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/smp_lock.h>
73 #include <linux/init.h>
74 #include <linux/module.h>
75 #include <linux/proc_fs.h>
76 #include <linux/security.h>
77 #include <linux/kmod.h>
78 #include <linux/namei.h>
79 #include <linux/buffer_head.h>
80 #include <linux/capability.h>
81 #include <linux/quotaops.h>
82 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
84 #include <asm/uaccess.h>
86 #define __DQUOT_PARANOIA
89 * There are two quota SMP locks. dq_list_lock protects all lists with quotas
90 * and quota formats and also dqstats structure containing statistics about the
91 * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
92 * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
93 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
94 * in inode_add_bytes() and inode_sub_bytes().
96 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
98 * Note that some things (eg. sb pointer, type, id) doesn't change during
99 * the life of the dquot structure and so needn't to be protected by a lock
101 * Any operation working on dquots via inode pointers must hold dqptr_sem. If
102 * operation is just reading pointers from inode (or not using them at all) the
103 * read lock is enough. If pointers are altered function must hold write lock
104 * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
105 * for altering the flag i_mutex is also needed). If operation is holding
106 * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
108 * This locking assures that:
109 * a) update/access to dquot pointers in inode is serialized
110 * b) everyone is guarded against invalidate_dquots()
112 * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
113 * from inodes (dquot_alloc_space() and such don't check the dq_lock).
114 * Currently dquot is locked only when it is being read to memory (or space for
115 * it is being allocated) on the first dqget() and when it is being released on
116 * the last dqput(). The allocation and release oparations are serialized by
117 * the dq_lock and by checking the use count in dquot_release(). Write
118 * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
119 * spinlock to internal buffers before writing.
121 * Lock ordering (including related VFS locks) is the following:
122 * i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
124 * i_mutex on quota files is special (it's below dqio_mutex)
127 static DEFINE_SPINLOCK(dq_list_lock
);
128 DEFINE_SPINLOCK(dq_data_lock
);
130 static char *quotatypes
[] = INITQFNAMES
;
131 static struct quota_format_type
*quota_formats
; /* List of registered formats */
132 static struct quota_module_name module_names
[] = INIT_QUOTA_MODULE_NAMES
;
134 /* SLAB cache for dquot structures */
135 static struct kmem_cache
*dquot_cachep
;
137 int register_quota_format(struct quota_format_type
*fmt
)
139 spin_lock(&dq_list_lock
);
140 fmt
->qf_next
= quota_formats
;
142 spin_unlock(&dq_list_lock
);
146 void unregister_quota_format(struct quota_format_type
*fmt
)
148 struct quota_format_type
**actqf
;
150 spin_lock(&dq_list_lock
);
151 for (actqf
= "a_formats
; *actqf
&& *actqf
!= fmt
; actqf
= &(*actqf
)->qf_next
);
153 *actqf
= (*actqf
)->qf_next
;
154 spin_unlock(&dq_list_lock
);
157 static struct quota_format_type
*find_quota_format(int id
)
159 struct quota_format_type
*actqf
;
161 spin_lock(&dq_list_lock
);
162 for (actqf
= quota_formats
; actqf
&& actqf
->qf_fmt_id
!= id
; actqf
= actqf
->qf_next
);
163 if (!actqf
|| !try_module_get(actqf
->qf_owner
)) {
166 spin_unlock(&dq_list_lock
);
168 for (qm
= 0; module_names
[qm
].qm_fmt_id
&& module_names
[qm
].qm_fmt_id
!= id
; qm
++);
169 if (!module_names
[qm
].qm_fmt_id
|| request_module(module_names
[qm
].qm_mod_name
))
172 spin_lock(&dq_list_lock
);
173 for (actqf
= quota_formats
; actqf
&& actqf
->qf_fmt_id
!= id
; actqf
= actqf
->qf_next
);
174 if (actqf
&& !try_module_get(actqf
->qf_owner
))
177 spin_unlock(&dq_list_lock
);
181 static void put_quota_format(struct quota_format_type
*fmt
)
183 module_put(fmt
->qf_owner
);
187 * Dquot List Management:
188 * The quota code uses three lists for dquot management: the inuse_list,
189 * free_dquots, and dquot_hash[] array. A single dquot structure may be
190 * on all three lists, depending on its current state.
192 * All dquots are placed to the end of inuse_list when first created, and this
193 * list is used for invalidate operation, which must look at every dquot.
195 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
196 * and this list is searched whenever we need an available dquot. Dquots are
197 * removed from the list as soon as they are used again, and
198 * dqstats.free_dquots gives the number of dquots on the list. When
199 * dquot is invalidated it's completely released from memory.
201 * Dquots with a specific identity (device, type and id) are placed on
202 * one of the dquot_hash[] hash chains. The provides an efficient search
203 * mechanism to locate a specific dquot.
206 static LIST_HEAD(inuse_list
);
207 static LIST_HEAD(free_dquots
);
208 static unsigned int dq_hash_bits
, dq_hash_mask
;
209 static struct hlist_head
*dquot_hash
;
211 struct dqstats dqstats
;
213 static void dqput(struct dquot
*dquot
);
215 static inline unsigned int
216 hashfn(const struct super_block
*sb
, unsigned int id
, int type
)
220 tmp
= (((unsigned long)sb
>>L1_CACHE_SHIFT
) ^ id
) * (MAXQUOTAS
- type
);
221 return (tmp
+ (tmp
>> dq_hash_bits
)) & dq_hash_mask
;
225 * Following list functions expect dq_list_lock to be held
227 static inline void insert_dquot_hash(struct dquot
*dquot
)
229 struct hlist_head
*head
= dquot_hash
+ hashfn(dquot
->dq_sb
, dquot
->dq_id
, dquot
->dq_type
);
230 hlist_add_head(&dquot
->dq_hash
, head
);
233 static inline void remove_dquot_hash(struct dquot
*dquot
)
235 hlist_del_init(&dquot
->dq_hash
);
238 static inline struct dquot
*find_dquot(unsigned int hashent
, struct super_block
*sb
, unsigned int id
, int type
)
240 struct hlist_node
*node
;
243 hlist_for_each (node
, dquot_hash
+hashent
) {
244 dquot
= hlist_entry(node
, struct dquot
, dq_hash
);
245 if (dquot
->dq_sb
== sb
&& dquot
->dq_id
== id
&& dquot
->dq_type
== type
)
251 /* Add a dquot to the tail of the free list */
252 static inline void put_dquot_last(struct dquot
*dquot
)
254 list_add_tail(&dquot
->dq_free
, &free_dquots
);
255 dqstats
.free_dquots
++;
258 static inline void remove_free_dquot(struct dquot
*dquot
)
260 if (list_empty(&dquot
->dq_free
))
262 list_del_init(&dquot
->dq_free
);
263 dqstats
.free_dquots
--;
266 static inline void put_inuse(struct dquot
*dquot
)
268 /* We add to the back of inuse list so we don't have to restart
269 * when traversing this list and we block */
270 list_add_tail(&dquot
->dq_inuse
, &inuse_list
);
271 dqstats
.allocated_dquots
++;
274 static inline void remove_inuse(struct dquot
*dquot
)
276 dqstats
