2 * Implementation of the diskquota system for the LINUX operating system. QUOTA
3 * is implemented using the BSD system call interface as the means of
4 * communication with the user level. This file contains the generic routines
5 * called by the different filesystems on allocation of an inode or block.
6 * These routines take care of the administration needed to have a consistent
7 * diskquota tracking system. The ideas of both user and group quotas are based
8 * on the Melbourne quota system as used on BSD derived systems. The internal
9 * implementation is based on one of the several variants of the LINUX
10 * inode-subsystem with added complexity of the diskquota system.
12 * Author: Marco van Wieringen <mvw@planets.elm.net>
14 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 * Revised list management to avoid races
17 * -- Bill Hawes, <whawes@star.net>, 9/98
19 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20 * As the consequence the locking was moved from dquot_decr_...(),
21 * dquot_incr_...() to calling functions.
22 * invalidate_dquots() now writes modified dquots.
23 * Serialized quota_off() and quota_on() for mount point.
24 * Fixed a few bugs in grow_dquots().
25 * Fixed deadlock in write_dquot() - we no longer account quotas on
27 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
28 * add_dquot_ref() restarts after blocking
29 * Added check for bogus uid and fixed check for group in quotactl.
30 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32 * Used struct list_head instead of own list struct
33 * Invalidation of referenced dquots is no longer possible
34 * Improved free_dquots list management
35 * Quota and i_blocks are now updated in one place to avoid races
36 * Warnings are now delayed so we won't block in critical section
37 * Write updated not to require dquot lock
38 * Jan Kara, <jack@suse.cz>, 9/2000
40 * Added dynamic quota structure allocation
41 * Jan Kara <jack@suse.cz> 12/2000
43 * Rewritten quota interface. Implemented new quota format and
44 * formats registering.
45 * Jan Kara, <jack@suse.cz>, 2001,2002
48 * Jan Kara, <jack@suse.cz>, 10/2002
50 * Added journalled quota support, fix lock inversion problems
51 * Jan Kara, <jack@suse.cz>, 2003,2004
53 * (C) Copyright 1994 - 1997 Marco van Wieringen
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
59 #include <linux/mount.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/cred.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/capability.h>
79 #include <linux/quotaops.h>
80 #include "../internal.h" /* ugh */
82 #include <linux/uaccess.h>
85 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
87 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
88 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
89 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
90 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
91 * modifications of quota state (on quotaon and quotaoff) and readers who care
92 * about latest values take it as well.
94 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
95 * dq_list_lock > dq_state_lock
97 * Note that some things (eg. sb pointer, type, id) doesn't change during
98 * the life of the dquot structure and so needn't to be protected by a lock
100 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
101 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
102 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
103 * inode and before dropping dquot references to avoid use of dquots after
104 * they are freed. dq_data_lock is used to serialize the pointer setting and
105 * clearing operations.
106 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
107 * inode is a quota file). Functions adding pointers from inode to dquots have
108 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
109 * have to do all pointer modifications before dropping dq_data_lock. This makes
110 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
111 * then drops all pointers to dquots from an inode.
113 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
114 * memory (or space for it is being allocated) on the first dqget(), when it is
115 * being written out, and when it is being released on the last dqput(). The
116 * allocation and release operations are serialized by the dq_lock and by
117 * checking the use count in dquot_release().
119 * Lock ordering (including related VFS locks) is the following:
120 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
123 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
124 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
125 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
126 EXPORT_SYMBOL(dq_data_lock);
127 DEFINE_STATIC_SRCU(dquot_srcu);
129 void __quota_error(struct super_block *sb, const char *func,
130 const char *fmt, ...)
132 if (printk_ratelimit()) {
134 struct va_format vaf;
141 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
142 sb->s_id, func, &vaf);
147 EXPORT_SYMBOL(__quota_error);
149 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
150 static char *quotatypes[] = INITQFNAMES;
152 static struct quota_format_type *quota_formats; /* List of registered formats */
153 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
155 /* SLAB cache for dquot structures */
156 static struct kmem_cache *dquot_cachep;
158 int register_quota_format(struct quota_format_type *fmt)
160 spin_lock(&dq_list_lock);
161 fmt->qf_next = quota_formats;
163 spin_unlock(&dq_list_lock);
166 EXPORT_SYMBOL(register_quota_format);
168 void unregister_quota_format(struct quota_format_type *fmt)
170 struct quota_format_type **actqf;
172 spin_lock(&dq_list_lock);
173 for (actqf = "a_formats; *actqf && *actqf != fmt;
174 actqf = &(*actqf)->qf_next)
177 *actqf = (*actqf)->qf_next;
178 spin_unlock(&dq_list_lock);
180 EXPORT_SYMBOL(unregister_quota_format);
182 static struct quota_format_type *find_quota_format(int id)
184 struct quota_format_type *actqf;
186 spin_lock(&dq_list_lock);
187 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
188 actqf = actqf->qf_next)
190 if (!actqf || !try_module_get(actqf->qf_owner)) {
193 spin_unlock(&dq_list_lock);
195 for (qm = 0; module_names[qm].qm_fmt_id &&
196 module_names[qm].qm_fmt_id != id; qm++)
198 if (!module_names[qm].qm_fmt_id ||
199 request_module(module_names[qm].qm_mod_name))
202 spin_lock(&dq_list_lock);
203 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
204 actqf = actqf->qf_next)
206 if (actqf && !try_module_get(actqf->qf_owner))
209 spin_unlock(&dq_list_lock);
213 static void put_quota_format(struct quota_format_type *fmt)
215 module_put(fmt->qf_owner);
219 * Dquot List Management:
220 * The quota code uses three lists for dquot management: the inuse_list,
221 * free_dquots, and dquot_hash[] array. A single dquot structure may be
222 * on all three lists, depending on its current state.
224 * All dquots are placed to the end of inuse_list when first created, and this
225 * list is used for invalidate operation, which must look at every dquot.
227 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
228 * and this list is searched whenever we need an available dquot. Dquots are
229 * removed from the list as soon as they are used again, and
230 * dqstats.free_dquots gives the number of dquots on the list. When
231 * dquot is invalidated it's completely released from memory.
233 * Dquots with a specific identity (device, type and id) are placed on
234 * one of the dquot_hash[] hash chains. The provides an efficient search
235 * mechanism to locate a specific dquot.
238 static LIST_HEAD(inuse_list);
239 static LIST_HEAD(free_dquots);
240 static unsigned int dq_hash_bits, dq_hash_mask;
241 static struct hlist_head *dquot_hash;
243 struct dqstats dqstats;
244 EXPORT_SYMBOL(dqstats);
246 static qsize_t inode_get_rsv_space(struct inode *inode);
247 static int __dquot_initialize(struct inode *inode, int type);
249 static inline unsigned int
250 hashfn(const struct super_block *sb, struct kqid qid)
252 unsigned int id = from_kqid(&init_user_ns, qid);
256 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
257 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
261 * Following list functions expect dq_list_lock to be held
263 static inline void insert_dquot_hash(struct dquot *dquot)
265 struct hlist_head *head;
266 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
267 hlist_add_head(&dquot->dq_hash, head);
270 static inline void remove_dquot_hash(struct dquot *dquot)
272 hlist_del_init(&dquot->dq_hash);
275 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
278 struct hlist_node *node;
281 hlist_for_each (node, dquot_hash+hashent) {
282 dquot = hlist_entry(node, struct dquot, dq_hash);
283 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
289 /* Add a dquot to the tail of the free list */
290 static inline void put_dquot_last(struct dquot *dquot)
292 list_add_tail(&dquot->dq_free, &free_dquots);
293 dqstats_inc(DQST_FREE_DQUOTS);
296 static inline void remove_free_dquot(struct dquot *dquot)
298 if (list_empty(&dquot->dq_free))
300 list_del_init(&dquot->dq_free);
301 dqstats_dec(DQST_FREE_DQUOTS);
304 static inline void put_inuse(struct dquot *dquot)
306 /* We add to the back of inuse list so we don't have to restart
307 * when traversing this list and we block */
308 list_add_tail(&dquot->dq_inuse, &inuse_list);
309 dqstats_inc(DQST_ALLOC_DQUOTS);
312 static inline void remove_inuse(struct dquot *dquot)
314 dqstats_dec(DQST_ALLOC_DQUOTS);
315 list_del(&dquot->dq_inuse);
318 * End of list functions needing dq_list_lock
321 static void wait_on_dquot(struct dquot *dquot)
323 mutex_lock(&dquot->dq_lock);
324 mutex_unlock(&dquot->dq_lock);
327 static inline int dquot_dirty(struct dquot *dquot)
329 return test_bit(DQ_MOD_B, &dquot->dq_flags);
332 static inline int mark_dquot_dirty(struct dquot *dquot)
334 return dquot->dq_sb->dq_op->mark_dirty(dquot);
337 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
338 int dquot_mark_dquot_dirty(struct dquot *dquot)
342 /* If quota is dirty already, we don't have to acquire dq_list_lock */
343 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
346 spin_lock(&dq_list_lock);
347 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
348 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
349 info[dquot->dq_id.type].dqi_dirty_list);
352 spin_unlock(&dq_list_lock);
355 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
357 /* Dirtify all the dquots - this can block when journalling */
358 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
363 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
365 /* Even in case of error we have to continue */
366 ret = mark_dquot_dirty(dquot[cnt]);
373 static inline void dqput_all(struct dquot **dquot)
377 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
381 /* This function needs dq_list_lock */
382 static inline int clear_dquot_dirty(struct dquot *dquot)
384 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
386 list_del_init(&dquot->dq_dirty);
390 void mark_info_dirty(struct super_block *sb, int type)
392 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
394 EXPORT_SYMBOL(mark_info_dirty);
397 * Read dquot from disk and alloc space for it
400 int dquot_acquire(struct dquot *dquot)
402 int ret = 0, ret2 = 0;
403 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
405 mutex_lock(&dquot->dq_lock);
406 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
407 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
410 /* Make sure flags update is visible after dquot has been filled */
411 smp_mb__before_atomic();
412 set_bit(DQ_READ_B, &dquot->dq_flags);
413 /* Instantiate dquot if needed */
414 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
415 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
416 /* Write the info if needed */
417 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
418 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
419 dquot->dq_sb, dquot->dq_id.type);
429 * Make sure flags update is visible after on-disk struct has been
430 * allocated. Paired with smp_rmb() in dqget().
