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 static inline int dquot_dirty(struct dquot *dquot)
294 return test_bit(DQ_MOD_B, &dquot->dq_flags);
297 static inline int mark_dquot_dirty(struct dquot *dquot)
299 return dquot->dq_sb->dq_op->mark_dirty(dquot);
302 int dquot_mark_dquot_dirty(struct dquot *dquot)
304 spin_lock(&dq_list_lock);
305 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
306 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
307 info[dquot->dq_type].dqi_dirty_list);
308 spin_unlock(&dq_list_lock);
312 /* This function needs dq_list_lock */
313 static inline int clear_dquot_dirty(struct dquot *dquot)
315 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
317 list_del_init(&dquot->dq_dirty);
321 void mark_info_dirty(struct super_block *sb, int type)
323 set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
325 EXPORT_SYMBOL(mark_info_dirty);
328 * Read dquot from disk and alloc space for it
331 int dquot_acquire(struct dquot *dquot)
333 int ret = 0, ret2 = 0;
334 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
336 mutex_lock(&dquot->dq_lock);
337 mutex_lock(&dqopt->dqio_mutex);
338 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
339 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
342 set_bit(DQ_READ_B, &dquot->dq_flags);
343 /* Instantiate dquot if needed */
344 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
345 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
346 /* Write the info if needed */
347 if (info_dirty(&dqopt->info[dquot->dq_type]))
348 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
356 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
358 mutex_unlock(&dqopt->dqio_mutex);
359 mutex_unlock(&dquot->dq_lock);
364 * Write dquot to disk
366 int dquot_commit(struct dquot *dquot)
368 int ret = 0, ret2 = 0;
369 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
371 mutex_lock(&dqopt->dqio_mutex);
372 spin_lock(&dq_list_lock);
373 if (!clear_dquot_dirty(dquot)) {
374 spin_unlock(&dq_list_lock);
377 spin_unlock(&dq_list_lock);
378 /* Inactive dquot can be only if there was error during read/init
379 * => we have better not writing it */
380 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
381 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
382 if (info_dirty(&dqopt->info[dquot->dq_type]))
383 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
388 mutex_unlock(&dqopt->dqio_mutex);
395 int dquot_release(struct dquot *dquot)
397 int ret = 0, ret2 = 0;
398 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
400 mutex_lock(&dquot->dq_lock);
401 /* Check whether we are not racing with some other dqget() */
402 if (atomic_read(&dquot->dq_count) > 1)
404 mutex_lock(&dqopt->dqio_mutex);
405 if (dqopt->ops[dquot->dq_type]->release_dqblk) {
406 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
408 if (info_dirty(&dqopt->info[dquot->dq_type]))
409 ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
413 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
414 mutex_unlock(&dqopt->dqio_mutex);
416 mutex_unlock(&dquot->dq_lock);
420 /* Invalidate all dquots on the list. Note that this function is called after
421 * quota is disabled and pointers from inodes removed so there cannot be new
422 * quota users. There can still be some users of quotas due to inodes being
423 * just deleted or pruned by prune_icache() (those are not attached to any
424 * list). We have to wait for such users.
426 static void invalidate_dquots(struct super_block *sb, int type)
428 struct dquot *dquot, *tmp;
431 spin_lock(&dq_list_lock);
432 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
433 if (dquot->dq_sb != sb)
435 if (dquot->dq_type != type)
437 /* Wait for dquot users */
438 if (atomic_read(&dquot->dq_count)) {
441 atomic_inc(&dquot->dq_count);
442 prepare_to_wait(&dquot->dq_wait_unused, &wait,
443 TASK_UNINTERRUPTIBLE);
444 spin_unlock(&dq_list_lock);
445 /* Once dqput() wakes us up, we know it's time to free
447 * IMPORTANT: we rely on the fact that there is always
448 * at most one process waiting for dquot to free.
449 * Otherwise dq_count would be > 1 and we would never
452 if (atomic_read(&dquot->dq_count) > 1)
454 finish_wait(&dquot->dq_wait_unused, &wait);
456 /* At this moment dquot() need not exist (it could be
457 * reclaimed by prune_dqcache(). Hence we must
462 * Quota now has no users and it has been written on last
465 remove_dquot_hash(dquot);
466 remove_free_dquot(dquot);
468 kmem_cache_free(dquot_cachep, dquot);
470 spin_unlock(&dq_list_lock);
473 int vfs_quota_sync(struct super_block *sb, int type)
475 struct list_head *dirty;
477 struct quota_info *dqopt = sb_dqopt(sb);
480 mutex_lock(&dqopt->dqonoff_mutex);
481 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
482 if (type != -1 && cnt != type)
484 if (!sb_has_quota_enabled(sb, cnt))
486 spin_lock(&dq_list_lock);
487 dirty = &dqopt->info[cnt].dqi_dirty_list;
488 while (!list_empty(dirty)) {
489 dquot = list_first_entry(dirty, struct dquot, dq_dirty);
490 /* Dirty and inactive can be only bad dquot... */
491 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
492 clear_dquot_dirty(dquot);
495 /* Now we have active dquot from which someone is
496 * holding reference so we can safely just increase
498 atomic_inc(&dquot->dq_count);
500 spin_unlock(&dq_list_lock);
501 sb->dq_op->write_dquot(dquot);
503 spin_lock(&dq_list_lock);
505 spin_unlock(&dq_list_lock);
508 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
509 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
510 && info_dirty(&dqopt->info[cnt]))
511 sb->dq_op->write_info(sb, cnt);
512 spin_lock(&dq_list_lock);
514 spin_unlock(&dq_list_lock);
515 mutex_unlock(&dqopt->dqonoff_mutex);
520 /* Free unused dquots from cache */
521 static void prune_dqcache(int count)
523 struct list_head *head;
526 head = free_dquots.prev;
527 while (head != &free_dquots && count) {
528 dquot = list_entry(head, struct dquot, dq_free);
529 remove_dquot_hash(dquot);
530 remove_free_dquot(dquot);
532 kmem_cache_free(dquot_cachep, dquot);
534 head = free_dquots.prev;
539 * This is called from kswapd when we think we need some
543 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
546 spin_lock(&dq_list_lock);
548 spin_unlock(&dq_list_lock);
550 return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
553 static struct shrinker dqcache_shrinker = {
554 .shrink = shrink_dqcache_memory,
555 .seeks = DEFAULT_SEEKS,
559 * Put reference to dquot
560 * NOTE: If you change this function please check whether dqput_blocks() works right...
