quota: Move locking into clear_dquot_dirty()
[sfrench/cifs-2.6.git] / fs / quota / dquot.c
1 /*
2  * Implementation of the diskquota system for the LINUX operating system. QUOTA
3  * is implemented using the BSD system call interface as the means of
4  * communication with the user level. This file contains the generic routines
5  * called by the different filesystems on allocation of an inode or block.
6  * These routines take care of the administration needed to have a consistent
7  * diskquota tracking system. The ideas of both user and group quotas are based
8  * on the Melbourne quota system as used on BSD derived systems. The internal
9  * implementation is based on one of the several variants of the LINUX
10  * inode-subsystem with added complexity of the diskquota system.
11  * 
12  * Author:      Marco van Wieringen <mvw@planets.elm.net>
13  *
14  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
15  *
16  *              Revised list management to avoid races
17  *              -- Bill Hawes, <whawes@star.net>, 9/98
18  *
19  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20  *              As the consequence the locking was moved from dquot_decr_...(),
21  *              dquot_incr_...() to calling functions.
22  *              invalidate_dquots() now writes modified dquots.
23  *              Serialized quota_off() and quota_on() for mount point.
24  *              Fixed a few bugs in grow_dquots().
25  *              Fixed deadlock in write_dquot() - we no longer account quotas on
26  *              quota files
27  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
28  *              add_dquot_ref() restarts after blocking
29  *              Added check for bogus uid and fixed check for group in quotactl.
30  *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
31  *
32  *              Used struct list_head instead of own list struct
33  *              Invalidation of referenced dquots is no longer possible
34  *              Improved free_dquots list management
35  *              Quota and i_blocks are now updated in one place to avoid races
36  *              Warnings are now delayed so we won't block in critical section
37  *              Write updated not to require dquot lock
38  *              Jan Kara, <jack@suse.cz>, 9/2000
39  *
40  *              Added dynamic quota structure allocation
41  *              Jan Kara <jack@suse.cz> 12/2000
42  *
43  *              Rewritten quota interface. Implemented new quota format and
44  *              formats registering.
45  *              Jan Kara, <jack@suse.cz>, 2001,2002
46  *
47  *              New SMP locking.
48  *              Jan Kara, <jack@suse.cz>, 10/2002
49  *
50  *              Added journalled quota support, fix lock inversion problems
51  *              Jan Kara, <jack@suse.cz>, 2003,2004
52  *
53  * (C) Copyright 1994 - 1997 Marco van Wieringen 
54  */
55
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
58 #include <linux/fs.h>
59 #include <linux/mount.h>
60 #include <linux/mm.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/cred.h>
76 #include <linux/kmod.h>
77 #include <linux/namei.h>
78 #include <linux/capability.h>
79 #include <linux/quotaops.h>
80 #include "../internal.h" /* ugh */
81
82 #include <linux/uaccess.h>
83
84 /*
85  * There are three quota SMP locks. dq_list_lock protects all lists with quotas
86  * and quota formats.
87  * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
88  * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
89  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
90  * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
91  * modifications of quota state (on quotaon and quotaoff) and readers who care
92  * about latest values take it as well.
93  *
94  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
95  *   dq_list_lock > dq_state_lock
96  *
97  * Note that some things (eg. sb pointer, type, id) doesn't change during
98  * the life of the dquot structure and so needn't to be protected by a lock
99  *
100  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
101  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
102  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
103  * inode and before dropping dquot references to avoid use of dquots after
104  * they are freed. dq_data_lock is used to serialize the pointer setting and
105  * clearing operations.
106  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
107  * inode is a quota file). Functions adding pointers from inode to dquots have
108  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
109  * have to do all pointer modifications before dropping dq_data_lock. This makes
110  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
111  * then drops all pointers to dquots from an inode.
112  *
113  * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
114  * memory (or space for it is being allocated) on the first dqget(), when it is
115  * being written out, and when it is being released on the last dqput(). The
116  * allocation and release operations are serialized by the dq_lock and by
117  * checking the use count in dquot_release().
118  *
119  * Lock ordering (including related VFS locks) is the following:
120  *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
121  */
122
123 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
124 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
125 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
126 EXPORT_SYMBOL(dq_data_lock);
127 DEFINE_STATIC_SRCU(dquot_srcu);
128
129 void __quota_error(struct super_block *sb, const char *func,
130                    const char *fmt, ...)
131 {
132         if (printk_ratelimit()) {
133                 va_list args;
134                 struct va_format vaf;
135
136                 va_start(args, fmt);
137
138                 vaf.fmt = fmt;
139                 vaf.va = &args;
140
141                 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
142                        sb->s_id, func, &vaf);
143
144                 va_end(args);
145         }
146 }
147 EXPORT_SYMBOL(__quota_error);
148
149 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
150 static char *quotatypes[] = INITQFNAMES;
151 #endif
152 static struct quota_format_type *quota_formats; /* List of registered formats */
153 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
154
155 /* SLAB cache for dquot structures */
156 static struct kmem_cache *dquot_cachep;
157
158 int register_quota_format(struct quota_format_type *fmt)
159 {
160         spin_lock(&dq_list_lock);
161         fmt->qf_next = quota_formats;
162         quota_formats = fmt;
163         spin_unlock(&dq_list_lock);
164         return 0;
165 }
166 EXPORT_SYMBOL(register_quota_format);
167
168 void unregister_quota_format(struct quota_format_type *fmt)
169 {
170         struct quota_format_type **actqf;
171
172         spin_lock(&dq_list_lock);
173         for (actqf = &quota_formats; *actqf && *actqf != fmt;
174              actqf = &(*actqf)->qf_next)
175                 ;
176         if (*actqf)
177                 *actqf = (*actqf)->qf_next;
178         spin_unlock(&dq_list_lock);
179 }
180 EXPORT_SYMBOL(unregister_quota_format);
181
182 static struct quota_format_type *find_quota_format(int id)
183 {
184         struct quota_format_type *actqf;
185
186         spin_lock(&dq_list_lock);
187         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
188              actqf = actqf->qf_next)
189                 ;
190         if (!actqf || !try_module_get(actqf->qf_owner)) {
191                 int qm;
192
193                 spin_unlock(&dq_list_lock);
194                 
195                 for (qm = 0; module_names[qm].qm_fmt_id &&
196                              module_names[qm].qm_fmt_id != id; qm++)
197                         ;
198                 if (!module_names[qm].qm_fmt_id ||
199                     request_module(module_names[qm].qm_mod_name))
200                         return NULL;
201
202                 spin_lock(&dq_list_lock);
203                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
204                      actqf = actqf->qf_next)
205                         ;
206                 if (actqf && !try_module_get(actqf->qf_owner))
207                         actqf = NULL;
208         }
209         spin_unlock(&dq_list_lock);
210         return actqf;
211 }
212
213 static void put_quota_format(struct quota_format_type *fmt)
214 {
215         module_put(fmt->qf_owner);
216 }
217
218 /*
219  * Dquot List Management:
220  * The quota code uses three lists for dquot management: the inuse_list,
221  * free_dquots, and dquot_hash[] array. A single dquot structure may be
222  * on all three lists, depending on its current state.
223  *
224  * All dquots are placed to the end of inuse_list when first created, and this
225  * list is used for invalidate operation, which must look at every dquot.
226  *
227  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
228  * and this list is searched whenever we need an available dquot.  Dquots are
229  * removed from the list as soon as they are used again, and
230  * dqstats.free_dquots gives the number of dquots on the list. When
231  * dquot is invalidated it's completely released from memory.
232  *
233  * Dquots with a specific identity (device, type and id) are placed on
234  * one of the dquot_hash[] hash chains. The provides an efficient search
235  * mechanism to locate a specific dquot.
236  */
237
238 static LIST_HEAD(inuse_list);
239 static LIST_HEAD(free_dquots);
240 static unsigned int dq_hash_bits, dq_hash_mask;
241 static struct hlist_head *dquot_hash;
242
243 struct dqstats dqstats;
244 EXPORT_SYMBOL(dqstats);
245
246 static qsize_t inode_get_rsv_space(struct inode *inode);
247 static int __dquot_initialize(struct inode *inode, int type);
248
249 static inline unsigned int
250 hashfn(const struct super_block *sb, struct kqid qid)
251 {
252         unsigned int id = from_kqid(&init_user_ns, qid);
253         int type = qid.type;
254         unsigned long tmp;
255
256         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
257         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
258 }
259
260 /*
261  * Following list functions expect dq_list_lock to be held
262  */
263 static inline void insert_dquot_hash(struct dquot *dquot)
264 {
265         struct hlist_head *head;
266         head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
267         hlist_add_head(&dquot->dq_hash, head);
268 }
269
270 static inline void remove_dquot_hash(struct dquot *dquot)
271 {
272         hlist_del_init(&dquot->dq_hash);
273 }
274
275 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
276                                 struct kqid qid)
277 {
278         struct hlist_node *node;
279         struct dquot *dquot;
280
281         hlist_for_each (node, dquot_hash+hashent) {
282                 dquot = hlist_entry(node, struct dquot, dq_hash);
283                 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
284                         return dquot;
285         }
286         return NULL;
287 }
288
289 /* Add a dquot to the tail of the free list */
290 static inline void put_dquot_last(struct dquot *dquot)
291 {
292         list_add_tail(&dquot->dq_free, &free_dquots);
293         dqstats_inc(DQST_FREE_DQUOTS);
294 }
295
296 static inline void remove_free_dquot(struct dquot *dquot)
297 {
298         if (list_empty(&dquot->dq_free))
299                 return;
300         list_del_init(&dquot->dq_free);
301         dqstats_dec(DQST_FREE_DQUOTS);
302 }
303
304 static inline void put_inuse(struct dquot *dquot)
305 {
306         /* We add to the back of inuse list so we don't have to restart
307          * when traversing this list and we block */
308         list_add_tail(&dquot->dq_inuse, &inuse_list);
309         dqstats_inc(DQST_ALLOC_DQUOTS);
310 }
311
312 static inline void remove_inuse(struct dquot *dquot)
313 {
314         dqstats_dec(DQST_ALLOC_DQUOTS);
315         list_del(&dquot->dq_inuse);
316 }
317 /*
318  * End of list functions needing dq_list_lock
319  */
320
321 static void wait_on_dquot(struct dquot *dquot)
322 {
323         mutex_lock(&dquot->dq_lock);
324         mutex_unlock(&dquot->dq_lock);
325 }
326
327 static inline int dquot_dirty(struct dquot *dquot)
328 {
329         return test_bit(DQ_MOD_B, &dquot->dq_flags);
330 }
331
332 static inline int mark_dquot_dirty(struct dquot *dquot)
333 {
334         return dquot->dq_sb->dq_op->mark_dirty(dquot);
335 }
336
337 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
338 int dquot_mark_dquot_dirty(struct dquot *dquot)
339 {
340         int ret = 1;
341
342         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
343                 return 0;
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 static inline int clear_dquot_dirty(struct dquot *dquot)
385 {
386         spin_lock(&dq_list_lock);
387         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
388                 spin_unlock(&dq_list_lock);
389                 return 0;
390         }
391         list_del_init(&dquot->dq_dirty);
392         spin_unlock(&dq_list_lock);
393         return 1;
394 }
395
396 void mark_info_dirty(struct super_block *sb, int type)
397 {
398         spin_lock(&dq_data_lock);
399         sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
400         spin_unlock(&dq_data_lock);
401 }
402 EXPORT_SYMBOL(mark_info_dirty);
403
404 /*
405  *      Read dquot from disk and alloc space for it
406  */
407
408 int dquot_acquire(struct dquot *dquot)
409 {
410         int ret = 0, ret2 = 0;
411         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
412
413         mutex_lock(&dquot->dq_lock);
414         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
415                 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
416         if (ret < 0)
417                 goto out_iolock;
418         /* Make sure flags update is visible after dquot has been filled */
419         smp_mb__before_atomic();
420         set_bit(DQ_READ_B, &dquot->dq_flags);
421         /* Instantiate dquot if needed */
422         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
423                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
424                 /* Write the info if needed */
425                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
426                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
427                                         dquot->dq_sb, dquot->dq_id.type);
428                 }
429                 if (ret < 0)
430                         goto out_iolock;
431                 if (ret2 < 0) {
432                         ret = ret2;
433                         goto out_iolock;
434                 }
435         }
436         /*
437          * Make sure flags update is visible after on-disk struct has been
438          * allocated. Paired with smp_rmb() in dqget().