432 smp_mb__before_atomic();
433 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
435 mutex_unlock(&dquot->dq_lock);
438 EXPORT_SYMBOL(dquot_acquire);
441 * Write dquot to disk
443 int dquot_commit(struct dquot *dquot)
446 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
448 mutex_lock(&dquot->dq_lock);
449 spin_lock(&dq_list_lock);
450 if (!clear_dquot_dirty(dquot)) {
451 spin_unlock(&dq_list_lock);
454 spin_unlock(&dq_list_lock);
455 /* Inactive dquot can be only if there was error during read/init
456 * => we have better not writing it */
457 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
458 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
462 mutex_unlock(&dquot->dq_lock);
465 EXPORT_SYMBOL(dquot_commit);
470 int dquot_release(struct dquot *dquot)
472 int ret = 0, ret2 = 0;
473 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
475 mutex_lock(&dquot->dq_lock);
476 /* Check whether we are not racing with some other dqget() */
477 if (atomic_read(&dquot->dq_count) > 1)
479 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
480 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
482 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
483 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
484 dquot->dq_sb, dquot->dq_id.type);
489 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
491 mutex_unlock(&dquot->dq_lock);
494 EXPORT_SYMBOL(dquot_release);
496 void dquot_destroy(struct dquot *dquot)
498 kmem_cache_free(dquot_cachep, dquot);
500 EXPORT_SYMBOL(dquot_destroy);
502 static inline void do_destroy_dquot(struct dquot *dquot)
504 dquot->dq_sb->dq_op->destroy_dquot(dquot);
507 /* Invalidate all dquots on the list. Note that this function is called after
508 * quota is disabled and pointers from inodes removed so there cannot be new
509 * quota users. There can still be some users of quotas due to inodes being
510 * just deleted or pruned by prune_icache() (those are not attached to any
511 * list) or parallel quotactl call. We have to wait for such users.
513 static void invalidate_dquots(struct super_block *sb, int type)
515 struct dquot *dquot, *tmp;
518 spin_lock(&dq_list_lock);
519 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
520 if (dquot->dq_sb != sb)
522 if (dquot->dq_id.type != type)
524 /* Wait for dquot users */
525 if (atomic_read(&dquot->dq_count)) {
529 prepare_to_wait(&dquot->dq_wait_unused, &wait,
530 TASK_UNINTERRUPTIBLE);
531 spin_unlock(&dq_list_lock);
532 /* Once dqput() wakes us up, we know it's time to free
534 * IMPORTANT: we rely on the fact that there is always
535 * at most one process waiting for dquot to free.
536 * Otherwise dq_count would be > 1 and we would never
539 if (atomic_read(&dquot->dq_count) > 1)
541 finish_wait(&dquot->dq_wait_unused, &wait);
543 /* At this moment dquot() need not exist (it could be
544 * reclaimed by prune_dqcache(). Hence we must
549 * Quota now has no users and it has been written on last
552 remove_dquot_hash(dquot);
553 remove_free_dquot(dquot);
555 do_destroy_dquot(dquot);
557 spin_unlock(&dq_list_lock);
560 /* Call callback for every active dquot on given filesystem */
561 int dquot_scan_active(struct super_block *sb,
562 int (*fn)(struct dquot *dquot, unsigned long priv),
565 struct dquot *dquot, *old_dquot = NULL;
568 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
570 spin_lock(&dq_list_lock);
571 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
572 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
574 if (dquot->dq_sb != sb)
576 /* Now we have active dquot so we can just increase use count */
577 atomic_inc(&dquot->dq_count);
578 spin_unlock(&dq_list_lock);
579 dqstats_inc(DQST_LOOKUPS);
583 * ->release_dquot() can be racing with us. Our reference
584 * protects us from new calls to it so just wait for any
585 * outstanding call and recheck the DQ_ACTIVE_B after that.
587 wait_on_dquot(dquot);
588 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
589 ret = fn(dquot, priv);
593 spin_lock(&dq_list_lock);
594 /* We are safe to continue now because our dquot could not
595 * be moved out of the inuse list while we hold the reference */
597 spin_unlock(&dq_list_lock);
602 EXPORT_SYMBOL(dquot_scan_active);
604 /* Write all dquot structures to quota files */
605 int dquot_writeback_dquots(struct super_block *sb, int type)
607 struct list_head *dirty;
609 struct quota_info *dqopt = sb_dqopt(sb);
613 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
615 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
616 if (type != -1 && cnt != type)
618 if (!sb_has_quota_active(sb, cnt))
620 spin_lock(&dq_list_lock);
621 dirty = &dqopt->info[cnt].dqi_dirty_list;
622 while (!list_empty(dirty)) {
623 dquot = list_first_entry(dirty, struct dquot,
625 /* Dirty and inactive can be only bad dquot... */
626 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
627 clear_dquot_dirty(dquot);
630 /* Now we have active dquot from which someone is
631 * holding reference so we can safely just increase
634 spin_unlock(&dq_list_lock);
635 dqstats_inc(DQST_LOOKUPS);
636 err = sb->dq_op->write_dquot(dquot);
640 spin_lock(&dq_list_lock);
642 spin_unlock(&dq_list_lock);
645 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
646 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
647 && info_dirty(&dqopt->info[cnt]))
648 sb->dq_op->write_info(sb, cnt);
649 dqstats_inc(DQST_SYNCS);
653 EXPORT_SYMBOL(dquot_writeback_dquots);
655 /* Write all dquot structures to disk and make them visible from userspace */
656 int dquot_quota_sync(struct super_block *sb, int type)
658 struct quota_info *dqopt = sb_dqopt(sb);
662 ret = dquot_writeback_dquots(sb, type);
665 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
668 /* This is not very clever (and fast) but currently I don't know about
669 * any other simple way of getting quota data to disk and we must get
670 * them there for userspace to be visible... */
671 if (sb->s_op->sync_fs)
672 sb->s_op->sync_fs(sb, 1);
673 sync_blockdev(sb->s_bdev);
676 * Now when everything is written we can discard the pagecache so
677 * that userspace sees the changes.