561 * MUST be called with either dqptr_sem or dqonoff_mutex held
563 static void dqput(struct dquot *dquot)
567 #ifdef __DQUOT_PARANOIA
568 if (!atomic_read(&dquot->dq_count)) {
569 printk("VFS: dqput: trying to free free dquot\n");
570 printk("VFS: device %s, dquot of %s %d\n",
572 quotatypes[dquot->dq_type],
578 spin_lock(&dq_list_lock);
580 spin_unlock(&dq_list_lock);
582 spin_lock(&dq_list_lock);
583 if (atomic_read(&dquot->dq_count) > 1) {
584 /* We have more than one user... nothing to do */
585 atomic_dec(&dquot->dq_count);
586 /* Releasing dquot during quotaoff phase? */
587 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
588 atomic_read(&dquot->dq_count) == 1)
589 wake_up(&dquot->dq_wait_unused);
590 spin_unlock(&dq_list_lock);
593 /* Need to release dquot? */
594 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
595 spin_unlock(&dq_list_lock);
596 /* Commit dquot before releasing */
597 dquot->dq_sb->dq_op->write_dquot(dquot);
600 /* Clear flag in case dquot was inactive (something bad happened) */
601 clear_dquot_dirty(dquot);
602 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
603 spin_unlock(&dq_list_lock);
604 dquot->dq_sb->dq_op->release_dquot(dquot);
607 atomic_dec(&dquot->dq_count);
608 #ifdef __DQUOT_PARANOIA
610 BUG_ON(!list_empty(&dquot->dq_free));
612 put_dquot_last(dquot);
613 spin_unlock(&dq_list_lock);
616 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
620 dquot = kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
624 mutex_init(&dquot->dq_lock);
625 INIT_LIST_HEAD(&dquot->dq_free);
626 INIT_LIST_HEAD(&dquot->dq_inuse);
627 INIT_HLIST_NODE(&dquot->dq_hash);
628 INIT_LIST_HEAD(&dquot->dq_dirty);
629 init_waitqueue_head(&dquot->dq_wait_unused);
631 dquot->dq_type = type;
632 atomic_set(&dquot->dq_count, 1);
638 * Get reference to dquot
639 * MUST be called with either dqptr_sem or dqonoff_mutex held
641 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
643 unsigned int hashent = hashfn(sb, id, type);
644 struct dquot *dquot, *empty = NODQUOT;
646 if (!sb_has_quota_enabled(sb, type))
649 spin_lock(&dq_list_lock);
650 if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
651 if (empty == NODQUOT) {
652 spin_unlock(&dq_list_lock);
653 if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
654 schedule(); /* Try to wait for a moment... */
659 /* all dquots go on the inuse_list */
661 /* hash it first so it can be found */
662 insert_dquot_hash(dquot);
664 spin_unlock(&dq_list_lock);
666 if (!atomic_read(&dquot->dq_count))
667 remove_free_dquot(dquot);
668 atomic_inc(&dquot->dq_count);
669 dqstats.cache_hits++;
671 spin_unlock(&dq_list_lock);
673 kmem_cache_free(dquot_cachep, empty);
675 /* Wait for dq_lock - after this we know that either dquot_release() is already
676 * finished or it will be canceled due to dq_count > 1 test */
677 wait_on_dquot(dquot);
678 /* Read the dquot and instantiate it (everything done only if needed) */
679 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
683 #ifdef __DQUOT_PARANOIA
684 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
690 static int dqinit_needed(struct inode *inode, int type)
694 if (IS_NOQUOTA(inode))
697 return inode->i_dquot[type] == NODQUOT;
698 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
699 if (inode->i_dquot[cnt] == NODQUOT)
704 /* This routine is guarded by dqonoff_mutex mutex */
705 static void add_dquot_ref(struct super_block *sb, int type)
707 struct inode *inode, *old_inode = NULL;
709 spin_lock(&inode_lock);
710 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
711 if (!atomic_read(&inode->i_writecount))
713 if (!dqinit_needed(inode, type))
715 if (inode->i_state & (I_FREEING|I_WILL_FREE))
719 spin_unlock(&inode_lock);
722 sb->dq_op->initialize(inode, type);
723 /* We hold a reference to 'inode' so it couldn't have been
724 * removed from s_inodes list while we dropped the inode_lock.
