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)
699 struct inode *inode, *old_inode = NULL;
701 spin_lock(&inode_lock);
702 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
703 if (!atomic_read(&inode->i_writecount))
705 if (!dqinit_needed(inode, type))
707 if (inode->i_state & (I_FREEING|I_WILL_FREE))
711 spin_unlock(&inode_lock);
714 sb->dq_op->initialize(inode, type);
715 /* We hold a reference to 'inode' so it couldn't have been
716 * removed from s_inodes list while we dropped the inode_lock.
717 * We cannot iput the inode now as we can be holding the last
718 * reference and we cannot iput it under inode_lock. So we
719 * keep the reference and iput it later. */
721 spin_lock(&inode_lock);
723 spin_unlock(&inode_lock);
727 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
728 static inline int dqput_blocks(struct dquot *dquot)
730 if (atomic_read(&dquot->dq_count) <= 1)
735 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
736 /* We can't race with anybody because we hold dqptr_sem for writing... */
737 static int remove_inode_dquot_ref(struct inode *inode, int type,
738 struct list_head *tofree_head)
740 struct dquot *dquot = inode->i_dquot[type];
742 inode->i_dquot[type] = NODQUOT;
743 if (dquot != NODQUOT) {
744 if (dqput_blocks(dquot)) {
745 #ifdef __DQUOT_PARANOIA
746 if (atomic_read(&dquot->dq_count) != 1)
747 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
749 spin_lock(&dq_list_lock);
750 list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
751 spin_unlock(&dq_list_lock);
755 dqput(dquot); /* We have guaranteed we won't block */
760 /* Free list of dquots - called from inode.c */
761 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
762 static void put_dquot_list(struct list_head *tofree_head)
764 struct list_head *act_head;
767 act_head = tofree_head->next;
768 /* So now we have dquots on the list... Just free them */
769 while (act_head != tofree_head) {
770 dquot = list_entry(act_head, struct dquot, dq_free);
771 act_head = act_head->next;
772 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
777 static void remove_dquot_ref(struct super_block *sb, int type,
778 struct list_head *tofree_head)
782 spin_lock(&inode_lock);
783 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
784 if (!IS_NOQUOTA(inode))
785 remove_inode_dquot_ref(inode, type, tofree_head);
787 spin_unlock(&inode_lock);
790 /* Gather all references from inodes and drop them */
791 static void drop_dquot_ref(struct super_block *sb, int type)
793 LIST_HEAD(tofree_head);
796 down_write(&sb_dqopt(sb)->dqptr_sem);
797 remove_dquot_ref(sb, type, &tofree_head);
798 up_write(&sb_dqopt(sb)->dqptr_sem);
799 put_dquot_list(&tofree_head);
803 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
805 dquot->dq_dqb.dqb_curinodes += number;
808 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
810 dquot->dq_dqb.dqb_curspace += number;
813 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
815 if (dquot->dq_dqb.dqb_curinodes > number)
816 dquot->dq_dqb.dqb_curinodes -= number;
818 dquot->dq_dqb.dqb_curinodes = 0;
819 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
820 dquot->dq_dqb.dqb_itime = (time_t) 0;
821 clear_bit(DQ_INODES_B, &dquot->dq_flags);
824 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
826 if (dquot->dq_dqb.dqb_curspace > number)
827 dquot->dq_dqb.dqb_curspace -= number;
829 dquot->dq_dqb.dqb_curspace = 0;
830 if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
831 dquot->dq_dqb.dqb_btime = (time_t) 0;
832 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
835 static int warning_issued(struct dquot *dquot, const int warntype)
837 int flag = (warntype == QUOTA_NL_BHARDWARN ||
838 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
839 ((warntype == QUOTA_NL_IHARDWARN ||
840 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
844 return test_and_set_bit(flag, &dquot->dq_flags);
847 #ifdef CONFIG_PRINT_QUOTA_WARNING
848 static int flag_print_warnings = 1;
850 static inline int need_print_warning(struct dquot *dquot)
852 if (!flag_print_warnings)
855 switch (dquot->dq_type) {
857 return current->fsuid == dquot->dq_id;
859 return in_group_p(dquot->dq_id);
864 /* Print warning to user which exceeded quota */
865 static void print_warning(struct dquot *dquot, const int warntype)
868 struct tty_struct *tty;
870 if (!need_print_warning(dquot))
873 mutex_lock(&tty_mutex);
874 tty = get_current_tty();
877 tty_write_message(tty, dquot->dq_sb->s_id);
878 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
879 tty_write_message(tty, ": warning, ");
881 tty_write_message(tty, ": write failed, ");
882 tty_write_message(tty, quotatypes[dquot->dq_type]);