439          */
440         smp_mb__before_atomic();
441         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
442 out_iolock:
443         mutex_unlock(&dquot->dq_lock);
444         return ret;
445 }
446 EXPORT_SYMBOL(dquot_acquire);
447
448 /*
449  *      Write dquot to disk
450  */
451 int dquot_commit(struct dquot *dquot)
452 {
453         int ret = 0;
454         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
455
456         mutex_lock(&dquot->dq_lock);
457         if (!clear_dquot_dirty(dquot))
458                 goto out_lock;
459         /* Inactive dquot can be only if there was error during read/init
460          * => we have better not writing it */
461         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
462                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
463         else
464                 ret = -EIO;
465 out_lock:
466         mutex_unlock(&dquot->dq_lock);
467         return ret;
468 }
469 EXPORT_SYMBOL(dquot_commit);
470
471 /*
472  *      Release dquot
473  */
474 int dquot_release(struct dquot *dquot)
475 {
476         int ret = 0, ret2 = 0;
477         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
478
479         mutex_lock(&dquot->dq_lock);
480         /* Check whether we are not racing with some other dqget() */
481         if (atomic_read(&dquot->dq_count) > 1)
482                 goto out_dqlock;
483         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
484                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
485                 /* Write the info */
486                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
487                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
488                                                 dquot->dq_sb, dquot->dq_id.type);
489                 }
490                 if (ret >= 0)
491                         ret = ret2;
492         }
493         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
494 out_dqlock:
495         mutex_unlock(&dquot->dq_lock);
496         return ret;
497 }
498 EXPORT_SYMBOL(dquot_release);
499
500 void dquot_destroy(struct dquot *dquot)
501 {
502         kmem_cache_free(dquot_cachep, dquot);
503 }
504 EXPORT_SYMBOL(dquot_destroy);
505
506 static inline void do_destroy_dquot(struct dquot *dquot)
507 {
508         dquot->dq_sb->dq_op->destroy_dquot(dquot);
509 }
510
511 /* Invalidate all dquots on the list. Note that this function is called after
512  * quota is disabled and pointers from inodes removed so there cannot be new
513  * quota users. There can still be some users of quotas due to inodes being
514  * just deleted or pruned by prune_icache() (those are not attached to any
515  * list) or parallel quotactl call. We have to wait for such users.
516  */
517 static void invalidate_dquots(struct super_block *sb, int type)
518 {
519         struct dquot *dquot, *tmp;
520
521 restart:
522         spin_lock(&dq_list_lock);
523         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
524                 if (dquot->dq_sb != sb)
525                         continue;
526                 if (dquot->dq_id.type != type)
527                         continue;
528                 /* Wait for dquot users */
529                 if (atomic_read(&dquot->dq_count)) {
530                         DEFINE_WAIT(wait);
531
532                         dqgrab(dquot);
533                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
534                                         TASK_UNINTERRUPTIBLE);
535                         spin_unlock(&dq_list_lock);
536                         /* Once dqput() wakes us up, we know it's time to free
537                          * the dquot.
538                          * IMPORTANT: we rely on the fact that there is always
539                          * at most one process waiting for dquot to free.
540                          * Otherwise dq_count would be > 1 and we would never
541                          * wake up.
542                          */
543                         if (atomic_read(&dquot->dq_count) > 1)
544                                 schedule();
545                         finish_wait(&dquot->dq_wait_unused, &wait);
546                         dqput(dquot);
547                         /* At this moment dquot() need not exist (it could be
548                          * reclaimed by prune_dqcache(). Hence we must
549                          * restart. */
550                         goto restart;
551                 }
552                 /*
553                  * Quota now has no users and it has been written on last
554                  * dqput()
555                  */
556                 remove_dquot_hash(dquot);
557                 remove_free_dquot(dquot);
558                 remove_inuse(dquot);
559                 do_destroy_dquot(dquot);
560         }
561         spin_unlock(&dq_list_lock);
562 }
563
564 /* Call callback for every active dquot on given filesystem */
565 int dquot_scan_active(struct super_block *sb,
566                       int (*fn)(struct dquot *dquot, unsigned long priv),
567                       unsigned long priv)
568 {
569         struct dquot *dquot, *old_dquot = NULL;
570         int ret = 0;
571
572         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
573
574         spin_lock(&dq_list_lock);
575         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
576                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
577                         continue;
578                 if (dquot->dq_sb != sb)
579                         continue;
580                 /* Now we have active dquot so we can just increase use count */
581                 atomic_inc(&dquot->dq_count);
582                 spin_unlock(&dq_list_lock);
583                 dqstats_inc(DQST_LOOKUPS);
584                 dqput(old_dquot);
585                 old_dquot = dquot;
586                 /*
587                  * ->release_dquot() can be racing with us. Our reference
588                  * protects us from new calls to it so just wait for any
589                  * outstanding call and recheck the DQ_ACTIVE_B after that.
590                  */
591                 wait_on_dquot(dquot);
592                 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
593                         ret = fn(dquot, priv);
594                         if (ret < 0)
595                                 goto out;
596                 }
597                 spin_lock(&dq_list_lock);
598                 /* We are safe to continue now because our dquot could not
599                  * be moved out of the inuse list while we hold the reference */
600         }
601         spin_unlock(&dq_list_lock);
602 out:
603         dqput(old_dquot);
604         return ret;
605 }
606 EXPORT_SYMBOL(dquot_scan_active);
607
608 /* Write all dquot structures to quota files */
609 int dquot_writeback_dquots(struct super_block *sb, int type)
610 {
611         struct list_head *dirty;
612         struct dquot *dquot;
613         struct quota_info *dqopt = sb_dqopt(sb);
614         int cnt;
615         int err, ret = 0;
616
617         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
618
619         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
620                 if (type != -1 && cnt != type)
621                         continue;
622                 if (!sb_has_quota_active(sb, cnt))
623                         continue;
624                 spin_lock(&dq_list_lock);
625                 dirty = &dqopt->info[cnt].dqi_dirty_list;
626                 while (!list_empty(dirty)) {
627                         dquot = list_first_entry(dirty, struct dquot,
628                                                  dq_dirty);
629
630                         WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
631
632                         /* Now we have active dquot from which someone is
633                          * holding reference so we can safely just increase
634                          * use count */
635                         dqgrab(dquot);
636                         spin_unlock(&dq_list_lock);
637                         dqstats_inc(DQST_LOOKUPS);
638                         err = sb->dq_op->write_dquot(dquot);
639                         if (!ret && err)
640                                 ret = err;
641                         dqput(dquot);
642                         spin_lock(&dq_list_lock);
643                 }
644                 spin_unlock(&dq_list_lock);
645         }
646
647         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
648                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
649                     && info_dirty(&dqopt->info[cnt]))
650                         sb->dq_op->write_info(sb, cnt);
651         dqstats_inc(DQST_SYNCS);
652
653         return ret;
654 }
655 EXPORT_SYMBOL(dquot_writeback_dquots);
656
657 /* Write all dquot structures to disk and make them visible from userspace */
658 int dquot_quota_sync(struct super_block *sb, int type)
659 {
660         struct quota_info *dqopt = sb_dqopt(sb);
661         int cnt;
662         int ret;
663
664         ret = dquot_writeback_dquots(sb, type);
665         if (ret)
666                 return ret;
667         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
668                 return 0;
669
670         /* This is not very clever (and fast) but currently I don't know about
671          * any other simple way of getting quota data to disk and we must get
672          * them there for userspace to be visible... */
673         if (sb->s_op->sync_fs)
674                 sb->s_op->sync_fs(sb, 1);
675         sync_blockdev(sb->s_bdev);
676
677         /*
678          * Now when everything is written we can discard the pagecache so
679          * that userspace sees the changes.
680          */
681         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
682                 if (type != -1 && cnt != type)
683                         continue;
684                 if (!sb_has_quota_active(sb, cnt))
685                         continue;
686                 inode_lock(dqopt->files[cnt]);
687                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
688                 inode_unlock(dqopt->files[cnt]);
689         }
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 (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                         clear_dquot_dirty(dquot);
775                 }
776                 goto we_slept;
777         }
778         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
779                 spin_unlock(&dq_list_lock);
780                 dquot->dq_sb->dq_op->release_dquot(dquot);
781                 goto we_slept;
782         }
783         atomic_dec(&dquot->dq_count);
784 #ifdef CONFIG_QUOTA_DEBUG
785         /* sanity check */
786         BUG_ON(!list_empty(&dquot->dq_free));
787 #endif
788         put_dquot_last(dquot);
789         spin_unlock(&dq_list_lock);
790 }
791 EXPORT_SYMBOL(dqput);
792
793 struct dquot *dquot_alloc(struct super_block *sb, int type)
794 {
795         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
796 }
797 EXPORT_SYMBOL(dquot_alloc);
798
799 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
800 {
801         struct dquot *dquot;
802
803         dquot = sb->dq_op->alloc_dquot(sb, type);
804         if(!dquot)
805                 return NULL;
806
807         mutex_init(&dquot->dq_lock);
808         INIT_LIST_HEAD(&dquot->dq_free);
809         INIT_LIST_HEAD(&dquot->dq_inuse);
810         INIT_HLIST_NODE(&dquot->dq_hash);
811         INIT_LIST_HEAD(&dquot->dq_dirty);
812         init_waitqueue_head(&dquot->dq_wait_unused);
813         dquot->dq_sb = sb;
814         dquot->dq_id = make_kqid_invalid(type);
815         atomic_set(&dquot->dq_count, 1);
816
817         return dquot;
818 }
819
820 /*
821  * Get reference to dquot
822  *
823  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
824  * destroying our dquot by:
825  *   a) checking for quota flags under dq_list_lock and
826  *   b) getting a reference to dquot before we release dq_list_lock
827  */
828 struct dquot *dqget(struct super_block *sb, struct kqid qid)
829 {
830         unsigned int hashent = hashfn(sb, qid);
831         struct dquot *dquot, *empty = NULL;
832
833         if (!qid_has_mapping(sb->s_user_ns, qid))
834                 return ERR_PTR(-EINVAL);
835
836         if (!sb_has_quota_active(sb, qid.type))
837                 return ERR_PTR(-ESRCH);
838 we_slept:
839         spin_lock(&dq_list_lock);
840         spin_lock(&dq_state_lock);
841         if (!sb_has_quota_active(sb, qid.type)) {
842                 spin_unlock(&dq_state_lock);
843                 spin_unlock(&dq_list_lock);
844                 dquot = ERR_PTR(-ESRCH);
845                 goto out;
846         }
847         spin_unlock(&dq_state_lock);
848
849         dquot = find_dquot(hashent, sb, qid);
850         if (!dquot) {
851                 if (!empty) {
852                         spin_unlock(&dq_list_lock);
853                         empty = get_empty_dquot(sb, qid.type);
854                         if (!empty)
855                                 schedule();     /* Try to wait for a moment... */
856                         goto we_slept;
857                 }
858                 dquot = empty;
859                 empty = NULL;
860                 dquot->dq_id = qid;
861                 /* all dquots go on the inuse_list */
862                 put_inuse(dquot);
863                 /* hash it first so it can be found */
864                 insert_dquot_hash(dquot);
865                 spin_unlock(&dq_list_lock);
866                 dqstats_inc(DQST_LOOKUPS);
867         } else {
868                 if (!atomic_read(&dquot->dq_count))
869                         remove_free_dquot(dquot);
870                 atomic_inc(&dquot->dq_count);
871                 spin_unlock(&dq_list_lock);
872                 dqstats_inc(DQST_CACHE_HITS);
873                 dqstats_inc(DQST_LOOKUPS);
874         }
875         /* Wait for dq_lock - after this we know that either dquot_release() is
876          * already finished or it will be canceled due to dq_count > 1 test */
877         wait_on_dquot(dquot);
878         /* Read the dquot / allocate space in quota file */
879         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
880                 int err;
881
882                 err = sb->dq_op->acquire_dquot(dquot);
883                 if (err < 0) {
884                         dqput(dquot);
885                         dquot = ERR_PTR(err);
886                         goto out;
887                 }
888         }
889         /*
890          * Make sure following reads see filled structure - paired with
891          * smp_mb__before_atomic() in dquot_acquire().
