quota: Push dqio_sem down to ->release_dqblk()
[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 quota is dirty already, we don't have to acquire dq_list_lock */
343         if (test_bit(DQ_MOD_B, &dquot->dq_flags))
344                 return 1;
345
346         spin_lock(&dq_list_lock);
347         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
348                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
349                                 info[dquot->dq_id.type].dqi_dirty_list);
350                 ret = 0;
351         }
352         spin_unlock(&dq_list_lock);
353         return ret;
354 }
355 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
356
357 /* Dirtify all the dquots - this can block when journalling */
358 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
359 {
360         int ret, err, cnt;
361
362         ret = err = 0;
363         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
364                 if (dquot[cnt])
365                         /* Even in case of error we have to continue */
366                         ret = mark_dquot_dirty(dquot[cnt]);
367                 if (!err)
368                         err = ret;
369         }
370         return err;
371 }
372
373 static inline void dqput_all(struct dquot **dquot)
374 {
375         unsigned int cnt;
376
377         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
378                 dqput(dquot[cnt]);
379 }
380
381 /* This function needs dq_list_lock */
382 static inline int clear_dquot_dirty(struct dquot *dquot)
383 {
384         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
385                 return 0;
386         list_del_init(&dquot->dq_dirty);
387         return 1;
388 }
389
390 void mark_info_dirty(struct super_block *sb, int type)
391 {
392         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
393 }
394 EXPORT_SYMBOL(mark_info_dirty);
395
396 /*
397  *      Read dquot from disk and alloc space for it
398  */
399
400 int dquot_acquire(struct dquot *dquot)
401 {
402         int ret = 0, ret2 = 0;
403         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
404
405         mutex_lock(&dquot->dq_lock);
406         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
407                 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
408         if (ret < 0)
409                 goto out_iolock;
410         /* Make sure flags update is visible after dquot has been filled */
411         smp_mb__before_atomic();
412         set_bit(DQ_READ_B, &dquot->dq_flags);
413         /* Instantiate dquot if needed */
414         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
415                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
416                 /* Write the info if needed */
417                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
418                         down_write(&dqopt->dqio_sem);
419                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
420                                         dquot->dq_sb, dquot->dq_id.type);
421                         up_write(&dqopt->dqio_sem);
422                 }
423                 if (ret < 0)
424                         goto out_iolock;
425                 if (ret2 < 0) {
426                         ret = ret2;
427                         goto out_iolock;
428                 }
429         }
430         /*
431          * Make sure flags update is visible after on-disk struct has been
432          * allocated. Paired with smp_rmb() in dqget().
433          */
434         smp_mb__before_atomic();
435         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
436 out_iolock:
437         mutex_unlock(&dquot->dq_lock);
438         return ret;
439 }
440 EXPORT_SYMBOL(dquot_acquire);
441
442 /*
443  *      Write dquot to disk
444  */
445 int dquot_commit(struct dquot *dquot)
446 {
447         int ret = 0;
448         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
449
450         mutex_lock(&dquot->dq_lock);
451         spin_lock(&dq_list_lock);
452         if (!clear_dquot_dirty(dquot)) {
453                 spin_unlock(&dq_list_lock);
454                 goto out_lock;
455         }
456         spin_unlock(&dq_list_lock);
457         /* Inactive dquot can be only if there was error during read/init
458          * => we have better not writing it */
459         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
460                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
461         else
462                 ret = -EIO;
463 out_lock:
464         mutex_unlock(&dquot->dq_lock);
465         return ret;
466 }
467 EXPORT_SYMBOL(dquot_commit);