884 case QUOTA_NL_IHARDWARN:
885 msg = " file limit reached.\r\n";
887 case QUOTA_NL_ISOFTLONGWARN:
888 msg = " file quota exceeded too long.\r\n";
890 case QUOTA_NL_ISOFTWARN:
891 msg = " file quota exceeded.\r\n";
893 case QUOTA_NL_BHARDWARN:
894 msg = " block limit reached.\r\n";
896 case QUOTA_NL_BSOFTLONGWARN:
897 msg = " block quota exceeded too long.\r\n";
899 case QUOTA_NL_BSOFTWARN:
900 msg = " block quota exceeded.\r\n";
903 tty_write_message(tty, msg);
905 mutex_unlock(&tty_mutex);
909 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
911 /* Netlink family structure for quota */
912 static struct genl_family quota_genl_family = {
913 .id = GENL_ID_GENERATE,
917 .maxattr = QUOTA_NL_A_MAX,
920 /* Send warning to userspace about user which exceeded quota */
921 static void send_warning(const struct dquot *dquot, const char warntype)
927 int msg_size = 4 * nla_total_size(sizeof(u32)) +
928 2 * nla_total_size(sizeof(u64));
930 /* We have to allocate using GFP_NOFS as we are called from a
931 * filesystem performing write and thus further recursion into
932 * the fs to free some data could cause deadlocks. */
933 skb = genlmsg_new(msg_size, GFP_NOFS);
936 "VFS: Not enough memory to send quota warning.\n");
939 msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
940 "a_genl_family, 0, QUOTA_NL_C_WARNING);
943 "VFS: Cannot store netlink header in quota warning.\n");
946 ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
949 ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
952 ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
955 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
956 MAJOR(dquot->dq_sb->s_dev));
959 ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
960 MINOR(dquot->dq_sb->s_dev));
963 ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
966 genlmsg_end(skb, msg_head);
968 ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
969 if (ret < 0 && ret != -ESRCH)
971 "VFS: Failed to send notification message: %d\n", ret);
974 printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
980 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
984 for (i = 0; i < MAXQUOTAS; i++)
985 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
986 !warning_issued(dquots[i], warntype[i])) {
987 #ifdef CONFIG_PRINT_QUOTA_WARNING
988 print_warning(dquots[i], warntype[i]);
990 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
991 send_warning(dquots[i], warntype[i]);
996 static inline char ignore_hardlimit(struct dquot *dquot)
998 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1000 return capable(CAP_SYS_RESOURCE) &&
1001 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1004 /* needs dq_data_lock */
1005 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
1007 *warntype = QUOTA_NL_NOWARN;
1008 if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1011 if (dquot->dq_dqb.dqb_ihardlimit &&
1012 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1013 !ignore_hardlimit(dquot)) {
1014 *warntype = QUOTA_NL_IHARDWARN;
1018 if (dquot->dq_dqb.dqb_isoftlimit &&
1019 (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1020 dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1021 !ignore_hardlimit(dquot)) {
1022 *warntype = QUOTA_NL_ISOFTLONGWARN;
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 == 0) {
1029 *warntype = QUOTA_NL_ISOFTWARN;
1030 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1036 /* needs dq_data_lock */
1037 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1039 *warntype = QUOTA_NL_NOWARN;
1040 if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1043 if (dquot->dq_dqb.dqb_bhardlimit &&
1044 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1045 !ignore_hardlimit(dquot)) {
1047 *warntype = QUOTA_NL_BHARDWARN;
1051 if (dquot->dq_dqb.dqb_bsoftlimit &&
1052 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1053 dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1054 !ignore_hardlimit(dquot)) {
1056 *warntype = QUOTA_NL_BSOFTLONGWARN;
1060 if (dquot->dq_dqb.dqb_bsoftlimit &&
1061 toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1062 dquot->dq_dqb.dqb_btime == 0) {
1064 *warntype = QUOTA_NL_BSOFTWARN;
1065 dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1069 * We don't allow preallocation to exceed softlimit so exceeding will
1079 * Initialize quota pointers in inode
1080 * Transaction must be started at entry
1082 int dquot_initialize(struct inode *inode, int type)
1084 unsigned int id = 0;
1087 /* First test before acquiring mutex - solves deadlocks when we
1088 * re-enter the quota code and are already holding the mutex */
1089 if (IS_NOQUOTA(inode))
