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 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
131 void __quota_error(struct super_block *sb, const char *func,
132 const char *fmt, ...)
134 if (printk_ratelimit()) {
136 struct va_format vaf;
143 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
144 sb->s_id, func, &vaf);
149 EXPORT_SYMBOL(__quota_error);
151 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
152 static char *quotatypes[] = INITQFNAMES;
154 static struct quota_format_type *quota_formats; /* List of registered formats */
155 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
157 /* SLAB cache for dquot structures */
158 static struct kmem_cache *dquot_cachep;
160 int register_quota_format(struct quota_format_type *fmt)
162 spin_lock(&dq_list_lock);
163 fmt->qf_next = quota_formats;
165 spin_unlock(&dq_list_lock);
168 EXPORT_SYMBOL(register_quota_format);
170 void unregister_quota_format(struct quota_format_type *fmt)
172 struct quota_format_type **actqf;
174 spin_lock(&dq_list_lock);
175 for (actqf = "a_formats; *actqf && *actqf != fmt;
176 actqf = &(*actqf)->qf_next)
179 *actqf = (*actqf)->qf_next;
180 spin_unlock(&dq_list_lock);
182 EXPORT_SYMBOL(unregister_quota_format);
184 static struct quota_format_type *find_quota_format(int id)
186 struct quota_format_type *actqf;
188 spin_lock(&dq_list_lock);
189 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
190 actqf = actqf->qf_next)
192 if (!actqf || !try_module_get(actqf->qf_owner)) {
195 spin_unlock(&dq_list_lock);
197 for (qm = 0; module_names[qm].qm_fmt_id &&
198 module_names[qm].qm_fmt_id != id; qm++)
200 if (!module_names[qm].qm_fmt_id ||
201 request_module(module_names[qm].qm_mod_name))
204 spin_lock(&dq_list_lock);
205 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
206 actqf = actqf->qf_next)
208 if (actqf && !try_module_get(actqf->qf_owner))
211 spin_unlock(&dq_list_lock);
215 static void put_quota_format(struct quota_format_type *fmt)
217 module_put(fmt->qf_owner);
221 * Dquot List Management:
222 * The quota code uses three lists for dquot management: the inuse_list,
223 * free_dquots, and dquot_hash[] array. A single dquot structure may be
224 * on all three lists, depending on its current state.
226 * All dquots are placed to the end of inuse_list when first created, and this
227 * list is used for invalidate operation, which must look at every dquot.
229 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
230 * and this list is searched whenever we need an available dquot. Dquots are
231 * removed from the list as soon as they are used again, and
232 * dqstats.free_dquots gives the number of dquots on the list. When
233 * dquot is invalidated it's completely released from memory.
235 * Dquots with a specific identity (device, type and id) are placed on
236 * one of the dquot_hash[] hash chains. The provides an efficient search
237 * mechanism to locate a specific dquot.
240 static LIST_HEAD(inuse_list);
241 static LIST_HEAD(free_dquots);
242 static unsigned int dq_hash_bits, dq_hash_mask;
243 static struct hlist_head *dquot_hash;
245 struct dqstats dqstats;
246 EXPORT_SYMBOL(dqstats);
248 static qsize_t inode_get_rsv_space(struct inode *inode);
249 static int __dquot_initialize(struct inode *inode, int type);
251 static inline unsigned int
252 hashfn(const struct super_block *sb, struct kqid qid)
254 unsigned int id = from_kqid(&init_user_ns, qid);
258 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
259 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
263 * Following list functions expect dq_list_lock to be held
265 static inline void insert_dquot_hash(struct dquot *dquot)
267 struct hlist_head *head;
268 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
269 hlist_add_head(&dquot->dq_hash, head);
272 static inline void remove_dquot_hash(struct dquot *dquot)
274 hlist_del_init(&dquot->dq_hash);
277 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
280 struct hlist_node *node;
283 hlist_for_each (node, dquot_hash+hashent) {
284 dquot = hlist_entry(node, struct dquot, dq_hash);
285 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
291 /* Add a dquot to the tail of the free list */
292 static inline void put_dquot_last(struct dquot *dquot)
294 list_add_tail(&dquot->dq_free, &free_dquots);
295 dqstats_inc(DQST_FREE_DQUOTS);
298 static inline void remove_free_dquot(struct dquot *dquot)
300 if (list_empty(&dquot->dq_free))
302 list_del_init(&dquot->dq_free);
303 dqstats_dec(DQST_FREE_DQUOTS);
306 static inline void put_inuse(struct dquot *dquot)
308 /* We add to the back of inuse list so we don't have to restart
309 * when traversing this list and we block */
310 list_add_tail(&dquot->dq_inuse, &inuse_list);
311 dqstats_inc(DQST_ALLOC_DQUOTS);
314 static inline void remove_inuse(struct dquot *dquot)
316 dqstats_dec(DQST_ALLOC_DQUOTS);
317 list_del(&dquot->dq_inuse);
320 * End of list functions needing dq_list_lock
323 static void wait_on_dquot(struct dquot *dquot)
325 mutex_lock(&dquot->dq_lock);
326 mutex_unlock(&dquot->dq_lock);
329 static inline int dquot_dirty(struct dquot *dquot)
331 return test_bit(DQ_MOD_B, &dquot->dq_flags);
334 static inline int mark_dquot_dirty(struct dquot *dquot)
336 return dquot->dq_sb->dq_op->mark_dirty(dquot);
339 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
340 int dquot_mark_dquot_dirty(struct dquot *dquot)
344 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
347 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
348 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
350 /* If quota is dirty already, we don't have to acquire dq_list_lock */
351 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
354 spin_lock(&dq_list_lock);
355 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
356 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
357 info[dquot->dq_id.type].dqi_dirty_list);
360 spin_unlock(&dq_list_lock);
363 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
365 /* Dirtify all the dquots - this can block when journalling */
366 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
371 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
373 /* Even in case of error we have to continue */
374 ret = mark_dquot_dirty(dquot[cnt]);
381 static inline void dqput_all(struct dquot **dquot)
385 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
389 static inline int clear_dquot_dirty(struct dquot *dquot)
391 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
392 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
394 spin_lock(&dq_list_lock);
395 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
396 spin_unlock(&dq_list_lock);
399 list_del_init(&dquot->dq_dirty);
400 spin_unlock(&dq_list_lock);
404 void mark_info_dirty(struct super_block *sb, int type)
406 spin_lock(&dq_data_lock);
407 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
408 spin_unlock(&dq_data_lock);
410 EXPORT_SYMBOL(mark_info_dirty);
413 * Read dquot from disk and alloc space for it
416 int dquot_acquire(struct dquot *dquot)
418 int ret = 0, ret2 = 0;
419 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
421 mutex_lock(&dquot->dq_lock);
422 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
423 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
426 /* Make sure flags update is visible after dquot has been filled */
427 smp_mb__before_atomic();
428 set_bit(DQ_READ_B, &dquot->dq_flags);
429 /* Instantiate dquot if needed */
430 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
431 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
432 /* Write the info if needed */
433 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
434 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
435 dquot->dq_sb, dquot->dq_id.type);
445 * Make sure flags update is visible after on-disk struct has been
446 * allocated. Paired with smp_rmb() in dqget().
448 smp_mb__before_atomic();
449 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
451 mutex_unlock(&dquot->dq_lock);
454 EXPORT_SYMBOL(dquot_acquire);
457 * Write dquot to disk
459 int dquot_commit(struct dquot *dquot)
462 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
464 mutex_lock(&dquot->dq_lock);
465 if (!clear_dquot_dirty(dquot))
467 /* Inactive dquot can be only if there was error during read/init
468 * => we have better not writing it */
469 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
470 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
474 mutex_unlock(&dquot->dq_lock);
477 EXPORT_SYMBOL(dquot_commit);
482 int dquot_release(struct dquot *dquot)
484 int ret = 0, ret2 = 0;
485 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
487 mutex_lock(&dquot->dq_lock);
488 /* Check whether we are not racing with some other dqget() */
489 if (atomic_read(&dquot->dq_count) > 1)
491 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
492 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
494 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
495 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
496 dquot->dq_sb, dquot->dq_id.type);
501 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
503 mutex_unlock(&dquot->dq_lock);
506 EXPORT_SYMBOL(dquot_release);
508 void dquot_destroy(struct dquot *dquot)
510 kmem_cache_free(dquot_cachep, dquot);
512 EXPORT_SYMBOL(dquot_destroy);
514 static inline void do_destroy_dquot(struct dquot *dquot)
516 dquot->dq_sb->dq_op->destroy_dquot(dquot);
519 /* Invalidate all dquots on the list. Note that this function is called after
520 * quota is disabled and pointers from inodes removed so there cannot be new
521 * quota users. There can still be some users of quotas due to inodes being
522 * just deleted or pruned by prune_icache() (those are not attached to any
523 * list) or parallel quotactl call. We have to wait for such users.