468
469 /*
470  *      Release dquot
471  */
472 int dquot_release(struct dquot *dquot)
473 {
474         int ret = 0, ret2 = 0;
475         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
476
477         mutex_lock(&dquot->dq_lock);
478         /* Check whether we are not racing with some other dqget() */
479         if (atomic_read(&dquot->dq_count) > 1)
480                 goto out_dqlock;
481         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
482                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
483                 /* Write the info */
484                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
485                         down_write(&dqopt->dqio_sem);
486                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
487                                                 dquot->dq_sb, dquot->dq_id.type);
488                         up_write(&dqopt->dqio_sem);
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                         /* Dirty and inactive can be only bad dquot... */
630                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
631                                 clear_dquot_dirty(dquot);
632                                 continue;
633                         }
634                         /* Now we have active dquot from which someone is
635                          * holding reference so we can safely just increase
636                          * use count */
637                         dqgrab(dquot);
638                         spin_unlock(&dq_list_lock);
639                         dqstats_inc(DQST_LOOKUPS);
640                         err = sb->dq_op->write_dquot(dquot);
641                         if (!ret && err)
642                                 ret = err;
643                         dqput(dquot);
644                         spin_lock(&dq_list_lock);
645                 }
646                 spin_unlock(&dq_list_lock);
647         }
648
649         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
650                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
651                     && info_dirty(&dqopt->info[cnt]))
652                         sb->dq_op->write_info(sb, cnt);
653         dqstats_inc(DQST_SYNCS);
654
655         return ret;
656 }
657 EXPORT_SYMBOL(dquot_writeback_dquots);
658
659 /* Write all dquot structures to disk and make them visible from userspace */
660 int dquot_quota_sync(struct super_block *sb, int type)
661 {
662         struct quota_info *dqopt = sb_dqopt(sb);
663         int cnt;
664         int ret;
665
666         ret = dquot_writeback_dquots(sb, type);
667         if (ret)
668                 return ret;
669         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
670                 return 0;
671
672         /* This is not very clever (and fast) but currently I don't know about
673          * any other simple way of getting quota data to disk and we must get
674          * them there for userspace to be visible... */
675         if (sb->s_op->sync_fs)
676                 sb->s_op->sync_fs(sb, 1);
677         sync_blockdev(sb->s_bdev);
678
679         /*
680          * Now when everything is written we can discard the pagecache so
681          * that userspace sees the changes.
682          */
683         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
684                 if (type != -1 && cnt != type)
685                         continue;
686                 if (!sb_has_quota_active(sb, cnt))
687                         continue;
688                 inode_lock(dqopt->files[cnt]);
689                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
690                 inode_unlock(dqopt->files[cnt]);
691         }
692
693         return 0;
694 }
695 EXPORT_SYMBOL(dquot_quota_sync);
696
697 static unsigned long
698 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
699 {
700         struct list_head *head;
701         struct dquot *dquot;
702         unsigned long freed = 0;
703
704         spin_lock(&dq_list_lock);
705         head = free_dquots.prev;
706         while (head != &free_dquots && sc->nr_to_scan) {
707                 dquot = list_entry(head, struct dquot, dq_free);
708                 remove_dquot_hash(dquot);
709                 remove_free_dquot(dquot);
710                 remove_inuse(dquot);
711                 do_destroy_dquot(dquot);
712                 sc->nr_to_scan--;
713                 freed++;
714                 head = free_dquots.prev;
715         }
716         spin_unlock(&dq_list_lock);
717         return freed;
718 }
719
720 static unsigned long
