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 * Version: $Id: dquot.c,v 6.3 1996/11/17 18:35:34 mvw Exp mvw $
14 * Author: Marco van Wieringen <mvw@planets.elm.net>
16 * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
18 * Revised list management to avoid races
19 * -- Bill Hawes, <whawes@star.net>, 9/98
21 * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
22 * As the consequence the locking was moved from dquot_decr_...(),
23 * dquot_incr_...() to calling functions.
24 * invalidate_dquots() now writes modified dquots.
25 * Serialized quota_off() and quota_on() for mount point.
26 * Fixed a few bugs in grow_dquots().
27 * Fixed deadlock in write_dquot() - we no longer account quotas on
29 * remove_dquot_ref() moved to inode.c - it now traverses through inodes
30 * add_dquot_ref() restarts after blocking
31 * Added check for bogus uid and fixed check for group in quotactl.
32 * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
34 * Used struct list_head instead of own list struct
35 * Invalidation of referenced dquots is no longer possible
36 * Improved free_dquots list management
37 * Quota and i_blocks are now updated in one place to avoid races
38 * Warnings are now delayed so we won't block in critical section
39 * Write updated not to require dquot lock
40 * Jan Kara, <jack@suse.cz>, 9/2000
42 * Added dynamic quota structure allocation
43 * Jan Kara <jack@suse.cz> 12/2000
45 * Rewritten quota interface. Implemented new quota format and
46 * formats registering.
47 * Jan Kara, <jack@suse.cz>, 2001,2002
50 * Jan Kara, <jack@suse.cz>, 10/2002
52 * Added journalled quota support, fix lock inversion problems
53 * Jan Kara, <jack@suse.cz>, 2003,2004
55 * (C) Copyright 1994 - 1997 Marco van Wieringen
58 #include <linux/errno.h>
59 #include <linux/kernel.h>
61 #include <linux/mount.h>
63 #include <linux/time.h>
64 #include <linux/types.h>
65 #include <linux/string.h>
66 #include <linux/fcntl.h>
67 #include <linux/stat.h>
68 #include <linux/tty.h>
69 #include <linux/file.h>
70 #include <linux/slab.h>
71 #include <linux/sysctl.h>
72 #include <linux/init.h>
73 #include <linux/module.h>
74 #include <linux/proc_fs.h>
75 #include <linux/security.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/buffer_head.h>
79 #include <linux/capability.h>
80 #include <linux/quotaops.h>
81 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
82 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
83 #include <net/netlink.h>
84 #include <net/genetlink.h>
87 #include <asm/uaccess.h>
89 #define __DQUOT_PARANOIA
92 * There are two quota SMP locks. dq_list_lock protects all lists with quotas
93 * and quota formats and also dqstats structure containing statistics about the
94 * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
95 * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
96 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
97 * in inode_add_bytes() and inode_sub_bytes().
99 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
101 * Note that some things (eg. sb pointer, type, id) doesn't change during
102 * the life of the dquot structure and so needn't to be protected by a lock
104 * Any operation working on dquots via inode pointers must hold dqptr_sem. If
105 * operation is just reading pointers from inode (or not using them at all) the
106 * read lock is enough. If pointers are altered function must hold write lock
107 * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
108 * for altering the flag i_mutex is also needed). If operation is holding
109 * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
111 * This locking assures that:
112 * a) update/access to dquot pointers in inode is serialized
113 * b) everyone is guarded against invalidate_dquots()
115 * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
116 * from inodes (dquot_alloc_space() and such don't check the dq_lock).
117 * Currently dquot is locked only when it is being read to memory (or space for
118 * it is being allocated) on the first dqget() and when it is being released on
119 * the last dqput(). The allocation and release oparations are serialized by
120 * the dq_lock and by checking the use count in dquot_release(). Write
121 * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
122 * spinlock to internal buffers before writing.
124 * Lock ordering (including related VFS locks) is the following:
125 * i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
127 * i_mutex on quota files is special (it's below dqio_mutex)
130 static DEFINE_SPINLOCK(dq_list_lock);
131 DEFINE_SPINLOCK(dq_data_lock);
133 static char *quotatypes[] = INITQFNAMES;
134 static struct quota_format_type *quota_formats; /* List of registered formats */
135 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
137 /* SLAB cache for dquot structures */
138 static struct kmem_cache *dquot_cachep;
140 int register_quota_format(struct quota_format_type *fmt)
142 spin_lock(&dq_list_lock);
143 fmt->qf_next = quota_formats;
145 spin_unlock(&dq_list_lock);
149 void unregister_quota_format(struct quota_format_type *fmt)
151 struct quota_format_type **actqf;
153 spin_lock(&dq_list_lock);
154 for (actqf = "a_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
156 *actqf = (*actqf)->qf_next;
157 spin_unlock(&dq_list_lock);
160 static struct quota_format_type *find_quota_format(int id)
162 struct quota_format_type *actqf;
164 spin_lock(&dq_list_lock);
165 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
166 if (!actqf || !try_module_get(actqf->qf_owner)) {
169 spin_unlock(&dq_list_lock);
171 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
172 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
175 spin_lock(&dq_list_lock);
176 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
177 if (actqf && !try_module_get(actqf->qf_owner))
180 spin_unlock(&dq_list_lock);
184 static void put_quota_format(struct quota_format_type *fmt)
186 module_put(fmt->qf_owner);
190 * Dquot List Management:
191 * The quota code uses three lists for dquot management: the inuse_list,
192 * free_dquots, and dquot_hash[] array. A single dquot structure may be
193 * on all three lists, depending on its current state.
195 * All dquots are placed to the end of inuse_list when first created, and this
196 * list is used for invalidate operation, which must look at every dquot.
198 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
199 * and this list is searched whenever we need an available dquot. Dquots are
200 * removed from the list as soon as they are used again, and
201 * dqstats.free_dquots gives the number of dquots on the list. When
202 * dquot is invalidated it's completely released from memory.
204 * Dquots with a specific identity (device, type and id) are placed on
205 * one of the dquot_hash[] hash chains. The provides an efficient search
206 * mechanism to locate a specific dquot.
