quota: Refactor dquot_transfer code so that OCFS2 can pass in its references
[linux-2.6.git] / fs / quota / dquot.c
1 /*
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.
11  * 
12  * Author:      Marco van Wieringen <mvw@planets.elm.net>
13  *
14  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
15  *
16  *              Revised list management to avoid races
17  *              -- Bill Hawes, <whawes@star.net>, 9/98
18  *
19  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20  *              As the consequence the locking was moved from dquot_decr_...(),
21  *              dquot_incr_...() to calling functions.
22  *              invalidate_dquots() now writes modified dquots.
23  *              Serialized quota_off() and quota_on() for mount point.
24  *              Fixed a few bugs in grow_dquots().
25  *              Fixed deadlock in write_dquot() - we no longer account quotas on
26  *              quota files
27  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
28  *              add_dquot_ref() restarts after blocking
29  *              Added check for bogus uid and fixed check for group in quotactl.
30  *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
31  *
32  *              Used struct list_head instead of own list struct
33  *              Invalidation of referenced dquots is no longer possible
34  *              Improved free_dquots list management
35  *              Quota and i_blocks are now updated in one place to avoid races
36  *              Warnings are now delayed so we won't block in critical section
37  *              Write updated not to require dquot lock
38  *              Jan Kara, <jack@suse.cz>, 9/2000
39  *
40  *              Added dynamic quota structure allocation
41  *              Jan Kara <jack@suse.cz> 12/2000
42  *
43  *              Rewritten quota interface. Implemented new quota format and
44  *              formats registering.
45  *              Jan Kara, <jack@suse.cz>, 2001,2002
46  *
47  *              New SMP locking.
48  *              Jan Kara, <jack@suse.cz>, 10/2002
49  *
50  *              Added journalled quota support, fix lock inversion problems
51  *              Jan Kara, <jack@suse.cz>, 2003,2004
52  *
53  * (C) Copyright 1994 - 1997 Marco van Wieringen 
54  */
55
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
58 #include <linux/fs.h>
59 #include <linux/mount.h>
60 #include <linux/mm.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/kmod.h>
75 #include <linux/namei.h>
76 #include <linux/buffer_head.h>
77 #include <linux/capability.h>
78 #include <linux/quotaops.h>
79 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
80
81 #include <asm/uaccess.h>
82
83 /*
84  * There are three quota SMP locks. dq_list_lock protects all lists with quotas
85  * and quota formats.
86  * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
87  * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
88  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
89  * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
90  * modifications of quota state (on quotaon and quotaoff) and readers who care
91  * about latest values take it as well.
92  *
93  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
94  *   dq_list_lock > dq_state_lock
95  *
96  * Note that some things (eg. sb pointer, type, id) doesn't change during
97  * the life of the dquot structure and so needn't to be protected by a lock
98  *
99  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
100  * operation is just reading pointers from inode (or not using them at all) the
101  * read lock is enough. If pointers are altered function must hold write lock.
102  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
103  * inode is a quota file). Functions adding pointers from inode to dquots have
104  * to check this flag under dqptr_sem and then (if S_NOQUOTA is not set) they
105  * have to do all pointer modifications before dropping dqptr_sem. This makes
106  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
107  * then drops all pointers to dquots from an inode.
108  *
109  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
110  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
111  * Currently dquot is locked only when it is being read to memory (or space for
112  * it is being allocated) on the first dqget() and when it is being released on
113  * the last dqput(). The allocation and release oparations are serialized by
114  * the dq_lock and by checking the use count in dquot_release().  Write
115  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
116  * spinlock to internal buffers before writing.
117  *
118  * Lock ordering (including related VFS locks) is the following:
119  *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
120  *   dqio_mutex
121  * The lock ordering of dqptr_sem imposed by quota code is only dqonoff_sem >
122  * dqptr_sem. But filesystem has to count with the fact that functions such as
123  * dquot_alloc_space() acquire dqptr_sem and they usually have to be called
124  * from inside a transaction to keep filesystem consistency after a crash. Also
125  * filesystems usually want to do some IO on dquot from ->mark_dirty which is
126  * called with dqptr_sem held.
127  * i_mutex on quota files is special (it's below dqio_mutex)
128  */
129
130 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
131 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
132 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
133 EXPORT_SYMBOL(dq_data_lock);
134
135 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
136 static char *quotatypes[] = INITQFNAMES;
137 #endif
138 static struct quota_format_type *quota_formats; /* List of registered formats */
139 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
140
141 /* SLAB cache for dquot structures */
142 static struct kmem_cache *dquot_cachep;
143
144 int register_quota_format(struct quota_format_type *fmt)
145 {
146         spin_lock(&dq_list_lock);
147         fmt->qf_next = quota_formats;
148         quota_formats = fmt;
149         spin_unlock(&dq_list_lock);
150         return 0;
151 }
152 EXPORT_SYMBOL(register_quota_format);
153
154 void unregister_quota_format(struct quota_format_type *fmt)
155 {
156         struct quota_format_type **actqf;
157
158         spin_lock(&dq_list_lock);
159         for (actqf = &quota_formats; *actqf && *actqf != fmt;
160              actqf = &(*actqf)->qf_next)
161                 ;
162         if (*actqf)
163                 *actqf = (*actqf)->qf_next;
164         spin_unlock(&dq_list_lock);
165 }
166 EXPORT_SYMBOL(unregister_quota_format);
167
168 static struct quota_format_type *find_quota_format(int id)
169 {
170         struct quota_format_type *actqf;
171
172         spin_lock(&dq_list_lock);
173         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
174              actqf = actqf->qf_next)
175                 ;
176         if (!actqf || !try_module_get(actqf->qf_owner)) {
177                 int qm;
178
179                 spin_unlock(&dq_list_lock);
180                 
181                 for (qm = 0; module_names[qm].qm_fmt_id &&
182                              module_names[qm].qm_fmt_id != id; qm++)
183                         ;
184                 if (!module_names[qm].qm_fmt_id ||
185                     request_module(module_names[qm].qm_mod_name))
186                         return NULL;
187
188                 spin_lock(&dq_list_lock);
189                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
190                      actqf = actqf->qf_next)
191                         ;
192                 if (actqf && !try_module_get(actqf->qf_owner))
193                         actqf = NULL;
194         }
195         spin_unlock(&dq_list_lock);
196         return actqf;
197 }
198
199 static void put_quota_format(struct quota_format_type *fmt)
200 {
201         module_put(fmt->qf_owner);
202 }
203
204 /*
205  * Dquot List Management:
206  * The quota code uses three lists for dquot management: the inuse_list,
207  * free_dquots, and dquot_hash[] array. A single dquot structure may be
208  * on all three lists, depending on its current state.
209  *
210  * All dquots are placed to the end of inuse_list when first created, and this
211  * list is used for invalidate operation, which must look at every dquot.
212  *
213  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
214  * and this list is searched whenever we need an available dquot.  Dquots are
215  * removed from the list as soon as they are used again, and
216  * dqstats.free_dquots gives the number of dquots on the list. When
217  * dquot is invalidated it's completely released from memory.
218  *
219  * Dquots with a specific identity (device, type and id) are placed on
220  * one of the dquot_hash[] hash chains. The provides an efficient search
221  * mechanism to locate a specific dquot.
222  */
223
224 static LIST_HEAD(inuse_list);
225 static LIST_HEAD(free_dquots);
226 static unsigned int dq_hash_bits, dq_hash_mask;
227 static struct hlist_head *dquot_hash;
228
229 struct dqstats dqstats;
230 EXPORT_SYMBOL(dqstats);
231 #ifdef CONFIG_SMP
232 struct dqstats *dqstats_pcpu;
233 EXPORT_SYMBOL(dqstats_pcpu);
234 #endif
235
236 static qsize_t inode_get_rsv_space(struct inode *inode);
237 static void __dquot_initialize(struct inode *inode, int type);
238
239 static inline unsigned int
240 hashfn(const struct super_block *sb, unsigned int id, int type)
241 {
242         unsigned long tmp;
243
244         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
245         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
246 }
247
248 /*
249  * Following list functions expect dq_list_lock to be held
250  */
251 static inline void insert_dquot_hash(struct dquot *dquot)
252 {
253         struct hlist_head *head;
254         head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
255         hlist_add_head(&dquot->dq_hash, head);
256 }
257
258 static inline void remove_dquot_hash(struct dquot *dquot)
259 {
260         hlist_del_init(&dquot->dq_hash);
261 }
262
263 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
264                                 unsigned int id, int type)
265 {
266         struct hlist_node *node;
267         struct dquot *dquot;
268
269         hlist_for_each (node, dquot_hash+hashent) {
270                 dquot = hlist_entry(node, struct dquot, dq_hash);
271                 if (dquot->dq_sb == sb && dquot->dq_id == id &&
272                     dquot->dq_type == type)
273                         return dquot;
274         }
275         return NULL;
276 }
277
278 /* Add a dquot to the tail of the free list */
279 static inline void put_dquot_last(struct dquot *dquot)
280 {
281         list_add_tail(&dquot->dq_free, &free_dquots);
282         dqstats_inc(DQST_FREE_DQUOTS);
283 }
284
285 static inline void remove_free_dquot(struct dquot *dquot)
286 {
287         if (list_empty(&dquot->dq_free))
288                 return;
289         list_del_init(&dquot->dq_free);
290         dqstats_dec(DQST_FREE_DQUOTS);
291 }
292
293 static inline void put_inuse(struct dquot *dquot)
294 {
295         /* We add to the back of inuse list so we don't have to restart
296          * when traversing this list and we block */
297         list_add_tail(&dquot->dq_inuse, &inuse_list);
298         dqstats_inc(DQST_ALLOC_DQUOTS);
299 }
300
301 static inline void remove_inuse(struct dquot *dquot)
302 {
303         dqstats_dec(DQST_ALLOC_DQUOTS);
304         list_del(&dquot->dq_inuse);
305 }
306 /*
307  * End of list functions needing dq_list_lock
308  */
309
310 static void wait_on_dquot(struct dquot *dquot)
311 {
312         mutex_lock(&dquot->dq_lock);
313         mutex_unlock(&dquot->dq_lock);
314 }
315
316 static inline int dquot_dirty(struct dquot *dquot)
317 {
318         return test_bit(DQ_MOD_B, &dquot->dq_flags);
319 }
320
321 static inline int mark_dquot_dirty(struct dquot *dquot)
322 {
323         return dquot->dq_sb->dq_op->mark_dirty(dquot);
324 }
325
326 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
327 int dquot_mark_dquot_dirty(struct dquot *dquot)
328 {
329         int ret = 1;
330
331         /* If quota is dirty already, we don't have to acquire dq_list_lock */
332         if (test_bit(DQ_MOD_B, &dquot->dq_flags))
333                 return 1;
334
335         spin_lock(&dq_list_lock);
336         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
337                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
338                                 info[dquot->dq_type].dqi_dirty_list);
339                 ret = 0;
340         }
341         spin_unlock(&dq_list_lock);
342         return ret;
343 }
344 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
345
346 /* Dirtify all the dquots - this can block when journalling */
347 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
348 {
349         int ret, err, cnt;
350
351         ret = err = 0;
352         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
353                 if (dquot[cnt])
354                         /* Even in case of error we have to continue */
355                         ret = mark_dquot_dirty(dquot[cnt]);
356                 if (!err)
357                         err = ret;
358         }
359         return err;
360 }
361
362 static inline void dqput_all(struct dquot **dquot)
363 {
364         unsigned int cnt;
365
366         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
367                 dqput(dquot[cnt]);
368 }
369
370 /* This function needs dq_list_lock */
371 static inline int clear_dquot_dirty(struct dquot *dquot)
372 {
373         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