725 * We cannot iput the inode now as we can be holding the last
726 * reference and we cannot iput it under inode_lock. So we
727 * keep the reference and iput it later. */
729 spin_lock(&inode_lock);
731 spin_unlock(&inode_lock);
735 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
736 static inline int dqput_blocks(struct dquot *dquot)
738 if (atomic_read(&dquot->dq_count) <= 1)
743 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
744 /* We can't race with anybody because we hold dqptr_sem for writing... */
745 static int remove_inode_dquot_ref(struct inode *inode, int type,
746 struct list_head *tofree_head)
748 struct dquot *dquot = inode->i_dquot[type];
750 inode->i_dquot[type] = NODQUOT;
751 if (dquot != NODQUOT) {
752 if (dqput_blocks(dquot)) {
753 #ifdef __DQUOT_PARANOIA
754 if (atomic_read(&dquot->dq_count) != 1)
755 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
757 spin_lock(&dq_list_lock);
758 list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
759 spin_unlock(&dq_list_lock);
763 dqput(dquot); /* We have guaranteed we won't block */
768 /* Free list of dquots - called from inode.c */
769 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
770 static void put_dquot_list(struct list_head *tofree_head)
772 struct list_head *act_head;
775 act_head = tofree_head->next;
776 /* So now we have dquots on the list... Just free them */
777 while (act_head != tofree_head) {
778 dquot = list_entry(act_head, struct dquot, dq_free);
779 act_head = act_head->next;
780 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
785 static void remove_dquot_ref(struct super_block *sb, int type,
786 struct list_head *tofree_head)
790 spin_lock(&inode_lock);
791 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
792 if (!IS_NOQUOTA(inode))
793 remove_inode_dquot_ref(inode, type, tofree_head);
795 spin_unlock(&inode_lock);
798 /* Gather all references from inodes and drop them */
799 static void drop_dquot_ref(struct super_block *sb, int type)
801 LIST_HEAD(tofree_head);
804 down_write(&sb_dqopt(sb)->dqptr_sem);
805 remove_dquot_ref(sb, type, &tofree_head);
806 up_write(&sb_dqopt(sb)->dqptr_sem);
807 put_dquot_list(&tofree_head);
811 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
813 dquot->dq_dqb.dqb_curinodes += number;
816 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
818 dquot->dq_dqb.dqb_curspace += number;
821 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
823 if (dquot->dq_dqb.dqb_curinodes > number)
824 dquot->dq_dqb.dqb_curinodes -= number;
826 dquot->dq_dqb.dqb_curinodes = 0;
827 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
828 dquot->dq_dqb.dqb_itime = (time_t) 0;
829 clear_bit(DQ_INODES_B, &dquot->dq_flags);
832 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
834 if (dquot->dq_dqb.dqb_curspace > number)
835 dquot->dq_dqb.dqb_curspace -= number;
837 dquot->dq_dqb.dqb_curspace = 0;
838 if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
839 dquot->dq_dqb.dqb_btime = (time_t) 0;
840 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
843 static int warning_issued(struct dquot *dquot, const int warntype)
845 int flag = (warntype == QUOTA_NL_BHARDWARN ||
846 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
847 ((warntype == QUOTA_NL_IHARDWARN ||
848 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
852 return test_and_set_bit(flag, &dquot->dq_flags);
855 #ifdef CONFIG_PRINT_QUOTA_WARNING
856 static int flag_print_warnings = 1;
858 static inline int need_print_warning(struct dquot *dquot)
860 if (!flag_print_warnings)
863 switch (dquot->dq_type) {
865 return current->fsuid == dquot->dq_id;
867 return in_group_p(dquot->dq_id);
872 /* Print warning to user which exceeded quota */
873 static void print_warning(struct dquot *dquot, const int warntype)
876 struct tty_struct *tty;
878 if (!need_print_warning(dquot))
881 mutex_lock(&tty_mutex);
882 tty = get_current_tty();
885 tty_write_message(tty, dquot->dq_sb->s_id);
886 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
887 tty_write_message(tty, ": warning, ");
889 tty_write_message(tty, ": write failed, ");
890 tty_write_message(tty, quotatypes[dquot->dq_type]);
892 case QUOTA_NL_IHARDWARN:
893 msg = " file limit reached.\r\n";
895 case QUOTA_NL_ISOFTLONGWARN:
896 msg = " file quota exceeded too long.\r\n";
898 case QUOTA_NL_ISOFTWARN:
899 msg = " file quota exceeded.\r\n";
901 case QUOTA_NL_BHARDWARN:
902 msg = " block limit reached.\r\n";
904 case QUOTA_NL_BSOFTLONGWARN:
905 msg = " block quota exceeded too long.\r\n";
907 case QUOTA_NL_BSOFTWARN:
908 msg = " block quota exceeded.\r\n";
911 tty_write_message(tty, msg);
913 mutex_unlock(&tty_mutex);
917 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
919 /* Netlink family structure for quota */
920 static struct genl_family quota_genl_family = {
921 .id = GENL_ID_GENERATE,
925 .maxattr = QUOTA_NL_A_MAX,
928 /* Send warning to userspace about user which exceeded quota */
929 static void send_warning(const struct dquot *dquot, const char warntype)
935 int msg_size = 4 * nla_total_size(sizeof(u32)) +
936 2 * nla_total_size(sizeof(u64));
938 /* We have to allocate using GFP_NOFS as we are called from a
939 * filesystem performing write and thus further recursion into
940 * the fs to free some data could cause deadlocks. */
941 skb = genlmsg_new(msg_size, GFP_NOFS);
944 "VFS: Not enough memory to send quota warning.\n");
947 msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
948 "a_genl_family, 0, QUOTA_NL_C_WARNING);
951 "VFS: Cannot store netlink header in quota warning.\n");
954 ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
957 ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
960 ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
963 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
964 MAJOR(dquot->dq_sb->s_dev));
967 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
968 MINOR(dquot->dq_sb->s_dev));
971 ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
974 genlmsg_end(skb, msg_head);
976 ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
