These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / 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/sched.h>
75 #include <linux/kmod.h>
76 #include <linux/namei.h>
77 #include <linux/capability.h>
78 #include <linux/quotaops.h>
79 #include "../internal.h" /* ugh */
80
81 #include <linux/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  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
100  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
101  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
102  * inode and before dropping dquot references to avoid use of dquots after
103  * they are freed. dq_data_lock is used to serialize the pointer setting and
104  * clearing operations.
105  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
106  * inode is a quota file). Functions adding pointers from inode to dquots have
107  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
108  * have to do all pointer modifications before dropping dq_data_lock. This makes
109  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
110  * then drops all pointers to dquots from an inode.
111  *
112  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
113  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
114  * Currently dquot is locked only when it is being read to memory (or space for
115  * it is being allocated) on the first dqget() and when it is being released on
116  * the last dqput(). The allocation and release oparations are serialized by
117  * the dq_lock and by checking the use count in dquot_release().  Write
118  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
119  * spinlock to internal buffers before writing.
120  *
121  * Lock ordering (including related VFS locks) is the following:
122  *   dqonoff_mutex > i_mutex > journal_lock > dquot->dq_lock > dqio_mutex
123  * dqonoff_mutex > i_mutex comes from dquot_quota_sync, dquot_enable, etc.
124  */
125
126 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
128 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
129 EXPORT_SYMBOL(dq_data_lock);
130 DEFINE_STATIC_SRCU(dquot_srcu);
131
132 void __quota_error(struct super_block *sb, const char *func,
133                    const char *fmt, ...)
134 {
135         if (printk_ratelimit()) {
136                 va_list args;
137                 struct va_format vaf;
138
139                 va_start(args, fmt);
140
141                 vaf.fmt = fmt;
142                 vaf.va = &args;
143
144                 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
145                        sb->s_id, func, &vaf);
146
147                 va_end(args);
148         }
149 }
150 EXPORT_SYMBOL(__quota_error);
151
152 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
153 static char *quotatypes[] = INITQFNAMES;
154 #endif
155 static struct quota_format_type *quota_formats; /* List of registered formats */
156 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
157
158 /* SLAB cache for dquot structures */
159 static struct kmem_cache *dquot_cachep;
160
161 int register_quota_format(struct quota_format_type *fmt)
162 {
163         spin_lock(&dq_list_lock);
164         fmt->qf_next = quota_formats;
165         quota_formats = fmt;
166         spin_unlock(&dq_list_lock);
167         return 0;
168 }
169 EXPORT_SYMBOL(register_quota_format);
170
171 void unregister_quota_format(struct quota_format_type *fmt)
172 {
173         struct quota_format_type **actqf;
174
175         spin_lock(&dq_list_lock);
176         for (actqf = &quota_formats; *actqf && *actqf != fmt;
177              actqf = &(*actqf)->qf_next)
178                 ;
179         if (*actqf)
180                 *actqf = (*actqf)->qf_next;
181         spin_unlock(&dq_list_lock);
182 }
183 EXPORT_SYMBOL(unregister_quota_format);
184
185 static struct quota_format_type *find_quota_format(int id)
186 {
187         struct quota_format_type *actqf;
188
189         spin_lock(&dq_list_lock);
190         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
191              actqf = actqf->qf_next)
192                 ;
193         if (!actqf || !try_module_get(actqf->qf_owner)) {
194                 int qm;
195
196                 spin_unlock(&dq_list_lock);
197                 
198                 for (qm = 0; module_names[qm].qm_fmt_id &&
199                              module_names[qm].qm_fmt_id != id; qm++)
200                         ;
201                 if (!module_names[qm].qm_fmt_id ||
202                     request_module(module_names[qm].qm_mod_name))
203                         return NULL;
204
205                 spin_lock(&dq_list_lock);
206                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
207                      actqf = actqf->qf_next)
208                         ;
209                 if (actqf && !try_module_get(actqf->qf_owner))
210                         actqf = NULL;
211         }
212         spin_unlock(&dq_list_lock);
213         return actqf;
214 }
215
216 static void put_quota_format(struct quota_format_type *fmt)
217 {
218         module_put(fmt->qf_owner);
219 }
220
221 /*
222  * Dquot List Management:
223  * The quota code uses three lists for dquot management: the inuse_list,
224  * free_dquots, and dquot_hash[] array. A single dquot structure may be
225  * on all three lists, depending on its current state.
226  *
227  * All dquots are placed to the end of inuse_list when first created, and this
228  * list is used for invalidate operation, which must look at every dquot.
229  *
230  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
231  * and this list is searched whenever we need an available dquot.  Dquots are
232  * removed from the list as soon as they are used again, and
233  * dqstats.free_dquots gives the number of dquots on the list. When
234  * dquot is invalidated it's completely released from memory.
235  *
236  * Dquots with a specific identity (device, type and id) are placed on
237  * one of the dquot_hash[] hash chains. The provides an efficient search
238  * mechanism to locate a specific dquot.
239  */
240
241 static LIST_HEAD(inuse_list);
242 static LIST_HEAD(free_dquots);
243 static unsigned int dq_hash_bits, dq_hash_mask;
244 static struct hlist_head *dquot_hash;
245
246 struct dqstats dqstats;
247 EXPORT_SYMBOL(dqstats);
248
249 static qsize_t inode_get_rsv_space(struct inode *inode);
250 static int __dquot_initialize(struct inode *inode, int type);
251
252 static inline unsigned int
253 hashfn(const struct super_block *sb, struct kqid qid)
254 {
255         unsigned int id = from_kqid(&init_user_ns, qid);
256         int type = qid.type;
257         unsigned long tmp;
258
259         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
260         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
261 }
262
263 /*
264  * Following list functions expect dq_list_lock to be held
265  */
266 static inline void insert_dquot_hash(struct dquot *dquot)
267 {
268         struct hlist_head *head;
269         head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
270         hlist_add_head(&dquot->dq_hash, head);
271 }
272
273 static inline void remove_dquot_hash(struct dquot *dquot)
274 {
275         hlist_del_init(&dquot->dq_hash);
276 }
277
278 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
279                                 struct kqid qid)
280 {
281         struct hlist_node *node;
282         struct dquot *dquot;
283
284         hlist_for_each (node, dquot_hash+hashent) {
285                 dquot = hlist_entry(node, struct dquot, dq_hash);
286                 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
287                         return dquot;
288         }
289         return NULL;
290 }
291
292 /* Add a dquot to the tail of the free list */
293 static inline void put_dquot_last(struct dquot *dquot)
294 {
295         list_add_tail(&dquot->dq_free, &free_dquots);
296         dqstats_inc(DQST_FREE_DQUOTS);
297 }
298
299 static inline void remove_free_dquot(struct dquot *dquot)
300 {
301         if (list_empty(&dquot->dq_free))
302                 return;
303         list_del_init(&dquot->dq_free);
304         dqstats_dec(DQST_FREE_DQUOTS);
305 }
306
307 static inline void put_inuse(struct dquot *dquot)
308 {
309         /* We add to the back of inuse list so we don't have to restart
310          * when traversing this list and we block */
311         list_add_tail(&dquot->dq_inuse, &inuse_list);
312         dqstats_inc(DQST_ALLOC_DQUOTS);
313 }
314
315 static inline void remove_inuse(struct dquot *dquot)
316 {
317         dqstats_dec(DQST_ALLOC_DQUOTS);
318         list_del(&dquot->dq_inuse);
319 }
320 /*
321  * End of list functions needing dq_list_lock
322  */
323
324 static void wait_on_dquot(struct dquot *dquot)
325 {
326         mutex_lock(&dquot->dq_lock);
327         mutex_unlock(&dquot->dq_lock);
328 }
329
330 static inline int dquot_dirty(struct dquot *dquot)
331 {
332         return test_bit(DQ_MOD_B, &dquot->dq_flags);
333 }
334
335 static inline int mark_dquot_dirty(struct dquot *dquot)
336 {
337         return dquot->dq_sb->dq_op->mark_dirty(dquot);
338 }
339
340 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
341 int dquot_mark_dquot_dirty(struct dquot *dquot)
342 {
343         int ret = 1;
344
345         /* If quota is dirty already, we don't have to acquire dq_list_lock */
346         if (test_bit(DQ_MOD_B, &dquot->dq_flags))
347                 return 1;
348
349         spin_lock(&dq_list_lock);
350         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
351                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
352                                 info[dquot->dq_id.type].dqi_dirty_list);
353                 ret = 0;
354         }
355         spin_unlock(&dq_list_lock);
356         return ret;
357 }
358 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
359
360 /* Dirtify all the dquots - this can block when journalling */
361 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
362 {
363         int ret, err, cnt;
364
365         ret = err = 0;
366         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
367                 if (dquot[cnt])
368                         /* Even in case of error we have to continue */
369                         ret = mark_dquot_dirty(dquot[cnt]);
370                 if (!err)
371                         err = ret;
372         }
373         return err;
374 }
375
376 static inline void dqput_all(struct dquot **dquot)
377 {
378         unsigned int cnt;
379
380         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
381                 dqput(dquot[cnt]);
382 }
383
384 /* This function needs dq_list_lock */
385 static inline int clear_dquot_dirty(struct dquot *dquot)
386 {
387         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
388                 return 0;
389         list_del_init(&dquot->dq_dirty);
390         return 1;
391 }
392
393 void mark_info_dirty(struct super_block *sb, int type)
394 {
395         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
396 }
397 EXPORT_SYMBOL(mark_info_dirty);
398
399 /*
400  *      Read dquot from disk and alloc space for it
401  */
402
403 int dquot_acquire(struct dquot *dquot)
404 {
405         int ret = 0, ret2 = 0;
406         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
407
408         mutex_lock(&dquot->dq_lock);
409         mutex_lock(&dqopt->dqio_mutex);
410         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
411                 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
412         if (ret < 0)
413                 goto out_iolock;
414         set_bit(DQ_READ_B, &dquot->dq_flags);
415         /* Instantiate dquot if needed */
416         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
417                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
418                 /* Write the info if needed */
419                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
420                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
421                                         dquot->dq_sb, dquot->dq_id.type);
422                 }
423                 if (ret < 0)
424                         goto out_iolock;
425                 if (ret2 < 0) {
426                         ret = ret2;
427                         goto out_iolock;
428                 }
429         }
430         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
431 out_iolock:
432         mutex_unlock(&dqopt->dqio_mutex);
433         mutex_unlock(&dquot->dq_lock);
434         return ret;
435 }
436 EXPORT_SYMBOL(dquot_acquire);
437
438 /*
439  *      Write dquot to disk
440  */
441 int dquot_commit(struct dquot *dquot)
442 {
443         int ret = 0;
444         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
445
446         mutex_lock(&dqopt->dqio_mutex);
447         spin_lock(&dq_list_lock);
448         if (!clear_dquot_dirty(dquot)) {
449                 spin_unlock(&dq_list_lock);
450                 goto out_sem;
451         }
452         spin_unlock(&dq_list_lock);
453         /* Inactive dquot can be only if there was error during read/init
454          * => we have better not writing it */
455         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
456                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
457         else
458                 ret = -EIO;
459 out_sem:
460         mutex_unlock(&dqopt->dqio_mutex);
461         return ret;
462 }
463 EXPORT_SYMBOL(dquot_commit);
464
465 /*
466  *      Release dquot
467  */
468 int dquot_release(struct dquot *dquot)
469 {
470         int ret = 0, ret2 = 0;
471         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
472
473         mutex_lock(&dquot->dq_lock);
474         /* Check whether we are not racing with some other dqget() */
475         if (atomic_read(&dquot->dq_count) > 1)
476                 goto out_dqlock;
477         mutex_lock(&dqopt->dqio_mutex);
478         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
479                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
480                 /* Write the info */
481                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
482                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
483                                                 dquot->dq_sb, dquot->dq_id.type);
484                 }
485                 if (ret >= 0)
486                         ret = ret2;
487         }
488         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
489         mutex_unlock(&dqopt->dqio_mutex);
490 out_dqlock:
491         mutex_unlock(&dquot->dq_lock);
492         return ret;
493 }
494 EXPORT_SYMBOL(dquot_release);
495
496 void dquot_destroy(struct dquot *dquot)
497 {
498         kmem_cache_free(dquot_cachep, dquot);
499 }
500 EXPORT_SYMBOL(dquot_destroy);
501
502 static inline void do_destroy_dquot(struct dquot *dquot)
503 {
504         dquot->dq_sb->dq_op->destroy_dquot(dquot);
505 }
506
507 /* Invalidate all dquots on the list. Note that this function is called after
508  * quota is disabled and pointers from inodes removed so there cannot be new
509  * quota users. There can still be some users of quotas due to inodes being
510  * just deleted or pruned by prune_icache() (those are not attached to any
511  * list) or parallel quotactl call. We have to wait for such users.