525 static void invalidate_dquots(struct super_block *sb, int type)
527 struct dquot *dquot, *tmp;
530 spin_lock(&dq_list_lock);
531 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
532 if (dquot->dq_sb != sb)
534 if (dquot->dq_id.type != type)
536 /* Wait for dquot users */
537 if (atomic_read(&dquot->dq_count)) {
539 spin_unlock(&dq_list_lock);
541 * Once dqput() wakes us up, we know it's time to free
543 * IMPORTANT: we rely on the fact that there is always
544 * at most one process waiting for dquot to free.
545 * Otherwise dq_count would be > 1 and we would never
548 wait_event(dquot_ref_wq,
549 atomic_read(&dquot->dq_count) == 1);
551 /* At this moment dquot() need not exist (it could be
552 * reclaimed by prune_dqcache(). Hence we must
557 * Quota now has no users and it has been written on last
560 remove_dquot_hash(dquot);
561 remove_free_dquot(dquot);
563 do_destroy_dquot(dquot);
565 spin_unlock(&dq_list_lock);
568 /* Call callback for every active dquot on given filesystem */
569 int dquot_scan_active(struct super_block *sb,
570 int (*fn)(struct dquot *dquot, unsigned long priv),
573 struct dquot *dquot, *old_dquot = NULL;
576 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
578 spin_lock(&dq_list_lock);
579 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
580 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
582 if (dquot->dq_sb != sb)
584 /* Now we have active dquot so we can just increase use count */
585 atomic_inc(&dquot->dq_count);
586 spin_unlock(&dq_list_lock);
587 dqstats_inc(DQST_LOOKUPS);
591 * ->release_dquot() can be racing with us. Our reference
592 * protects us from new calls to it so just wait for any
593 * outstanding call and recheck the DQ_ACTIVE_B after that.
595 wait_on_dquot(dquot);
596 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
597 ret = fn(dquot, priv);
601 spin_lock(&dq_list_lock);
602 /* We are safe to continue now because our dquot could not
603 * be moved out of the inuse list while we hold the reference */
605 spin_unlock(&dq_list_lock);
610 EXPORT_SYMBOL(dquot_scan_active);
612 /* Write all dquot structures to quota files */
613 int dquot_writeback_dquots(struct super_block *sb, int type)
615 struct list_head *dirty;
617 struct quota_info *dqopt = sb_dqopt(sb);
621 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
623 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
624 if (type != -1 && cnt != type)
626 if (!sb_has_quota_active(sb, cnt))
628 spin_lock(&dq_list_lock);
629 dirty = &dqopt->info[cnt].dqi_dirty_list;
630 while (!list_empty(dirty)) {
631 dquot = list_first_entry(dirty, struct dquot,
634 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
636 /* Now we have active dquot from which someone is
637 * holding reference so we can safely just increase
640 spin_unlock(&dq_list_lock);
641 dqstats_inc(DQST_LOOKUPS);
642 err = sb->dq_op->write_dquot(dquot);
646 spin_lock(&dq_list_lock);
648 spin_unlock(&dq_list_lock);
651 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
652 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
653 && info_dirty(&dqopt->info[cnt]))
654 sb->dq_op->write_info(sb, cnt);
655 dqstats_inc(DQST_SYNCS);
659 EXPORT_SYMBOL(dquot_writeback_dquots);
661 /* Write all dquot structures to disk and make them visible from userspace */
662 int dquot_quota_sync(struct super_block *sb, int type)
664 struct quota_info *dqopt = sb_dqopt(sb);
668 ret = dquot_writeback_dquots(sb, type);
671 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
674 /* This is not very clever (and fast) but currently I don't know about
675 * any other simple way of getting quota data to disk and we must get
676 * them there for userspace to be visible... */
677 if (sb->s_op->sync_fs)
678 sb->s_op->sync_fs(sb, 1);
679 sync_blockdev(sb->s_bdev);
682 * Now when everything is written we can discard the pagecache so
683 * that userspace sees the changes.
685 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
686 if (type != -1 && cnt != type)
688 if (!sb_has_quota_active(sb, cnt))
690 inode_lock(dqopt->files[cnt]);
691 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
692 inode_unlock(dqopt->files[cnt]);
697 EXPORT_SYMBOL(dquot_quota_sync);
700 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
702 struct list_head *head;
704 unsigned long freed = 0;
706 spin_lock(&dq_list_lock);
707 head = free_dquots.prev;
708 while (head != &free_dquots && sc->nr_to_scan) {
709 dquot = list_entry(head, struct dquot, dq_free);
710 remove_dquot_hash(dquot);
711 remove_free_dquot(dquot);
713 do_destroy_dquot(dquot);
716 head = free_dquots.prev;
718 spin_unlock(&dq_list_lock);
723 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
725 return vfs_pressure_ratio(
726 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
729 static struct shrinker dqcache_shrinker = {
730 .count_objects = dqcache_shrink_count,
731 .scan_objects = dqcache_shrink_scan,
732 .seeks = DEFAULT_SEEKS,
736 * Put reference to dquot
738 void dqput(struct dquot *dquot)
744 #ifdef CONFIG_QUOTA_DEBUG
745 if (!atomic_read(&dquot->dq_count)) {
746 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
747 quotatypes[dquot->dq_id.type],
748 from_kqid(&init_user_ns, dquot->dq_id));
752 dqstats_inc(DQST_DROPS);
754 spin_lock(&dq_list_lock);
755 if (atomic_read(&dquot->dq_count) > 1) {
756 /* We have more than one user... nothing to do */
757 atomic_dec(&dquot->dq_count);
758 /* Releasing dquot during quotaoff phase? */
759 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
760 atomic_read(&dquot->dq_count) == 1)
761 wake_up(&dquot_ref_wq);
762 spin_unlock(&dq_list_lock);
765 /* Need to release dquot? */
766 if (dquot_dirty(dquot)) {
767 spin_unlock(&dq_list_lock);
768 /* Commit dquot before releasing */
769 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
771 quota_error(dquot->dq_sb, "Can't write quota structure"
772 " (error %d). Quota may get out of sync!",
775 * We clear dirty bit anyway, so that we avoid
778 clear_dquot_dirty(dquot);
782 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
783 spin_unlock(&dq_list_lock);
784 dquot->dq_sb->dq_op->release_dquot(dquot);
787 atomic_dec(&dquot->dq_count);
788 #ifdef CONFIG_QUOTA_DEBUG
790 BUG_ON(!list_empty(&dquot->dq_free));
792 put_dquot_last(dquot);
793 spin_unlock(&dq_list_lock);
795 EXPORT_SYMBOL(dqput);
797 struct dquot *dquot_alloc(struct super_block *sb, int type)
799 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
801 EXPORT_SYMBOL(dquot_alloc);
803 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
807 dquot = sb->dq_op->alloc_dquot(sb, type);
811 mutex_init(&dquot->dq_lock);
812 INIT_LIST_HEAD(&dquot->dq_free);
813 INIT_LIST_HEAD(&dquot->dq_inuse);
814 INIT_HLIST_NODE(&dquot->dq_hash);
815 INIT_LIST_HEAD(&dquot->dq_dirty);
817 dquot->dq_id = make_kqid_invalid(type);
818 atomic_set(&dquot->dq_count, 1);
824 * Get reference to dquot
826 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
827 * destroying our dquot by:
828 * a) checking for quota flags under dq_list_lock and
829 * b) getting a reference to dquot before we release dq_list_lock
831 struct dquot *dqget(struct super_block *sb, struct kqid qid)
833 unsigned int hashent = hashfn(sb, qid);
834 struct dquot *dquot, *empty = NULL;
836 if (!qid_has_mapping(sb->s_user_ns, qid))
837 return ERR_PTR(-EINVAL);
839 if (!sb_has_quota_active(sb, qid.type))
840 return ERR_PTR(-ESRCH);
842 spin_lock(&dq_list_lock);
843 spin_lock(&dq_state_lock);
844 if (!sb_has_quota_active(sb, qid.type)) {
845 spin_unlock(&dq_state_lock);
846 spin_unlock(&dq_list_lock);
847 dquot = ERR_PTR(-ESRCH);
850 spin_unlock(&dq_state_lock);
852 dquot = find_dquot(hashent, sb, qid);
855 spin_unlock(&dq_list_lock);
856 empty = get_empty_dquot(sb, qid.type);
858 schedule(); /* Try to wait for a moment... */
864 /* all dquots go on the inuse_list */
866 /* hash it first so it can be found */
867 insert_dquot_hash(dquot);
868 spin_unlock(&dq_list_lock);
869 dqstats_inc(DQST_LOOKUPS);
871 if (!atomic_read(&dquot->dq_count))
872 remove_free_dquot(dquot);
873 atomic_inc(&dquot->dq_count);
874 spin_unlock(&dq_list_lock);
875 dqstats_inc(DQST_CACHE_HITS);
876 dqstats_inc(DQST_LOOKUPS);
878 /* Wait for dq_lock - after this we know that either dquot_release() is
879 * already finished or it will be canceled due to dq_count > 1 test */
880 wait_on_dquot(dquot);
881 /* Read the dquot / allocate space in quota file */
882 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
885 err = sb->dq_op->acquire_dquot(dquot);
888 dquot = ERR_PTR(err);
893 * Make sure following reads see filled structure - paired with
894 * smp_mb__before_atomic() in dquot_acquire().