.allocated_dquots
--;
277 list_del(&dquot
->dq_inuse
);
280 * End of list functions needing dq_list_lock
283 static void wait_on_dquot(struct dquot
*dquot
)
285 mutex_lock(&dquot
->dq_lock
);
286 mutex_unlock(&dquot
->dq_lock
);
289 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
291 int dquot_mark_dquot_dirty(struct dquot
*dquot
)
293 spin_lock(&dq_list_lock
);
294 if (!test_and_set_bit(DQ_MOD_B
, &dquot
->dq_flags
))
295 list_add(&dquot
->dq_dirty
, &sb_dqopt(dquot
->dq_sb
)->
296 info
[dquot
->dq_type
].dqi_dirty_list
);
297 spin_unlock(&dq_list_lock
);
301 /* This function needs dq_list_lock */
302 static inline int clear_dquot_dirty(struct dquot
*dquot
)
304 if (!test_and_clear_bit(DQ_MOD_B
, &dquot
->dq_flags
))
306 list_del_init(&dquot
->dq_dirty
);
310 void mark_info_dirty(struct super_block
*sb
, int type
)
312 set_bit(DQF_INFO_DIRTY_B
, &sb_dqopt(sb
)->info
[type
].dqi_flags
);
314 EXPORT_SYMBOL(mark_info_dirty
);
317 * Read dquot from disk and alloc space for it
320 int dquot_acquire(struct dquot
*dquot
)
322 int ret
= 0, ret2
= 0;
323 struct quota_info
*dqopt
= sb_dqopt(dquot
->dq_sb
);
325 mutex_lock(&dquot
->dq_lock
);
326 mutex_lock(&dqopt
->dqio_mutex
);
327 if (!test_bit(DQ_READ_B
, &dquot
->dq_flags
))
328 ret
= dqopt
->ops
[dquot
->dq_type
]->read_dqblk(dquot
);
331 set_bit(DQ_READ_B
, &dquot
->dq_flags
);
332 /* Instantiate dquot if needed */
333 if (!test_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
) && !dquot
->dq_off
) {
334 ret
= dqopt
->ops
[dquot
->dq_type
]->commit_dqblk(dquot
);
335 /* Write the info if needed */
336 if (info_dirty(&dqopt
->info
[dquot
->dq_type
]))
337 ret2
= dqopt
->ops
[dquot
->dq_type
]->write_file_info(dquot
->dq_sb
, dquot
->dq_type
);
345 set_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
);
347 mutex_unlock(&dqopt
->dqio_mutex
);
348 mutex_unlock(&dquot
->dq_lock
);
353 * Write dquot to disk
355 int dquot_commit(struct dquot
*dquot
)
357 int ret
= 0, ret2
= 0;
358 struct quota_info
*dqopt
= sb_dqopt(dquot
->dq_sb
);
360 mutex_lock(&dqopt
->dqio_mutex
);
361 spin_lock(&dq_list_lock
);
362 if (!clear_dquot_dirty(dquot
)) {
363 spin_unlock(&dq_list_lock
);
366 spin_unlock(&dq_list_lock
);
367 /* Inactive dquot can be only if there was error during read/init
368 * => we have better not writing it */
369 if (test_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
)) {
370 ret
= dqopt
->ops
[dquot
->dq_type
]->commit_dqblk(dquot
);
371 if (info_dirty(&dqopt
->info
[dquot
->dq_type
]))
372 ret2
= dqopt
->ops
[dquot
->dq_type
]->write_file_info(dquot
->dq_sb
, dquot
->dq_type
);
377 mutex_unlock(&dqopt
->dqio_mutex
);
384 int dquot_release(struct dquot
*dquot
)
386 int ret
= 0, ret2
= 0;
387 struct quota_info
*dqopt
= sb_dqopt(dquot
->dq_sb
);
389 mutex_lock(&dquot
->dq_lock
);
390 /* Check whether we are not racing with some other dqget() */
391 if (atomic_read(&dquot
->dq_count
) > 1)
393 mutex_lock(&dqopt
->dqio_mutex
);
394 if (dqopt
->ops
[dquot
->dq_type
]->release_dqblk
) {
395 ret
= dqopt
->ops
[dquot
->dq_type
]->release_dqblk(dquot
);
397 if (info_dirty(&dqopt
->info
[dquot
->dq_type
]))
398 ret2
= dqopt
->ops
[dquot
->dq_type
]->write_file_info(dquot
->dq_sb
, dquot
->dq_type
);
402 clear_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
);
403 mutex_unlock(&dqopt
->dqio_mutex
);
405 mutex_unlock(&dquot
->dq_lock
);
409 /* Invalidate all dquots on the list. Note that this function is called after
410 * quota is disabled and pointers from inodes removed so there cannot be new
411 * quota users. There can still be some users of quotas due to inodes being
412 * just deleted or pruned by prune_icache() (those are not attached to any
413 * list). We have to wait for such users.
415 static void invalidate_dquots(struct super_block
*sb
, int type
)
417 struct dquot
*dquot
, *tmp
;
420 spin_lock(&dq_list_lock
);
421 list_for_each_entry_safe(dquot
, tmp
, &inuse_list
, dq_inuse
) {
422 if (dquot
->dq_sb
!= sb
)
424 if (dquot
->dq_type
!= type
)
426 /* Wait for dquot users */
427 if (atomic_read(&dquot
->dq_count
)) {
430 atomic_inc(&dquot
->dq_count
);
431 prepare_to_wait(&dquot
->dq_wait_unused
, &wait
,
432 TASK_UNINTERRUPTIBLE
);
433 spin_unlock(&dq_list_lock
);
434 /* Once dqput() wakes us up, we know it's time to free
436 * IMPORTANT: we rely on the fact that there is always
437 * at most one process waiting for dquot to free.
438 * Otherwise dq_count would be > 1 and we would never
441 if (atomic_read(&dquot
->dq_count
) > 1)
443 finish_wait(&dquot
->dq_wait_unused
, &wait
);
445 /* At this moment dquot() need not exist (it could be
446 * reclaimed by prune_dqcache(). Hence we must
451 * Quota now has no users and it has been written on last
454 remove_dquot_hash(dquot
);
455 remove_free_dquot(dquot
);
457 kmem_cache_free(dquot_cachep
, dquot
);
459 spin_unlock(&dq_list_lock
);
462 int vfs_quota_sync(struct super_block
*sb
, int type
)
464 struct list_head
*dirty
;
466 struct quota_info
*dqopt
= sb_dqopt(sb
);
469 mutex_lock(&dqopt
->dqonoff_mutex
);
470 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
471 if (type
!= -1 && cnt
!= type
)
473 if (!sb_has_quota_enabled(sb
, cnt
))
475 spin_lock(&dq_list_lock
);
476 dirty
= &dqopt
->info
[cnt
].dqi_dirty_list
;
477 while (!list_empty(dirty
)) {
478 dquot
= list_entry(dirty
->next
, struct dquot
, dq_dirty
);
479 /* Dirty and inactive can be only bad dquot... */
480 if (!test_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
)) {
481 clear_dquot_dirty(dquot
);
484 /* Now we have active dquot from which someone is
485 * holding reference so we can safely just increase
487 atomic_inc(&dquot
->dq_count
);
489 spin_unlock(&dq_list_lock
);
490 sb
->dq_op
->write_dquot(dquot
);
492 spin_lock(&dq_list_lock
);
494 spin_unlock(&dq_list_lock
);
497 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
498 if ((cnt
== type
|| type
== -1) && sb_has_quota_enabled(sb
, cnt
)
499 && info_dirty(&dqopt
->info
[cnt
]))
500 sb
->dq_op
->write_info(sb
, cnt
);
501 spin_lock(&dq_list_lock
);
503 spin_unlock(&dq_list_lock
);
504 mutex_unlock(&dqopt
->dqonoff_mutex
);
509 /* Free unused dquots from cache */
510 static void prune_dqcache(int count
)
512 struct list_head
*head
;
515 head
= free_dquots
.prev
;
516 while (head
!= &free_dquots
&& count
) {
517 dquot
= list_entry(head
, struct dquot
, dq_free
);
518 remove_dquot_hash(dquot
);
519 remove_free_dquot(dquot
);
521 kmem_cache_free(dquot_cachep
, dquot
);
523 head
= free_dquots
.prev
;
528 * This is called from kswapd when we think we need some
532 static int shrink_dqcache_memory(int nr
, gfp_t gfp_mask
)
535 spin_lock(&dq_list_lock
);
537 spin_unlock(&dq_list_lock
);
539 return (dqstats
.free_dquots
/ 100) * sysctl_vfs_cache_pressure
;
543 * Put reference to dquot
544 * NOTE: If you change this function please check whether dqput_blocks() works right...
