generic routine for setting common part of quota file information */
2797 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2799 struct mem_dqinfo *mi;
2802 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2803 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2805 if (!sb_has_quota_active(sb, type))
2807 mi = sb_dqopt(sb)->info + type;
2808 if (ii->i_fieldmask & QC_FLAGS) {
2809 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2810 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2813 spin_lock(&dq_data_lock);
2814 if (ii->i_fieldmask & QC_SPC_TIMER)
2815 mi->dqi_bgrace = ii->i_spc_timelimit;
2816 if (ii->i_fieldmask & QC_INO_TIMER)
2817 mi->dqi_igrace = ii->i_ino_timelimit;
2818 if (ii->i_fieldmask & QC_FLAGS) {
2819 if (ii->i_flags & QCI_ROOT_SQUASH)
2820 mi->dqi_flags |= DQF_ROOT_SQUASH;
2822 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2824 spin_unlock(&dq_data_lock);
2825 mark_info_dirty(sb, type);
2826 /* Force write to disk */
2827 sb->dq_op->write_info(sb, type);
2830 EXPORT_SYMBOL(dquot_set_dqinfo);
2832 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2833 .quota_enable = dquot_quota_enable,
2834 .quota_disable = dquot_quota_disable,
2835 .quota_sync = dquot_quota_sync,
2836 .get_state = dquot_get_state,
2837 .set_info = dquot_set_dqinfo,
2838 .get_dqblk = dquot_get_dqblk,
2839 .get_nextdqblk = dquot_get_next_dqblk,
2840 .set_dqblk = dquot_set_dqblk
2842 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2844 static int do_proc_dqstats(struct ctl_table *table, int write,
2845 void *buffer, size_t *lenp, loff_t *ppos)
2847 unsigned int type = (unsigned long *)table->data - dqstats.stat;
2848 s64 value = percpu_counter_sum(&dqstats.counter[type]);
2850 /* Filter negative values for non-monotonic counters */
2851 if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2852 type == DQST_FREE_DQUOTS))
2855 /* Update global table */
2856 dqstats.stat[type] = value;
2857 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2860 static struct ctl_table fs_dqstats_table[] = {
2862 .procname = "lookups",
2863 .data = &dqstats.stat[DQST_LOOKUPS],
2864 .maxlen = sizeof(unsigned long),
2866 .proc_handler = do_proc_dqstats,
2869 .procname = "drops",
2870 .data = &dqstats.stat[DQST_DROPS],
2871 .maxlen = sizeof(unsigned long),
2873 .proc_handler = do_proc_dqstats,
2876 .procname = "reads",
2877 .data = &dqstats.stat[DQST_READS],
2878 .maxlen = sizeof(unsigned long),
2880 .proc_handler = do_proc_dqstats,
2883 .procname = "writes",
2884 .data = &dqstats.stat[DQST_WRITES],
2885 .maxlen = sizeof(unsigned long),
2887 .proc_handler = do_proc_dqstats,
2890 .procname = "cache_hits",
2891 .data = &dqstats.stat[DQST_CACHE_HITS],
2892 .maxlen = sizeof(unsigned long),
2894 .proc_handler = do_proc_dqstats,
2897 .procname = "allocated_dquots",
2898 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2899 .maxlen = sizeof(unsigned long),
2901 .proc_handler = do_proc_dqstats,
2904 .procname = "free_dquots",
2905 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2906 .maxlen = sizeof(unsigned long),
2908 .proc_handler = do_proc_dqstats,
2911 .procname = "syncs",
2912 .data = &dqstats.stat[DQST_SYNCS],
2913 .maxlen = sizeof(unsigned long),
2915 .proc_handler = do_proc_dqstats,
2917 #ifdef CONFIG_PRINT_QUOTA_WARNING
2919 .procname = "warnings",
2920 .data = &flag_print_warnings,
2921 .maxlen = sizeof(int),
2923 .proc_handler = proc_dointvec,
2929 static struct ctl_table fs_table[] = {
2931 .procname = "quota",
2933 .child = fs_dqstats_table,
2938 static struct ctl_table sys_table[] = {
2947 static int __init dquot_init(void)
2950 unsigned long nr_hash, order;
2952 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2954 register_sysctl_table(sys_table);
2956 dquot_cachep = kmem_cache_create("dquot",
2957 sizeof(struct dquot), sizeof(unsigned long) * 4,
2958 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2959 SLAB_MEM_SPREAD|SLAB_PANIC),
2963 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2965 panic("Cannot create dquot hash table");
2967 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2968 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2970 panic("Cannot create dquot stat counters");
2973 /* Find power-of-two hlist_heads which can fit into allocation */
2974 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2975 dq_hash_bits = ilog2(nr_hash);
2977 nr_hash = 1UL << dq_hash_bits;
2978 dq_hash_mask = nr_hash - 1;
2979 for (i = 0; i < nr_hash; i++)
2980 INIT_HLIST_HEAD(dquot_hash + i);
2982 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2983 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2985 if (register_shrinker(&dqcache_shrinker))
2986 panic("Cannot register dquot shrinker");
2990 fs_initcall(dquot_init);