679 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
680 if (type != -1 && cnt != type)
682 if (!sb_has_quota_active(sb, cnt))
684 inode_lock(dqopt->files[cnt]);
685 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
686 inode_unlock(dqopt->files[cnt]);
691 EXPORT_SYMBOL(dquot_quota_sync);
694 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
696 struct list_head *head;
698 unsigned long freed = 0;
700 spin_lock(&dq_list_lock);
701 head = free_dquots.prev;
702 while (head != &free_dquots && sc->nr_to_scan) {
703 dquot = list_entry(head, struct dquot, dq_free);
704 remove_dquot_hash(dquot);
705 remove_free_dquot(dquot);
707 do_destroy_dquot(dquot);
710 head = free_dquots.prev;
712 spin_unlock(&dq_list_lock);
717 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
719 return vfs_pressure_ratio(
720 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
723 static struct shrinker dqcache_shrinker = {
724 .count_objects = dqcache_shrink_count,
725 .scan_objects = dqcache_shrink_scan,
726 .seeks = DEFAULT_SEEKS,
730 * Put reference to dquot
732 void dqput(struct dquot *dquot)
738 #ifdef CONFIG_QUOTA_DEBUG
739 if (!atomic_read(&dquot->dq_count)) {
740 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
741 quotatypes[dquot->dq_id.type],
742 from_kqid(&init_user_ns, dquot->dq_id));
746 dqstats_inc(DQST_DROPS);
748 spin_lock(&dq_list_lock);
749 if (atomic_read(&dquot->dq_count) > 1) {
750 /* We have more than one user... nothing to do */
751 atomic_dec(&dquot->dq_count);
752 /* Releasing dquot during quotaoff phase? */
753 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
754 atomic_read(&dquot->dq_count) == 1)
755 wake_up(&dquot->dq_wait_unused);
756 spin_unlock(&dq_list_lock);
759 /* Need to release dquot? */
760 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
761 spin_unlock(&dq_list_lock);
762 /* Commit dquot before releasing */
763 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
765 quota_error(dquot->dq_sb, "Can't write quota structure"
766 " (error %d). Quota may get out of sync!",
769 * We clear dirty bit anyway, so that we avoid
772 spin_lock(&dq_list_lock);
773 clear_dquot_dirty(dquot);
774 spin_unlock(&dq_list_lock);
778 /* Clear flag in case dquot was inactive (something bad happened) */
779 clear_dquot_dirty(dquot);
780 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
781 spin_unlock(&dq_list_lock);
782 dquot->dq_sb->dq_op->release_dquot(dquot);
785 atomic_dec(&dquot->dq_count);
786 #ifdef CONFIG_QUOTA_DEBUG
788 BUG_ON(!list_empty(&dquot->dq_free));
790 put_dquot_last(dquot);
791 spin_unlock(&dq_list_lock);
793 EXPORT_SYMBOL(dqput);
795 struct dquot *dquot_alloc(struct super_block *sb, int type)
797 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
799 EXPORT_SYMBOL(dquot_alloc);
801 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
805 dquot = sb->dq_op->alloc_dquot(sb, type);
809 mutex_init(&dquot->dq_lock);
810 INIT_LIST_HEAD(&dquot->dq_free);
811 INIT_LIST_HEAD(&dquot->dq_inuse);
812 INIT_HLIST_NODE(&dquot->dq_hash);
813 INIT_LIST_HEAD(&dquot->dq_dirty);
814 init_waitqueue_head(&dquot->dq_wait_unused);
816 dquot->dq_id = make_kqid_invalid(type);
817 atomic_set(&dquot->dq_count, 1);
823 * Get reference to dquot
825 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
826 * destroying our dquot by:
827 * a) checking for quota flags under dq_list_lock and
828 * b) getting a reference to dquot before we release dq_list_lock
830 struct dquot *dqget(struct super_block *sb, struct kqid qid)
832 unsigned int hashent = hashfn(sb, qid);
833 struct dquot *dquot, *empty = NULL;
835 if (!qid_has_mapping(sb->s_user_ns, qid))
836 return ERR_PTR(-EINVAL);
838 if (!sb_has_quota_active(sb, qid.type))
839 return ERR_PTR(-ESRCH);
841 spin_lock(&dq_list_lock);
842 spin_lock(&dq_state_lock);
843 if (!sb_has_quota_active(sb, qid.type)) {
844 spin_unlock(&dq_state_lock);
845 spin_unlock(&dq_list_lock);
846 dquot = ERR_PTR(-ESRCH);
849 spin_unlock(&dq_state_lock);
851 dquot = find_dquot(hashent, sb, qid);
854 spin_unlock(&dq_list_lock);
855 empty = get_empty_dquot(sb, qid.type);
857 schedule(); /* Try to wait for a moment... */
863 /* all dquots go on the inuse_list */
865 /* hash it first so it can be found */
866 insert_dquot_hash(dquot);
867 spin_unlock(&dq_list_lock);
868 dqstats_inc(DQST_LOOKUPS);
870 if (!atomic_read(&dquot->dq_count))
871 remove_free_dquot(dquot);
872 atomic_inc(&dquot->dq_count);
873 spin_unlock(&dq_list_lock);
874 dqstats_inc(DQST_CACHE_HITS);
875 dqstats_inc(DQST_LOOKUPS);
877 /* Wait for dq_lock - after this we know that either dquot_release() is
878 * already finished or it will be canceled due to dq_count > 1 test */
879 wait_on_dquot(dquot);
880 /* Read the dquot / allocate space in quota file */
881 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
884 err = sb->dq_op->acquire_dquot(dquot);
887 dquot = ERR_PTR(err);
892 * Make sure following reads see filled structure - paired with
893 * smp_mb__before_atomic() in dquot_acquire().
896 #ifdef CONFIG_QUOTA_DEBUG
897 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
901 do_destroy_dquot(empty);
905 EXPORT_SYMBOL(dqget);
907 static inline struct dquot **i_dquot(struct inode *inode)
909 return inode->i_sb->s_op->get_dquots(inode);
912 static int dqinit_needed(struct inode *inode, int type)
914 struct dquot * const *dquots;
917 if (IS_NOQUOTA(inode))
920 dquots = i_dquot(inode);
922 return !dquots[type];
923 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
929 /* This routine is guarded by s_umount semaphore */
930 static void add_dquot_ref(struct super_block *sb, int type)
932 struct inode *inode, *old_inode = NULL;
933 #ifdef CONFIG_QUOTA_DEBUG
937 spin_lock(&sb->s_inode_list_lock);
938 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
939 spin_lock(&inode->i_lock);
940 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
941 !atomic_read(&inode->i_writecount) ||
942 !dqinit_needed(inode, type)) {
943 spin_unlock(&inode->i_lock);
947 spin_unlock(&inode->i_lock);
948 spin_unlock(&sb->s_inode_list_lock);
950 #ifdef CONFIG_QUOTA_DEBUG
951 if (unlikely(inode_get_rsv_space(inode) > 0))
955 __dquot_initialize(inode, type);
958 * We hold a reference to 'inode' so it couldn't have been
959 * removed from s_inodes list while we dropped the
960 * s_inode_list_lock. We cannot iput the inode now as we can be
961 * holding the last reference and we cannot iput it under
962 * s_inode_list_lock. So we keep the reference and iput it
966 spin_lock(&sb->s_inode_list_lock);
968 spin_unlock(&sb->s_inode_list_lock);
971 #ifdef CONFIG_QUOTA_DEBUG
973 quota_error(sb, "Writes happened before quota was turned on "
974 "thus quota information is probably inconsistent. "
975 "Please run quotacheck(8)");
981 * Remove references to dquots from inode and add dquot to list for freeing
982 * if we have the last reference to dquot
984 static void remove_inode_dquot_ref(struct inode *inode, int type,
985 struct list_head *tofree_head)
987 struct dquot **dquots = i_dquot(inode);
988 struct dquot *dquot = dquots[type];
994 if (list_empty(&dquot->dq_free)) {
996 * The inode still has reference to dquot so it can't be in the
999 spin_lock(&dq_list_lock);
1000 list_add(&dquot->dq_free, tofree_head);
1001 spin_unlock(&dq_list_lock);
1004 * Dquot is already in a list to put so we won't drop the last
1012 * Free list of dquots
1013 * Dquots are removed from inodes and no new references can be got so we are
1014 * the only ones holding reference
1016 static void put_dquot_list(struct list_head *tofree_head)
1018 struct list_head *act_head;
1019 struct dquot *dquot;
1021 act_head = tofree_head->next;
1022 while (act_head != tofree_head) {
1023 dquot = list_entry(act_head, struct dquot, dq_free);
1024 act_head = act_head->next;
1025 /* Remove dquot from the list so we won't have problems... */
1026 list_del_init(&dquot->dq_free);
1031 static void remove_dquot_ref(struct super_block *sb, int type,
1032 struct list_head *tofree_head)
1034 struct inode *inode;
1037 spin_lock(&sb->s_inode_list_lock);
1038 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1040 * We have to scan also I_NEW inodes because they can already
1041 * have quota pointer initialized. Luckily, we need to touch
1042 * only quota pointers and these have separate locking
1045 spin_lock(&dq_data_lock);
1046 if (!IS_NOQUOTA(inode)) {
1047 if (unlikely(inode_get_rsv_space(inode) > 0))
1049 remove_inode_dquot_ref(inode, type, tofree_head);
1051 spin_unlock(&dq_data_lock);
1053 spin_unlock(&sb->s_inode_list_lock);
1054 #ifdef CONFIG_QUOTA_DEBUG
1056 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1057 " was disabled thus quota information is probably "
1058 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1063 /* Gather all references from inodes and drop them */
1064 static void drop_dquot_ref(struct super_block *sb, int type)
1066 LIST_HEAD(tofree_head);
1069 remove_dquot_ref(sb, type, &tofree_head);
1070 synchronize_srcu(&dquot_srcu);
1071 put_dquot_list(&tofree_head);
1075 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1077 dquot->dq_dqb.dqb_curinodes += number;
1080 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1082 dquot->dq_dqb.dqb_curspace += number;
1085 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1087 dquot->dq_dqb.dqb_rsvspace += number;
1091 * Claim reserved quota space
1093 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1095 if (dquot->dq_dqb.dqb_rsvspace < number) {
1097 number = dquot->dq_dqb.dqb_rsvspace;
1099 dquot->dq_dqb.dqb_curspace += number;