512  */
513 static void invalidate_dquots(struct super_block *sb, int type)
514 {
515         struct dquot *dquot, *tmp;
516
517 restart:
518         spin_lock(&dq_list_lock);
519         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
520                 if (dquot->dq_sb != sb)
521                         continue;
522                 if (dquot->dq_id.type != type)
523                         continue;
524                 /* Wait for dquot users */
525                 if (atomic_read(&dquot->dq_count)) {
526                         DEFINE_WAIT(wait);
527
528                         dqgrab(dquot);
529                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
530                                         TASK_UNINTERRUPTIBLE);
531                         spin_unlock(&dq_list_lock);
532                         /* Once dqput() wakes us up, we know it's time to free
533                          * the dquot.
534                          * IMPORTANT: we rely on the fact that there is always
535                          * at most one process waiting for dquot to free.
536                          * Otherwise dq_count would be > 1 and we would never
537                          * wake up.
538                          */
539                         if (atomic_read(&dquot->dq_count) > 1)
540                                 schedule();
541                         finish_wait(&dquot->dq_wait_unused, &wait);
542                         dqput(dquot);
543                         /* At this moment dquot() need not exist (it could be
544                          * reclaimed by prune_dqcache(). Hence we must
545                          * restart. */
546                         goto restart;
547                 }
548                 /*
549                  * Quota now has no users and it has been written on last
550                  * dqput()
551                  */
552                 remove_dquot_hash(dquot);
553                 remove_free_dquot(dquot);
554                 remove_inuse(dquot);
555                 do_destroy_dquot(dquot);
556         }
557         spin_unlock(&dq_list_lock);
558 }
559
560 /* Call callback for every active dquot on given filesystem */
561 int dquot_scan_active(struct super_block *sb,
562                       int (*fn)(struct dquot *dquot, unsigned long priv),
563                       unsigned long priv)
564 {
565         struct dquot *dquot, *old_dquot = NULL;
566         int ret = 0;
567
568         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
569         spin_lock(&dq_list_lock);
570         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
571                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
572                         continue;
573                 if (dquot->dq_sb != sb)
574                         continue;
575                 /* Now we have active dquot so we can just increase use count */
576                 atomic_inc(&dquot->dq_count);
577                 spin_unlock(&dq_list_lock);
578                 dqstats_inc(DQST_LOOKUPS);
579                 dqput(old_dquot);
580                 old_dquot = dquot;
581                 /*
582                  * ->release_dquot() can be racing with us. Our reference
583                  * protects us from new calls to it so just wait for any
584                  * outstanding call and recheck the DQ_ACTIVE_B after that.
585                  */
586                 wait_on_dquot(dquot);
587                 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
588                         ret = fn(dquot, priv);
589                         if (ret < 0)
590                                 goto out;
591                 }
592                 spin_lock(&dq_list_lock);
593                 /* We are safe to continue now because our dquot could not
594                  * be moved out of the inuse list while we hold the reference */
595         }
596         spin_unlock(&dq_list_lock);
597 out:
598         dqput(old_dquot);
599         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
600         return ret;
601 }
602 EXPORT_SYMBOL(dquot_scan_active);
603
604 /* Write all dquot structures to quota files */
605 int dquot_writeback_dquots(struct super_block *sb, int type)
606 {
607         struct list_head *dirty;
608         struct dquot *dquot;
609         struct quota_info *dqopt = sb_dqopt(sb);
610         int cnt;
611         int err, ret = 0;
612
613         mutex_lock(&dqopt->dqonoff_mutex);
614         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
615                 if (type != -1 && cnt != type)
616                         continue;
617                 if (!sb_has_quota_active(sb, cnt))
618                         continue;
619                 spin_lock(&dq_list_lock);
620                 dirty = &dqopt->info[cnt].dqi_dirty_list;
621                 while (!list_empty(dirty)) {
622                         dquot = list_first_entry(dirty, struct dquot,
623                                                  dq_dirty);
624                         /* Dirty and inactive can be only bad dquot... */
625                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
626                                 clear_dquot_dirty(dquot);
627                                 continue;
628                         }
629                         /* Now we have active dquot from which someone is
630                          * holding reference so we can safely just increase
631                          * use count */
632                         dqgrab(dquot);
633                         spin_unlock(&dq_list_lock);
634                         dqstats_inc(DQST_LOOKUPS);
635                         err = sb->dq_op->write_dquot(dquot);
636                         if (!ret && err)
637                                 ret = err;
638                         dqput(dquot);
639                         spin_lock(&dq_list_lock);
640                 }
641                 spin_unlock(&dq_list_lock);
642         }
643
644         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
645                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
646                     && info_dirty(&dqopt->info[cnt]))
647                         sb->dq_op->write_info(sb, cnt);
648         dqstats_inc(DQST_SYNCS);
649         mutex_unlock(&dqopt->dqonoff_mutex);
650
651         return ret;
652 }
653 EXPORT_SYMBOL(dquot_writeback_dquots);
654
655 /* Write all dquot structures to disk and make them visible from userspace */
656 int dquot_quota_sync(struct super_block *sb, int type)
657 {
658         struct quota_info *dqopt = sb_dqopt(sb);
659         int cnt;
660         int ret;
661
662         ret = dquot_writeback_dquots(sb, type);
663         if (ret)
664                 return ret;
665         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
666                 return 0;
667
668         /* This is not very clever (and fast) but currently I don't know about
669          * any other simple way of getting quota data to disk and we must get
670          * them there for userspace to be visible... */
671         if (sb->s_op->sync_fs)
672                 sb->s_op->sync_fs(sb, 1);
673         sync_blockdev(sb->s_bdev);
674
675         /*
676          * Now when everything is written we can discard the pagecache so
677          * that userspace sees the changes.
678          */
679         mutex_lock(&dqopt->dqonoff_mutex);
680         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
681                 if (type != -1 && cnt != type)
682                         continue;
683                 if (!sb_has_quota_active(sb, cnt))
684                         continue;
685                 mutex_lock(&dqopt->files[cnt]->i_mutex);
686                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
687                 mutex_unlock(&dqopt->files[cnt]->i_mutex);
688         }
689         mutex_unlock(&dqopt->dqonoff_mutex);
690
691         return 0;
692 }
693 EXPORT_SYMBOL(dquot_quota_sync);
694
695 static unsigned long
696 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
697 {
698         struct list_head *head;
699         struct dquot *dquot;
700         unsigned long freed = 0;
701
702         spin_lock(&dq_list_lock);
703         head = free_dquots.prev;
704         while (head != &free_dquots && sc->nr_to_scan) {
705                 dquot = list_entry(head, struct dquot, dq_free);
706                 remove_dquot_hash(dquot);
707                 remove_free_dquot(dquot);
708                 remove_inuse(dquot);
709                 do_destroy_dquot(dquot);
710                 sc->nr_to_scan--;
711                 freed++;
712                 head = free_dquots.prev;
713         }
714         spin_unlock(&dq_list_lock);
715         return freed;
716 }
717
718 static unsigned long
719 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
720 {
721         return vfs_pressure_ratio(
722         percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
723 }
724
725 static struct shrinker dqcache_shrinker = {
726         .count_objects = dqcache_shrink_count,
727         .scan_objects = dqcache_shrink_scan,
728         .seeks = DEFAULT_SEEKS,
729 };
730
731 /*
732  * Put reference to dquot
733  */
734 void dqput(struct dquot *dquot)
735 {
736         int ret;
737
738         if (!dquot)
739                 return;
740 #ifdef CONFIG_QUOTA_DEBUG
741         if (!atomic_read(&dquot->dq_count)) {
742                 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
743                             quotatypes[dquot->dq_id.type],
744                             from_kqid(&init_user_ns, dquot->dq_id));
745                 BUG();
746         }
747 #endif
748         dqstats_inc(DQST_DROPS);
749 we_slept:
750         spin_lock(&dq_list_lock);
751         if (atomic_read(&dquot->dq_count) > 1) {
752                 /* We have more than one user... nothing to do */
753                 atomic_dec(&dquot->dq_count);
754                 /* Releasing dquot during quotaoff phase? */
755                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
756                     atomic_read(&dquot->dq_count) == 1)
757                         wake_up(&dquot->dq_wait_unused);
758                 spin_unlock(&dq_list_lock);
759                 return;
760         }
761         /* Need to release dquot? */
762         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
763                 spin_unlock(&dq_list_lock);
764                 /* Commit dquot before releasing */
765                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
766                 if (ret < 0) {
767                         quota_error(dquot->dq_sb, "Can't write quota structure"
768                                     " (error %d). Quota may get out of sync!",
769                                     ret);
770                         /*
771                          * We clear dirty bit anyway, so that we avoid
772                          * infinite loop here
773                          */
774                         spin_lock(&dq_list_lock);
775                         clear_dquot_dirty(dquot);
776                         spin_unlock(&dq_list_lock);
777                 }
778                 goto we_slept;
779         }
780         /* Clear flag in case dquot was inactive (something bad happened) */
781         clear_dquot_dirty(dquot);
782         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
783                 spin_unlock(&dq_list_lock);
784                 dquot->dq_sb->dq_op->release_dquot(dquot);
785                 goto we_slept;
786         }
787         atomic_dec(&dquot->dq_count);
788 #ifdef CONFIG_QUOTA_DEBUG
789         /* sanity check */
790         BUG_ON(!list_empty(&dquot->dq_free));
791 #endif
792         put_dquot_last(dquot);
793         spin_unlock(&dq_list_lock);
794 }
795 EXPORT_SYMBOL(dqput);
796
797 struct dquot *dquot_alloc(struct super_block *sb, int type)
798 {
799         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
800 }
801 EXPORT_SYMBOL(dquot_alloc);
802
803 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
804 {
805         struct dquot *dquot;
806
807         dquot = sb->dq_op->alloc_dquot(sb, type);
808         if(!dquot)
809                 return NULL;
810
811         mutex_init(&dquot->dq_lock);
812         INIT_LIST_HEAD(&dquot->dq_free);
813         INIT_LIST_HEAD(&dquot->dq_inuse);
814         INIT_HLIST_NODE(&dquot->dq_hash);
815         INIT_LIST_HEAD(&dquot->dq_dirty);
816         init_waitqueue_head(&dquot->dq_wait_unused);
817         dquot->dq_sb = sb;
818         dquot->dq_id = make_kqid_invalid(type);
819         atomic_set(&dquot->dq_count, 1);
820
821         return dquot;
822 }
823
824 /*
825  * Get reference to dquot
826  *
827  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
828  * destroying our dquot by:
829  *   a) checking for quota flags under dq_list_lock and
830  *   b) getting a reference to dquot before we release dq_list_lock
831  */
832 struct dquot *dqget(struct super_block *sb, struct kqid qid)
833 {
834         unsigned int hashent = hashfn(sb, qid);
835         struct dquot *dquot, *empty = NULL;
836
837         if (!sb_has_quota_active(sb, qid.type))
838                 return ERR_PTR(-ESRCH);
839 we_slept:
840         spin_lock(&dq_list_lock);
841         spin_lock(&dq_state_lock);
842         if (!sb_has_quota_active(sb, qid.type)) {
843                 spin_unlock(&dq_state_lock);
844                 spin_unlock(&dq_list_lock);
845                 dquot = ERR_PTR(-ESRCH);
846                 goto out;
847         }
848         spin_unlock(&dq_state_lock);
849
850         dquot = find_dquot(hashent, sb, qid);
851         if (!dquot) {
852                 if (!empty) {
853                         spin_unlock(&dq_list_lock);
854                         empty = get_empty_dquot(sb, qid.type);
855                         if (!empty)
856                                 schedule();     /* Try to wait for a moment... */
857                         goto we_slept;