1091 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1092 /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1093 if (IS_NOQUOTA(inode))
1095 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1096 if (type != -1 && cnt != type)
1098 if (inode->i_dquot[cnt] == NODQUOT) {
1107 inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1111 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1116 * Release all quotas referenced by inode
1117 * Transaction must be started at an entry
1119 int dquot_drop(struct inode *inode)
1123 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1124 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1125 if (inode->i_dquot[cnt] != NODQUOT) {
1126 dqput(inode->i_dquot[cnt]);
1127 inode->i_dquot[cnt] = NODQUOT;
1130 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1135 * Following four functions update i_blocks+i_bytes fields and
1136 * quota information (together with appropriate checks)
1137 * NOTE: We absolutely rely on the fact that caller dirties
1138 * the inode (usually macros in quotaops.h care about this) and
1139 * holds a handle for the current transaction so that dquot write and
1140 * inode write go into the same transaction.
1144 * This operation can block, but only after everything is updated
1146 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1148 int cnt, ret = NO_QUOTA;
1149 char warntype[MAXQUOTAS];
1151 /* First test before acquiring mutex - solves deadlocks when we
1152 * re-enter the quota code and are already holding the mutex */
1153 if (IS_NOQUOTA(inode)) {
1155 inode_add_bytes(inode, number);
1158 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1159 warntype[cnt] = QUOTA_NL_NOWARN;
1161 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1162 if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */
1163 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1166 spin_lock(&dq_data_lock);
1167 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1168 if (inode->i_dquot[cnt] == NODQUOT)
1170 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1173 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1174 if (inode->i_dquot[cnt] == NODQUOT)
1176 dquot_incr_space(inode->i_dquot[cnt], number);
1178 inode_add_bytes(inode, number);
1181 spin_unlock(&dq_data_lock);
1182 if (ret == QUOTA_OK)
1183 /* Dirtify all the dquots - this can block when journalling */
1184 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1185 if (inode->i_dquot[cnt])
1186 mark_dquot_dirty(inode->i_dquot[cnt]);
1187 flush_warnings(inode->i_dquot, warntype);
1188 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1193 * This operation can block, but only after everything is updated
1195 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1197 int cnt, ret = NO_QUOTA;
1198 char warntype[MAXQUOTAS];
1200 /* First test before acquiring mutex - solves deadlocks when we
1201 * re-enter the quota code and are already holding the mutex */
1202 if (IS_NOQUOTA(inode))
1204 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1205 warntype[cnt] = QUOTA_NL_NOWARN;
1206 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1207 if (IS_NOQUOTA(inode)) {
1208 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1211 spin_lock(&dq_data_lock);
1212 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1213 if (inode->i_dquot[cnt] == NODQUOT)
1215 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1219 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1220 if (inode->i_dquot[cnt] == NODQUOT)
1222 dquot_incr_inodes(inode->i_dquot[cnt], number);
1226 spin_unlock(&dq_data_lock);
1227 if (ret == QUOTA_OK)
1228 /* Dirtify all the dquots - this can block when journalling */
1229 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1230 if (inode->i_dquot[cnt])
1231 mark_dquot_dirty(inode->i_dquot[cnt]);
1232 flush_warnings(inode->i_dquot, warntype);
1233 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1238 * This operation can block, but only after everything is updated
1240 int dquot_free_space(struct inode *inode, qsize_t number)
1244 /* First test before acquiring mutex - solves deadlocks when we
1245 * re-enter the quota code and are already holding the mutex */
1246 if (IS_NOQUOTA(inode)) {
1248 inode_sub_bytes(inode, number);
1251 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1252 /* Now recheck reliably when holding dqptr_sem */
1253 if (IS_NOQUOTA(inode)) {
1254 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1257 spin_lock(&dq_data_lock);
1258 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1259 if (inode->i_dquot[cnt] == NODQUOT)
1261 dquot_decr_space(inode->i_dquot[cnt], number);
1263 inode_sub_bytes(inode, number);
1264 spin_unlock(&dq_data_lock);