897 #ifdef CONFIG_QUOTA_DEBUG
898 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
902 do_destroy_dquot(empty);
906 EXPORT_SYMBOL(dqget);
908 static inline struct dquot **i_dquot(struct inode *inode)
910 return inode->i_sb->s_op->get_dquots(inode);
913 static int dqinit_needed(struct inode *inode, int type)
915 struct dquot * const *dquots;
918 if (IS_NOQUOTA(inode))
921 dquots = i_dquot(inode);
923 return !dquots[type];
924 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
930 /* This routine is guarded by s_umount semaphore */
931 static void add_dquot_ref(struct super_block *sb, int type)
933 struct inode *inode, *old_inode = NULL;
934 #ifdef CONFIG_QUOTA_DEBUG
938 spin_lock(&sb->s_inode_list_lock);
939 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
940 spin_lock(&inode->i_lock);
941 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
942 !atomic_read(&inode->i_writecount) ||
943 !dqinit_needed(inode, type)) {
944 spin_unlock(&inode->i_lock);
948 spin_unlock(&inode->i_lock);
949 spin_unlock(&sb->s_inode_list_lock);
951 #ifdef CONFIG_QUOTA_DEBUG
952 if (unlikely(inode_get_rsv_space(inode) > 0))
956 __dquot_initialize(inode, type);
959 * We hold a reference to 'inode' so it couldn't have been
960 * removed from s_inodes list while we dropped the
961 * s_inode_list_lock. We cannot iput the inode now as we can be
962 * holding the last reference and we cannot iput it under
963 * s_inode_list_lock. So we keep the reference and iput it
967 spin_lock(&sb->s_inode_list_lock);
969 spin_unlock(&sb->s_inode_list_lock);
972 #ifdef CONFIG_QUOTA_DEBUG
974 quota_error(sb, "Writes happened before quota was turned on "
975 "thus quota information is probably inconsistent. "
976 "Please run quotacheck(8)");
982 * Remove references to dquots from inode and add dquot to list for freeing
983 * if we have the last reference to dquot
985 static void remove_inode_dquot_ref(struct inode *inode, int type,
986 struct list_head *tofree_head)
988 struct dquot **dquots = i_dquot(inode);
989 struct dquot *dquot = dquots[type];
995 if (list_empty(&dquot->dq_free)) {
997 * The inode still has reference to dquot so it can't be in the
1000 spin_lock(&dq_list_lock);
1001 list_add(&dquot->dq_free, tofree_head);
1002 spin_unlock(&dq_list_lock);
1005 * Dquot is already in a list to put so we won't drop the last
1013 * Free list of dquots
1014 * Dquots are removed from inodes and no new references can be got so we are
1015 * the only ones holding reference
1017 static void put_dquot_list(struct list_head *tofree_head)
1019 struct list_head *act_head;
1020 struct dquot *dquot;
1022 act_head = tofree_head->next;
1023 while (act_head != tofree_head) {
1024 dquot = list_entry(act_head, struct dquot, dq_free);
1025 act_head = act_head->next;
1026 /* Remove dquot from the list so we won't have problems... */
1027 list_del_init(&dquot->dq_free);
1032 static void remove_dquot_ref(struct super_block *sb, int type,
1033 struct list_head *tofree_head)
1035 struct inode *inode;
1038 spin_lock(&sb->s_inode_list_lock);
1039 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1041 * We have to scan also I_NEW inodes because they can already
1042 * have quota pointer initialized. Luckily, we need to touch
1043 * only quota pointers and these have separate locking
1046 spin_lock(&dq_data_lock);
1047 if (!IS_NOQUOTA(inode)) {
1048 if (unlikely(inode_get_rsv_space(inode) > 0))
1050 remove_inode_dquot_ref(inode, type, tofree_head);
1052 spin_unlock(&dq_data_lock);
1054 spin_unlock(&sb->s_inode_list_lock);
1055 #ifdef CONFIG_QUOTA_DEBUG
1057 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1058 " was disabled thus quota information is probably "
1059 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1064 /* Gather all references from inodes and drop them */
1065 static void drop_dquot_ref(struct super_block *sb, int type)
1067 LIST_HEAD(tofree_head);
1070 remove_dquot_ref(sb, type, &tofree_head);
1071 synchronize_srcu(&dquot_srcu);
1072 put_dquot_list(&tofree_head);
1076 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1078 dquot->dq_dqb.dqb_curinodes += number;
1081 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1083 dquot->dq_dqb.dqb_curspace += number;
1086 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1088 dquot->dq_dqb.dqb_rsvspace += number;
1092 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1094 if (dquot->dq_dqb.dqb_rsvspace >= number)
1095 dquot->dq_dqb.dqb_rsvspace -= number;
1098 dquot->dq_dqb.dqb_rsvspace = 0;
1102 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1104 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1105 dquot->dq_dqb.dqb_curinodes >= number)
1106 dquot->dq_dqb.dqb_curinodes -= number;
1108 dquot->dq_dqb.dqb_curinodes = 0;
1109 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1110 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1111 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1114 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1116 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1117 dquot->dq_dqb.dqb_curspace >= number)
1118 dquot->dq_dqb.dqb_curspace -= number;
1120 dquot->dq_dqb.dqb_curspace = 0;
1121 if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1122 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1123 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1127 struct super_block *w_sb;
1128 struct kqid w_dq_id;
1132 static int warning_issued(struct dquot *dquot, const int warntype)
1134 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1135 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1136 ((warntype == QUOTA_NL_IHARDWARN ||
1137 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1141 return test_and_set_bit(flag, &dquot->dq_flags);
1144 #ifdef CONFIG_PRINT_QUOTA_WARNING
1145 static int flag_print_warnings = 1;
1147 static int need_print_warning(struct dquot_warn *warn)
1149 if (!flag_print_warnings)
1152 switch (warn->w_dq_id.type) {
1154 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1156 return in_group_p(warn->w_dq_id.gid);
1163 /* Print warning to user which exceeded quota */
1164 static void print_warning(struct dquot_warn *warn)
1167 struct tty_struct *tty;
1168 int warntype = warn->w_type;
1170 if (warntype == QUOTA_NL_IHARDBELOW ||
1171 warntype == QUOTA_NL_ISOFTBELOW ||
1172 warntype == QUOTA_NL_BHARDBELOW ||
1173 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1176 tty = get_current_tty();
1179 tty_write_message(tty, warn->w_sb->s_id);
1180 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1181 tty_write_message(tty, ": warning, ");
1183 tty_write_message(tty, ": write failed, ");
1184 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1186 case QUOTA_NL_IHARDWARN:
1187 msg = " file limit reached.\r\n";
1189 case QUOTA_NL_ISOFTLONGWARN:
1190 msg = " file quota exceeded too long.\r\n";
1192 case QUOTA_NL_ISOFTWARN:
1193 msg = " file quota exceeded.\r\n";
1195 case QUOTA_NL_BHARDWARN:
1196 msg = " block limit reached.\r\n";
1198 case QUOTA_NL_BSOFTLONGWARN:
1199 msg = " block quota exceeded too long.\r\n";
1201 case QUOTA_NL_BSOFTWARN:
1202 msg = " block quota exceeded.\r\n";
1205 tty_write_message(tty, msg);
1210 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1213 if (warning_issued(dquot, warntype))
1215 warn->w_type = warntype;
1216 warn->w_sb = dquot->dq_sb;
1217 warn->w_dq_id = dquot->dq_id;
1221 * Write warnings to the console and send warning messages over netlink.