1100 dquot->dq_dqb.dqb_rsvspace -= number;
1103 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1105 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1106 number = dquot->dq_dqb.dqb_curspace;
1107 dquot->dq_dqb.dqb_rsvspace += number;
1108 dquot->dq_dqb.dqb_curspace -= number;
1112 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1114 if (dquot->dq_dqb.dqb_rsvspace >= number)
1115 dquot->dq_dqb.dqb_rsvspace -= number;
1118 dquot->dq_dqb.dqb_rsvspace = 0;
1122 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1124 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1125 dquot->dq_dqb.dqb_curinodes >= number)
1126 dquot->dq_dqb.dqb_curinodes -= number;
1128 dquot->dq_dqb.dqb_curinodes = 0;
1129 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1130 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1131 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1134 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1136 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1137 dquot->dq_dqb.dqb_curspace >= number)
1138 dquot->dq_dqb.dqb_curspace -= number;
1140 dquot->dq_dqb.dqb_curspace = 0;
1141 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1142 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1143 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1147 struct super_block *w_sb;
1148 struct kqid w_dq_id;
1152 static int warning_issued(struct dquot *dquot, const int warntype)
1154 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1155 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1156 ((warntype == QUOTA_NL_IHARDWARN ||
1157 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1161 return test_and_set_bit(flag, &dquot->dq_flags);
1164 #ifdef CONFIG_PRINT_QUOTA_WARNING
1165 static int flag_print_warnings = 1;
1167 static int need_print_warning(struct dquot_warn *warn)
1169 if (!flag_print_warnings)
1172 switch (warn->w_dq_id.type) {
1174 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1176 return in_group_p(warn->w_dq_id.gid);
1183 /* Print warning to user which exceeded quota */
1184 static void print_warning(struct dquot_warn *warn)
1187 struct tty_struct *tty;
1188 int warntype = warn->w_type;
1190 if (warntype == QUOTA_NL_IHARDBELOW ||
1191 warntype == QUOTA_NL_ISOFTBELOW ||
1192 warntype == QUOTA_NL_BHARDBELOW ||
1193 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1196 tty = get_current_tty();
1199 tty_write_message(tty, warn->w_sb->s_id);
1200 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1201 tty_write_message(tty, ": warning, ");
1203 tty_write_message(tty, ": write failed, ");
1204 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1206 case QUOTA_NL_IHARDWARN:
1207 msg = " file limit reached.\r\n";
1209 case QUOTA_NL_ISOFTLONGWARN:
1210 msg = " file quota exceeded too long.\r\n";
1212 case QUOTA_NL_ISOFTWARN:
1213 msg = " file quota exceeded.\r\n";
1215 case QUOTA_NL_BHARDWARN:
1216 msg = " block limit reached.\r\n";
1218 case QUOTA_NL_BSOFTLONGWARN:
1219 msg = " block quota exceeded too long.\r\n";
1221 case QUOTA_NL_BSOFTWARN:
1222 msg = " block quota exceeded.\r\n";
1225 tty_write_message(tty, msg);
1230 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1233 if (warning_issued(dquot, warntype))
1235 warn->w_type = warntype;
1236 warn->w_sb = dquot->dq_sb;
1237 warn->w_dq_id = dquot->dq_id;
1241 * Write warnings to the console and send warning messages over netlink.
1243 * Note that this function can call into tty and networking code.
1245 static void flush_warnings(struct dquot_warn *warn)
1249 for (i = 0; i < MAXQUOTAS; i++) {
1250 if (warn[i].w_type == QUOTA_NL_NOWARN)
1252 #ifdef CONFIG_PRINT_QUOTA_WARNING
1253 print_warning(&warn[i]);
1255 quota_send_warning(warn[i].w_dq_id,
1256 warn[i].w_sb->s_dev, warn[i].w_type);
1260 static int ignore_hardlimit(struct dquot *dquot)
1262 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1264 return capable(CAP_SYS_RESOURCE) &&
1265 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1266 !(info->dqi_flags & DQF_ROOT_SQUASH));
1269 /* needs dq_data_lock */
1270 static int check_idq(struct dquot *dquot, qsize_t inodes,
1271 struct dquot_warn *warn)
1273 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1275 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1276 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1279 if (dquot->dq_dqb.dqb_ihardlimit &&
1280 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1281 !ignore_hardlimit(dquot)) {
1282 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);
1295 if (dquot->dq_dqb.dqb_isoftlimit &&
1296 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1297 dquot->dq_dqb.dqb_itime == 0) {
1298 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1299 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1300 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1306 /* needs dq_data_lock */
1307 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1308 struct dquot_warn *warn)
1311 struct super_block *sb = dquot->dq_sb;
1313 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1314 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1317 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1320 if (dquot->dq_dqb.dqb_bhardlimit &&
1321 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1322 !ignore_hardlimit(dquot)) {
1324 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1328 if (dquot->dq_dqb.dqb_bsoftlimit &&
1329 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1330 dquot->dq_dqb.dqb_btime &&
1331 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1332 !ignore_hardlimit(dquot)) {
1334 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1338 if (dquot->dq_dqb.dqb_bsoftlimit &&
1339 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1340 dquot->dq_dqb.dqb_btime == 0) {
1342 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1343 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1344 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1348 * We don't allow preallocation to exceed softlimit so exceeding will
1357 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1361 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1362 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1363 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1364 return QUOTA_NL_NOWARN;
1366 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1367 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1368 return QUOTA_NL_ISOFTBELOW;
1369 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1370 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1371 return QUOTA_NL_IHARDBELOW;
1372 return QUOTA_NL_NOWARN;
1375 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1377 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1378 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1379 return QUOTA_NL_NOWARN;
1381 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1382 return QUOTA_NL_BSOFTBELOW;
1383 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1384 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1385 return QUOTA_NL_BHARDBELOW;
1386 return QUOTA_NL_NOWARN;
1389 static int dquot_active(const struct inode *inode)
1391 struct super_block *sb = inode->i_sb;
1393 if (IS_NOQUOTA(inode))
1395 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1399 * Initialize quota pointers in inode
1401 * It is better to call this function outside of any transaction as it
1402 * might need a lot of space in journal for dquot structure allocation.
1404 static int __dquot_initialize(struct inode *inode, int type)
1406 int cnt, init_needed = 0;
1407 struct dquot **dquots, *got[MAXQUOTAS] = {};
1408 struct super_block *sb = inode->i_sb;
1412 if (!dquot_active(inode))
1415 dquots = i_dquot(inode);
1417 /* First get references to structures we might need. */
1418 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1422 struct dquot *dquot;
1424 if (type != -1 && cnt != type)
1427 * The i_dquot should have been initialized in most cases,
1428 * we check it without locking here to avoid unnecessary
1429 * dqget()/dqput() calls.
1434 if (!sb_has_quota_active(sb, cnt))
1441 qid = make_kqid_uid(inode->i_uid);
1444 qid = make_kqid_gid(inode->i_gid);
1447 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1450 qid = make_kqid_projid(projid);
1453 dquot = dqget(sb, qid);
1454 if (IS_ERR(dquot)) {
1455 /* We raced with somebody turning quotas off... */
1456 if (PTR_ERR(dquot) != -ESRCH) {
1457 ret = PTR_ERR(dquot);
1465 /* All required i_dquot has been initialized */
1469 spin_lock(&dq_data_lock);
1470 if (IS_NOQUOTA(inode))
1472 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1473 if (type != -1 && cnt != type)
1475 /* Avoid races with quotaoff() */
1476 if (!sb_has_quota_active(sb, cnt))
1478 /* We could race with quotaon or dqget() could have failed */
1482 dquots[cnt] = got[cnt];
1485 * Make quota reservation system happy if someone
1486 * did a write before quota was turned on
1488 rsv = inode_get_rsv_space(inode);
1490 dquot_resv_space(dquots[cnt], rsv);
1494 spin_unlock(&dq_data_lock);
1496 /* Drop unused references */
1502 int dquot_initialize(struct inode *inode)
1504 return __dquot_initialize(inode, -1);
1506 EXPORT_SYMBOL(dquot_initialize);
1508 bool dquot_initialize_needed(struct inode *inode)
1510 struct dquot **dquots;
1513 if (!dquot_active(inode))
1516 dquots = i_dquot(inode);
1517 for (i = 0; i < MAXQUOTAS; i++)
1518 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1522 EXPORT_SYMBOL(dquot_initialize_needed);
1525 * Release all quotas referenced by inode.