721 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
722 {
723         return vfs_pressure_ratio(
724         percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
725 }
726
727 static struct shrinker dqcache_shrinker = {
728         .count_objects = dqcache_shrink_count,
729         .scan_objects = dqcache_shrink_scan,
730         .seeks = DEFAULT_SEEKS,
731 };
732
733 /*
734  * Put reference to dquot
735  */
736 void dqput(struct dquot *dquot)
737 {
738         int ret;
739
740         if (!dquot)
741                 return;
742 #ifdef CONFIG_QUOTA_DEBUG
743         if (!atomic_read(&dquot->dq_count)) {
744                 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
745                             quotatypes[dquot->dq_id.type],
746                             from_kqid(&init_user_ns, dquot->dq_id));
747                 BUG();
748         }
749 #endif
750         dqstats_inc(DQST_DROPS);
751 we_slept:
752         spin_lock(&dq_list_lock);
753         if (atomic_read(&dquot->dq_count) > 1) {
754                 /* We have more than one user... nothing to do */
755                 atomic_dec(&dquot->dq_count);
756                 /* Releasing dquot during quotaoff phase? */
757                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
758                     atomic_read(&dquot->dq_count) == 1)
759                         wake_up(&dquot->dq_wait_unused);
760                 spin_unlock(&dq_list_lock);
761                 return;
762         }
763         /* Need to release dquot? */
764         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
765                 spin_unlock(&dq_list_lock);
766                 /* Commit dquot before releasing */
767                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
768                 if (ret < 0) {
769                         quota_error(dquot->dq_sb, "Can't write quota structure"
770                                     " (error %d). Quota may get out of sync!",
771                                     ret);
772                         /*
773                          * We clear dirty bit anyway, so that we avoid
774                          * infinite loop here
775                          */
776                         spin_lock(&dq_list_lock);
777                         clear_dquot_dirty(dquot);
778                         spin_unlock(&dq_list_lock);
779                 }
780                 goto we_slept;
781         }
782         /* Clear flag in case dquot was inactive (something bad happened) */
783         clear_dquot_dirty(dquot);
784         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
785                 spin_unlock(&dq_list_lock);
786                 dquot->dq_sb->dq_op->release_dquot(dquot);
787                 goto we_slept;
788         }
789         atomic_dec(&dquot->dq_count);
790 #ifdef CONFIG_QUOTA_DEBUG
791         /* sanity check */
792         BUG_ON(!list_empty(&dquot->dq_free));
793 #endif
794         put_dquot_last(dquot);
795         spin_unlock(&dq_list_lock);
796 }
797 EXPORT_SYMBOL(dqput);
798
799 struct dquot *dquot_alloc(struct super_block *sb, int type)
800 {
801         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
802 }
803 EXPORT_SYMBOL(dquot_alloc);
804
805 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
806 {
807         struct dquot *dquot;
808
809         dquot = sb->dq_op->alloc_dquot(sb, type);
810         if(!dquot)
811                 return NULL;
812
813         mutex_init(&dquot->dq_lock);
814         INIT_LIST_HEAD(&dquot->dq_free);
815         INIT_LIST_HEAD(&dquot->dq_inuse);
816         INIT_HLIST_NODE(&dquot->dq_hash);
817         INIT_LIST_HEAD(&dquot->dq_dirty);
818         init_waitqueue_head(&dquot->dq_wait_unused);
819         dquot->dq_sb = sb;
820         dquot->dq_id = make_kqid_invalid(type);
821         atomic_set(&dquot->dq_count, 1);
822
823         return dquot;
824 }
825
826 /*
827  * Get reference to dquot
828  *
829  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
830  * destroying our dquot by:
831  *   a) checking for quota flags under dq_list_lock and
832  *   b) getting a reference to dquot before we release dq_list_lock
833  */
834 struct dquot *dqget(struct super_block *sb, struct kqid qid)
835 {
836         unsigned int hashent = hashfn(sb, qid);
837         struct dquot *dquot, *empty = NULL;
838
839         if (!qid_has_mapping(sb->s_user_ns, qid))
840                 return ERR_PTR(-EINVAL);
841
842         if (!sb_has_quota_active(sb, qid.type))
843                 return ERR_PTR(-ESRCH);
844 we_slept:
845         spin_lock(&dq_list_lock);
846         spin_lock(&dq_state_lock);
847         if (!sb_has_quota_active(sb, qid.type)) {
