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 five quota SMP locks:
86 * * dq_list_lock protects all lists with quotas and quota formats.
87 * * dquot->dq_dqb_lock protects data from dq_dqb
88 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
89 * consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
90 * dquot_transfer() can stabilize amount it transfers
91 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
92 * pointers in the inode
93 * * dq_state_lock protects modifications of quota state (on quotaon and
94 * quotaoff) and readers who care about latest values take it as well.
96 * The spinlock ordering is hence:
97 * dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
98 * dq_list_lock > dq_state_lock
100 * Note that some things (eg. sb pointer, type, id) doesn't change during
101 * the life of the dquot structure and so needn't to be protected by a lock
103 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
104 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
105 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
106 * inode and before dropping dquot references to avoid use of dquots after
107 * they are freed. dq_data_lock is used to serialize the pointer setting and
108 * clearing operations.
109 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
110 * inode is a quota file). Functions adding pointers from inode to dquots have
111 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
112 * have to do all pointer modifications before dropping dq_data_lock. This makes
113 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
114 * then drops all pointers to dquots from an inode.
116 * Each dquot has its dq_lock mutex. Dquot is locked when it is being read to
117 * memory (or space for it is being allocated) on the first dqget(), when it is
118 * being written out, and when it is being released on the last dqput(). The
119 * allocation and release operations are serialized by the dq_lock and by
120 * checking the use count in dquot_release().
122 * Lock ordering (including related VFS locks) is the following:
123 * s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
126 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
128 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
129 EXPORT_SYMBOL(dq_data_lock);
130 DEFINE_STATIC_SRCU(dquot_srcu);
132 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
134 void __quota_error(struct super_block *sb, const char *func,
135 const char *fmt, ...)
137 if (printk_ratelimit()) {
139 struct va_format vaf;
146 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
147 sb->s_id, func, &vaf);
152 EXPORT_SYMBOL(__quota_error);
154 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
155 static char *quotatypes[] = INITQFNAMES;
157 static struct quota_format_type *quota_formats; /* List of registered formats */
158 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
160 /* SLAB cache for dquot structures */
161 static struct kmem_cache *dquot_cachep;
163 int register_quota_format(struct quota_format_type *fmt)
165 spin_lock(&dq_list_lock);
166 fmt->qf_next = quota_formats;
168 spin_unlock(&dq_list_lock);
171 EXPORT_SYMBOL(register_quota_format);
173 void unregister_quota_format(struct quota_format_type *fmt)
175 struct quota_format_type **actqf;
177 spin_lock(&dq_list_lock);
178 for (actqf = "a_formats; *actqf && *actqf != fmt;
179 actqf = &(*actqf)->qf_next)
182 *actqf = (*actqf)->qf_next;
183 spin_unlock(&dq_list_lock);
185 EXPORT_SYMBOL(unregister_quota_format);
187 static struct quota_format_type *find_quota_format(int id)
189 struct quota_format_type *actqf;
191 spin_lock(&dq_list_lock);
192 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
193 actqf = actqf->qf_next)
195 if (!actqf || !try_module_get(actqf->qf_owner)) {
198 spin_unlock(&dq_list_lock);
200 for (qm = 0; module_names[qm].qm_fmt_id &&
201 module_names[qm].qm_fmt_id != id; qm++)
203 if (!module_names[qm].qm_fmt_id ||
204 request_module(module_names[qm].qm_mod_name))
207 spin_lock(&dq_list_lock);
208 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
209 actqf = actqf->qf_next)
211 if (actqf && !try_module_get(actqf->qf_owner))
214 spin_unlock(&dq_list_lock);
218 static void put_quota_format(struct quota_format_type *fmt)
220 module_put(fmt->qf_owner);
224 * Dquot List Management:
225 * The quota code uses three lists for dquot management: the inuse_list,
226 * free_dquots, and dquot_hash[] array. A single dquot structure may be
227 * on all three lists, depending on its current state.
229 * All dquots are placed to the end of inuse_list when first created, and this
230 * list is used for invalidate operation, which must look at every dquot.
232 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
233 * and this list is searched whenever we need an available dquot. Dquots are
234 * removed from the list as soon as they are used again, and
235 * dqstats.free_dquots gives the number of dquots on the list. When
236 * dquot is invalidated it's completely released from memory.
238 * Dquots with a specific identity (device, type and id) are placed on
239 * one of the dquot_hash[] hash chains. The provides an efficient search
240 * mechanism to locate a specific dquot.
243 static LIST_HEAD(inuse_list);
244 static LIST_HEAD(free_dquots);
245 static unsigned int dq_hash_bits, dq_hash_mask;
246 static struct hlist_head *dquot_hash;
248 struct dqstats dqstats;
249 EXPORT_SYMBOL(dqstats);
251 static qsize_t inode_get_rsv_space(struct inode *inode);
252 static qsize_t __inode_get_rsv_space(struct inode *inode);
253 static int __dquot_initialize(struct inode *inode, int type);
255 static inline unsigned int
256 hashfn(const struct super_block *sb, struct kqid qid)
258 unsigned int id = from_kqid(&init_user_ns, qid);
262 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
263 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
267 * Following list functions expect dq_list_lock to be held
269 static inline void insert_dquot_hash(struct dquot *dquot)
271 struct hlist_head *head;
272 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
273 hlist_add_head(&dquot->dq_hash, head);
276 static inline void remove_dquot_hash(struct dquot *dquot)
278 hlist_del_init(&dquot->dq_hash);
281 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
284 struct hlist_node *node;
287 hlist_for_each (node, dquot_hash+hashent) {
288 dquot = hlist_entry(node, struct dquot, dq_hash);
289 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
295 /* Add a dquot to the tail of the free list */
296 static inline void put_dquot_last(struct dquot *dquot)
298 list_add_tail(&dquot->dq_free, &free_dquots);
299 dqstats_inc(DQST_FREE_DQUOTS);
302 static inline void remove_free_dquot(struct dquot *dquot)
304 if (list_empty(&dquot->dq_free))
306 list_del_init(&dquot->dq_free);
307 dqstats_dec(DQST_FREE_DQUOTS);
310 static inline void put_inuse(struct dquot *dquot)
312 /* We add to the back of inuse list so we don't have to restart
313 * when traversing this list and we block */
314 list_add_tail(&dquot->dq_inuse, &inuse_list);
315 dqstats_inc(DQST_ALLOC_DQUOTS);
318 static inline void remove_inuse(struct dquot *dquot)
320 dqstats_dec(DQST_ALLOC_DQUOTS);
321 list_del(&dquot->dq_inuse);
324 * End of list functions needing dq_list_lock
327 static void wait_on_dquot(struct dquot *dquot)
329 mutex_lock(&dquot->dq_lock);
330 mutex_unlock(&dquot->dq_lock);
333 static inline int dquot_dirty(struct dquot *dquot)
335 return test_bit(DQ_MOD_B, &dquot->dq_flags);
338 static inline int mark_dquot_dirty(struct dquot *dquot)
340 return dquot->dq_sb->dq_op->mark_dirty(dquot);
343 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
344 int dquot_mark_dquot_dirty(struct dquot *dquot)
348 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
351 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
352 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
354 /* If quota is dirty already, we don't have to acquire dq_list_lock */
355 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
358 spin_lock(&dq_list_lock);
359 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
360 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
361 info[dquot->dq_id.type].dqi_dirty_list);
364 spin_unlock(&dq_list_lock);
367 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
369 /* Dirtify all the dquots - this can block when journalling */
370 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
375 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
377 /* Even in case of error we have to continue */
378 ret = mark_dquot_dirty(dquot[cnt]);
385 static inline void dqput_all(struct dquot **dquot)
389 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
393 static inline int clear_dquot_dirty(struct dquot *dquot)
395 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
396 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
398 spin_lock(&dq_list_lock);
399 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
400 spin_unlock(&dq_list_lock);
403 list_del_init(&dquot->dq_dirty);
404 spin_unlock(&dq_list_lock);
408 void mark_info_dirty(struct super_block *sb, int type)
410 spin_lock(&dq_data_lock);
411 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
412 spin_unlock(&dq_data_lock);
414 EXPORT_SYMBOL(mark_info_dirty);
417 * Read dquot from disk and alloc space for it
420 int dquot_acquire(struct dquot *dquot)
422 int ret = 0, ret2 = 0;
423 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
425 mutex_lock(&dquot->dq_lock);
426 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
427 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
430 /* Make sure flags update is visible after dquot has been filled */
431 smp_mb__before_atomic();
432 set_bit(DQ_READ_B, &dquot->dq_flags);
433 /* Instantiate dquot if needed */
434 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
435 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
436 /* Write the info if needed */
437 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
438 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
439 dquot->dq_sb, dquot->dq_id.type);
449 * Make sure flags update is visible after on-disk struct has been
450 * allocated. Paired with smp_rmb() in dqget().