545 * MUST be called with either dqptr_sem or dqonoff_mutex held
547 static void dqput(struct dquot
*dquot
)
551 #ifdef __DQUOT_PARANOIA
552 if (!atomic_read(&dquot
->dq_count
)) {
553 printk("VFS: dqput: trying to free free dquot\n");
554 printk("VFS: device %s, dquot of %s %d\n",
556 quotatypes
[dquot
->dq_type
],
562 spin_lock(&dq_list_lock
);
564 spin_unlock(&dq_list_lock
);
566 spin_lock(&dq_list_lock
);
567 if (atomic_read(&dquot
->dq_count
) > 1) {
568 /* We have more than one user... nothing to do */
569 atomic_dec(&dquot
->dq_count
);
570 /* Releasing dquot during quotaoff phase? */
571 if (!sb_has_quota_enabled(dquot
->dq_sb
, dquot
->dq_type
) &&
572 atomic_read(&dquot
->dq_count
) == 1)
573 wake_up(&dquot
->dq_wait_unused
);
574 spin_unlock(&dq_list_lock
);
577 /* Need to release dquot? */
578 if (test_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
) && dquot_dirty(dquot
)) {
579 spin_unlock(&dq_list_lock
);
580 /* Commit dquot before releasing */
581 dquot
->dq_sb
->dq_op
->write_dquot(dquot
);
584 /* Clear flag in case dquot was inactive (something bad happened) */
585 clear_dquot_dirty(dquot
);
586 if (test_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
)) {
587 spin_unlock(&dq_list_lock
);
588 dquot
->dq_sb
->dq_op
->release_dquot(dquot
);
591 atomic_dec(&dquot
->dq_count
);
592 #ifdef __DQUOT_PARANOIA
594 BUG_ON(!list_empty(&dquot
->dq_free
));
596 put_dquot_last(dquot
);
597 spin_unlock(&dq_list_lock
);
600 static struct dquot
*get_empty_dquot(struct super_block
*sb
, int type
)
604 dquot
= kmem_cache_zalloc(dquot_cachep
, GFP_NOFS
);
608 mutex_init(&dquot
->dq_lock
);
609 INIT_LIST_HEAD(&dquot
->dq_free
);
610 INIT_LIST_HEAD(&dquot
->dq_inuse
);
611 INIT_HLIST_NODE(&dquot
->dq_hash
);
612 INIT_LIST_HEAD(&dquot
->dq_dirty
);
613 init_waitqueue_head(&dquot
->dq_wait_unused
);
615 dquot
->dq_type
= type
;
616 atomic_set(&dquot
->dq_count
, 1);
622 * Get reference to dquot
623 * MUST be called with either dqptr_sem or dqonoff_mutex held
625 static struct dquot
*dqget(struct super_block
*sb
, unsigned int id
, int type
)
627 unsigned int hashent
= hashfn(sb
, id
, type
);
628 struct dquot
*dquot
, *empty
= NODQUOT
;
630 if (!sb_has_quota_enabled(sb
, type
))
633 spin_lock(&dq_list_lock
);
634 if ((dquot
= find_dquot(hashent
, sb
, id
, type
)) == NODQUOT
) {
635 if (empty
== NODQUOT
) {
636 spin_unlock(&dq_list_lock
);
637 if ((empty
= get_empty_dquot(sb
, type
)) == NODQUOT
)
638 schedule(); /* Try to wait for a moment... */
643 /* all dquots go on the inuse_list */
645 /* hash it first so it can be found */
646 insert_dquot_hash(dquot
);
648 spin_unlock(&dq_list_lock
);
650 if (!atomic_read(&dquot
->dq_count
))
651 remove_free_dquot(dquot
);
652 atomic_inc(&dquot
->dq_count
);
653 dqstats
.cache_hits
++;
655 spin_unlock(&dq_list_lock
);
657 kmem_cache_free(dquot_cachep
, empty
);
659 /* Wait for dq_lock - after this we know that either dquot_release() is already
660 * finished or it will be canceled due to dq_count > 1 test */
661 wait_on_dquot(dquot
);
662 /* Read the dquot and instantiate it (everything done only if needed) */
663 if (!test_bit(DQ_ACTIVE_B
, &dquot
->dq_flags
) && sb
->dq_op
->acquire_dquot(dquot
) < 0) {
667 #ifdef __DQUOT_PARANOIA
668 BUG_ON(!dquot
->dq_sb
); /* Has somebody invalidated entry under us? */
674 static int dqinit_needed(struct inode
*inode
, int type
)
678 if (IS_NOQUOTA(inode
))
681 return inode
->i_dquot
[type
] == NODQUOT
;
682 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
683 if (inode
->i_dquot
[cnt
] == NODQUOT
)
688 /* This routine is guarded by dqonoff_mutex mutex */
689 static void add_dquot_ref(struct super_block
*sb
, int type
)
694 spin_lock(&inode_lock
);
695 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
696 if (!atomic_read(&inode
->i_writecount
))
698 if (!dqinit_needed(inode
, type
))
700 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
))
704 spin_unlock(&inode_lock
);
706 sb
->dq_op
->initialize(inode
, type
);
708 /* As we may have blocked we had better restart... */
711 spin_unlock(&inode_lock
);
714 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
715 static inline int dqput_blocks(struct dquot
*dquot
)
717 if (atomic_read(&dquot
->dq_count
) <= 1)
722 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
723 /* We can't race with anybody because we hold dqptr_sem for writing... */
724 static int remove_inode_dquot_ref(struct inode
*inode
, int type
,
725 struct list_head
*tofree_head
)
727 struct dquot
*dquot
= inode
->i_dquot
[type
];
729 inode
->i_dquot
[type
] = NODQUOT
;
730 if (dquot
!= NODQUOT
) {
731 if (dqput_blocks(dquot
)) {
732 #ifdef __DQUOT_PARANOIA
733 if (atomic_read(&dquot
->dq_count
) != 1)
734 printk(KERN_WARNING
"VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot
->dq_count
));
736 spin_lock(&dq_list_lock
);
737 list_add(&dquot
->dq_free
, tofree_head
); /* As dquot must have currently users it can't be on the free list... */
738 spin_unlock(&dq_list_lock
);
742 dqput(dquot
); /* We have guaranteed we won't block */
747 /* Free list of dquots - called from inode.c */
748 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
749 static void put_dquot_list(struct list_head
*tofree_head
)
751 struct list_head
*act_head
;
754 act_head
= tofree_head
->next
;
755 /* So now we have dquots on the list... Just free them */
756 while (act_head
!= tofree_head
) {
757 dquot
= list_entry(act_head
, struct dquot
, dq_free
);
758 act_head
= act_head
->next
;
759 list_del_init(&dquot
->dq_free
); /* Remove dquot from the list so we won't have problems... */
764 static void remove_dquot_ref(struct super_block
*sb
, int type
,
765 struct list_head
*tofree_head
)
769 spin_lock(&inode_lock
);
770 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
771 if (!IS_NOQUOTA(inode
))
772 remove_inode_dquot_ref(inode
, type
, tofree_head
);
774 spin_unlock(&inode_lock
);
777 /* Gather all references from inodes and drop them */
778 static void drop_dquot_ref(struct super_block
*sb
, int type
)
780 LIST_HEAD(tofree_head
);
783 down_write(&sb_dqopt(sb
)->dqptr_sem
);
784 remove_dquot_ref(sb
, type
, &tofree_head
);
785 up_write(&sb_dqopt(sb
)->dqptr_sem
);
786 put_dquot_list(&tofree_head
);
790 static inline void dquot_incr_inodes(struct dquot
*dquot
, unsigned long number
)
792 dquot
->dq_dqb