977 if (ret < 0 && ret != -ESRCH)
979 "VFS: Failed to send notification message: %d\n", ret);
982 printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
988 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
992 for (i = 0; i < MAXQUOTAS; i++)
993 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
994 !warning_issued(dquots[i], warntype[i])) {
995 #ifdef CONFIG_PRINT_QUOTA_WARNING
996 print_warning(dquots[i], warntype[i]);
998 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
999 send_warning(dquots[i], warntype[i]);
1004 static inline char ignore_hardlimit(struct dquot *dquot)
1006 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1008 return capable(CAP_SYS_RESOURCE) &&
1009 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1012 /* needs dq_data_lock */
1013 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
1015 *warntype = QUOTA_NL_NOWARN;
1016 if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1019 if (dquot->dq_dqb.dqb_ihardlimit &&
1020 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1021 !ignore_hardlimit(dquot)) {
1022 *warntype = QUOTA_NL_IHARDWARN;
1026 if (dquot->dq_dqb.dqb_isoftlimit &&
1027 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1028 dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1029 !ignore_hardlimit(dquot)) {
1030 *warntype = QUOTA_NL_ISOFTLONGWARN;
1034 if (dquot->dq_dqb.dqb_isoftlimit &&
1035 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1036 dquot->dq_dqb.dqb_itime == 0) {
1037 *warntype = QUOTA_NL_ISOFTWARN;
1038 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1044 /* needs dq_data_lock */
1045 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1047 *warntype = QUOTA_NL_NOWARN;
1048 if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1051 if (dquot->dq_dqb.dqb_bhardlimit &&
1052 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1053 !ignore_hardlimit(dquot)) {
1055 *warntype = QUOTA_NL_BHARDWARN;
1059 if (dquot->dq_dqb.dqb_bsoftlimit &&
1060 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1061 dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1062 !ignore_hardlimit(dquot)) {
1064 *warntype = QUOTA_NL_BSOFTLONGWARN;
1068 if (dquot->dq_dqb.dqb_bsoftlimit &&
1069 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1070 dquot->dq_dqb.dqb_btime == 0) {
1072 *warntype = QUOTA_NL_BSOFTWARN;
1073 dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1077 * We don't allow preallocation to exceed softlimit so exceeding will
1087 * Initialize quota pointers in inode
1088 * Transaction must be started at entry
1090 int dquot_initialize(struct inode *inode, int type)
1092 unsigned int id = 0;
1095 /* First test before acquiring mutex - solves deadlocks when we
1096 * re-enter the quota code and are already holding the mutex */
1097 if (IS_NOQUOTA(inode))
1099 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1100 /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1101 if (IS_NOQUOTA(inode))
1103 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1104 if (type != -1 && cnt != type)
1106 if (inode->i_dquot[cnt] == NODQUOT) {
1115 inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1119 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1124 * Release all quotas referenced by inode
1125 * Transaction must be started at an entry
1127 int dquot_drop(struct inode *inode)
1131 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1132 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1133 if (inode->i_dquot[cnt] != NODQUOT) {
1134 dqput(inode->i_dquot[cnt]);
1135 inode->i_dquot[cnt] = NODQUOT;
1138 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1143 * Following four functions update i_blocks+i_bytes fields and
1144 * quota information (together with appropriate checks)
1145 * NOTE: We absolutely rely on the fact that caller dirties
1146 * the inode (usually macros in quotaops.h care about this) and
1147 * holds a handle for the current transaction so that dquot write and
1148 * inode write go into the same transaction.
1152 * This operation can block, but only after everything is updated
1154 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1156 int cnt, ret = NO_QUOTA;
1157 char warntype[MAXQUOTAS];
1159 /* First test before acquiring mutex - solves deadlocks when we
1160 * re-enter the quota code and are already holding the mutex */
1161 if (IS_NOQUOTA(inode)) {
1163 inode_add_bytes(inode, number);
1166 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1167 warntype[cnt] = QUOTA_NL_NOWARN;
1169 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1170 if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */
1171 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1174 spin_lock(&dq_data_lock);
1175 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1176 if (inode->i_dquot[cnt] == NODQUOT)
1178 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1181 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1182 if (inode->i_dquot[cnt] == NODQUOT)
1184 dquot_incr_space(inode->i_dquot[cnt], number);
1186 inode_add_bytes(inode, number);
1189 spin_unlock(&dq_data_lock);
1190 if (ret == QUOTA_OK)
1191 /* Dirtify all the dquots - this can block when journalling */
1192 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1193 if (inode->i_dquot[cnt])
1194 mark_dquot_dirty(inode->i_dquot[cnt]);
1195 flush_warnings(inode->i_dquot, warntype);
1196 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1201 * This operation can block, but only after everything is updated
1203 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1205 int cnt, ret = NO_QUOTA;
1206 char warntype[MAXQUOTAS];
1208 /* First test before acquiring mutex - solves deadlocks when we
1209 * re-enter the quota code and are already holding the mutex */
1210 if (IS_NOQUOTA(inode))
1212 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1213 warntype[cnt] = QUOTA_NL_NOWARN;
1214 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1215 if (IS_NOQUOTA(inode)) {
1216 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1219 spin_lock(&dq_data_lock);