452 smp_mb__before_atomic();
453 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
455 mutex_unlock(&dquot->dq_lock);
458 EXPORT_SYMBOL(dquot_acquire);
461 * Write dquot to disk
463 int dquot_commit(struct dquot *dquot)
466 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
468 mutex_lock(&dquot->dq_lock);
469 if (!clear_dquot_dirty(dquot))
471 /* Inactive dquot can be only if there was error during read/init
472 * => we have better not writing it */
473 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
474 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
478 mutex_unlock(&dquot->dq_lock);
481 EXPORT_SYMBOL(dquot_commit);
486 int dquot_release(struct dquot *dquot)
488 int ret = 0, ret2 = 0;
489 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
491 mutex_lock(&dquot->dq_lock);
492 /* Check whether we are not racing with some other dqget() */
493 if (atomic_read(&dquot->dq_count) > 1)
495 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
496 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
498 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
499 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
500 dquot->dq_sb, dquot->dq_id.type);
505 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
507 mutex_unlock(&dquot->dq_lock);
510 EXPORT_SYMBOL(dquot_release);
512 void dquot_destroy(struct dquot *dquot)
514 kmem_cache_free(dquot_cachep, dquot);
516 EXPORT_SYMBOL(dquot_destroy);
518 static inline void do_destroy_dquot(struct dquot *dquot)
520 dquot->dq_sb->dq_op->destroy_dquot(dquot);
523 /* Invalidate all dquots on the list. Note that this function is called after
524 * quota is disabled and pointers from inodes removed so there cannot be new
525 * quota users. There can still be some users of quotas due to inodes being
526 * just deleted or pruned by prune_icache() (those are not attached to any
527 * list) or parallel quotactl call. We have to wait for such users.
529 static void invalidate_dquots(struct super_block *sb, int type)
531 struct dquot *dquot, *tmp;
534 spin_lock(&dq_list_lock);
535 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
536 if (dquot->dq_sb != sb)
538 if (dquot->dq_id.type != type)
540 /* Wait for dquot users */
541 if (atomic_read(&dquot->dq_count)) {
543 spin_unlock(&dq_list_lock);
545 * Once dqput() wakes us up, we know it's time to free
547 * IMPORTANT: we rely on the fact that there is always
548 * at most one process waiting for dquot to free.
549 * Otherwise dq_count would be > 1 and we would never
552 wait_event(dquot_ref_wq,
553 atomic_read(&dquot->dq_count) == 1);
555 /* At this moment dquot() need not exist (it could be
556 * reclaimed by prune_dqcache(). Hence we must
561 * Quota now has no users and it has been written on last
564 remove_dquot_hash(dquot);
565 remove_free_dquot(dquot);
567 do_destroy_dquot(dquot);
569 spin_unlock(&dq_list_lock);
572 /* Call callback for every active dquot on given filesystem */
573 int dquot_scan_active(struct super_block *sb,
574 int (*fn)(struct dquot *dquot, unsigned long priv),
577 struct dquot *dquot, *old_dquot = NULL;
580 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
582 spin_lock(&dq_list_lock);
583 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
584 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
586 if (dquot->dq_sb != sb)
588 /* Now we have active dquot so we can just increase use count */
589 atomic_inc(&dquot->dq_count);
590 spin_unlock(&dq_list_lock);
591 dqstats_inc(DQST_LOOKUPS);
595 * ->release_dquot() can be racing with us. Our reference
596 * protects us from new calls to it so just wait for any
597 * outstanding call and recheck the DQ_ACTIVE_B after that.
599 wait_on_dquot(dquot);
600 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
601 ret = fn(dquot, priv);
605 spin_lock(&dq_list_lock);
606 /* We are safe to continue now because our dquot could not
607 * be moved out of the inuse list while we hold the reference */
609 spin_unlock(&dq_list_lock);
614 EXPORT_SYMBOL(dquot_scan_active);
616 /* Write all dquot structures to quota files */
617 int dquot_writeback_dquots(struct super_block *sb, int type)
619 struct list_head *dirty;
621 struct quota_info *dqopt = sb_dqopt(sb);
625 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
627 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
628 if (type != -1 && cnt != type)
630 if (!sb_has_quota_active(sb, cnt))
632 spin_lock(&dq_list_lock);
633 dirty = &dqopt->info[cnt].dqi_dirty_list;
634 while (!list_empty(dirty)) {
635 dquot = list_first_entry(dirty, struct dquot,
638 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
640 /* Now we have active dquot from which someone is
641 * holding reference so we can safely just increase
644 spin_unlock(&dq_list_lock);
645 dqstats_inc(DQST_LOOKUPS);
646 err = sb->dq_op->write_dquot(dquot);
650 spin_lock(&dq_list_lock);
652 spin_unlock(&dq_list_lock);
655 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
656 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
657 && info_dirty(&dqopt->info[cnt]))
658 sb->dq_op->write_info(sb, cnt);
659 dqstats_inc(DQST_SYNCS);
663 EXPORT_SYMBOL(dquot_writeback_dquots);
665 /* Write all dquot structures to disk and make them visible from userspace */
666 int dquot_quota_sync(struct super_block *sb, int type)
668 struct quota_info *dqopt = sb_dqopt(sb);
672 ret = dquot_writeback_dquots(sb, type);
675 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
678 /* This is not very clever (and fast) but currently I don't know about
679 * any other simple way of getting quota data to disk and we must get
680 * them there for userspace to be visible... */
681 if (sb->s_op->sync_fs)
682 sb->s_op->sync_fs(sb, 1);
683 sync_blockdev(sb->s_bdev);
686 * Now when everything is written we can discard the pagecache so
687 * that userspace sees the changes.
689 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
690 if (type != -1 && cnt != type)
692 if (!sb_has_quota_active(sb, cnt))
694 inode_lock(dqopt->files[cnt]);
695 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
696 inode_unlock(dqopt->files[cnt]);
701 EXPORT_SYMBOL(dquot_quota_sync);
704 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
706 struct list_head *head;
708 unsigned long freed = 0;
710 spin_lock(&dq_list_lock);
711 head = free_dquots.prev;
712 while (head != &free_dquots && sc->nr_to_scan) {
713 dquot = list_entry(head, struct dquot, dq_free);
714 remove_dquot_hash(dquot);
715 remove_free_dquot(dquot);
717 do_destroy_dquot(dquot);
720 head = free_dquots.prev;
722 spin_unlock(&dq_list_lock);
727 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
729 return vfs_pressure_ratio(
730 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
733 static struct shrinker dqcache_shrinker = {
734 .count_objects = dqcache_shrink_count,
735 .scan_objects = dqcache_shrink_scan,
736 .seeks = DEFAULT_SEEKS,
740 * Put reference to dquot
742 void dqput(struct dquot *dquot)
748 #ifdef CONFIG_QUOTA_DEBUG
749 if (!atomic_read(&dquot->dq_count)) {
750 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
751 quotatypes[dquot->dq_id.type],
752 from_kqid(&init_user_ns, dquot->dq_id));
756 dqstats_inc(DQST_DROPS);
758 spin_lock(&dq_list_lock);
759 if (atomic_read(&dquot->dq_count) > 1) {
760 /* We have more than one user... nothing to do */
761 atomic_dec(&dquot->dq_count);
762 /* Releasing dquot during quotaoff phase? */
763 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
764 atomic_read(&dquot->dq_count) == 1)
765 wake_up(&dquot_ref_wq);
766 spin_unlock(&dq_list_lock);
769 /* Need to release dquot? */
770 if (dquot_dirty(dquot)) {
771 spin_unlock(&dq_list_lock);
772 /* Commit dquot before releasing */
773 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
775 quota_error(dquot->dq_sb, "Can't write quota structure"
776 " (error %d). Quota may get out of sync!",
779 * We clear dirty bit anyway, so that we avoid
782 clear_dquot_dirty(dquot);
786 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
787 spin_unlock(&dq_list_lock);
788 dquot->dq_sb->dq_op->release_dquot(dquot);
791 atomic_dec(&dquot->dq_count);
792 #ifdef CONFIG_QUOTA_DEBUG
794 BUG_ON(!list_empty(&dquot->dq_free));
796 put_dquot_last(dquot);
797 spin_unlock(&dq_list_lock);
799 EXPORT_SYMBOL(dqput);
801 struct dquot *dquot_alloc(struct super_block *sb, int type)
803 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
805 EXPORT_SYMBOL(dquot_alloc);
807 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
811 dquot = sb->dq_op->alloc_dquot(sb, type);
815 mutex_init(&dquot->dq_lock);
816 INIT_LIST_HEAD(&dquot->dq_free);
817 INIT_LIST_HEAD(&dquot->dq_inuse);
818 INIT_HLIST_NODE(&dquot->dq_hash);
819 INIT_LIST_HEAD(&dquot->dq_dirty);
821 dquot->dq_id = make_kqid_invalid(type);
822 atomic_set(&dquot->dq_count, 1);
823 spin_lock_init(&dquot->dq_dqb_lock);
829 * Get reference to dquot
831 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
832 * destroying our dquot by:
833 * a) checking for quota flags under dq_list_lock and
834 * b) getting a reference to dquot before we release dq_list_lock
836 struct dquot *dqget(struct super_block *sb, struct kqid qid)
838 unsigned int hashent = hashfn(sb, qid);
839 struct dquot *dquot, *empty = NULL;
841 if (!qid_has_mapping(sb->s_user_ns, qid))
842 return ERR_PTR(-EINVAL);
844 if (!sb_has_quota_active(sb, qid.type))
845 return ERR_PTR(-ESRCH);
847 spin_lock(&dq_list_lock);
848 spin_lock(&dq_state_lock);
849 if (!sb_has_quota_active(sb, qid.type)) {
850 spin_unlock(&dq_state_lock);
851 spin_unlock(&dq_list_lock);
852 dquot = ERR_PTR(-ESRCH);
855 spin_unlock(&dq_state_lock);
857 dquot = find_dquot(hashent, sb, qid);
860 spin_unlock(&dq_list_lock);
861 empty = get_empty_dquot(sb, qid.type);
863 schedule(); /* Try to wait for a moment... */
869 /* all dquots go on the inuse_list */
871 /* hash it first so it can be found */
872 insert_dquot_hash(dquot);
873 spin_unlock(&dq_list_lock);
874 dqstats_inc(DQST_LOOKUPS);
876 if (!atomic_read(&dquot->dq_count))
877 remove_free_dquot(dquot);
878 atomic_inc(&dquot->dq_count);
879 spin_unlock(&dq_list_lock);
880 dqstats_inc(DQST_CACHE_HITS);
881 dqstats_inc(DQST_LOOKUPS);
883 /* Wait for dq_lock - after this we know that either dquot_release() is
884 * already finished or it will be canceled due to dq_count > 1 test */
885 wait_on_dquot(dquot);
886 /* Read the dquot / allocate space in quota file */
887 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
890 err = sb->dq_op->acquire_dquot(dquot);
893 dquot = ERR_PTR(err);
898 * Make sure following reads see filled structure - paired with
899 * smp_mb__before_atomic() in dquot_acquire().