.dqb_curinodes
+= number
;
795 static inline void dquot_incr_space(struct dquot
*dquot
, qsize_t number
)
797 dquot
->dq_dqb
.dqb_curspace
+= number
;
800 static inline void dquot_decr_inodes(struct dquot
*dquot
, unsigned long number
)
802 if (dquot
->dq_dqb
.dqb_curinodes
> number
)
803 dquot
->dq_dqb
.dqb_curinodes
-= number
;
805 dquot
->dq_dqb
.dqb_curinodes
= 0;
806 if (dquot
->dq_dqb
.dqb_curinodes
<= dquot
->dq_dqb
.dqb_isoftlimit
)
807 dquot
->dq_dqb
.dqb_itime
= (time_t) 0;
808 clear_bit(DQ_INODES_B
, &dquot
->dq_flags
);
811 static inline void dquot_decr_space(struct dquot
*dquot
, qsize_t number
)
813 if (dquot
->dq_dqb
.dqb_curspace
> number
)
814 dquot
->dq_dqb
.dqb_curspace
-= number
;
816 dquot
->dq_dqb
.dqb_curspace
= 0;
817 if (toqb(dquot
->dq_dqb
.dqb_curspace
) <= dquot
->dq_dqb
.dqb_bsoftlimit
)
818 dquot
->dq_dqb
.dqb_btime
= (time_t) 0;
819 clear_bit(DQ_BLKS_B
, &dquot
->dq_flags
);
822 static int flag_print_warnings
= 1;
824 static inline int need_print_warning(struct dquot
*dquot
)
826 if (!flag_print_warnings
)
829 switch (dquot
->dq_type
) {
831 return current
->fsuid
== dquot
->dq_id
;
833 return in_group_p(dquot
->dq_id
);
838 /* Values of warnings */
841 #define ISOFTLONGWARN 2
844 #define BSOFTLONGWARN 5
847 /* Print warning to user which exceeded quota */
848 static void print_warning(struct dquot
*dquot
, const char warntype
)
851 struct tty_struct
*tty
;
852 int flag
= (warntype
== BHARDWARN
|| warntype
== BSOFTLONGWARN
) ? DQ_BLKS_B
:
853 ((warntype
== IHARDWARN
|| warntype
== ISOFTLONGWARN
) ? DQ_INODES_B
: 0);
855 if (!need_print_warning(dquot
) || (flag
&& test_and_set_bit(flag
, &dquot
->dq_flags
)))
858 mutex_lock(&tty_mutex
);
859 tty
= get_current_tty();
862 tty_write_message(tty
, dquot
->dq_sb
->s_id
);
863 if (warntype
== ISOFTWARN
|| warntype
== BSOFTWARN
)
864 tty_write_message(tty
, ": warning, ");
866 tty_write_message(tty
, ": write failed, ");
867 tty_write_message(tty
, quotatypes
[dquot
->dq_type
]);
870 msg
= " file limit reached.\r\n";
873 msg
= " file quota exceeded too long.\r\n";
876 msg
= " file quota exceeded.\r\n";
879 msg
= " block limit reached.\r\n";
882 msg
= " block quota exceeded too long.\r\n";
885 msg
= " block quota exceeded.\r\n";
888 tty_write_message(tty
, msg
);
890 mutex_unlock(&tty_mutex
);
893 static inline void flush_warnings(struct dquot
**dquots
, char *warntype
)
897 for (i
= 0; i
< MAXQUOTAS
; i
++)
898 if (dquots
[i
] != NODQUOT
&& warntype
[i
] != NOWARN
)
899 print_warning(dquots
[i
], warntype
[i
]);
902 static inline char ignore_hardlimit(struct dquot
*dquot
)
904 struct mem_dqinfo
*info
= &sb_dqopt(dquot
->dq_sb
)->info
[dquot
->dq_type
];
906 return capable(CAP_SYS_RESOURCE
) &&
907 (info
->dqi_format
->qf_fmt_id
!= QFMT_VFS_OLD
|| !(info
->dqi_flags
& V1_DQF_RSQUASH
));
910 /* needs dq_data_lock */
911 static int check_idq(struct dquot
*dquot
, ulong inodes
, char *warntype
)
914 if (inodes
<= 0 || test_bit(DQ_FAKE_B
, &dquot
->dq_flags
))
917 if (dquot
->dq_dqb
.dqb_ihardlimit
&&
918 (dquot
->dq_dqb
.dqb_curinodes
+ inodes
) > dquot
->dq_dqb
.dqb_ihardlimit
&&
919 !ignore_hardlimit(dquot
)) {
920 *warntype
= IHARDWARN
;
924 if (dquot
->dq_dqb
.dqb_isoftlimit
&&
925 (dquot
->dq_dqb
.dqb_curinodes
+ inodes
) > dquot
->dq_dqb
.dqb_isoftlimit
&&
926 dquot
->dq_dqb
.dqb_itime
&& get_seconds() >= dquot
->dq_dqb
.dqb_itime
&&
927 !ignore_hardlimit(dquot
)) {
928 *warntype
= ISOFTLONGWARN
;
932 if (dquot
->dq_dqb
.dqb_isoftlimit
&&
933 (dquot
->dq_dqb
.dqb_curinodes
+ inodes
) > dquot
->dq_dqb
.dqb_isoftlimit
&&
934 dquot
->dq_dqb
.dqb_itime
== 0) {
935 *warntype
= ISOFTWARN
;
936 dquot
->dq_dqb
.dqb_itime
= get_seconds() + sb_dqopt(dquot
->dq_sb
)->info
[dquot
->dq_type
].dqi_igrace
;
942 /* needs dq_data_lock */
943 static int check_bdq(struct dquot
*dquot
, qsize_t space
, int prealloc
, char *warntype
)
946 if (space
<= 0 || test_bit(DQ_FAKE_B
, &dquot
->dq_flags
))
949 if (dquot
->dq_dqb
.dqb_bhardlimit
&&
950 toqb(dquot
->dq_dqb
.dqb_curspace
+ space
) > dquot
->dq_dqb
.dqb_bhardlimit
&&
951 !ignore_hardlimit(dquot
)) {
953 *warntype
= BHARDWARN
;
957 if (dquot
->dq_dqb
.dqb_bsoftlimit
&&
958 toqb(dquot
->dq_dqb
.dqb_curspace
+ space
) > dquot
->dq_dqb
.dqb_bsoftlimit
&&
959 dquot
->dq_dqb
.dqb_btime
&& get_seconds() >= dquot
->dq_dqb
.dqb_btime
&&
960 !ignore_hardlimit(dquot
)) {
962 *warntype
= BSOFTLONGWARN
;
966 if (dquot
->dq_dqb
.dqb_bsoftlimit
&&
967 toqb(dquot
->dq_dqb
.dqb_curspace
+ space
) > dquot
->dq_dqb
.dqb_bsoftlimit
&&
968 dquot
->dq_dqb
.dqb_btime
== 0) {
970 *warntype
= BSOFTWARN
;
971 dquot
->dq_dqb
.dqb_btime
= get_seconds() + sb_dqopt(dquot
->dq_sb
)->info
[dquot
->dq_type
].dqi_bgrace
;
975 * We don't allow preallocation to exceed softlimit so exceeding will
985 * Initialize quota pointers in inode
986 * Transaction must be started at entry
988 int dquot_initialize(struct inode
*inode
, int type
)
993 /* First test before acquiring mutex - solves deadlocks when we
994 * re-enter the quota code and are already holding the mutex */
995 if (IS_NOQUOTA(inode
))
997 down_write(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
998 /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
999 if (IS_NOQUOTA(inode
))
1001 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1002 if (type
!= -1 && cnt
!= type
)
1004 if (inode
->i_dquot
[cnt
] == NODQUOT
) {
1013 inode
->i_dquot
[cnt
] = dqget(inode
->i_sb
, id
, cnt
);
1017 up_write(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1022 * Release all quotas referenced by inode
1023 * Transaction must be started at an entry
1025 int dquot_drop(struct inode
*inode
)
1029 down_write(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1030 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1031 if (inode
->i_dquot
[cnt
] != NODQUOT
) {
1032 dqput(inode
->i_dquot
[cnt
]);
1033 inode
->i_dquot
[cnt
] = NODQUOT
;
1036 up_write(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1041 * Following four functions update i_blocks+i_bytes fields and
1042 * quota information (together with appropriate checks)
1043 * NOTE: We absolutely rely on the fact that caller dirties
1044 * the inode (usually macros in quotaops.h care about this) and
1045 * holds a handle for the current transaction so that dquot write and
1046 * inode write go into the same transaction.