1265 /* Dirtify all the dquots - this can block when journalling */
1266 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1267 if (inode->i_dquot[cnt])
1268 mark_dquot_dirty(inode->i_dquot[cnt]);
1269 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1274 * This operation can block, but only after everything is updated
1276 int dquot_free_inode(const struct inode *inode, unsigned long number)
1280 /* First test before acquiring mutex - solves deadlocks when we
1281 * re-enter the quota code and are already holding the mutex */
1282 if (IS_NOQUOTA(inode))
1284 down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1285 /* Now recheck reliably when holding dqptr_sem */
1286 if (IS_NOQUOTA(inode)) {
1287 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1290 spin_lock(&dq_data_lock);
1291 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1292 if (inode->i_dquot[cnt] == NODQUOT)
1294 dquot_decr_inodes(inode->i_dquot[cnt], number);
1296 spin_unlock(&dq_data_lock);
1297 /* Dirtify all the dquots - this can block when journalling */
1298 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1299 if (inode->i_dquot[cnt])
1300 mark_dquot_dirty(inode->i_dquot[cnt]);
1301 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1306 * Transfer the number of inode and blocks from one diskquota to an other.
1308 * This operation can block, but only after everything is updated
1309 * A transaction must be started when entering this function.
1311 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1314 struct dquot *transfer_from[MAXQUOTAS];
1315 struct dquot *transfer_to[MAXQUOTAS];
1316 int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1317 chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1318 char warntype[MAXQUOTAS];
1320 /* First test before acquiring mutex - solves deadlocks when we
1321 * re-enter the quota code and are already holding the mutex */
1322 if (IS_NOQUOTA(inode))
1324 /* Clear the arrays */
1325 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1326 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1327 warntype[cnt] = QUOTA_NL_NOWARN;
1329 down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1330 /* Now recheck reliably when holding dqptr_sem */
1331 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1332 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1335 /* First build the transfer_to list - here we can block on
1336 * reading/instantiating of dquots. We know that the transaction for
1337 * us was already started so we don't violate lock ranking here */
1338 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1343 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1348 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1352 spin_lock(&dq_data_lock);
1353 space = inode_get_bytes(inode);
1354 /* Build the transfer_from list and check the limits */
1355 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1356 if (transfer_to[cnt] == NODQUOT)
1358 transfer_from[cnt] = inode->i_dquot[cnt];
1359 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1360 check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1365 * Finally perform the needed transfer from transfer_from to transfer_to
1367 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1369 * Skip changes for same uid or gid or for turned off quota-type.
1371 if (transfer_to[cnt] == NODQUOT)
1374 /* Due to IO error we might not have transfer_from[] structure */
1375 if (transfer_from[cnt]) {
1376 dquot_decr_inodes(transfer_from[cnt], 1);
1377 dquot_decr_space(transfer_from[cnt], space);
1380 dquot_incr_inodes(transfer_to[cnt], 1);
1381 dquot_incr_space(transfer_to[cnt], space);
1383 inode->i_dquot[cnt] = transfer_to[cnt];
1387 spin_unlock(&dq_data_lock);
1388 /* Dirtify all the dquots - this can block when journalling */
1389 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1390 if (transfer_from[cnt])
1391 mark_dquot_dirty(transfer_from[cnt]);
1392 if (transfer_to[cnt])
1393 mark_dquot_dirty(transfer_to[cnt]);
1395 flush_warnings(transfer_to, warntype);
1397 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1398 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1399 dqput(transfer_from[cnt]);
1400 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1401 dqput(transfer_to[cnt]);
1403 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1408 * Write info of quota file to disk
1410 int dquot_commit_info(struct super_block *sb, int type)
1413 struct quota_info *dqopt = sb_dqopt(sb);
1415 mutex_lock(&dqopt->dqio_mutex);
1416 ret = dqopt->ops[type]->write_file_info(sb, type);
1417 mutex_unlock(&dqopt->dqio_mutex);
1422 * Definitions of diskquota operations.