1223 * Note that this function can call into tty and networking code.
1225 static void flush_warnings(struct dquot_warn *warn)
1229 for (i = 0; i < MAXQUOTAS; i++) {
1230 if (warn[i].w_type == QUOTA_NL_NOWARN)
1232 #ifdef CONFIG_PRINT_QUOTA_WARNING
1233 print_warning(&warn[i]);
1235 quota_send_warning(warn[i].w_dq_id,
1236 warn[i].w_sb->s_dev, warn[i].w_type);
1240 static int ignore_hardlimit(struct dquot *dquot)
1242 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1244 return capable(CAP_SYS_RESOURCE) &&
1245 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1246 !(info->dqi_flags & DQF_ROOT_SQUASH));
1249 /* needs dq_data_lock */
1250 static int check_idq(struct dquot *dquot, qsize_t inodes,
1251 struct dquot_warn *warn)
1253 qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1255 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1256 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1259 if (dquot->dq_dqb.dqb_ihardlimit &&
1260 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1261 !ignore_hardlimit(dquot)) {
1262 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1266 if (dquot->dq_dqb.dqb_isoftlimit &&
1267 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1268 dquot->dq_dqb.dqb_itime &&
1269 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1270 !ignore_hardlimit(dquot)) {
1271 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1275 if (dquot->dq_dqb.dqb_isoftlimit &&
1276 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1277 dquot->dq_dqb.dqb_itime == 0) {
1278 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1279 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1280 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1286 /* needs dq_data_lock */
1287 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1288 struct dquot_warn *warn)
1291 struct super_block *sb = dquot->dq_sb;
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
1300 if (dquot->dq_dqb.dqb_bhardlimit &&
1301 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1302 !ignore_hardlimit(dquot)) {
1304 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1308 if (dquot->dq_dqb.dqb_bsoftlimit &&
1309 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1310 dquot->dq_dqb.dqb_btime &&
1311 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1312 !ignore_hardlimit(dquot)) {
1314 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1318 if (dquot->dq_dqb.dqb_bsoftlimit &&
1319 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1320 dquot->dq_dqb.dqb_btime == 0) {
1322 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1323 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1324 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1328 * We don't allow preallocation to exceed softlimit so exceeding will
1337 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1341 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1342 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1343 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1344 return QUOTA_NL_NOWARN;
1346 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1347 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1348 return QUOTA_NL_ISOFTBELOW;
1349 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1350 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1351 return QUOTA_NL_IHARDBELOW;
1352 return QUOTA_NL_NOWARN;
1355 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1357 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1358 dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1359 return QUOTA_NL_NOWARN;
1361 if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1362 return QUOTA_NL_BSOFTBELOW;
1363 if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1364 dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1365 return QUOTA_NL_BHARDBELOW;
1366 return QUOTA_NL_NOWARN;
1369 static int dquot_active(const struct inode *inode)
1371 struct super_block *sb = inode->i_sb;
1373 if (IS_NOQUOTA(inode))
1375 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1379 * Initialize quota pointers in inode
1381 * It is better to call this function outside of any transaction as it
1382 * might need a lot of space in journal for dquot structure allocation.
1384 static int __dquot_initialize(struct inode *inode, int type)
1386 int cnt, init_needed = 0;
1387 struct dquot **dquots, *got[MAXQUOTAS] = {};
1388 struct super_block *sb = inode->i_sb;
1392 if (!dquot_active(inode))
1395 dquots = i_dquot(inode);
1397 /* First get references to structures we might need. */
1398 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1402 struct dquot *dquot;
1404 if (type != -1 && cnt != type)
1407 * The i_dquot should have been initialized in most cases,
1408 * we check it without locking here to avoid unnecessary
1409 * dqget()/dqput() calls.
1414 if (!sb_has_quota_active(sb, cnt))
1421 qid = make_kqid_uid(inode->i_uid);
1424 qid = make_kqid_gid(inode->i_gid);
1427 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1430 qid = make_kqid_projid(projid);
1433 dquot = dqget(sb, qid);
1434 if (IS_ERR(dquot)) {
1435 /* We raced with somebody turning quotas off... */
1436 if (PTR_ERR(dquot) != -ESRCH) {
1437 ret = PTR_ERR(dquot);
1445 /* All required i_dquot has been initialized */
1449 spin_lock(&dq_data_lock);
1450 if (IS_NOQUOTA(inode))
1452 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1453 if (type != -1 && cnt != type)
1455 /* Avoid races with quotaoff() */
1456 if (!sb_has_quota_active(sb, cnt))
1458 /* We could race with quotaon or dqget() could have failed */
1462 dquots[cnt] = got[cnt];
1465 * Make quota reservation system happy if someone
1466 * did a write before quota was turned on
1468 rsv = inode_get_rsv_space(inode);
1470 dquot_resv_space(dquots[cnt], rsv);
1474 spin_unlock(&dq_data_lock);
1476 /* Drop unused references */
1482 int dquot_initialize(struct inode *inode)
1484 return __dquot_initialize(inode, -1);
1486 EXPORT_SYMBOL(dquot_initialize);
1488 bool dquot_initialize_needed(struct inode *inode)
1490 struct dquot **dquots;
1493 if (!dquot_active(inode))
1496 dquots = i_dquot(inode);
1497 for (i = 0; i < MAXQUOTAS; i++)
1498 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1502 EXPORT_SYMBOL(dquot_initialize_needed);
1505 * Release all quotas referenced by inode.
1507 * This function only be called on inode free or converting
1508 * a file to quota file, no other users for the i_dquot in
1509 * both cases, so we needn't call synchronize_srcu() after
1512 static void __dquot_drop(struct inode *inode)
1515 struct dquot **dquots = i_dquot(inode);
1516 struct dquot *put[MAXQUOTAS];
1518 spin_lock(&dq_data_lock);
1519 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1520 put[cnt] = dquots[cnt];
1523 spin_unlock(&dq_data_lock);
1527 void dquot_drop(struct inode *inode)
1529 struct dquot * const *dquots;
1532 if (IS_NOQUOTA(inode))
1536 * Test before calling to rule out calls from proc and such
1537 * where we are not allowed to block. Note that this is
1538 * actually reliable test even without the lock - the caller
1539 * must assure that nobody can come after the DQUOT_DROP and
1540 * add quota pointers back anyway.
1542 dquots = i_dquot(inode);
1543 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1548 if (cnt < MAXQUOTAS)
1549 __dquot_drop(inode);
1551 EXPORT_SYMBOL(dquot_drop);
1554 * inode_reserved_space is managed internally by quota, and protected by
1555 * i_lock similar to i_blocks+i_bytes.
1557 static qsize_t *inode_reserved_space(struct inode * inode)
1559 /* Filesystem must explicitly define it's own method in order to use
1560 * quota reservation interface */
1561 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1562 return inode->i_sb->dq_op->get_reserved_space(inode);
1565 static qsize_t inode_get_rsv_space(struct inode *inode)
1569 if (!inode->i_sb->dq_op->get_reserved_space)
1571 spin_lock(&inode->i_lock);
1572 ret = *inode_reserved_space(inode);
1573 spin_unlock(&inode->i_lock);
1578 * This functions updates i_blocks+i_bytes fields and quota information
1579 * (together with appropriate checks).