374                 return 0;
375         list_del_init(&dquot->dq_dirty);
376         return 1;
377 }
378
379 void mark_info_dirty(struct super_block *sb, int type)
380 {
381         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
382 }
383 EXPORT_SYMBOL(mark_info_dirty);
384
385 /*
386  *      Read dquot from disk and alloc space for it
387  */
388
389 int dquot_acquire(struct dquot *dquot)
390 {
391         int ret = 0, ret2 = 0;
392         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
393
394         mutex_lock(&dquot->dq_lock);
395         mutex_lock(&dqopt->dqio_mutex);
396         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
397                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
398         if (ret < 0)
399                 goto out_iolock;
400         set_bit(DQ_READ_B, &dquot->dq_flags);
401         /* Instantiate dquot if needed */
402         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
403                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
404                 /* Write the info if needed */
405                 if (info_dirty(&dqopt->info[dquot->dq_type])) {
406                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
407                                                 dquot->dq_sb, dquot->dq_type);
408                 }
409                 if (ret < 0)
410                         goto out_iolock;
411                 if (ret2 < 0) {
412                         ret = ret2;
413                         goto out_iolock;
414                 }
415         }
416         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
417 out_iolock:
418         mutex_unlock(&dqopt->dqio_mutex);
419         mutex_unlock(&dquot->dq_lock);
420         return ret;
421 }
422 EXPORT_SYMBOL(dquot_acquire);
423
424 /*
425  *      Write dquot to disk
426  */
427 int dquot_commit(struct dquot *dquot)
428 {
429         int ret = 0, ret2 = 0;
430         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
431
432         mutex_lock(&dqopt->dqio_mutex);
433         spin_lock(&dq_list_lock);
434         if (!clear_dquot_dirty(dquot)) {
435                 spin_unlock(&dq_list_lock);
436                 goto out_sem;
437         }
438         spin_unlock(&dq_list_lock);
439         /* Inactive dquot can be only if there was error during read/init
440          * => we have better not writing it */
441         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
442                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
443                 if (info_dirty(&dqopt->info[dquot->dq_type])) {
444                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
445                                                 dquot->dq_sb, dquot->dq_type);
446                 }
447                 if (ret >= 0)
448                         ret = ret2;
449         }
450 out_sem:
451         mutex_unlock(&dqopt->dqio_mutex);
452         return ret;
453 }
454 EXPORT_SYMBOL(dquot_commit);
455
456 /*
457  *      Release dquot
458  */
459 int dquot_release(struct dquot *dquot)
460 {
461         int ret = 0, ret2 = 0;
462         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
463
464         mutex_lock(&dquot->dq_lock);
465         /* Check whether we are not racing with some other dqget() */
466         if (atomic_read(&dquot->dq_count) > 1)
467                 goto out_dqlock;
468         mutex_lock(&dqopt->dqio_mutex);
469         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
470                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
471                 /* Write the info */
472                 if (info_dirty(&dqopt->info[dquot->dq_type])) {
473                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
474                                                 dquot->dq_sb, dquot->dq_type);
475                 }
476                 if (ret >= 0)
477                         ret = ret2;
478         }
479         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
480         mutex_unlock(&dqopt->dqio_mutex);
481 out_dqlock:
482         mutex_unlock(&dquot->dq_lock);
483         return ret;
484 }
485 EXPORT_SYMBOL(dquot_release);
486
487 void dquot_destroy(struct dquot *dquot)
488 {
489         kmem_cache_free(dquot_cachep, dquot);
490 }
491 EXPORT_SYMBOL(dquot_destroy);
492
493 static inline void do_destroy_dquot(struct dquot *dquot)
494 {
495         dquot->dq_sb->dq_op->destroy_dquot(dquot);
496 }
497
498 /* Invalidate all dquots on the list. Note that this function is called after
499  * quota is disabled and pointers from inodes removed so there cannot be new
500  * quota users. There can still be some users of quotas due to inodes being
501  * just deleted or pruned by prune_icache() (those are not attached to any
502  * list) or parallel quotactl call. We have to wait for such users.
503  */
504 static void invalidate_dquots(struct super_block *sb, int type)
505 {
506         struct dquot *dquot, *tmp;
507
508 restart:
509         spin_lock(&dq_list_lock);
510         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
511                 if (dquot->dq_sb != sb)
512                         continue;
513                 if (dquot->dq_type != type)
514                         continue;
515                 /* Wait for dquot users */
516                 if (atomic_read(&dquot->dq_count)) {
517                         DEFINE_WAIT(wait);
518
519                         atomic_inc(&dquot->dq_count);
520                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
521                                         TASK_UNINTERRUPTIBLE);
522                         spin_unlock(&dq_list_lock);
523                         /* Once dqput() wakes us up, we know it's time to free
524                          * the dquot.
525                          * IMPORTANT: we rely on the fact that there is always
526                          * at most one process waiting for dquot to free.
527                          * Otherwise dq_count would be > 1 and we would never
528                          * wake up.
529                          */
530                         if (atomic_read(&dquot->dq_count) > 1)
531                                 schedule();
532                         finish_wait(&dquot->dq_wait_unused, &wait);
533                         dqput(dquot);
534                         /* At this moment dquot() need not exist (it could be
535                          * reclaimed by prune_dqcache(). Hence we must
536                          * restart. */
537                         goto restart;
538                 }
539                 /*
540                  * Quota now has no users and it has been written on last
541                  * dqput()
542                  */
543                 remove_dquot_hash(dquot);
544                 remove_free_dquot(dquot);
545                 remove_inuse(dquot);
546                 do_destroy_dquot(dquot);
547         }
548         spin_unlock(&dq_list_lock);
549 }
550
551 /* Call callback for every active dquot on given filesystem */
552 int dquot_scan_active(struct super_block *sb,
553                       int (*fn)(struct dquot *dquot, unsigned long priv),
554                       unsigned long priv)
555 {
556         struct dquot *dquot, *old_dquot = NULL;
557         int ret = 0;
558
559         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
560         spin_lock(&dq_list_lock);
561         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
562                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
563                         continue;
564                 if (dquot->dq_sb != sb)
565                         continue;
566                 /* Now we have active dquot so we can just increase use count */
567                 atomic_inc(&dquot->dq_count);
568                 spin_unlock(&dq_list_lock);
569                 dqstats_inc(DQST_LOOKUPS);
570                 dqput(old_dquot);
571                 old_dquot = dquot;
572                 ret = fn(dquot, priv);
573                 if (ret < 0)
574                         goto out;
575                 spin_lock(&dq_list_lock);
576                 /* We are safe to continue now because our dquot could not
577                  * be moved out of the inuse list while we hold the reference */
578         }
579         spin_unlock(&dq_list_lock);
580 out:
581         dqput(old_dquot);
582         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
583         return ret;
584 }
585 EXPORT_SYMBOL(dquot_scan_active);
586
587 int vfs_quota_sync(struct super_block *sb, int type, int wait)
588 {
589         struct list_head *dirty;
590         struct dquot *dquot;
591         struct quota_info *dqopt = sb_dqopt(sb);
592         int cnt;
593
594         mutex_lock(&dqopt->dqonoff_mutex);
595         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
596                 if (type != -1 && cnt != type)
597                         continue;
598                 if (!sb_has_quota_active(sb, cnt))
599                         continue;
600                 spin_lock(&dq_list_lock);
601                 dirty = &dqopt->info[cnt].dqi_dirty_list;
602                 while (!list_empty(dirty)) {
603                         dquot = list_first_entry(dirty, struct dquot,
604                                                  dq_dirty);
605                         /* Dirty and inactive can be only bad dquot... */
606                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
607                                 clear_dquot_dirty(dquot);
608                                 continue;
609                         }
610                         /* Now we have active dquot from which someone is
611                          * holding reference so we can safely just increase
612                          * use count */
613                         atomic_inc(&dquot->dq_count);
614                         spin_unlock(&dq_list_lock);
615                         dqstats_inc(DQST_LOOKUPS);
616                         sb->dq_op->write_dquot(dquot);
617                         dqput(dquot);
618                         spin_lock(&dq_list_lock);
619                 }
620                 spin_unlock(&dq_list_lock);
621         }
622
623         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
624                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
625                     && info_dirty(&dqopt->info[cnt]))
626                         sb->dq_op->write_info(sb, cnt);
627         dqstats_inc(DQST_SYNCS);
628         mutex_unlock(&dqopt->dqonoff_mutex);
629
630         if (!wait || (sb_dqopt(sb)->flags & DQUOT_QUOTA_SYS_FILE))
631                 return 0;
632
633         /* This is not very clever (and fast) but currently I don't know about
634          * any other simple way of getting quota data to disk and we must get
635          * them there for userspace to be visible... */
636         if (sb->s_op->sync_fs)
637                 sb->s_op->sync_fs(sb, 1);
638         sync_blockdev(sb->s_bdev);
639
640         /*
641          * Now when everything is written we can discard the pagecache so
642          * that userspace sees the changes.
643          */
644         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
645         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
646                 if (type != -1 && cnt != type)
647                         continue;
648                 if (!sb_has_quota_active(sb, cnt))
649                         continue;
650                 mutex_lock_nested(&sb_dqopt(sb)->files[cnt]->i_mutex,
651                                   I_MUTEX_QUOTA);
652                 truncate_inode_pages(&sb_dqopt(sb)->files[cnt]->i_data, 0);
653                 mutex_unlock(&sb_dqopt(sb)->files[cnt]->i_mutex);
654         }
655         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
656
657         return 0;
658 }
659 EXPORT_SYMBOL(vfs_quota_sync);
660
661 /* Free unused dquots from cache */
662 static void prune_dqcache(int count)
663 {
664         struct list_head *head;
665         struct dquot *dquot;
666
667         head = free_dquots.prev;
668         while (head != &free_dquots && count) {
669                 dquot = list_entry(head, struct dquot, dq_free);
670                 remove_dquot_hash(dquot);
671                 remove_free_dquot(dquot);
672                 remove_inuse(dquot);
673                 do_destroy_dquot(dquot);
674                 count--;
675                 head = free_dquots.prev;
676         }
677 }
678
679 static int dqstats_read(unsigned int type)
680 {
681         int count = 0;
682 #ifdef CONFIG_SMP
683         int cpu;
684         for_each_possible_cpu(cpu)
685                 count += per_cpu_ptr(dqstats_pcpu, cpu)->stat[type];
686         /* Statistics reading is racy, but absolute accuracy isn't required */
687         if (count < 0)
688                 count = 0;
689 #else
690         count = dqstats.stat[type];
691 #endif
692         return count;
693 }
694
695 /*
696  * This is called from kswapd when we think we need some
697  * more memory
698  */
699
700 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
701 {
702         if (nr) {
703                 spin_lock(&dq_list_lock);
704                 prune_dqcache(nr);
705                 spin_unlock(&dq_list_lock);
706         }
707         return (dqstats_read(DQST_FREE_DQUOTS)/100) * sysctl_vfs_cache_pressure;
708 }
709
710 static struct shrinker dqcache_shrinker = {
711         .shrink = shrink_dqcache_memory,
712         .seeks = DEFAULT_SEEKS,
713 };
714
715 /*
716  * Put reference to dquot
717  * NOTE: If you change this function please check whether dqput_blocks() works right...