902 #ifdef CONFIG_QUOTA_DEBUG
903 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
907 do_destroy_dquot(empty);
911 EXPORT_SYMBOL(dqget);
913 static inline struct dquot **i_dquot(struct inode *inode)
915 return inode->i_sb->s_op->get_dquots(inode);
918 static int dqinit_needed(struct inode *inode, int type)
920 struct dquot * const *dquots;
923 if (IS_NOQUOTA(inode))
926 dquots = i_dquot(inode);
928 return !dquots[type];
929 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
935 /* This routine is guarded by s_umount semaphore */
936 static void add_dquot_ref(struct super_block *sb, int type)
938 struct inode *inode, *old_inode = NULL;
939 #ifdef CONFIG_QUOTA_DEBUG
943 spin_lock(&sb->s_inode_list_lock);
944 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
945 spin_lock(&inode->i_lock);
946 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
947 !atomic_read(&inode->i_writecount) ||
948 !dqinit_needed(inode, type)) {
949 spin_unlock(&inode->i_lock);
953 spin_unlock(&inode->i_lock);
954 spin_unlock(&sb->s_inode_list_lock);
956 #ifdef CONFIG_QUOTA_DEBUG
957 if (unlikely(inode_get_rsv_space(inode) > 0))
961 __dquot_initialize(inode, type);
964 * We hold a reference to 'inode' so it couldn't have been
965 * removed from s_inodes list while we dropped the
966 * s_inode_list_lock. We cannot iput the inode now as we can be
967 * holding the last reference and we cannot iput it under
968 * s_inode_list_lock. So we keep the reference and iput it
972 spin_lock(&sb->s_inode_list_lock);
974 spin_unlock(&sb->s_inode_list_lock);
977 #ifdef CONFIG_QUOTA_DEBUG
979 quota_error(sb, "Writes happened before quota was turned on "
980 "thus quota information is probably inconsistent. "
981 "Please run quotacheck(8)");
987 * Remove references to dquots from inode and add dquot to list for freeing
988 * if we have the last reference to dquot
990 static void remove_inode_dquot_ref(struct inode *inode, int type,
991 struct list_head *tofree_head)
993 struct dquot **dquots = i_dquot(inode);
994 struct dquot *dquot = dquots[type];
1000 if (list_empty(&dquot->dq_free)) {
1002 * The inode still has reference to dquot so it can't be in the
1005 spin_lock(&dq_list_lock);
1006 list_add(&dquot->dq_free, tofree_head);
1007 spin_unlock(&dq_list_lock);
1010 * Dquot is already in a list to put so we won't drop the last
1018 * Free list of dquots
1019 * Dquots are removed from inodes and no new references can be got so we are
1020 * the only ones holding reference
1022 static void put_dquot_list(struct list_head *tofree_head)
1024 struct list_head *act_head;
1025 struct dquot *dquot;
1027 act_head = tofree_head->next;
1028 while (act_head != tofree_head) {
1029 dquot = list_entry(act_head, struct dquot, dq_free);
1030 act_head = act_head->next;
1031 /* Remove dquot from the list so we won't have problems... */
1032 list_del_init(&dquot->dq_free);
1037 static void remove_dquot_ref(struct super_block *sb, int type,
1038 struct list_head *tofree_head)
1040 struct inode *inode;
1043 spin_lock(&sb->s_inode_list_lock);
1044 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1046 * We have to scan also I_NEW inodes because they can already
1047 * have quota pointer initialized. Luckily, we need to touch
1048 * only quota pointers and these have separate locking
1051 spin_lock(&dq_data_lock);
1052 if (!IS_NOQUOTA(inode)) {
1053 if (unlikely(inode_get_rsv_space(inode) > 0))
1055 remove_inode_dquot_ref(inode, type, tofree_head);
1057 spin_unlock(&dq_data_lock);
1059 spin_unlock(&sb->s_inode_list_lock);
1060 #ifdef CONFIG_QUOTA_DEBUG
1062 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1063 " was disabled thus quota information is probably "
1064 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1069 /* Gather all references from inodes and drop them */
1070 static void drop_dquot_ref(struct super_block *sb, int type)
1072 LIST_HEAD(tofree_head);
1075 remove_dquot_ref(sb, type, &tofree_head);
1076 synchronize_srcu(&dquot_srcu);
1077 put_dquot_list(&tofree_head);
1082 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1084 if (dquot->dq_dqb.dqb_rsvspace >= number)
1085 dquot->dq_dqb.dqb_rsvspace -= number;
1088 dquot->dq_dqb.dqb_rsvspace = 0;
1092 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1094 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1095 dquot->dq_dqb.dqb_curinodes >= number)
1096 dquot->dq_dqb.dqb_curinodes -= number;
1098 dquot->dq_dqb.dqb_curinodes = 0;
1099 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1100 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1101 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1104 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1106 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1107 dquot->dq_dqb.dqb_curspace >= number)
1108 dquot->dq_dqb.dqb_curspace -= number;
1110 dquot->dq_dqb.dqb_curspace = 0;
1111 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1112 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1113 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1117 struct super_block *w_sb;
1118 struct kqid w_dq_id;
1122 static int warning_issued(struct dquot *dquot, const int warntype)
1124 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1125 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1126 ((warntype == QUOTA_NL_IHARDWARN ||
1127 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1131 return test_and_set_bit(flag, &dquot->dq_flags);
1134 #ifdef CONFIG_PRINT_QUOTA_WARNING
1135 static int flag_print_warnings = 1;
1137 static int need_print_warning(struct dquot_warn *warn)
1139 if (!flag_print_warnings)
1142 switch (warn->w_dq_id.type) {
1144 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1146 return in_group_p(warn->w_dq_id.gid);
1153 /* Print warning to user which exceeded quota */
1154 static void print_warning(struct dquot_warn *warn)
1157 struct tty_struct *tty;
1158 int warntype = warn->w_type;
1160 if (warntype == QUOTA_NL_IHARDBELOW ||
1161 warntype == QUOTA_NL_ISOFTBELOW ||
1162 warntype == QUOTA_NL_BHARDBELOW ||
1163 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1166 tty = get_current_tty();
1169 tty_write_message(tty, warn->w_sb->s_id);
1170 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1171 tty_write_message(tty, ": warning, ");
1173 tty_write_message(tty, ": write failed, ");
1174 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1176 case QUOTA_NL_IHARDWARN:
1177 msg = " file limit reached.\r\n";
1179 case QUOTA_NL_ISOFTLONGWARN:
1180 msg = " file quota exceeded too long.\r\n";
1182 case QUOTA_NL_ISOFTWARN:
1183 msg = " file quota exceeded.\r\n";
1185 case QUOTA_NL_BHARDWARN:
1186 msg = " block limit reached.\r\n";
1188 case QUOTA_NL_BSOFTLONGWARN:
1189 msg = " block quota exceeded too long.\r\n";
1191 case QUOTA_NL_BSOFTWARN:
1192 msg = " block quota exceeded.\r\n";
1195 tty_write_message(tty, msg);
1200 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1203 if (warning_issued(dquot, warntype))
1205 warn->w_type = warntype;
1206 warn->w_sb = dquot->dq_sb;
1207 warn->w_dq_id = dquot->dq_id;
1211 * Write warnings to the console and send warning messages over netlink.