892          */
893         smp_rmb();
894 #ifdef CONFIG_QUOTA_DEBUG
895         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
896 #endif
897 out:
898         if (empty)
899                 do_destroy_dquot(empty);
900
901         return dquot;
902 }
903 EXPORT_SYMBOL(dqget);
904
905 static inline struct dquot **i_dquot(struct inode *inode)
906 {
907         return inode->i_sb->s_op->get_dquots(inode);
908 }
909
910 static int dqinit_needed(struct inode *inode, int type)
911 {
912         struct dquot * const *dquots;
913         int cnt;
914
915         if (IS_NOQUOTA(inode))
916                 return 0;
917
918         dquots = i_dquot(inode);
919         if (type != -1)
920                 return !dquots[type];
921         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
922                 if (!dquots[cnt])
923                         return 1;
924         return 0;
925 }
926
927 /* This routine is guarded by s_umount semaphore */
928 static void add_dquot_ref(struct super_block *sb, int type)
929 {
930         struct inode *inode, *old_inode = NULL;
931 #ifdef CONFIG_QUOTA_DEBUG
932         int reserved = 0;
933 #endif
934
935         spin_lock(&sb->s_inode_list_lock);
936         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
937                 spin_lock(&inode->i_lock);
938                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
939                     !atomic_read(&inode->i_writecount) ||
940                     !dqinit_needed(inode, type)) {
941                         spin_unlock(&inode->i_lock);
942                         continue;
943                 }
944                 __iget(inode);
945                 spin_unlock(&inode->i_lock);
946                 spin_unlock(&sb->s_inode_list_lock);
947
948 #ifdef CONFIG_QUOTA_DEBUG
949                 if (unlikely(inode_get_rsv_space(inode) > 0))
950                         reserved = 1;
951 #endif
952                 iput(old_inode);
953                 __dquot_initialize(inode, type);
954
955                 /*
956                  * We hold a reference to 'inode' so it couldn't have been
957                  * removed from s_inodes list while we dropped the
958                  * s_inode_list_lock. We cannot iput the inode now as we can be
959                  * holding the last reference and we cannot iput it under
960                  * s_inode_list_lock. So we keep the reference and iput it
961                  * later.
962                  */
963                 old_inode = inode;
964                 spin_lock(&sb->s_inode_list_lock);
965         }
966         spin_unlock(&sb->s_inode_list_lock);
967         iput(old_inode);
968
969 #ifdef CONFIG_QUOTA_DEBUG
970         if (reserved) {
971                 quota_error(sb, "Writes happened before quota was turned on "
972                         "thus quota information is probably inconsistent. "
973                         "Please run quotacheck(8)");
974         }
975 #endif
976 }
977
978 /*
979  * Remove references to dquots from inode and add dquot to list for freeing
980  * if we have the last reference to dquot
981  */
982 static void remove_inode_dquot_ref(struct inode *inode, int type,
983                                    struct list_head *tofree_head)
984 {
985         struct dquot **dquots = i_dquot(inode);
986         struct dquot *dquot = dquots[type];
987
988         if (!dquot)
989                 return;
990
991         dquots[type] = NULL;
992         if (list_empty(&dquot->dq_free)) {
993                 /*
994                  * The inode still has reference to dquot so it can't be in the
995                  * free list
996                  */
997                 spin_lock(&dq_list_lock);
998                 list_add(&dquot->dq_free, tofree_head);
999                 spin_unlock(&dq_list_lock);
1000         } else {
1001                 /*
1002                  * Dquot is already in a list to put so we won't drop the last
1003                  * reference here.
1004                  */
1005                 dqput(dquot);
1006         }
1007 }
1008
1009 /*
1010  * Free list of dquots
1011  * Dquots are removed from inodes and no new references can be got so we are
1012  * the only ones holding reference
1013  */
1014 static void put_dquot_list(struct list_head *tofree_head)
1015 {
1016         struct list_head *act_head;
1017         struct dquot *dquot;
1018
1019         act_head = tofree_head->next;
1020         while (act_head != tofree_head) {
1021                 dquot = list_entry(act_head, struct dquot, dq_free);
1022                 act_head = act_head->next;
1023                 /* Remove dquot from the list so we won't have problems... */
1024                 list_del_init(&dquot->dq_free);
1025                 dqput(dquot);
1026         }
1027 }
1028
1029 static void remove_dquot_ref(struct super_block *sb, int type,
1030                 struct list_head *tofree_head)
1031 {
1032         struct inode *inode;
1033         int reserved = 0;
1034
1035         spin_lock(&sb->s_inode_list_lock);
1036         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1037                 /*
1038                  *  We have to scan also I_NEW inodes because they can already
1039                  *  have quota pointer initialized. Luckily, we need to touch
1040                  *  only quota pointers and these have separate locking
1041                  *  (dq_data_lock).
1042                  */
1043                 spin_lock(&dq_data_lock);
1044                 if (!IS_NOQUOTA(inode)) {
1045                         if (unlikely(inode_get_rsv_space(inode) > 0))
1046                                 reserved = 1;
1047                         remove_inode_dquot_ref(inode, type, tofree_head);
1048                 }
1049                 spin_unlock(&dq_data_lock);
1050         }
1051         spin_unlock(&sb->s_inode_list_lock);
1052 #ifdef CONFIG_QUOTA_DEBUG
1053         if (reserved) {
1054                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1055                         " was disabled thus quota information is probably "
1056                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1057         }
1058 #endif
1059 }
1060
1061 /* Gather all references from inodes and drop them */
1062 static void drop_dquot_ref(struct super_block *sb, int type)
1063 {
1064         LIST_HEAD(tofree_head);
1065
1066         if (sb->dq_op) {
1067                 remove_dquot_ref(sb, type, &tofree_head);
1068                 synchronize_srcu(&dquot_srcu);
1069                 put_dquot_list(&tofree_head);
1070         }
1071 }
1072
1073 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1074 {
1075         dquot->dq_dqb.dqb_curinodes += number;
1076 }
1077
1078 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1079 {
1080         dquot->dq_dqb.dqb_curspace += number;
1081 }
1082
1083 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1084 {
1085         dquot->dq_dqb.dqb_rsvspace += number;
1086 }
1087
1088 /*
1089  * Claim reserved quota space
1090  */
1091 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1092 {
1093         if (dquot->dq_dqb.dqb_rsvspace < number) {
1094                 WARN_ON_ONCE(1);
1095                 number = dquot->dq_dqb.dqb_rsvspace;
1096         }
1097         dquot->dq_dqb.dqb_curspace += number;
1098         dquot->dq_dqb.dqb_rsvspace -= number;
1099 }
1100
1101 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1102 {
1103         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1104                 number = dquot->dq_dqb.dqb_curspace;
1105         dquot->dq_dqb.dqb_rsvspace += number;
1106         dquot->dq_dqb.dqb_curspace -= number;
1107 }
1108
1109 static inline
1110 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1111 {
1112         if (dquot->dq_dqb.dqb_rsvspace >= number)
1113                 dquot->dq_dqb.dqb_rsvspace -= number;
1114         else {
1115                 WARN_ON_ONCE(1);
1116                 dquot->dq_dqb.dqb_rsvspace = 0;
1117         }
1118 }
1119
1120 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1121 {
1122         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1123             dquot->dq_dqb.dqb_curinodes >= number)
1124                 dquot->dq_dqb.dqb_curinodes -= number;
1125         else
1126                 dquot->dq_dqb.dqb_curinodes = 0;
1127         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1128                 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1129         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1130 }
1131
1132 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1133 {
1134         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1135             dquot->dq_dqb.dqb_curspace >= number)
1136                 dquot->dq_dqb.dqb_curspace -= number;
1137         else
1138                 dquot->dq_dqb.dqb_curspace = 0;
1139         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1140                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1141         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1142 }
1143
1144 struct dquot_warn {
1145         struct super_block *w_sb;
1146         struct kqid w_dq_id;
1147         short w_type;
1148 };
1149
1150 static int warning_issued(struct dquot *dquot, const int warntype)
1151 {
1152         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1153                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1154                 ((warntype == QUOTA_NL_IHARDWARN ||
1155                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1156
1157         if (!flag)
1158                 return 0;
1159         return test_and_set_bit(flag, &dquot->dq_flags);
1160 }
1161
1162 #ifdef CONFIG_PRINT_QUOTA_WARNING
1163 static int flag_print_warnings = 1;
1164
1165 static int need_print_warning(struct dquot_warn *warn)
1166 {
1167         if (!flag_print_warnings)
1168                 return 0;
1169
1170         switch (warn->w_dq_id.type) {
1171                 case USRQUOTA:
1172                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1173                 case GRPQUOTA:
1174                         return in_group_p(warn->w_dq_id.gid);
1175                 case PRJQUOTA:
1176                         return 1;
1177         }
1178         return 0;
1179 }
1180
1181 /* Print warning to user which exceeded quota */
1182 static void print_warning(struct dquot_warn *warn)
1183 {
1184         char *msg = NULL;
1185         struct tty_struct *tty;
1186         int warntype = warn->w_type;
1187
1188         if (warntype == QUOTA_NL_IHARDBELOW ||
1189             warntype == QUOTA_NL_ISOFTBELOW ||
1190             warntype == QUOTA_NL_BHARDBELOW ||
1191             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1192                 return;
1193
1194         tty = get_current_tty();
1195         if (!tty)
1196                 return;
1197         tty_write_message(tty, warn->w_sb->s_id);
1198         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1199                 tty_write_message(tty, ": warning, ");
1200         else
1201                 tty_write_message(tty, ": write failed, ");
1202         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1203         switch (warntype) {
1204                 case QUOTA_NL_IHARDWARN:
1205                         msg = " file limit reached.\r\n";
1206                         break;
1207                 case QUOTA_NL_ISOFTLONGWARN:
1208                         msg = " file quota exceeded too long.\r\n";
1209                         break;
1210                 case QUOTA_NL_ISOFTWARN:
1211                         msg = " file quota exceeded.\r\n";
1212                         break;
1213                 case QUOTA_NL_BHARDWARN:
1214                         msg = " block limit reached.\r\n";
1215                         break;
1216                 case QUOTA_NL_BSOFTLONGWARN:
1217                         msg = " block quota exceeded too long.\r\n";
1218                         break;
1219                 case QUOTA_NL_BSOFTWARN:
1220                         msg = " block quota exceeded.\r\n";
1221                         break;
1222         }
1223         tty_write_message(tty, msg);
1224         tty_kref_put(tty);
1225 }
1226 #endif
1227
1228 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1229                             int warntype)
1230 {
1231         if (warning_issued(dquot, warntype))
1232                 return;
1233         warn->w_type = warntype;
1234         warn->w_sb = dquot->dq_sb;
1235         warn->w_dq_id = dquot->dq_id;
1236 }
1237
1238 /*
1239  * Write warnings to the console and send warning messages over netlink.