718  */
719 void dqput(struct dquot *dquot)
720 {
721         int ret;
722
723         if (!dquot)
724                 return;
725 #ifdef CONFIG_QUOTA_DEBUG
726         if (!atomic_read(&dquot->dq_count)) {
727                 printk("VFS: dqput: trying to free free dquot\n");
728                 printk("VFS: device %s, dquot of %s %d\n",
729                         dquot->dq_sb->s_id,
730                         quotatypes[dquot->dq_type],
731                         dquot->dq_id);
732                 BUG();
733         }
734 #endif
735         dqstats_inc(DQST_DROPS);
736 we_slept:
737         spin_lock(&dq_list_lock);
738         if (atomic_read(&dquot->dq_count) > 1) {
739                 /* We have more than one user... nothing to do */
740                 atomic_dec(&dquot->dq_count);
741                 /* Releasing dquot during quotaoff phase? */
742                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_type) &&
743                     atomic_read(&dquot->dq_count) == 1)
744                         wake_up(&dquot->dq_wait_unused);
745                 spin_unlock(&dq_list_lock);
746                 return;
747         }
748         /* Need to release dquot? */
749         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
750                 spin_unlock(&dq_list_lock);
751                 /* Commit dquot before releasing */
752                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
753                 if (ret < 0) {
754                         printk(KERN_ERR "VFS: cannot write quota structure on "
755                                 "device %s (error %d). Quota may get out of "
756                                 "sync!\n", dquot->dq_sb->s_id, ret);
757                         /*
758                          * We clear dirty bit anyway, so that we avoid
759                          * infinite loop here
760                          */
761                         spin_lock(&dq_list_lock);
762                         clear_dquot_dirty(dquot);
763                         spin_unlock(&dq_list_lock);
764                 }
765                 goto we_slept;
766         }
767         /* Clear flag in case dquot was inactive (something bad happened) */
768         clear_dquot_dirty(dquot);
769         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
770                 spin_unlock(&dq_list_lock);
771                 dquot->dq_sb->dq_op->release_dquot(dquot);
772                 goto we_slept;
773         }
774         atomic_dec(&dquot->dq_count);
775 #ifdef CONFIG_QUOTA_DEBUG
776         /* sanity check */
777         BUG_ON(!list_empty(&dquot->dq_free));
778 #endif
779         put_dquot_last(dquot);
780         spin_unlock(&dq_list_lock);
781 }
782 EXPORT_SYMBOL(dqput);
783
784 struct dquot *dquot_alloc(struct super_block *sb, int type)
785 {
786         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
787 }
788 EXPORT_SYMBOL(dquot_alloc);
789
790 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
791 {
792         struct dquot *dquot;
793
794         dquot = sb->dq_op->alloc_dquot(sb, type);
795         if(!dquot)
796                 return NULL;
797
798         mutex_init(&dquot->dq_lock);
799         INIT_LIST_HEAD(&dquot->dq_free);
800         INIT_LIST_HEAD(&dquot->dq_inuse);
801         INIT_HLIST_NODE(&dquot->dq_hash);
802         INIT_LIST_HEAD(&dquot->dq_dirty);
803         init_waitqueue_head(&dquot->dq_wait_unused);
804         dquot->dq_sb = sb;
805         dquot->dq_type = type;
806         atomic_set(&dquot->dq_count, 1);
807
808         return dquot;
809 }
810
811 /*
812  * Get reference to dquot
813  *
814  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
815  * destroying our dquot by:
816  *   a) checking for quota flags under dq_list_lock and
817  *   b) getting a reference to dquot before we release dq_list_lock
818  */
819 struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
820 {
821         unsigned int hashent = hashfn(sb, id, type);
822         struct dquot *dquot = NULL, *empty = NULL;
823
824         if (!sb_has_quota_active(sb, type))
825                 return NULL;
826 we_slept:
827         spin_lock(&dq_list_lock);
828         spin_lock(&dq_state_lock);
829         if (!sb_has_quota_active(sb, type)) {
830                 spin_unlock(&dq_state_lock);
831                 spin_unlock(&dq_list_lock);
832                 goto out;
833         }
834         spin_unlock(&dq_state_lock);
835
836         dquot = find_dquot(hashent, sb, id, type);
837         if (!dquot) {
838                 if (!empty) {
839                         spin_unlock(&dq_list_lock);
840                         empty = get_empty_dquot(sb, type);
841                         if (!empty)
842                                 schedule();     /* Try to wait for a moment... */
843                         goto we_slept;
844                 }
845                 dquot = empty;
846                 empty = NULL;
847                 dquot->dq_id = id;
848                 /* all dquots go on the inuse_list */
849                 put_inuse(dquot);
850                 /* hash it first so it can be found */
851                 insert_dquot_hash(dquot);
852                 spin_unlock(&dq_list_lock);
853                 dqstats_inc(DQST_LOOKUPS);
854         } else {
855                 if (!atomic_read(&dquot->dq_count))
856                         remove_free_dquot(dquot);
857                 atomic_inc(&dquot->dq_count);
858                 spin_unlock(&dq_list_lock);
859                 dqstats_inc(DQST_CACHE_HITS);
860                 dqstats_inc(DQST_LOOKUPS);
861         }
862         /* Wait for dq_lock - after this we know that either dquot_release() is
863          * already finished or it will be canceled due to dq_count > 1 test */
864         wait_on_dquot(dquot);
865         /* Read the dquot / allocate space in quota file */
866         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
867             sb->dq_op->acquire_dquot(dquot) < 0) {
868                 dqput(dquot);
869                 dquot = NULL;
870                 goto out;
871         }
872 #ifdef CONFIG_QUOTA_DEBUG
873         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
874 #endif
875 out:
876         if (empty)
877                 do_destroy_dquot(empty);
878
879         return dquot;
880 }
881 EXPORT_SYMBOL(dqget);
882
883 static int dqinit_needed(struct inode *inode, int type)
884 {
885         int cnt;
886
887         if (IS_NOQUOTA(inode))
888                 return 0;
889         if (type != -1)
890                 return !inode->i_dquot[type];
891         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
892                 if (!inode->i_dquot[cnt])
893                         return 1;
894         return 0;
895 }
896
897 /* This routine is guarded by dqonoff_mutex mutex */
898 static void add_dquot_ref(struct super_block *sb, int type)
899 {
900         struct inode *inode, *old_inode = NULL;
901 #ifdef CONFIG_QUOTA_DEBUG
902         int reserved = 0;
903 #endif
904
905         spin_lock(&inode_lock);
906         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
907                 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
908                         continue;
909 #ifdef CONFIG_QUOTA_DEBUG
910                 if (unlikely(inode_get_rsv_space(inode) > 0))
911                         reserved = 1;
912 #endif
913                 if (!atomic_read(&inode->i_writecount))
914                         continue;
915                 if (!dqinit_needed(inode, type))
916                         continue;
917
918                 __iget(inode);
919                 spin_unlock(&inode_lock);
920
921                 iput(old_inode);
922                 __dquot_initialize(inode, type);
923                 /* We hold a reference to 'inode' so it couldn't have been
924                  * removed from s_inodes list while we dropped the inode_lock.
925                  * We cannot iput the inode now as we can be holding the last
926                  * reference and we cannot iput it under inode_lock. So we
927                  * keep the reference and iput it later. */
928                 old_inode = inode;
929                 spin_lock(&inode_lock);
930         }
931         spin_unlock(&inode_lock);
932         iput(old_inode);
933
934 #ifdef CONFIG_QUOTA_DEBUG
935         if (reserved) {
936                 printk(KERN_WARNING "VFS (%s): Writes happened before quota"
937                         " was turned on thus quota information is probably "
938                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
939         }
940 #endif
941 }
942
943 /*
944  * Return 0 if dqput() won't block.
945  * (note that 1 doesn't necessarily mean blocking)
946  */
947 static inline int dqput_blocks(struct dquot *dquot)
948 {
949         if (atomic_read(&dquot->dq_count) <= 1)
950                 return 1;
951         return 0;
952 }
953
954 /*
955  * Remove references to dquots from inode and add dquot to list for freeing
956  * if we have the last referece to dquot
957  * We can't race with anybody because we hold dqptr_sem for writing...
958  */
959 static int remove_inode_dquot_ref(struct inode *inode, int type,
960                                   struct list_head *tofree_head)
961 {
962         struct dquot *dquot = inode->i_dquot[type];
963
964         inode->i_dquot[type] = NULL;
965         if (dquot) {
966                 if (dqput_blocks(dquot)) {
967 #ifdef CONFIG_QUOTA_DEBUG
968                         if (atomic_read(&dquot->dq_count) != 1)
969                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
970 #endif
971                         spin_lock(&dq_list_lock);
972                         /* As dquot must have currently users it can't be on
973                          * the free list... */
974                         list_add(&dquot->dq_free, tofree_head);
975                         spin_unlock(&dq_list_lock);
976                         return 1;
977                 }
978                 else
979                         dqput(dquot);   /* We have guaranteed we won't block */
980         }
981         return 0;
982 }
983
984 /*
985  * Free list of dquots
986  * Dquots are removed from inodes and no new references can be got so we are
987  * the only ones holding reference
988  */
989 static void put_dquot_list(struct list_head *tofree_head)
990 {
991         struct list_head *act_head;
992         struct dquot *dquot;
993
994         act_head = tofree_head->next;
995         while (act_head != tofree_head) {
996                 dquot = list_entry(act_head, struct dquot, dq_free);
997                 act_head = act_head->next;
998                 /* Remove dquot from the list so we won't have problems... */
999                 list_del_init(&dquot->dq_free);
1000                 dqput(dquot);
1001         }
1002 }
1003
1004 static void remove_dquot_ref(struct super_block *sb, int type,
1005                 struct list_head *tofree_head)
1006 {
1007         struct inode *inode;
1008
1009         spin_lock(&inode_lock);
1010         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1011                 /*
1012                  *  We have to scan also I_NEW inodes because they can already
1013                  *  have quota pointer initialized. Luckily, we need to touch
1014                  *  only quota pointers and these have separate locking
1015                  *  (dqptr_sem).