1220 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1221 if (inode->i_dquot[cnt] == NODQUOT)
1223 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1227 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1228 if (inode->i_dquot[cnt] == NODQUOT)
1230 dquot_incr_inodes(inode->i_dquot[cnt], number);
1234 spin_unlock(&dq_data_lock);
1235 if (ret == QUOTA_OK)
1236 /* Dirtify all the dquots - this can block when journalling */
1237 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1238 if (inode->i_dquot[cnt])
1239 mark_dquot_dirty(inode->i_dquot[cnt]);
1240 flush_warnings(inode->i_dquot, warntype);
1241 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1246 * This operation can block, but only after everything is updated
1248 int dquot_free_space(struct inode *inode, qsize_t number)
1252 /* First test before acquiring mutex - solves deadlocks when we
1253 * re-enter the quota code and are already holding the mutex */
1254 if (IS_NOQUOTA(inode)) {
1256 inode_sub_bytes(inode, number);
1259 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1260 /* Now recheck reliably when holding dqptr_sem */
1261 if (IS_NOQUOTA(inode)) {
1262 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1265 spin_lock(&dq_data_lock);
1266 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1267 if (inode->i_dquot[cnt] == NODQUOT)
1269 dquot_decr_space(inode->i_dquot[cnt], number);
1271 inode_sub_bytes(inode, number);
1272 spin_unlock(&dq_data_lock);
1273 /* Dirtify all the dquots - this can block when journalling */
1274 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1275 if (inode->i_dquot[cnt])
1276 mark_dquot_dirty(inode->i_dquot[cnt]);
1277 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1282 * This operation can block, but only after everything is updated
1284 int dquot_free_inode(const struct inode *inode, unsigned long number)
1288 /* First test before acquiring mutex - solves deadlocks when we
1289 * re-enter the quota code and are already holding the mutex */
1290 if (IS_NOQUOTA(inode))
1292 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1293 /* Now recheck reliably when holding dqptr_sem */
1294 if (IS_NOQUOTA(inode)) {
1295 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1298 spin_lock(&dq_data_lock);
1299 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1300 if (inode->i_dquot[cnt] == NODQUOT)
1302 dquot_decr_inodes(inode->i_dquot[cnt], number);
1304 spin_unlock(&dq_data_lock);
1305 /* Dirtify all the dquots - this can block when journalling */
1306 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1307 if (inode->i_dquot[cnt])
1308 mark_dquot_dirty(inode->i_dquot[cnt]);
1309 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1314 * Transfer the number of inode and blocks from one diskquota to an other.
1316 * This operation can block, but only after everything is updated
1317 * A transaction must be started when entering this function.
1319 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1322 struct dquot *transfer_from[MAXQUOTAS];
1323 struct dquot *transfer_to[MAXQUOTAS];
1324 int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1325 chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1326 char warntype[MAXQUOTAS];
1328 /* First test before acquiring mutex - solves deadlocks when we
1329 * re-enter the quota code and are already holding the mutex */
1330 if (IS_NOQUOTA(inode))
1332 /* Clear the arrays */
1333 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1334 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1335 warntype[cnt] = QUOTA_NL_NOWARN;
1337 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1338 /* Now recheck reliably when holding dqptr_sem */
1339 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1340 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1343 /* First build the transfer_to list - here we can block on
1344 * reading/instantiating of dquots. We know that the transaction for
1345 * us was already started so we don't violate lock ranking here */
1346 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1351 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1356 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1360 spin_lock(&dq_data_lock);
1361 space = inode_get_bytes(inode);
1362 /* Build the transfer_from list and check the limits */
1363 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1364 if (transfer_to[cnt] == NODQUOT)
1366 transfer_from[cnt] = inode->i_dquot[cnt];
1367 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1368 check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1373 * Finally perform the needed transfer from transfer_from to transfer_to
1375 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1377 * Skip changes for same uid or gid or for turned off quota-type.
1379 if (transfer_to[cnt] == NODQUOT)
1382 /* Due to IO error we might not have transfer_from[] structure */
1383 if (transfer_from[cnt]) {
1384 dquot_decr_inodes(transfer_from[cnt], 1);
1385 dquot_decr_space(transfer_from[cnt], space);
1388 dquot_incr_inodes(transfer_to[cnt], 1);
1389 dquot_incr_space(transfer_to[cnt], space);
1391 inode->i_dquot[cnt] = transfer_to[cnt];
1395 spin_unlock(&dq_data_lock);
1396 /* Dirtify all the dquots - this can block when journalling */
1397 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1398 if (transfer_from[cnt])
1399 mark_dquot_dirty(transfer_from[cnt]);
1400 if (transfer_to[cnt])
1401 mark_dquot_dirty(transfer_to[cnt]);
1403 flush_warnings(transfer_to, warntype);
1405 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1406 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1407 dqput(transfer_from[cnt]);
1408 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1409 dqput(transfer_to[cnt]);
1411 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1416 * Write info of quota file to disk
1418 int dquot_commit_info(struct super_block *sb, int type)
1421 struct quota_info *dqopt = sb_dqopt(sb);
1423 mutex_lock(&dqopt->dqio_mutex);
1424 ret = dqopt->ops[type]->write_file_info(sb, type);
1425 mutex_unlock(&dqopt->dqio_mutex);
1430 * Definitions of diskquota operations.