209 static LIST_HEAD(inuse_list);
210 static LIST_HEAD(free_dquots);
211 static unsigned int dq_hash_bits, dq_hash_mask;
212 static struct hlist_head *dquot_hash;
214 struct dqstats dqstats;
216 static void dqput(struct dquot *dquot);
218 static inline unsigned int
219 hashfn(const struct super_block *sb, unsigned int id, int type)
223 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
224 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
228 * Following list functions expect dq_list_lock to be held
230 static inline void insert_dquot_hash(struct dquot *dquot)
232 struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
233 hlist_add_head(&dquot->dq_hash, head);
236 static inline void remove_dquot_hash(struct dquot *dquot)
238 hlist_del_init(&dquot->dq_hash);
241 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
243 struct hlist_node *node;
246 hlist_for_each (node, dquot_hash+hashent) {
247 dquot = hlist_entry(node, struct dquot, dq_hash);
248 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
254 /* Add a dquot to the tail of the free list */
255 static inline void put_dquot_last(struct dquot *dquot)
257 list_add_tail(&dquot->dq_free, &free_dquots);
258 dqstats.free_dquots++;
261 static inline void remove_free_dquot(struct dquot *dquot)
263 if (list_empty(&dquot->dq_free))
265 list_del_init(&dquot->dq_free);
266 dqstats.free_dquots--;
269 static inline void put_inuse(struct dquot *dquot)
271 /* We add to the back of inuse list so we don't have to restart
272 * when traversing this list and we block */
273 list_add_tail(&dquot->dq_inuse, &inuse_list);
274 dqstats.allocated_dquots++;
277 static inline void remove_inuse(struct dquot *dquot)
279 dqstats.allocated_dquots--;
280 list_del(&dquot->dq_inuse);
283 * End of list functions needing dq_list_lock
286 static void wait_on_dquot(struct dquot *dquot)
288 mutex_lock(&dquot->dq_lock);
289 mutex_unlock(&dquot->dq_lock);
292 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
294 int dquot_mark_dquot_dirty(struct dquot *dquot)
296 spin_lock(&dq_list_lock);
297 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
298 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
299 info[dquot->dq_type].dqi_dirty_list);
300 spin_unlock(&dq_list_lock);
304 /* This function needs dq_list_lock */
305 static inline int clear_dquot_dirty(struct dquot *dquot)
307 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
309 list_del_init(&dquot->dq_dirty);
313 void mark_info_dirty(struct super_block *sb, int type)
315 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
317 EXPORT_SYMBOL(mark_info_dirty);
320 * Read dquot from disk and alloc space for it
323 int dquot_acquire(struct dquot *dquot)
325 int ret = 0, ret2 = 0;
326 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
328 mutex_lock(&dquot->dq_lock);
329 mutex_lock(&dqopt->dqio_mutex);
330 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
331 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
334 set_bit(DQ_READ_B, &dquot->dq_flags);
335 /* Instantiate dquot if needed */
336 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
337 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
338 /* Write the info if needed */
339 if (info_dirty(&dqopt->info[dquot->dq_type]))
340 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
348 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
350 mutex_unlock(&dqopt->dqio_mutex);
351 mutex_unlock(&dquot->dq_lock);
356 * Write dquot to disk
358 int dquot_commit(struct dquot *dquot)
360 int ret = 0, ret2 = 0;
361 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
363 mutex_lock(&dqopt->dqio_mutex);
364 spin_lock(&dq_list_lock);
365 if (!clear_dquot_dirty(dquot)) {
366 spin_unlock(&dq_list_lock);
369 spin_unlock(&dq_list_lock);
370 /* Inactive dquot can be only if there was error during read/init
371 * => we have better not writing it */
372 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
373 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
374 if (info_dirty(&dqopt->info[dquot->dq_type]))
375 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
380 mutex_unlock(&dqopt->dqio_mutex);
387 int dquot_release(struct dquot *dquot)
389 int ret = 0, ret2 = 0;
390 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
392 mutex_lock(&dquot->dq_lock);
393 /* Check whether we are not racing with some other dqget() */
394 if (atomic_read(&dquot->dq_count) > 1)
396 mutex_lock(&dqopt->dqio_mutex);
397 if (dqopt->ops[dquot->dq_type]->release_dqblk) {
398 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
400 if (info_dirty(&dqopt->info[dquot->dq_type]))
401 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
405 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
406 mutex_unlock(&dqopt->dqio_mutex);
408 mutex_unlock(&dquot->dq_lock);
412 /* Invalidate all dquots on the list. Note that this function is called after
413 * quota is disabled and pointers from inodes removed so there cannot be new
414 * quota users. There can still be some users of quotas due to inodes being
415 * just deleted or pruned by prune_icache() (those are not attached to any
416 * list). We have to wait for such users.
418 static void invalidate_dquots(struct super_block *sb, int type)
420 struct dquot *dquot, *tmp;
423 spin_lock(&dq_list_lock);
424 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
425 if (dquot->dq_sb != sb)
427 if (dquot->dq_type != type)
429 /* Wait for dquot users */
430 if (atomic_read(&dquot->dq_count)) {
433 atomic_inc(&dquot->dq_count);
434 prepare_to_wait(&dquot->dq_wait_unused, &wait,
435 TASK_UNINTERRUPTIBLE);
436 spin_unlock(&dq_list_lock);
437 /* Once dqput() wakes us up, we know it's time to free
439 * IMPORTANT: we rely on the fact that there is always
440 * at most one process waiting for dquot to free.