1424 struct dquot_operations dquot_operations = {
1425 .initialize = dquot_initialize,
1427 .alloc_space = dquot_alloc_space,
1428 .alloc_inode = dquot_alloc_inode,
1429 .free_space = dquot_free_space,
1430 .free_inode = dquot_free_inode,
1431 .transfer = dquot_transfer,
1432 .write_dquot = dquot_commit,
1433 .acquire_dquot = dquot_acquire,
1434 .release_dquot = dquot_release,
1435 .mark_dirty = dquot_mark_dquot_dirty,
1436 .write_info = dquot_commit_info
1439 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1443 dqopt->flags |= DQUOT_USR_ENABLED;
1446 dqopt->flags |= DQUOT_GRP_ENABLED;
1451 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1455 dqopt->flags &= ~DQUOT_USR_ENABLED;
1458 dqopt->flags &= ~DQUOT_GRP_ENABLED;
1464 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1466 int vfs_quota_off(struct super_block *sb, int type)
1469 struct quota_info *dqopt = sb_dqopt(sb);
1470 struct inode *toputinode[MAXQUOTAS];
1472 /* We need to serialize quota_off() for device */
1473 mutex_lock(&dqopt->dqonoff_mutex);
1474 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1475 toputinode[cnt] = NULL;
1476 if (type != -1 && cnt != type)
1478 if (!sb_has_quota_enabled(sb, cnt))
1480 reset_enable_flags(dqopt, cnt);
1482 /* Note: these are blocking operations */
1483 drop_dquot_ref(sb, cnt);
1484 invalidate_dquots(sb, cnt);
1486 * Now all dquots should be invalidated, all writes done so we should be only
1487 * users of the info. No locks needed.
1489 if (info_dirty(&dqopt->info[cnt]))
1490 sb->dq_op->write_info(sb, cnt);
1491 if (dqopt->ops[cnt]->free_file_info)
1492 dqopt->ops[cnt]->free_file_info(sb, cnt);
1493 put_quota_format(dqopt->info[cnt].dqi_format);
1495 toputinode[cnt] = dqopt->files[cnt];
1496 dqopt->files[cnt] = NULL;
1497 dqopt->info[cnt].dqi_flags = 0;
1498 dqopt->info[cnt].dqi_igrace = 0;
1499 dqopt->info[cnt].dqi_bgrace = 0;
1500 dqopt->ops[cnt] = NULL;
1502 mutex_unlock(&dqopt->dqonoff_mutex);
1503 /* Sync the superblock so that buffers with quota data are written to
1504 * disk (and so userspace sees correct data afterwards). */
1505 if (sb->s_op->sync_fs)
1506 sb->s_op->sync_fs(sb, 1);
1507 sync_blockdev(sb->s_bdev);
1508 /* Now the quota files are just ordinary files and we can set the
1509 * inode flags back. Moreover we discard the pagecache so that
1510 * userspace sees the writes we did bypassing the pagecache. We
1511 * must also discard the blockdev buffers so that we see the
1512 * changes done by userspace on the next quotaon() */
1513 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1514 if (toputinode[cnt]) {
1515 mutex_lock(&dqopt->dqonoff_mutex);
1516 /* If quota was reenabled in the meantime, we have
1518 if (!sb_has_quota_enabled(sb, cnt)) {
1519 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1520 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1521 S_NOATIME | S_NOQUOTA);
1522 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1523 mutex_unlock(&toputinode[cnt]->i_mutex);
1524 mark_inode_dirty(toputinode[cnt]);
1525 iput(toputinode[cnt]);
1527 mutex_unlock(&dqopt->dqonoff_mutex);
1530 invalidate_bdev(sb->s_bdev);
1535 * Turn quotas on on a device
1538 /* Helper function when we already have the inode */
1539 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1541 struct quota_format_type *fmt = find_quota_format(format_id);
1542 struct super_block *sb = inode->i_sb;
1543 struct quota_info *dqopt = sb_dqopt(sb);
1549 if (!S_ISREG(inode->i_mode)) {
1553 if (IS_RDONLY(inode)) {
1557 if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1562 /* As we bypass the pagecache we must now flush the inode so that