1527 * This function only be called on inode free or converting
1528 * a file to quota file, no other users for the i_dquot in
1529 * both cases, so we needn't call synchronize_srcu() after
1532 static void __dquot_drop(struct inode *inode)
1535 struct dquot **dquots = i_dquot(inode);
1536 struct dquot *put[MAXQUOTAS];
1538 spin_lock(&dq_data_lock);
1539 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1540 put[cnt] = dquots[cnt];
1543 spin_unlock(&dq_data_lock);
1547 void dquot_drop(struct inode *inode)
1549 struct dquot * const *dquots;
1552 if (IS_NOQUOTA(inode))
1556 * Test before calling to rule out calls from proc and such
1557 * where we are not allowed to block. Note that this is
1558 * actually reliable test even without the lock - the caller
1559 * must assure that nobody can come after the DQUOT_DROP and
1560 * add quota pointers back anyway.
1562 dquots = i_dquot(inode);
1563 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1568 if (cnt < MAXQUOTAS)
1569 __dquot_drop(inode);
1571 EXPORT_SYMBOL(dquot_drop);
1574 * inode_reserved_space is managed internally by quota, and protected by
1575 * i_lock similar to i_blocks+i_bytes.
1577 static qsize_t *inode_reserved_space(struct inode * inode)
1579 /* Filesystem must explicitly define it's own method in order to use
1580 * quota reservation interface */
1581 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1582 return inode->i_sb->dq_op->get_reserved_space(inode);
1585 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1587 spin_lock(&inode->i_lock);
1588 *inode_reserved_space(inode) += number;
1589 spin_unlock(&inode->i_lock);
1591 EXPORT_SYMBOL(inode_add_rsv_space);
1593 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1595 spin_lock(&inode->i_lock);
1596 *inode_reserved_space(inode) -= number;
1597 __inode_add_bytes(inode, number);
1598 spin_unlock(&inode->i_lock);
1600 EXPORT_SYMBOL(inode_claim_rsv_space);
1602 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1604 spin_lock(&inode->i_lock);
1605 *inode_reserved_space(inode) += number;
1606 __inode_sub_bytes(inode, number);
1607 spin_unlock(&inode->i_lock);
1609 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1611 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1613 spin_lock(&inode->i_lock);
1614 *inode_reserved_space(inode) -= number;
1615 spin_unlock(&inode->i_lock);
1617 EXPORT_SYMBOL(inode_sub_rsv_space);
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_reserved_space(inode);
1627 spin_unlock(&inode->i_lock);
1631 static void inode_incr_space(struct inode *inode, qsize_t number,
1635 inode_add_rsv_space(inode, number);
1637 inode_add_bytes(inode, number);
1640 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1643 inode_sub_rsv_space(inode, number);
1645 inode_sub_bytes(inode, number);
1649 * This functions updates i_blocks+i_bytes fields and quota information
1650 * (together with appropriate checks).
1652 * NOTE: We absolutely rely on the fact that caller dirties the inode
1653 * (usually helpers in quotaops.h care about this) and holds a handle for
1654 * the current transaction so that dquot write and inode write go into the
1659 * This operation can block, but only after everything is updated
1661 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1663 int cnt, ret = 0, index;
1664 struct dquot_warn warn[MAXQUOTAS];
1665 int reserve = flags & DQUOT_SPACE_RESERVE;
1666 struct dquot **dquots;
1668 if (!dquot_active(inode)) {
1669 inode_incr_space(inode, number, reserve);
1673 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1674 warn[cnt].w_type = QUOTA_NL_NOWARN;
1676 dquots = i_dquot(inode);
1677 index = srcu_read_lock(&dquot_srcu);
1678 spin_lock(&dq_data_lock);
1679 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1682 ret = check_bdq(dquots[cnt], number,
1683 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1684 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1685 spin_unlock(&dq_data_lock);
1686 goto out_flush_warn;
1689 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1693 dquot_resv_space(dquots[cnt], number);
1695 dquot_incr_space(dquots[cnt], number);
1697 inode_incr_space(inode, number, reserve);
1698 spin_unlock(&dq_data_lock);
1701 goto out_flush_warn;
1702 mark_all_dquot_dirty(dquots);
1704 srcu_read_unlock(&dquot_srcu, index);
1705 flush_warnings(warn);
1709 EXPORT_SYMBOL(__dquot_alloc_space);
1712 * This operation can block, but only after everything is updated
1714 int dquot_alloc_inode(struct inode *inode)
1716 int cnt, ret = 0, index;
1717 struct dquot_warn warn[MAXQUOTAS];
1718 struct dquot * const *dquots;
1720 if (!dquot_active(inode))
1722 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1723 warn[cnt].w_type = QUOTA_NL_NOWARN;
1725 dquots = i_dquot(inode);
1726 index = srcu_read_lock(&dquot_srcu);
1727 spin_lock(&dq_data_lock);
1728 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1731 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1736 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1739 dquot_incr_inodes(dquots[cnt], 1);
1743 spin_unlock(&dq_data_lock);
1745 mark_all_dquot_dirty(dquots);
1746 srcu_read_unlock(&dquot_srcu, index);
1747 flush_warnings(warn);
1750 EXPORT_SYMBOL(dquot_alloc_inode);
1753 * Convert in-memory reserved quotas to real consumed quotas
1755 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1757 struct dquot **dquots;
1760 if (!dquot_active(inode)) {
1761 inode_claim_rsv_space(inode, number);
1765 dquots = i_dquot(inode);
1766 index = srcu_read_lock(&dquot_srcu);
1767 spin_lock(&dq_data_lock);
1768 /* Claim reserved quotas to allocated quotas */
1769 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1771 dquot_claim_reserved_space(dquots[cnt], number);
1773 /* Update inode bytes */
1774 inode_claim_rsv_space(inode, number);
1775 spin_unlock(&dq_data_lock);
1776 mark_all_dquot_dirty(dquots);
1777 srcu_read_unlock(&dquot_srcu, index);
1780 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1783 * Convert allocated space back to in-memory reserved quotas
1785 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1787 struct dquot **dquots;
1790 if (!dquot_active(inode)) {
1791 inode_reclaim_rsv_space(inode, number);
1795 dquots = i_dquot(inode);
1796 index = srcu_read_lock(&dquot_srcu);
1797 spin_lock(&dq_data_lock);
1798 /* Claim reserved quotas to allocated quotas */
1799 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1801 dquot_reclaim_reserved_space(dquots[cnt], number);
1803 /* Update inode bytes */
1804 inode_reclaim_rsv_space(inode, number);
1805 spin_unlock(&dq_data_lock);
1806 mark_all_dquot_dirty(dquots);
1807 srcu_read_unlock(&dquot_srcu, index);
1810 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1813 * This operation can block, but only after everything is updated
1815 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1818 struct dquot_warn warn[MAXQUOTAS];
1819 struct dquot **dquots;
1820 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1822 if (!dquot_active(inode)) {
1823 inode_decr_space(inode, number, reserve);
1827 dquots = i_dquot(inode);
1828 index = srcu_read_lock(&dquot_srcu);
1829 spin_lock(&dq_data_lock);
1830 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1833 warn[cnt].w_type = QUOTA_NL_NOWARN;
1836 wtype = info_bdq_free(dquots[cnt], number);
1837 if (wtype != QUOTA_NL_NOWARN)
1838 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1840 dquot_free_reserved_space(dquots[cnt], number);
1842 dquot_decr_space(dquots[cnt], number);
1844 inode_decr_space(inode, number, reserve);
1845 spin_unlock(&dq_data_lock);
1849 mark_all_dquot_dirty(dquots);
1851 srcu_read_unlock(&dquot_srcu, index);
1852 flush_warnings(warn);
1854 EXPORT_SYMBOL(__dquot_free_space);
1857 * This operation can block, but only after everything is updated
1859 void dquot_free_inode(struct inode *inode)
1862 struct dquot_warn warn[MAXQUOTAS];
1863 struct dquot * const *dquots;
1866 if (!dquot_active(inode))
1869 dquots = i_dquot(inode);
1870 index = srcu_read_lock(&dquot_srcu);
1871 spin_lock(&dq_data_lock);
1872 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1875 warn[cnt].w_type = QUOTA_NL_NOWARN;
1878 wtype = info_idq_free(dquots[cnt], 1);
1879 if (wtype != QUOTA_NL_NOWARN)
1880 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1881 dquot_decr_inodes(dquots[cnt], 1);
1883 spin_unlock(&dq_data_lock);
1884 mark_all_dquot_dirty(dquots);
1885 srcu_read_unlock(&dquot_srcu, index);
1886 flush_warnings(warn);
1888 EXPORT_SYMBOL(dquot_free_inode);
1891 * Transfer the number of inode and blocks from one diskquota to an other.
1892 * On success, dquot references in transfer_to are consumed and references
1893 * to original dquots that need to be released are placed there. On failure,
1894 * references are kept untouched.
1896 * This operation can block, but only after everything is updated
1897 * A transaction must be started when entering this function.
1899 * We are holding reference on transfer_from & transfer_to, no need to
1900 * protect them by srcu_read_lock().