441 * Otherwise dq_count would be > 1 and we would never
444 if (atomic_read(&dquot->dq_count) > 1)
446 finish_wait(&dquot->dq_wait_unused, &wait);
448 /* At this moment dquot() need not exist (it could be
449 * reclaimed by prune_dqcache(). Hence we must
454 * Quota now has no users and it has been written on last
457 remove_dquot_hash(dquot);
458 remove_free_dquot(dquot);
460 kmem_cache_free(dquot_cachep, dquot);
462 spin_unlock(&dq_list_lock);
465 int vfs_quota_sync(struct super_block *sb, int type)
467 struct list_head *dirty;
469 struct quota_info *dqopt = sb_dqopt(sb);
472 mutex_lock(&dqopt->dqonoff_mutex);
473 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
474 if (type != -1 && cnt != type)
476 if (!sb_has_quota_enabled(sb, cnt))
478 spin_lock(&dq_list_lock);
479 dirty = &dqopt->info[cnt].dqi_dirty_list;
480 while (!list_empty(dirty)) {
481 dquot = list_first_entry(dirty, struct dquot, dq_dirty);
482 /* Dirty and inactive can be only bad dquot... */
483 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
484 clear_dquot_dirty(dquot);
487 /* Now we have active dquot from which someone is
488 * holding reference so we can safely just increase
490 atomic_inc(&dquot->dq_count);
492 spin_unlock(&dq_list_lock);
493 sb->dq_op->write_dquot(dquot);
495 spin_lock(&dq_list_lock);
497 spin_unlock(&dq_list_lock);
500 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
501 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
502 && info_dirty(&dqopt->info[cnt]))
503 sb->dq_op->write_info(sb, cnt);
504 spin_lock(&dq_list_lock);
506 spin_unlock(&dq_list_lock);
507 mutex_unlock(&dqopt->dqonoff_mutex);
512 /* Free unused dquots from cache */
513 static void prune_dqcache(int count)
515 struct list_head *head;
518 head = free_dquots.prev;
519 while (head != &free_dquots && count) {
520 dquot = list_entry(head, struct dquot, dq_free);
521 remove_dquot_hash(dquot);
522 remove_free_dquot(dquot);
524 kmem_cache_free(dquot_cachep, dquot);
526 head = free_dquots.prev;
531 * This is called from kswapd when we think we need some
535 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
538 spin_lock(&dq_list_lock);
540 spin_unlock(&dq_list_lock);
542 return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
545 static struct shrinker dqcache_shrinker = {
546 .shrink = shrink_dqcache_memory,
547 .seeks = DEFAULT_SEEKS,
551 * Put reference to dquot
552 * NOTE: If you change this function please check whether dqput_blocks() works right...
553 * MUST be called with either dqptr_sem or dqonoff_mutex held
555 static void dqput(struct dquot *dquot)
559 #ifdef __DQUOT_PARANOIA
560 if (!atomic_read(&dquot->dq_count)) {
561 printk("VFS: dqput: trying to free free dquot\n");
562 printk("VFS: device %s, dquot of %s %d\n",
564 quotatypes[dquot->dq_type],
570 spin_lock(&dq_list_lock);
572 spin_unlock(&dq_list_lock);
574 spin_lock(&dq_list_lock);
575 if (atomic_read(&dquot->dq_count) > 1) {
576 /* We have more than one user... nothing to do */
577 atomic_dec(&dquot->dq_count);
578 /* Releasing dquot during quotaoff phase? */
579 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
580 atomic_read(&dquot->dq_count) == 1)
581 wake_up(&dquot->dq_wait_unused);
582 spin_unlock(&dq_list_lock);
585 /* Need to release dquot? */
586 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
587 spin_unlock(&dq_list_lock);
588 /* Commit dquot before releasing */
589 dquot->dq_sb->dq_op->write_dquot(dquot);
592 /* Clear flag in case dquot was inactive (something bad happened) */
593 clear_dquot_dirty(dquot);
594 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
595 spin_unlock(&dq_list_lock);
596 dquot->dq_sb->dq_op->release_dquot(dquot);
599 atomic_dec(&dquot->dq_count);
600 #ifdef __DQUOT_PARANOIA
602 BUG_ON(!list_empty(&dquot->dq_free));
604 put_dquot_last(dquot);
605 spin_unlock(&dq_list_lock);
608 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
612 dquot = kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
616 mutex_init(&dquot->dq_lock);
617 INIT_LIST_HEAD(&dquot->dq_free);
618 INIT_LIST_HEAD(&dquot->dq_inuse);
619 INIT_HLIST_NODE(&dquot->dq_hash);
620 INIT_LIST_HEAD(&dquot->dq_dirty);
621 init_waitqueue_head(&dquot->dq_wait_unused);
623 dquot->dq_type = type;
624 atomic_set(&dquot->dq_count, 1);
630 * Get reference to dquot
631 * MUST be called with either dqptr_sem or dqonoff_mutex held
633 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
635 unsigned int hashent = hashfn(sb, id, type);
636 struct dquot *dquot, *empty = NODQUOT;
638 if (!sb_has_quota_enabled(sb, type))
641 spin_lock(&dq_list_lock);
642 if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
643 if (empty == NODQUOT) {
644 spin_unlock(&dq_list_lock);
645 if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
646 schedule(); /* Try to wait for a moment... */
651 /* all dquots go on the inuse_list */
653 /* hash it first so it can be found */
654 insert_dquot_hash(dquot);
656 spin_unlock(&dq_list_lock);
658 if (!atomic_read(&dquot->dq_count))
659 remove_free_dquot(dquot);
660 atomic_inc(&dquot->dq_count);
661 dqstats.cache_hits++;
663 spin_unlock(&dq_list_lock);
665 kmem_cache_free(dquot_cachep, empty);
667 /* Wait for dq_lock - after this we know that either dquot_release() is already
668 * finished or it will be canceled due to dq_count > 1 test */
669 wait_on_dquot(dquot);
670 /* Read the dquot and instantiate it (everything done only if needed) */
671 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
675 #ifdef __DQUOT_PARANOIA
676 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
682 static int dqinit_needed(struct inode *inode, int type)
686 if (IS_NOQUOTA(inode))
689 return inode->i_dquot[type] == NODQUOT;
690 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
691 if (inode->i_dquot[cnt] == NODQUOT)
696 /* This routine is guarded by dqonoff_mutex mutex */
697 static void add_dquot_ref(struct super_block *sb, int type)
702 spin_lock(&inode_lock);
703 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
704 if (!atomic_read(&inode->i_writecount))
706 if (!dqinit_needed(inode, type))
708 if (inode->i_state & (I_FREEING|I_WILL_FREE))
712 spin_unlock(&inode_lock);
714 sb->dq_op->initialize(inode, type);
716 /* As we may have blocked we had better restart... */
719 spin_unlock(&inode_lock);
722 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
723 static inline int dqput_blocks(struct dquot *dquot)
725 if (atomic_read(&dquot->dq_count) <= 1)
730 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
731 /* We can't race with anybody because we hold dqptr_sem for writing... */
732 static int remove_inode_dquot_ref(struct inode *inode, int type,
733 struct list_head *tofree_head)
735 struct dquot *dquot = inode->i_dquot[type];
737 inode->i_dquot[type] = NODQUOT;
738 if (dquot != NODQUOT) {
739 if (dqput_blocks(dquot)) {
740 #ifdef __DQUOT_PARANOIA
741 if (atomic_read(&dquot->dq_count) != 1)
742 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