858                 }
859                 dquot = empty;
860                 empty = NULL;
861                 dquot->dq_id = qid;
862                 /* all dquots go on the inuse_list */
863                 put_inuse(dquot);
864                 /* hash it first so it can be found */
865                 insert_dquot_hash(dquot);
866                 spin_unlock(&dq_list_lock);
867                 dqstats_inc(DQST_LOOKUPS);
868         } else {
869                 if (!atomic_read(&dquot->dq_count))
870                         remove_free_dquot(dquot);
871                 atomic_inc(&dquot->dq_count);
872                 spin_unlock(&dq_list_lock);
873                 dqstats_inc(DQST_CACHE_HITS);
874                 dqstats_inc(DQST_LOOKUPS);
875         }
876         /* Wait for dq_lock - after this we know that either dquot_release() is
877          * already finished or it will be canceled due to dq_count > 1 test */
878         wait_on_dquot(dquot);
879         /* Read the dquot / allocate space in quota file */
880         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
881                 int err;
882
883                 err = sb->dq_op->acquire_dquot(dquot);
884                 if (err < 0) {
885                         dqput(dquot);
886                         dquot = ERR_PTR(err);
887                         goto out;
888                 }
889         }
890 #ifdef CONFIG_QUOTA_DEBUG
891         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
892 #endif
893 out:
894         if (empty)
895                 do_destroy_dquot(empty);
896
897         return dquot;
898 }
899 EXPORT_SYMBOL(dqget);
900
901 static inline struct dquot **i_dquot(struct inode *inode)
902 {
903         return inode->i_sb->s_op->get_dquots(inode);
904 }
905
906 static int dqinit_needed(struct inode *inode, int type)
907 {
908         struct dquot * const *dquots;
909         int cnt;
910
911         if (IS_NOQUOTA(inode))
912                 return 0;
913
914         dquots = i_dquot(inode);
915         if (type != -1)
916                 return !dquots[type];
917         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
918                 if (!dquots[cnt])
919                         return 1;
920         return 0;
921 }
922
923 /* This routine is guarded by dqonoff_mutex mutex */
924 static void add_dquot_ref(struct super_block *sb, int type)
925 {
926         struct inode *inode, *old_inode = NULL;
927 #ifdef CONFIG_QUOTA_DEBUG
928         int reserved = 0;
929 #endif
930
931         spin_lock(&sb->s_inode_list_lock);
932         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
933                 spin_lock(&inode->i_lock);
934                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
935                     !atomic_read(&inode->i_writecount) ||
936                     !dqinit_needed(inode, type)) {
937                         spin_unlock(&inode->i_lock);
938                         continue;
939                 }
940                 __iget(inode);
941                 spin_unlock(&inode->i_lock);
942                 spin_unlock(&sb->s_inode_list_lock);
943
944 #ifdef CONFIG_QUOTA_DEBUG
945                 if (unlikely(inode_get_rsv_space(inode) > 0))
946                         reserved = 1;
947 #endif
948                 iput(old_inode);
949                 __dquot_initialize(inode, type);
950
951                 /*
952                  * We hold a reference to 'inode' so it couldn't have been
953                  * removed from s_inodes list while we dropped the
954                  * s_inode_list_lock. We cannot iput the inode now as we can be
955                  * holding the last reference and we cannot iput it under
956                  * s_inode_list_lock. So we keep the reference and iput it
957                  * later.
958                  */
959                 old_inode = inode;
960                 spin_lock(&sb->s_inode_list_lock);
961         }
962         spin_unlock(&sb->s_inode_list_lock);
963         iput(old_inode);
964
965 #ifdef CONFIG_QUOTA_DEBUG
966         if (reserved) {
967                 quota_error(sb, "Writes happened before quota was turned on "
968                         "thus quota information is probably inconsistent. "
969                         "Please run quotacheck(8)");
970         }
971 #endif
972 }
973
974 /*
975  * Remove references to dquots from inode and add dquot to list for freeing
976  * if we have the last reference to dquot
977  */
978 static void remove_inode_dquot_ref(struct inode *inode, int type,
979                                    struct list_head *tofree_head)
980 {
981         struct dquot **dquots = i_dquot(inode);
982         struct dquot *dquot = dquots[type];
983
984         if (!dquot)
985                 return;
986
987         dquots[type] = NULL;
988         if (list_empty(&dquot->dq_free)) {
989                 /*
990                  * The inode still has reference to dquot so it can't be in the
991                  * free list
992                  */
993                 spin_lock(&dq_list_lock);
994                 list_add(&dquot->dq_free, tofree_head);
995                 spin_unlock(&dq_list_lock);
996         } else {
997                 /*
998                  * Dquot is already in a list to put so we won't drop the last
999                  * reference here.
1000                  */
1001                 dqput(dquot);
1002         }
1003 }
1004
1005 /*
1006  * Free list of dquots
1007  * Dquots are removed from inodes and no new references can be got so we are
1008  * the only ones holding reference
1009  */
1010 static void put_dquot_list(struct list_head *tofree_head)
1011 {
1012         struct list_head *act_head;
1013         struct dquot *dquot;
1014
1015         act_head = tofree_head->next;
1016         while (act_head != tofree_head) {
1017                 dquot = list_entry(act_head, struct dquot, dq_free);
1018                 act_head = act_head->next;
1019                 /* Remove dquot from the list so we won't have problems... */
1020                 list_del_init(&dquot->dq_free);
1021                 dqput(dquot);
1022         }
1023 }
1024
1025 static void remove_dquot_ref(struct super_block *sb, int type,
1026                 struct list_head *tofree_head)
1027 {
1028         struct inode *inode;
1029         int reserved = 0;
1030
1031         spin_lock(&sb->s_inode_list_lock);
1032         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1033                 /*
1034                  *  We have to scan also I_NEW inodes because they can already
1035                  *  have quota pointer initialized. Luckily, we need to touch
1036                  *  only quota pointers and these have separate locking
1037                  *  (dq_data_lock).
1038                  */
1039                 spin_lock(&dq_data_lock);
1040                 if (!IS_NOQUOTA(inode)) {
1041                         if (unlikely(inode_get_rsv_space(inode) > 0))
1042                                 reserved = 1;
1043                         remove_inode_dquot_ref(inode, type, tofree_head);
1044                 }
1045                 spin_unlock(&dq_data_lock);
1046         }
1047         spin_unlock(&sb->s_inode_list_lock);
1048 #ifdef CONFIG_QUOTA_DEBUG
1049         if (reserved) {
1050                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1051                         " was disabled thus quota information is probably "
1052                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1053         }
1054 #endif
1055 }
1056
1057 /* Gather all references from inodes and drop them */
1058 static void drop_dquot_ref(struct super_block *sb, int type)
1059 {
1060         LIST_HEAD(tofree_head);
1061
1062         if (sb->dq_op) {
1063                 remove_dquot_ref(sb, type, &tofree_head);
1064                 synchronize_srcu(&dquot_srcu);
1065                 put_dquot_list(&tofree_head);
1066         }
1067 }
1068
1069 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1070 {
1071         dquot->dq_dqb.dqb_curinodes += number;
1072 }
1073
1074 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1075 {
1076         dquot->dq_dqb.dqb_curspace += number;
1077 }
1078
1079 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1080 {
1081         dquot->dq_dqb.dqb_rsvspace += number;
1082 }
1083
1084 /*
1085  * Claim reserved quota space
1086  */
1087 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1088 {
1089         if (dquot->dq_dqb.dqb_rsvspace < number) {
1090                 WARN_ON_ONCE(1);
1091                 number = dquot->dq_dqb.dqb_rsvspace;
1092         }
1093         dquot->dq_dqb.dqb_curspace += number;
1094         dquot->dq_dqb.dqb_rsvspace -= number;
1095 }
1096
1097 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1098 {
1099         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1100                 number = dquot->dq_dqb.dqb_curspace;
1101         dquot->dq_dqb.dqb_rsvspace += number;
1102         dquot->dq_dqb.dqb_curspace -= number;
1103 }
1104
1105 static inline
1106 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1107 {
1108         if (dquot->dq_dqb.dqb_rsvspace >= number)
1109                 dquot->dq_dqb.dqb_rsvspace -= number;
1110         else {
1111                 WARN_ON_ONCE(1);
1112                 dquot->dq_dqb.dqb_rsvspace = 0;
1113         }
1114 }
1115
1116 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1117 {
1118         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1119             dquot->dq_dqb.dqb_curinodes >= number)
1120                 dquot->dq_dqb.dqb_curinodes -= number;
1121         else
1122                 dquot->dq_dqb.dqb_curinodes = 0;
1123         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1124                 dquot->dq_dqb.dqb_itime = (time_t) 0;
1125         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1126 }
1127
1128 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1129 {
1130         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1131             dquot->dq_dqb.dqb_curspace >= number)
1132                 dquot->dq_dqb.dqb_curspace -= number;
1133         else
1134                 dquot->dq_dqb.dqb_curspace = 0;
1135         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1136                 dquot->dq_dqb.dqb_btime = (time_t) 0;
1137         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1138 }
1139
1140 struct dquot_warn {
1141         struct super_block *w_sb;
1142         struct kqid w_dq_id;
1143         short w_type;
1144 };
1145
1146 static int warning_issued(struct dquot *dquot, const int warntype)
1147 {
1148         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1149                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1150                 ((warntype == QUOTA_NL_IHARDWARN ||
1151                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1152
1153         if (!flag)
1154                 return 0;
1155         return test_and_set_bit(flag, &dquot->dq_flags);
1156 }
1157
1158 #ifdef CONFIG_PRINT_QUOTA_WARNING
1159 static int flag_print_warnings = 1;
1160
1161 static int need_print_warning(struct dquot_warn *warn)
1162 {
1163         if (!flag_print_warnings)
1164                 return 0;
1165
1166         switch (warn->w_dq_id.type) {
1167                 case USRQUOTA:
1168                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1169                 case GRPQUOTA:
1170                         return in_group_p(warn->w_dq_id.gid);
1171                 case PRJQUOTA:
1172                         return 1;
1173         }
1174         return 0;
1175 }
1176
1177 /* Print warning to user which exceeded quota */
1178 static void print_warning(struct dquot_warn *warn)
1179 {
1180         char *msg = NULL;
1181         struct tty_struct *tty;
1182         int warntype = warn->w_type;
1183
1184         if (warntype == QUOTA_NL_IHARDBELOW ||
1185             warntype == QUOTA_NL_ISOFTBELOW ||
1186             warntype == QUOTA_NL_BHARDBELOW ||
1187             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1188                 return;
1189
1190         tty = get_current_tty();
1191         if (!tty)
1192                 return;
1193         tty_write_message(tty, warn->w_sb->s_id);
1194         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1195                 tty_write_message(tty, ": warning, ");
1196         else
1197                 tty_write_message(tty, ": write failed, ");
1198         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1199         switch (warntype) {
1200                 case QUOTA_NL_IHARDWARN:
1201                         msg = " file limit reached.\r\n";
1202                         break;
1203                 case QUOTA_NL_ISOFTLONGWARN:
1204                         msg = " file quota exceeded too long.\r\n";
1205                         break;
1206                 case QUOTA_NL_ISOFTWARN:
1207                         msg = " file quota exceeded.\r\n";
1208                         break;
1209                 case QUOTA_NL_BHARDWARN:
1210                         msg = " block limit reached.\r\n";
1211                         break;
1212                 case QUOTA_NL_BSOFTLONGWARN:
1213                         msg = " block quota exceeded too long.\r\n";
1214                         break;
1215                 case QUOTA_NL_BSOFTWARN:
1216                         msg = " block quota exceeded.\r\n";
1217                         break;
1218         }
1219         tty_write_message(tty, msg);
1220         tty_kref_put(tty);
1221 }
1222 #endif
1223
1224 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1225                             int warntype)
1226 {
1227         if (warning_issued(dquot, warntype))
1228                 return;
1229         warn->w_type = warntype;
1230         warn->w_sb = dquot->dq_sb;
1231         warn->w_dq_id = dquot->dq_id;
1232 }
1233
1234 /*
1235  * Write warnings to the console and send warning messages over netlink.