1432 struct dquot_operations dquot_operations = {
1433 .initialize = dquot_initialize,
1435 .alloc_space = dquot_alloc_space,
1436 .alloc_inode = dquot_alloc_inode,
1437 .free_space = dquot_free_space,
1438 .free_inode = dquot_free_inode,
1439 .transfer = dquot_transfer,
1440 .write_dquot = dquot_commit,
1441 .acquire_dquot = dquot_acquire,
1442 .release_dquot = dquot_release,
1443 .mark_dirty = dquot_mark_dquot_dirty,
1444 .write_info = dquot_commit_info
1447 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1451 dqopt->flags |= DQUOT_USR_ENABLED;
1452 dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1455 dqopt->flags |= DQUOT_GRP_ENABLED;
1456 dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1461 static inline void reset_enable_flags(struct quota_info *dqopt, int type,
1466 dqopt->flags &= ~DQUOT_USR_ENABLED;
1468 dqopt->flags |= DQUOT_USR_SUSPENDED;
1470 dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1473 dqopt->flags &= ~DQUOT_GRP_ENABLED;
1475 dqopt->flags |= DQUOT_GRP_SUSPENDED;
1477 dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1484 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1486 int vfs_quota_off(struct super_block *sb, int type, int remount)
1489 struct quota_info *dqopt = sb_dqopt(sb);
1490 struct inode *toputinode[MAXQUOTAS];
1492 /* We need to serialize quota_off() for device */
1493 mutex_lock(&dqopt->dqonoff_mutex);
1494 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1495 toputinode[cnt] = NULL;
1496 if (type != -1 && cnt != type)
1498 /* If we keep inodes of quota files after remount and quotaoff
1499 * is called, drop kept inodes. */
1500 if (!remount && sb_has_quota_suspended(sb, cnt)) {
1501 iput(dqopt->files[cnt]);
1502 dqopt->files[cnt] = NULL;
1503 reset_enable_flags(dqopt, cnt, 0);
1506 if (!sb_has_quota_enabled(sb, cnt))
1508 reset_enable_flags(dqopt, cnt, remount);
1510 /* Note: these are blocking operations */
1511 drop_dquot_ref(sb, cnt);
1512 invalidate_dquots(sb, cnt);
1514 * Now all dquots should be invalidated, all writes done so we should be only
1515 * users of the info. No locks needed.
1517 if (info_dirty(&dqopt->info[cnt]))
1518 sb->dq_op->write_info(sb, cnt);
1519 if (dqopt->ops[cnt]->free_file_info)
1520 dqopt->ops[cnt]->free_file_info(sb, cnt);
1521 put_quota_format(dqopt->info[cnt].dqi_format);
1523 toputinode[cnt] = dqopt->files[cnt];
1525 dqopt->files[cnt] = NULL;
1526 dqopt->info[cnt].dqi_flags = 0;
1527 dqopt->info[cnt].dqi_igrace = 0;
1528 dqopt->info[cnt].dqi_bgrace = 0;
1529 dqopt->ops[cnt] = NULL;
1531 mutex_unlock(&dqopt->dqonoff_mutex);
1532 /* Sync the superblock so that buffers with quota data are written to
1533 * disk (and so userspace sees correct data afterwards). */
1534 if (sb->s_op->sync_fs)
1535 sb->s_op->sync_fs(sb, 1);
1536 sync_blockdev(sb->s_bdev);
1537 /* Now the quota files are just ordinary files and we can set the
1538 * inode flags back. Moreover we discard the pagecache so that
1539 * userspace sees the writes we did bypassing the pagecache. We
1540 * must also discard the blockdev buffers so that we see the
1541 * changes done by userspace on the next quotaon() */
1542 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1543 if (toputinode[cnt]) {
1544 mutex_lock(&dqopt->dqonoff_mutex);
1545 /* If quota was reenabled in the meantime, we have
1547 if (!sb_has_quota_enabled(sb, cnt)) {
1548 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1549 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1550 S_NOATIME | S_NOQUOTA);
1551 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1552 mutex_unlock(&toputinode[cnt]->i_mutex);
1553 mark_inode_dirty(toputinode[cnt]);
1555 mutex_unlock(&dqopt->dqonoff_mutex);
1556 /* On remount RO, we keep the inode pointer so that we
1557 * can reenable quota on the subsequent remount RW.
1558 * But we have better not keep inode pointer when there
1559 * is pending delete on the quota file... */
1561 iput(toputinode[cnt]);
1562 else if (!toputinode[cnt]->i_nlink)
1566 invalidate_bdev(sb->s_bdev);
1571 * Turn quotas on on a device
1574 /* Helper function when we already have the inode */
1575 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1577 struct quota_format_type *fmt = find_quota_format(format_id);
1578 struct super_block *sb = inode->i_sb;
1579 struct quota_info *dqopt = sb_dqopt(sb);
1585 if (!S_ISREG(inode->i_mode)) {
1589 if (IS_RDONLY(inode)) {
1593 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1598 /* As we bypass the pagecache we must now flush the inode so that
1599 * we see all the changes from userspace... */
1600 write_inode_now(inode, 1);
1601 /* And now flush the block cache so that kernel sees the changes */
1602 invalidate_bdev(sb->s_bdev);
1603 mutex_lock(&inode->i_mutex);
1604 mutex_lock(&dqopt->dqonoff_mutex);
1605 if (sb_has_quota_enabled(sb, type) ||
1606 sb_has_quota_suspended(sb, type)) {
1610 /* We don't want quota and atime on quota files (deadlocks possible)
1611 * Also nobody should write to the file - we use special IO operations
1612 * which ignore the immutable bit. */
1613 down_write(&dqopt->dqptr_sem);
1614 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1615 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1616 up_write(&dqopt->dqptr_sem);
1617 sb->dq_op->drop(inode);
1620 dqopt->files[type] = igrab(inode);
1621 if (!dqopt->files[type])
1624 if (!fmt->qf_ops->check_quota_file(sb, type))
1627 dqopt->ops[type] = fmt->qf_ops;
1628 dqopt->info[type].dqi_format = fmt;
1629 dqopt->info[type].dqi_fmt_id = format_id;
1630 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1631 mutex_lock(&dqopt->dqio_mutex);
1632 if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1633 mutex_unlock(&dqopt->dqio_mutex);
1636 mutex_unlock(&dqopt->dqio_mutex);
1637 mutex_unlock(&inode->i_mutex);
1638 set_enable_flags(dqopt, type);
1640 add_dquot_ref(sb, type);
1641 mutex_unlock(&dqopt->dqonoff_mutex);
1646 dqopt->files[type] = NULL;
1649 mutex_unlock(&dqopt->dqonoff_mutex);
1650 if (oldflags != -1) {
1651 down_write(&dqopt->dqptr_sem);
1652 /* Set the flags back (in the case of accidental quotaon()
1653 * on a wrong file we don't want to mess up the flags) */
1654 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1655 inode->i_flags |= oldflags;
1656 up_write(&dqopt->dqptr_sem);
1658 mutex_unlock(&inode->i_mutex);
1660 put_quota_format(fmt);
1665 /* Reenable quotas on remount RW */
1666 static int vfs_quota_on_remount(struct super_block *sb, int type)
1668 struct quota_info *dqopt = sb_dqopt(sb);
1669 struct inode *inode;
1672 mutex_lock(&dqopt->dqonoff_mutex);
1673 if (!sb_has_quota_suspended(sb, type)) {
1674 mutex_unlock(&dqopt->dqonoff_mutex);
1677 BUG_ON(sb_has_quota_enabled(sb, type));
1679 inode = dqopt->files[type];
1680 dqopt->files[type] = NULL;
1681 reset_enable_flags(dqopt, type, 0);
1682 mutex_unlock(&dqopt->dqonoff_mutex);
1684 ret = vfs_quota_on_inode(inode, type, dqopt->info[type].dqi_fmt_id);
1690 /* Actual function called from quotactl() */
1691 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path,
1694 struct nameidata nd;
1698 return vfs_quota_on_remount(sb, type);
1700 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1703 error = security_quota_on(nd.path.dentry);
1706 /* Quota file not on the same filesystem? */
1707 if (nd.path.mnt->mnt_sb != sb)
1710 error = vfs_quota_on_inode(nd.path.dentry->d_inode, type,
1718 * This function is used when filesystem needs to initialize quotas
1719 * during mount time.