1902 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1904 qsize_t space, cur_space;
1905 qsize_t rsv_space = 0;
1906 qsize_t inode_usage = 1;
1907 struct dquot *transfer_from[MAXQUOTAS] = {};
1909 char is_valid[MAXQUOTAS] = {};
1910 struct dquot_warn warn_to[MAXQUOTAS];
1911 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1912 struct dquot_warn warn_from_space[MAXQUOTAS];
1914 if (IS_NOQUOTA(inode))
1917 if (inode->i_sb->dq_op->get_inode_usage) {
1918 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1923 /* Initialize the arrays */
1924 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1925 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1926 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1927 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1930 spin_lock(&dq_data_lock);
1931 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1932 spin_unlock(&dq_data_lock);
1935 cur_space = inode_get_bytes(inode);
1936 rsv_space = inode_get_rsv_space(inode);
1937 space = cur_space + rsv_space;
1938 /* Build the transfer_from list and check the limits */
1939 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1941 * Skip changes for same uid or gid or for turned off quota-type.
1943 if (!transfer_to[cnt])
1945 /* Avoid races with quotaoff() */
1946 if (!sb_has_quota_active(inode->i_sb, cnt))
1949 transfer_from[cnt] = i_dquot(inode)[cnt];
1950 ret = check_idq(transfer_to[cnt], inode_usage, &warn_to[cnt]);
1953 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1959 * Finally perform the needed transfer from transfer_from to transfer_to
1961 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1964 /* Due to IO error we might not have transfer_from[] structure */
1965 if (transfer_from[cnt]) {
1967 wtype = info_idq_free(transfer_from[cnt], inode_usage);
1968 if (wtype != QUOTA_NL_NOWARN)
1969 prepare_warning(&warn_from_inodes[cnt],
1970 transfer_from[cnt], wtype);
1971 wtype = info_bdq_free(transfer_from[cnt], space);
1972 if (wtype != QUOTA_NL_NOWARN)
1973 prepare_warning(&warn_from_space[cnt],
1974 transfer_from[cnt], wtype);
1975 dquot_decr_inodes(transfer_from[cnt], inode_usage);
1976 dquot_decr_space(transfer_from[cnt], cur_space);
1977 dquot_free_reserved_space(transfer_from[cnt],
1981 dquot_incr_inodes(transfer_to[cnt], inode_usage);
1982 dquot_incr_space(transfer_to[cnt], cur_space);
1983 dquot_resv_space(transfer_to[cnt], rsv_space);
1985 i_dquot(inode)[cnt] = transfer_to[cnt];
1987 spin_unlock(&dq_data_lock);
1989 mark_all_dquot_dirty(transfer_from);
1990 mark_all_dquot_dirty(transfer_to);
1991 flush_warnings(warn_to);
1992 flush_warnings(warn_from_inodes);
1993 flush_warnings(warn_from_space);
1994 /* Pass back references to put */
1995 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1997 transfer_to[cnt] = transfer_from[cnt];
2000 spin_unlock(&dq_data_lock);
2001 flush_warnings(warn_to);
2004 EXPORT_SYMBOL(__dquot_transfer);
2006 /* Wrapper for transferring ownership of an inode for uid/gid only
2007 * Called from FSXXX_setattr()
2009 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2011 struct dquot *transfer_to[MAXQUOTAS] = {};
2012 struct dquot *dquot;
2013 struct super_block *sb = inode->i_sb;
2016 if (!dquot_active(inode))
2019 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2020 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2021 if (IS_ERR(dquot)) {
2022 if (PTR_ERR(dquot) != -ESRCH) {
2023 ret = PTR_ERR(dquot);
2028 transfer_to[USRQUOTA] = dquot;
2030 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2031 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2032 if (IS_ERR(dquot)) {
2033 if (PTR_ERR(dquot) != -ESRCH) {
2034 ret = PTR_ERR(dquot);
2039 transfer_to[GRPQUOTA] = dquot;
2041 ret = __dquot_transfer(inode, transfer_to);
2043 dqput_all(transfer_to);
2046 EXPORT_SYMBOL(dquot_transfer);
2049 * Write info of quota file to disk
2051 int dquot_commit_info(struct super_block *sb, int type)
2053 struct quota_info *dqopt = sb_dqopt(sb);
2055 return dqopt->ops[type]->write_file_info(sb, type);
2057 EXPORT_SYMBOL(dquot_commit_info);
2059 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2061 struct quota_info *dqopt = sb_dqopt(sb);
2063 if (!sb_has_quota_active(sb, qid->type))
2065 if (!dqopt->ops[qid->type]->get_next_id)
2067 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2069 EXPORT_SYMBOL(dquot_get_next_id);
2072 * Definitions of diskquota operations.
2074 const struct dquot_operations dquot_operations = {
2075 .write_dquot = dquot_commit,
2076 .acquire_dquot = dquot_acquire,
2077 .release_dquot = dquot_release,
2078 .mark_dirty = dquot_mark_dquot_dirty,
2079 .write_info = dquot_commit_info,
2080 .alloc_dquot = dquot_alloc,
2081 .destroy_dquot = dquot_destroy,
2082 .get_next_id = dquot_get_next_id,
2084 EXPORT_SYMBOL(dquot_operations);
2087 * Generic helper for ->open on filesystems supporting disk quotas.
2089 int dquot_file_open(struct inode *inode, struct file *file)
2093 error = generic_file_open(inode, file);
2094 if (!error && (file->f_mode & FMODE_WRITE))
2095 dquot_initialize(inode);
2098 EXPORT_SYMBOL(dquot_file_open);
2101 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2103 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2106 struct quota_info *dqopt = sb_dqopt(sb);
2107 struct inode *toputinode[MAXQUOTAS];
2109 /* s_umount should be held in exclusive mode */
2110 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2111 up_read(&sb->s_umount);
2113 /* Cannot turn off usage accounting without turning off limits, or
2114 * suspend quotas and simultaneously turn quotas off. */
2115 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2116 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2117 DQUOT_USAGE_ENABLED)))
2121 * Skip everything if there's nothing to do. We have to do this because
2122 * sometimes we are called when fill_super() failed and calling
2123 * sync_fs() in such cases does no good.
2125 if (!sb_any_quota_loaded(sb))
2128 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2129 toputinode[cnt] = NULL;
2130 if (type != -1 && cnt != type)
2132 if (!sb_has_quota_loaded(sb, cnt))
2135 if (flags & DQUOT_SUSPENDED) {
2136 spin_lock(&dq_state_lock);
2138 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2139 spin_unlock(&dq_state_lock);
2141 spin_lock(&dq_state_lock);
2142 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2143 /* Turning off suspended quotas? */
2144 if (!sb_has_quota_loaded(sb, cnt) &&
2145 sb_has_quota_suspended(sb, cnt)) {
2146 dqopt->flags &= ~dquot_state_flag(
2147 DQUOT_SUSPENDED, cnt);
2148 spin_unlock(&dq_state_lock);
2149 iput(dqopt->files[cnt]);
2150 dqopt->files[cnt] = NULL;
2153 spin_unlock(&dq_state_lock);
2156 /* We still have to keep quota loaded? */
2157 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2160 /* Note: these are blocking operations */
2161 drop_dquot_ref(sb, cnt);
2162 invalidate_dquots(sb, cnt);
2164 * Now all dquots should be invalidated, all writes done so we
2165 * should be only users of the info. No locks needed.
2167 if (info_dirty(&dqopt->info[cnt]))
2168 sb->dq_op->write_info(sb, cnt);
2169 if (dqopt->ops[cnt]->free_file_info)
2170 dqopt->ops[cnt]->free_file_info(sb, cnt);
2171 put_quota_format(dqopt->info[cnt].dqi_format);
2173 toputinode[cnt] = dqopt->files[cnt];
2174 if (!sb_has_quota_loaded(sb, cnt))
2175 dqopt->files[cnt] = NULL;
2176 dqopt->info[cnt].dqi_flags = 0;
2177 dqopt->info[cnt].dqi_igrace = 0;
2178 dqopt->info[cnt].dqi_bgrace = 0;
2179 dqopt->ops[cnt] = NULL;
2182 /* Skip syncing and setting flags if quota files are hidden */
2183 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2186 /* Sync the superblock so that buffers with quota data are written to
2187 * disk (and so userspace sees correct data afterwards). */
2188 if (sb->s_op->sync_fs)
2189 sb->s_op->sync_fs(sb, 1);
2190 sync_blockdev(sb->s_bdev);
2191 /* Now the quota files are just ordinary files and we can set the
2192 * inode flags back. Moreover we discard the pagecache so that
2193 * userspace sees the writes we did bypassing the pagecache. We
2194 * must also discard the blockdev buffers so that we see the
2195 * changes done by userspace on the next quotaon() */
2196 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2197 /* This can happen when suspending quotas on remount-ro... */
2198 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2199 inode_lock(toputinode[cnt]);
2200 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2201 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2202 inode_unlock(toputinode[cnt]);
2203 mark_inode_dirty_sync(toputinode[cnt]);
2206 invalidate_bdev(sb->s_bdev);
2208 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2209 if (toputinode[cnt]) {
2210 /* On remount RO, we keep the inode pointer so that we
2211 * can reenable quota on the subsequent remount RW. We
2212 * have to check 'flags' variable and not use sb_has_
2213 * function because another quotaon / quotaoff could
2214 * change global state before we got here. We refuse
2215 * to suspend quotas when there is pending delete on
2216 * the quota file... */
2217 if (!(flags & DQUOT_SUSPENDED))
2218 iput(toputinode[cnt]);
2219 else if (!toputinode[cnt]->i_nlink)
2224 EXPORT_SYMBOL(dquot_disable);
2226 int dquot_quota_off(struct super_block *sb, int type)
2228 return dquot_disable(sb, type,
2229 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2231 EXPORT_SYMBOL(dquot_quota_off);
2234 * Turn quotas on on a device
2238 * Helper function to turn quotas on when we already have the inode of
2239 * quota file and no quota information is loaded.