1581 * NOTE: We absolutely rely on the fact that caller dirties the inode
1582 * (usually helpers in quotaops.h care about this) and holds a handle for
1583 * the current transaction so that dquot write and inode write go into the
1588 * This operation can block, but only after everything is updated
1590 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1592 int cnt, ret = 0, index;
1593 struct dquot_warn warn[MAXQUOTAS];
1594 int reserve = flags & DQUOT_SPACE_RESERVE;
1595 struct dquot **dquots;
1597 if (!dquot_active(inode)) {
1599 spin_lock(&inode->i_lock);
1600 *inode_reserved_space(inode) += number;
1601 spin_unlock(&inode->i_lock);
1603 inode_add_bytes(inode, number);
1608 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1609 warn[cnt].w_type = QUOTA_NL_NOWARN;
1611 dquots = i_dquot(inode);
1612 index = srcu_read_lock(&dquot_srcu);
1613 spin_lock(&dq_data_lock);
1614 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1617 ret = check_bdq(dquots[cnt], number,
1618 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1619 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1620 spin_unlock(&dq_data_lock);
1621 goto out_flush_warn;
1624 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1628 dquot_resv_space(dquots[cnt], number);
1630 dquot_incr_space(dquots[cnt], number);
1633 spin_lock(&inode->i_lock);
1634 *inode_reserved_space(inode) += number;
1635 spin_unlock(&inode->i_lock);
1637 inode_add_bytes(inode, number);
1639 spin_unlock(&dq_data_lock);
1642 goto out_flush_warn;
1643 mark_all_dquot_dirty(dquots);
1645 srcu_read_unlock(&dquot_srcu, index);
1646 flush_warnings(warn);
1650 EXPORT_SYMBOL(__dquot_alloc_space);
1653 * This operation can block, but only after everything is updated
1655 int dquot_alloc_inode(struct inode *inode)
1657 int cnt, ret = 0, index;
1658 struct dquot_warn warn[MAXQUOTAS];
1659 struct dquot * const *dquots;
1661 if (!dquot_active(inode))
1663 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1664 warn[cnt].w_type = QUOTA_NL_NOWARN;
1666 dquots = i_dquot(inode);
1667 index = srcu_read_lock(&dquot_srcu);
1668 spin_lock(&dq_data_lock);
1669 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1672 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1677 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1680 dquot_incr_inodes(dquots[cnt], 1);
1684 spin_unlock(&dq_data_lock);
1686 mark_all_dquot_dirty(dquots);
1687 srcu_read_unlock(&dquot_srcu, index);
1688 flush_warnings(warn);
1691 EXPORT_SYMBOL(dquot_alloc_inode);
1694 * Convert in-memory reserved quotas to real consumed quotas
1696 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1698 struct dquot **dquots;
1701 if (!dquot_active(inode)) {
1702 spin_lock(&inode->i_lock);
1703 *inode_reserved_space(inode) -= number;
1704 __inode_add_bytes(inode, number);
1705 spin_unlock(&inode->i_lock);
1709 dquots = i_dquot(inode);
1710 index = srcu_read_lock(&dquot_srcu);
1711 spin_lock(&dq_data_lock);
1712 /* Claim reserved quotas to allocated quotas */
1713 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1715 struct dquot *dquot = dquots[cnt];
1717 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1718 number = dquot->dq_dqb.dqb_rsvspace;
1719 dquot->dq_dqb.dqb_curspace += number;
1720 dquot->dq_dqb.dqb_rsvspace -= number;
1723 /* Update inode bytes */
1724 spin_lock(&inode->i_lock);
1725 *inode_reserved_space(inode) -= number;
1726 __inode_add_bytes(inode, number);
1727 spin_unlock(&inode->i_lock);
1728 spin_unlock(&dq_data_lock);
1729 mark_all_dquot_dirty(dquots);
1730 srcu_read_unlock(&dquot_srcu, index);
1733 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1736 * Convert allocated space back to in-memory reserved quotas
1738 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1740 struct dquot **dquots;
1743 if (!dquot_active(inode)) {
1744 spin_lock(&inode->i_lock);
1745 *inode_reserved_space(inode) += number;
1746 __inode_sub_bytes(inode, number);
1747 spin_unlock(&inode->i_lock);
1751 dquots = i_dquot(inode);
1752 index = srcu_read_lock(&dquot_srcu);
1753 spin_lock(&dq_data_lock);
1754 /* Claim reserved quotas to allocated quotas */
1755 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1757 struct dquot *dquot = dquots[cnt];
1759 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1760 number = dquot->dq_dqb.dqb_curspace;
1761 dquot->dq_dqb.dqb_rsvspace += number;
1762 dquot->dq_dqb.dqb_curspace -= number;
1765 /* Update inode bytes */
1766 spin_lock(&inode->i_lock);
1767 *inode_reserved_space(inode) += number;
1768 __inode_sub_bytes(inode, number);
1769 spin_unlock(&inode->i_lock);
1770 spin_unlock(&dq_data_lock);
1771 mark_all_dquot_dirty(dquots);
1772 srcu_read_unlock(&dquot_srcu, index);
1775 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1778 * This operation can block, but only after everything is updated
1780 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1783 struct dquot_warn warn[MAXQUOTAS];
1784 struct dquot **dquots;
1785 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1787 if (!dquot_active(inode)) {
1789 spin_lock(&inode->i_lock);
1790 *inode_reserved_space(inode) -= number;
1791 spin_unlock(&inode->i_lock);
1793 inode_sub_bytes(inode, number);
1798 dquots = i_dquot(inode);
1799 index = srcu_read_lock(&dquot_srcu);
1800 spin_lock(&dq_data_lock);
1801 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1804 warn[cnt].w_type = QUOTA_NL_NOWARN;
1807 wtype = info_bdq_free(dquots[cnt], number);
1808 if (wtype != QUOTA_NL_NOWARN)
1809 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1811 dquot_free_reserved_space(dquots[cnt], number);
1813 dquot_decr_space(dquots[cnt], number);
1816 spin_lock(&inode->i_lock);
1817 *inode_reserved_space(inode) -= number;
1818 spin_unlock(&inode->i_lock);
1820 inode_sub_bytes(inode, number);
1822 spin_unlock(&dq_data_lock);
1826 mark_all_dquot_dirty(dquots);
1828 srcu_read_unlock(&dquot_srcu, index);
1829 flush_warnings(warn);
1831 EXPORT_SYMBOL(__dquot_free_space);
1834 * This operation can block, but only after everything is updated
1836 void dquot_free_inode(struct inode *inode)
1839 struct dquot_warn warn[MAXQUOTAS];
1840 struct dquot * const *dquots;
1843 if (!dquot_active(inode))
1846 dquots = i_dquot(inode);
1847 index = srcu_read_lock(&dquot_srcu);
1848 spin_lock(&dq_data_lock);
1849 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1852 warn[cnt].w_type = QUOTA_NL_NOWARN;
1855 wtype = info_idq_free(dquots[cnt], 1);
1856 if (wtype != QUOTA_NL_NOWARN)
1857 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1858 dquot_decr_inodes(dquots[cnt], 1);
1860 spin_unlock(&dq_data_lock);
1861 mark_all_dquot_dirty(dquots);
1862 srcu_read_unlock(&dquot_srcu, index);
1863 flush_warnings(warn);
1865 EXPORT_SYMBOL(dquot_free_inode);
1868 * Transfer the number of inode and blocks from one diskquota to an other.
1869 * On success, dquot references in transfer_to are consumed and references
1870 * to original dquots that need to be released are placed there. On failure,
1871 * references are kept untouched.
1873 * This operation can block, but only after everything is updated
1874 * A transaction must be started when entering this function.
1876 * We are holding reference on transfer_from & transfer_to, no need to
1877 * protect them by srcu_read_lock().
1879 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1881 qsize_t space, cur_space;
1882 qsize_t rsv_space = 0;
1883 qsize_t inode_usage = 1;
1884 struct dquot *transfer_from[MAXQUOTAS] = {};
1886 char is_valid[MAXQUOTAS] = {};
1887 struct dquot_warn warn_to[MAXQUOTAS];
1888 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1889 struct dquot_warn warn_from_space[MAXQUOTAS];
1891 if (IS_NOQUOTA(inode))
1894 if (inode->i_sb->dq_op->get_inode_usage) {
1895 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1900 /* Initialize the arrays */
1901 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1902 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1903 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1904 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1907 spin_lock(&dq_data_lock);
1908 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1909 spin_unlock(&dq_data_lock);
1912 cur_space = inode_get_bytes(inode);
1913 rsv_space = inode_get_rsv_space(inode);
1914 space = cur_space + rsv_space;
1915 /* Build the transfer_from list and check the limits */
1916 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1918 * Skip changes for same uid or gid or for turned off quota-type.