1240  *
1241  * Note that this function can call into tty and networking code.
1242  */
1243 static void flush_warnings(struct dquot_warn *warn)
1244 {
1245         int i;
1246
1247         for (i = 0; i < MAXQUOTAS; i++) {
1248                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1249                         continue;
1250 #ifdef CONFIG_PRINT_QUOTA_WARNING
1251                 print_warning(&warn[i]);
1252 #endif
1253                 quota_send_warning(warn[i].w_dq_id,
1254                                    warn[i].w_sb->s_dev, warn[i].w_type);
1255         }
1256 }
1257
1258 static int ignore_hardlimit(struct dquot *dquot)
1259 {
1260         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1261
1262         return capable(CAP_SYS_RESOURCE) &&
1263                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1264                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1265 }
1266
1267 /* needs dq_data_lock */
1268 static int check_idq(struct dquot *dquot, qsize_t inodes,
1269                      struct dquot_warn *warn)
1270 {
1271         qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1272
1273         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1274             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1275                 return 0;
1276
1277         if (dquot->dq_dqb.dqb_ihardlimit &&
1278             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1279             !ignore_hardlimit(dquot)) {
1280                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1281                 return -EDQUOT;
1282         }
1283
1284         if (dquot->dq_dqb.dqb_isoftlimit &&
1285             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1286             dquot->dq_dqb.dqb_itime &&
1287             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1288             !ignore_hardlimit(dquot)) {
1289                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1290                 return -EDQUOT;
1291         }
1292
1293         if (dquot->dq_dqb.dqb_isoftlimit &&
1294             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1295             dquot->dq_dqb.dqb_itime == 0) {
1296                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1297                 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1298                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1299         }
1300
1301         return 0;
1302 }
1303
1304 /* needs dq_data_lock */
1305 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1306                      struct dquot_warn *warn)
1307 {
1308         qsize_t tspace;
1309         struct super_block *sb = dquot->dq_sb;
1310
1311         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1312             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1313                 return 0;
1314
1315         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1316                 + space;
1317
1318         if (dquot->dq_dqb.dqb_bhardlimit &&
1319             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1320             !ignore_hardlimit(dquot)) {
1321                 if (!prealloc)
1322                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1323                 return -EDQUOT;
1324         }
1325
1326         if (dquot->dq_dqb.dqb_bsoftlimit &&
1327             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1328             dquot->dq_dqb.dqb_btime &&
1329             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1330             !ignore_hardlimit(dquot)) {
1331                 if (!prealloc)
1332                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1333                 return -EDQUOT;
1334         }
1335
1336         if (dquot->dq_dqb.dqb_bsoftlimit &&
1337             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1338             dquot->dq_dqb.dqb_btime == 0) {
1339                 if (!prealloc) {
1340                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1341                         dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1342                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1343                 }
1344                 else
1345                         /*
1346                          * We don't allow preallocation to exceed softlimit so exceeding will
1347                          * be always printed
1348                          */
1349                         return -EDQUOT;
1350         }
1351
1352         return 0;
1353 }
1354
1355 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1356 {
1357         qsize_t newinodes;
1358
1359         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1360             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1361             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1362                 return QUOTA_NL_NOWARN;
1363
1364         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1365         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1366                 return QUOTA_NL_ISOFTBELOW;
1367         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1368             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1369                 return QUOTA_NL_IHARDBELOW;
1370         return QUOTA_NL_NOWARN;
1371 }
1372
1373 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1374 {
1375         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1376             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1377                 return QUOTA_NL_NOWARN;
1378
1379         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1380                 return QUOTA_NL_BSOFTBELOW;
1381         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1382             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1383                 return QUOTA_NL_BHARDBELOW;
1384         return QUOTA_NL_NOWARN;
1385 }
1386
1387 static int dquot_active(const struct inode *inode)
1388 {
1389         struct super_block *sb = inode->i_sb;
1390
1391         if (IS_NOQUOTA(inode))
1392                 return 0;
1393         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1394 }
1395
1396 /*
1397  * Initialize quota pointers in inode
1398  *
1399  * It is better to call this function outside of any transaction as it
1400  * might need a lot of space in journal for dquot structure allocation.
1401  */
1402 static int __dquot_initialize(struct inode *inode, int type)
1403 {
1404         int cnt, init_needed = 0;
1405         struct dquot **dquots, *got[MAXQUOTAS] = {};
1406         struct super_block *sb = inode->i_sb;
1407         qsize_t rsv;
1408         int ret = 0;
1409
1410         if (!dquot_active(inode))
1411                 return 0;
1412
1413         dquots = i_dquot(inode);
1414
1415         /* First get references to structures we might need. */
1416         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1417                 struct kqid qid;
1418                 kprojid_t projid;
1419                 int rc;
1420                 struct dquot *dquot;
1421
1422                 if (type != -1 && cnt != type)
1423                         continue;
1424                 /*
1425                  * The i_dquot should have been initialized in most cases,
1426                  * we check it without locking here to avoid unnecessary
1427                  * dqget()/dqput() calls.
1428                  */
1429                 if (dquots[cnt])
1430                         continue;
1431
1432                 if (!sb_has_quota_active(sb, cnt))
1433                         continue;
1434
1435                 init_needed = 1;
1436
1437                 switch (cnt) {
1438                 case USRQUOTA:
1439                         qid = make_kqid_uid(inode->i_uid);
1440                         break;
1441                 case GRPQUOTA:
1442                         qid = make_kqid_gid(inode->i_gid);
1443                         break;
1444                 case PRJQUOTA:
1445                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1446                         if (rc)
1447                                 continue;
1448                         qid = make_kqid_projid(projid);
1449                         break;
1450                 }
1451                 dquot = dqget(sb, qid);
1452                 if (IS_ERR(dquot)) {
1453                         /* We raced with somebody turning quotas off... */
1454                         if (PTR_ERR(dquot) != -ESRCH) {
1455                                 ret = PTR_ERR(dquot);
1456                                 goto out_put;
1457                         }
1458                         dquot = NULL;
1459                 }
1460                 got[cnt] = dquot;
1461         }
1462
1463         /* All required i_dquot has been initialized */
1464         if (!init_needed)
1465                 return 0;
1466
1467         spin_lock(&dq_data_lock);
1468         if (IS_NOQUOTA(inode))
1469                 goto out_lock;
1470         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1471                 if (type != -1 && cnt != type)
1472                         continue;
1473                 /* Avoid races with quotaoff() */
1474                 if (!sb_has_quota_active(sb, cnt))
1475                         continue;
1476                 /* We could race with quotaon or dqget() could have failed */
1477                 if (!got[cnt])
1478                         continue;
1479                 if (!dquots[cnt]) {
1480                         dquots[cnt] = got[cnt];
1481                         got[cnt] = NULL;
1482                         /*
1483                          * Make quota reservation system happy if someone
1484                          * did a write before quota was turned on
1485                          */
1486                         rsv = inode_get_rsv_space(inode);
1487                         if (unlikely(rsv))
1488                                 dquot_resv_space(dquots[cnt], rsv);
1489                 }
1490         }
1491 out_lock:
1492         spin_unlock(&dq_data_lock);
1493 out_put:
1494         /* Drop unused references */
1495         dqput_all(got);
1496
1497         return ret;
1498 }
1499
1500 int dquot_initialize(struct inode *inode)
1501 {
1502         return __dquot_initialize(inode, -1);
1503 }
1504 EXPORT_SYMBOL(dquot_initialize);
1505
1506 bool dquot_initialize_needed(struct inode *inode)
1507 {
1508         struct dquot **dquots;
1509         int i;
1510
1511         if (!dquot_active(inode))
1512                 return false;
1513
1514         dquots = i_dquot(inode);
1515         for (i = 0; i < MAXQUOTAS; i++)
1516                 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1517                         return true;
1518         return false;
1519 }
1520 EXPORT_SYMBOL(dquot_initialize_needed);
1521
1522 /*
1523  * Release all quotas referenced by inode.
1524  *
1525  * This function only be called on inode free or converting
1526  * a file to quota file, no other users for the i_dquot in
1527  * both cases, so we needn't call synchronize_srcu() after
1528  * clearing i_dquot.
1529  */
1530 static void __dquot_drop(struct inode *inode)
1531 {
1532         int cnt;
1533         struct dquot **dquots = i_dquot(inode);
1534         struct dquot *put[MAXQUOTAS];
1535
1536         spin_lock(&dq_data_lock);
1537         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1538                 put[cnt] = dquots[cnt];
1539                 dquots[cnt] = NULL;
1540         }
1541         spin_unlock(&dq_data_lock);
1542         dqput_all(put);
1543 }
1544
1545 void dquot_drop(struct inode *inode)
1546 {
1547         struct dquot * const *dquots;
1548         int cnt;
1549
1550         if (IS_NOQUOTA(inode))
1551                 return;
1552
1553         /*
1554          * Test before calling to rule out calls from proc and such
1555          * where we are not allowed to block. Note that this is
1556          * actually reliable test even without the lock - the caller
1557          * must assure that nobody can come after the DQUOT_DROP and
1558          * add quota pointers back anyway.
1559          */
1560         dquots = i_dquot(inode);
1561         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1562                 if (dquots[cnt])
1563                         break;
1564         }
1565
1566         if (cnt < MAXQUOTAS)
1567                 __dquot_drop(inode);
1568 }
1569 EXPORT_SYMBOL(dquot_drop);
1570
1571 /*
1572  * inode_reserved_space is managed internally by quota, and protected by
1573  * i_lock similar to i_blocks+i_bytes.