1236  *
1237  * Note that this function can call into tty and networking code.
1238  */
1239 static void flush_warnings(struct dquot_warn *warn)
1240 {
1241         int i;
1242
1243         for (i = 0; i < MAXQUOTAS; i++) {
1244                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1245                         continue;
1246 #ifdef CONFIG_PRINT_QUOTA_WARNING
1247                 print_warning(&warn[i]);
1248 #endif
1249                 quota_send_warning(warn[i].w_dq_id,
1250                                    warn[i].w_sb->s_dev, warn[i].w_type);
1251         }
1252 }
1253
1254 static int ignore_hardlimit(struct dquot *dquot)
1255 {
1256         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1257
1258         return capable(CAP_SYS_RESOURCE) &&
1259                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1260                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1261 }
1262
1263 /* needs dq_data_lock */
1264 static int check_idq(struct dquot *dquot, qsize_t inodes,
1265                      struct dquot_warn *warn)
1266 {
1267         qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1268
1269         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1270             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1271                 return 0;
1272
1273         if (dquot->dq_dqb.dqb_ihardlimit &&
1274             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1275             !ignore_hardlimit(dquot)) {
1276                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1277                 return -EDQUOT;
1278         }
1279
1280         if (dquot->dq_dqb.dqb_isoftlimit &&
1281             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1282             dquot->dq_dqb.dqb_itime &&
1283             get_seconds() >= dquot->dq_dqb.dqb_itime &&
1284             !ignore_hardlimit(dquot)) {
1285                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1286                 return -EDQUOT;
1287         }
1288
1289         if (dquot->dq_dqb.dqb_isoftlimit &&
1290             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1291             dquot->dq_dqb.dqb_itime == 0) {
1292                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1293                 dquot->dq_dqb.dqb_itime = get_seconds() +
1294                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1295         }
1296
1297         return 0;
1298 }
1299
1300 /* needs dq_data_lock */
1301 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1302                      struct dquot_warn *warn)
1303 {
1304         qsize_t tspace;
1305         struct super_block *sb = dquot->dq_sb;
1306
1307         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1308             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1309                 return 0;
1310
1311         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1312                 + space;
1313
1314         if (dquot->dq_dqb.dqb_bhardlimit &&
1315             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1316             !ignore_hardlimit(dquot)) {
1317                 if (!prealloc)
1318                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1319                 return -EDQUOT;
1320         }
1321
1322         if (dquot->dq_dqb.dqb_bsoftlimit &&
1323             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1324             dquot->dq_dqb.dqb_btime &&
1325             get_seconds() >= dquot->dq_dqb.dqb_btime &&
1326             !ignore_hardlimit(dquot)) {
1327                 if (!prealloc)
1328                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1329                 return -EDQUOT;
1330         }
1331
1332         if (dquot->dq_dqb.dqb_bsoftlimit &&
1333             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1334             dquot->dq_dqb.dqb_btime == 0) {
1335                 if (!prealloc) {
1336                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1337                         dquot->dq_dqb.dqb_btime = get_seconds() +
1338                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1339                 }
1340                 else
1341                         /*
1342                          * We don't allow preallocation to exceed softlimit so exceeding will
1343                          * be always printed
1344                          */
1345                         return -EDQUOT;
1346         }
1347
1348         return 0;
1349 }
1350
1351 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1352 {
1353         qsize_t newinodes;
1354
1355         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1356             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1357             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1358                 return QUOTA_NL_NOWARN;
1359
1360         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1361         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1362                 return QUOTA_NL_ISOFTBELOW;
1363         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1364             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1365                 return QUOTA_NL_IHARDBELOW;
1366         return QUOTA_NL_NOWARN;
1367 }
1368
1369 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1370 {
1371         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1372             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1373                 return QUOTA_NL_NOWARN;
1374
1375         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1376                 return QUOTA_NL_BSOFTBELOW;
1377         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1378             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1379                 return QUOTA_NL_BHARDBELOW;
1380         return QUOTA_NL_NOWARN;
1381 }
1382
1383 static int dquot_active(const struct inode *inode)
1384 {
1385         struct super_block *sb = inode->i_sb;
1386
1387         if (IS_NOQUOTA(inode))
1388                 return 0;
1389         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1390 }
1391
1392 /*
1393  * Initialize quota pointers in inode
1394  *
1395  * It is better to call this function outside of any transaction as it
1396  * might need a lot of space in journal for dquot structure allocation.
1397  */
1398 static int __dquot_initialize(struct inode *inode, int type)
1399 {
1400         int cnt, init_needed = 0;
1401         struct dquot **dquots, *got[MAXQUOTAS];
1402         struct super_block *sb = inode->i_sb;
1403         qsize_t rsv;
1404         int ret = 0;
1405
1406         if (!dquot_active(inode))
1407                 return 0;
1408
1409         dquots = i_dquot(inode);
1410
1411         /* First get references to structures we might need. */
1412         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1413                 struct kqid qid;
1414                 kprojid_t projid;
1415                 int rc;
1416                 struct dquot *dquot;
1417
1418                 got[cnt] = NULL;
1419                 if (type != -1 && cnt != type)
1420                         continue;
1421                 /*
1422                  * The i_dquot should have been initialized in most cases,
1423                  * we check it without locking here to avoid unnecessary
1424                  * dqget()/dqput() calls.
1425                  */
1426                 if (dquots[cnt])
1427                         continue;
1428
1429                 if (!sb_has_quota_active(sb, cnt))
1430                         continue;
1431
1432                 init_needed = 1;
1433
1434                 switch (cnt) {
1435                 case USRQUOTA:
1436                         qid = make_kqid_uid(inode->i_uid);
1437                         break;
1438                 case GRPQUOTA:
1439                         qid = make_kqid_gid(inode->i_gid);
1440                         break;
1441                 case PRJQUOTA:
1442                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1443                         if (rc)
1444                                 continue;
1445                         qid = make_kqid_projid(projid);
1446                         break;
1447                 }
1448                 dquot = dqget(sb, qid);
1449                 if (IS_ERR(dquot)) {
1450                         /* We raced with somebody turning quotas off... */
1451                         if (PTR_ERR(dquot) != -ESRCH) {
1452                                 ret = PTR_ERR(dquot);
1453                                 goto out_put;
1454                         }
1455                         dquot = NULL;
1456                 }
1457                 got[cnt] = dquot;
1458         }
1459
1460         /* All required i_dquot has been initialized */
1461         if (!init_needed)
1462                 return 0;
1463
1464         spin_lock(&dq_data_lock);
1465         if (IS_NOQUOTA(inode))
1466                 goto out_lock;
1467         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1468                 if (type != -1 && cnt != type)
1469                         continue;
1470                 /* Avoid races with quotaoff() */
1471                 if (!sb_has_quota_active(sb, cnt))
1472                         continue;
1473                 /* We could race with quotaon or dqget() could have failed */
1474                 if (!got[cnt])
1475                         continue;
1476                 if (!dquots[cnt]) {
1477                         dquots[cnt] = got[cnt];
1478                         got[cnt] = NULL;
1479                         /*
1480                          * Make quota reservation system happy if someone
1481                          * did a write before quota was turned on
1482                          */
1483                         rsv = inode_get_rsv_space(inode);
1484                         if (unlikely(rsv))
1485                                 dquot_resv_space(dquots[cnt], rsv);
1486                 }
1487         }
1488 out_lock:
1489         spin_unlock(&dq_data_lock);
1490 out_put:
1491         /* Drop unused references */
1492         dqput_all(got);
1493
1494         return ret;
1495 }
1496
1497 int dquot_initialize(struct inode *inode)
1498 {
1499         return __dquot_initialize(inode, -1);
1500 }
1501 EXPORT_SYMBOL(dquot_initialize);
1502
1503 /*
1504  * Release all quotas referenced by inode.
1505  *
1506  * This function only be called on inode free or converting
1507  * a file to quota file, no other users for the i_dquot in
1508  * both cases, so we needn't call synchronize_srcu() after
1509  * clearing i_dquot.
1510  */
1511 static void __dquot_drop(struct inode *inode)
1512 {
1513         int cnt;
1514         struct dquot **dquots = i_dquot(inode);
1515         struct dquot *put[MAXQUOTAS];
1516
1517         spin_lock(&dq_data_lock);
1518         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1519                 put[cnt] = dquots[cnt];
1520                 dquots[cnt] = NULL;
1521         }
1522         spin_unlock(&dq_data_lock);
1523         dqput_all(put);
1524 }
1525
1526 void dquot_drop(struct inode *inode)
1527 {
1528         struct dquot * const *dquots;
1529         int cnt;
1530
1531         if (IS_NOQUOTA(inode))
1532                 return;
1533
1534         /*
1535          * Test before calling to rule out calls from proc and such
1536          * where we are not allowed to block. Note that this is
1537          * actually reliable test even without the lock - the caller
1538          * must assure that nobody can come after the DQUOT_DROP and
1539          * add quota pointers back anyway.
1540          */
1541         dquots = i_dquot(inode);
1542         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1543                 if (dquots[cnt])
1544                         break;
1545         }
1546
1547         if (cnt < MAXQUOTAS)
1548                 __dquot_drop(inode);
1549 }
1550 EXPORT_SYMBOL(dquot_drop);
1551
1552 /*
1553  * inode_reserved_space is managed internally by quota, and protected by
1554  * i_lock similar to i_blocks+i_bytes.