1920 if (!transfer_to[cnt])
1922 /* Avoid races with quotaoff() */
1923 if (!sb_has_quota_active(inode->i_sb, cnt))
1926 transfer_from[cnt] = i_dquot(inode)[cnt];
1927 ret = check_idq(transfer_to[cnt], inode_usage, &warn_to[cnt]);
1930 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1936 * Finally perform the needed transfer from transfer_from to transfer_to
1938 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1941 /* Due to IO error we might not have transfer_from[] structure */
1942 if (transfer_from[cnt]) {
1944 wtype = info_idq_free(transfer_from[cnt], inode_usage);
1945 if (wtype != QUOTA_NL_NOWARN)
1946 prepare_warning(&warn_from_inodes[cnt],
1947 transfer_from[cnt], wtype);
1948 wtype = info_bdq_free(transfer_from[cnt], space);
1949 if (wtype != QUOTA_NL_NOWARN)
1950 prepare_warning(&warn_from_space[cnt],
1951 transfer_from[cnt], wtype);
1952 dquot_decr_inodes(transfer_from[cnt], inode_usage);
1953 dquot_decr_space(transfer_from[cnt], cur_space);
1954 dquot_free_reserved_space(transfer_from[cnt],
1958 dquot_incr_inodes(transfer_to[cnt], inode_usage);
1959 dquot_incr_space(transfer_to[cnt], cur_space);
1960 dquot_resv_space(transfer_to[cnt], rsv_space);
1962 i_dquot(inode)[cnt] = transfer_to[cnt];
1964 spin_unlock(&dq_data_lock);
1966 mark_all_dquot_dirty(transfer_from);
1967 mark_all_dquot_dirty(transfer_to);
1968 flush_warnings(warn_to);
1969 flush_warnings(warn_from_inodes);
1970 flush_warnings(warn_from_space);
1971 /* Pass back references to put */
1972 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1974 transfer_to[cnt] = transfer_from[cnt];
1977 spin_unlock(&dq_data_lock);
1978 flush_warnings(warn_to);
1981 EXPORT_SYMBOL(__dquot_transfer);
1983 /* Wrapper for transferring ownership of an inode for uid/gid only
1984 * Called from FSXXX_setattr()
1986 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1988 struct dquot *transfer_to[MAXQUOTAS] = {};
1989 struct dquot *dquot;
1990 struct super_block *sb = inode->i_sb;
1993 if (!dquot_active(inode))
1996 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
1997 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
1998 if (IS_ERR(dquot)) {
1999 if (PTR_ERR(dquot) != -ESRCH) {
2000 ret = PTR_ERR(dquot);
2005 transfer_to[USRQUOTA] = dquot;
2007 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2008 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2009 if (IS_ERR(dquot)) {
2010 if (PTR_ERR(dquot) != -ESRCH) {
2011 ret = PTR_ERR(dquot);
2016 transfer_to[GRPQUOTA] = dquot;
2018 ret = __dquot_transfer(inode, transfer_to);
2020 dqput_all(transfer_to);
2023 EXPORT_SYMBOL(dquot_transfer);
2026 * Write info of quota file to disk
2028 int dquot_commit_info(struct super_block *sb, int type)
2030 struct quota_info *dqopt = sb_dqopt(sb);
2032 return dqopt->ops[type]->write_file_info(sb, type);
2034 EXPORT_SYMBOL(dquot_commit_info);
2036 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2038 struct quota_info *dqopt = sb_dqopt(sb);
2040 if (!sb_has_quota_active(sb, qid->type))
2042 if (!dqopt->ops[qid->type]->get_next_id)
2044 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2046 EXPORT_SYMBOL(dquot_get_next_id);
2049 * Definitions of diskquota operations.
2051 const struct dquot_operations dquot_operations = {
2052 .write_dquot = dquot_commit,
2053 .acquire_dquot = dquot_acquire,
2054 .release_dquot = dquot_release,
2055 .mark_dirty = dquot_mark_dquot_dirty,
2056 .write_info = dquot_commit_info,
2057 .alloc_dquot = dquot_alloc,
2058 .destroy_dquot = dquot_destroy,
2059 .get_next_id = dquot_get_next_id,
2061 EXPORT_SYMBOL(dquot_operations);
2064 * Generic helper for ->open on filesystems supporting disk quotas.
2066 int dquot_file_open(struct inode *inode, struct file *file)
2070 error = generic_file_open(inode, file);
2071 if (!error && (file->f_mode & FMODE_WRITE))
2072 dquot_initialize(inode);
2075 EXPORT_SYMBOL(dquot_file_open);
2078 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2080 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2083 struct quota_info *dqopt = sb_dqopt(sb);
2084 struct inode *toputinode[MAXQUOTAS];
2086 /* s_umount should be held in exclusive mode */
2087 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2088 up_read(&sb->s_umount);
2090 /* Cannot turn off usage accounting without turning off limits, or
2091 * suspend quotas and simultaneously turn quotas off. */
2092 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2093 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2094 DQUOT_USAGE_ENABLED)))
2098 * Skip everything if there's nothing to do. We have to do this because
2099 * sometimes we are called when fill_super() failed and calling
2100 * sync_fs() in such cases does no good.
2102 if (!sb_any_quota_loaded(sb))
2105 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2106 toputinode[cnt] = NULL;
2107 if (type != -1 && cnt != type)
2109 if (!sb_has_quota_loaded(sb, cnt))
2112 if (flags & DQUOT_SUSPENDED) {
2113 spin_lock(&dq_state_lock);
2115 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2116 spin_unlock(&dq_state_lock);
2118 spin_lock(&dq_state_lock);
2119 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2120 /* Turning off suspended quotas? */
2121 if (!sb_has_quota_loaded(sb, cnt) &&
2122 sb_has_quota_suspended(sb, cnt)) {
2123 dqopt->flags &= ~dquot_state_flag(
2124 DQUOT_SUSPENDED, cnt);
2125 spin_unlock(&dq_state_lock);
2126 iput(dqopt->files[cnt]);
2127 dqopt->files[cnt] = NULL;
2130 spin_unlock(&dq_state_lock);
2133 /* We still have to keep quota loaded? */
2134 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2137 /* Note: these are blocking operations */
2138 drop_dquot_ref(sb, cnt);
2139 invalidate_dquots(sb, cnt);
2141 * Now all dquots should be invalidated, all writes done so we
2142 * should be only users of the info. No locks needed.
2144 if (info_dirty(&dqopt->info[cnt]))
2145 sb->dq_op->write_info(sb, cnt);
2146 if (dqopt->ops[cnt]->free_file_info)
2147 dqopt->ops[cnt]->free_file_info(sb, cnt);
2148 put_quota_format(dqopt->info[cnt].dqi_format);
2150 toputinode[cnt] = dqopt->files[cnt];
2151 if (!sb_has_quota_loaded(sb, cnt))
2152 dqopt->files[cnt] = NULL;
2153 dqopt->info[cnt].dqi_flags = 0;
2154 dqopt->info[cnt].dqi_igrace = 0;
2155 dqopt->info[cnt].dqi_bgrace = 0;
2156 dqopt->ops[cnt] = NULL;
2159 /* Skip syncing and setting flags if quota files are hidden */
2160 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2163 /* Sync the superblock so that buffers with quota data are written to
2164 * disk (and so userspace sees correct data afterwards). */
2165 if (sb->s_op->sync_fs)
2166 sb->s_op->sync_fs(sb, 1);
2167 sync_blockdev(sb->s_bdev);
2168 /* Now the quota files are just ordinary files and we can set the
2169 * inode flags back. Moreover we discard the pagecache so that
2170 * userspace sees the writes we did bypassing the pagecache. We
2171 * must also discard the blockdev buffers so that we see the
2172 * changes done by userspace on the next quotaon() */
2173 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2174 /* This can happen when suspending quotas on remount-ro... */
2175 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2176 inode_lock(toputinode[cnt]);
2177 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2178 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2179 inode_unlock(toputinode[cnt]);
2180 mark_inode_dirty_sync(toputinode[cnt]);
2183 invalidate_bdev(sb->s_bdev);
2185 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2186 if (toputinode[cnt]) {
2187 /* On remount RO, we keep the inode pointer so that we
2188 * can reenable quota on the subsequent remount RW. We
2189 * have to check 'flags' variable and not use sb_has_
2190 * function because another quotaon / quotaoff could
2191 * change global state before we got here. We refuse
2192 * to suspend quotas when there is pending delete on
2193 * the quota file... */
2194 if (!(flags & DQUOT_SUSPENDED))
2195 iput(toputinode[cnt]);
2196 else if (!toputinode[cnt]->i_nlink)
2201 EXPORT_SYMBOL(dquot_disable);
2203 int dquot_quota_off(struct super_block *sb, int type)
2205 return dquot_disable(sb, type,
2206 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2208 EXPORT_SYMBOL(dquot_quota_off);
2211 * Turn quotas on on a device
2215 * Helper function to turn quotas on when we already have the inode of
2216 * quota file and no quota information is loaded.