848                 spin_unlock(&dq_state_lock);
849                 spin_unlock(&dq_list_lock);
850                 dquot = ERR_PTR(-ESRCH);
851                 goto out;
852         }
853         spin_unlock(&dq_state_lock);
854
855         dquot = find_dquot(hashent, sb, qid);
856         if (!dquot) {
857                 if (!empty) {
858                         spin_unlock(&dq_list_lock);
859                         empty = get_empty_dquot(sb, qid.type);
860                         if (!empty)
861                                 schedule();     /* Try to wait for a moment... */
862                         goto we_slept;
863                 }
864                 dquot = empty;
865                 empty = NULL;
866                 dquot->dq_id = qid;
867                 /* all dquots go on the inuse_list */
868                 put_inuse(dquot);
869                 /* hash it first so it can be found */
870                 insert_dquot_hash(dquot);
871                 spin_unlock(&dq_list_lock);
872                 dqstats_inc(DQST_LOOKUPS);
873         } else {
874                 if (!atomic_read(&dquot->dq_count))
875                         remove_free_dquot(dquot);
876                 atomic_inc(&dquot->dq_count);
877                 spin_unlock(&dq_list_lock);
878                 dqstats_inc(DQST_CACHE_HITS);
879                 dqstats_inc(DQST_LOOKUPS);
880         }
881         /* Wait for dq_lock - after this we know that either dquot_release() is
882          * already finished or it will be canceled due to dq_count > 1 test */
883         wait_on_dquot(dquot);
884         /* Read the dquot / allocate space in quota file */
885         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
886                 int err;
887
888                 err = sb->dq_op->acquire_dquot(dquot);
889                 if (err < 0) {
890                         dqput(dquot);
891                         dquot = ERR_PTR(err);
892                         goto out;
893                 }
894         }
895         /*
896          * Make sure following reads see filled structure - paired with
897          * smp_mb__before_atomic() in dquot_acquire().
898          */
899         smp_rmb();
900 #ifdef CONFIG_QUOTA_DEBUG
901         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
902 #endif
903 out:
904         if (empty)
905                 do_destroy_dquot(empty);
906
907         return dquot;
908 }
909 EXPORT_SYMBOL(dqget);
910
911 static inline struct dquot **i_dquot(struct inode *inode)
912 {
913         return inode->i_sb->s_op->get_dquots(inode);
914 }
915
916 static int dqinit_needed(struct inode *inode, int type)
917 {
918         struct dquot * const *dquots;
919         int cnt;
920
921         if (IS_NOQUOTA(inode))
922                 return 0;
923
924         dquots = i_dquot(inode);
925         if (type != -1)
926                 return !dquots[type];
927         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
928                 if (!dquots[cnt])
929                         return 1;
930         return 0;
931 }
932
933 /* This routine is guarded by s_umount semaphore */
934 static void add_dquot_ref(struct super_block *sb, int type)
935 {
936         struct inode *inode, *old_inode = NULL;
937 #ifdef CONFIG_QUOTA_DEBUG
938         int reserved = 0;
939 #endif
940
941         spin_lock(&sb->s_inode_list_lock);
942         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
943                 spin_lock(&inode->i_lock);
944                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
945                     !atomic_read(&inode->i_writecount) ||
946                     !dqinit_needed(inode, type)) {
947                         spin_unlock(&inode->i_lock);
948                         continue;
949                 }
950                 __iget(inode);
951                 spin_unlock(&inode->i_lock);
952                 spin_unlock(&sb->s_inode_list_lock);
953
954 #ifdef CONFIG_QUOTA_DEBUG
955                 if (unlikely(inode_get_rsv_space(inode) > 0))
956                         reserved = 1;
957 #endif
958                 iput(old_inode);
959                 __dquot_initialize(inode, type);
960
961                 /*
962                  * We hold a reference to 'inode' so it couldn't have been
963                  * removed from s_inodes list while we dropped the
964                  * s_inode_list_lock. We cannot iput the inode now as we can be
965                  * holding the last reference and we cannot iput it under
966                  * s_inode_list_lock. So we keep the reference and iput it
967                  * later.