744 spin_lock(&dq_list_lock);
745 list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
746 spin_unlock(&dq_list_lock);
750 dqput(dquot); /* We have guaranteed we won't block */
755 /* Free list of dquots - called from inode.c */
756 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
757 static void put_dquot_list(struct list_head *tofree_head)
759 struct list_head *act_head;
762 act_head = tofree_head->next;
763 /* So now we have dquots on the list... Just free them */
764 while (act_head != tofree_head) {
765 dquot = list_entry(act_head, struct dquot, dq_free);
766 act_head = act_head->next;
767 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
772 static void remove_dquot_ref(struct super_block *sb, int type,
773 struct list_head *tofree_head)
777 spin_lock(&inode_lock);
778 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
779 if (!IS_NOQUOTA(inode))
780 remove_inode_dquot_ref(inode, type, tofree_head);
782 spin_unlock(&inode_lock);
785 /* Gather all references from inodes and drop them */
786 static void drop_dquot_ref(struct super_block *sb, int type)
788 LIST_HEAD(tofree_head);
791 down_write(&sb_dqopt(sb)->dqptr_sem);
792 remove_dquot_ref(sb, type, &tofree_head);
793 up_write(&sb_dqopt(sb)->dqptr_sem);
794 put_dquot_list(&tofree_head);
798 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
800 dquot->dq_dqb.dqb_curinodes += number;
803 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
805 dquot->dq_dqb.dqb_curspace += number;
808 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
810 if (dquot->dq_dqb.dqb_curinodes > number)
811 dquot->dq_dqb.dqb_curinodes -= number;
813 dquot->dq_dqb.dqb_curinodes = 0;
814 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
815 dquot->dq_dqb.dqb_itime = (time_t) 0;
816 clear_bit(DQ_INODES_B, &dquot->dq_flags);
819 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
821 if (dquot->dq_dqb.dqb_curspace > number)
822 dquot->dq_dqb.dqb_curspace -= number;
824 dquot->dq_dqb.dqb_curspace = 0;
825 if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
826 dquot->dq_dqb.dqb_btime = (time_t) 0;
827 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
830 #ifdef CONFIG_PRINT_QUOTA_WARNING
831 static int flag_print_warnings = 1;
833 static inline int need_print_warning(struct dquot *dquot)
835 if (!flag_print_warnings)
838 switch (dquot->dq_type) {
840 return current->fsuid == dquot->dq_id;
842 return in_group_p(dquot->dq_id);
847 /* Print warning to user which exceeded quota */
848 static void print_warning(struct dquot *dquot, const char warntype)
851 struct tty_struct *tty;
852 int flag = (warntype == QUOTA_NL_BHARDWARN ||
853 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
854 ((warntype == QUOTA_NL_IHARDWARN ||
855 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
857 if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
860 mutex_lock(&tty_mutex);
861 tty = get_current_tty();
864 tty_write_message(tty, dquot->dq_sb->s_id);
865 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
866 tty_write_message(tty, ": warning, ");
868 tty_write_message(tty, ": write failed, ");
869 tty_write_message(tty, quotatypes[dquot->dq_type]);
871 case QUOTA_NL_IHARDWARN:
872 msg = " file limit reached.\r\n";
874 case QUOTA_NL_ISOFTLONGWARN:
875 msg = " file quota exceeded too long.\r\n";
877 case QUOTA_NL_ISOFTWARN:
878 msg = " file quota exceeded.\r\n";
880 case QUOTA_NL_BHARDWARN:
881 msg = " block limit reached.\r\n";
883 case QUOTA_NL_BSOFTLONGWARN:
884 msg = " block quota exceeded too long.\r\n";
886 case QUOTA_NL_BSOFTWARN:
887 msg = " block quota exceeded.\r\n";
890 tty_write_message(tty, msg);
892 mutex_unlock(&tty_mutex);
896 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
898 /* Netlink family structure for quota */
899 static struct genl_family quota_genl_family = {
900 .id = GENL_ID_GENERATE,
904 .maxattr = QUOTA_NL_A_MAX,
907 /* Send warning to userspace about user which exceeded quota */
908 static void send_warning(const struct dquot *dquot, const char warntype)
914 int msg_size = 4 * nla_total_size(sizeof(u32)) +
915 2 * nla_total_size(sizeof(u64));
917 /* We have to allocate using GFP_NOFS as we are called from a
918 * filesystem performing write and thus further recursion into
919 * the fs to free some data could cause deadlocks. */
920 skb = genlmsg_new(msg_size, GFP_NOFS);
923 "VFS: Not enough memory to send quota warning.\n");
926 msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
927 "a_genl_family, 0, QUOTA_NL_C_WARNING);
930 "VFS: Cannot store netlink header in quota warning.\n");
933 ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
936 ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
939 ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
942 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
943 MAJOR(dquot->dq_sb->s_dev));
946 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
947 MINOR(dquot->dq_sb->s_dev));
950 ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
953 genlmsg_end(skb, msg_head);
955 ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
956 if (ret < 0 && ret != -ESRCH)
958 "VFS: Failed to send notification message: %d\n", ret);
961 printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
967 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
971 for (i = 0; i < MAXQUOTAS; i++)
972 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN) {
973 #ifdef CONFIG_PRINT_QUOTA_WARNING
974 print_warning(dquots[i], warntype[i]);
976 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
977 send_warning(dquots[i], warntype[i]);
982 static inline char ignore_hardlimit(struct dquot *dquot)
984 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
986 return capable(CAP_SYS_RESOURCE) &&
987 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
990 /* needs dq_data_lock */
991 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
993 *warntype = QUOTA_NL_NOWARN;
994 if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
997 if (dquot->dq_dqb.dqb_ihardlimit &&
998 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
999 !ignore_hardlimit(dquot)) {
1000 *warntype = QUOTA_NL_IHARDWARN;
1004 if (dquot->dq_dqb.dqb_isoftlimit &&
1005 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1006 dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1007 !ignore_hardlimit(dquot)) {
1008 *warntype = QUOTA_NL_ISOFTLONGWARN;
1012 if (dquot->dq_dqb.dqb_isoftlimit &&
1013 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1014 dquot->dq_dqb.dqb_itime == 0) {
1015 *warntype = QUOTA_NL_ISOFTWARN;
1016 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1022 /* needs dq_data_lock */
1023 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1025 *warntype = QUOTA_NL_NOWARN;
1026 if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1029 if (dquot->dq_dqb.dqb_bhardlimit &&
1030 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1031 !ignore_hardlimit(dquot)) {