2218 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2221 struct quota_format_type *fmt = find_quota_format(format_id);
2222 struct super_block *sb = inode->i_sb;
2223 struct quota_info *dqopt = sb_dqopt(sb);
2228 if (!S_ISREG(inode->i_mode)) {
2232 if (IS_RDONLY(inode)) {
2236 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2237 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2241 /* Filesystems outside of init_user_ns not yet supported */
2242 if (sb->s_user_ns != &init_user_ns) {
2246 /* Usage always has to be set... */
2247 if (!(flags & DQUOT_USAGE_ENABLED)) {
2251 if (sb_has_quota_loaded(sb, type)) {
2256 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2257 /* As we bypass the pagecache we must now flush all the
2258 * dirty data and invalidate caches so that kernel sees
2259 * changes from userspace. It is not enough to just flush
2260 * the quota file since if blocksize < pagesize, invalidation
2261 * of the cache could fail because of other unrelated dirty
2263 sync_filesystem(sb);
2264 invalidate_bdev(sb->s_bdev);
2267 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2268 /* We don't want quota and atime on quota files (deadlocks
2269 * possible) Also nobody should write to the file - we use
2270 * special IO operations which ignore the immutable bit. */
2272 inode->i_flags |= S_NOQUOTA;
2273 inode_unlock(inode);
2275 * When S_NOQUOTA is set, remove dquot references as no more
2276 * references can be added
2278 __dquot_drop(inode);
2282 dqopt->files[type] = igrab(inode);
2283 if (!dqopt->files[type])
2284 goto out_file_flags;
2286 if (!fmt->qf_ops->check_quota_file(sb, type))
2289 dqopt->ops[type] = fmt->qf_ops;
2290 dqopt->info[type].dqi_format = fmt;
2291 dqopt->info[type].dqi_fmt_id = format_id;
2292 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2293 error = dqopt->ops[type]->read_file_info(sb, type);
2296 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2297 spin_lock(&dq_data_lock);
2298 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2299 spin_unlock(&dq_data_lock);
2301 spin_lock(&dq_state_lock);
2302 dqopt->flags |= dquot_state_flag(flags, type);
2303 spin_unlock(&dq_state_lock);
2305 add_dquot_ref(sb, type);
2310 dqopt->files[type] = NULL;
2314 inode->i_flags &= ~S_NOQUOTA;
2315 inode_unlock(inode);
2317 put_quota_format(fmt);
2322 /* Reenable quotas on remount RW */
2323 int dquot_resume(struct super_block *sb, int type)
2325 struct quota_info *dqopt = sb_dqopt(sb);
2326 struct inode *inode;
2330 /* s_umount should be held in exclusive mode */
2331 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2332 up_read(&sb->s_umount);
2334 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2335 if (type != -1 && cnt != type)
2337 if (!sb_has_quota_suspended(sb, cnt))
2340 inode = dqopt->files[cnt];
2341 dqopt->files[cnt] = NULL;
2342 spin_lock(&dq_state_lock);
2343 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2344 DQUOT_LIMITS_ENABLED,
2346 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2347 spin_unlock(&dq_state_lock);
2349 flags = dquot_generic_flag(flags, cnt);
2350 ret = vfs_load_quota_inode(inode, cnt,
2351 dqopt->info[cnt].dqi_fmt_id, flags);
2357 EXPORT_SYMBOL(dquot_resume);
2359 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2360 const struct path *path)
2362 int error = security_quota_on(path->dentry);
2365 /* Quota file not on the same filesystem? */
2366 if (path->dentry->d_sb != sb)
2369 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2370 format_id, DQUOT_USAGE_ENABLED |
2371 DQUOT_LIMITS_ENABLED);
2374 EXPORT_SYMBOL(dquot_quota_on);
2377 * More powerful function for turning on quotas allowing setting
2378 * of individual quota flags
2380 int dquot_enable(struct inode *inode, int type, int format_id,
2383 struct super_block *sb = inode->i_sb;
2385 /* Just unsuspend quotas? */
2386 BUG_ON(flags & DQUOT_SUSPENDED);
2387 /* s_umount should be held in exclusive mode */
2388 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2389 up_read(&sb->s_umount);
2393 /* Just updating flags needed? */
2394 if (sb_has_quota_loaded(sb, type)) {
2395 if (flags & DQUOT_USAGE_ENABLED &&
2396 sb_has_quota_usage_enabled(sb, type))
2398 if (flags & DQUOT_LIMITS_ENABLED &&
2399 sb_has_quota_limits_enabled(sb, type))
2401 spin_lock(&dq_state_lock);
2402 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2403 spin_unlock(&dq_state_lock);
2407 return vfs_load_quota_inode(inode, type, format_id, flags);
2409 EXPORT_SYMBOL(dquot_enable);
2412 * This function is used when filesystem needs to initialize quotas
2413 * during mount time.
2415 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2416 int format_id, int type)
2418 struct dentry *dentry;
2421 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2423 return PTR_ERR(dentry);
2425 if (d_really_is_negative(dentry)) {
2430 error = security_quota_on(dentry);
2432 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2433 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2439 EXPORT_SYMBOL(dquot_quota_on_mount);
2441 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2445 struct quota_info *dqopt = sb_dqopt(sb);
2447 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2449 /* Accounting cannot be turned on while fs is mounted */
2450 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2453 for (type = 0; type < MAXQUOTAS; type++) {
2454 if (!(flags & qtype_enforce_flag(type)))
2456 /* Can't enforce without accounting */
2457 if (!sb_has_quota_usage_enabled(sb, type))
2459 ret = dquot_enable(dqopt->files[type], type,
2460 dqopt->info[type].dqi_fmt_id,
2461 DQUOT_LIMITS_ENABLED);
2467 /* Backout enforcement enablement we already did */
2468 for (type--; type >= 0; type--) {
2469 if (flags & qtype_enforce_flag(type))
2470 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2472 /* Error code translation for better compatibility with XFS */
2478 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2482 struct quota_info *dqopt = sb_dqopt(sb);
2484 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2487 * We don't support turning off accounting via quotactl. In principle
2488 * quota infrastructure can do this but filesystems don't expect
2489 * userspace to be able to do it.