1016                  */
1017                 if (!IS_NOQUOTA(inode))
1018                         remove_inode_dquot_ref(inode, type, tofree_head);
1019         }
1020         spin_unlock(&inode_lock);
1021 }
1022
1023 /* Gather all references from inodes and drop them */
1024 static void drop_dquot_ref(struct super_block *sb, int type)
1025 {
1026         LIST_HEAD(tofree_head);
1027
1028         if (sb->dq_op) {
1029                 down_write(&sb_dqopt(sb)->dqptr_sem);
1030                 remove_dquot_ref(sb, type, &tofree_head);
1031                 up_write(&sb_dqopt(sb)->dqptr_sem);
1032                 put_dquot_list(&tofree_head);
1033         }
1034 }
1035
1036 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1037 {
1038         dquot->dq_dqb.dqb_curinodes += number;
1039 }
1040
1041 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1042 {
1043         dquot->dq_dqb.dqb_curspace += number;
1044 }
1045
1046 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1047 {
1048         dquot->dq_dqb.dqb_rsvspace += number;
1049 }
1050
1051 /*
1052  * Claim reserved quota space
1053  */
1054 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1055 {
1056         if (dquot->dq_dqb.dqb_rsvspace < number) {
1057                 WARN_ON_ONCE(1);
1058                 number = dquot->dq_dqb.dqb_rsvspace;
1059         }
1060         dquot->dq_dqb.dqb_curspace += number;
1061         dquot->dq_dqb.dqb_rsvspace -= number;
1062 }
1063
1064 static inline
1065 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1066 {
1067         if (dquot->dq_dqb.dqb_rsvspace >= number)
1068                 dquot->dq_dqb.dqb_rsvspace -= number;
1069         else {
1070                 WARN_ON_ONCE(1);
1071                 dquot->dq_dqb.dqb_rsvspace = 0;
1072         }
1073 }
1074
1075 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1076 {
1077         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1078             dquot->dq_dqb.dqb_curinodes >= number)
1079                 dquot->dq_dqb.dqb_curinodes -= number;
1080         else
1081                 dquot->dq_dqb.dqb_curinodes = 0;
1082         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1083                 dquot->dq_dqb.dqb_itime = (time_t) 0;
1084         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1085 }
1086
1087 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1088 {
1089         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1090             dquot->dq_dqb.dqb_curspace >= number)
1091                 dquot->dq_dqb.dqb_curspace -= number;
1092         else
1093                 dquot->dq_dqb.dqb_curspace = 0;
1094         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1095                 dquot->dq_dqb.dqb_btime = (time_t) 0;
1096         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1097 }
1098
1099 static int warning_issued(struct dquot *dquot, const int warntype)
1100 {
1101         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1102                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1103                 ((warntype == QUOTA_NL_IHARDWARN ||
1104                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1105
1106         if (!flag)
1107                 return 0;
1108         return test_and_set_bit(flag, &dquot->dq_flags);
1109 }
1110
1111 #ifdef CONFIG_PRINT_QUOTA_WARNING
1112 static int flag_print_warnings = 1;
1113
1114 static int need_print_warning(struct dquot *dquot)
1115 {
1116         if (!flag_print_warnings)
1117                 return 0;
1118
1119         switch (dquot->dq_type) {
1120                 case USRQUOTA:
1121                         return current_fsuid() == dquot->dq_id;
1122                 case GRPQUOTA:
1123                         return in_group_p(dquot->dq_id);
1124         }
1125         return 0;
1126 }
1127
1128 /* Print warning to user which exceeded quota */
1129 static void print_warning(struct dquot *dquot, const int warntype)
1130 {
1131         char *msg = NULL;
1132         struct tty_struct *tty;
1133
1134         if (warntype == QUOTA_NL_IHARDBELOW ||
1135             warntype == QUOTA_NL_ISOFTBELOW ||
1136             warntype == QUOTA_NL_BHARDBELOW ||
1137             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
1138                 return;
1139
1140         tty = get_current_tty();
1141         if (!tty)
1142                 return;
1143         tty_write_message(tty, dquot->dq_sb->s_id);
1144         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1145                 tty_write_message(tty, ": warning, ");
1146         else
1147                 tty_write_message(tty, ": write failed, ");
1148         tty_write_message(tty, quotatypes[dquot->dq_type]);
1149         switch (warntype) {
1150                 case QUOTA_NL_IHARDWARN:
1151                         msg = " file limit reached.\r\n";
1152                         break;
1153                 case QUOTA_NL_ISOFTLONGWARN:
1154                         msg = " file quota exceeded too long.\r\n";
1155                         break;
1156                 case QUOTA_NL_ISOFTWARN:
1157                         msg = " file quota exceeded.\r\n";
1158                         break;
1159                 case QUOTA_NL_BHARDWARN:
1160                         msg = " block limit reached.\r\n";
1161                         break;
1162                 case QUOTA_NL_BSOFTLONGWARN:
1163                         msg = " block quota exceeded too long.\r\n";
1164                         break;
1165                 case QUOTA_NL_BSOFTWARN:
1166                         msg = " block quota exceeded.\r\n";
1167                         break;
1168         }
1169         tty_write_message(tty, msg);
1170         tty_kref_put(tty);
1171 }
1172 #endif
1173
1174 /*
1175  * Write warnings to the console and send warning messages over netlink.
1176  *
1177  * Note that this function can sleep.
1178  */
1179 static void flush_warnings(struct dquot *const *dquots, char *warntype)
1180 {
1181         struct dquot *dq;
1182         int i;
1183
1184         for (i = 0; i < MAXQUOTAS; i++) {
1185                 dq = dquots[i];
1186                 if (dq && warntype[i] != QUOTA_NL_NOWARN &&
1187                     !warning_issued(dq, warntype[i])) {
1188 #ifdef CONFIG_PRINT_QUOTA_WARNING
1189                         print_warning(dq, warntype[i]);
1190 #endif
1191                         quota_send_warning(dq->dq_type, dq->dq_id,
1192                                            dq->dq_sb->s_dev, warntype[i]);
1193                 }
1194         }
1195 }
1196
1197 static int ignore_hardlimit(struct dquot *dquot)
1198 {
1199         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1200
1201         return capable(CAP_SYS_RESOURCE) &&
1202                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1203                 !(info->dqi_flags & V1_DQF_RSQUASH));
1204 }
1205
1206 /* needs dq_data_lock */
1207 static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1208 {
1209         qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1210
1211         *warntype = QUOTA_NL_NOWARN;
1212         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1213             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1214                 return 0;
1215
1216         if (dquot->dq_dqb.dqb_ihardlimit &&
1217             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1218             !ignore_hardlimit(dquot)) {
1219                 *warntype = QUOTA_NL_IHARDWARN;
1220                 return -EDQUOT;
1221         }
1222
1223         if (dquot->dq_dqb.dqb_isoftlimit &&
1224             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1225             dquot->dq_dqb.dqb_itime &&
1226             get_seconds() >= dquot->dq_dqb.dqb_itime &&
1227             !ignore_hardlimit(dquot)) {
1228                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1229                 return -EDQUOT;
1230         }
1231
1232         if (dquot->dq_dqb.dqb_isoftlimit &&
1233             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1234             dquot->dq_dqb.dqb_itime == 0) {
1235                 *warntype = QUOTA_NL_ISOFTWARN;
1236                 dquot->dq_dqb.dqb_itime = get_seconds() +
1237                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1238         }
1239
1240         return 0;
1241 }
1242
1243 /* needs dq_data_lock */
1244 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1245 {
1246         qsize_t tspace;
1247         struct super_block *sb = dquot->dq_sb;
1248
1249         *warntype = QUOTA_NL_NOWARN;
1250         if (!sb_has_quota_limits_enabled(sb, dquot->dq_type) ||
1251             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1252                 return 0;
1253
1254         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1255                 + space;
1256
1257         if (dquot->dq_dqb.dqb_bhardlimit &&
1258             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1259             !ignore_hardlimit(dquot)) {
1260                 if (!prealloc)
1261                         *warntype = QUOTA_NL_BHARDWARN;
1262                 return -EDQUOT;
1263         }
1264
1265         if (dquot->dq_dqb.dqb_bsoftlimit &&
1266             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1267             dquot->dq_dqb.dqb_btime &&
1268             get_seconds() >= dquot->dq_dqb.dqb_btime &&
1269             !ignore_hardlimit(dquot)) {
1270                 if (!prealloc)
1271                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1272                 return -EDQUOT;
1273         }
1274
1275         if (dquot->dq_dqb.dqb_bsoftlimit &&
1276             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1277             dquot->dq_dqb.dqb_btime == 0) {
1278                 if (!prealloc) {
1279                         *warntype = QUOTA_NL_BSOFTWARN;
1280                         dquot->dq_dqb.dqb_btime = get_seconds() +
1281                             sb_dqopt(sb)->info[dquot->dq_type].dqi_bgrace;
1282                 }
1283                 else
1284                         /*
1285                          * We don't allow preallocation to exceed softlimit so exceeding will
1286                          * be always printed
1287                          */
1288                         return -EDQUOT;
1289         }
1290
1291         return 0;
1292 }
1293
1294 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1295 {
1296         qsize_t newinodes;
1297
1298         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1299             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1300             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type))
1301                 return QUOTA_NL_NOWARN;
1302
1303         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1304         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1305                 return QUOTA_NL_ISOFTBELOW;
1306         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1307             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1308                 return QUOTA_NL_IHARDBELOW;
1309         return QUOTA_NL_NOWARN;
1310 }
1311
1312 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1313 {
1314         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1315             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1316                 return QUOTA_NL_NOWARN;
1317
1318         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1319                 return QUOTA_NL_BSOFTBELOW;
1320         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1321             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1322                 return QUOTA_NL_BHARDBELOW;
1323         return QUOTA_NL_NOWARN;
1324 }
1325
1326 /*
1327  * Initialize quota pointers in inode
1328  *
1329  * We do things in a bit complicated way but by that we avoid calling
1330  * dqget() and thus filesystem callbacks under dqptr_sem.
1331  *
1332  * It is better to call this function outside of any transaction as it
1333  * might need a lot of space in journal for dquot structure allocation.
1334  */
1335 static void __dquot_initialize(struct inode *inode, int type)
1336 {
1337         unsigned int id = 0;
1338         int cnt;
1339         struct dquot *got[MAXQUOTAS];
1340         struct super_block *sb = inode->i_sb;
1341         qsize_t rsv;
1342
1343         /* First test before acquiring mutex - solves deadlocks when we
1344          * re-enter the quota code and are already holding the mutex */
1345         if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1346                 return;
1347
1348         /* First get references to structures we might need. */
1349         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1350                 got[cnt] = NULL;
1351                 if (type != -1 && cnt != type)
1352                         continue;
1353                 switch (cnt) {
1354                 case USRQUOTA:
1355                         id = inode->i_uid;
1356                         break;
1357                 case GRPQUOTA:
1358                         id = inode->i_gid;
1359                         break;
1360                 }
1361                 got[cnt] = dqget(sb, id, cnt);
1362         }
1363
1364         down_write(&sb_dqopt(sb)->dqptr_sem);
1365         if (IS_NOQUOTA(inode))
1366                 goto out_err;
1367         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1368                 if (type != -1 && cnt != type)
1369                         continue;
1370                 /* Avoid races with quotaoff() */
1371                 if (!sb_has_quota_active(sb, cnt))
1372                         continue;
1373                 if (!inode->i_dquot[cnt]) {
1374                         inode->i_dquot[cnt] = got[cnt];
1375                         got[cnt] = NULL;
1376                         /*
1377                          * Make quota reservation system happy if someone
1378                          * did a write before quota was turned on
1379                          */
1380                         rsv = inode_get_rsv_space(inode);
1381                         if (unlikely(rsv))
1382                                 dquot_resv_space(inode->i_dquot[cnt], rsv);
1383                 }
1384         }
1385 out_err:
1386         up_write(&sb_dqopt(sb)->dqptr_sem);
1387         /* Drop unused references */
1388         dqput_all(got);
1389 }
1390
1391 void dquot_initialize(struct inode *inode)
1392 {
1393         __dquot_initialize(inode, -1);
1394 }
1395 EXPORT_SYMBOL(dquot_initialize);
1396
1397 /*
1398  *      Release all quotas referenced by inode
1399  */
1400 static void __dquot_drop(struct inode *inode)
1401 {
1402         int cnt;
1403         struct dquot *put[MAXQUOTAS];
1404
1405         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1406         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1407                 put[cnt] = inode->i_dquot[cnt];
1408                 inode->i_dquot[cnt] = NULL;
1409         }
1410         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1411         dqput_all(put);
1412 }
1413
1414 void dquot_drop(struct inode *inode)
1415 {
1416         int cnt;
1417
1418         if (IS_NOQUOTA(inode))
1419                 return;
1420
1421         /*
1422          * Test before calling to rule out calls from proc and such
1423          * where we are not allowed to block. Note that this is
1424          * actually reliable test even without the lock - the caller
1425          * must assure that nobody can come after the DQUOT_DROP and
1426          * add quota pointers back anyway.