1033 *warntype = QUOTA_NL_BHARDWARN;
1037 if (dquot->dq_dqb.dqb_bsoftlimit &&
1038 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1039 dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1040 !ignore_hardlimit(dquot)) {
1042 *warntype = QUOTA_NL_BSOFTLONGWARN;
1046 if (dquot->dq_dqb.dqb_bsoftlimit &&
1047 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1048 dquot->dq_dqb.dqb_btime == 0) {
1050 *warntype = QUOTA_NL_BSOFTWARN;
1051 dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1055 * We don't allow preallocation to exceed softlimit so exceeding will
1065 * Initialize quota pointers in inode
1066 * Transaction must be started at entry
1068 int dquot_initialize(struct inode *inode, int type)
1070 unsigned int id = 0;
1073 /* First test before acquiring mutex - solves deadlocks when we
1074 * re-enter the quota code and are already holding the mutex */
1075 if (IS_NOQUOTA(inode))
1077 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1078 /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1079 if (IS_NOQUOTA(inode))
1081 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1082 if (type != -1 && cnt != type)
1084 if (inode->i_dquot[cnt] == NODQUOT) {
1093 inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1097 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1102 * Release all quotas referenced by inode
1103 * Transaction must be started at an entry
1105 int dquot_drop(struct inode *inode)
1109 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1110 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1111 if (inode->i_dquot[cnt] != NODQUOT) {
1112 dqput(inode->i_dquot[cnt]);
1113 inode->i_dquot[cnt] = NODQUOT;
1116 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1121 * Following four functions update i_blocks+i_bytes fields and
1122 * quota information (together with appropriate checks)
1123 * NOTE: We absolutely rely on the fact that caller dirties
1124 * the inode (usually macros in quotaops.h care about this) and
1125 * holds a handle for the current transaction so that dquot write and
1126 * inode write go into the same transaction.
1130 * This operation can block, but only after everything is updated
1132 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1134 int cnt, ret = NO_QUOTA;
1135 char warntype[MAXQUOTAS];
1137 /* First test before acquiring mutex - solves deadlocks when we
1138 * re-enter the quota code and are already holding the mutex */
1139 if (IS_NOQUOTA(inode)) {
1141 inode_add_bytes(inode, number);
1144 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1145 warntype[cnt] = QUOTA_NL_NOWARN;
1147 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1148 if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */
1149 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1152 spin_lock(&dq_data_lock);
1153 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1154 if (inode->i_dquot[cnt] == NODQUOT)
1156 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1159 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1160 if (inode->i_dquot[cnt] == NODQUOT)
1162 dquot_incr_space(inode->i_dquot[cnt], number);
1164 inode_add_bytes(inode, number);
1167 spin_unlock(&dq_data_lock);
1168 if (ret == QUOTA_OK)
1169 /* Dirtify all the dquots - this can block when journalling */
1170 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1171 if (inode->i_dquot[cnt])
1172 mark_dquot_dirty(inode->i_dquot[cnt]);
1173 flush_warnings(inode->i_dquot, warntype);
1174 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1179 * This operation can block, but only after everything is updated
1181 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1183 int cnt, ret = NO_QUOTA;
1184 char warntype[MAXQUOTAS];
1186 /* First test before acquiring mutex - solves deadlocks when we
1187 * re-enter the quota code and are already holding the mutex */
1188 if (IS_NOQUOTA(inode))
1190 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1191 warntype[cnt] = QUOTA_NL_NOWARN;
1192 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1193 if (IS_NOQUOTA(inode)) {
1194 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1197 spin_lock(&dq_data_lock);
1198 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1199 if (inode->i_dquot[cnt] == NODQUOT)
1201 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1205 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1206 if (inode->i_dquot[cnt] == NODQUOT)
1208 dquot_incr_inodes(inode->i_dquot[cnt], number);
1212 spin_unlock(&dq_data_lock);
1213 if (ret == QUOTA_OK)
1214 /* Dirtify all the dquots - this can block when journalling */
1215 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1216 if (inode->i_dquot[cnt])
1217 mark_dquot_dirty(inode->i_dquot[cnt]);
1218 flush_warnings(inode->i_dquot, warntype);
1219 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1224 * This operation can block, but only after everything is updated
1226 int dquot_free_space(struct inode *inode, qsize_t number)
1230 /* First test before acquiring mutex - solves deadlocks when we
1231 * re-enter the quota code and are already holding the mutex */
1232 if (IS_NOQUOTA(inode)) {
1234 inode_sub_bytes(inode, number);
1237 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1238 /* Now recheck reliably when holding dqptr_sem */
1239 if (IS_NOQUOTA(inode)) {
1240 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1243 spin_lock(&dq_data_lock);
1244 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1245 if (inode->i_dquot[cnt] == NODQUOT)
1247 dquot_decr_space(inode->i_dquot[cnt], number);
1249 inode_sub_bytes(inode, number);
1250 spin_unlock(&dq_data_lock);
1251 /* Dirtify all the dquots - this can block when journalling */
1252 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1253 if (inode->i_dquot[cnt])
1254 mark_dquot_dirty(inode->i_dquot[cnt]);
1255 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1260 * This operation can block, but only after everything is updated
1262 int dquot_free_inode(const struct inode *inode, unsigned long number)
1266 /* First test before acquiring mutex - solves deadlocks when we
1267 * re-enter the quota code and are already holding the mutex */
1268 if (IS_NOQUOTA(inode))
1270 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1271 /* Now recheck reliably when holding dqptr_sem */
1272 if (IS_NOQUOTA(inode)) {
1273 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1276 spin_lock(&dq_data_lock);
1277 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1278 if (inode->i_dquot[cnt] == NODQUOT)
1280 dquot_decr_inodes(inode->i_dquot[cnt], number);
1282 spin_unlock(&dq_data_lock);
1283 /* Dirtify all the dquots - this can block when journalling */
1284 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1285 if (inode->i_dquot[cnt])
1286 mark_dquot_dirty(inode->i_dquot[cnt]);
1287 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1292 * Transfer the number of inode and blocks from one diskquota to an other.