2492 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2495 /* Filter out limits not enabled */
2496 for (type = 0; type < MAXQUOTAS; type++)
2497 if (!sb_has_quota_limits_enabled(sb, type))
2498 flags &= ~qtype_enforce_flag(type);
2502 for (type = 0; type < MAXQUOTAS; type++) {
2503 if (flags & qtype_enforce_flag(type)) {
2504 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2511 /* Backout enforcement disabling we already did */
2512 for (type--; type >= 0; type--) {
2513 if (flags & qtype_enforce_flag(type))
2514 dquot_enable(dqopt->files[type], type,
2515 dqopt->info[type].dqi_fmt_id,
2516 DQUOT_LIMITS_ENABLED);
2521 /* Generic routine for getting common part of quota structure */
2522 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2524 struct mem_dqblk *dm = &dquot->dq_dqb;
2526 memset(di, 0, sizeof(*di));
2527 spin_lock(&dq_data_lock);
2528 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2529 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2530 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2531 di->d_ino_softlimit = dm->dqb_isoftlimit;
2532 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2533 di->d_ino_count = dm->dqb_curinodes;
2534 di->d_spc_timer = dm->dqb_btime;
2535 di->d_ino_timer = dm->dqb_itime;
2536 spin_unlock(&dq_data_lock);
2539 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2540 struct qc_dqblk *di)
2542 struct dquot *dquot;
2544 dquot = dqget(sb, qid);
2546 return PTR_ERR(dquot);
2547 do_get_dqblk(dquot, di);
2552 EXPORT_SYMBOL(dquot_get_dqblk);
2554 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2555 struct qc_dqblk *di)
2557 struct dquot *dquot;
2560 if (!sb->dq_op->get_next_id)
2562 err = sb->dq_op->get_next_id(sb, qid);
2565 dquot = dqget(sb, *qid);
2567 return PTR_ERR(dquot);
2568 do_get_dqblk(dquot, di);
2573 EXPORT_SYMBOL(dquot_get_next_dqblk);
2575 #define VFS_QC_MASK \
2576 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2577 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2578 QC_SPC_TIMER | QC_INO_TIMER)
2580 /* Generic routine for setting common part of quota structure */
2581 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2583 struct mem_dqblk *dm = &dquot->dq_dqb;
2584 int check_blim = 0, check_ilim = 0;
2585 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2587 if (di->d_fieldmask & ~VFS_QC_MASK)
2590 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2591 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2592 ((di->d_fieldmask & QC_SPC_HARD) &&
2593 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2594 ((di->d_fieldmask & QC_INO_SOFT) &&
2595 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2596 ((di->d_fieldmask & QC_INO_HARD) &&
2597 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2600 spin_lock(&dq_data_lock);
2601 if (di->d_fieldmask & QC_SPACE) {
2602 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2604 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2607 if (di->d_fieldmask & QC_SPC_SOFT)
2608 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2609 if (di->d_fieldmask & QC_SPC_HARD)
2610 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2611 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2613 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2616 if (di->d_fieldmask & QC_INO_COUNT) {
2617 dm->dqb_curinodes = di->d_ino_count;
2619 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2622 if (di->d_fieldmask & QC_INO_SOFT)
2623 dm->dqb_isoftlimit = di->d_ino_softlimit;
2624 if (di->d_fieldmask & QC_INO_HARD)
2625 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2626 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2628 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2631 if (di->d_fieldmask & QC_SPC_TIMER) {
2632 dm->dqb_btime = di->d_spc_timer;
2634 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2637 if (di->d_fieldmask & QC_INO_TIMER) {
2638 dm->dqb_itime = di->d_ino_timer;
2640 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2644 if (!dm->dqb_bsoftlimit ||
2645 dm->dqb_curspace < dm->dqb_bsoftlimit) {
2647 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2648 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2649 /* Set grace only if user hasn't provided his own... */
2650 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2653 if (!dm->dqb_isoftlimit ||
2654 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2656 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2657 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2658 /* Set grace only if user hasn't provided his own... */
2659 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2661 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2663 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2665 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2666 spin_unlock(&dq_data_lock);
2667 mark_dquot_dirty(dquot);
2672 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2673 struct qc_dqblk *di)
2675 struct dquot *dquot;
2678 dquot = dqget(sb, qid);
2679 if (IS_ERR(dquot)) {
2680 rc = PTR_ERR(dquot);
2683 rc = do_set_dqblk(dquot, di);
2688 EXPORT_SYMBOL(dquot_set_dqblk);
2690 /* Generic routine for getting common part of quota file information */
2691 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2693 struct mem_dqinfo *mi;
2694 struct qc_type_state *tstate;
2695 struct quota_info *dqopt = sb_dqopt(sb);
2698 memset(state, 0, sizeof(*state));
2699 for (type = 0; type < MAXQUOTAS; type++) {
2700 if (!sb_has_quota_active(sb, type))
2702 tstate = state->s_state + type;
2703 mi = sb_dqopt(sb)->info + type;
2704 tstate->flags = QCI_ACCT_ENABLED;
2705 spin_lock(&dq_data_lock);
2706 if (mi->dqi_flags & DQF_SYS_FILE)
2707 tstate->flags |= QCI_SYSFILE;
2708 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2709 tstate->flags |= QCI_ROOT_SQUASH;
2710 if (sb_has_quota_limits_enabled(sb, type))
2711 tstate->flags |= QCI_LIMITS_ENFORCED;
2712 tstate->spc_timelimit = mi->dqi_bgrace;
2713 tstate->ino_timelimit = mi->dqi_igrace;
2714 tstate->ino = dqopt->files[type]->i_ino;
2715 tstate->blocks = dqopt->files[type]->i_blocks;
2716 tstate->nextents = 1; /* We don't know... */
2717 spin_unlock(&dq_data_lock);
2721 EXPORT_SYMBOL(dquot_get_state);
2723 /* Generic routine for setting common part of quota file information */
2724 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2726 struct mem_dqinfo *mi;
2729 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2730 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2732 if (!sb_has_quota_active(sb, type))
2734 mi = sb_dqopt(sb)->info + type;
2735 if (ii->i_fieldmask & QC_FLAGS) {
2736 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2737 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2740 spin_lock(&dq_data_lock);
2741 if (ii->i_fieldmask & QC_SPC_TIMER)
2742 mi->dqi_bgrace = ii->i_spc_timelimit;
2743 if (ii->i_fieldmask & QC_INO_TIMER)
2744 mi->dqi_igrace = ii->i_ino_timelimit;
2745 if (ii->i_fieldmask & QC_FLAGS) {
2746 if (ii->i_flags & QCI_ROOT_SQUASH)
2747 mi->dqi_flags |= DQF_ROOT_SQUASH;
2749 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2751 spin_unlock(&dq_data_lock);
2752 mark_info_dirty(sb, type);
2753 /* Force write to disk */
2754 sb->dq_op->write_info(sb, type);
2757 EXPORT_SYMBOL(dquot_set_dqinfo);
2759 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2760 .quota_enable = dquot_quota_enable,
2761 .quota_disable = dquot_quota_disable,
2762 .quota_sync = dquot_quota_sync,
2763 .get_state = dquot_get_state,
2764 .set_info = dquot_set_dqinfo,
2765 .get_dqblk = dquot_get_dqblk,
2766 .get_nextdqblk = dquot_get_next_dqblk,
2767 .set_dqblk = dquot_set_dqblk
2769 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2771 static int do_proc_dqstats(struct ctl_table *table, int write,
2772 void __user *buffer, size_t *lenp, loff_t *ppos)
2774 unsigned int type = (int *)table->data - dqstats.stat;
2776 /* Update global table */
2777 dqstats.stat[type] =
2778 percpu_counter_sum_positive(&dqstats.counter[type]);
2779 return proc_dointvec(table, write, buffer, lenp, ppos);
2782 static struct ctl_table fs_dqstats_table[] = {
2784 .procname = "lookups",
2785 .data = &dqstats.stat[DQST_LOOKUPS],
2786 .maxlen = sizeof(int),
2788 .proc_handler = do_proc_dqstats,
2791 .procname = "drops",
2792 .data = &dqstats.stat[DQST_DROPS],
2793 .maxlen = sizeof(int),
2795 .proc_handler = do_proc_dqstats,
2798 .procname = "reads",
2799 .data = &dqstats.stat[DQST_READS],
2800 .maxlen = sizeof(int),
2802 .proc_handler = do_proc_dqstats,
2805 .procname = "writes",
2806 .data = &dqstats.stat[DQST_WRITES],
2807 .maxlen = sizeof(int),
2809 .proc_handler = do_proc_dqstats,
2812 .procname = "cache_hits",
2813 .data = &dqstats.stat[DQST_CACHE_HITS],
2814 .maxlen = sizeof(int),
2816 .proc_handler = do_proc_dqstats,
2819 .procname = "allocated_dquots",
2820 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2821 .maxlen = sizeof(int),
2823 .proc_handler = do_proc_dqstats,
2826 .procname = "free_dquots",
2827 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2828 .maxlen = sizeof(int),
2830 .proc_handler = do_proc_dqstats,
2833 .procname = "syncs",
2834 .data = &dqstats.stat[DQST_SYNCS],
2835 .maxlen = sizeof(int),
2837 .proc_handler = do_proc_dqstats,
2839 #ifdef CONFIG_PRINT_QUOTA_WARNING
2841 .procname = "warnings",
2842 .data = &flag_print_warnings,
2843 .maxlen = sizeof(int),
2845 .proc_handler = proc_dointvec,
2851 static struct ctl_table fs_table[] = {
2853 .procname = "quota",
2855 .child = fs_dqstats_table,
2860 static struct ctl_table sys_table[] = {
2869 static int __init dquot_init(void)
2872 unsigned long nr_hash, order;
2874 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2876 register_sysctl_table(sys_table);
2878 dquot_cachep = kmem_cache_create("dquot",
2879 sizeof(struct dquot), sizeof(unsigned long) * 4,
2880 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2881 SLAB_MEM_SPREAD|SLAB_PANIC),
2885 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2887 panic("Cannot create dquot hash table");
2889 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2890 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2892 panic("Cannot create dquot stat counters");
2895 /* Find power-of-two hlist_heads which can fit into allocation */
2896 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2900 } while (nr_hash >> dq_hash_bits);
2903 nr_hash = 1UL << dq_hash_bits;
2904 dq_hash_mask = nr_hash - 1;
2905 for (i = 0; i < nr_hash; i++)
2906 INIT_HLIST_HEAD(dquot_hash + i);
2908 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2909 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2911 register_shrinker(&dqcache_shrinker);
2915 fs_initcall(dquot_init);