1050 * This operation can block, but only after everything is updated
1052 int dquot_alloc_space(struct inode
*inode
, qsize_t number
, int warn
)
1054 int cnt
, ret
= NO_QUOTA
;
1055 char warntype
[MAXQUOTAS
];
1057 /* First test before acquiring mutex - solves deadlocks when we
1058 * re-enter the quota code and are already holding the mutex */
1059 if (IS_NOQUOTA(inode
)) {
1061 inode_add_bytes(inode
, number
);
1064 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
1065 warntype
[cnt
] = NOWARN
;
1067 down_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1068 if (IS_NOQUOTA(inode
)) { /* Now we can do reliable test... */
1069 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1072 spin_lock(&dq_data_lock
);
1073 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1074 if (inode
->i_dquot
[cnt
] == NODQUOT
)
1076 if (check_bdq(inode
->i_dquot
[cnt
], number
, warn
, warntype
+cnt
) == NO_QUOTA
)
1079 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1080 if (inode
->i_dquot
[cnt
] == NODQUOT
)
1082 dquot_incr_space(inode
->i_dquot
[cnt
], number
);
1084 inode_add_bytes(inode
, number
);
1087 spin_unlock(&dq_data_lock
);
1088 if (ret
== QUOTA_OK
)
1089 /* Dirtify all the dquots - this can block when journalling */
1090 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
1091 if (inode
->i_dquot
[cnt
])
1092 mark_dquot_dirty(inode
->i_dquot
[cnt
]);
1093 flush_warnings(inode
->i_dquot
, warntype
);
1094 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1099 * This operation can block, but only after everything is updated
1101 int dquot_alloc_inode(const struct inode
*inode
, unsigned long number
)
1103 int cnt
, ret
= NO_QUOTA
;
1104 char warntype
[MAXQUOTAS
];
1106 /* First test before acquiring mutex - solves deadlocks when we
1107 * re-enter the quota code and are already holding the mutex */
1108 if (IS_NOQUOTA(inode
))
1110 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
1111 warntype
[cnt
] = NOWARN
;
1112 down_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1113 if (IS_NOQUOTA(inode
)) {
1114 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1117 spin_lock(&dq_data_lock
);
1118 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1119 if (inode
->i_dquot
[cnt
] == NODQUOT
)
1121 if (check_idq(inode
->i_dquot
[cnt
], number
, warntype
+cnt
) == NO_QUOTA
)
1125 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1126 if (inode
->i_dquot
[cnt
] == NODQUOT
)
1128 dquot_incr_inodes(inode
->i_dquot
[cnt
], number
);
1132 spin_unlock(&dq_data_lock
);
1133 if (ret
== QUOTA_OK
)
1134 /* Dirtify all the dquots - this can block when journalling */
1135 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
1136 if (inode
->i_dquot
[cnt
])
1137 mark_dquot_dirty(inode
->i_dquot
[cnt
]);
1138 flush_warnings((struct dquot
**)inode
->i_dquot
, warntype
);
1139 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1144 * This operation can block, but only after everything is updated
1146 int dquot_free_space(struct inode
*inode
, qsize_t number
)
1150 /* First test before acquiring mutex - solves deadlocks when we
1151 * re-enter the quota code and are already holding the mutex */
1152 if (IS_NOQUOTA(inode
)) {
1154 inode_sub_bytes(inode
, number
);
1157 down_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1158 /* Now recheck reliably when holding dqptr_sem */
1159 if (IS_NOQUOTA(inode
)) {
1160 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1163 spin_lock(&dq_data_lock
);
1164 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1165 if (inode
->i_dquot
[cnt
] == NODQUOT
)
1167 dquot_decr_space(inode
->i_dquot
[cnt
], number
);
1169 inode_sub_bytes(inode
, number
);
1170 spin_unlock(&dq_data_lock
);
1171 /* Dirtify all the dquots - this can block when journalling */
1172 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
1173 if (inode
->i_dquot
[cnt
])
1174 mark_dquot_dirty(inode
->i_dquot
[cnt
]);
1175 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1180 * This operation can block, but only after everything is updated
1182 int dquot_free_inode(const struct inode
*inode
, unsigned long number
)
1186 /* First test before acquiring mutex - solves deadlocks when we
1187 * re-enter the quota code and are already holding the mutex */
1188 if (IS_NOQUOTA(inode
))
1190 down_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1191 /* Now recheck reliably when holding dqptr_sem */
1192 if (IS_NOQUOTA(inode
)) {
1193 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1196 spin_lock(&dq_data_lock
);
1197 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1198 if (inode
->i_dquot
[cnt
] == NODQUOT
)
1200 dquot_decr_inodes(inode
->i_dquot
[cnt
], number
);
1202 spin_unlock(&dq_data_lock
);
1203 /* Dirtify all the dquots - this can block when journalling */
1204 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
1205 if (inode
->i_dquot
[cnt
])
1206 mark_dquot_dirty(inode
->i_dquot
[cnt
]);
1207 up_read(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1212 * Transfer the number of inode and blocks from one diskquota to an other.
1214 * This operation can block, but only after everything is updated
1215 * A transaction must be started when entering this function.
1217 int dquot_transfer(struct inode
*inode
, struct iattr
*iattr
)
1220 struct dquot
*transfer_from
[MAXQUOTAS
];
1221 struct dquot
*transfer_to
[MAXQUOTAS
];
1222 int cnt
, ret
= NO_QUOTA
, chuid
= (iattr
->ia_valid
& ATTR_UID
) && inode
->i_uid
!= iattr
->ia_uid
,
1223 chgid
= (iattr
->ia_valid
& ATTR_GID
) && inode
->i_gid
!= iattr
->ia_gid
;
1224 char warntype
[MAXQUOTAS
];
1226 /* First test before acquiring mutex - solves deadlocks when we
1227 * re-enter the quota code and are already holding the mutex */
1228 if (IS_NOQUOTA(inode
))
1230 /* Clear the arrays */
1231 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1232 transfer_to
[cnt
] = transfer_from
[cnt
] = NODQUOT
;
1233 warntype
[cnt
] = NOWARN
;
1235 down_write(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1236 /* Now recheck reliably when holding dqptr_sem */
1237 if (IS_NOQUOTA(inode
)) { /* File without quota accounting? */
1238 up_write(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1241 /* First build the transfer_to list - here we can block on
1242 * reading/instantiating of dquots. We know that the transaction for
1243 * us was already started so we don't violate lock ranking here */
1244 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1249 transfer_to
[cnt
] = dqget(inode
->i_sb
, iattr
->ia_uid
, cnt
);
1254 transfer_to
[cnt
] = dqget(inode
->i_sb
, iattr
->ia_gid
, cnt
);
1258 spin_lock(&dq_data_lock
);
1259 space
= inode_get_bytes(inode
);
1260 /* Build the transfer_from list and check the limits */
1261 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1262 if (transfer_to
[cnt
] == NODQUOT
)
1264 transfer_from
[cnt
] = inode
->i_dquot
[cnt
];
1265 if (check_idq(transfer_to
[cnt
], 1, warntype
+cnt
) == NO_QUOTA
||
1266 check_bdq(transfer_to
[cnt
], space
, 0, warntype
+cnt
) == NO_QUOTA
)
1271 * Finally perform the needed transfer from transfer_from to transfer_to
1273 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1275 * Skip changes for same uid or gid or for turned off quota-type.