1563 * we see all the changes from userspace... */
1564 write_inode_now(inode, 1);
1565 /* And now flush the block cache so that kernel sees the changes */
1566 invalidate_bdev(sb->s_bdev);
1567 mutex_lock(&inode->i_mutex);
1568 mutex_lock(&dqopt->dqonoff_mutex);
1569 if (sb_has_quota_enabled(sb, type)) {
1573 /* We don't want quota and atime on quota files (deadlocks possible)
1574 * Also nobody should write to the file - we use special IO operations
1575 * which ignore the immutable bit. */
1576 down_write(&dqopt->dqptr_sem);
1577 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1578 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1579 up_write(&dqopt->dqptr_sem);
1580 sb->dq_op->drop(inode);
1583 dqopt->files[type] = igrab(inode);
1584 if (!dqopt->files[type])
1587 if (!fmt->qf_ops->check_quota_file(sb, type))
1590 dqopt->ops[type] = fmt->qf_ops;
1591 dqopt->info[type].dqi_format = fmt;
1592 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1593 mutex_lock(&dqopt->dqio_mutex);
1594 if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1595 mutex_unlock(&dqopt->dqio_mutex);
1598 mutex_unlock(&dqopt->dqio_mutex);
1599 mutex_unlock(&inode->i_mutex);
1600 set_enable_flags(dqopt, type);
1602 add_dquot_ref(sb, type);
1603 mutex_unlock(&dqopt->dqonoff_mutex);
1608 dqopt->files[type] = NULL;
1611 mutex_unlock(&dqopt->dqonoff_mutex);
1612 if (oldflags != -1) {
1613 down_write(&dqopt->dqptr_sem);
1614 /* Set the flags back (in the case of accidental quotaon()
1615 * on a wrong file we don't want to mess up the flags) */
1616 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1617 inode->i_flags |= oldflags;
1618 up_write(&dqopt->dqptr_sem);
1620 mutex_unlock(&inode->i_mutex);
1622 put_quota_format(fmt);
1627 /* Actual function called from quotactl() */
1628 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1630 struct nameidata nd;
1633 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1636 error = security_quota_on(nd.path.dentry);
1639 /* Quota file not on the same filesystem? */
1640 if (nd.path.mnt->mnt_sb != sb)
1643 error = vfs_quota_on_inode(nd.path.dentry->d_inode, type,
1651 * This function is used when filesystem needs to initialize quotas
1652 * during mount time.
1654 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1655 int format_id, int type)
1657 struct dentry *dentry;
1660 dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1662 return PTR_ERR(dentry);
1664 if (!dentry->d_inode) {
1669 error = security_quota_on(dentry);
1671 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1678 /* Generic routine for getting common part of quota structure */
1679 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1681 struct mem_dqblk *dm = &dquot->dq_dqb;
1683 spin_lock(&dq_data_lock);
1684 di->dqb_bhardlimit = dm->dqb_bhardlimit;
1685 di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1686 di->dqb_curspace = dm->dqb_curspace;
1687 di->dqb_ihardlimit = dm->dqb_ihardlimit;
1688 di->dqb_isoftlimit = dm->dqb_isoftlimit;
1689 di->dqb_curinodes = dm->dqb_curinodes;
1690 di->dqb_btime = dm->dqb_btime;
1691 di->dqb_itime = dm->dqb_itime;
1692 di->dqb_valid = QIF_ALL;
1693 spin_unlock(&dq_data_lock);
1696 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1698 struct dquot *dquot;
1700 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1701 if (!(dquot = dqget(sb, id, type))) {
1702 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1705 do_get_dqblk(dquot, di);
1707 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1711 /* Generic routine for setting common part of quota structure */
1712 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1714 struct mem_dqblk *dm = &dquot->dq_dqb;