968                  */
969                 old_inode = inode;
970                 spin_lock(&sb->s_inode_list_lock);
971         }
972         spin_unlock(&sb->s_inode_list_lock);
973         iput(old_inode);
974
975 #ifdef CONFIG_QUOTA_DEBUG
976         if (reserved) {
977                 quota_error(sb, "Writes happened before quota was turned on "
978                         "thus quota information is probably inconsistent. "
979                         "Please run quotacheck(8)");
980         }
981 #endif
982 }
983
984 /*
985  * Remove references to dquots from inode and add dquot to list for freeing
986  * if we have the last reference to dquot
987  */
988 static void remove_inode_dquot_ref(struct inode *inode, int type,
989                                    struct list_head *tofree_head)
990 {
991         struct dquot **dquots = i_dquot(inode);
992         struct dquot *dquot = dquots[type];
993
994         if (!dquot)
995                 return;
996
997         dquots[type] = NULL;
998         if (list_empty(&dquot->dq_free)) {
999                 /*
1000                  * The inode still has reference to dquot so it can't be in the
1001                  * free list
1002                  */
1003                 spin_lock(&dq_list_lock);
1004                 list_add(&dquot->dq_free, tofree_head);
1005                 spin_unlock(&dq_list_lock);
1006         } else {
1007                 /*
1008                  * Dquot is already in a list to put so we won't drop the last
1009                  * reference here.
1010                  */
1011                 dqput(dquot);
1012         }
1013 }
1014
1015 /*
1016  * Free list of dquots
1017  * Dquots are removed from inodes and no new references can be got so we are
1018  * the only ones holding reference
1019  */
1020 static void put_dquot_list(struct list_head *tofree_head)
1021 {
1022         struct list_head *act_head;
1023         struct dquot *dquot;
1024
1025         act_head = tofree_head->next;
1026         while (act_head != tofree_head) {
1027                 dquot = list_entry(act_head, struct dquot, dq_free);
1028                 act_head = act_head->next;
1029                 /* Remove dquot from the list so we won't have problems... */
1030                 list_del_init(&dquot->dq_free);
1031                 dqput(dquot);
1032         }
1033 }
1034
1035 static void remove_dquot_ref(struct super_block *sb, int type,
1036                 struct list_head *tofree_head)
1037 {
1038         struct inode *inode;
1039         int reserved = 0;
1040
1041         spin_lock(&sb->s_inode_list_lock);
1042         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1043                 /*
1044                  *  We have to scan also I_NEW inodes because they can already
1045                  *  have quota pointer initialized. Luckily, we need to touch
1046                  *  only quota pointers and these have separate locking
1047                  *  (dq_data_lock).