1277 if (transfer_to
[cnt
] == NODQUOT
)
1280 /* Due to IO error we might not have transfer_from[] structure */
1281 if (transfer_from
[cnt
]) {
1282 dquot_decr_inodes(transfer_from
[cnt
], 1);
1283 dquot_decr_space(transfer_from
[cnt
], space
);
1286 dquot_incr_inodes(transfer_to
[cnt
], 1);
1287 dquot_incr_space(transfer_to
[cnt
], space
);
1289 inode
->i_dquot
[cnt
] = transfer_to
[cnt
];
1293 spin_unlock(&dq_data_lock
);
1294 /* Dirtify all the dquots - this can block when journalling */
1295 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1296 if (transfer_from
[cnt
])
1297 mark_dquot_dirty(transfer_from
[cnt
]);
1298 if (transfer_to
[cnt
])
1299 mark_dquot_dirty(transfer_to
[cnt
]);
1301 flush_warnings(transfer_to
, warntype
);
1303 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1304 if (ret
== QUOTA_OK
&& transfer_from
[cnt
] != NODQUOT
)
1305 dqput(transfer_from
[cnt
]);
1306 if (ret
== NO_QUOTA
&& transfer_to
[cnt
] != NODQUOT
)
1307 dqput(transfer_to
[cnt
]);
1309 up_write(&sb_dqopt(inode
->i_sb
)->dqptr_sem
);
1314 * Write info of quota file to disk
1316 int dquot_commit_info(struct super_block
*sb
, int type
)
1319 struct quota_info
*dqopt
= sb_dqopt(sb
);
1321 mutex_lock(&dqopt
->dqio_mutex
);
1322 ret
= dqopt
->ops
[type
]->write_file_info(sb
, type
);
1323 mutex_unlock(&dqopt
->dqio_mutex
);
1328 * Definitions of diskquota operations.
1330 struct dquot_operations dquot_operations
= {
1331 .initialize
= dquot_initialize
,
1333 .alloc_space
= dquot_alloc_space
,
1334 .alloc_inode
= dquot_alloc_inode
,
1335 .free_space
= dquot_free_space
,
1336 .free_inode
= dquot_free_inode
,
1337 .transfer
= dquot_transfer
,
1338 .write_dquot
= dquot_commit
,
1339 .acquire_dquot
= dquot_acquire
,
1340 .release_dquot
= dquot_release
,
1341 .mark_dirty
= dquot_mark_dquot_dirty
,
1342 .write_info
= dquot_commit_info
1345 static inline void set_enable_flags(struct quota_info
*dqopt
, int type
)
1349 dqopt
->flags
|= DQUOT_USR_ENABLED
;
1352 dqopt
->flags
|= DQUOT_GRP_ENABLED
;
1357 static inline void reset_enable_flags(struct quota_info
*dqopt
, int type
)
1361 dqopt
->flags
&= ~DQUOT_USR_ENABLED
;
1364 dqopt
->flags
&= ~DQUOT_GRP_ENABLED
;
1370 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1372 int vfs_quota_off(struct super_block
*sb
, int type
)
1375 struct quota_info
*dqopt
= sb_dqopt(sb
);
1376 struct inode
*toputinode
[MAXQUOTAS
];
1378 /* We need to serialize quota_off() for device */
1379 mutex_lock(&dqopt
->dqonoff_mutex
);
1380 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++) {
1381 toputinode
[cnt
] = NULL
;
1382 if (type
!= -1 && cnt
!= type
)
1384 if (!sb_has_quota_enabled(sb
, cnt
))
1386 reset_enable_flags(dqopt
, cnt
);
1388 /* Note: these are blocking operations */
1389 drop_dquot_ref(sb
, cnt
);
1390 invalidate_dquots(sb
, cnt
);
1392 * Now all dquots should be invalidated, all writes done so we should be only
1393 * users of the info. No locks needed.
1395 if (info_dirty(&dqopt
->info
[cnt
]))
1396 sb
->dq_op
->write_info(sb
, cnt
);
1397 if (dqopt
->ops
[cnt
]->free_file_info
)
1398 dqopt
->ops
[cnt
]->free_file_info(sb
, cnt
);
1399 put_quota_format(dqopt
->info
[cnt
].dqi_format
);
1401 toputinode
[cnt
] = dqopt
->files
[cnt
];
1402 dqopt
->files
[cnt
] = NULL
;
1403 dqopt
->info
[cnt
].dqi_flags
= 0;
1404 dqopt
->info
[cnt
].dqi_igrace
= 0;
1405 dqopt
->info
[cnt
].dqi_bgrace
= 0;
1406 dqopt
->ops
[cnt
] = NULL
;
1408 mutex_unlock(&dqopt
->dqonoff_mutex
);
1409 /* Sync the superblock so that buffers with quota data are written to
1410 * disk (and so userspace sees correct data afterwards). */
1411 if (sb
->s_op
->sync_fs
)
1412 sb
->s_op
->sync_fs(sb
, 1);
1413 sync_blockdev(sb
->s_bdev
);
1414 /* Now the quota files are just ordinary files and we can set the
1415 * inode flags back. Moreover we discard the pagecache so that
1416 * userspace sees the writes we did bypassing the pagecache. We
1417 * must also discard the blockdev buffers so that we see the
1418 * changes done by userspace on the next quotaon() */
1419 for (cnt
= 0; cnt
< MAXQUOTAS
; cnt
++)
1420 if (toputinode
[cnt
]) {
1421 mutex_lock(&dqopt
->dqonoff_mutex
);
1422 /* If quota was reenabled in the meantime, we have
1424 if (!sb_has_quota_enabled(sb
, cnt
)) {
1425 mutex_lock(&toputinode
[cnt
]->i_mutex
);
1426 toputinode
[cnt
]->i_flags
&= ~(S_IMMUTABLE
|
1427 S_NOATIME
| S_NOQUOTA
);
1428 truncate_inode_pages(&toputinode
[cnt
]->i_data
, 0);
1429 mutex_unlock(&toputinode
[cnt
]->i_mutex
);
1430 mark_inode_dirty(toputinode
[cnt
]);
1431 iput(toputinode
[cnt
]);
1433 mutex_unlock(&dqopt
->dqonoff_mutex
);
1436 invalidate_bdev(sb
->s_bdev
);
1441 * Turn quotas on on a device
1444 /* Helper function when we already have the inode */
1445 static int vfs_quota_on_inode(struct inode
*inode
, int type
, int format_id
)
1447 struct quota_format_type
*fmt
= find_quota_format(format_id
);
1448 struct super_block
*sb
= inode
->i_sb
;
1449 struct quota_info
*dqopt
= sb_dqopt(sb
);
1455 if (!S_ISREG(inode
->i_mode
)) {
1459 if (IS_RDONLY(inode
)) {
1463 if (!sb
->s_op
->quota_write
|| !sb
->s_op
->quota_read
) {
1468 /* As we bypass the pagecache we must now flush the inode so that
1469 * we see all the changes from userspace... */
1470 write_inode_now(inode
, 1);
1471 /* And now flush the block cache so that kernel sees the changes */
1472 invalidate_bdev(sb
->s_bdev
);
1473 mutex_lock(&inode
->i_mutex
);
1474 mutex_lock(&dqopt
->dqonoff_mutex
);
1475 if (sb_has_quota_enabled(sb
, type
)) {
1479 /* We don't want quota and atime on quota files (deadlocks possible)
1480 * Also nobody should write to the file - we use special IO operations
1481 * which ignore the immutable bit. */
1482 down_write(&dqopt
->dqptr_sem
);
1483 oldflags
= inode
->i_flags
& (S_NOATIME
| S_IMMUTABLE
| S_NOQUOTA
);
1484 inode
->i_flags
|= S_NOQUOTA
| S_NOATIME
| S_IMMUTABLE
;
1485 up_write(&dqopt
->dqptr_sem
);
1486 sb
->dq_op
->drop(inode
);
1489 dqopt
->files
[type
] = igrab(inode
);
1490 if (!dqopt
->files
[type
])
1493 if (!fmt
->qf_ops
->check_quota_file(sb
, type
))
1496 dqopt
->ops
[type
] = fmt
->qf_ops
;
1497 dqopt
->info
[type
].dqi_format
= fmt
;
1498 INIT_LIST_HEAD(&dqopt
->info
[type
].dqi_dirty_list
);
1499 mutex_lock(&dqopt
->dqio_mutex
);
1500 if ((error
= dqopt
->ops
[type
]->read_file_info(sb
, type
)) < 0) {
1501 mutex_unlock(&dqopt
->dqio_mutex
);
1504 mutex_unlock(&dqopt
->dqio_mutex
);
1505 mutex_unlock(&inode
->i_mutex
);
1506 set_enable_flags(dqopt
, type
);
1508 add_dquot_ref(sb
, type
);
1509 mutex_unlock(&dqopt
->dqonoff_mutex
);
1514 dqopt
->files
[type
] = NULL
;
1517 mutex_unlock(&dqopt
->dqonoff_mutex
);
1518 if (oldflags
!= -1) {
1519 down_write(&dqopt
->dqptr_sem
);
1520 /* Set the flags back (in the case of accidental quotaon()
1521 * on a wrong file we don't want to mess up the flags) */
1522 inode
->i_flags
&= ~(S_NOATIME
| S_NOQUOTA
| S_IMMUTABLE
);
1523 inode
->i_flags
|= oldflags
;
1524 up_write(&dqopt
->dqptr_sem
);
1526 mutex_unlock(&inode
->i_mutex
);
1528 put_quota_format(fmt
);
1533 /* Actual function called from quotactl() */
1534 int vfs_quota_on(struct super_block
*sb
, int type
, int format_id
, char *path
)
1536 struct nameidata nd
;
1539 error
= path_lookup(path
, LOOKUP_FOLLOW
, &nd
);
1542 error
= security_quota_on(nd
.dentry
);
1545 /* Quota file not on the same filesystem? */
1546 if (nd
.mnt
->mnt_sb
!= sb
)
1549 error
= vfs_quota_on_inode(nd
.dentry
->d_inode
, type
, format_id
);
1556 * This function is used when filesystem needs to initialize quotas
1557 * during mount time.