1555  */
1556 static qsize_t *inode_reserved_space(struct inode * inode)
1557 {
1558         /* Filesystem must explicitly define it's own method in order to use
1559          * quota reservation interface */
1560         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1561         return inode->i_sb->dq_op->get_reserved_space(inode);
1562 }
1563
1564 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1565 {
1566         spin_lock(&inode->i_lock);
1567         *inode_reserved_space(inode) += number;
1568         spin_unlock(&inode->i_lock);
1569 }
1570 EXPORT_SYMBOL(inode_add_rsv_space);
1571
1572 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1573 {
1574         spin_lock(&inode->i_lock);
1575         *inode_reserved_space(inode) -= number;
1576         __inode_add_bytes(inode, number);
1577         spin_unlock(&inode->i_lock);
1578 }
1579 EXPORT_SYMBOL(inode_claim_rsv_space);
1580
1581 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1582 {
1583         spin_lock(&inode->i_lock);
1584         *inode_reserved_space(inode) += number;
1585         __inode_sub_bytes(inode, number);
1586         spin_unlock(&inode->i_lock);
1587 }
1588 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1589
1590 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1591 {
1592         spin_lock(&inode->i_lock);
1593         *inode_reserved_space(inode) -= number;
1594         spin_unlock(&inode->i_lock);
1595 }
1596 EXPORT_SYMBOL(inode_sub_rsv_space);
1597
1598 static qsize_t inode_get_rsv_space(struct inode *inode)
1599 {
1600         qsize_t ret;
1601
1602         if (!inode->i_sb->dq_op->get_reserved_space)
1603                 return 0;
1604         spin_lock(&inode->i_lock);
1605         ret = *inode_reserved_space(inode);
1606         spin_unlock(&inode->i_lock);
1607         return ret;
1608 }
1609
1610 static void inode_incr_space(struct inode *inode, qsize_t number,
1611                                 int reserve)
1612 {
1613         if (reserve)
1614                 inode_add_rsv_space(inode, number);
1615         else
1616                 inode_add_bytes(inode, number);
1617 }
1618
1619 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1620 {
1621         if (reserve)
1622                 inode_sub_rsv_space(inode, number);
1623         else
1624                 inode_sub_bytes(inode, number);
1625 }
1626
1627 /*
1628  * This functions updates i_blocks+i_bytes fields and quota information
1629  * (together with appropriate checks).
1630  *
1631  * NOTE: We absolutely rely on the fact that caller dirties the inode
1632  * (usually helpers in quotaops.h care about this) and holds a handle for
1633  * the current transaction so that dquot write and inode write go into the
1634  * same transaction.
1635  */
1636
1637 /*
1638  * This operation can block, but only after everything is updated
1639  */
1640 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1641 {
1642         int cnt, ret = 0, index;
1643         struct dquot_warn warn[MAXQUOTAS];
1644         int reserve = flags & DQUOT_SPACE_RESERVE;
1645         struct dquot **dquots;
1646
1647         if (!dquot_active(inode)) {
1648                 inode_incr_space(inode, number, reserve);
1649                 goto out;
1650         }
1651
1652         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1653                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1654
1655         dquots = i_dquot(inode);
1656         index = srcu_read_lock(&dquot_srcu);
1657         spin_lock(&dq_data_lock);
1658         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1659                 if (!dquots[cnt])
1660                         continue;
1661                 ret = check_bdq(dquots[cnt], number,
1662                                 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1663                 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1664                         spin_unlock(&dq_data_lock);
1665                         goto out_flush_warn;
1666                 }
1667         }
1668         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1669                 if (!dquots[cnt])
1670                         continue;
1671                 if (reserve)
1672                         dquot_resv_space(dquots[cnt], number);
1673                 else
1674                         dquot_incr_space(dquots[cnt], number);
1675         }
1676         inode_incr_space(inode, number, reserve);
1677         spin_unlock(&dq_data_lock);
1678
1679         if (reserve)
1680                 goto out_flush_warn;
1681         mark_all_dquot_dirty(dquots);
1682 out_flush_warn:
1683         srcu_read_unlock(&dquot_srcu, index);
1684         flush_warnings(warn);
1685 out:
1686         return ret;
1687 }
1688 EXPORT_SYMBOL(__dquot_alloc_space);
1689
1690 /*
1691  * This operation can block, but only after everything is updated
1692  */
1693 int dquot_alloc_inode(struct inode *inode)
1694 {
1695         int cnt, ret = 0, index;
1696         struct dquot_warn warn[MAXQUOTAS];
1697         struct dquot * const *dquots;
1698
1699         if (!dquot_active(inode))
1700                 return 0;
1701         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1702                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1703
1704         dquots = i_dquot(inode);
1705         index = srcu_read_lock(&dquot_srcu);
1706         spin_lock(&dq_data_lock);
1707         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1708                 if (!dquots[cnt])
1709                         continue;
1710                 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1711                 if (ret)
1712                         goto warn_put_all;
1713         }
1714
1715         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1716                 if (!dquots[cnt])
1717                         continue;
1718                 dquot_incr_inodes(dquots[cnt], 1);
1719         }
1720
1721 warn_put_all:
1722         spin_unlock(&dq_data_lock);
1723         if (ret == 0)
1724                 mark_all_dquot_dirty(dquots);
1725         srcu_read_unlock(&dquot_srcu, index);
1726         flush_warnings(warn);
1727         return ret;
1728 }
1729 EXPORT_SYMBOL(dquot_alloc_inode);
1730
1731 /*
1732  * Convert in-memory reserved quotas to real consumed quotas
1733  */
1734 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1735 {
1736         struct dquot **dquots;
1737         int cnt, index;
1738
1739         if (!dquot_active(inode)) {
1740                 inode_claim_rsv_space(inode, number);
1741                 return 0;
1742         }
1743
1744         dquots = i_dquot(inode);
1745         index = srcu_read_lock(&dquot_srcu);
1746         spin_lock(&dq_data_lock);
1747         /* Claim reserved quotas to allocated quotas */
1748         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1749                 if (dquots[cnt])
1750                         dquot_claim_reserved_space(dquots[cnt], number);
1751         }
1752         /* Update inode bytes */
1753         inode_claim_rsv_space(inode, number);
1754         spin_unlock(&dq_data_lock);
1755         mark_all_dquot_dirty(dquots);
1756         srcu_read_unlock(&dquot_srcu, index);
1757         return 0;
1758 }
1759 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1760
1761 /*
1762  * Convert allocated space back to in-memory reserved quotas
1763  */
1764 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1765 {
1766         struct dquot **dquots;
1767         int cnt, index;
1768
1769         if (!dquot_active(inode)) {
1770                 inode_reclaim_rsv_space(inode, number);
1771                 return;
1772         }
1773
1774         dquots = i_dquot(inode);
1775         index = srcu_read_lock(&dquot_srcu);
1776         spin_lock(&dq_data_lock);
1777         /* Claim reserved quotas to allocated quotas */
1778         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1779                 if (dquots[cnt])
1780                         dquot_reclaim_reserved_space(dquots[cnt], number);
1781         }
1782         /* Update inode bytes */
1783         inode_reclaim_rsv_space(inode, number);
1784         spin_unlock(&dq_data_lock);
1785         mark_all_dquot_dirty(dquots);
1786         srcu_read_unlock(&dquot_srcu, index);
1787         return;
1788 }
1789 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1790
1791 /*
1792  * This operation can block, but only after everything is updated
1793  */
1794 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1795 {
1796         unsigned int cnt;
1797         struct dquot_warn warn[MAXQUOTAS];
1798         struct dquot **dquots;
1799         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1800
1801         if (!dquot_active(inode)) {
1802                 inode_decr_space(inode, number, reserve);
1803                 return;
1804         }
1805
1806         dquots = i_dquot(inode);
1807         index = srcu_read_lock(&dquot_srcu);
1808         spin_lock(&dq_data_lock);
1809         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1810                 int wtype;
1811
1812                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1813                 if (!dquots[cnt])
1814                         continue;
1815                 wtype = info_bdq_free(dquots[cnt], number);
1816                 if (wtype != QUOTA_NL_NOWARN)
1817                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1818                 if (reserve)
1819                         dquot_free_reserved_space(dquots[cnt], number);
1820                 else
1821                         dquot_decr_space(dquots[cnt], number);
1822         }
1823         inode_decr_space(inode, number, reserve);
1824         spin_unlock(&dq_data_lock);
1825
1826         if (reserve)
1827                 goto out_unlock;
1828         mark_all_dquot_dirty(dquots);
1829 out_unlock:
1830         srcu_read_unlock(&dquot_srcu, index);
1831         flush_warnings(warn);
1832 }
1833 EXPORT_SYMBOL(__dquot_free_space);
1834
1835 /*
1836  * This operation can block, but only after everything is updated
1837  */
1838 void dquot_free_inode(struct inode *inode)
1839 {
1840         unsigned int cnt;
1841         struct dquot_warn warn[MAXQUOTAS];
1842         struct dquot * const *dquots;
1843         int index;
1844
1845         if (!dquot_active(inode))
1846                 return;
1847
1848         dquots = i_dquot(inode);
1849         index = srcu_read_lock(&dquot_srcu);
1850         spin_lock(&dq_data_lock);
1851         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1852                 int wtype;
1853
1854                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1855                 if (!dquots[cnt])
1856                         continue;
1857                 wtype = info_idq_free(dquots[cnt], 1);
1858                 if (wtype != QUOTA_NL_NOWARN)
1859                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1860                 dquot_decr_inodes(dquots[cnt], 1);
1861         }
1862         spin_unlock(&dq_data_lock);
1863         mark_all_dquot_dirty(dquots);
1864         srcu_read_unlock(&dquot_srcu, index);
1865         flush_warnings(warn);
1866 }
1867 EXPORT_SYMBOL(dquot_free_inode);
1868
1869 /*
1870  * Transfer the number of inode and blocks from one diskquota to an other.
1871  * On success, dquot references in transfer_to are consumed and references
1872  * to original dquots that need to be released are placed there. On failure,
1873  * references are kept untouched.
1874  *
1875  * This operation can block, but only after everything is updated
1876  * A transaction must be started when entering this function.
1877  *
1878  * We are holding reference on transfer_from & transfer_to, no need to
1879  * protect them by srcu_read_lock().
1880  */
1881 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1882 {
1883         qsize_t space, cur_space;
1884         qsize_t rsv_space = 0;
1885         struct dquot *transfer_from[MAXQUOTAS] = {};
1886         int cnt, ret = 0;
1887         char is_valid[MAXQUOTAS] = {};
1888         struct dquot_warn warn_to[MAXQUOTAS];
1889         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1890         struct dquot_warn warn_from_space[MAXQUOTAS];
1891
1892         if (IS_NOQUOTA(inode))
1893                 return 0;
1894         /* Initialize the arrays */
1895         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1896                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1897                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1898                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1899         }
1900
1901         spin_lock(&dq_data_lock);
1902         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1903                 spin_unlock(&dq_data_lock);
1904                 return 0;
1905         }
1906         cur_space = inode_get_bytes(inode);
1907         rsv_space = inode_get_rsv_space(inode);
1908         space = cur_space + rsv_space;
1909         /* Build the transfer_from list and check the limits */
1910         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1911                 /*
1912                  * Skip changes for same uid or gid or for turned off quota-type.