1427          */
1428         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1429                 if (inode->i_dquot[cnt])
1430                         break;
1431         }
1432
1433         if (cnt < MAXQUOTAS)
1434                 __dquot_drop(inode);
1435 }
1436 EXPORT_SYMBOL(dquot_drop);
1437
1438 /*
1439  * inode_reserved_space is managed internally by quota, and protected by
1440  * i_lock similar to i_blocks+i_bytes.
1441  */
1442 static qsize_t *inode_reserved_space(struct inode * inode)
1443 {
1444         /* Filesystem must explicitly define it's own method in order to use
1445          * quota reservation interface */
1446         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1447         return inode->i_sb->dq_op->get_reserved_space(inode);
1448 }
1449
1450 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1451 {
1452         spin_lock(&inode->i_lock);
1453         *inode_reserved_space(inode) += number;
1454         spin_unlock(&inode->i_lock);
1455 }
1456 EXPORT_SYMBOL(inode_add_rsv_space);
1457
1458 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1459 {
1460         spin_lock(&inode->i_lock);
1461         *inode_reserved_space(inode) -= number;
1462         __inode_add_bytes(inode, number);
1463         spin_unlock(&inode->i_lock);
1464 }
1465 EXPORT_SYMBOL(inode_claim_rsv_space);
1466
1467 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1468 {
1469         spin_lock(&inode->i_lock);
1470         *inode_reserved_space(inode) -= number;
1471         spin_unlock(&inode->i_lock);
1472 }
1473 EXPORT_SYMBOL(inode_sub_rsv_space);
1474
1475 static qsize_t inode_get_rsv_space(struct inode *inode)
1476 {
1477         qsize_t ret;
1478
1479         if (!inode->i_sb->dq_op->get_reserved_space)
1480                 return 0;
1481         spin_lock(&inode->i_lock);
1482         ret = *inode_reserved_space(inode);
1483         spin_unlock(&inode->i_lock);
1484         return ret;
1485 }
1486
1487 static void inode_incr_space(struct inode *inode, qsize_t number,
1488                                 int reserve)
1489 {
1490         if (reserve)
1491                 inode_add_rsv_space(inode, number);
1492         else
1493                 inode_add_bytes(inode, number);
1494 }
1495
1496 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1497 {
1498         if (reserve)
1499                 inode_sub_rsv_space(inode, number);
1500         else
1501                 inode_sub_bytes(inode, number);
1502 }
1503
1504 /*
1505  * This functions updates i_blocks+i_bytes fields and quota information
1506  * (together with appropriate checks).
1507  *
1508  * NOTE: We absolutely rely on the fact that caller dirties the inode
1509  * (usually helpers in quotaops.h care about this) and holds a handle for
1510  * the current transaction so that dquot write and inode write go into the
1511  * same transaction.
1512  */
1513
1514 /*
1515  * This operation can block, but only after everything is updated
1516  */
1517 int __dquot_alloc_space(struct inode *inode, qsize_t number,
1518                 int warn, int reserve)
1519 {
1520         int cnt, ret = 0;
1521         char warntype[MAXQUOTAS];
1522
1523         /*
1524          * First test before acquiring mutex - solves deadlocks when we
1525          * re-enter the quota code and are already holding the mutex
1526          */
1527         if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1528                 inode_incr_space(inode, number, reserve);
1529                 goto out;
1530         }
1531
1532         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1533         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1534                 warntype[cnt] = QUOTA_NL_NOWARN;
1535
1536         spin_lock(&dq_data_lock);
1537         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1538                 if (!inode->i_dquot[cnt])
1539                         continue;
1540                 ret = check_bdq(inode->i_dquot[cnt], number, !warn,
1541                                 warntype+cnt);
1542                 if (ret) {
1543                         spin_unlock(&dq_data_lock);
1544                         goto out_flush_warn;
1545                 }
1546         }
1547         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1548                 if (!inode->i_dquot[cnt])
1549                         continue;
1550                 if (reserve)
1551                         dquot_resv_space(inode->i_dquot[cnt], number);
1552                 else
1553                         dquot_incr_space(inode->i_dquot[cnt], number);
1554         }
1555         inode_incr_space(inode, number, reserve);
1556         spin_unlock(&dq_data_lock);
1557
1558         if (reserve)
1559                 goto out_flush_warn;
1560         mark_all_dquot_dirty(inode->i_dquot);
1561 out_flush_warn:
1562         flush_warnings(inode->i_dquot, warntype);
1563         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1564 out:
1565         return ret;
1566 }
1567 EXPORT_SYMBOL(__dquot_alloc_space);
1568
1569 /*
1570  * This operation can block, but only after everything is updated
1571  */
1572 int dquot_alloc_inode(const struct inode *inode)
1573 {
1574         int cnt, ret = 0;
1575         char warntype[MAXQUOTAS];
1576
1577         /* First test before acquiring mutex - solves deadlocks when we
1578          * re-enter the quota code and are already holding the mutex */
1579         if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1580                 return 0;
1581         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1582                 warntype[cnt] = QUOTA_NL_NOWARN;
1583         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1584         spin_lock(&dq_data_lock);
1585         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1586                 if (!inode->i_dquot[cnt])
1587                         continue;
1588                 ret = check_idq(inode->i_dquot[cnt], 1, warntype + cnt);
1589                 if (ret)
1590                         goto warn_put_all;
1591         }
1592
1593         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1594                 if (!inode->i_dquot[cnt])
1595                         continue;
1596                 dquot_incr_inodes(inode->i_dquot[cnt], 1);
1597         }
1598
1599 warn_put_all:
1600         spin_unlock(&dq_data_lock);
1601         if (ret == 0)
1602                 mark_all_dquot_dirty(inode->i_dquot);
1603         flush_warnings(inode->i_dquot, warntype);
1604         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1605         return ret;
1606 }
1607 EXPORT_SYMBOL(dquot_alloc_inode);
1608
1609 /*
1610  * Convert in-memory reserved quotas to real consumed quotas
1611  */
1612 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1613 {
1614         int cnt;
1615
1616         if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1617                 inode_claim_rsv_space(inode, number);
1618                 return 0;
1619         }
1620
1621         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1622         spin_lock(&dq_data_lock);
1623         /* Claim reserved quotas to allocated quotas */
1624         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1625                 if (inode->i_dquot[cnt])
1626                         dquot_claim_reserved_space(inode->i_dquot[cnt],
1627                                                         number);
1628         }
1629         /* Update inode bytes */
1630         inode_claim_rsv_space(inode, number);
1631         spin_unlock(&dq_data_lock);
1632         mark_all_dquot_dirty(inode->i_dquot);
1633         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1634         return 0;
1635 }
1636 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1637
1638 /*
1639  * This operation can block, but only after everything is updated
1640  */
1641 void __dquot_free_space(struct inode *inode, qsize_t number, int reserve)
1642 {
1643         unsigned int cnt;
1644         char warntype[MAXQUOTAS];
1645
1646         /* First test before acquiring mutex - solves deadlocks when we
1647          * re-enter the quota code and are already holding the mutex */
1648         if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode)) {
1649                 inode_decr_space(inode, number, reserve);
1650                 return;
1651         }
1652
1653         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1654         spin_lock(&dq_data_lock);
1655         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1656                 if (!inode->i_dquot[cnt])
1657                         continue;
1658                 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1659                 if (reserve)
1660                         dquot_free_reserved_space(inode->i_dquot[cnt], number);
1661                 else
1662                         dquot_decr_space(inode->i_dquot[cnt], number);
1663         }
1664         inode_decr_space(inode, number, reserve);
1665         spin_unlock(&dq_data_lock);
1666
1667         if (reserve)
1668                 goto out_unlock;
1669         mark_all_dquot_dirty(inode->i_dquot);
1670 out_unlock:
1671         flush_warnings(inode->i_dquot, warntype);
1672         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1673 }
1674 EXPORT_SYMBOL(__dquot_free_space);
1675
1676 /*
1677  * This operation can block, but only after everything is updated
1678  */
1679 void dquot_free_inode(const struct inode *inode)
1680 {
1681         unsigned int cnt;
1682         char warntype[MAXQUOTAS];
1683
1684         /* First test before acquiring mutex - solves deadlocks when we
1685          * re-enter the quota code and are already holding the mutex */
1686         if (!sb_any_quota_active(inode->i_sb) || IS_NOQUOTA(inode))
1687                 return;
1688
1689         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1690         spin_lock(&dq_data_lock);
1691         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1692                 if (!inode->i_dquot[cnt])
1693                         continue;
1694                 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], 1);
1695                 dquot_decr_inodes(inode->i_dquot[cnt], 1);
1696         }
1697         spin_unlock(&dq_data_lock);
1698         mark_all_dquot_dirty(inode->i_dquot);
1699         flush_warnings(inode->i_dquot, warntype);
1700         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1701 }
1702 EXPORT_SYMBOL(dquot_free_inode);
1703
1704 /*
1705  * Transfer the number of inode and blocks from one diskquota to an other.
1706  * On success, dquot references in transfer_to are consumed and references
1707  * to original dquots that need to be released are placed there. On failure,
1708  * references are kept untouched.
1709  *
1710  * This operation can block, but only after everything is updated
1711  * A transaction must be started when entering this function.
1712  *
1713  */
1714 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1715 {
1716         qsize_t space, cur_space;
1717         qsize_t rsv_space = 0;
1718         struct dquot *transfer_from[MAXQUOTAS] = {};
1719         int cnt, ret = 0;
1720         char warntype_to[MAXQUOTAS];
1721         char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1722
1723         /* First test before acquiring mutex - solves deadlocks when we
1724          * re-enter the quota code and are already holding the mutex */
1725         if (IS_NOQUOTA(inode))
1726                 return 0;
1727         /* Initialize the arrays */
1728         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1729                 warntype_to[cnt] = QUOTA_NL_NOWARN;
1730         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1731         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1732                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1733                 return 0;
1734         }
1735         spin_lock(&dq_data_lock);
1736         cur_space = inode_get_bytes(inode);
1737         rsv_space = inode_get_rsv_space(inode);
1738         space = cur_space + rsv_space;
1739         /* Build the transfer_from list and check the limits */
1740         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1741                 if (!transfer_to[cnt])
1742                         continue;
1743                 transfer_from[cnt] = inode->i_dquot[cnt];
1744                 ret = check_idq(transfer_to[cnt], 1, warntype_to + cnt);
1745                 if (ret)
1746                         goto over_quota;
1747                 ret = check_bdq(transfer_to[cnt], space, 0, warntype_to + cnt);
1748                 if (ret)
1749                         goto over_quota;
1750         }
1751
1752         /*
1753          * Finally perform the needed transfer from transfer_from to transfer_to
1754          */
1755         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1756                 /*
1757                  * Skip changes for same uid or gid or for turned off quota-type.