1213 * Note that this function can call into tty and networking code.
1215 static void flush_warnings(struct dquot_warn *warn)
1219 for (i = 0; i < MAXQUOTAS; i++) {
1220 if (warn[i].w_type == QUOTA_NL_NOWARN)
1222 #ifdef CONFIG_PRINT_QUOTA_WARNING
1223 print_warning(&warn[i]);
1225 quota_send_warning(warn[i].w_dq_id,
1226 warn[i].w_sb->s_dev, warn[i].w_type);
1230 static int ignore_hardlimit(struct dquot *dquot)
1232 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1234 return capable(CAP_SYS_RESOURCE) &&
1235 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1236 !(info->dqi_flags & DQF_ROOT_SQUASH));
1239 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1240 struct dquot_warn *warn)
1245 spin_lock(&dquot->dq_dqb_lock);
1246 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1247 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1248 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1251 if (dquot->dq_dqb.dqb_ihardlimit &&
1252 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1253 !ignore_hardlimit(dquot)) {
1254 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1259 if (dquot->dq_dqb.dqb_isoftlimit &&
1260 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1261 dquot->dq_dqb.dqb_itime &&
1262 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1263 !ignore_hardlimit(dquot)) {
1264 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1269 if (dquot->dq_dqb.dqb_isoftlimit &&
1270 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1271 dquot->dq_dqb.dqb_itime == 0) {
1272 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1273 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1274 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1277 dquot->dq_dqb.dqb_curinodes = newinodes;
1280 spin_unlock(&dquot->dq_dqb_lock);
1284 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1285 qsize_t rsv_space, unsigned int flags,
1286 struct dquot_warn *warn)
1289 struct super_block *sb = dquot->dq_sb;
1292 spin_lock(&dquot->dq_dqb_lock);
1293 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1294 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1297 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1298 + space + rsv_space;
1300 if (flags & DQUOT_SPACE_NOFAIL)
1303 if (dquot->dq_dqb.dqb_bhardlimit &&
1304 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1305 !ignore_hardlimit(dquot)) {
1306 if (flags & DQUOT_SPACE_WARN)
1307 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1312 if (dquot->dq_dqb.dqb_bsoftlimit &&
1313 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1314 dquot->dq_dqb.dqb_btime &&
1315 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1316 !ignore_hardlimit(dquot)) {
1317 if (flags & DQUOT_SPACE_WARN)
1318 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1323 if (dquot->dq_dqb.dqb_bsoftlimit &&
1324 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1325 dquot->dq_dqb.dqb_btime == 0) {
1326 if (flags & DQUOT_SPACE_WARN) {
1327 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1328 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1329 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1332 * We don't allow preallocation to exceed softlimit so exceeding will
1340 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1341 dquot->dq_dqb.dqb_curspace += space;
1343 spin_unlock(&dquot->dq_dqb_lock);
1347 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1351 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1352 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1353 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1354 return QUOTA_NL_NOWARN;
1356 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1357 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1358 return QUOTA_NL_ISOFTBELOW;
1359 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1360 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1361 return QUOTA_NL_IHARDBELOW;
1362 return QUOTA_NL_NOWARN;
1365 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1367 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1368 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1369 return QUOTA_NL_NOWARN;
1371 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1372 return QUOTA_NL_BSOFTBELOW;
1373 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1374 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1375 return QUOTA_NL_BHARDBELOW;
1376 return QUOTA_NL_NOWARN;
1379 static int dquot_active(const struct inode *inode)
1381 struct super_block *sb = inode->i_sb;
1383 if (IS_NOQUOTA(inode))
1385 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1389 * Initialize quota pointers in inode
1391 * It is better to call this function outside of any transaction as it
1392 * might need a lot of space in journal for dquot structure allocation.
1394 static int __dquot_initialize(struct inode *inode, int type)
1396 int cnt, init_needed = 0;
1397 struct dquot **dquots, *got[MAXQUOTAS] = {};
1398 struct super_block *sb = inode->i_sb;
1402 if (!dquot_active(inode))
1405 dquots = i_dquot(inode);
1407 /* First get references to structures we might need. */
1408 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1412 struct dquot *dquot;
1414 if (type != -1 && cnt != type)
1417 * The i_dquot should have been initialized in most cases,
1418 * we check it without locking here to avoid unnecessary
1419 * dqget()/dqput() calls.
1424 if (!sb_has_quota_active(sb, cnt))
1431 qid = make_kqid_uid(inode->i_uid);
1434 qid = make_kqid_gid(inode->i_gid);
1437 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1440 qid = make_kqid_projid(projid);
1443 dquot = dqget(sb, qid);
1444 if (IS_ERR(dquot)) {
1445 /* We raced with somebody turning quotas off... */
1446 if (PTR_ERR(dquot) != -ESRCH) {
1447 ret = PTR_ERR(dquot);
1455 /* All required i_dquot has been initialized */
1459 spin_lock(&dq_data_lock);
1460 if (IS_NOQUOTA(inode))
1462 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1463 if (type != -1 && cnt != type)
1465 /* Avoid races with quotaoff() */
1466 if (!sb_has_quota_active(sb, cnt))
1468 /* We could race with quotaon or dqget() could have failed */
1472 dquots[cnt] = got[cnt];
1475 * Make quota reservation system happy if someone
1476 * did a write before quota was turned on
1478 rsv = inode_get_rsv_space(inode);
1479 if (unlikely(rsv)) {
1480 spin_lock(&inode->i_lock);
1481 /* Get reservation again under proper lock */
1482 rsv = __inode_get_rsv_space(inode);
1483 spin_lock(&dquots[cnt]->dq_dqb_lock);
1484 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1485 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1486 spin_unlock(&inode->i_lock);
1491 spin_unlock(&dq_data_lock);
1493 /* Drop unused references */
1499 int dquot_initialize(struct inode *inode)
1501 return __dquot_initialize(inode, -1);
1503 EXPORT_SYMBOL(dquot_initialize);
1505 bool dquot_initialize_needed(struct inode *inode)
1507 struct dquot **dquots;
1510 if (!dquot_active(inode))
1513 dquots = i_dquot(inode);
1514 for (i = 0; i < MAXQUOTAS; i++)
1515 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1519 EXPORT_SYMBOL(dquot_initialize_needed);
1522 * Release all quotas referenced by inode.
1524 * This function only be called on inode free or converting
1525 * a file to quota file, no other users for the i_dquot in
1526 * both cases, so we needn't call synchronize_srcu() after
1529 static void __dquot_drop(struct inode *inode)
1532 struct dquot **dquots = i_dquot(inode);
1533 struct dquot *put[MAXQUOTAS];
1535 spin_lock(&dq_data_lock);
1536 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1537 put[cnt] = dquots[cnt];
1540 spin_unlock(&dq_data_lock);
1544 void dquot_drop(struct inode *inode)
1546 struct dquot * const *dquots;
1549 if (IS_NOQUOTA(inode))
1553 * Test before calling to rule out calls from proc and such
1554 * where we are not allowed to block. Note that this is
1555 * actually reliable test even without the lock - the caller
1556 * must assure that nobody can come after the DQUOT_DROP and
1557 * add quota pointers back anyway.
1559 dquots = i_dquot(inode);
1560 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1565 if (cnt < MAXQUOTAS)
1566 __dquot_drop(inode);
1568 EXPORT_SYMBOL(dquot_drop);
1571 * inode_reserved_space is managed internally by quota, and protected by
1572 * i_lock similar to i_blocks+i_bytes.
1574 static qsize_t *inode_reserved_space(struct inode * inode)
1576 /* Filesystem must explicitly define it's own method in order to use
1577 * quota reservation interface */
1578 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1579 return inode->i_sb->dq_op->get_reserved_space(inode);
1582 static qsize_t __inode_get_rsv_space(struct inode *inode)
1584 if (!inode->i_sb->dq_op->get_reserved_space)
1586 return *inode_reserved_space(inode);
1589 static qsize_t inode_get_rsv_space(struct inode *inode)
1593 if (!inode->i_sb->dq_op->get_reserved_space)
1595 spin_lock(&inode->i_lock);
1596 ret = __inode_get_rsv_space(inode);
1597 spin_unlock(&inode->i_lock);
1602 * This functions updates i_blocks+i_bytes fields and quota information
1603 * (together with appropriate checks).