1913                  */
1914                 if (!transfer_to[cnt])
1915                         continue;
1916                 /* Avoid races with quotaoff() */
1917                 if (!sb_has_quota_active(inode->i_sb, cnt))
1918                         continue;
1919                 is_valid[cnt] = 1;
1920                 transfer_from[cnt] = i_dquot(inode)[cnt];
1921                 ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1922                 if (ret)
1923                         goto over_quota;
1924                 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1925                 if (ret)
1926                         goto over_quota;
1927         }
1928
1929         /*
1930          * Finally perform the needed transfer from transfer_from to transfer_to
1931          */
1932         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1933                 if (!is_valid[cnt])
1934                         continue;
1935                 /* Due to IO error we might not have transfer_from[] structure */
1936                 if (transfer_from[cnt]) {
1937                         int wtype;
1938                         wtype = info_idq_free(transfer_from[cnt], 1);
1939                         if (wtype != QUOTA_NL_NOWARN)
1940                                 prepare_warning(&warn_from_inodes[cnt],
1941                                                 transfer_from[cnt], wtype);
1942                         wtype = info_bdq_free(transfer_from[cnt], space);
1943                         if (wtype != QUOTA_NL_NOWARN)
1944                                 prepare_warning(&warn_from_space[cnt],
1945                                                 transfer_from[cnt], wtype);
1946                         dquot_decr_inodes(transfer_from[cnt], 1);
1947                         dquot_decr_space(transfer_from[cnt], cur_space);
1948                         dquot_free_reserved_space(transfer_from[cnt],
1949                                                   rsv_space);
1950                 }
1951
1952                 dquot_incr_inodes(transfer_to[cnt], 1);
1953                 dquot_incr_space(transfer_to[cnt], cur_space);
1954                 dquot_resv_space(transfer_to[cnt], rsv_space);
1955
1956                 i_dquot(inode)[cnt] = transfer_to[cnt];
1957         }
1958         spin_unlock(&dq_data_lock);
1959
1960         mark_all_dquot_dirty(transfer_from);
1961         mark_all_dquot_dirty(transfer_to);
1962         flush_warnings(warn_to);
1963         flush_warnings(warn_from_inodes);
1964         flush_warnings(warn_from_space);
1965         /* Pass back references to put */
1966         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1967                 if (is_valid[cnt])
1968                         transfer_to[cnt] = transfer_from[cnt];
1969         return 0;
1970 over_quota:
1971         spin_unlock(&dq_data_lock);
1972         flush_warnings(warn_to);
1973         return ret;
1974 }
1975 EXPORT_SYMBOL(__dquot_transfer);
1976
1977 /* Wrapper for transferring ownership of an inode for uid/gid only
1978  * Called from FSXXX_setattr()
1979  */
1980 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1981 {
1982         struct dquot *transfer_to[MAXQUOTAS] = {};
1983         struct dquot *dquot;
1984         struct super_block *sb = inode->i_sb;
1985         int ret;
1986
1987         if (!dquot_active(inode))
1988                 return 0;
1989
1990         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
1991                 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
1992                 if (IS_ERR(dquot)) {
1993                         if (PTR_ERR(dquot) != -ESRCH) {
1994                                 ret = PTR_ERR(dquot);
1995                                 goto out_put;
1996                         }
1997                         dquot = NULL;
1998                 }
1999                 transfer_to[USRQUOTA] = dquot;
2000         }
2001         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2002                 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2003                 if (IS_ERR(dquot)) {
2004                         if (PTR_ERR(dquot) != -ESRCH) {
2005                                 ret = PTR_ERR(dquot);
2006                                 goto out_put;
2007                         }
2008                         dquot = NULL;
2009                 }
2010                 transfer_to[GRPQUOTA] = dquot;
2011         }
2012         ret = __dquot_transfer(inode, transfer_to);
2013 out_put:
2014         dqput_all(transfer_to);
2015         return ret;
2016 }
2017 EXPORT_SYMBOL(dquot_transfer);
2018
2019 /*
2020  * Write info of quota file to disk
2021  */
2022 int dquot_commit_info(struct super_block *sb, int type)
2023 {
2024         int ret;
2025         struct quota_info *dqopt = sb_dqopt(sb);
2026
2027         mutex_lock(&dqopt->dqio_mutex);
2028         ret = dqopt->ops[type]->write_file_info(sb, type);
2029         mutex_unlock(&dqopt->dqio_mutex);
2030         return ret;
2031 }
2032 EXPORT_SYMBOL(dquot_commit_info);
2033
2034 /*
2035  * Definitions of diskquota operations.
2036  */
2037 const struct dquot_operations dquot_operations = {
2038         .write_dquot    = dquot_commit,
2039         .acquire_dquot  = dquot_acquire,
2040         .release_dquot  = dquot_release,
2041         .mark_dirty     = dquot_mark_dquot_dirty,
2042         .write_info     = dquot_commit_info,
2043         .alloc_dquot    = dquot_alloc,
2044         .destroy_dquot  = dquot_destroy,
2045 };
2046 EXPORT_SYMBOL(dquot_operations);
2047
2048 /*
2049  * Generic helper for ->open on filesystems supporting disk quotas.
2050  */
2051 int dquot_file_open(struct inode *inode, struct file *file)
2052 {
2053         int error;
2054
2055         error = generic_file_open(inode, file);
2056         if (!error && (file->f_mode & FMODE_WRITE))
2057                 dquot_initialize(inode);
2058         return error;
2059 }
2060 EXPORT_SYMBOL(dquot_file_open);
2061
2062 /*
2063  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2064  */
2065 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2066 {
2067         int cnt, ret = 0;
2068         struct quota_info *dqopt = sb_dqopt(sb);
2069         struct inode *toputinode[MAXQUOTAS];
2070
2071         /* Cannot turn off usage accounting without turning off limits, or
2072          * suspend quotas and simultaneously turn quotas off. */
2073         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2074             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2075             DQUOT_USAGE_ENABLED)))
2076                 return -EINVAL;
2077
2078         /* We need to serialize quota_off() for device */
2079         mutex_lock(&dqopt->dqonoff_mutex);
2080
2081         /*
2082          * Skip everything if there's nothing to do. We have to do this because
2083          * sometimes we are called when fill_super() failed and calling
2084          * sync_fs() in such cases does no good.
2085          */
2086         if (!sb_any_quota_loaded(sb)) {
2087                 mutex_unlock(&dqopt->dqonoff_mutex);
2088                 return 0;
2089         }
2090         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2091                 toputinode[cnt] = NULL;
2092                 if (type != -1 && cnt != type)
2093                         continue;
2094                 if (!sb_has_quota_loaded(sb, cnt))
2095                         continue;
2096
2097                 if (flags & DQUOT_SUSPENDED) {
2098                         spin_lock(&dq_state_lock);
2099                         dqopt->flags |=
2100                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2101                         spin_unlock(&dq_state_lock);
2102                 } else {
2103                         spin_lock(&dq_state_lock);
2104                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2105                         /* Turning off suspended quotas? */
2106                         if (!sb_has_quota_loaded(sb, cnt) &&
2107                             sb_has_quota_suspended(sb, cnt)) {
2108                                 dqopt->flags &= ~dquot_state_flag(
2109                                                         DQUOT_SUSPENDED, cnt);
2110                                 spin_unlock(&dq_state_lock);
2111                                 iput(dqopt->files[cnt]);
2112                                 dqopt->files[cnt] = NULL;
2113                                 continue;
2114                         }
2115                         spin_unlock(&dq_state_lock);
2116                 }
2117
2118                 /* We still have to keep quota loaded? */
2119                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2120                         continue;
2121
2122                 /* Note: these are blocking operations */
2123                 drop_dquot_ref(sb, cnt);
2124                 invalidate_dquots(sb, cnt);
2125                 /*
2126                  * Now all dquots should be invalidated, all writes done so we
2127                  * should be only users of the info. No locks needed.
2128                  */
2129                 if (info_dirty(&dqopt->info[cnt]))
2130                         sb->dq_op->write_info(sb, cnt);
2131                 if (dqopt->ops[cnt]->free_file_info)
2132                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2133                 put_quota_format(dqopt->info[cnt].dqi_format);
2134
2135                 toputinode[cnt] = dqopt->files[cnt];
2136                 if (!sb_has_quota_loaded(sb, cnt))
2137                         dqopt->files[cnt] = NULL;
2138                 dqopt->info[cnt].dqi_flags = 0;
2139                 dqopt->info[cnt].dqi_igrace = 0;
2140                 dqopt->info[cnt].dqi_bgrace = 0;
2141                 dqopt->ops[cnt] = NULL;
2142         }
2143         mutex_unlock(&dqopt->dqonoff_mutex);
2144
2145         /* Skip syncing and setting flags if quota files are hidden */
2146         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2147                 goto put_inodes;
2148
2149         /* Sync the superblock so that buffers with quota data are written to
2150          * disk (and so userspace sees correct data afterwards). */
2151         if (sb->s_op->sync_fs)
2152                 sb->s_op->sync_fs(sb, 1);
2153         sync_blockdev(sb->s_bdev);
2154         /* Now the quota files are just ordinary files and we can set the
2155          * inode flags back. Moreover we discard the pagecache so that
2156          * userspace sees the writes we did bypassing the pagecache. We
2157          * must also discard the blockdev buffers so that we see the
2158          * changes done by userspace on the next quotaon() */
2159         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2160                 if (toputinode[cnt]) {
2161                         mutex_lock(&dqopt->dqonoff_mutex);
2162                         /* If quota was reenabled in the meantime, we have
2163                          * nothing to do */
2164                         if (!sb_has_quota_loaded(sb, cnt)) {
2165                                 mutex_lock(&toputinode[cnt]->i_mutex);
2166                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2167                                   S_NOATIME | S_NOQUOTA);
2168                                 truncate_inode_pages(&toputinode[cnt]->i_data,
2169                                                      0);
2170                                 mutex_unlock(&toputinode[cnt]->i_mutex);
2171                                 mark_inode_dirty_sync(toputinode[cnt]);
2172                         }
2173                         mutex_unlock(&dqopt->dqonoff_mutex);
2174                 }
2175         if (sb->s_bdev)
2176                 invalidate_bdev(sb->s_bdev);
2177 put_inodes:
2178         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2179                 if (toputinode[cnt]) {
2180                         /* On remount RO, we keep the inode pointer so that we
2181                          * can reenable quota on the subsequent remount RW. We
2182                          * have to check 'flags' variable and not use sb_has_
2183                          * function because another quotaon / quotaoff could
2184                          * change global state before we got here. We refuse
2185                          * to suspend quotas when there is pending delete on
2186                          * the quota file... */
2187                         if (!(flags & DQUOT_SUSPENDED))
2188                                 iput(toputinode[cnt]);
2189                         else if (!toputinode[cnt]->i_nlink)
2190                                 ret = -EBUSY;
2191                 }
2192         return ret;
2193 }
2194 EXPORT_SYMBOL(dquot_disable);
2195
2196 int dquot_quota_off(struct super_block *sb, int type)
2197 {
2198         return dquot_disable(sb, type,
2199                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2200 }
2201 EXPORT_SYMBOL(dquot_quota_off);
2202
2203 /*
2204  *      Turn quotas on on a device
2205  */
2206
2207 /*
2208  * Helper function to turn quotas on when we already have the inode of
2209  * quota file and no quota information is loaded.