1758                  */
1759                 if (!transfer_to[cnt])
1760                         continue;
1761
1762                 /* Due to IO error we might not have transfer_from[] structure */
1763                 if (transfer_from[cnt]) {
1764                         warntype_from_inodes[cnt] =
1765                                 info_idq_free(transfer_from[cnt], 1);
1766                         warntype_from_space[cnt] =
1767                                 info_bdq_free(transfer_from[cnt], space);
1768                         dquot_decr_inodes(transfer_from[cnt], 1);
1769                         dquot_decr_space(transfer_from[cnt], cur_space);
1770                         dquot_free_reserved_space(transfer_from[cnt],
1771                                                   rsv_space);
1772                 }
1773
1774                 dquot_incr_inodes(transfer_to[cnt], 1);
1775                 dquot_incr_space(transfer_to[cnt], cur_space);
1776                 dquot_resv_space(transfer_to[cnt], rsv_space);
1777
1778                 inode->i_dquot[cnt] = transfer_to[cnt];
1779         }
1780         spin_unlock(&dq_data_lock);
1781         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1782
1783         mark_all_dquot_dirty(transfer_from);
1784         mark_all_dquot_dirty(transfer_to);
1785         /* Pass back references to put */
1786         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1787                 transfer_to[cnt] = transfer_from[cnt];
1788 warn:
1789         flush_warnings(transfer_to, warntype_to);
1790         flush_warnings(transfer_from, warntype_from_inodes);
1791         flush_warnings(transfer_from, warntype_from_space);
1792         return ret;
1793 over_quota:
1794         spin_unlock(&dq_data_lock);
1795         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1796         goto warn;
1797 }
1798 EXPORT_SYMBOL(__dquot_transfer);
1799
1800 /* Wrapper for transferring ownership of an inode for uid/gid only
1801  * Called from FSXXX_setattr()
1802  */
1803 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1804 {
1805         struct dquot *transfer_to[MAXQUOTAS] = {};
1806         struct super_block *sb = inode->i_sb;
1807         int ret;
1808
1809         if (!sb_any_quota_active(sb) || IS_NOQUOTA(inode))
1810                 return 0;
1811
1812         if (iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid)
1813                 transfer_to[USRQUOTA] = dqget(sb, iattr->ia_uid, USRQUOTA);
1814         if (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)
1815                 transfer_to[GRPQUOTA] = dqget(sb, iattr->ia_uid, GRPQUOTA);
1816
1817         ret = __dquot_transfer(inode, transfer_to);
1818         dqput_all(transfer_to);
1819         return ret;
1820 }
1821 EXPORT_SYMBOL(dquot_transfer);
1822
1823 /*
1824  * Write info of quota file to disk
1825  */
1826 int dquot_commit_info(struct super_block *sb, int type)
1827 {
1828         int ret;
1829         struct quota_info *dqopt = sb_dqopt(sb);
1830
1831         mutex_lock(&dqopt->dqio_mutex);
1832         ret = dqopt->ops[type]->write_file_info(sb, type);
1833         mutex_unlock(&dqopt->dqio_mutex);
1834         return ret;
1835 }
1836 EXPORT_SYMBOL(dquot_commit_info);
1837
1838 /*
1839  * Definitions of diskquota operations.
1840  */
1841 const struct dquot_operations dquot_operations = {
1842         .write_dquot    = dquot_commit,
1843         .acquire_dquot  = dquot_acquire,
1844         .release_dquot  = dquot_release,
1845         .mark_dirty     = dquot_mark_dquot_dirty,
1846         .write_info     = dquot_commit_info,
1847         .alloc_dquot    = dquot_alloc,
1848         .destroy_dquot  = dquot_destroy,
1849 };
1850
1851 /*
1852  * Generic helper for ->open on filesystems supporting disk quotas.
1853  */
1854 int dquot_file_open(struct inode *inode, struct file *file)
1855 {
1856         int error;
1857
1858         error = generic_file_open(inode, file);
1859         if (!error && (file->f_mode & FMODE_WRITE))
1860                 dquot_initialize(inode);
1861         return error;
1862 }
1863 EXPORT_SYMBOL(dquot_file_open);
1864
1865 /*
1866  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1867  */
1868 int vfs_quota_disable(struct super_block *sb, int type, unsigned int flags)
1869 {
1870         int cnt, ret = 0;
1871         struct quota_info *dqopt = sb_dqopt(sb);
1872         struct inode *toputinode[MAXQUOTAS];
1873
1874         /* Cannot turn off usage accounting without turning off limits, or
1875          * suspend quotas and simultaneously turn quotas off. */
1876         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
1877             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
1878             DQUOT_USAGE_ENABLED)))
1879                 return -EINVAL;
1880
1881         /* We need to serialize quota_off() for device */
1882         mutex_lock(&dqopt->dqonoff_mutex);
1883
1884         /*
1885          * Skip everything if there's nothing to do. We have to do this because
1886          * sometimes we are called when fill_super() failed and calling
1887          * sync_fs() in such cases does no good.
1888          */
1889         if (!sb_any_quota_loaded(sb)) {
1890                 mutex_unlock(&dqopt->dqonoff_mutex);
1891                 return 0;
1892         }
1893         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1894                 toputinode[cnt] = NULL;
1895                 if (type != -1 && cnt != type)
1896                         continue;
1897                 if (!sb_has_quota_loaded(sb, cnt))
1898                         continue;
1899
1900                 if (flags & DQUOT_SUSPENDED) {
1901                         spin_lock(&dq_state_lock);
1902                         dqopt->flags |=
1903                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
1904                         spin_unlock(&dq_state_lock);
1905                 } else {
1906                         spin_lock(&dq_state_lock);
1907                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
1908                         /* Turning off suspended quotas? */
1909                         if (!sb_has_quota_loaded(sb, cnt) &&
1910                             sb_has_quota_suspended(sb, cnt)) {
1911                                 dqopt->flags &= ~dquot_state_flag(
1912                                                         DQUOT_SUSPENDED, cnt);
1913                                 spin_unlock(&dq_state_lock);
1914                                 iput(dqopt->files[cnt]);
1915                                 dqopt->files[cnt] = NULL;
1916                                 continue;
1917                         }
1918                         spin_unlock(&dq_state_lock);
1919                 }
1920
1921                 /* We still have to keep quota loaded? */
1922                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
1923                         continue;
1924
1925                 /* Note: these are blocking operations */
1926                 drop_dquot_ref(sb, cnt);
1927                 invalidate_dquots(sb, cnt);
1928                 /*
1929                  * Now all dquots should be invalidated, all writes done so we
1930                  * should be only users of the info. No locks needed.
1931                  */
1932                 if (info_dirty(&dqopt->info[cnt]))
1933                         sb->dq_op->write_info(sb, cnt);
1934                 if (dqopt->ops[cnt]->free_file_info)
1935                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1936                 put_quota_format(dqopt->info[cnt].dqi_format);
1937
1938                 toputinode[cnt] = dqopt->files[cnt];
1939                 if (!sb_has_quota_loaded(sb, cnt))
1940                         dqopt->files[cnt] = NULL;
1941                 dqopt->info[cnt].dqi_flags = 0;
1942                 dqopt->info[cnt].dqi_igrace = 0;
1943                 dqopt->info[cnt].dqi_bgrace = 0;
1944                 dqopt->ops[cnt] = NULL;
1945         }
1946         mutex_unlock(&dqopt->dqonoff_mutex);
1947
1948         /* Skip syncing and setting flags if quota files are hidden */
1949         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
1950                 goto put_inodes;
1951
1952         /* Sync the superblock so that buffers with quota data are written to
1953          * disk (and so userspace sees correct data afterwards). */
1954         if (sb->s_op->sync_fs)
1955                 sb->s_op->sync_fs(sb, 1);
1956         sync_blockdev(sb->s_bdev);
1957         /* Now the quota files are just ordinary files and we can set the
1958          * inode flags back. Moreover we discard the pagecache so that
1959          * userspace sees the writes we did bypassing the pagecache. We
1960          * must also discard the blockdev buffers so that we see the
1961          * changes done by userspace on the next quotaon() */
1962         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1963                 if (toputinode[cnt]) {
1964                         mutex_lock(&dqopt->dqonoff_mutex);
1965                         /* If quota was reenabled in the meantime, we have
1966                          * nothing to do */
1967                         if (!sb_has_quota_loaded(sb, cnt)) {
1968                                 mutex_lock_nested(&toputinode[cnt]->i_mutex,
1969                                                   I_MUTEX_QUOTA);
1970                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1971                                   S_NOATIME | S_NOQUOTA);
1972                                 truncate_inode_pages(&toputinode[cnt]->i_data,
1973                                                      0);
1974                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1975                                 mark_inode_dirty(toputinode[cnt]);
1976                         }
1977                         mutex_unlock(&dqopt->dqonoff_mutex);
1978                 }
1979         if (sb->s_bdev)
1980                 invalidate_bdev(sb->s_bdev);
1981 put_inodes:
1982         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1983                 if (toputinode[cnt]) {
1984                         /* On remount RO, we keep the inode pointer so that we
1985                          * can reenable quota on the subsequent remount RW. We
1986                          * have to check 'flags' variable and not use sb_has_
1987                          * function because another quotaon / quotaoff could
1988                          * change global state before we got here. We refuse
1989                          * to suspend quotas when there is pending delete on
1990                          * the quota file... */
1991                         if (!(flags & DQUOT_SUSPENDED))
1992                                 iput(toputinode[cnt]);
1993                         else if (!toputinode[cnt]->i_nlink)
1994                                 ret = -EBUSY;
1995                 }
1996         return ret;
1997 }
1998 EXPORT_SYMBOL(vfs_quota_disable);
1999
2000 int vfs_quota_off(struct super_block *sb, int type, int remount)
2001 {
2002         return vfs_quota_disable(sb, type, remount ? DQUOT_SUSPENDED :
2003                                  (DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED));
2004 }
2005 EXPORT_SYMBOL(vfs_quota_off);
2006 /*
2007  *      Turn quotas on on a device
2008  */
2009
2010 /*
2011  * Helper function to turn quotas on when we already have the inode of
2012  * quota file and no quota information is loaded.