1559 int vfs_quota_on_mount(struct super_block
*sb
, char *qf_name
,
1560 int format_id
, int type
)
1562 struct dentry
*dentry
;
1565 dentry
= lookup_one_len(qf_name
, sb
->s_root
, strlen(qf_name
));
1567 return PTR_ERR(dentry
);
1569 if (!dentry
->d_inode
) {
1574 error
= security_quota_on(dentry
);
1576 error
= vfs_quota_on_inode(dentry
->d_inode
, type
, format_id
);
1583 /* Generic routine for getting common part of quota structure */
1584 static void do_get_dqblk(struct dquot
*dquot
, struct if_dqblk
*di
)
1586 struct mem_dqblk
*dm
= &dquot
->dq_dqb
;
1588 spin_lock(&dq_data_lock
);
1589 di
->dqb_bhardlimit
= dm
->dqb_bhardlimit
;
1590 di
->dqb_bsoftlimit
= dm
->dqb_bsoftlimit
;
1591 di
->dqb_curspace
= dm
->dqb_curspace
;
1592 di
->dqb_ihardlimit
= dm
->dqb_ihardlimit
;
1593 di
->dqb_isoftlimit
= dm
->dqb_isoftlimit
;
1594 di
->dqb_curinodes
= dm
->dqb_curinodes
;
1595 di
->dqb_btime
= dm
->dqb_btime
;
1596 di
->dqb_itime
= dm
->dqb_itime
;
1597 di
->dqb_valid
= QIF_ALL
;
1598 spin_unlock(&dq_data_lock
);
1601 int vfs_get_dqblk(struct super_block
*sb
, int type
, qid_t id
, struct if_dqblk
*di
)
1603 struct dquot
*dquot
;
1605 mutex_lock(&sb_dqopt(sb
)->dqonoff_mutex
);
1606 if (!(dquot
= dqget(sb
, id
, type
))) {
1607 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1610 do_get_dqblk(dquot
, di
);
1612 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1616 /* Generic routine for setting common part of quota structure */
1617 static void do_set_dqblk(struct dquot
*dquot
, struct if_dqblk
*di
)
1619 struct mem_dqblk
*dm
= &dquot
->dq_dqb
;
1620 int check_blim
= 0, check_ilim
= 0;
1622 spin_lock(&dq_data_lock
);
1623 if (di
->dqb_valid
& QIF_SPACE
) {
1624 dm
->dqb_curspace
= di
->dqb_curspace
;
1627 if (di
->dqb_valid
& QIF_BLIMITS
) {
1628 dm
->dqb_bsoftlimit
= di
->dqb_bsoftlimit
;
1629 dm
->dqb_bhardlimit
= di
->dqb_bhardlimit
;
1632 if (di
->dqb_valid
& QIF_INODES
) {
1633 dm
->dqb_curinodes
= di
->dqb_curinodes
;
1636 if (di
->dqb_valid
& QIF_ILIMITS
) {
1637 dm
->dqb_isoftlimit
= di
->dqb_isoftlimit
;
1638 dm
->dqb_ihardlimit
= di
->dqb_ihardlimit
;
1641 if (di
->dqb_valid
& QIF_BTIME
)
1642 dm
->dqb_btime
= di
->dqb_btime
;
1643 if (di
->dqb_valid
& QIF_ITIME
)
1644 dm
->dqb_itime
= di
->dqb_itime
;
1647 if (!dm
->dqb_bsoftlimit
|| toqb(dm
->dqb_curspace
) < dm
->dqb_bsoftlimit
) {
1649 clear_bit(DQ_BLKS_B
, &dquot
->dq_flags
);
1651 else if (!(di
->dqb_valid
& QIF_BTIME
)) /* Set grace only if user hasn't provided his own... */
1652 dm
->dqb_btime
= get_seconds() + sb_dqopt(dquot
->dq_sb
)->info
[dquot
->dq_type
].dqi_bgrace
;
1655 if (!dm
->dqb_isoftlimit
|| dm
->dqb_curinodes
< dm
->dqb_isoftlimit
) {
1657 clear_bit(DQ_INODES_B
, &dquot
->dq_flags
);
1659 else if (!(di
->dqb_valid
& QIF_ITIME
)) /* Set grace only if user hasn't provided his own... */
1660 dm
->dqb_itime
= get_seconds() + sb_dqopt(dquot
->dq_sb
)->info
[dquot
->dq_type
].dqi_igrace
;
1662 if (dm
->dqb_bhardlimit
|| dm
->dqb_bsoftlimit
|| dm
->dqb_ihardlimit
|| dm
->dqb_isoftlimit
)
1663 clear_bit(DQ_FAKE_B
, &dquot
->dq_flags
);
1665 set_bit(DQ_FAKE_B
, &dquot
->dq_flags
);
1666 spin_unlock(&dq_data_lock
);
1667 mark_dquot_dirty(dquot
);
1670 int vfs_set_dqblk(struct super_block
*sb
, int type
, qid_t id
, struct if_dqblk
*di
)
1672 struct dquot
*dquot
;
1674 mutex_lock(&sb_dqopt(sb
)->dqonoff_mutex
);
1675 if (!(dquot
= dqget(sb
, id
, type
))) {
1676 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1679 do_set_dqblk(dquot
, di
);
1681 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1685 /* Generic routine for getting common part of quota file information */
1686 int vfs_get_dqinfo(struct super_block
*sb
, int type
, struct if_dqinfo
*ii
)
1688 struct mem_dqinfo
*mi
;
1690 mutex_lock(&sb_dqopt(sb
)->dqonoff_mutex
);
1691 if (!sb_has_quota_enabled(sb
, type
)) {
1692 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1695 mi
= sb_dqopt(sb
)->info
+ type
;
1696 spin_lock(&dq_data_lock
);
1697 ii
->dqi_bgrace
= mi
->dqi_bgrace
;
1698 ii
->dqi_igrace
= mi
->dqi_igrace
;
1699 ii
->dqi_flags
= mi
->dqi_flags
& DQF_MASK
;
1700 ii
->dqi_valid
= IIF_ALL
;
1701 spin_unlock(&dq_data_lock
);
1702 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1706 /* Generic routine for setting common part of quota file information */
1707 int vfs_set_dqinfo(struct super_block
*sb
, int type
, struct if_dqinfo
*ii
)
1709 struct mem_dqinfo
*mi
;
1711 mutex_lock(&sb_dqopt(sb
)->dqonoff_mutex
);
1712 if (!sb_has_quota_enabled(sb
, type
)) {