1605 * NOTE: We absolutely rely on the fact that caller dirties the inode
1606 * (usually helpers in quotaops.h care about this) and holds a handle for
1607 * the current transaction so that dquot write and inode write go into the
1612 * This operation can block, but only after everything is updated
1614 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1616 int cnt, ret = 0, index;
1617 struct dquot_warn warn[MAXQUOTAS];
1618 int reserve = flags & DQUOT_SPACE_RESERVE;
1619 struct dquot **dquots;
1621 if (!dquot_active(inode)) {
1623 spin_lock(&inode->i_lock);
1624 *inode_reserved_space(inode) += number;
1625 spin_unlock(&inode->i_lock);
1627 inode_add_bytes(inode, number);
1632 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1633 warn[cnt].w_type = QUOTA_NL_NOWARN;
1635 dquots = i_dquot(inode);
1636 index = srcu_read_lock(&dquot_srcu);
1637 spin_lock(&inode->i_lock);
1638 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1641 if (flags & DQUOT_SPACE_RESERVE) {
1642 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1645 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1649 /* Back out changes we already did */
1650 for (cnt--; cnt >= 0; cnt--) {
1653 spin_lock(&dquots[cnt]->dq_dqb_lock);
1654 if (flags & DQUOT_SPACE_RESERVE) {
1655 dquots[cnt]->dq_dqb.dqb_rsvspace -=
1658 dquots[cnt]->dq_dqb.dqb_curspace -=
1661 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1663 spin_unlock(&inode->i_lock);
1664 goto out_flush_warn;
1668 *inode_reserved_space(inode) += number;
1670 __inode_add_bytes(inode, number);
1671 spin_unlock(&inode->i_lock);
1674 goto out_flush_warn;
1675 mark_all_dquot_dirty(dquots);
1677 srcu_read_unlock(&dquot_srcu, index);
1678 flush_warnings(warn);
1682 EXPORT_SYMBOL(__dquot_alloc_space);
1685 * This operation can block, but only after everything is updated
1687 int dquot_alloc_inode(struct inode *inode)
1689 int cnt, ret = 0, index;
1690 struct dquot_warn warn[MAXQUOTAS];
1691 struct dquot * const *dquots;
1693 if (!dquot_active(inode))
1695 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1696 warn[cnt].w_type = QUOTA_NL_NOWARN;
1698 dquots = i_dquot(inode);
1699 index = srcu_read_lock(&dquot_srcu);
1700 spin_lock(&inode->i_lock);
1701 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1704 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1706 for (cnt--; cnt >= 0; cnt--) {
1709 /* Back out changes we already did */
1710 spin_lock(&dquots[cnt]->dq_dqb_lock);
1711 dquots[cnt]->dq_dqb.dqb_curinodes--;
1712 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1719 spin_unlock(&inode->i_lock);
1721 mark_all_dquot_dirty(dquots);
1722 srcu_read_unlock(&dquot_srcu, index);
1723 flush_warnings(warn);
1726 EXPORT_SYMBOL(dquot_alloc_inode);
1729 * Convert in-memory reserved quotas to real consumed quotas
1731 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1733 struct dquot **dquots;
1736 if (!dquot_active(inode)) {
1737 spin_lock(&inode->i_lock);
1738 *inode_reserved_space(inode) -= number;
1739 __inode_add_bytes(inode, number);
1740 spin_unlock(&inode->i_lock);
1744 dquots = i_dquot(inode);
1745 index = srcu_read_lock(&dquot_srcu);
1746 spin_lock(&inode->i_lock);
1747 /* Claim reserved quotas to allocated quotas */
1748 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1750 struct dquot *dquot = dquots[cnt];
1752 spin_lock(&dquot->dq_dqb_lock);
1753 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1754 number = dquot->dq_dqb.dqb_rsvspace;
1755 dquot->dq_dqb.dqb_curspace += number;
1756 dquot->dq_dqb.dqb_rsvspace -= number;
1757 spin_unlock(&dquot->dq_dqb_lock);
1760 /* Update inode bytes */
1761 *inode_reserved_space(inode) -= number;
1762 __inode_add_bytes(inode, number);
1763 spin_unlock(&inode->i_lock);
1764 mark_all_dquot_dirty(dquots);
1765 srcu_read_unlock(&dquot_srcu, index);
1768 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1771 * Convert allocated space back to in-memory reserved quotas
1773 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1775 struct dquot **dquots;
1778 if (!dquot_active(inode)) {
1779 spin_lock(&inode->i_lock);
1780 *inode_reserved_space(inode) += number;
1781 __inode_sub_bytes(inode, number);
1782 spin_unlock(&inode->i_lock);
1786 dquots = i_dquot(inode);
1787 index = srcu_read_lock(&dquot_srcu);
1788 spin_lock(&inode->i_lock);
1789 /* Claim reserved quotas to allocated quotas */
1790 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1792 struct dquot *dquot = dquots[cnt];
1794 spin_lock(&dquot->dq_dqb_lock);
1795 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1796 number = dquot->dq_dqb.dqb_curspace;
1797 dquot->dq_dqb.dqb_rsvspace += number;
1798 dquot->dq_dqb.dqb_curspace -= number;
1799 spin_unlock(&dquot->dq_dqb_lock);
1802 /* Update inode bytes */
1803 *inode_reserved_space(inode) += number;
1804 __inode_sub_bytes(inode, number);
1805 spin_unlock(&inode->i_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)) {
1824 spin_lock(&inode->i_lock);
1825 *inode_reserved_space(inode) -= number;
1826 spin_unlock(&inode->i_lock);
1828 inode_sub_bytes(inode, number);
1833 dquots = i_dquot(inode);
1834 index = srcu_read_lock(&dquot_srcu);
1835 spin_lock(&inode->i_lock);
1836 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1839 warn[cnt].w_type = QUOTA_NL_NOWARN;
1842 spin_lock(&dquots[cnt]->dq_dqb_lock);
1843 wtype = info_bdq_free(dquots[cnt], number);
1844 if (wtype != QUOTA_NL_NOWARN)
1845 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1847 dquot_free_reserved_space(dquots[cnt], number);
1849 dquot_decr_space(dquots[cnt], number);
1850 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1853 *inode_reserved_space(inode) -= number;
1855 __inode_sub_bytes(inode, number);
1856 spin_unlock(&inode->i_lock);
1860 mark_all_dquot_dirty(dquots);
1862 srcu_read_unlock(&dquot_srcu, index);
1863 flush_warnings(warn);
1865 EXPORT_SYMBOL(__dquot_free_space);
1868 * This operation can block, but only after everything is updated
1870 void dquot_free_inode(struct inode *inode)
1873 struct dquot_warn warn[MAXQUOTAS];
1874 struct dquot * const *dquots;
1877 if (!dquot_active(inode))
1880 dquots = i_dquot(inode);
1881 index = srcu_read_lock(&dquot_srcu);
1882 spin_lock(&inode->i_lock);
1883 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1886 warn[cnt].w_type = QUOTA_NL_NOWARN;
1889 spin_lock(&dquots[cnt]->dq_dqb_lock);
1890 wtype = info_idq_free(dquots[cnt], 1);
1891 if (wtype != QUOTA_NL_NOWARN)
1892 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1893 dquot_decr_inodes(dquots[cnt], 1);
1894 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1896 spin_unlock(&inode->i_lock);
1897 mark_all_dquot_dirty(dquots);
1898 srcu_read_unlock(&dquot_srcu, index);
1899 flush_warnings(warn);
1901 EXPORT_SYMBOL(dquot_free_inode);
1904 * Transfer the number of inode and blocks from one diskquota to an other.
1905 * On success, dquot references in transfer_to are consumed and references
1906 * to original dquots that need to be released are placed there. On failure,
1907 * references are kept untouched.
1909 * This operation can block, but only after everything is updated
1910 * A transaction must be started when entering this function.
1912 * We are holding reference on transfer_from & transfer_to, no need to
1913 * protect them by srcu_read_lock().
1915 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1918 qsize_t rsv_space = 0;
1919 qsize_t inode_usage = 1;
1920 struct dquot *transfer_from[MAXQUOTAS] = {};
1922 char is_valid[MAXQUOTAS] = {};
1923 struct dquot_warn warn_to[MAXQUOTAS];
1924 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1925 struct dquot_warn warn_from_space[MAXQUOTAS];
1927 if (IS_NOQUOTA(inode))
1930 if (inode->i_sb->dq_op->get_inode_usage) {
1931 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1936 /* Initialize the arrays */
1937 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1938 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1939 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1940 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1943 spin_lock(&dq_data_lock);
1944 spin_lock(&inode->i_lock);
1945 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1946 spin_unlock(&inode->i_lock);
1947 spin_unlock(&dq_data_lock);
1950 cur_space = __inode_get_bytes(inode);
1951 rsv_space = __inode_get_rsv_space(inode);
1953 * Build the transfer_from list, check limits, and update usage in
1954 * the target structures.