2241 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2244 struct quota_format_type *fmt = find_quota_format(format_id);
2245 struct super_block *sb = inode->i_sb;
2246 struct quota_info *dqopt = sb_dqopt(sb);
2251 if (!S_ISREG(inode->i_mode)) {
2255 if (IS_RDONLY(inode)) {
2259 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2260 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2264 /* Filesystems outside of init_user_ns not yet supported */
2265 if (sb->s_user_ns != &init_user_ns) {
2269 /* Usage always has to be set... */
2270 if (!(flags & DQUOT_USAGE_ENABLED)) {
2274 if (sb_has_quota_loaded(sb, type)) {
2279 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2280 /* As we bypass the pagecache we must now flush all the
2281 * dirty data and invalidate caches so that kernel sees
2282 * changes from userspace. It is not enough to just flush
2283 * the quota file since if blocksize < pagesize, invalidation
2284 * of the cache could fail because of other unrelated dirty
2286 sync_filesystem(sb);
2287 invalidate_bdev(sb->s_bdev);
2290 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2291 /* We don't want quota and atime on quota files (deadlocks
2292 * possible) Also nobody should write to the file - we use
2293 * special IO operations which ignore the immutable bit. */
2295 inode->i_flags |= S_NOQUOTA;
2296 inode_unlock(inode);
2298 * When S_NOQUOTA is set, remove dquot references as no more
2299 * references can be added
2301 __dquot_drop(inode);
2305 dqopt->files[type] = igrab(inode);
2306 if (!dqopt->files[type])
2307 goto out_file_flags;
2309 if (!fmt->qf_ops->check_quota_file(sb, type))
2312 dqopt->ops[type] = fmt->qf_ops;
2313 dqopt->info[type].dqi_format = fmt;
2314 dqopt->info[type].dqi_fmt_id = format_id;
2315 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2316 error = dqopt->ops[type]->read_file_info(sb, type);
2319 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2320 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2321 spin_lock(&dq_state_lock);
2322 dqopt->flags |= dquot_state_flag(flags, type);
2323 spin_unlock(&dq_state_lock);
2325 add_dquot_ref(sb, type);
2330 dqopt->files[type] = NULL;
2334 inode->i_flags &= ~S_NOQUOTA;
2335 inode_unlock(inode);
2337 put_quota_format(fmt);
2342 /* Reenable quotas on remount RW */
2343 int dquot_resume(struct super_block *sb, int type)
2345 struct quota_info *dqopt = sb_dqopt(sb);
2346 struct inode *inode;
2350 /* s_umount should be held in exclusive mode */
2351 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2352 up_read(&sb->s_umount);
2354 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2355 if (type != -1 && cnt != type)
2357 if (!sb_has_quota_suspended(sb, cnt))
2360 inode = dqopt->files[cnt];
2361 dqopt->files[cnt] = NULL;
2362 spin_lock(&dq_state_lock);
2363 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2364 DQUOT_LIMITS_ENABLED,
2366 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2367 spin_unlock(&dq_state_lock);
2369 flags = dquot_generic_flag(flags, cnt);
2370 ret = vfs_load_quota_inode(inode, cnt,
2371 dqopt->info[cnt].dqi_fmt_id, flags);
2377 EXPORT_SYMBOL(dquot_resume);
2379 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2380 const struct path *path)
2382 int error = security_quota_on(path->dentry);
2385 /* Quota file not on the same filesystem? */
2386 if (path->dentry->d_sb != sb)
2389 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2390 format_id, DQUOT_USAGE_ENABLED |
2391 DQUOT_LIMITS_ENABLED);
2394 EXPORT_SYMBOL(dquot_quota_on);
2397 * More powerful function for turning on quotas allowing setting
2398 * of individual quota flags
2400 int dquot_enable(struct inode *inode, int type, int format_id,
2403 struct super_block *sb = inode->i_sb;
2405 /* Just unsuspend quotas? */
2406 BUG_ON(flags & DQUOT_SUSPENDED);
2407 /* s_umount should be held in exclusive mode */
2408 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2409 up_read(&sb->s_umount);
2413 /* Just updating flags needed? */
2414 if (sb_has_quota_loaded(sb, type)) {
2415 if (flags & DQUOT_USAGE_ENABLED &&
2416 sb_has_quota_usage_enabled(sb, type))
2418 if (flags & DQUOT_LIMITS_ENABLED &&
2419 sb_has_quota_limits_enabled(sb, type))
2421 spin_lock(&dq_state_lock);
2422 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2423 spin_unlock(&dq_state_lock);
2427 return vfs_load_quota_inode(inode, type, format_id, flags);
2429 EXPORT_SYMBOL(dquot_enable);
2432 * This function is used when filesystem needs to initialize quotas
2433 * during mount time.
2435 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2436 int format_id, int type)
2438 struct dentry *dentry;
2441 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2443 return PTR_ERR(dentry);
2445 if (d_really_is_negative(dentry)) {
2450 error = security_quota_on(dentry);
2452 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2453 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2459 EXPORT_SYMBOL(dquot_quota_on_mount);
2461 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2465 struct quota_info *dqopt = sb_dqopt(sb);
2467 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2469 /* Accounting cannot be turned on while fs is mounted */
2470 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2473 for (type = 0; type < MAXQUOTAS; type++) {
2474 if (!(flags & qtype_enforce_flag(type)))
2476 /* Can't enforce without accounting */
2477 if (!sb_has_quota_usage_enabled(sb, type))
2479 ret = dquot_enable(dqopt->files[type], type,
2480 dqopt->info[type].dqi_fmt_id,
2481 DQUOT_LIMITS_ENABLED);
2487 /* Backout enforcement enablement we already did */
2488 for (type--; type >= 0; type--) {
2489 if (flags & qtype_enforce_flag(type))
2490 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2492 /* Error code translation for better compatibility with XFS */
2498 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2502 struct quota_info *dqopt = sb_dqopt(sb);
2504 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2507 * We don't support turning off accounting via quotactl. In principle
2508 * quota infrastructure can do this but filesystems don't expect
2509 * userspace to be able to do it.