1574  */
1575 static qsize_t *inode_reserved_space(struct inode * inode)
1576 {
1577         /* Filesystem must explicitly define it's own method in order to use
1578          * quota reservation interface */
1579         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1580         return inode->i_sb->dq_op->get_reserved_space(inode);
1581 }
1582
1583 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1584 {
1585         spin_lock(&inode->i_lock);
1586         *inode_reserved_space(inode) += number;
1587         spin_unlock(&inode->i_lock);
1588 }
1589 EXPORT_SYMBOL(inode_add_rsv_space);
1590
1591 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1592 {
1593         spin_lock(&inode->i_lock);
1594         *inode_reserved_space(inode) -= number;
1595         __inode_add_bytes(inode, number);
1596         spin_unlock(&inode->i_lock);
1597 }
1598 EXPORT_SYMBOL(inode_claim_rsv_space);
1599
1600 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1601 {
1602         spin_lock(&inode->i_lock);
1603         *inode_reserved_space(inode) += number;
1604         __inode_sub_bytes(inode, number);
1605         spin_unlock(&inode->i_lock);
1606 }
1607 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1608
1609 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1610 {
1611         spin_lock(&inode->i_lock);
1612         *inode_reserved_space(inode) -= number;
1613         spin_unlock(&inode->i_lock);
1614 }
1615 EXPORT_SYMBOL(inode_sub_rsv_space);
1616
1617 static qsize_t inode_get_rsv_space(struct inode *inode)
1618 {
1619         qsize_t ret;
1620
1621         if (!inode->i_sb->dq_op->get_reserved_space)
1622                 return 0;
1623         spin_lock(&inode->i_lock);
1624         ret = *inode_reserved_space(inode);
1625         spin_unlock(&inode->i_lock);
1626         return ret;
1627 }
1628
1629 static void inode_incr_space(struct inode *inode, qsize_t number,
1630                                 int reserve)
1631 {
1632         if (reserve)
1633                 inode_add_rsv_space(inode, number);
1634         else
1635                 inode_add_bytes(inode, number);
1636 }
1637
1638 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1639 {
1640         if (reserve)
1641                 inode_sub_rsv_space(inode, number);
1642         else
1643                 inode_sub_bytes(inode, number);
1644 }
1645
1646 /*
1647  * This functions updates i_blocks+i_bytes fields and quota information
1648  * (together with appropriate checks).
1649  *
1650  * NOTE: We absolutely rely on the fact that caller dirties the inode
1651  * (usually helpers in quotaops.h care about this) and holds a handle for
1652  * the current transaction so that dquot write and inode write go into the
1653  * same transaction.
1654  */
1655
1656 /*
1657  * This operation can block, but only after everything is updated
1658  */
1659 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1660 {
1661         int cnt, ret = 0, index;
1662         struct dquot_warn warn[MAXQUOTAS];
1663         int reserve = flags & DQUOT_SPACE_RESERVE;
1664         struct dquot **dquots;
1665
1666         if (!dquot_active(inode)) {
1667                 inode_incr_space(inode, number, reserve);
1668                 goto out;
1669         }
1670
1671         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1672                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1673
1674         dquots = i_dquot(inode);
1675         index = srcu_read_lock(&dquot_srcu);
1676         spin_lock(&dq_data_lock);
1677         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1678                 if (!dquots[cnt])
1679                         continue;
1680                 ret = check_bdq(dquots[cnt], number,
1681                                 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1682                 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1683                         spin_unlock(&dq_data_lock);
1684                         goto out_flush_warn;
1685                 }
1686         }
1687         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1688                 if (!dquots[cnt])
1689                         continue;
1690                 if (reserve)
1691                         dquot_resv_space(dquots[cnt], number);
1692                 else
1693                         dquot_incr_space(dquots[cnt], number);
1694         }
1695         inode_incr_space(inode, number, reserve);
1696         spin_unlock(&dq_data_lock);
1697
1698         if (reserve)
1699                 goto out_flush_warn;
1700         mark_all_dquot_dirty(dquots);
1701 out_flush_warn:
1702         srcu_read_unlock(&dquot_srcu, index);
1703         flush_warnings(warn);
1704 out:
1705         return ret;
1706 }
1707 EXPORT_SYMBOL(__dquot_alloc_space);
1708
1709 /*
1710  * This operation can block, but only after everything is updated
1711  */
1712 int dquot_alloc_inode(struct inode *inode)
1713 {
1714         int cnt, ret = 0, index;
1715         struct dquot_warn warn[MAXQUOTAS];
1716         struct dquot * const *dquots;
1717
1718         if (!dquot_active(inode))
1719                 return 0;
1720         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1721                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1722
1723         dquots = i_dquot(inode);
1724         index = srcu_read_lock(&dquot_srcu);
1725         spin_lock(&dq_data_lock);
1726         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1727                 if (!dquots[cnt])
1728                         continue;
1729                 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1730                 if (ret)
1731                         goto warn_put_all;
1732         }
1733
1734         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1735                 if (!dquots[cnt])
1736                         continue;
1737                 dquot_incr_inodes(dquots[cnt], 1);
1738         }
1739
1740 warn_put_all:
1741         spin_unlock(&dq_data_lock);
1742         if (ret == 0)
1743                 mark_all_dquot_dirty(dquots);
1744         srcu_read_unlock(&dquot_srcu, index);
1745         flush_warnings(warn);
1746         return ret;
1747 }
1748 EXPORT_SYMBOL(dquot_alloc_inode);
1749
1750 /*
1751  * Convert in-memory reserved quotas to real consumed quotas
1752  */
1753 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1754 {
1755         struct dquot **dquots;
1756         int cnt, index;
1757
1758         if (!dquot_active(inode)) {
1759                 inode_claim_rsv_space(inode, number);
1760                 return 0;
1761         }
1762
1763         dquots = i_dquot(inode);
1764         index = srcu_read_lock(&dquot_srcu);
1765         spin_lock(&dq_data_lock);
1766         /* Claim reserved quotas to allocated quotas */
1767         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1768                 if (dquots[cnt])
1769                         dquot_claim_reserved_space(dquots[cnt], number);
1770         }
1771         /* Update inode bytes */
1772         inode_claim_rsv_space(inode, number);
1773         spin_unlock(&dq_data_lock);
1774         mark_all_dquot_dirty(dquots);
1775         srcu_read_unlock(&dquot_srcu, index);
1776         return 0;
1777 }
1778 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1779
1780 /*
1781  * Convert allocated space back to in-memory reserved quotas
1782  */
1783 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1784 {
1785         struct dquot **dquots;
1786         int cnt, index;
1787
1788         if (!dquot_active(inode)) {
1789                 inode_reclaim_rsv_space(inode, number);
1790                 return;
1791         }
1792
1793         dquots = i_dquot(inode);
1794         index = srcu_read_lock(&dquot_srcu);
1795         spin_lock(&dq_data_lock);
1796         /* Claim reserved quotas to allocated quotas */
1797         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1798                 if (dquots[cnt])
1799                         dquot_reclaim_reserved_space(dquots[cnt], number);
1800         }
1801         /* Update inode bytes */
1802         inode_reclaim_rsv_space(inode, number);
1803         spin_unlock(&dq_data_lock);
1804         mark_all_dquot_dirty(dquots);
1805         srcu_read_unlock(&dquot_srcu, index);
1806         return;
1807 }
1808 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1809
1810 /*
1811  * This operation can block, but only after everything is updated
1812  */
1813 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1814 {
1815         unsigned int cnt;
1816         struct dquot_warn warn[MAXQUOTAS];
1817         struct dquot **dquots;
1818         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1819
1820         if (!dquot_active(inode)) {
1821                 inode_decr_space(inode, number, reserve);
1822                 return;
1823         }
1824
1825         dquots = i_dquot(inode);
1826         index = srcu_read_lock(&dquot_srcu);
1827         spin_lock(&dq_data_lock);
1828         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1829                 int wtype;
1830
1831                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1832                 if (!dquots[cnt])
1833                         continue;
1834                 wtype = info_bdq_free(dquots[cnt], number);
1835                 if (wtype != QUOTA_NL_NOWARN)
1836                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1837                 if (reserve)
1838                         dquot_free_reserved_space(dquots[cnt], number);
1839                 else
1840                         dquot_decr_space(dquots[cnt], number);
1841         }
1842         inode_decr_space(inode, number, reserve);
1843         spin_unlock(&dq_data_lock);
1844
1845         if (reserve)
1846                 goto out_unlock;
1847         mark_all_dquot_dirty(dquots);
1848 out_unlock:
1849         srcu_read_unlock(&dquot_srcu, index);
1850         flush_warnings(warn);
1851 }
1852 EXPORT_SYMBOL(__dquot_free_space);
1853
1854 /*
1855  * This operation can block, but only after everything is updated
1856  */
1857 void dquot_free_inode(struct inode *inode)
1858 {
1859         unsigned int cnt;
1860         struct dquot_warn warn[MAXQUOTAS];
1861         struct dquot * const *dquots;
1862         int index;
1863
1864         if (!dquot_active(inode))
1865                 return;
1866
1867         dquots = i_dquot(inode);
1868         index = srcu_read_lock(&dquot_srcu);
1869         spin_lock(&dq_data_lock);
1870         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1871                 int wtype;
1872
1873                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1874                 if (!dquots[cnt])
1875                         continue;
1876                 wtype = info_idq_free(dquots[cnt], 1);
1877                 if (wtype != QUOTA_NL_NOWARN)
1878                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1879                 dquot_decr_inodes(dquots[cnt], 1);
1880         }
1881         spin_unlock(&dq_data_lock);
1882         mark_all_dquot_dirty(dquots);
1883         srcu_read_unlock(&dquot_srcu, index);
1884         flush_warnings(warn);
1885 }
1886 EXPORT_SYMBOL(dquot_free_inode);
1887
1888 /*
1889  * Transfer the number of inode and blocks from one diskquota to an other.
1890  * On success, dquot references in transfer_to are consumed and references
1891  * to original dquots that need to be released are placed there. On failure,
1892  * references are kept untouched.
1893  *
1894  * This operation can block, but only after everything is updated
1895  * A transaction must be started when entering this function.
1896  *
1897  * We are holding reference on transfer_from & transfer_to, no need to
1898  * protect them by srcu_read_lock().
1899  */
1900 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1901 {
1902         qsize_t space, cur_space;
1903         qsize_t rsv_space = 0;
1904         qsize_t inode_usage = 1;
1905         struct dquot *transfer_from[MAXQUOTAS] = {};
1906         int cnt, ret = 0;
1907         char is_valid[MAXQUOTAS] = {};
1908         struct dquot_warn warn_to[MAXQUOTAS];
1909         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1910         struct dquot_warn warn_from_space[MAXQUOTAS];
1911
1912         if (IS_NOQUOTA(inode))
1913                 return 0;
1914
1915         if (inode->i_sb->dq_op->get_inode_usage) {
1916                 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1917                 if (ret)
1918                         return ret;
1919         }
1920
1921         /* Initialize the arrays */
1922         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1923                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1924                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1925                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1926         }
1927
1928         spin_lock(&dq_data_lock);
1929         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1930                 spin_unlock(&dq_data_lock);
1931                 return 0;
1932         }
1933         cur_space = inode_get_bytes(inode);
1934         rsv_space = inode_get_rsv_space(inode);
1935         space = cur_space + rsv_space;
1936         /* Build the transfer_from list and check the limits */
1937         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1938                 /*
1939                  * Skip changes for same uid or gid or for turned off quota-type.