1956 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1958 * Skip changes for same uid or gid or for turned off quota-type.
1960 if (!transfer_to[cnt])
1962 /* Avoid races with quotaoff() */
1963 if (!sb_has_quota_active(inode->i_sb, cnt))
1966 transfer_from[cnt] = i_dquot(inode)[cnt];
1967 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1971 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space, 0,
1974 dquot_decr_inodes(transfer_to[cnt], inode_usage);
1979 /* Decrease usage for source structures and update quota pointers */
1980 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1983 /* Due to IO error we might not have transfer_from[] structure */
1984 if (transfer_from[cnt]) {
1987 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
1988 wtype = info_idq_free(transfer_from[cnt], inode_usage);
1989 if (wtype != QUOTA_NL_NOWARN)
1990 prepare_warning(&warn_from_inodes[cnt],
1991 transfer_from[cnt], wtype);
1992 wtype = info_bdq_free(transfer_from[cnt],
1993 cur_space + rsv_space);
1994 if (wtype != QUOTA_NL_NOWARN)
1995 prepare_warning(&warn_from_space[cnt],
1996 transfer_from[cnt], wtype);
1997 dquot_decr_inodes(transfer_from[cnt], inode_usage);
1998 dquot_decr_space(transfer_from[cnt], cur_space);
1999 dquot_free_reserved_space(transfer_from[cnt],
2001 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2003 i_dquot(inode)[cnt] = transfer_to[cnt];
2005 spin_unlock(&inode->i_lock);
2006 spin_unlock(&dq_data_lock);
2008 mark_all_dquot_dirty(transfer_from);
2009 mark_all_dquot_dirty(transfer_to);
2010 flush_warnings(warn_to);
2011 flush_warnings(warn_from_inodes);
2012 flush_warnings(warn_from_space);
2013 /* Pass back references to put */
2014 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2016 transfer_to[cnt] = transfer_from[cnt];
2019 /* Back out changes we already did */
2020 for (cnt--; cnt >= 0; cnt--) {
2023 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2024 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2025 dquot_decr_space(transfer_to[cnt], cur_space);
2026 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2027 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2029 spin_unlock(&inode->i_lock);
2030 spin_unlock(&dq_data_lock);
2031 flush_warnings(warn_to);
2034 EXPORT_SYMBOL(__dquot_transfer);
2036 /* Wrapper for transferring ownership of an inode for uid/gid only
2037 * Called from FSXXX_setattr()
2039 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2041 struct dquot *transfer_to[MAXQUOTAS] = {};
2042 struct dquot *dquot;
2043 struct super_block *sb = inode->i_sb;
2046 if (!dquot_active(inode))
2049 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2050 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2051 if (IS_ERR(dquot)) {
2052 if (PTR_ERR(dquot) != -ESRCH) {
2053 ret = PTR_ERR(dquot);
2058 transfer_to[USRQUOTA] = dquot;
2060 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2061 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2062 if (IS_ERR(dquot)) {
2063 if (PTR_ERR(dquot) != -ESRCH) {
2064 ret = PTR_ERR(dquot);
2069 transfer_to[GRPQUOTA] = dquot;
2071 ret = __dquot_transfer(inode, transfer_to);
2073 dqput_all(transfer_to);
2076 EXPORT_SYMBOL(dquot_transfer);
2079 * Write info of quota file to disk
2081 int dquot_commit_info(struct super_block *sb, int type)
2083 struct quota_info *dqopt = sb_dqopt(sb);
2085 return dqopt->ops[type]->write_file_info(sb, type);
2087 EXPORT_SYMBOL(dquot_commit_info);
2089 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2091 struct quota_info *dqopt = sb_dqopt(sb);
2093 if (!sb_has_quota_active(sb, qid->type))
2095 if (!dqopt->ops[qid->type]->get_next_id)
2097 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2099 EXPORT_SYMBOL(dquot_get_next_id);
2102 * Definitions of diskquota operations.
2104 const struct dquot_operations dquot_operations = {
2105 .write_dquot = dquot_commit,
2106 .acquire_dquot = dquot_acquire,
2107 .release_dquot = dquot_release,
2108 .mark_dirty = dquot_mark_dquot_dirty,
2109 .write_info = dquot_commit_info,
2110 .alloc_dquot = dquot_alloc,
2111 .destroy_dquot = dquot_destroy,
2112 .get_next_id = dquot_get_next_id,
2114 EXPORT_SYMBOL(dquot_operations);
2117 * Generic helper for ->open on filesystems supporting disk quotas.
2119 int dquot_file_open(struct inode *inode, struct file *file)
2123 error = generic_file_open(inode, file);
2124 if (!error && (file->f_mode & FMODE_WRITE))
2125 dquot_initialize(inode);
2128 EXPORT_SYMBOL(dquot_file_open);
2131 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2133 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2136 struct quota_info *dqopt = sb_dqopt(sb);
2137 struct inode *toputinode[MAXQUOTAS];
2139 /* s_umount should be held in exclusive mode */
2140 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2141 up_read(&sb->s_umount);
2143 /* Cannot turn off usage accounting without turning off limits, or
2144 * suspend quotas and simultaneously turn quotas off. */
2145 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2146 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2147 DQUOT_USAGE_ENABLED)))
2151 * Skip everything if there's nothing to do. We have to do this because
2152 * sometimes we are called when fill_super() failed and calling
2153 * sync_fs() in such cases does no good.
2155 if (!sb_any_quota_loaded(sb))
2158 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2159 toputinode[cnt] = NULL;
2160 if (type != -1 && cnt != type)
2162 if (!sb_has_quota_loaded(sb, cnt))
2165 if (flags & DQUOT_SUSPENDED) {
2166 spin_lock(&dq_state_lock);
2168 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2169 spin_unlock(&dq_state_lock);
2171 spin_lock(&dq_state_lock);
2172 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2173 /* Turning off suspended quotas? */
2174 if (!sb_has_quota_loaded(sb, cnt) &&
2175 sb_has_quota_suspended(sb, cnt)) {
2176 dqopt->flags &= ~dquot_state_flag(
2177 DQUOT_SUSPENDED, cnt);
2178 spin_unlock(&dq_state_lock);
2179 iput(dqopt->files[cnt]);
2180 dqopt->files[cnt] = NULL;
2183 spin_unlock(&dq_state_lock);
2186 /* We still have to keep quota loaded? */
2187 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2190 /* Note: these are blocking operations */
2191 drop_dquot_ref(sb, cnt);
2192 invalidate_dquots(sb, cnt);
2194 * Now all dquots should be invalidated, all writes done so we
2195 * should be only users of the info. No locks needed.
2197 if (info_dirty(&dqopt->info[cnt]))
2198 sb->dq_op->write_info(sb, cnt);
2199 if (dqopt->ops[cnt]->free_file_info)
2200 dqopt->ops[cnt]->free_file_info(sb, cnt);
2201 put_quota_format(dqopt->info[cnt].dqi_format);
2203 toputinode[cnt] = dqopt->files[cnt];
2204 if (!sb_has_quota_loaded(sb, cnt))
2205 dqopt->files[cnt] = NULL;
2206 dqopt->info[cnt].dqi_flags = 0;
2207 dqopt->info[cnt].dqi_igrace = 0;
2208 dqopt->info[cnt].dqi_bgrace = 0;
2209 dqopt->ops[cnt] = NULL;
2212 /* Skip syncing and setting flags if quota files are hidden */
2213 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2216 /* Sync the superblock so that buffers with quota data are written to
2217 * disk (and so userspace sees correct data afterwards). */
2218 if (sb->s_op->sync_fs)
2219 sb->s_op->sync_fs(sb, 1);
2220 sync_blockdev(sb->s_bdev);
2221 /* Now the quota files are just ordinary files and we can set the
2222 * inode flags back. Moreover we discard the pagecache so that
2223 * userspace sees the writes we did bypassing the pagecache. We
2224 * must also discard the blockdev buffers so that we see the
2225 * changes done by userspace on the next quotaon() */
2226 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2227 /* This can happen when suspending quotas on remount-ro... */
2228 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2229 inode_lock(toputinode[cnt]);
2230 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2231 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2232 inode_unlock(toputinode[cnt]);
2233 mark_inode_dirty_sync(toputinode[cnt]);
2236 invalidate_bdev(sb->s_bdev);
2238 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2239 if (toputinode[cnt]) {
2240 /* On remount RO, we keep the inode pointer so that we
2241 * can reenable quota on the subsequent remount RW. We
2242 * have to check 'flags' variable and not use sb_has_
2243 * function because another quotaon / quotaoff could
2244 * change global state before we got here. We refuse
2245 * to suspend quotas when there is pending delete on
2246 * the quota file... */
2247 if (!(flags & DQUOT_SUSPENDED))
2248 iput(toputinode[cnt]);
2249 else if (!toputinode[cnt]->i_nlink)
2254 EXPORT_SYMBOL(dquot_disable);
2256 int dquot_quota_off(struct super_block *sb, int type)
2258 return dquot_disable(sb, type,
2259 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2261 EXPORT_SYMBOL(dquot_quota_off);
2264 * Turn quotas on on a device
2268 * Helper function to turn quotas on when we already have the inode of
2269 * quota file and no quota information is loaded.