1715 int check_blim = 0, check_ilim = 0;
1717 spin_lock(&dq_data_lock);
1718 if (di->dqb_valid & QIF_SPACE) {
1719 dm->dqb_curspace = di->dqb_curspace;
1722 if (di->dqb_valid & QIF_BLIMITS) {
1723 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1724 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1727 if (di->dqb_valid & QIF_INODES) {
1728 dm->dqb_curinodes = di->dqb_curinodes;
1731 if (di->dqb_valid & QIF_ILIMITS) {
1732 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1733 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1736 if (di->dqb_valid & QIF_BTIME)
1737 dm->dqb_btime = di->dqb_btime;
1738 if (di->dqb_valid & QIF_ITIME)
1739 dm->dqb_itime = di->dqb_itime;
1742 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1744 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1746 else if (!(di->dqb_valid & QIF_BTIME)) /* Set grace only if user hasn't provided his own... */
1747 dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1750 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1752 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1754 else if (!(di->dqb_valid & QIF_ITIME)) /* Set grace only if user hasn't provided his own... */
1755 dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1757 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1758 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1760 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1761 spin_unlock(&dq_data_lock);
1762 mark_dquot_dirty(dquot);
1765 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1767 struct dquot *dquot;
1769 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1770 if (!(dquot = dqget(sb, id, type))) {
1771 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1774 do_set_dqblk(dquot, di);
1776 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1780 /* Generic routine for getting common part of quota file information */
1781 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1783 struct mem_dqinfo *mi;
1785 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1786 if (!sb_has_quota_enabled(sb, type)) {
1787 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1790 mi = sb_dqopt(sb)->info + type;
1791 spin_lock(&dq_data_lock);
1792 ii->dqi_bgrace = mi->dqi_bgrace;
1793 ii->dqi_igrace = mi->dqi_igrace;
1794 ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1795 ii->dqi_valid = IIF_ALL;
1796 spin_unlock(&dq_data_lock);
1797 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1801 /* Generic routine for setting common part of quota file information */
1802 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1804 struct mem_dqinfo *mi;
1806 mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1807 if (!sb_has_quota_enabled(sb, type)) {
1808 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1811 mi = sb_dqopt(sb)->info + type;
1812 spin_lock(&dq_data_lock);
1813 if (ii->dqi_valid & IIF_BGRACE)
1814 mi->dqi_bgrace = ii->dqi_bgrace;
1815 if (ii->dqi_valid & IIF_IGRACE)
1816 mi->dqi_igrace = ii->dqi_igrace;
1817 if (ii->dqi_valid & IIF_FLAGS)
1818 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1819 spin_unlock(&dq_data_lock);
1820 mark_info_dirty(sb, type);
1821 /* Force write to disk */
1822 sb->dq_op->write_info(sb, type);
1823 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1827 struct quotactl_ops vfs_quotactl_ops = {
1828 .quota_on = vfs_quota_on,
1829 .quota_off = vfs_quota_off,
1830 .quota_sync = vfs_quota_sync,
1831 .get_info = vfs_get_dqinfo,
1832 .set_info = vfs_set_dqinfo,
1833 .get_dqblk = vfs_get_dqblk,
1834 .set_dqblk = vfs_set_dqblk
1837 static ctl_table fs_dqstats_table[] = {