1940                  */
1941                 if (!transfer_to[cnt])
1942                         continue;
1943                 /* Avoid races with quotaoff() */
1944                 if (!sb_has_quota_active(inode->i_sb, cnt))
1945                         continue;
1946                 is_valid[cnt] = 1;
1947                 transfer_from[cnt] = i_dquot(inode)[cnt];
1948                 ret = check_idq(transfer_to[cnt], inode_usage, &warn_to[cnt]);
1949                 if (ret)
1950                         goto over_quota;
1951                 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1952                 if (ret)
1953                         goto over_quota;
1954         }
1955
1956         /*
1957          * Finally perform the needed transfer from transfer_from to transfer_to
1958          */
1959         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1960                 if (!is_valid[cnt])
1961                         continue;
1962                 /* Due to IO error we might not have transfer_from[] structure */
1963                 if (transfer_from[cnt]) {
1964                         int wtype;
1965                         wtype = info_idq_free(transfer_from[cnt], inode_usage);
1966                         if (wtype != QUOTA_NL_NOWARN)
1967                                 prepare_warning(&warn_from_inodes[cnt],
1968                                                 transfer_from[cnt], wtype);
1969                         wtype = info_bdq_free(transfer_from[cnt], space);
1970                         if (wtype != QUOTA_NL_NOWARN)
1971                                 prepare_warning(&warn_from_space[cnt],
1972                                                 transfer_from[cnt], wtype);
1973                         dquot_decr_inodes(transfer_from[cnt], inode_usage);
1974                         dquot_decr_space(transfer_from[cnt], cur_space);
1975                         dquot_free_reserved_space(transfer_from[cnt],
1976                                                   rsv_space);
1977                 }
1978
1979                 dquot_incr_inodes(transfer_to[cnt], inode_usage);
1980                 dquot_incr_space(transfer_to[cnt], cur_space);
1981                 dquot_resv_space(transfer_to[cnt], rsv_space);
1982
1983                 i_dquot(inode)[cnt] = transfer_to[cnt];
1984         }
1985         spin_unlock(&dq_data_lock);
1986
1987         mark_all_dquot_dirty(transfer_from);
1988         mark_all_dquot_dirty(transfer_to);
1989         flush_warnings(warn_to);
1990         flush_warnings(warn_from_inodes);
1991         flush_warnings(warn_from_space);
1992         /* Pass back references to put */
1993         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1994                 if (is_valid[cnt])
1995                         transfer_to[cnt] = transfer_from[cnt];
1996         return 0;
1997 over_quota:
1998         spin_unlock(&dq_data_lock);
1999         flush_warnings(warn_to);
2000         return ret;
2001 }
2002 EXPORT_SYMBOL(__dquot_transfer);
2003
2004 /* Wrapper for transferring ownership of an inode for uid/gid only
2005  * Called from FSXXX_setattr()
2006  */
2007 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2008 {
2009         struct dquot *transfer_to[MAXQUOTAS] = {};
2010         struct dquot *dquot;
2011         struct super_block *sb = inode->i_sb;
2012         int ret;
2013
2014         if (!dquot_active(inode))
2015                 return 0;
2016
2017         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2018                 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2019                 if (IS_ERR(dquot)) {
2020                         if (PTR_ERR(dquot) != -ESRCH) {
2021                                 ret = PTR_ERR(dquot);
2022                                 goto out_put;
2023                         }
2024                         dquot = NULL;
2025                 }
2026                 transfer_to[USRQUOTA] = dquot;
2027         }
2028         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2029                 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2030                 if (IS_ERR(dquot)) {
2031                         if (PTR_ERR(dquot) != -ESRCH) {
2032                                 ret = PTR_ERR(dquot);
2033                                 goto out_put;
2034                         }
2035                         dquot = NULL;
2036                 }
2037                 transfer_to[GRPQUOTA] = dquot;
2038         }
2039         ret = __dquot_transfer(inode, transfer_to);
2040 out_put:
2041         dqput_all(transfer_to);
2042         return ret;
2043 }
2044 EXPORT_SYMBOL(dquot_transfer);
2045
2046 /*
2047  * Write info of quota file to disk
2048  */
2049 int dquot_commit_info(struct super_block *sb, int type)
2050 {
2051         struct quota_info *dqopt = sb_dqopt(sb);
2052
2053         return dqopt->ops[type]->write_file_info(sb, type);
2054 }
2055 EXPORT_SYMBOL(dquot_commit_info);
2056
2057 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2058 {
2059         struct quota_info *dqopt = sb_dqopt(sb);
2060
2061         if (!sb_has_quota_active(sb, qid->type))
2062                 return -ESRCH;
2063         if (!dqopt->ops[qid->type]->get_next_id)
2064                 return -ENOSYS;
2065         return dqopt->ops[qid->type]->get_next_id(sb, qid);
2066 }
2067 EXPORT_SYMBOL(dquot_get_next_id);
2068
2069 /*
2070  * Definitions of diskquota operations.
2071  */
2072 const struct dquot_operations dquot_operations = {
2073         .write_dquot    = dquot_commit,
2074         .acquire_dquot  = dquot_acquire,
2075         .release_dquot  = dquot_release,
2076         .mark_dirty     = dquot_mark_dquot_dirty,
2077         .write_info     = dquot_commit_info,
2078         .alloc_dquot    = dquot_alloc,
2079         .destroy_dquot  = dquot_destroy,
2080         .get_next_id    = dquot_get_next_id,
2081 };
2082 EXPORT_SYMBOL(dquot_operations);
2083
2084 /*
2085  * Generic helper for ->open on filesystems supporting disk quotas.
2086  */
2087 int dquot_file_open(struct inode *inode, struct file *file)
2088 {
2089         int error;
2090
2091         error = generic_file_open(inode, file);
2092         if (!error && (file->f_mode & FMODE_WRITE))
2093                 dquot_initialize(inode);
2094         return error;
2095 }
2096 EXPORT_SYMBOL(dquot_file_open);
2097
2098 /*
2099  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2100  */
2101 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2102 {
2103         int cnt, ret = 0;
2104         struct quota_info *dqopt = sb_dqopt(sb);
2105         struct inode *toputinode[MAXQUOTAS];
2106
2107         /* s_umount should be held in exclusive mode */
2108         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2109                 up_read(&sb->s_umount);
2110
2111         /* Cannot turn off usage accounting without turning off limits, or
2112          * suspend quotas and simultaneously turn quotas off. */
2113         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2114             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2115             DQUOT_USAGE_ENABLED)))
2116                 return -EINVAL;
2117
2118         /*
2119          * Skip everything if there's nothing to do. We have to do this because
2120          * sometimes we are called when fill_super() failed and calling
2121          * sync_fs() in such cases does no good.
2122          */
2123         if (!sb_any_quota_loaded(sb))
2124                 return 0;
2125
2126         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2127                 toputinode[cnt] = NULL;
2128                 if (type != -1 && cnt != type)
2129                         continue;
2130                 if (!sb_has_quota_loaded(sb, cnt))
2131                         continue;
2132
2133                 if (flags & DQUOT_SUSPENDED) {
2134                         spin_lock(&dq_state_lock);
2135                         dqopt->flags |=
2136                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2137                         spin_unlock(&dq_state_lock);
2138                 } else {
2139                         spin_lock(&dq_state_lock);
2140                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2141                         /* Turning off suspended quotas? */
2142                         if (!sb_has_quota_loaded(sb, cnt) &&
2143                             sb_has_quota_suspended(sb, cnt)) {
2144                                 dqopt->flags &= ~dquot_state_flag(
2145                                                         DQUOT_SUSPENDED, cnt);
2146                                 spin_unlock(&dq_state_lock);
2147                                 iput(dqopt->files[cnt]);
2148                                 dqopt->files[cnt] = NULL;
2149                                 continue;
2150                         }
2151                         spin_unlock(&dq_state_lock);
2152                 }
2153
2154                 /* We still have to keep quota loaded? */
2155                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2156                         continue;
2157
2158                 /* Note: these are blocking operations */
2159                 drop_dquot_ref(sb, cnt);
2160                 invalidate_dquots(sb, cnt);
2161                 /*
2162                  * Now all dquots should be invalidated, all writes done so we
2163                  * should be only users of the info. No locks needed.
2164                  */
2165                 if (info_dirty(&dqopt->info[cnt]))
2166                         sb->dq_op->write_info(sb, cnt);
2167                 if (dqopt->ops[cnt]->free_file_info)
2168                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2169                 put_quota_format(dqopt->info[cnt].dqi_format);
2170
2171                 toputinode[cnt] = dqopt->files[cnt];
2172                 if (!sb_has_quota_loaded(sb, cnt))
2173                         dqopt->files[cnt] = NULL;
2174                 dqopt->info[cnt].dqi_flags = 0;
2175                 dqopt->info[cnt].dqi_igrace = 0;
2176                 dqopt->info[cnt].dqi_bgrace = 0;
2177                 dqopt->ops[cnt] = NULL;
2178         }
2179
2180         /* Skip syncing and setting flags if quota files are hidden */
2181         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2182                 goto put_inodes;
2183
2184         /* Sync the superblock so that buffers with quota data are written to
2185          * disk (and so userspace sees correct data afterwards). */
2186         if (sb->s_op->sync_fs)
2187                 sb->s_op->sync_fs(sb, 1);
2188         sync_blockdev(sb->s_bdev);
2189         /* Now the quota files are just ordinary files and we can set the
2190          * inode flags back. Moreover we discard the pagecache so that
2191          * userspace sees the writes we did bypassing the pagecache. We
2192          * must also discard the blockdev buffers so that we see the
2193          * changes done by userspace on the next quotaon() */
2194         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2195                 /* This can happen when suspending quotas on remount-ro... */
2196                 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2197                         inode_lock(toputinode[cnt]);
2198                         toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2199                         truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2200                         inode_unlock(toputinode[cnt]);
2201                         mark_inode_dirty_sync(toputinode[cnt]);
2202                 }
2203         if (sb->s_bdev)
2204                 invalidate_bdev(sb->s_bdev);
2205 put_inodes:
2206         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2207                 if (toputinode[cnt]) {
2208                         /* On remount RO, we keep the inode pointer so that we
2209                          * can reenable quota on the subsequent remount RW. We
2210                          * have to check 'flags' variable and not use sb_has_
2211                          * function because another quotaon / quotaoff could
2212                          * change global state before we got here. We refuse
2213                          * to suspend quotas when there is pending delete on
2214                          * the quota file... */
2215                         if (!(flags & DQUOT_SUSPENDED))
2216                                 iput(toputinode[cnt]);
2217                         else if (!toputinode[cnt]->i_nlink)
2218                                 ret = -EBUSY;
2219                 }
2220         return ret;
2221 }
2222 EXPORT_SYMBOL(dquot_disable);
2223
2224 int dquot_quota_off(struct super_block *sb, int type)
2225 {
2226         return dquot_disable(sb, type,
2227                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2228 }
2229 EXPORT_SYMBOL(dquot_quota_off);
2230
2231 /*
2232  *      Turn quotas on on a device
2233  */
2234
2235 /*
2236  * Helper function to turn quotas on when we already have the inode of
2237  * quota file and no quota information is loaded.