2210  */
2211 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2212         unsigned int flags)
2213 {
2214         struct quota_format_type *fmt = find_quota_format(format_id);
2215         struct super_block *sb = inode->i_sb;
2216         struct quota_info *dqopt = sb_dqopt(sb);
2217         int error;
2218         int oldflags = -1;
2219
2220         if (!fmt)
2221                 return -ESRCH;
2222         if (!S_ISREG(inode->i_mode)) {
2223                 error = -EACCES;
2224                 goto out_fmt;
2225         }
2226         if (IS_RDONLY(inode)) {
2227                 error = -EROFS;
2228                 goto out_fmt;
2229         }
2230         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2231             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2232                 error = -EINVAL;
2233                 goto out_fmt;
2234         }
2235         /* Usage always has to be set... */
2236         if (!(flags & DQUOT_USAGE_ENABLED)) {
2237                 error = -EINVAL;
2238                 goto out_fmt;
2239         }
2240
2241         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2242                 /* As we bypass the pagecache we must now flush all the
2243                  * dirty data and invalidate caches so that kernel sees
2244                  * changes from userspace. It is not enough to just flush
2245                  * the quota file since if blocksize < pagesize, invalidation
2246                  * of the cache could fail because of other unrelated dirty
2247                  * data */
2248                 sync_filesystem(sb);
2249                 invalidate_bdev(sb->s_bdev);
2250         }
2251         mutex_lock(&dqopt->dqonoff_mutex);
2252         if (sb_has_quota_loaded(sb, type)) {
2253                 error = -EBUSY;
2254                 goto out_lock;
2255         }
2256
2257         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2258                 /* We don't want quota and atime on quota files (deadlocks
2259                  * possible) Also nobody should write to the file - we use
2260                  * special IO operations which ignore the immutable bit. */
2261                 mutex_lock(&inode->i_mutex);
2262                 oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2263                                              S_NOQUOTA);
2264                 inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2265                 mutex_unlock(&inode->i_mutex);
2266                 /*
2267                  * When S_NOQUOTA is set, remove dquot references as no more
2268                  * references can be added
2269                  */
2270                 __dquot_drop(inode);
2271         }
2272
2273         error = -EIO;
2274         dqopt->files[type] = igrab(inode);
2275         if (!dqopt->files[type])
2276                 goto out_lock;
2277         error = -EINVAL;
2278         if (!fmt->qf_ops->check_quota_file(sb, type))
2279                 goto out_file_init;
2280
2281         dqopt->ops[type] = fmt->qf_ops;
2282         dqopt->info[type].dqi_format = fmt;
2283         dqopt->info[type].dqi_fmt_id = format_id;
2284         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2285         mutex_lock(&dqopt->dqio_mutex);
2286         error = dqopt->ops[type]->read_file_info(sb, type);
2287         if (error < 0) {
2288                 mutex_unlock(&dqopt->dqio_mutex);
2289                 goto out_file_init;
2290         }
2291         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2292                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2293         mutex_unlock(&dqopt->dqio_mutex);
2294         spin_lock(&dq_state_lock);
2295         dqopt->flags |= dquot_state_flag(flags, type);
2296         spin_unlock(&dq_state_lock);
2297
2298         add_dquot_ref(sb, type);
2299         mutex_unlock(&dqopt->dqonoff_mutex);
2300
2301         return 0;
2302
2303 out_file_init:
2304         dqopt->files[type] = NULL;
2305         iput(inode);
2306 out_lock:
2307         if (oldflags != -1) {
2308                 mutex_lock(&inode->i_mutex);
2309                 /* Set the flags back (in the case of accidental quotaon()
2310                  * on a wrong file we don't want to mess up the flags) */
2311                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2312                 inode->i_flags |= oldflags;
2313                 mutex_unlock(&inode->i_mutex);
2314         }
2315         mutex_unlock(&dqopt->dqonoff_mutex);
2316 out_fmt:
2317         put_quota_format(fmt);
2318
2319         return error; 
2320 }
2321
2322 /* Reenable quotas on remount RW */
2323 int dquot_resume(struct super_block *sb, int type)
2324 {
2325         struct quota_info *dqopt = sb_dqopt(sb);
2326         struct inode *inode;
2327         int ret = 0, cnt;
2328         unsigned int flags;
2329
2330         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2331                 if (type != -1 && cnt != type)
2332                         continue;
2333
2334                 mutex_lock(&dqopt->dqonoff_mutex);
2335                 if (!sb_has_quota_suspended(sb, cnt)) {
2336                         mutex_unlock(&dqopt->dqonoff_mutex);
2337                         continue;
2338                 }
2339                 inode = dqopt->files[cnt];
2340                 dqopt->files[cnt] = NULL;
2341                 spin_lock(&dq_state_lock);
2342                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2343                                                         DQUOT_LIMITS_ENABLED,
2344                                                         cnt);
2345                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2346                 spin_unlock(&dq_state_lock);
2347                 mutex_unlock(&dqopt->dqonoff_mutex);
2348
2349                 flags = dquot_generic_flag(flags, cnt);
2350                 ret = vfs_load_quota_inode(inode, cnt,
2351                                 dqopt->info[cnt].dqi_fmt_id, flags);
2352                 iput(inode);
2353         }
2354
2355         return ret;
2356 }
2357 EXPORT_SYMBOL(dquot_resume);
2358
2359 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2360                    struct path *path)
2361 {
2362         int error = security_quota_on(path->dentry);
2363         if (error)
2364                 return error;
2365         /* Quota file not on the same filesystem? */
2366         if (path->dentry->d_sb != sb)
2367                 error = -EXDEV;
2368         else
2369                 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2370                                              format_id, DQUOT_USAGE_ENABLED |
2371                                              DQUOT_LIMITS_ENABLED);
2372         return error;
2373 }
2374 EXPORT_SYMBOL(dquot_quota_on);
2375
2376 /*
2377  * More powerful function for turning on quotas allowing setting
2378  * of individual quota flags
2379  */
2380 int dquot_enable(struct inode *inode, int type, int format_id,
2381                  unsigned int flags)
2382 {
2383         int ret = 0;
2384         struct super_block *sb = inode->i_sb;
2385         struct quota_info *dqopt = sb_dqopt(sb);
2386
2387         /* Just unsuspend quotas? */
2388         BUG_ON(flags & DQUOT_SUSPENDED);
2389
2390         if (!flags)
2391                 return 0;
2392         /* Just updating flags needed? */
2393         if (sb_has_quota_loaded(sb, type)) {
2394                 mutex_lock(&dqopt->dqonoff_mutex);
2395                 /* Now do a reliable test... */
2396                 if (!sb_has_quota_loaded(sb, type)) {
2397                         mutex_unlock(&dqopt->dqonoff_mutex);
2398                         goto load_quota;
2399                 }
2400                 if (flags & DQUOT_USAGE_ENABLED &&
2401                     sb_has_quota_usage_enabled(sb, type)) {
2402                         ret = -EBUSY;
2403                         goto out_lock;
2404                 }
2405                 if (flags & DQUOT_LIMITS_ENABLED &&
2406                     sb_has_quota_limits_enabled(sb, type)) {
2407                         ret = -EBUSY;
2408                         goto out_lock;
2409                 }
2410                 spin_lock(&dq_state_lock);
2411                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2412                 spin_unlock(&dq_state_lock);
2413 out_lock:
2414                 mutex_unlock(&dqopt->dqonoff_mutex);
2415                 return ret;
2416         }
2417
2418 load_quota:
2419         return vfs_load_quota_inode(inode, type, format_id, flags);
2420 }
2421 EXPORT_SYMBOL(dquot_enable);
2422
2423 /*
2424  * This function is used when filesystem needs to initialize quotas
2425  * during mount time.
2426  */
2427 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2428                 int format_id, int type)
2429 {
2430         struct dentry *dentry;
2431         int error;
2432
2433         mutex_lock(&d_inode(sb->s_root)->i_mutex);
2434         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2435         mutex_unlock(&d_inode(sb->s_root)->i_mutex);
2436         if (IS_ERR(dentry))
2437                 return PTR_ERR(dentry);
2438
2439         if (d_really_is_negative(dentry)) {
2440                 error = -ENOENT;
2441                 goto out;
2442         }
2443
2444         error = security_quota_on(dentry);
2445         if (!error)
2446                 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2447                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2448
2449 out:
2450         dput(dentry);
2451         return error;
2452 }
2453 EXPORT_SYMBOL(dquot_quota_on_mount);
2454
2455 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2456 {
2457         int ret;
2458         int type;
2459         struct quota_info *dqopt = sb_dqopt(sb);
2460
2461         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2462                 return -ENOSYS;
2463         /* Accounting cannot be turned on while fs is mounted */
2464         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2465         if (!flags)
2466                 return -EINVAL;
2467         for (type = 0; type < MAXQUOTAS; type++) {
2468                 if (!(flags & qtype_enforce_flag(type)))
2469                         continue;
2470                 /* Can't enforce without accounting */
2471                 if (!sb_has_quota_usage_enabled(sb, type))
2472                         return -EINVAL;
2473                 ret = dquot_enable(dqopt->files[type], type,
2474                                    dqopt->info[type].dqi_fmt_id,
2475                                    DQUOT_LIMITS_ENABLED);
2476                 if (ret < 0)
2477                         goto out_err;
2478         }
2479         return 0;
2480 out_err:
2481         /* Backout enforcement enablement we already did */
2482         for (type--; type >= 0; type--)  {
2483                 if (flags & qtype_enforce_flag(type))
2484                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2485         }
2486         /* Error code translation for better compatibility with XFS */
2487         if (ret == -EBUSY)
2488                 ret = -EEXIST;
2489         return ret;
2490 }
2491
2492 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2493 {
2494         int ret;
2495         int type;
2496         struct quota_info *dqopt = sb_dqopt(sb);
2497
2498         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2499                 return -ENOSYS;
2500         /*
2501          * We don't support turning off accounting via quotactl. In principle
2502          * quota infrastructure can do this but filesystems don't expect
2503          * userspace to be able to do it.