1294 * This operation can block, but only after everything is updated
1295 * A transaction must be started when entering this function.
1297 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1300 struct dquot *transfer_from[MAXQUOTAS];
1301 struct dquot *transfer_to[MAXQUOTAS];
1302 int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1303 chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1304 char warntype[MAXQUOTAS];
1306 /* First test before acquiring mutex - solves deadlocks when we
1307 * re-enter the quota code and are already holding the mutex */
1308 if (IS_NOQUOTA(inode))
1310 /* Clear the arrays */
1311 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1312 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1313 warntype[cnt] = QUOTA_NL_NOWARN;
1315 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1316 /* Now recheck reliably when holding dqptr_sem */
1317 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1318 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1321 /* First build the transfer_to list - here we can block on
1322 * reading/instantiating of dquots. We know that the transaction for
1323 * us was already started so we don't violate lock ranking here */
1324 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1329 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1334 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1338 spin_lock(&dq_data_lock);
1339 space = inode_get_bytes(inode);
1340 /* Build the transfer_from list and check the limits */
1341 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1342 if (transfer_to[cnt] == NODQUOT)
1344 transfer_from[cnt] = inode->i_dquot[cnt];
1345 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1346 check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1351 * Finally perform the needed transfer from transfer_from to transfer_to
1353 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1355 * Skip changes for same uid or gid or for turned off quota-type.
1357 if (transfer_to[cnt] == NODQUOT)
1360 /* Due to IO error we might not have transfer_from[] structure */
1361 if (transfer_from[cnt]) {
1362 dquot_decr_inodes(transfer_from[cnt], 1);
1363 dquot_decr_space(transfer_from[cnt], space);
1366 dquot_incr_inodes(transfer_to[cnt], 1);
1367 dquot_incr_space(transfer_to[cnt], space);
1369 inode->i_dquot[cnt] = transfer_to[cnt];
1373 spin_unlock(&dq_data_lock);
1374 /* Dirtify all the dquots - this can block when journalling */
1375 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1376 if (transfer_from[cnt])
1377 mark_dquot_dirty(transfer_from[cnt]);
1378 if (transfer_to[cnt])
1379 mark_dquot_dirty(transfer_to[cnt]);
1381 flush_warnings(transfer_to, warntype);
1383 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1384 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1385 dqput(transfer_from[cnt]);
1386 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1387 dqput(transfer_to[cnt]);
1389 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1394 * Write info of quota file to disk
1396 int dquot_commit_info(struct super_block *sb, int type)
1399 struct quota_info *dqopt = sb_dqopt(sb);
1401 mutex_lock(&dqopt->dqio_mutex);
1402 ret = dqopt->ops[type]->write_file_info(sb, type);
1403 mutex_unlock(&dqopt->dqio_mutex);
1408 * Definitions of diskquota operations.
1410 struct dquot_operations dquot_operations = {
1411 .initialize = dquot_initialize,
1413 .alloc_space = dquot_alloc_space,
1414 .alloc_inode = dquot_alloc_inode,
1415 .free_space = dquot_free_space,
1416 .free_inode = dquot_free_inode,
1417 .transfer = dquot_transfer,
1418 .write_dquot = dquot_commit,
1419 .acquire_dquot = dquot_acquire,
1420 .release_dquot = dquot_release,
1421 .mark_dirty = dquot_mark_dquot_dirty,
1422 .write_info = dquot_commit_info
1425 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1429 dqopt->flags |= DQUOT_USR_ENABLED;
1432 dqopt->flags |= DQUOT_GRP_ENABLED;
1437 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1441 dqopt->flags &= ~DQUOT_USR_ENABLED;
1444 dqopt->flags &= ~DQUOT_GRP_ENABLED;
1450 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1452 int vfs_quota_off(struct super_block *sb, int type)
1455 struct quota_info *dqopt = sb_dqopt(sb);
1456 struct inode *toputinode[MAXQUOTAS];
1458 /* We need to serialize quota_off() for device */
1459 mutex_lock(&dqopt->dqonoff_mutex);
1460 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1461 toputinode[cnt] = NULL;
1462 if (type != -1 && cnt != type)
1464 if (!sb_has_quota_enabled(sb, cnt))
1466 reset_enable_flags(dqopt, cnt);
1468 /* Note: these are blocking operations */
1469 drop_dquot_ref(sb, cnt);
1470 invalidate_dquots(sb, cnt);
1472 * Now all dquots should be invalidated, all writes done so we should be only
1473 * users of the info. No locks needed.