1048                  */
1049                 spin_lock(&dq_data_lock);
1050                 if (!IS_NOQUOTA(inode)) {
1051                         if (unlikely(inode_get_rsv_space(inode) > 0))
1052                                 reserved = 1;
1053                         remove_inode_dquot_ref(inode, type, tofree_head);
1054                 }
1055                 spin_unlock(&dq_data_lock);
1056         }
1057         spin_unlock(&sb->s_inode_list_lock);
1058 #ifdef CONFIG_QUOTA_DEBUG
1059         if (reserved) {
1060                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1061                         " was disabled thus quota information is probably "
1062                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1063         }
1064 #endif
1065 }
1066
1067 /* Gather all references from inodes and drop them */
1068 static void drop_dquot_ref(struct super_block *sb, int type)
1069 {
1070         LIST_HEAD(tofree_head);
1071
1072         if (sb->dq_op) {
1073                 remove_dquot_ref(sb, type, &tofree_head);
1074                 synchronize_srcu(&dquot_srcu);
1075                 put_dquot_list(&tofree_head);
1076         }
1077 }
1078
1079 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1080 {
1081         dquot->dq_dqb.dqb_curinodes += number;
1082 }
1083
1084 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1085 {
1086         dquot->dq_dqb.dqb_curspace += number;
1087 }
1088
1089 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1090 {
1091         dquot->dq_dqb.dqb_rsvspace += number;
1092 }
1093
1094 /*
1095  * Claim reserved quota space
1096  */
1097 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1098 {
1099         if (dquot->dq_dqb.dqb_rsvspace < number) {
1100                 WARN_ON_ONCE(1);
1101                 number = dquot->dq_dqb.dqb_rsvspace;
1102         }
1103         dquot->dq_dqb.dqb_curspace += number;
1104         dquot->dq_dqb.dqb_rsvspace -= number;
1105 }
1106
1107 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1108 {
1109         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1110                 number = dquot->dq_dqb.dqb_curspace;
1111         dquot->dq_dqb.dqb_rsvspace += number;
1112         dquot->dq_dqb.dqb_curspace -= number;
1113 }
1114
1115 static inline
1116 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1117 {
1118         if (dquot->dq_dqb.dqb_rsvspace >= number)
1119                 dquot->dq_dqb.dqb_rsvspace -= number;
1120         else {
1121                 WARN_ON_ONCE(1);
1122                 dquot->dq_dqb.dqb_rsvspace = 0;
1123         }
1124 }
1125
1126 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1127 {
1128         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1129             dquot->dq_dqb.dqb_curinodes >= number)
1130                 dquot->dq_dqb.dqb_curinodes -= number;
1131         else
1132                 dquot->dq_dqb.dqb_curinodes = 0;
1133         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1134                 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1135         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1136 }
1137
1138 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1139 {
1140         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1141             dquot->dq_dqb.dqb_curspace >= number)
1142                 dquot->dq_dqb.dqb_curspace -= number;
1143         else
1144                 dquot->dq_dqb.dqb_curspace = 0;
1145         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1146                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1147         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1148 }
1149
1150 struct dquot_warn {
1151         struct super_block *w_sb;
1152         struct kqid w_dq_id;
1153         short w_type;
1154 };
1155
1156 static int warning_issued(struct dquot *dquot, const int warntype)
1157 {
1158         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1159                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1160                 ((warntype == QUOTA_NL_IHARDWARN ||
1161                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1162
1163         if (!flag)
1164                 return 0;
1165         return test_and_set_bit(flag, &dquot->dq_flags);
1166 }
1167
1168 #ifdef CONFIG_PRINT_QUOTA_WARNING
1169 static int flag_print_warnings = 1;
1170
1171 static int need_print_warning(struct dquot_warn *warn)
1172 {
1173         if (!flag_print_warnings)
1174                 return 0;
1175
1176         switch (warn->w_dq_id.type) {
1177                 case USRQUOTA:
1178                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1179                 case GRPQUOTA:
1180                         return in_group_p(warn->w_dq_id.gid);
1181                 case PRJQUOTA:
1182                         return 1;
1183         }
1184         return 0;
1185 }
1186
1187 /* Print warning to user which exceeded quota */
1188 static void print_warning(struct dquot_warn *warn)
1189 {
1190         char *msg = NULL;
1191         struct tty_struct *tty;
1192         int warntype = warn->w_type;
1193
1194         if (warntype == QUOTA_NL_IHARDBELOW ||
1195             warntype == QUOTA_NL_ISOFTBELOW ||
1196             warntype == QUOTA_NL_BHARDBELOW ||
1197             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1198                 return;
1199
1200         tty = get_current_tty();
1201         if (!tty)
1202                 return;