2512 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2515 /* Filter out limits not enabled */
2516 for (type = 0; type < MAXQUOTAS; type++)
2517 if (!sb_has_quota_limits_enabled(sb, type))
2518 flags &= ~qtype_enforce_flag(type);
2522 for (type = 0; type < MAXQUOTAS; type++) {
2523 if (flags & qtype_enforce_flag(type)) {
2524 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2531 /* Backout enforcement disabling we already did */
2532 for (type--; type >= 0; type--) {
2533 if (flags & qtype_enforce_flag(type))
2534 dquot_enable(dqopt->files[type], type,
2535 dqopt->info[type].dqi_fmt_id,
2536 DQUOT_LIMITS_ENABLED);
2541 /* Generic routine for getting common part of quota structure */
2542 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2544 struct mem_dqblk *dm = &dquot->dq_dqb;
2546 memset(di, 0, sizeof(*di));
2547 spin_lock(&dq_data_lock);
2548 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2549 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2550 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2551 di->d_ino_softlimit = dm->dqb_isoftlimit;
2552 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2553 di->d_ino_count = dm->dqb_curinodes;
2554 di->d_spc_timer = dm->dqb_btime;
2555 di->d_ino_timer = dm->dqb_itime;
2556 spin_unlock(&dq_data_lock);
2559 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2560 struct qc_dqblk *di)
2562 struct dquot *dquot;
2564 dquot = dqget(sb, qid);
2566 return PTR_ERR(dquot);
2567 do_get_dqblk(dquot, di);
2572 EXPORT_SYMBOL(dquot_get_dqblk);
2574 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2575 struct qc_dqblk *di)
2577 struct dquot *dquot;
2580 if (!sb->dq_op->get_next_id)
2582 err = sb->dq_op->get_next_id(sb, qid);
2585 dquot = dqget(sb, *qid);
2587 return PTR_ERR(dquot);
2588 do_get_dqblk(dquot, di);
2593 EXPORT_SYMBOL(dquot_get_next_dqblk);
2595 #define VFS_QC_MASK \
2596 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2597 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2598 QC_SPC_TIMER | QC_INO_TIMER)
2600 /* Generic routine for setting common part of quota structure */
2601 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2603 struct mem_dqblk *dm = &dquot->dq_dqb;
2604 int check_blim = 0, check_ilim = 0;
2605 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2607 if (di->d_fieldmask & ~VFS_QC_MASK)
2610 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2611 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2612 ((di->d_fieldmask & QC_SPC_HARD) &&
2613 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2614 ((di->d_fieldmask & QC_INO_SOFT) &&
2615 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2616 ((di->d_fieldmask & QC_INO_HARD) &&
2617 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2620 spin_lock(&dq_data_lock);
2621 if (di->d_fieldmask & QC_SPACE) {
2622 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2624 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2627 if (di->d_fieldmask & QC_SPC_SOFT)
2628 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2629 if (di->d_fieldmask & QC_SPC_HARD)
2630 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2631 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2633 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2636 if (di->d_fieldmask & QC_INO_COUNT) {
2637 dm->dqb_curinodes = di->d_ino_count;
2639 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2642 if (di->d_fieldmask & QC_INO_SOFT)
2643 dm->dqb_isoftlimit = di->d_ino_softlimit;
2644 if (di->d_fieldmask & QC_INO_HARD)
2645 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2646 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2648 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2651 if (di->d_fieldmask & QC_SPC_TIMER) {
2652 dm->dqb_btime = di->d_spc_timer;
2654 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2657 if (di->d_fieldmask & QC_INO_TIMER) {
2658 dm->dqb_itime = di->d_ino_timer;
2660 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2664 if (!dm->dqb_bsoftlimit ||
2665 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2667 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2668 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2669 /* Set grace only if user hasn't provided his own... */
2670 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2673 if (!dm->dqb_isoftlimit ||
2674 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2676 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2677 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2678 /* Set grace only if user hasn't provided his own... */
2679 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2681 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2683 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2685 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2686 spin_unlock(&dq_data_lock);
2687 mark_dquot_dirty(dquot);
2692 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2693 struct qc_dqblk *di)
2695 struct dquot *dquot;
2698 dquot = dqget(sb, qid);
2699 if (IS_ERR(dquot)) {
2700 rc = PTR_ERR(dquot);
2703 rc = do_set_dqblk(dquot, di);
2708 EXPORT_SYMBOL(dquot_set_dqblk);
2710 /* Generic routine for getting common part of quota file information */
2711 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2713 struct mem_dqinfo *mi;
2714 struct qc_type_state *tstate;
2715 struct quota_info *dqopt = sb_dqopt(sb);
2718 memset(state, 0, sizeof(*state));
2719 for (type = 0; type < MAXQUOTAS; type++) {
2720 if (!sb_has_quota_active(sb, type))
2722 tstate = state->s_state + type;
2723 mi = sb_dqopt(sb)->info + type;
2724 tstate->flags = QCI_ACCT_ENABLED;
2725 spin_lock(&dq_data_lock);
2726 if (mi->dqi_flags & DQF_SYS_FILE)
2727 tstate->flags |= QCI_SYSFILE;
2728 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2729 tstate->flags |= QCI_ROOT_SQUASH;
2730 if (sb_has_quota_limits_enabled(sb, type))
2731 tstate->flags |= QCI_LIMITS_ENFORCED;
2732 tstate->spc_timelimit = mi->dqi_bgrace;
2733 tstate->ino_timelimit = mi->dqi_igrace;
2734 tstate->ino = dqopt->files[type]->i_ino;
2735 tstate->blocks = dqopt->files[type]->i_blocks;
2736 tstate->nextents = 1; /* We don't know... */
2737 spin_unlock(&dq_data_lock);
2741 EXPORT_SYMBOL(dquot_get_state);
2743 /* Generic routine for setting common part of quota file information */
2744 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2746 struct mem_dqinfo *mi;
2749 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2750 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2752 if (!sb_has_quota_active(sb, type))
2754 mi = sb_dqopt(sb)->info + type;
2755 if (ii->i_fieldmask & QC_FLAGS) {
2756 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2757 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2760 spin_lock(&dq_data_lock);
2761 if (ii->i_fieldmask & QC_SPC_TIMER)
2762 mi->dqi_bgrace = ii->i_spc_timelimit;
2763 if (ii->i_fieldmask & QC_INO_TIMER)
2764 mi->dqi_igrace = ii->i_ino_timelimit;
2765 if (ii->i_fieldmask & QC_FLAGS) {
2766 if (ii->i_flags & QCI_ROOT_SQUASH)
2767 mi->dqi_flags |= DQF_ROOT_SQUASH;
2769 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2771 spin_unlock(&dq_data_lock);
2772 mark_info_dirty(sb, type);
2773 /* Force write to disk */
2774 sb->dq_op->write_info(sb, type);
2777 EXPORT_SYMBOL(dquot_set_dqinfo);
2779 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2780 .quota_enable = dquot_quota_enable,
2781 .quota_disable = dquot_quota_disable,
2782 .quota_sync = dquot_quota_sync,
2783 .get_state = dquot_get_state,
2784 .set_info = dquot_set_dqinfo,
2785 .get_dqblk = dquot_get_dqblk,
2786 .get_nextdqblk = dquot_get_next_dqblk,
2787 .set_dqblk = dquot_set_dqblk
2789 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2791 static int do_proc_dqstats(struct ctl_table *table, int write,
2792 void __user *buffer, size_t *lenp, loff_t *ppos)
2794 unsigned int type = (int *)table->data - dqstats.stat;
2796 /* Update global table */
2797 dqstats.stat[type] =
2798 percpu_counter_sum_positive(&dqstats.counter[type]);
2799 return proc_dointvec(table, write, buffer, lenp, ppos);
2802 static struct ctl_table fs_dqstats_table[] = {
2804 .procname = "lookups",
2805 .data = &dqstats.stat[DQST_LOOKUPS],
2806 .maxlen = sizeof(int),
2808 .proc_handler = do_proc_dqstats,
2811 .procname = "drops",
2812 .data = &dqstats.stat[DQST_DROPS],
2813 .maxlen = sizeof(int),
2815 .proc_handler = do_proc_dqstats,
2818 .procname = "reads",
2819 .data = &dqstats.stat[DQST_READS],
2820 .maxlen = sizeof(int),
2822 .proc_handler = do_proc_dqstats,
2825 .procname = "writes",
2826 .data = &dqstats.stat[DQST_WRITES],
2827 .maxlen = sizeof(int),
2829 .proc_handler = do_proc_dqstats,
2832 .procname = "cache_hits",
2833 .data = &dqstats.stat[DQST_CACHE_HITS],
2834 .maxlen = sizeof(int),
2836 .proc_handler = do_proc_dqstats,
2839 .procname = "allocated_dquots",
2840 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2841 .maxlen = sizeof(int),
2843 .proc_handler = do_proc_dqstats,
2846 .procname = "free_dquots",
2847 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2848 .maxlen = sizeof(int),
2850 .proc_handler = do_proc_dqstats,
2853 .procname = "syncs",
2854 .data = &dqstats.stat[DQST_SYNCS],
2855 .maxlen = sizeof(int),
2857 .proc_handler = do_proc_dqstats,
2859 #ifdef CONFIG_PRINT_QUOTA_WARNING
2861 .procname = "warnings",
2862 .data = &flag_print_warnings,
2863 .maxlen = sizeof(int),
2865 .proc_handler = proc_dointvec,
2871 static struct ctl_table fs_table[] = {
2873 .procname = "quota",
2875 .child = fs_dqstats_table,
2880 static struct ctl_table sys_table[] = {
2889 static int __init dquot_init(void)
2892 unsigned long nr_hash, order;
2894 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2896 register_sysctl_table(sys_table);
2898 dquot_cachep = kmem_cache_create("dquot",
2899 sizeof(struct dquot), sizeof(unsigned long) * 4,
2900 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2901 SLAB_MEM_SPREAD|SLAB_PANIC),
2905 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2907 panic("Cannot create dquot hash table");
2909 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2910 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2912 panic("Cannot create dquot stat counters");
2915 /* Find power-of-two hlist_heads which can fit into allocation */
2916 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2920 } while (nr_hash >> dq_hash_bits);
2923 nr_hash = 1UL << dq_hash_bits;
2924 dq_hash_mask = nr_hash - 1;
2925 for (i = 0; i < nr_hash; i++)
2926 INIT_HLIST_HEAD(dquot_hash + i);
2928 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2929 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2931 register_shrinker(&dqcache_shrinker);
2935 fs_initcall(dquot_init);