1475 if (info_dirty(&dqopt->info[cnt]))
1476 sb->dq_op->write_info(sb, cnt);
1477 if (dqopt->ops[cnt]->free_file_info)
1478 dqopt->ops[cnt]->free_file_info(sb, cnt);
1479 put_quota_format(dqopt->info[cnt].dqi_format);
1481 toputinode[cnt] = dqopt->files[cnt];
1482 dqopt->files[cnt] = NULL;
1483 dqopt->info[cnt].dqi_flags = 0;
1484 dqopt->info[cnt].dqi_igrace = 0;
1485 dqopt->info[cnt].dqi_bgrace = 0;
1486 dqopt->ops[cnt] = NULL;
1488 mutex_unlock(&dqopt->dqonoff_mutex);
1489 /* Sync the superblock so that buffers with quota data are written to
1490 * disk (and so userspace sees correct data afterwards). */
1491 if (sb->s_op->sync_fs)
1492 sb->s_op->sync_fs(sb, 1);
1493 sync_blockdev(sb->s_bdev);
1494 /* Now the quota files are just ordinary files and we can set the
1495 * inode flags back. Moreover we discard the pagecache so that
1496 * userspace sees the writes we did bypassing the pagecache. We
1497 * must also discard the blockdev buffers so that we see the
1498 * changes done by userspace on the next quotaon() */
1499 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1500 if (toputinode[cnt]) {
1501 mutex_lock(&dqopt->dqonoff_mutex);
1502 /* If quota was reenabled in the meantime, we have
1504 if (!sb_has_quota_enabled(sb, cnt)) {
1505 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1506 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1507 S_NOATIME | S_NOQUOTA);
1508 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1509 mutex_unlock(&toputinode[cnt]->i_mutex);
1510 mark_inode_dirty(toputinode[cnt]);
1511 iput(toputinode[cnt]);
1513 mutex_unlock(&dqopt->dqonoff_mutex);
1516 invalidate_bdev(sb->s_bdev);
1521 * Turn quotas on on a device
1524 /* Helper function when we already have the inode */
1525 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1527 struct quota_format_type *fmt = find_quota_format(format_id);
1528 struct super_block *sb = inode->i_sb;
1529 struct quota_info *dqopt = sb_dqopt(sb);
1535 if (!S_ISREG(inode->i_mode)) {
1539 if (IS_RDONLY(inode)) {
1543 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1548 /* As we bypass the pagecache we must now flush the inode so that
1549 * we see all the changes from userspace... */
1550 write_inode_now(inode, 1);
1551 /* And now flush the block cache so that kernel sees the changes */
1552 invalidate_bdev(sb->s_bdev);
1553 mutex_lock(&inode->i_mutex);
1554 mutex_lock(&dqopt->dqonoff_mutex);
1555 if (sb_has_quota_enabled(sb, type)) {
1559 /* We don't want quota and atime on quota files (deadlocks possible)
1560 * Also nobody should write to the file - we use special IO operations
1561 * which ignore the immutable bit. */
1562 down_write(&dqopt->dqptr_sem);
1563 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1564 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1565 up_write(&dqopt->dqptr_sem);
1566 sb->dq_op->drop(inode);
1569 dqopt->files[type] = igrab(inode);
1570 if (!dqopt->files[type])
1573 if (!fmt->qf_ops->check_quota_file(sb, type))
1576 dqopt->ops[type] = fmt->qf_ops;
1577 dqopt->info[type].dqi_format = fmt;
1578 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1579 mutex_lock(&dqopt->dqio_mutex);
1580 if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1581 mutex_unlock(&dqopt->dqio_mutex);
1584 mutex_unlock(&dqopt->dqio_mutex);
1585 mutex_unlock(&inode->i_mutex);
1586 set_enable_flags(dqopt, type);
1588 add_dquot_ref(sb, type);
1589 mutex_unlock(&dqopt->dqonoff_mutex);
1594 dqopt->files[type] = NULL;
1597 mutex_unlock(&dqopt->dqonoff_mutex);
1598 if (oldflags != -1) {
1599 down_write(&dqopt->dqptr_sem);
1600 /* Set the flags back (in the case of accidental quotaon()
1601 * on a wrong file we don't want to mess up the flags) */
1602 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1603 inode->i_flags |= oldflags;
1604 up_write(&dqopt->dqptr_sem);
1606 mutex_unlock(&inode->i_mutex);
1608 put_quota_format(fmt);
1613 /* Actual function called from quotactl() */
1614 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1616 struct nameidata nd;
1619 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1622 error = security_quota_on(nd.dentry);
1625 /* Quota file not on the same filesystem? */
1626 if (nd.mnt->mnt_sb != sb)
1629 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1636 * This function is used when filesystem needs to initialize quotas
1637 * during mount time.
1639 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1640 int format_id, int type)
1642 struct dentry *dentry;
1645 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1647 return PTR_ERR(dentry);
1649 if (!dentry->d_inode) {
1654 error = security_quota_on(dentry);
1656 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1663 /* Generic routine for getting common part of quota structure */
1664 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1666 struct mem_dqblk *dm = &dquot->dq_dqb;
1668 spin_lock(&dq_data_lock);
1669 di->dqb_bhardlimit = dm->dqb_bhardlimit;
1670 di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1671 di->dqb_curspace = dm->dqb_curspace;
1672 di->dqb_ihardlimit = dm->dqb_ihardlimit;
1673 di->dqb_isoftlimit = dm->dqb_isoftlimit;
1674 di->dqb_curinodes = dm->dqb_curinodes;
1675 di->dqb_btime = dm->dqb_btime;
1676 di->dqb_itime = dm->dqb_itime;
1677 di->dqb_valid = QIF_ALL;
1678 spin_unlock(&dq_data_lock);
1681 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1683 struct dquot *dquot;
1685 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1686 if (!(dquot = dqget(sb, id, type))) {
1687 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1690 do_get_dqblk(dquot, di);
1692 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1696 /* Generic routine for setting common part of quota structure */
1697 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1699 struct mem_dqblk *dm = &dquot->dq_dqb;
1700 int check_blim = 0, check_ilim = 0;
1702 spin_lock(&dq_data_lock);
1703 if (di->dqb_valid & QIF_SPACE) {
1704 dm->dqb_curspace = di->dqb_curspace;
1707 if (di->dqb_valid & QIF_BLIMITS) {
1708 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1709 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1712 if (di->dqb_valid & QIF_INODES) {
1713 dm->dqb_curinodes = di->dqb_curinodes;
1716 if (di->dqb_valid & QIF_ILIMITS) {
1717 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1718 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1721 if (di->dqb_valid & QIF_BTIME)
1722 dm->dqb_btime = di->dqb_btime;
1723 if (di->dqb_valid & QIF_ITIME)
1724 dm->dqb_itime = di->dqb_itime;