1839 .ctl_name = FS_DQ_LOOKUPS,
1840 .procname = "lookups",
1841 .data = &dqstats.lookups,
1842 .maxlen = sizeof(int),
1844 .proc_handler = &proc_dointvec,
1847 .ctl_name = FS_DQ_DROPS,
1848 .procname = "drops",
1849 .data = &dqstats.drops,
1850 .maxlen = sizeof(int),
1852 .proc_handler = &proc_dointvec,
1855 .ctl_name = FS_DQ_READS,
1856 .procname = "reads",
1857 .data = &dqstats.reads,
1858 .maxlen = sizeof(int),
1860 .proc_handler = &proc_dointvec,
1863 .ctl_name = FS_DQ_WRITES,
1864 .procname = "writes",
1865 .data = &dqstats.writes,
1866 .maxlen = sizeof(int),
1868 .proc_handler = &proc_dointvec,
1871 .ctl_name = FS_DQ_CACHE_HITS,
1872 .procname = "cache_hits",
1873 .data = &dqstats.cache_hits,
1874 .maxlen = sizeof(int),
1876 .proc_handler = &proc_dointvec,
1879 .ctl_name = FS_DQ_ALLOCATED,
1880 .procname = "allocated_dquots",
1881 .data = &dqstats.allocated_dquots,
1882 .maxlen = sizeof(int),
1884 .proc_handler = &proc_dointvec,
1887 .ctl_name = FS_DQ_FREE,
1888 .procname = "free_dquots",
1889 .data = &dqstats.free_dquots,
1890 .maxlen = sizeof(int),
1892 .proc_handler = &proc_dointvec,
1895 .ctl_name = FS_DQ_SYNCS,
1896 .procname = "syncs",
1897 .data = &dqstats.syncs,
1898 .maxlen = sizeof(int),
1900 .proc_handler = &proc_dointvec,
1902 #ifdef CONFIG_PRINT_QUOTA_WARNING
1904 .ctl_name = FS_DQ_WARNINGS,
1905 .procname = "warnings",
1906 .data = &flag_print_warnings,
1907 .maxlen = sizeof(int),
1909 .proc_handler = &proc_dointvec,
1915 static ctl_table fs_table[] = {
1917 .ctl_name = FS_DQSTATS,
1918 .procname = "quota",
1920 .child = fs_dqstats_table,
1925 static ctl_table sys_table[] = {
1935 static int __init dquot_init(void)
1938 unsigned long nr_hash, order;
1940 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1942 register_sysctl_table(sys_table);
1944 dquot_cachep = kmem_cache_create("dquot",
1945 sizeof(struct dquot), sizeof(unsigned long) * 4,
1946 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1947 SLAB_MEM_SPREAD|SLAB_PANIC),
1951 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1953 panic("Cannot create dquot hash table");
1955 /* Find power-of-two hlist_heads which can fit into allocation */
1956 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1960 } while (nr_hash >> dq_hash_bits);
1963 nr_hash = 1UL << dq_hash_bits;
1964 dq_hash_mask = nr_hash - 1;
1965 for (i = 0; i < nr_hash; i++)
1966 INIT_HLIST_HEAD(dquot_hash + i);
1968 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1969 nr_hash, order, (PAGE_SIZE << order));
1971 register_shrinker(&dqcache_shrinker);
1973 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1974 if (genl_register_family("a_genl_family) != 0)
1975 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
1980 module_init(dquot_init);
1982 EXPORT_SYMBOL(register_quota_format);
1983 EXPORT_SYMBOL(unregister_quota_format);
1984 EXPORT_SYMBOL(dqstats);
1985 EXPORT_SYMBOL(dq_data_lock);
1986 EXPORT_SYMBOL(vfs_quota_on);
1987 EXPORT_SYMBOL(vfs_quota_on_mount);
1988 EXPORT_SYMBOL(vfs_quota_off);
1989 EXPORT_SYMBOL(vfs_quota_sync);
1990 EXPORT_SYMBOL(vfs_get_dqinfo);
1991 EXPORT_SYMBOL(vfs_set_dqinfo);
1992 EXPORT_SYMBOL(vfs_get_dqblk);
1993 EXPORT_SYMBOL(vfs_set_dqblk);
1994 EXPORT_SYMBOL(dquot_commit);
1995 EXPORT_SYMBOL(dquot_commit_info);
1996 EXPORT_SYMBOL(dquot_acquire);
1997 EXPORT_SYMBOL(dquot_release);
1998 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1999 EXPORT_SYMBOL(dquot_initialize);
2000 EXPORT_SYMBOL(dquot_drop);
2001 EXPORT_SYMBOL(dquot_alloc_space);
2002 EXPORT_SYMBOL(dquot_alloc_inode);
2003 EXPORT_SYMBOL(dquot_free_space);
2004 EXPORT_SYMBOL(dquot_free_inode);
2005 EXPORT_SYMBOL(dquot_transfer);