2013  */
2014 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2015         unsigned int flags)
2016 {
2017         struct quota_format_type *fmt = find_quota_format(format_id);
2018         struct super_block *sb = inode->i_sb;
2019         struct quota_info *dqopt = sb_dqopt(sb);
2020         int error;
2021         int oldflags = -1;
2022
2023         if (!fmt)
2024                 return -ESRCH;
2025         if (!S_ISREG(inode->i_mode)) {
2026                 error = -EACCES;
2027                 goto out_fmt;
2028         }
2029         if (IS_RDONLY(inode)) {
2030                 error = -EROFS;
2031                 goto out_fmt;
2032         }
2033         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
2034                 error = -EINVAL;
2035                 goto out_fmt;
2036         }
2037         /* Usage always has to be set... */
2038         if (!(flags & DQUOT_USAGE_ENABLED)) {
2039                 error = -EINVAL;
2040                 goto out_fmt;
2041         }
2042
2043         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2044                 /* As we bypass the pagecache we must now flush all the
2045                  * dirty data and invalidate caches so that kernel sees
2046                  * changes from userspace. It is not enough to just flush
2047                  * the quota file since if blocksize < pagesize, invalidation
2048                  * of the cache could fail because of other unrelated dirty
2049                  * data */
2050                 sync_filesystem(sb);
2051                 invalidate_bdev(sb->s_bdev);
2052         }
2053         mutex_lock(&dqopt->dqonoff_mutex);
2054         if (sb_has_quota_loaded(sb, type)) {
2055                 error = -EBUSY;
2056                 goto out_lock;
2057         }
2058
2059         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2060                 /* We don't want quota and atime on quota files (deadlocks
2061                  * possible) Also nobody should write to the file - we use
2062                  * special IO operations which ignore the immutable bit. */
2063                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2064                 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2065                                              S_NOQUOTA);
2066                 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2067                 mutex_unlock(&inode->i_mutex);
2068                 /*
2069                  * When S_NOQUOTA is set, remove dquot references as no more
2070                  * references can be added
2071                  */
2072                 __dquot_drop(inode);
2073         }
2074
2075         error = -EIO;
2076         dqopt->files[type] = igrab(inode);
2077         if (!dqopt->files[type])
2078                 goto out_lock;
2079         error = -EINVAL;
2080         if (!fmt->qf_ops->check_quota_file(sb, type))
2081                 goto out_file_init;
2082
2083         dqopt->ops[type] = fmt->qf_ops;
2084         dqopt->info[type].dqi_format = fmt;
2085         dqopt->info[type].dqi_fmt_id = format_id;
2086         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2087         mutex_lock(&dqopt->dqio_mutex);
2088         error = dqopt->ops[type]->read_file_info(sb, type);
2089         if (error < 0) {
2090                 mutex_unlock(&dqopt->dqio_mutex);
2091                 goto out_file_init;
2092         }
2093         mutex_unlock(&dqopt->dqio_mutex);
2094         spin_lock(&dq_state_lock);
2095         dqopt->flags |= dquot_state_flag(flags, type);
2096         spin_unlock(&dq_state_lock);
2097
2098         add_dquot_ref(sb, type);
2099         mutex_unlock(&dqopt->dqonoff_mutex);
2100
2101         return 0;
2102
2103 out_file_init:
2104         dqopt->files[type] = NULL;
2105         iput(inode);
2106 out_lock:
2107         if (oldflags != -1) {
2108                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2109                 /* Set the flags back (in the case of accidental quotaon()
2110                  * on a wrong file we don't want to mess up the flags) */
2111                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2112                 inode->i_flags |= oldflags;
2113                 mutex_unlock(&inode->i_mutex);
2114         }
2115         mutex_unlock(&dqopt->dqonoff_mutex);
2116 out_fmt:
2117         put_quota_format(fmt);
2118
2119         return error; 
2120 }
2121
2122 /* Reenable quotas on remount RW */
2123 static int vfs_quota_on_remount(struct super_block *sb, int type)
2124 {
2125         struct quota_info *dqopt = sb_dqopt(sb);
2126         struct inode *inode;
2127         int ret;
2128         unsigned int flags;
2129
2130         mutex_lock(&dqopt->dqonoff_mutex);
2131         if (!sb_has_quota_suspended(sb, type)) {
2132                 mutex_unlock(&dqopt->dqonoff_mutex);
2133                 return 0;
2134         }
2135         inode = dqopt->files[type];
2136         dqopt->files[type] = NULL;
2137         spin_lock(&dq_state_lock);
2138         flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2139                                                 DQUOT_LIMITS_ENABLED, type);
2140         dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, type);
2141         spin_unlock(&dq_state_lock);
2142         mutex_unlock(&dqopt->dqonoff_mutex);
2143
2144         flags = dquot_generic_flag(flags, type);
2145         ret = vfs_load_quota_inode(inode, type, dqopt->info[type].dqi_fmt_id,
2146                                    flags);
2147         iput(inode);
2148
2149         return ret;
2150 }
2151
2152 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
2153                       struct path *path)
2154 {
2155         int error = security_quota_on(path->dentry);
2156         if (error)
2157                 return error;
2158         /* Quota file not on the same filesystem? */
2159         if (path->mnt->mnt_sb != sb)
2160                 error = -EXDEV;
2161         else
2162                 error = vfs_load_quota_inode(path->dentry->d_inode, type,
2163                                              format_id, DQUOT_USAGE_ENABLED |
2164                                              DQUOT_LIMITS_ENABLED);
2165         return error;
2166 }
2167 EXPORT_SYMBOL(vfs_quota_on_path);
2168
2169 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
2170                  int remount)
2171 {
2172         struct path path;
2173         int error;
2174
2175         if (remount)
2176                 return vfs_quota_on_remount(sb, type);
2177
2178         error = kern_path(name, LOOKUP_FOLLOW, &path);
2179         if (!error) {
2180                 error = vfs_quota_on_path(sb, type, format_id, &path);
2181                 path_put(&path);
2182         }
2183         return error;
2184 }
2185 EXPORT_SYMBOL(vfs_quota_on);
2186
2187 /*
2188  * More powerful function for turning on quotas allowing setting
2189  * of individual quota flags
2190  */
2191 int vfs_quota_enable(struct inode *inode, int type, int format_id,
2192                 unsigned int flags)
2193 {
2194         int ret = 0;
2195         struct super_block *sb = inode->i_sb;
2196         struct quota_info *dqopt = sb_dqopt(sb);
2197
2198         /* Just unsuspend quotas? */
2199         if (flags & DQUOT_SUSPENDED)
2200                 return vfs_quota_on_remount(sb, type);
2201         if (!flags)
2202                 return 0;
2203         /* Just updating flags needed? */
2204         if (sb_has_quota_loaded(sb, type)) {
2205                 mutex_lock(&dqopt->dqonoff_mutex);
2206                 /* Now do a reliable test... */
2207                 if (!sb_has_quota_loaded(sb, type)) {
2208                         mutex_unlock(&dqopt->dqonoff_mutex);
2209                         goto load_quota;
2210                 }
2211                 if (flags & DQUOT_USAGE_ENABLED &&
2212                     sb_has_quota_usage_enabled(sb, type)) {
2213                         ret = -EBUSY;
2214                         goto out_lock;
2215                 }
2216                 if (flags & DQUOT_LIMITS_ENABLED &&
2217                     sb_has_quota_limits_enabled(sb, type)) {
2218                         ret = -EBUSY;
2219                         goto out_lock;
2220                 }
2221                 spin_lock(&dq_state_lock);
2222                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2223                 spin_unlock(&dq_state_lock);
2224 out_lock:
2225                 mutex_unlock(&dqopt->dqonoff_mutex);
2226                 return ret;
2227         }
2228
2229 load_quota:
2230         return vfs_load_quota_inode(inode, type, format_id, flags);
2231 }
2232 EXPORT_SYMBOL(vfs_quota_enable);
2233
2234 /*
2235  * This function is used when filesystem needs to initialize quotas
2236  * during mount time.
2237  */
2238 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
2239                 int format_id, int type)
2240 {
2241         struct dentry *dentry;
2242         int error;
2243
2244         mutex_lock(&sb->s_root->d_inode->i_mutex);
2245         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2246         mutex_unlock(&sb->s_root->d_inode->i_mutex);
2247         if (IS_ERR(dentry))
2248                 return PTR_ERR(dentry);
2249
2250         if (!dentry->d_inode) {
2251                 error = -ENOENT;
2252                 goto out;
2253         }
2254
2255         error = security_quota_on(dentry);
2256         if (!error)
2257                 error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
2258                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2259
2260 out:
2261         dput(dentry);
2262         return error;
2263 }
2264 EXPORT_SYMBOL(vfs_quota_on_mount);
2265
2266 /* Wrapper to turn on quotas when remounting rw */
2267 int vfs_dq_quota_on_remount(struct super_block *sb)
2268 {
2269         int cnt;
2270         int ret = 0, err;
2271
2272         if (!sb->s_qcop || !sb->s_qcop->quota_on)
2273                 return -ENOSYS;
2274         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2275                 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
2276                 if (err < 0 && !ret)
2277                         ret = err;
2278         }
2279         return ret;
2280 }
2281 EXPORT_SYMBOL(vfs_dq_quota_on_remount);
2282
2283 static inline qsize_t qbtos(qsize_t blocks)
2284 {
2285         return blocks << QIF_DQBLKSIZE_BITS;
2286 }
2287
2288 static inline qsize_t stoqb(qsize_t space)
2289 {
2290         return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
2291 }
2292
2293 /* Generic routine for getting common part of quota structure */
2294 static void do_get_dqblk(struct dquot *dquot, struct fs_disk_quota *di)
2295 {
2296         struct mem_dqblk *dm = &dquot->dq_dqb;
2297
2298         memset(di, 0, sizeof(*di));
2299         di->d_version = FS_DQUOT_VERSION;
2300         di->d_flags = dquot->dq_type == USRQUOTA ?