1727 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1729 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1731 else if (!(di->dqb_valid & QIF_BTIME)) /* Set grace only if user hasn't provided his own... */
1732 dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1735 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1737 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1739 else if (!(di->dqb_valid & QIF_ITIME)) /* Set grace only if user hasn't provided his own... */
1740 dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1742 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1743 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1745 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1746 spin_unlock(&dq_data_lock);
1747 mark_dquot_dirty(dquot);
1750 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1752 struct dquot *dquot;
1754 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1755 if (!(dquot = dqget(sb, id, type))) {
1756 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1759 do_set_dqblk(dquot, di);
1761 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1765 /* Generic routine for getting common part of quota file information */
1766 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1768 struct mem_dqinfo *mi;
1770 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1771 if (!sb_has_quota_enabled(sb, type)) {
1772 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1775 mi = sb_dqopt(sb)->info + type;
1776 spin_lock(&dq_data_lock);
1777 ii->dqi_bgrace = mi->dqi_bgrace;
1778 ii->dqi_igrace = mi->dqi_igrace;
1779 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1780 ii->dqi_valid = IIF_ALL;
1781 spin_unlock(&dq_data_lock);
1782 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1786 /* Generic routine for setting common part of quota file information */
1787 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1789 struct mem_dqinfo *mi;
1791 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1792 if (!sb_has_quota_enabled(sb, type)) {
1793 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1796 mi = sb_dqopt(sb)->info + type;
1797 spin_lock(&dq_data_lock);
1798 if (ii->dqi_valid & IIF_BGRACE)
1799 mi->dqi_bgrace = ii->dqi_bgrace;
1800 if (ii->dqi_valid & IIF_IGRACE)
1801 mi->dqi_igrace = ii->dqi_igrace;
1802 if (ii->dqi_valid & IIF_FLAGS)
1803 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1804 spin_unlock(&dq_data_lock);
1805 mark_info_dirty(sb, type);
1806 /* Force write to disk */
1807 sb->dq_op->write_info(sb, type);
1808 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1812 struct quotactl_ops vfs_quotactl_ops = {
1813 .quota_on = vfs_quota_on,
1814 .quota_off = vfs_quota_off,
1815 .quota_sync = vfs_quota_sync,
1816 .get_info = vfs_get_dqinfo,
1817 .set_info = vfs_set_dqinfo,
1818 .get_dqblk = vfs_get_dqblk,
1819 .set_dqblk = vfs_set_dqblk
1822 static ctl_table fs_dqstats_table[] = {
1824 .ctl_name = FS_DQ_LOOKUPS,
1825 .procname = "lookups",
1826 .data = &dqstats.lookups,
1827 .maxlen = sizeof(int),
1829 .proc_handler = &proc_dointvec,
1832 .ctl_name = FS_DQ_DROPS,
1833 .procname = "drops",
1834 .data = &dqstats.drops,
1835 .maxlen = sizeof(int),
1837 .proc_handler = &proc_dointvec,
1840 .ctl_name = FS_DQ_READS,
1841 .procname = "reads",
1842 .data = &dqstats.reads,
1843 .maxlen = sizeof(int),
1845 .proc_handler = &proc_dointvec,
1848 .ctl_name = FS_DQ_WRITES,
1849 .procname = "writes",
1850 .data = &dqstats.writes,
1851 .maxlen = sizeof(int),
1853 .proc_handler = &proc_dointvec,
1856 .ctl_name = FS_DQ_CACHE_HITS,
1857 .procname = "cache_hits",
1858 .data = &dqstats.cache_hits,
1859 .maxlen = sizeof(int),
1861 .proc_handler = &proc_dointvec,
1864 .ctl_name = FS_DQ_ALLOCATED,
1865 .procname = "allocated_dquots",
1866 .data = &dqstats.allocated_dquots,
1867 .maxlen = sizeof(int),
1869 .proc_handler = &proc_dointvec,
1872 .ctl_name = FS_DQ_FREE,
1873 .procname = "free_dquots",
1874 .data = &dqstats.free_dquots,
1875 .maxlen = sizeof(int),
1877 .proc_handler = &proc_dointvec,
1880 .ctl_name = FS_DQ_SYNCS,
1881 .procname = "syncs",
1882 .data = &dqstats.syncs,
1883 .maxlen = sizeof(int),
1885 .proc_handler = &proc_dointvec,
1887 #ifdef CONFIG_PRINT_QUOTA_WARNING
1889 .ctl_name = FS_DQ_WARNINGS,
1890 .procname = "warnings",
1891 .data = &flag_print_warnings,
1892 .maxlen = sizeof(int),
1894 .proc_handler = &proc_dointvec,
1900 static ctl_table fs_table[] = {
1902 .ctl_name = FS_DQSTATS,
1903 .procname = "quota",
1905 .child = fs_dqstats_table,
1910 static ctl_table sys_table[] = {
1920 static int __init dquot_init(void)
1923 unsigned long nr_hash, order;
1925 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1927 register_sysctl_table(sys_table);
1929 dquot_cachep = kmem_cache_create("dquot",
1930 sizeof(struct dquot), sizeof(unsigned long) * 4,
1931 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1932 SLAB_MEM_SPREAD|SLAB_PANIC),
1936 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1938 panic("Cannot create dquot hash table");
1940 /* Find power-of-two hlist_heads which can fit into allocation */
1941 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1945 } while (nr_hash >> dq_hash_bits);
1948 nr_hash = 1UL << dq_hash_bits;
1949 dq_hash_mask = nr_hash - 1;
1950 for (i = 0; i < nr_hash; i++)
1951 INIT_HLIST_HEAD(dquot_hash + i);
1953 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1954 nr_hash, order, (PAGE_SIZE << order));
1956 register_shrinker(&dqcache_shrinker);
1958 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1959 if (genl_register_family("a_genl_family) != 0)
1960 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
1965 module_init(dquot_init);
1967 EXPORT_SYMBOL(register_quota_format);
1968 EXPORT_SYMBOL(unregister_quota_format);
1969 EXPORT_SYMBOL(dqstats);
1970 EXPORT_SYMBOL(dq_data_lock);
1971 EXPORT_SYMBOL(vfs_quota_on);
1972 EXPORT_SYMBOL(vfs_quota_on_mount);
1973 EXPORT_SYMBOL(vfs_quota_off);
1974 EXPORT_SYMBOL(vfs_quota_sync);
1975 EXPORT_SYMBOL(vfs_get_dqinfo);
1976 EXPORT_SYMBOL(vfs_set_dqinfo);
1977 EXPORT_SYMBOL(vfs_get_dqblk);
1978 EXPORT_SYMBOL(vfs_set_dqblk);
1979 EXPORT_SYMBOL(dquot_commit);
1980 EXPORT_SYMBOL(dquot_commit_info);
1981 EXPORT_SYMBOL(dquot_acquire);
1982 EXPORT_SYMBOL(dquot_release);
1983 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1984 EXPORT_SYMBOL(dquot_initialize);
1985 EXPORT_SYMBOL(dquot_drop);
1986 EXPORT_SYMBOL(dquot_alloc_space);
1987 EXPORT_SYMBOL(dquot_alloc_inode);
1988 EXPORT_SYMBOL(dquot_free_space);
1989 EXPORT_SYMBOL(dquot_free_inode);
1990 EXPORT_SYMBOL(dquot_transfer);