2504          */
2505         if (flags &
2506                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2507                 return -EOPNOTSUPP;
2508
2509         /* Filter out limits not enabled */
2510         for (type = 0; type < MAXQUOTAS; type++)
2511                 if (!sb_has_quota_limits_enabled(sb, type))
2512                         flags &= ~qtype_enforce_flag(type);
2513         /* Nothing left? */
2514         if (!flags)
2515                 return -EEXIST;
2516         for (type = 0; type < MAXQUOTAS; type++) {
2517                 if (flags & qtype_enforce_flag(type)) {
2518                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2519                         if (ret < 0)
2520                                 goto out_err;
2521                 }
2522         }
2523         return 0;
2524 out_err:
2525         /* Backout enforcement disabling we already did */
2526         for (type--; type >= 0; type--)  {
2527                 if (flags & qtype_enforce_flag(type))
2528                         dquot_enable(dqopt->files[type], type,
2529                                      dqopt->info[type].dqi_fmt_id,
2530                                      DQUOT_LIMITS_ENABLED);
2531         }
2532         return ret;
2533 }
2534
2535 /* Generic routine for getting common part of quota structure */
2536 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2537 {
2538         struct mem_dqblk *dm = &dquot->dq_dqb;
2539
2540         memset(di, 0, sizeof(*di));
2541         spin_lock(&dq_data_lock);
2542         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2543         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2544         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2545         di->d_ino_softlimit = dm->dqb_isoftlimit;
2546         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2547         di->d_ino_count = dm->dqb_curinodes;
2548         di->d_spc_timer = dm->dqb_btime;
2549         di->d_ino_timer = dm->dqb_itime;
2550         spin_unlock(&dq_data_lock);
2551 }
2552
2553 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2554                     struct qc_dqblk *di)
2555 {
2556         struct dquot *dquot;
2557
2558         dquot = dqget(sb, qid);
2559         if (IS_ERR(dquot))
2560                 return PTR_ERR(dquot);
2561         do_get_dqblk(dquot, di);
2562         dqput(dquot);
2563
2564         return 0;
2565 }
2566 EXPORT_SYMBOL(dquot_get_dqblk);
2567
2568 #define VFS_QC_MASK \
2569         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2570          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2571          QC_SPC_TIMER | QC_INO_TIMER)
2572
2573 /* Generic routine for setting common part of quota structure */
2574 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2575 {
2576         struct mem_dqblk *dm = &dquot->dq_dqb;
2577         int check_blim = 0, check_ilim = 0;
2578         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2579
2580         if (di->d_fieldmask & ~VFS_QC_MASK)
2581                 return -EINVAL;
2582
2583         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2584              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2585             ((di->d_fieldmask & QC_SPC_HARD) &&
2586              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2587             ((di->d_fieldmask & QC_INO_SOFT) &&
2588              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2589             ((di->d_fieldmask & QC_INO_HARD) &&
2590              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2591                 return -ERANGE;
2592
2593         spin_lock(&dq_data_lock);
2594         if (di->d_fieldmask & QC_SPACE) {
2595                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2596                 check_blim = 1;
2597                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2598         }
2599
2600         if (di->d_fieldmask & QC_SPC_SOFT)
2601                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2602         if (di->d_fieldmask & QC_SPC_HARD)
2603                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2604         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2605                 check_blim = 1;
2606                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2607         }
2608
2609         if (di->d_fieldmask & QC_INO_COUNT) {
2610                 dm->dqb_curinodes = di->d_ino_count;
2611                 check_ilim = 1;
2612                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2613         }
2614
2615         if (di->d_fieldmask & QC_INO_SOFT)
2616                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2617         if (di->d_fieldmask & QC_INO_HARD)
2618                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2619         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2620                 check_ilim = 1;
2621                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2622         }
2623
2624         if (di->d_fieldmask & QC_SPC_TIMER) {
2625                 dm->dqb_btime = di->d_spc_timer;
2626                 check_blim = 1;
2627                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2628         }
2629
2630         if (di->d_fieldmask & QC_INO_TIMER) {
2631                 dm->dqb_itime = di->d_ino_timer;
2632                 check_ilim = 1;
2633                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2634         }
2635
2636         if (check_blim) {
2637                 if (!dm->dqb_bsoftlimit ||
2638                     dm->dqb_curspace < dm->dqb_bsoftlimit) {
2639                         dm->dqb_btime = 0;
2640                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2641                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2642                         /* Set grace only if user hasn't provided his own... */
2643                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2644         }
2645         if (check_ilim) {
2646                 if (!dm->dqb_isoftlimit ||
2647                     dm->dqb_curinodes < dm->dqb_isoftlimit) {
2648                         dm->dqb_itime = 0;
2649                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2650                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2651                         /* Set grace only if user hasn't provided his own... */
2652                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2653         }
2654         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2655             dm->dqb_isoftlimit)
2656                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2657         else
2658                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2659         spin_unlock(&dq_data_lock);
2660         mark_dquot_dirty(dquot);
2661
2662         return 0;
2663 }
2664
2665 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2666                   struct qc_dqblk *di)
2667 {
2668         struct dquot *dquot;
2669         int rc;
2670
2671         dquot = dqget(sb, qid);
2672         if (IS_ERR(dquot)) {
2673                 rc = PTR_ERR(dquot);
2674                 goto out;
2675         }
2676         rc = do_set_dqblk(dquot, di);
2677         dqput(dquot);
2678 out:
2679         return rc;
2680 }
2681 EXPORT_SYMBOL(dquot_set_dqblk);
2682
2683 /* Generic routine for getting common part of quota file information */
2684 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2685 {
2686         struct mem_dqinfo *mi;
2687         struct qc_type_state *tstate;
2688         struct quota_info *dqopt = sb_dqopt(sb);
2689         int type;
2690   
2691         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2692         memset(state, 0, sizeof(*state));
2693         for (type = 0; type < MAXQUOTAS; type++) {
2694                 if (!sb_has_quota_active(sb, type))
2695                         continue;
2696                 tstate = state->s_state + type;
2697                 mi = sb_dqopt(sb)->info + type;
2698                 tstate->flags = QCI_ACCT_ENABLED;
2699                 spin_lock(&dq_data_lock);
2700                 if (mi->dqi_flags & DQF_SYS_FILE)
2701                         tstate->flags |= QCI_SYSFILE;
2702                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2703                         tstate->flags |= QCI_ROOT_SQUASH;
2704                 if (sb_has_quota_limits_enabled(sb, type))
2705                         tstate->flags |= QCI_LIMITS_ENFORCED;
2706                 tstate->spc_timelimit = mi->dqi_bgrace;
2707                 tstate->ino_timelimit = mi->dqi_igrace;
2708                 tstate->ino = dqopt->files[type]->i_ino;
2709                 tstate->blocks = dqopt->files[type]->i_blocks;
2710                 tstate->nextents = 1;   /* We don't know... */
2711                 spin_unlock(&dq_data_lock);
2712         }
2713         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2714         return 0;
2715 }
2716 EXPORT_SYMBOL(dquot_get_state);
2717
2718 /* Generic routine for setting common part of quota file information */
2719 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2720 {
2721         struct mem_dqinfo *mi;
2722         int err = 0;
2723
2724         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2725             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2726                 return -EINVAL;
2727         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2728         if (!sb_has_quota_active(sb, type)) {
2729                 err = -ESRCH;
2730                 goto out;
2731         }
2732         mi = sb_dqopt(sb)->info + type;
2733         if (ii->i_fieldmask & QC_FLAGS) {
2734                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2735                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD)) {
2736                         err = -EINVAL;
2737                         goto out;
2738                 }
2739         }
2740         spin_lock(&dq_data_lock);
2741         if (ii->i_fieldmask & QC_SPC_TIMER)
2742                 mi->dqi_bgrace = ii->i_spc_timelimit;
2743         if (ii->i_fieldmask & QC_INO_TIMER)
2744                 mi->dqi_igrace = ii->i_ino_timelimit;
2745         if (ii->i_fieldmask & QC_FLAGS) {
2746                 if (ii->i_flags & QCI_ROOT_SQUASH)
2747                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2748                 else
2749                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2750         }
2751         spin_unlock(&dq_data_lock);
2752         mark_info_dirty(sb, type);
2753         /* Force write to disk */
2754         sb->dq_op->write_info(sb, type);
2755 out:
2756         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2757         return err;
2758 }
2759 EXPORT_SYMBOL(dquot_set_dqinfo);
2760
2761 const struct quotactl_ops dquot_quotactl_ops = {
2762         .quota_on       = dquot_quota_on,
2763         .quota_off      = dquot_quota_off,
2764         .quota_sync     = dquot_quota_sync,
2765         .get_state      = dquot_get_state,
2766         .set_info       = dquot_set_dqinfo,
2767         .get_dqblk      = dquot_get_dqblk,
2768         .set_dqblk      = dquot_set_dqblk
2769 };
2770 EXPORT_SYMBOL(dquot_quotactl_ops);
2771
2772 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2773         .quota_enable   = dquot_quota_enable,
2774         .quota_disable  = dquot_quota_disable,
2775         .quota_sync     = dquot_quota_sync,
2776         .get_state      = dquot_get_state,
2777         .set_info       = dquot_set_dqinfo,
2778         .get_dqblk      = dquot_get_dqblk,
2779         .set_dqblk      = dquot_set_dqblk
2780 };
2781 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2782
2783 static int do_proc_dqstats(struct ctl_table *table, int write,
2784                      void __user *buffer, size_t *lenp, loff_t *ppos)
2785 {
2786         unsigned int type = (int *)table->data - dqstats.stat;
2787
2788         /* Update global table */
2789         dqstats.stat[type] =
2790                         percpu_counter_sum_positive(&dqstats.counter[type]);
2791         return proc_dointvec(table, write, buffer, lenp, ppos);
2792 }
2793
2794 static struct ctl_table fs_dqstats_table[] = {
2795         {
2796                 .procname       = "lookups",
2797                 .data           = &dqstats.stat[DQST_LOOKUPS],
2798                 .maxlen         = sizeof(int),
2799                 .mode           = 0444,
2800                 .proc_handler   = do_proc_dqstats,
2801         },
2802         {
2803                 .procname       = "drops",
2804                 .data           = &dqstats.stat[DQST_DROPS],
2805                 .maxlen         = sizeof(int),
2806                 .mode           = 0444,
2807                 .proc_handler   = do_proc_dqstats,
2808         },
2809         {
2810                 .procname       = "reads",
2811                 .data           = &dqstats.stat[DQST_READS],
2812                 .maxlen         = sizeof(int),
2813                 .mode           = 0444,
2814                 .proc_handler   = do_proc_dqstats,
2815         },
2816         {
2817                 .procname       = "writes",
2818                 .data           = &dqstats.stat[DQST_WRITES],
2819                 .maxlen         = sizeof(int),
2820                 .mode           = 0444,
2821                 .proc_handler   = do_proc_dqstats,
2822         },
2823         {
2824                 .procname       = "cache_hits",
2825                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2826                 .maxlen         = sizeof(int),
2827                 .mode           = 0444,
2828                 .proc_handler   = do_proc_dqstats,
2829         },
2830         {
2831                 .procname       = "allocated_dquots",
2832                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2833                 .maxlen         = sizeof(int),
2834                 .mode           = 0444,
2835                 .proc_handler   = do_proc_dqstats,
2836         },
2837         {
2838                 .procname       = "free_dquots",
2839                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2840                 .maxlen         = sizeof(int),
2841                 .mode           = 0444,
2842                 .proc_handler   = do_proc_dqstats,
2843         },
2844         {
2845                 .procname       = "syncs",
2846                 .data           = &dqstats.stat[DQST_SYNCS],
2847                 .maxlen         = sizeof(int),
2848                 .mode           = 0444,
2849                 .proc_handler   = do_proc_dqstats,
2850         },
2851 #ifdef CONFIG_PRINT_QUOTA_WARNING
2852         {
2853                 .procname       = "warnings",
2854                 .data           = &flag_print_warnings,
2855                 .maxlen         = sizeof(int),
2856                 .mode           = 0644,
2857                 .proc_handler   = proc_dointvec,
2858         },
2859 #endif
2860         { },
2861 };
2862
2863 static struct ctl_table fs_table[] = {
2864         {
2865                 .procname       = "quota",
2866                 .mode           = 0555,
2867                 .child          = fs_dqstats_table,
2868         },
2869         { },
2870 };
2871
2872 static struct ctl_table sys_table[] = {
2873         {
2874                 .procname       = "fs",
2875                 .mode           = 0555,
2876                 .child          = fs_table,
2877         },
2878         { },
2879 };
2880
2881 static int __init dquot_init(void)
2882 {
2883         int i, ret;
2884         unsigned long nr_hash, order;
2885
2886         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2887
2888         register_sysctl_table(sys_table);
2889
2890         dquot_cachep = kmem_cache_create("dquot",
2891                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2892                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2893                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2894                         NULL);
2895
2896         order = 0;
2897         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2898         if (!dquot_hash)
2899                 panic("Cannot create dquot hash table");
2900
2901         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2902                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2903                 if (ret)
2904                         panic("Cannot create dquot stat counters");
2905         }
2906
2907         /* Find power-of-two hlist_heads which can fit into allocation */
2908         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2909         dq_hash_bits = 0;
2910         do {
2911                 dq_hash_bits++;
2912         } while (nr_hash >> dq_hash_bits);
2913         dq_hash_bits--;
2914
2915         nr_hash = 1UL << dq_hash_bits;
2916         dq_hash_mask = nr_hash - 1;
2917         for (i = 0; i < nr_hash; i++)
2918                 INIT_HLIST_HEAD(dquot_hash + i);
2919
2920         pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2921                 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2922
2923         register_shrinker(&dqcache_shrinker);
2924
2925         return 0;
2926 }
2927 module_init(dquot_init);