1713 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1716 mi
= sb_dqopt(sb
)->info
+ type
;
1717 spin_lock(&dq_data_lock
);
1718 if (ii
->dqi_valid
& IIF_BGRACE
)
1719 mi
->dqi_bgrace
= ii
->dqi_bgrace
;
1720 if (ii
->dqi_valid
& IIF_IGRACE
)
1721 mi
->dqi_igrace
= ii
->dqi_igrace
;
1722 if (ii
->dqi_valid
& IIF_FLAGS
)
1723 mi
->dqi_flags
= (mi
->dqi_flags
& ~DQF_MASK
) | (ii
->dqi_flags
& DQF_MASK
);
1724 spin_unlock(&dq_data_lock
);
1725 mark_info_dirty(sb
, type
);
1726 /* Force write to disk */
1727 sb
->dq_op
->write_info(sb
, type
);
1728 mutex_unlock(&sb_dqopt(sb
)->dqonoff_mutex
);
1732 struct quotactl_ops vfs_quotactl_ops
= {
1733 .quota_on
= vfs_quota_on
,
1734 .quota_off
= vfs_quota_off
,
1735 .quota_sync
= vfs_quota_sync
,
1736 .get_info
= vfs_get_dqinfo
,
1737 .set_info
= vfs_set_dqinfo
,
1738 .get_dqblk
= vfs_get_dqblk
,
1739 .set_dqblk
= vfs_set_dqblk
1742 static ctl_table fs_dqstats_table
[] = {
1744 .ctl_name
= FS_DQ_LOOKUPS
,
1745 .procname
= "lookups",
1746 .data
= &dqstats
.lookups
,
1747 .maxlen
= sizeof(int),
1749 .proc_handler
= &proc_dointvec
,
1752 .ctl_name
= FS_DQ_DROPS
,
1753 .procname
= "drops",
1754 .data
= &dqstats
.drops
,
1755 .maxlen
= sizeof(int),
1757 .proc_handler
= &proc_dointvec
,
1760 .ctl_name
= FS_DQ_READS
,
1761 .procname
= "reads",
1762 .data
= &dqstats
.reads
,
1763 .maxlen
= sizeof(int),
1765 .proc_handler
= &proc_dointvec
,
1768 .ctl_name
= FS_DQ_WRITES
,
1769 .procname
= "writes",
1770 .data
= &dqstats
.writes
,
1771 .maxlen
= sizeof(int),
1773 .proc_handler
= &proc_dointvec
,
1776 .ctl_name
= FS_DQ_CACHE_HITS
,
1777 .procname
= "cache_hits",
1778 .data
= &dqstats
.cache_hits
,
1779 .maxlen
= sizeof(int),
1781 .proc_handler
= &proc_dointvec
,
1784 .ctl_name
= FS_DQ_ALLOCATED
,
1785 .procname
= "allocated_dquots",
1786 .data
= &dqstats
.allocated_dquots
,
1787 .maxlen
= sizeof(int),
1789 .proc_handler
= &proc_dointvec
,
1792 .ctl_name
= FS_DQ_FREE
,
1793 .procname
= "free_dquots",
1794 .data
= &dqstats
.free_dquots
,
1795 .maxlen
= sizeof(int),
1797 .proc_handler
= &proc_dointvec
,
1800 .ctl_name
= FS_DQ_SYNCS
,
1801 .procname
= "syncs",
1802 .data
= &dqstats
.syncs
,
1803 .maxlen
= sizeof(int),
1805 .proc_handler
= &proc_dointvec
,
1808 .ctl_name
= FS_DQ_WARNINGS
,
1809 .procname
= "warnings",
1810 .data
= &flag_print_warnings
,
1811 .maxlen
= sizeof(int),
1813 .proc_handler
= &proc_dointvec
,
1818 static ctl_table fs_table
[] = {
1820 .ctl_name
= FS_DQSTATS
,
1821 .procname
= "quota",
1823 .child
= fs_dqstats_table
,
1828 static ctl_table sys_table
[] = {
1838 static int __init
dquot_init(void)
1841 unsigned long nr_hash
, order
;
1843 printk(KERN_NOTICE
"VFS: Disk quotas %s\n", __DQUOT_VERSION__
);
1845 register_sysctl_table(sys_table
);
1847 dquot_cachep
= kmem_cache_create("dquot",
1848 sizeof(struct dquot
), sizeof(unsigned long) * 4,
1849 (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
1850 SLAB_MEM_SPREAD
|SLAB_PANIC
),
1854 dquot_hash
= (struct hlist_head
*)__get_free_pages(GFP_ATOMIC
, order
);
1856 panic("Cannot create dquot hash table");
1858 /* Find power-of-two hlist_heads which can fit into allocation */
1859 nr_hash
= (1UL << order
) * PAGE_SIZE
/ sizeof(struct hlist_head
);
1863 } while (nr_hash
>> dq_hash_bits
);
1866 nr_hash
= 1UL << dq_hash_bits
;
1867 dq_hash_mask
= nr_hash
- 1;
1868 for (i
= 0; i
< nr_hash
; i
++)
1869 INIT_HLIST_HEAD(dquot_hash
+ i
);
1871 printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1872 nr_hash
, order
, (PAGE_SIZE
<< order
));
1874 set_shrinker(DEFAULT_SEEKS
, shrink_dqcache_memory
);
1878 module_init(dquot_init
);
1880 EXPORT_SYMBOL(register_quota_format
);
1881 EXPORT_SYMBOL(unregister_quota_format
);
1882 EXPORT_SYMBOL(dqstats
);
1883 EXPORT_SYMBOL(dq_data_lock
);
1884 EXPORT_SYMBOL(vfs_quota_on
);
1885 EXPORT_SYMBOL(vfs_quota_on_mount
);
1886 EXPORT_SYMBOL(vfs_quota_off
);
1887 EXPORT_SYMBOL(vfs_quota_sync
);
1888 EXPORT_SYMBOL(vfs_get_dqinfo
);
1889 EXPORT_SYMBOL(vfs_set_dqinfo
);
1890 EXPORT_SYMBOL(vfs_get_dqblk
);
1891 EXPORT_SYMBOL(vfs_set_dqblk
);
1892 EXPORT_SYMBOL(dquot_commit
);
1893 EXPORT_SYMBOL(dquot_commit_info
);
1894 EXPORT_SYMBOL(dquot_acquire
);
1895 EXPORT_SYMBOL(dquot_release
);
1896 EXPORT_SYMBOL(dquot_mark_dquot_dirty
);
1897 EXPORT_SYMBOL(dquot_initialize
);
1898 EXPORT_SYMBOL(dquot_drop
);
1899 EXPORT_SYMBOL(dquot_alloc_space
);
1900 EXPORT_SYMBOL(dquot_alloc_inode
);
1901 EXPORT_SYMBOL(dquot_free_space
);
1902 EXPORT_SYMBOL(dquot_free_inode
);
1903 EXPORT_SYMBOL(dquot_transfer
);