1203         tty_write_message(tty, warn->w_sb->s_id);
1204         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1205                 tty_write_message(tty, ": warning, ");
1206         else
1207                 tty_write_message(tty, ": write failed, ");
1208         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1209         switch (warntype) {
1210                 case QUOTA_NL_IHARDWARN:
1211                         msg = " file limit reached.\r\n";
1212                         break;
1213                 case QUOTA_NL_ISOFTLONGWARN:
1214                         msg = " file quota exceeded too long.\r\n";
1215                         break;
1216                 case QUOTA_NL_ISOFTWARN:
1217                         msg = " file quota exceeded.\r\n";
1218                         break;
1219                 case QUOTA_NL_BHARDWARN:
1220                         msg = " block limit reached.\r\n";
1221                         break;
1222                 case QUOTA_NL_BSOFTLONGWARN:
1223                         msg = " block quota exceeded too long.\r\n";
1224                         break;
1225                 case QUOTA_NL_BSOFTWARN:
1226                         msg = " block quota exceeded.\r\n";
1227                         break;
1228         }
1229         tty_write_message(tty, msg);
1230         tty_kref_put(tty);
1231 }
1232 #endif
1233
1234 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1235                             int warntype)
1236 {
1237         if (warning_issued(dquot, warntype))
1238                 return;
1239         warn->w_type = warntype;
1240         warn->w_sb = dquot->dq_sb;
1241         warn->w_dq_id = dquot->dq_id;
1242 }
1243
1244 /*
1245  * Write warnings to the console and send warning messages over netlink.
1246  *
1247  * Note that this function can call into tty and networking code.
1248  */
1249 static void flush_warnings(struct dquot_warn *warn)
1250 {
1251         int i;
1252
1253         for (i = 0; i < MAXQUOTAS; i++) {
1254                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1255                         continue;
1256 #ifdef CONFIG_PRINT_QUOTA_WARNING
1257                 print_warning(&warn[i]);
1258 #endif
1259                 quota_send_warning(warn[i].w_dq_id,
1260                                    warn[i].w_sb->s_dev, warn[i].w_type);
1261         }
1262 }
1263
1264 static int ignore_hardlimit(struct dquot *dquot)
1265 {
1266         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1267
1268         return capable(CAP_SYS_RESOURCE) &&
1269                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1270                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1271 }
1272
1273 /* needs dq_data_lock */
1274 static int check_idq(struct dquot *dquot, qsize_t inodes,
1275                      struct dquot_warn *warn)
1276 {
1277         qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1278
1279         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1280             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1281                 return 0;
1282
1283         if (dquot->dq_dqb.dqb_ihardlimit &&
1284             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1285             !ignore_hardlimit(dquot)) {
1286                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1287                 return -EDQUOT;
1288         }
1289
1290         if (dquot->dq_dqb.dqb_isoftlimit &&
1291             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1292             dquot->dq_dqb.dqb_itime &&
1293             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1294             !ignore_hardlimit(dquot)) {
1295                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1296                 return -EDQUOT;
1297         }
1298
1299         if (dquot->dq_dqb.dqb_isoftlimit &&
1300             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1301             dquot->dq_dqb.dqb_itime == 0) {
1302                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1303                 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1304                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1305         }
1306
1307         return 0;
1308 }
1309
1310 /* needs dq_data_lock */
1311 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1312                      struct dquot_warn *warn)
1313 {
1314         qsize_t tspace;
1315         struct super_block *sb = dquot->dq_sb;
1316
1317         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1318             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1319                 return 0;
1320
1321         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1322                 + space;
1323
1324         if (dquot->dq_dqb.dqb_bhardlimit &&
1325             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1326             !ignore_hardlimit(dquot)) {
1327                 if (!prealloc)
1328                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1329                 return -EDQUOT;
1330         }
1331
1332         if (dquot->dq_dqb.dqb_bsoftlimit &&
1333             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1334             dquot->dq_dqb.dqb_btime &&
1335             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1336             !ignore_hardlimit(dquot)) {
1337                 if (!prealloc)
1338                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1339                 return -EDQUOT;
1340         }