2271 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2274 struct quota_format_type *fmt = find_quota_format(format_id);
2275 struct super_block *sb = inode->i_sb;
2276 struct quota_info *dqopt = sb_dqopt(sb);
2281 if (!S_ISREG(inode->i_mode)) {
2285 if (IS_RDONLY(inode)) {
2289 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2290 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2294 /* Filesystems outside of init_user_ns not yet supported */
2295 if (sb->s_user_ns != &init_user_ns) {
2299 /* Usage always has to be set... */
2300 if (!(flags & DQUOT_USAGE_ENABLED)) {
2304 if (sb_has_quota_loaded(sb, type)) {
2309 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2310 /* As we bypass the pagecache we must now flush all the
2311 * dirty data and invalidate caches so that kernel sees
2312 * changes from userspace. It is not enough to just flush
2313 * the quota file since if blocksize < pagesize, invalidation
2314 * of the cache could fail because of other unrelated dirty
2316 sync_filesystem(sb);
2317 invalidate_bdev(sb->s_bdev);
2320 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2321 /* We don't want quota and atime on quota files (deadlocks
2322 * possible) Also nobody should write to the file - we use
2323 * special IO operations which ignore the immutable bit. */
2325 inode->i_flags |= S_NOQUOTA;
2326 inode_unlock(inode);
2328 * When S_NOQUOTA is set, remove dquot references as no more
2329 * references can be added
2331 __dquot_drop(inode);
2335 dqopt->files[type] = igrab(inode);
2336 if (!dqopt->files[type])
2337 goto out_file_flags;
2339 if (!fmt->qf_ops->check_quota_file(sb, type))
2342 dqopt->ops[type] = fmt->qf_ops;
2343 dqopt->info[type].dqi_format = fmt;
2344 dqopt->info[type].dqi_fmt_id = format_id;
2345 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2346 error = dqopt->ops[type]->read_file_info(sb, type);
2349 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2350 spin_lock(&dq_data_lock);
2351 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2352 spin_unlock(&dq_data_lock);
2354 spin_lock(&dq_state_lock);
2355 dqopt->flags |= dquot_state_flag(flags, type);
2356 spin_unlock(&dq_state_lock);
2358 add_dquot_ref(sb, type);
2363 dqopt->files[type] = NULL;
2367 inode->i_flags &= ~S_NOQUOTA;
2368 inode_unlock(inode);
2370 put_quota_format(fmt);
2375 /* Reenable quotas on remount RW */
2376 int dquot_resume(struct super_block *sb, int type)
2378 struct quota_info *dqopt = sb_dqopt(sb);
2379 struct inode *inode;
2383 /* s_umount should be held in exclusive mode */
2384 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2385 up_read(&sb->s_umount);
2387 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2388 if (type != -1 && cnt != type)
2390 if (!sb_has_quota_suspended(sb, cnt))
2393 inode = dqopt->files[cnt];
2394 dqopt->files[cnt] = NULL;
2395 spin_lock(&dq_state_lock);
2396 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2397 DQUOT_LIMITS_ENABLED,
2399 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2400 spin_unlock(&dq_state_lock);
2402 flags = dquot_generic_flag(flags, cnt);
2403 ret = vfs_load_quota_inode(inode, cnt,
2404 dqopt->info[cnt].dqi_fmt_id, flags);
2410 EXPORT_SYMBOL(dquot_resume);
2412 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2413 const struct path *path)
2415 int error = security_quota_on(path->dentry);
2418 /* Quota file not on the same filesystem? */
2419 if (path->dentry->d_sb != sb)
2422 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2423 format_id, DQUOT_USAGE_ENABLED |
2424 DQUOT_LIMITS_ENABLED);
2427 EXPORT_SYMBOL(dquot_quota_on);
2430 * More powerful function for turning on quotas allowing setting
2431 * of individual quota flags
2433 int dquot_enable(struct inode *inode, int type, int format_id,
2436 struct super_block *sb = inode->i_sb;
2438 /* Just unsuspend quotas? */
2439 BUG_ON(flags & DQUOT_SUSPENDED);
2440 /* s_umount should be held in exclusive mode */
2441 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2442 up_read(&sb->s_umount);
2446 /* Just updating flags needed? */
2447 if (sb_has_quota_loaded(sb, type)) {
2448 if (flags & DQUOT_USAGE_ENABLED &&
2449 sb_has_quota_usage_enabled(sb, type))
2451 if (flags & DQUOT_LIMITS_ENABLED &&
2452 sb_has_quota_limits_enabled(sb, type))
2454 spin_lock(&dq_state_lock);
2455 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2456 spin_unlock(&dq_state_lock);
2460 return vfs_load_quota_inode(inode, type, format_id, flags);
2462 EXPORT_SYMBOL(dquot_enable);
2465 * This function is used when filesystem needs to initialize quotas
2466 * during mount time.
2468 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2469 int format_id, int type)
2471 struct dentry *dentry;
2474 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2476 return PTR_ERR(dentry);
2478 if (d_really_is_negative(dentry)) {
2483 error = security_quota_on(dentry);
2485 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2486 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2492 EXPORT_SYMBOL(dquot_quota_on_mount);
2494 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2498 struct quota_info *dqopt = sb_dqopt(sb);
2500 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2502 /* Accounting cannot be turned on while fs is mounted */
2503 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2506 for (type = 0; type < MAXQUOTAS; type++) {
2507 if (!(flags & qtype_enforce_flag(type)))
2509 /* Can't enforce without accounting */
2510 if (!sb_has_quota_usage_enabled(sb, type))
2512 ret = dquot_enable(dqopt->files[type], type,
2513 dqopt->info[type].dqi_fmt_id,
2514 DQUOT_LIMITS_ENABLED);
2520 /* Backout enforcement enablement we already did */
2521 for (type--; type >= 0; type--) {
2522 if (flags & qtype_enforce_flag(type))
2523 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2525 /* Error code translation for better compatibility with XFS */
2531 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2535 struct quota_info *dqopt = sb_dqopt(sb);
2537 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2540 * We don't support turning off accounting via quotactl. In principle
2541 * quota infrastructure can do this but filesystems don't expect
2542 * userspace to be able to do it.