1721 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1722 int format_id, int type)
1724 struct dentry *dentry;
1727 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1729 return PTR_ERR(dentry);
1731 if (!dentry->d_inode) {
1736 error = security_quota_on(dentry);
1738 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1745 /* Generic routine for getting common part of quota structure */
1746 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1748 struct mem_dqblk *dm = &dquot->dq_dqb;
1750 spin_lock(&dq_data_lock);
1751 di->dqb_bhardlimit = dm->dqb_bhardlimit;
1752 di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1753 di->dqb_curspace = dm->dqb_curspace;
1754 di->dqb_ihardlimit = dm->dqb_ihardlimit;
1755 di->dqb_isoftlimit = dm->dqb_isoftlimit;
1756 di->dqb_curinodes = dm->dqb_curinodes;
1757 di->dqb_btime = dm->dqb_btime;
1758 di->dqb_itime = dm->dqb_itime;
1759 di->dqb_valid = QIF_ALL;
1760 spin_unlock(&dq_data_lock);
1763 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1765 struct dquot *dquot;
1767 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1768 if (!(dquot = dqget(sb, id, type))) {
1769 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1772 do_get_dqblk(dquot, di);
1774 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1778 /* Generic routine for setting common part of quota structure */
1779 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1781 struct mem_dqblk *dm = &dquot->dq_dqb;
1782 int check_blim = 0, check_ilim = 0;
1783 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1785 if ((di->dqb_valid & QIF_BLIMITS &&
1786 (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
1787 di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
1788 (di->dqb_valid & QIF_ILIMITS &&
1789 (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
1790 di->dqb_isoftlimit > dqi->dqi_maxilimit)))
1793 spin_lock(&dq_data_lock);
1794 if (di->dqb_valid & QIF_SPACE) {
1795 dm->dqb_curspace = di->dqb_curspace;
1798 if (di->dqb_valid & QIF_BLIMITS) {
1799 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1800 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1803 if (di->dqb_valid & QIF_INODES) {
1804 dm->dqb_curinodes = di->dqb_curinodes;
1807 if (di->dqb_valid & QIF_ILIMITS) {
1808 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1809 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1812 if (di->dqb_valid & QIF_BTIME)
1813 dm->dqb_btime = di->dqb_btime;
1814 if (di->dqb_valid & QIF_ITIME)
1815 dm->dqb_itime = di->dqb_itime;
1818 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1820 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1822 else if (!(di->dqb_valid & QIF_BTIME)) /* Set grace only if user hasn't provided his own... */
1823 dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
1826 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1828 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1830 else if (!(di->dqb_valid & QIF_ITIME)) /* Set grace only if user hasn't provided his own... */
1831 dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
1833 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1834 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1836 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1837 spin_unlock(&dq_data_lock);
1838 mark_dquot_dirty(dquot);
1843 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1845 struct dquot *dquot;
1848 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1849 if (!(dquot = dqget(sb, id, type))) {
1850 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1853 rc = do_set_dqblk(dquot, di);
1855 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1859 /* Generic routine for getting common part of quota file information */
1860 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1862 struct mem_dqinfo *mi;
1864 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1865 if (!sb_has_quota_enabled(sb, type)) {
1866 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1869 mi = sb_dqopt(sb)->info + type;
1870 spin_lock(&dq_data_lock);
1871 ii->dqi_bgrace = mi->dqi_bgrace;
1872 ii->dqi_igrace = mi->dqi_igrace;
1873 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1874 ii->dqi_valid = IIF_ALL;
1875 spin_unlock(&dq_data_lock);
1876 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1880 /* Generic routine for setting common part of quota file information */
1881 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1883 struct mem_dqinfo *mi;
1885 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1886 if (!sb_has_quota_enabled(sb, type)) {