2238  */
2239 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2240         unsigned int flags)
2241 {
2242         struct quota_format_type *fmt = find_quota_format(format_id);
2243         struct super_block *sb = inode->i_sb;
2244         struct quota_info *dqopt = sb_dqopt(sb);
2245         int error;
2246
2247         if (!fmt)
2248                 return -ESRCH;
2249         if (!S_ISREG(inode->i_mode)) {
2250                 error = -EACCES;
2251                 goto out_fmt;
2252         }
2253         if (IS_RDONLY(inode)) {
2254                 error = -EROFS;
2255                 goto out_fmt;
2256         }
2257         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2258             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2259                 error = -EINVAL;
2260                 goto out_fmt;
2261         }
2262         /* Filesystems outside of init_user_ns not yet supported */
2263         if (sb->s_user_ns != &init_user_ns) {
2264                 error = -EINVAL;
2265                 goto out_fmt;
2266         }
2267         /* Usage always has to be set... */
2268         if (!(flags & DQUOT_USAGE_ENABLED)) {
2269                 error = -EINVAL;
2270                 goto out_fmt;
2271         }
2272         if (sb_has_quota_loaded(sb, type)) {
2273                 error = -EBUSY;
2274                 goto out_fmt;
2275         }
2276
2277         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2278                 /* As we bypass the pagecache we must now flush all the
2279                  * dirty data and invalidate caches so that kernel sees
2280                  * changes from userspace. It is not enough to just flush
2281                  * the quota file since if blocksize < pagesize, invalidation
2282                  * of the cache could fail because of other unrelated dirty
2283                  * data */
2284                 sync_filesystem(sb);
2285                 invalidate_bdev(sb->s_bdev);
2286         }
2287
2288         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2289                 /* We don't want quota and atime on quota files (deadlocks
2290                  * possible) Also nobody should write to the file - we use
2291                  * special IO operations which ignore the immutable bit. */
2292                 inode_lock(inode);
2293                 inode->i_flags |= S_NOQUOTA;
2294                 inode_unlock(inode);
2295                 /*
2296                  * When S_NOQUOTA is set, remove dquot references as no more
2297                  * references can be added
2298                  */
2299                 __dquot_drop(inode);
2300         }
2301
2302         error = -EIO;
2303         dqopt->files[type] = igrab(inode);
2304         if (!dqopt->files[type])
2305                 goto out_file_flags;
2306         error = -EINVAL;
2307         if (!fmt->qf_ops->check_quota_file(sb, type))
2308                 goto out_file_init;
2309
2310         dqopt->ops[type] = fmt->qf_ops;
2311         dqopt->info[type].dqi_format = fmt;
2312         dqopt->info[type].dqi_fmt_id = format_id;
2313         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2314         error = dqopt->ops[type]->read_file_info(sb, type);
2315         if (error < 0)
2316                 goto out_file_init;
2317         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2318                 spin_lock(&dq_data_lock);
2319                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2320                 spin_unlock(&dq_data_lock);
2321         }
2322         spin_lock(&dq_state_lock);
2323         dqopt->flags |= dquot_state_flag(flags, type);
2324         spin_unlock(&dq_state_lock);
2325
2326         add_dquot_ref(sb, type);
2327
2328         return 0;
2329
2330 out_file_init:
2331         dqopt->files[type] = NULL;
2332         iput(inode);
2333 out_file_flags:
2334         inode_lock(inode);
2335         inode->i_flags &= ~S_NOQUOTA;
2336         inode_unlock(inode);
2337 out_fmt:
2338         put_quota_format(fmt);
2339
2340         return error; 
2341 }
2342
2343 /* Reenable quotas on remount RW */
2344 int dquot_resume(struct super_block *sb, int type)
2345 {
2346         struct quota_info *dqopt = sb_dqopt(sb);
2347         struct inode *inode;
2348         int ret = 0, cnt;
2349         unsigned int flags;
2350
2351         /* s_umount should be held in exclusive mode */
2352         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2353                 up_read(&sb->s_umount);
2354
2355         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2356                 if (type != -1 && cnt != type)
2357                         continue;
2358                 if (!sb_has_quota_suspended(sb, cnt))
2359                         continue;
2360
2361                 inode = dqopt->files[cnt];
2362                 dqopt->files[cnt] = NULL;
2363                 spin_lock(&dq_state_lock);
2364                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2365                                                         DQUOT_LIMITS_ENABLED,
2366                                                         cnt);
2367                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2368                 spin_unlock(&dq_state_lock);
2369
2370                 flags = dquot_generic_flag(flags, cnt);
2371                 ret = vfs_load_quota_inode(inode, cnt,
2372                                 dqopt->info[cnt].dqi_fmt_id, flags);
2373                 iput(inode);
2374         }
2375
2376         return ret;
2377 }
2378 EXPORT_SYMBOL(dquot_resume);
2379
2380 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2381                    const struct path *path)
2382 {
2383         int error = security_quota_on(path->dentry);
2384         if (error)
2385                 return error;
2386         /* Quota file not on the same filesystem? */
2387         if (path->dentry->d_sb != sb)
2388                 error = -EXDEV;
2389         else
2390                 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2391                                              format_id, DQUOT_USAGE_ENABLED |
2392                                              DQUOT_LIMITS_ENABLED);
2393         return error;
2394 }
2395 EXPORT_SYMBOL(dquot_quota_on);
2396
2397 /*
2398  * More powerful function for turning on quotas allowing setting
2399  * of individual quota flags
2400  */
2401 int dquot_enable(struct inode *inode, int type, int format_id,
2402                  unsigned int flags)
2403 {
2404         struct super_block *sb = inode->i_sb;
2405
2406         /* Just unsuspend quotas? */
2407         BUG_ON(flags & DQUOT_SUSPENDED);
2408         /* s_umount should be held in exclusive mode */
2409         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2410                 up_read(&sb->s_umount);
2411
2412         if (!flags)
2413                 return 0;
2414         /* Just updating flags needed? */
2415         if (sb_has_quota_loaded(sb, type)) {
2416                 if (flags & DQUOT_USAGE_ENABLED &&
2417                     sb_has_quota_usage_enabled(sb, type))
2418                         return -EBUSY;
2419                 if (flags & DQUOT_LIMITS_ENABLED &&
2420                     sb_has_quota_limits_enabled(sb, type))
2421                         return -EBUSY;
2422                 spin_lock(&dq_state_lock);
2423                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2424                 spin_unlock(&dq_state_lock);
2425                 return 0;
2426         }
2427
2428         return vfs_load_quota_inode(inode, type, format_id, flags);
2429 }
2430 EXPORT_SYMBOL(dquot_enable);
2431
2432 /*
2433  * This function is used when filesystem needs to initialize quotas
2434  * during mount time.
2435  */
2436 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2437                 int format_id, int type)
2438 {
2439         struct dentry *dentry;
2440         int error;
2441
2442         dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2443         if (IS_ERR(dentry))
2444                 return PTR_ERR(dentry);
2445
2446         if (d_really_is_negative(dentry)) {
2447                 error = -ENOENT;
2448                 goto out;
2449         }
2450
2451         error = security_quota_on(dentry);
2452         if (!error)
2453                 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2454                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2455
2456 out:
2457         dput(dentry);
2458         return error;
2459 }
2460 EXPORT_SYMBOL(dquot_quota_on_mount);
2461
2462 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2463 {
2464         int ret;
2465         int type;
2466         struct quota_info *dqopt = sb_dqopt(sb);
2467
2468         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2469                 return -ENOSYS;
2470         /* Accounting cannot be turned on while fs is mounted */
2471         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2472         if (!flags)
2473                 return -EINVAL;
2474         for (type = 0; type < MAXQUOTAS; type++) {
2475                 if (!(flags & qtype_enforce_flag(type)))
2476                         continue;
2477                 /* Can't enforce without accounting */
2478                 if (!sb_has_quota_usage_enabled(sb, type))
2479                         return -EINVAL;
2480                 ret = dquot_enable(dqopt->files[type], type,
2481                                    dqopt->info[type].dqi_fmt_id,
2482                                    DQUOT_LIMITS_ENABLED);
2483                 if (ret < 0)
2484                         goto out_err;
2485         }
2486         return 0;
2487 out_err:
2488         /* Backout enforcement enablement we already did */
2489         for (type--; type >= 0; type--)  {
2490                 if (flags & qtype_enforce_flag(type))
2491                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2492         }
2493         /* Error code translation for better compatibility with XFS */
2494         if (ret == -EBUSY)
2495                 ret = -EEXIST;
2496         return ret;
2497 }
2498
2499 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2500 {
2501         int ret;
2502         int type;
2503         struct quota_info *dqopt = sb_dqopt(sb);
2504
2505         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2506                 return -ENOSYS;
2507         /*
2508          * We don't support turning off accounting via quotactl. In principle
2509          * quota infrastructure can do this but filesystems don't expect
2510          * userspace to be able to do it.