2545 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2548 /* Filter out limits not enabled */
2549 for (type = 0; type < MAXQUOTAS; type++)
2550 if (!sb_has_quota_limits_enabled(sb, type))
2551 flags &= ~qtype_enforce_flag(type);
2555 for (type = 0; type < MAXQUOTAS; type++) {
2556 if (flags & qtype_enforce_flag(type)) {
2557 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2564 /* Backout enforcement disabling we already did */
2565 for (type--; type >= 0; type--) {
2566 if (flags & qtype_enforce_flag(type))
2567 dquot_enable(dqopt->files[type], type,
2568 dqopt->info[type].dqi_fmt_id,
2569 DQUOT_LIMITS_ENABLED);
2574 /* Generic routine for getting common part of quota structure */
2575 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2577 struct mem_dqblk *dm = &dquot->dq_dqb;
2579 memset(di, 0, sizeof(*di));
2580 spin_lock(&dquot->dq_dqb_lock);
2581 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2582 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2583 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2584 di->d_ino_softlimit = dm->dqb_isoftlimit;
2585 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2586 di->d_ino_count = dm->dqb_curinodes;
2587 di->d_spc_timer = dm->dqb_btime;
2588 di->d_ino_timer = dm->dqb_itime;
2589 spin_unlock(&dquot->dq_dqb_lock);
2592 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2593 struct qc_dqblk *di)
2595 struct dquot *dquot;
2597 dquot = dqget(sb, qid);
2599 return PTR_ERR(dquot);
2600 do_get_dqblk(dquot, di);
2605 EXPORT_SYMBOL(dquot_get_dqblk);
2607 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2608 struct qc_dqblk *di)
2610 struct dquot *dquot;
2613 if (!sb->dq_op->get_next_id)
2615 err = sb->dq_op->get_next_id(sb, qid);
2618 dquot = dqget(sb, *qid);
2620 return PTR_ERR(dquot);
2621 do_get_dqblk(dquot, di);
2626 EXPORT_SYMBOL(dquot_get_next_dqblk);
2628 #define VFS_QC_MASK \
2629 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2630 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2631 QC_SPC_TIMER | QC_INO_TIMER)
2633 /* Generic routine for setting common part of quota structure */
2634 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2636 struct mem_dqblk *dm = &dquot->dq_dqb;
2637 int check_blim = 0, check_ilim = 0;
2638 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2640 if (di->d_fieldmask & ~VFS_QC_MASK)
2643 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2644 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2645 ((di->d_fieldmask & QC_SPC_HARD) &&
2646 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2647 ((di->d_fieldmask & QC_INO_SOFT) &&
2648 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2649 ((di->d_fieldmask & QC_INO_HARD) &&
2650 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2653 spin_lock(&dquot->dq_dqb_lock);
2654 if (di->d_fieldmask & QC_SPACE) {
2655 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2657 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2660 if (di->d_fieldmask & QC_SPC_SOFT)
2661 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2662 if (di->d_fieldmask & QC_SPC_HARD)
2663 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2664 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2666 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2669 if (di->d_fieldmask & QC_INO_COUNT) {
2670 dm->dqb_curinodes = di->d_ino_count;
2672 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2675 if (di->d_fieldmask & QC_INO_SOFT)
2676 dm->dqb_isoftlimit = di->d_ino_softlimit;
2677 if (di->d_fieldmask & QC_INO_HARD)
2678 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2679 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2681 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2684 if (di->d_fieldmask & QC_SPC_TIMER) {
2685 dm->dqb_btime = di->d_spc_timer;
2687 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2690 if (di->d_fieldmask & QC_INO_TIMER) {
2691 dm->dqb_itime = di->d_ino_timer;
2693 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2697 if (!dm->dqb_bsoftlimit ||
2698 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2700 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2701 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2702 /* Set grace only if user hasn't provided his own... */
2703 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2706 if (!dm->dqb_isoftlimit ||
2707 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2709 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2710 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2711 /* Set grace only if user hasn't provided his own... */
2712 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2714 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2716 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2718 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2719 spin_unlock(&dquot->dq_dqb_lock);
2720 mark_dquot_dirty(dquot);
2725 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2726 struct qc_dqblk *di)
2728 struct dquot *dquot;
2731 dquot = dqget(sb, qid);
2732 if (IS_ERR(dquot)) {
2733 rc = PTR_ERR(dquot);
2736 rc = do_set_dqblk(dquot, di);
2741 EXPORT_SYMBOL(dquot_set_dqblk);
2743 /* Generic routine for getting common part of quota file information */
2744 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2746 struct mem_dqinfo *mi;
2747 struct qc_type_state *tstate;
2748 struct quota_info *dqopt = sb_dqopt(sb);
2751 memset(state, 0, sizeof(*state));
2752 for (type = 0; type < MAXQUOTAS; type++) {
2753 if (!sb_has_quota_active(sb, type))
2755 tstate = state->s_state + type;
2756 mi = sb_dqopt(sb)->info + type;
2757 tstate->flags = QCI_ACCT_ENABLED;
2758 spin_lock(&dq_data_lock);
2759 if (mi->dqi_flags & DQF_SYS_FILE)
2760 tstate->flags |= QCI_SYSFILE;
2761 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2762 tstate->flags |= QCI_ROOT_SQUASH;
2763 if (sb_has_quota_limits_enabled(sb, type))
2764 tstate->flags |= QCI_LIMITS_ENFORCED;
2765 tstate->spc_timelimit = mi->dqi_bgrace;
2766 tstate->ino_timelimit = mi->dqi_igrace;
2767 tstate->ino = dqopt->files[type]->i_ino;
2768 tstate->blocks = dqopt->files[type]->i_blocks;
2769 tstate->nextents = 1; /* We don't know... */
2770 spin_unlock(&dq_data_lock);
2774 EXPORT_SYMBOL(dquot_get_state);
2776 /* Generic routine for setting common part of quota file information */
2777 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2779 struct mem_dqinfo *mi;
2782 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2783 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2785 if (!sb_has_quota_active(sb, type))
2787 mi = sb_dqopt(sb)->info + type;
2788 if (ii->i_fieldmask & QC_FLAGS) {
2789 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2790 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2793 spin_lock(&dq_data_lock);
2794 if (ii->i_fieldmask & QC_SPC_TIMER)
2795 mi->dqi_bgrace = ii->i_spc_timelimit;
2796 if (ii->i_fieldmask & QC_INO_TIMER)
2797 mi->dqi_igrace = ii->i_ino_timelimit;
2798 if (ii->i_fieldmask & QC_FLAGS) {
2799 if (ii->i_flags & QCI_ROOT_SQUASH)
2800 mi->dqi_flags |= DQF_ROOT_SQUASH;
2802 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2804 spin_unlock(&dq_data_lock);
2805 mark_info_dirty(sb, type);
2806 /* Force write to disk */
2807 sb->dq_op->write_info(sb, type);
2810 EXPORT_SYMBOL(dquot_set_dqinfo);
2812 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2813 .quota_enable = dquot_quota_enable,
2814 .quota_disable = dquot_quota_disable,
2815 .quota_sync = dquot_quota_sync,
2816 .get_state = dquot_get_state,
2817 .set_info = dquot_set_dqinfo,
2818 .get_dqblk = dquot_get_dqblk,
2819 .get_nextdqblk = dquot_get_next_dqblk,
2820 .set_dqblk = dquot_set_dqblk
2822 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2824 static int do_proc_dqstats(struct ctl_table *table, int write,
2825 void __user *buffer, size_t *lenp, loff_t *ppos)
2827 unsigned int type = (int *)table->data - dqstats.stat;
2829 /* Update global table */
2830 dqstats.stat[type] =
2831 percpu_counter_sum_positive(&dqstats.counter[type]);
2832 return proc_dointvec(table, write, buffer, lenp, ppos);
2835 static struct ctl_table fs_dqstats_table[] = {
2837 .procname = "lookups",
2838 .data = &dqstats.stat[DQST_LOOKUPS],
2839 .maxlen = sizeof(int),
2841 .proc_handler = do_proc_dqstats,
2844 .procname = "drops",
2845 .data = &dqstats.stat[DQST_DROPS],
2846 .maxlen = sizeof(int),
2848 .proc_handler = do_proc_dqstats,
2851 .procname = "reads",
2852 .data = &dqstats.stat[DQST_READS],
2853 .maxlen = sizeof(int),
2855 .proc_handler = do_proc_dqstats,
2858 .procname = "writes",
2859 .data = &dqstats.stat[DQST_WRITES],
2860 .maxlen = sizeof(int),
2862 .proc_handler = do_proc_dqstats,
2865 .procname = "cache_hits",
2866 .data = &dqstats.stat[DQST_CACHE_HITS],
2867 .maxlen = sizeof(int),
2869 .proc_handler = do_proc_dqstats,
2872 .procname = "allocated_dquots",
2873 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2874 .maxlen = sizeof(int),
2876 .proc_handler = do_proc_dqstats,
2879 .procname = "free_dquots",
2880 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2881 .maxlen = sizeof(int),
2883 .proc_handler = do_proc_dqstats,
2886 .procname = "syncs",
2887 .data = &dqstats.stat[DQST_SYNCS],
2888 .maxlen = sizeof(int),
2890 .proc_handler = do_proc_dqstats,
2892 #ifdef CONFIG_PRINT_QUOTA_WARNING
2894 .procname = "warnings",
2895 .data = &flag_print_warnings,
2896 .maxlen = sizeof(int),
2898 .proc_handler = proc_dointvec,
2904 static struct ctl_table fs_table[] = {
2906 .procname = "quota",
2908 .child = fs_dqstats_table,
2913 static struct ctl_table sys_table[] = {
2922 static int __init dquot_init(void)
2925 unsigned long nr_hash, order;
2927 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2929 register_sysctl_table(sys_table);
2931 dquot_cachep = kmem_cache_create("dquot",
2932 sizeof(struct dquot), sizeof(unsigned long) * 4,
2933 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2934 SLAB_MEM_SPREAD|SLAB_PANIC),
2938 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2940 panic("Cannot create dquot hash table");
2942 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2943 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2945 panic("Cannot create dquot stat counters");
2948 /* Find power-of-two hlist_heads which can fit into allocation */
2949 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2953 } while (nr_hash >> dq_hash_bits);
2956 nr_hash = 1UL << dq_hash_bits;
2957 dq_hash_mask = nr_hash - 1;
2958 for (i = 0; i < nr_hash; i++)
2959 INIT_HLIST_HEAD(dquot_hash + i);
2961 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2962 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2964 register_shrinker(&dqcache_shrinker);
2968 fs_initcall(dquot_init);