1887 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1890 mi = sb_dqopt(sb)->info + type;
1891 spin_lock(&dq_data_lock);
1892 if (ii->dqi_valid & IIF_BGRACE)
1893 mi->dqi_bgrace = ii->dqi_bgrace;
1894 if (ii->dqi_valid & IIF_IGRACE)
1895 mi->dqi_igrace = ii->dqi_igrace;
1896 if (ii->dqi_valid & IIF_FLAGS)
1897 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1898 spin_unlock(&dq_data_lock);
1899 mark_info_dirty(sb, type);
1900 /* Force write to disk */
1901 sb->dq_op->write_info(sb, type);
1902 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1906 struct quotactl_ops vfs_quotactl_ops = {
1907 .quota_on = vfs_quota_on,
1908 .quota_off = vfs_quota_off,
1909 .quota_sync = vfs_quota_sync,
1910 .get_info = vfs_get_dqinfo,
1911 .set_info = vfs_set_dqinfo,
1912 .get_dqblk = vfs_get_dqblk,
1913 .set_dqblk = vfs_set_dqblk
1916 static ctl_table fs_dqstats_table[] = {
1918 .ctl_name = FS_DQ_LOOKUPS,
1919 .procname = "lookups",
1920 .data = &dqstats.lookups,
1921 .maxlen = sizeof(int),
1923 .proc_handler = &proc_dointvec,
1926 .ctl_name = FS_DQ_DROPS,
1927 .procname = "drops",
1928 .data = &dqstats.drops,
1929 .maxlen = sizeof(int),
1931 .proc_handler = &proc_dointvec,
1934 .ctl_name = FS_DQ_READS,
1935 .procname = "reads",
1936 .data = &dqstats.reads,
1937 .maxlen = sizeof(int),
1939 .proc_handler = &proc_dointvec,
1942 .ctl_name = FS_DQ_WRITES,
1943 .procname = "writes",
1944 .data = &dqstats.writes,
1945 .maxlen = sizeof(int),
1947 .proc_handler = &proc_dointvec,
1950 .ctl_name = FS_DQ_CACHE_HITS,
1951 .procname = "cache_hits",
1952 .data = &dqstats.cache_hits,
1953 .maxlen = sizeof(int),
1955 .proc_handler = &proc_dointvec,
1958 .ctl_name = FS_DQ_ALLOCATED,
1959 .procname = "allocated_dquots",
1960 .data = &dqstats.allocated_dquots,
1961 .maxlen = sizeof(int),
1963 .proc_handler = &proc_dointvec,
1966 .ctl_name = FS_DQ_FREE,
1967 .procname = "free_dquots",
1968 .data = &dqstats.free_dquots,
1969 .maxlen = sizeof(int),
1971 .proc_handler = &proc_dointvec,
1974 .ctl_name = FS_DQ_SYNCS,
1975 .procname = "syncs",
1976 .data = &dqstats.syncs,
1977 .maxlen = sizeof(int),
1979 .proc_handler = &proc_dointvec,
1981 #ifdef CONFIG_PRINT_QUOTA_WARNING
1983 .ctl_name = FS_DQ_WARNINGS,
1984 .procname = "warnings",
1985 .data = &flag_print_warnings,
1986 .maxlen = sizeof(int),
1988 .proc_handler = &proc_dointvec,
1994 static ctl_table fs_table[] = {
1996 .ctl_name = FS_DQSTATS,
1997 .procname = "quota",
1999 .child = fs_dqstats_table,
2004 static ctl_table sys_table[] = {
2014 static int __init dquot_init(void)
2017 unsigned long nr_hash, order;
2019 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2021 register_sysctl_table(sys_table);
2023 dquot_cachep = kmem_cache_create("dquot",
2024 sizeof(struct dquot), sizeof(unsigned long) * 4,
2025 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2026 SLAB_MEM_SPREAD|SLAB_PANIC),
2030 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2032 panic("Cannot create dquot hash table");
2034 /* Find power-of-two hlist_heads which can fit into allocation */
2035 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2039 } while (nr_hash >> dq_hash_bits);
2042 nr_hash = 1UL << dq_hash_bits;
2043 dq_hash_mask = nr_hash - 1;
2044 for (i = 0; i < nr_hash; i++)
2045 INIT_HLIST_HEAD(dquot_hash + i);
2047 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2048 nr_hash, order, (PAGE_SIZE << order));
2050 register_shrinker(&dqcache_shrinker);
2052 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2053 if (genl_register_family("a_genl_family) != 0)
2054 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2059 module_init(dquot_init);
2061 EXPORT_SYMBOL(register_quota_format);
2062 EXPORT_SYMBOL(unregister_quota_format);
2063 EXPORT_SYMBOL(dqstats);
2064 EXPORT_SYMBOL(dq_data_lock);
2065 EXPORT_SYMBOL(vfs_quota_on);
2066 EXPORT_SYMBOL(vfs_quota_on_mount);
2067 EXPORT_SYMBOL(vfs_quota_off);
2068 EXPORT_SYMBOL(vfs_quota_sync);
2069 EXPORT_SYMBOL(vfs_get_dqinfo);
2070 EXPORT_SYMBOL(vfs_set_dqinfo);
2071 EXPORT_SYMBOL(vfs_get_dqblk);
2072 EXPORT_SYMBOL(vfs_set_dqblk);
2073 EXPORT_SYMBOL(dquot_commit);
2074 EXPORT_SYMBOL(dquot_commit_info);
2075 EXPORT_SYMBOL(dquot_acquire);
2076 EXPORT_SYMBOL(dquot_release);
2077 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2078 EXPORT_SYMBOL(dquot_initialize);
2079 EXPORT_SYMBOL(dquot_drop);
2080 EXPORT_SYMBOL(dquot_alloc_space);
2081 EXPORT_SYMBOL(dquot_alloc_inode);
2082 EXPORT_SYMBOL(dquot_free_space);
2083 EXPORT_SYMBOL(dquot_free_inode);
2084 EXPORT_SYMBOL(dquot_transfer);