2511          */
2512         if (flags &
2513                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2514                 return -EOPNOTSUPP;
2515
2516         /* Filter out limits not enabled */
2517         for (type = 0; type < MAXQUOTAS; type++)
2518                 if (!sb_has_quota_limits_enabled(sb, type))
2519                         flags &= ~qtype_enforce_flag(type);
2520         /* Nothing left? */
2521         if (!flags)
2522                 return -EEXIST;
2523         for (type = 0; type < MAXQUOTAS; type++) {
2524                 if (flags & qtype_enforce_flag(type)) {
2525                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2526                         if (ret < 0)
2527                                 goto out_err;
2528                 }
2529         }
2530         return 0;
2531 out_err:
2532         /* Backout enforcement disabling we already did */
2533         for (type--; type >= 0; type--)  {
2534                 if (flags & qtype_enforce_flag(type))
2535                         dquot_enable(dqopt->files[type], type,
2536                                      dqopt->info[type].dqi_fmt_id,
2537                                      DQUOT_LIMITS_ENABLED);
2538         }
2539         return ret;
2540 }
2541
2542 /* Generic routine for getting common part of quota structure */
2543 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2544 {
2545         struct mem_dqblk *dm = &dquot->dq_dqb;
2546
2547         memset(di, 0, sizeof(*di));
2548         spin_lock(&dq_data_lock);
2549         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2550         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2551         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2552         di->d_ino_softlimit = dm->dqb_isoftlimit;
2553         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2554         di->d_ino_count = dm->dqb_curinodes;
2555         di->d_spc_timer = dm->dqb_btime;
2556         di->d_ino_timer = dm->dqb_itime;
2557         spin_unlock(&dq_data_lock);
2558 }
2559
2560 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2561                     struct qc_dqblk *di)
2562 {
2563         struct dquot *dquot;
2564
2565         dquot = dqget(sb, qid);
2566         if (IS_ERR(dquot))
2567                 return PTR_ERR(dquot);
2568         do_get_dqblk(dquot, di);
2569         dqput(dquot);
2570
2571         return 0;
2572 }
2573 EXPORT_SYMBOL(dquot_get_dqblk);
2574
2575 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2576                          struct qc_dqblk *di)
2577 {
2578         struct dquot *dquot;
2579         int err;
2580
2581         if (!sb->dq_op->get_next_id)
2582                 return -ENOSYS;
2583         err = sb->dq_op->get_next_id(sb, qid);
2584         if (err < 0)
2585                 return err;
2586         dquot = dqget(sb, *qid);
2587         if (IS_ERR(dquot))
2588                 return PTR_ERR(dquot);
2589         do_get_dqblk(dquot, di);
2590         dqput(dquot);
2591
2592         return 0;
2593 }
2594 EXPORT_SYMBOL(dquot_get_next_dqblk);
2595
2596 #define VFS_QC_MASK \
2597         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2598          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2599          QC_SPC_TIMER | QC_INO_TIMER)
2600
2601 /* Generic routine for setting common part of quota structure */
2602 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2603 {
2604         struct mem_dqblk *dm = &dquot->dq_dqb;
2605         int check_blim = 0, check_ilim = 0;
2606         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2607
2608         if (di->d_fieldmask & ~VFS_QC_MASK)
2609                 return -EINVAL;
2610
2611         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2612              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2613             ((di->d_fieldmask & QC_SPC_HARD) &&
2614              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2615             ((di->d_fieldmask & QC_INO_SOFT) &&
2616              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2617             ((di->d_fieldmask & QC_INO_HARD) &&
2618              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2619                 return -ERANGE;
2620
2621         spin_lock(&dq_data_lock);
2622         if (di->d_fieldmask & QC_SPACE) {
2623                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2624                 check_blim = 1;
2625                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2626         }
2627
2628         if (di->d_fieldmask & QC_SPC_SOFT)
2629                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2630         if (di->d_fieldmask & QC_SPC_HARD)
2631                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2632         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2633                 check_blim = 1;
2634                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2635         }
2636
2637         if (di->d_fieldmask & QC_INO_COUNT) {
2638                 dm->dqb_curinodes = di->d_ino_count;
2639                 check_ilim = 1;
2640                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2641         }
2642
2643         if (di->d_fieldmask & QC_INO_SOFT)
2644                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2645         if (di->d_fieldmask & QC_INO_HARD)
2646                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2647         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2648                 check_ilim = 1;
2649                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2650         }
2651
2652         if (di->d_fieldmask & QC_SPC_TIMER) {
2653                 dm->dqb_btime = di->d_spc_timer;
2654                 check_blim = 1;
2655                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2656         }
2657
2658         if (di->d_fieldmask & QC_INO_TIMER) {
2659                 dm->dqb_itime = di->d_ino_timer;
2660                 check_ilim = 1;
2661                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2662         }
2663
2664         if (check_blim) {
2665                 if (!dm->dqb_bsoftlimit ||
2666                     dm->dqb_curspace < dm->dqb_bsoftlimit) {
2667                         dm->dqb_btime = 0;
2668                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2669                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2670                         /* Set grace only if user hasn't provided his own... */
2671                         dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2672         }
2673         if (check_ilim) {
2674                 if (!dm->dqb_isoftlimit ||
2675                     dm->dqb_curinodes < dm->dqb_isoftlimit) {
2676                         dm->dqb_itime = 0;
2677                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2678                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2679                         /* Set grace only if user hasn't provided his own... */
2680                         dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2681         }
2682         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2683             dm->dqb_isoftlimit)
2684                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2685         else
2686                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2687         spin_unlock(&dq_data_lock);
2688         mark_dquot_dirty(dquot);
2689
2690         return 0;
2691 }
2692
2693 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2694                   struct qc_dqblk *di)
2695 {
2696         struct dquot *dquot;
2697         int rc;
2698
2699         dquot = dqget(sb, qid);
2700         if (IS_ERR(dquot)) {
2701                 rc = PTR_ERR(dquot);
2702                 goto out;
2703         }
2704         rc = do_set_dqblk(dquot, di);
2705         dqput(dquot);
2706 out:
2707         return rc;
2708 }
2709 EXPORT_SYMBOL(dquot_set_dqblk);
2710
2711 /* Generic routine for getting common part of quota file information */
2712 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2713 {
2714         struct mem_dqinfo *mi;
2715         struct qc_type_state *tstate;
2716         struct quota_info *dqopt = sb_dqopt(sb);
2717         int type;
2718   
2719         memset(state, 0, sizeof(*state));
2720         for (type = 0; type < MAXQUOTAS; type++) {
2721                 if (!sb_has_quota_active(sb, type))
2722                         continue;
2723                 tstate = state->s_state + type;
2724                 mi = sb_dqopt(sb)->info + type;
2725                 tstate->flags = QCI_ACCT_ENABLED;
2726                 spin_lock(&dq_data_lock);
2727                 if (mi->dqi_flags & DQF_SYS_FILE)
2728                         tstate->flags |= QCI_SYSFILE;
2729                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2730                         tstate->flags |= QCI_ROOT_SQUASH;
2731                 if (sb_has_quota_limits_enabled(sb, type))
2732                         tstate->flags |= QCI_LIMITS_ENFORCED;
2733                 tstate->spc_timelimit = mi->dqi_bgrace;
2734                 tstate->ino_timelimit = mi->dqi_igrace;
2735                 tstate->ino = dqopt->files[type]->i_ino;
2736                 tstate->blocks = dqopt->files[type]->i_blocks;
2737                 tstate->nextents = 1;   /* We don't know... */
2738                 spin_unlock(&dq_data_lock);
2739         }
2740         return 0;
2741 }
2742 EXPORT_SYMBOL(dquot_get_state);
2743
2744 /* Generic routine for setting common part of quota file information */
2745 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2746 {
2747         struct mem_dqinfo *mi;
2748         int err = 0;
2749
2750         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2751             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2752                 return -EINVAL;
2753         if (!sb_has_quota_active(sb, type))
2754                 return -ESRCH;
2755         mi = sb_dqopt(sb)->info + type;
2756         if (ii->i_fieldmask & QC_FLAGS) {
2757                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2758                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2759                         return -EINVAL;
2760         }
2761         spin_lock(&dq_data_lock);
2762         if (ii->i_fieldmask & QC_SPC_TIMER)
2763                 mi->dqi_bgrace = ii->i_spc_timelimit;
2764         if (ii->i_fieldmask & QC_INO_TIMER)
2765                 mi->dqi_igrace = ii->i_ino_timelimit;
2766         if (ii->i_fieldmask & QC_FLAGS) {
2767                 if (ii->i_flags & QCI_ROOT_SQUASH)
2768                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2769                 else
2770                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2771         }
2772         spin_unlock(&dq_data_lock);
2773         mark_info_dirty(sb, type);
2774         /* Force write to disk */
2775         sb->dq_op->write_info(sb, type);
2776         return err;
2777 }
2778 EXPORT_SYMBOL(dquot_set_dqinfo);
2779
2780 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2781         .quota_enable   = dquot_quota_enable,
2782         .quota_disable  = dquot_quota_disable,
2783         .quota_sync     = dquot_quota_sync,
2784         .get_state      = dquot_get_state,
2785         .set_info       = dquot_set_dqinfo,
2786         .get_dqblk      = dquot_get_dqblk,
2787         .get_nextdqblk  = dquot_get_next_dqblk,
2788         .set_dqblk      = dquot_set_dqblk
2789 };
2790 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2791
2792 static int do_proc_dqstats(struct ctl_table *table, int write,
2793                      void __user *buffer, size_t *lenp, loff_t *ppos)
2794 {
2795         unsigned int type = (int *)table->data - dqstats.stat;
2796
2797         /* Update global table */
2798         dqstats.stat[type] =
2799                         percpu_counter_sum_positive(&dqstats.counter[type]);
2800         return proc_dointvec(table, write, buffer, lenp, ppos);
2801 }
2802
2803 static struct ctl_table fs_dqstats_table[] = {
2804         {
2805                 .procname       = "lookups",
2806                 .data           = &dqstats.stat[DQST_LOOKUPS],
2807                 .maxlen         = sizeof(int),
2808                 .mode           = 0444,
2809                 .proc_handler   = do_proc_dqstats,
2810         },
2811         {
2812                 .procname       = "drops",
2813                 .data           = &dqstats.stat[DQST_DROPS],
2814                 .maxlen         = sizeof(int),
2815                 .mode           = 0444,
2816                 .proc_handler   = do_proc_dqstats,
2817         },
2818         {
2819                 .procname       = "reads",
2820                 .data           = &dqstats.stat[DQST_READS],
2821                 .maxlen         = sizeof(int),
2822                 .mode           = 0444,
2823                 .proc_handler   = do_proc_dqstats,
2824         },
2825         {
2826                 .procname       = "writes",
2827                 .data           = &dqstats.stat[DQST_WRITES],
2828                 .maxlen         = sizeof(int),
2829                 .mode           = 0444,
2830                 .proc_handler   = do_proc_dqstats,
2831         },
2832         {
2833                 .procname       = "cache_hits",
2834                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2835                 .maxlen         = sizeof(int),
2836                 .mode           = 0444,
2837                 .proc_handler   = do_proc_dqstats,
2838         },
2839         {
2840                 .procname       = "allocated_dquots",
2841                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2842                 .maxlen         = sizeof(int),
2843                 .mode           = 0444,
2844                 .proc_handler   = do_proc_dqstats,
2845         },
2846         {
2847                 .procname       = "free_dquots",
2848                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2849                 .maxlen         = sizeof(int),
2850                 .mode           = 0444,
2851                 .proc_handler   = do_proc_dqstats,
2852         },
2853         {
2854                 .procname       = "syncs",
2855                 .data           = &dqstats.stat[DQST_SYNCS],
2856                 .maxlen         = sizeof(int),
2857                 .mode           = 0444,
2858                 .proc_handler   = do_proc_dqstats,
2859         },
2860 #ifdef CONFIG_PRINT_QUOTA_WARNING
2861         {
2862                 .procname       = "warnings",
2863                 .data           = &flag_print_warnings,
2864                 .maxlen         = sizeof(int),
2865                 .mode           = 0644,
2866                 .proc_handler   = proc_dointvec,
2867         },
2868 #endif
2869         { },
2870 };
2871
2872 static struct ctl_table fs_table[] = {
2873         {
2874                 .procname       = "quota",
2875                 .mode           = 0555,
2876                 .child          = fs_dqstats_table,
2877         },
2878         { },
2879 };
2880
2881 static struct ctl_table sys_table[] = {
2882         {
2883                 .procname       = "fs",
2884                 .mode           = 0555,
2885                 .child          = fs_table,
2886         },
2887         { },
2888 };
2889
2890 static int __init dquot_init(void)
2891 {
2892         int i, ret;
2893         unsigned long nr_hash, order;
2894
2895         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2896
2897         register_sysctl_table(sys_table);
2898
2899         dquot_cachep = kmem_cache_create("dquot",
2900                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2901                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2902                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2903                         NULL);
2904
2905         order = 0;
2906         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2907         if (!dquot_hash)
2908                 panic("Cannot create dquot hash table");
2909
2910         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2911                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2912                 if (ret)
2913                         panic("Cannot create dquot stat counters");
2914         }
2915
2916         /* Find power-of-two hlist_heads which can fit into allocation */
2917         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2918         dq_hash_bits = 0;
2919         do {
2920                 dq_hash_bits++;
2921         } while (nr_hash >> dq_hash_bits);
2922         dq_hash_bits--;
2923
2924         nr_hash = 1UL << dq_hash_bits;
2925         dq_hash_mask = nr_hash - 1;
2926         for (i = 0; i < nr_hash; i++)
2927                 INIT_HLIST_HEAD(dquot_hash + i);
2928
2929         pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2930                 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2931
2932         register_shrinker(&dqcache_shrinker);
2933
2934         return 0;
2935 }
2936 fs_initcall(dquot_init);