2301                         XFS_USER_QUOTA : XFS_GROUP_QUOTA;
2302         di->d_id = dquot->dq_id;
2303
2304         spin_lock(&dq_data_lock);
2305         di->d_blk_hardlimit = stoqb(dm->dqb_bhardlimit);
2306         di->d_blk_softlimit = stoqb(dm->dqb_bsoftlimit);
2307         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2308         di->d_ino_softlimit = dm->dqb_isoftlimit;
2309         di->d_bcount = dm->dqb_curspace + dm->dqb_rsvspace;
2310         di->d_icount = dm->dqb_curinodes;
2311         di->d_btimer = dm->dqb_btime;
2312         di->d_itimer = dm->dqb_itime;
2313         spin_unlock(&dq_data_lock);
2314 }
2315
2316 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id,
2317                   struct fs_disk_quota *di)
2318 {
2319         struct dquot *dquot;
2320
2321         dquot = dqget(sb, id, type);
2322         if (!dquot)
2323                 return -ESRCH;
2324         do_get_dqblk(dquot, di);
2325         dqput(dquot);
2326
2327         return 0;
2328 }
2329 EXPORT_SYMBOL(vfs_get_dqblk);
2330
2331 #define VFS_FS_DQ_MASK \
2332         (FS_DQ_BCOUNT | FS_DQ_BSOFT | FS_DQ_BHARD | \
2333          FS_DQ_ICOUNT | FS_DQ_ISOFT | FS_DQ_IHARD | \
2334          FS_DQ_BTIMER | FS_DQ_ITIMER)
2335
2336 /* Generic routine for setting common part of quota structure */
2337 static int do_set_dqblk(struct dquot *dquot, struct fs_disk_quota *di)
2338 {
2339         struct mem_dqblk *dm = &dquot->dq_dqb;
2340         int check_blim = 0, check_ilim = 0;
2341         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
2342
2343         if (di->d_fieldmask & ~VFS_FS_DQ_MASK)
2344                 return -EINVAL;
2345
2346         if (((di->d_fieldmask & FS_DQ_BSOFT) &&
2347              (di->d_blk_softlimit > dqi->dqi_maxblimit)) ||
2348             ((di->d_fieldmask & FS_DQ_BHARD) &&
2349              (di->d_blk_hardlimit > dqi->dqi_maxblimit)) ||
2350             ((di->d_fieldmask & FS_DQ_ISOFT) &&
2351              (di->d_ino_softlimit > dqi->dqi_maxilimit)) ||
2352             ((di->d_fieldmask & FS_DQ_IHARD) &&
2353              (di->d_ino_hardlimit > dqi->dqi_maxilimit)))
2354                 return -ERANGE;
2355
2356         spin_lock(&dq_data_lock);
2357         if (di->d_fieldmask & FS_DQ_BCOUNT) {
2358                 dm->dqb_curspace = di->d_bcount - dm->dqb_rsvspace;
2359                 check_blim = 1;
2360                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2361         }
2362
2363         if (di->d_fieldmask & FS_DQ_BSOFT)
2364                 dm->dqb_bsoftlimit = qbtos(di->d_blk_softlimit);
2365         if (di->d_fieldmask & FS_DQ_BHARD)
2366                 dm->dqb_bhardlimit = qbtos(di->d_blk_hardlimit);
2367         if (di->d_fieldmask & (FS_DQ_BSOFT | FS_DQ_BHARD)) {
2368                 check_blim = 1;
2369                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2370         }
2371
2372         if (di->d_fieldmask & FS_DQ_ICOUNT) {
2373                 dm->dqb_curinodes = di->d_icount;
2374                 check_ilim = 1;
2375                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2376         }
2377
2378         if (di->d_fieldmask & FS_DQ_ISOFT)
2379                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2380         if (di->d_fieldmask & FS_DQ_IHARD)
2381                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2382         if (di->d_fieldmask & (FS_DQ_ISOFT | FS_DQ_IHARD)) {
2383                 check_ilim = 1;
2384                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2385         }
2386
2387         if (di->d_fieldmask & FS_DQ_BTIMER) {
2388                 dm->dqb_btime = di->d_btimer;
2389                 check_blim = 1;
2390                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2391         }
2392
2393         if (di->d_fieldmask & FS_DQ_ITIMER) {
2394                 dm->dqb_itime = di->d_itimer;
2395                 check_ilim = 1;
2396                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2397         }
2398
2399         if (check_blim) {
2400                 if (!dm->dqb_bsoftlimit ||
2401                     dm->dqb_curspace < dm->dqb_bsoftlimit) {
2402                         dm->dqb_btime = 0;
2403                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2404                 } else if (!(di->d_fieldmask & FS_DQ_BTIMER))
2405                         /* Set grace only if user hasn't provided his own... */
2406                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2407         }
2408         if (check_ilim) {
2409                 if (!dm->dqb_isoftlimit ||
2410                     dm->dqb_curinodes < dm->dqb_isoftlimit) {
2411                         dm->dqb_itime = 0;
2412                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2413                 } else if (!(di->d_fieldmask & FS_DQ_ITIMER))
2414                         /* Set grace only if user hasn't provided his own... */
2415                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2416         }
2417         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2418             dm->dqb_isoftlimit)
2419                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2420         else
2421                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2422         spin_unlock(&dq_data_lock);
2423         mark_dquot_dirty(dquot);
2424
2425         return 0;
2426 }
2427
2428 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id,
2429                   struct fs_disk_quota *di)
2430 {
2431         struct dquot *dquot;
2432         int rc;
2433
2434         dquot = dqget(sb, id, type);
2435         if (!dquot) {
2436                 rc = -ESRCH;
2437                 goto out;
2438         }
2439         rc = do_set_dqblk(dquot, di);
2440         dqput(dquot);
2441 out:
2442         return rc;
2443 }
2444 EXPORT_SYMBOL(vfs_set_dqblk);
2445
2446 /* Generic routine for getting common part of quota file information */
2447 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2448 {
2449         struct mem_dqinfo *mi;
2450   
2451         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2452         if (!sb_has_quota_active(sb, type)) {
2453                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2454                 return -ESRCH;
2455         }
2456         mi = sb_dqopt(sb)->info + type;
2457         spin_lock(&dq_data_lock);
2458         ii->dqi_bgrace = mi->dqi_bgrace;
2459         ii->dqi_igrace = mi->dqi_igrace;
2460         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2461         ii->dqi_valid = IIF_ALL;
2462         spin_unlock(&dq_data_lock);
2463         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2464         return 0;
2465 }
2466 EXPORT_SYMBOL(vfs_get_dqinfo);
2467
2468 /* Generic routine for setting common part of quota file information */
2469 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2470 {
2471         struct mem_dqinfo *mi;
2472         int err = 0;
2473
2474         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2475         if (!sb_has_quota_active(sb, type)) {
2476                 err = -ESRCH;
2477                 goto out;
2478         }
2479         mi = sb_dqopt(sb)->info + type;
2480         spin_lock(&dq_data_lock);
2481         if (ii->dqi_valid & IIF_BGRACE)
2482                 mi->dqi_bgrace = ii->dqi_bgrace;
2483         if (ii->dqi_valid & IIF_IGRACE)
2484                 mi->dqi_igrace = ii->dqi_igrace;
2485         if (ii->dqi_valid & IIF_FLAGS)
2486                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) |
2487                                 (ii->dqi_flags & DQF_MASK);
2488         spin_unlock(&dq_data_lock);
2489         mark_info_dirty(sb, type);
2490         /* Force write to disk */
2491         sb->dq_op->write_info(sb, type);
2492 out:
2493         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2494         return err;
2495 }
2496 EXPORT_SYMBOL(vfs_set_dqinfo);
2497
2498 const struct quotactl_ops vfs_quotactl_ops = {
2499         .quota_on       = vfs_quota_on,
2500         .quota_off      = vfs_quota_off,
2501         .quota_sync     = vfs_quota_sync,
2502         .get_info       = vfs_get_dqinfo,
2503         .set_info       = vfs_set_dqinfo,
2504         .get_dqblk      = vfs_get_dqblk,
2505         .set_dqblk      = vfs_set_dqblk
2506 };
2507
2508
2509 static int do_proc_dqstats(struct ctl_table *table, int write,
2510                      void __user *buffer, size_t *lenp, loff_t *ppos)
2511 {
2512 #ifdef CONFIG_SMP
2513         /* Update global table */
2514         unsigned int type = (int *)table->data - dqstats.stat;
2515         dqstats.stat[type] = dqstats_read(type);
2516 #endif
2517         return proc_dointvec(table, write, buffer, lenp, ppos);
2518 }
2519
2520 static ctl_table fs_dqstats_table[] = {
2521         {
2522                 .procname       = "lookups",
2523                 .data           = &dqstats.stat[DQST_LOOKUPS],
2524                 .maxlen         = sizeof(int),
2525                 .mode           = 0444,
2526                 .proc_handler   = do_proc_dqstats,
2527         },
2528         {
2529                 .procname       = "drops",
2530                 .data           = &dqstats.stat[DQST_DROPS],
2531                 .maxlen         = sizeof(int),
2532                 .mode           = 0444,
2533                 .proc_handler   = do_proc_dqstats,
2534         },
2535         {
2536                 .procname       = "reads",
2537                 .data           = &dqstats.stat[DQST_READS],
2538                 .maxlen         = sizeof(int),
2539                 .mode           = 0444,
2540                 .proc_handler   = do_proc_dqstats,
2541         },
2542         {
2543                 .procname       = "writes",
2544                 .data           = &dqstats.stat[DQST_WRITES],
2545                 .maxlen         = sizeof(int),
2546                 .mode           = 0444,
2547                 .proc_handler   = do_proc_dqstats,
2548         },
2549         {
2550                 .procname       = "cache_hits",
2551                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2552                 .maxlen         = sizeof(int),
2553                 .mode           = 0444,
2554                 .proc_handler   = do_proc_dqstats,
2555         },
2556         {
2557                 .procname       = "allocated_dquots",
2558                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2559                 .maxlen         = sizeof(int),
2560                 .mode           = 0444,
2561                 .proc_handler   = do_proc_dqstats,
2562         },
2563         {
2564                 .procname       = "free_dquots",
2565                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2566                 .maxlen         = sizeof(int),
2567                 .mode           = 0444,
2568                 .proc_handler   = do_proc_dqstats,
2569         },
2570         {
2571                 .procname       = "syncs",
2572                 .data           = &dqstats.stat[DQST_SYNCS],
2573                 .maxlen         = sizeof(int),
2574                 .mode           = 0444,
2575                 .proc_handler   = do_proc_dqstats,
2576         },
2577 #ifdef CONFIG_PRINT_QUOTA_WARNING
2578         {
2579                 .procname       = "warnings",
2580                 .data           = &flag_print_warnings,
2581                 .maxlen         = sizeof(int),
2582                 .mode           = 0644,
2583                 .proc_handler   = proc_dointvec,
2584         },
2585 #endif
2586         { },
2587 };
2588
2589 static ctl_table fs_table[] = {
2590         {
2591                 .procname       = "quota",
2592                 .mode           = 0555,
2593                 .child          = fs_dqstats_table,
2594         },
2595         { },
2596 };
2597
2598 static ctl_table sys_table[] = {
2599         {
2600                 .procname       = "fs",
2601                 .mode           = 0555,
2602                 .child          = fs_table,
2603         },
2604         { },
2605 };
2606
2607 static int __init dquot_init(void)
2608 {
2609         int i;
2610         unsigned long nr_hash, order;
2611
2612         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2613
2614         register_sysctl_table(sys_table);
2615
2616         dquot_cachep = kmem_cache_create("dquot",
2617                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2618                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2619                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2620                         NULL);
2621
2622         order = 0;
2623         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2624         if (!dquot_hash)
2625                 panic("Cannot create dquot hash table");
2626
2627 #ifdef CONFIG_SMP
2628         dqstats_pcpu = alloc_percpu(struct dqstats);
2629         if (!dqstats_pcpu)
2630                 panic("Cannot create dquot stats table");
2631 #endif
2632         memset(&dqstats, 0, sizeof(struct dqstats));
2633
2634         /* Find power-of-two hlist_heads which can fit into allocation */
2635         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2636         dq_hash_bits = 0;
2637         do {
2638                 dq_hash_bits++;
2639         } while (nr_hash >> dq_hash_bits);
2640         dq_hash_bits--;
2641
2642         nr_hash = 1UL << dq_hash_bits;
2643         dq_hash_mask = nr_hash - 1;
2644         for (i = 0; i < nr_hash; i++)
2645                 INIT_HLIST_HEAD(dquot_hash + i);
2646
2647         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2648                         nr_hash, order, (PAGE_SIZE << order));
2649
2650         register_shrinker(&dqcache_shrinker);
2651
2652         return 0;
2653 }
2654 module_init(dquot_init);