1341
1342         if (dquot->dq_dqb.dqb_bsoftlimit &&
1343             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1344             dquot->dq_dqb.dqb_btime == 0) {
1345                 if (!prealloc) {
1346                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1347                         dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1348                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1349                 }
1350                 else
1351                         /*
1352                          * We don't allow preallocation to exceed softlimit so exceeding will
1353                          * be always printed
1354                          */
1355                         return -EDQUOT;
1356         }
1357
1358         return 0;
1359 }
1360
1361 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1362 {
1363         qsize_t newinodes;
1364
1365         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1366             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1367             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1368                 return QUOTA_NL_NOWARN;
1369
1370         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1371         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1372                 return QUOTA_NL_ISOFTBELOW;
1373         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1374             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1375                 return QUOTA_NL_IHARDBELOW;
1376         return QUOTA_NL_NOWARN;
1377 }
1378
1379 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1380 {
1381         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1382             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1383                 return QUOTA_NL_NOWARN;
1384
1385         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1386                 return QUOTA_NL_BSOFTBELOW;
1387         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1388             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1389                 return QUOTA_NL_BHARDBELOW;
1390         return QUOTA_NL_NOWARN;
1391 }
1392
1393 static int dquot_active(const struct inode *inode)
1394 {
1395         struct super_block *sb = inode->i_sb;
1396
1397         if (IS_NOQUOTA(inode))
1398                 return 0;
1399         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1400 }
1401
1402 /*
1403  * Initialize quota pointers in inode
1404  *
1405  * It is better to call this function outside of any transaction as it
1406  * might need a lot of space in journal for dquot structure allocation.
1407  */
1408 static int __dquot_initialize(struct inode *inode, int type)
1409 {
1410         int cnt, init_needed = 0;
1411         struct dquot **dquots, *got[MAXQUOTAS] = {};
1412         struct super_block *sb = inode->i_sb;
1413         qsize_t rsv;
1414         int ret = 0;
1415
1416         if (!dquot_active(inode))
1417                 return 0;
1418
1419         dquots = i_dquot(inode);
1420
1421         /* First get references to structures we might need. */
1422         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1423                 struct kqid qid;
1424                 kprojid_t projid;
1425                 int rc;
1426                 struct dquot *dquot;
1427
1428                 if (type != -1 && cnt != type)
1429                         continue;
1430                 /*
1431                  * The i_dquot should have been initialized in most cases,
1432                  * we check it without locking here to avoid unnecessary
1433                  * dqget()/dqput() calls.
1434                  */
1435                 if (dquots[cnt])
1436                         continue;
1437
1438                 if (!sb_has_quota_active(sb, cnt))
1439                         continue;
1440
1441                 init_needed = 1;
1442
1443                 switch (cnt) {
1444                 case USRQUOTA:
1445                         qid = make_kqid_uid(inode->i_uid);
1446                         break;
1447                 case GRPQUOTA:
1448                         qid = make_kqid_gid(inode->i_gid);
1449                         break;
1450                 case PRJQUOTA:
1451                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1452                         if (rc)
1453                                 continue;
1454                         qid = make_kqid_projid(projid);
1455                         break;
1456                 }
1457                 dquot = dqget(sb, qid);
1458                 if (IS_ERR(dquot)) {
1459                         /* We raced with somebody turning quotas off... */
1460                         if (PTR_ERR(dquot) != -ESRCH) {
1461                                 ret = PTR_ERR(dquot);
1462                                 goto out_put;
1463                         }
1464                         dquot = NULL;
1465                 }
1466                 got[cnt] = dquot;
1467         }
1468
1469         /* All required i_dquot has been initialized */
1470         if (!init_needed)
1471                 return 0;
1472
1473         spin_lock(&dq_data_lock);
1474         if (IS_NOQUOTA(inode))
1475                 goto out_lock;
1476         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1477                 if (type != -1 && cnt != type)
1478                         continue;
1479                 /* Avoid races with quotaoff() */
1480                 if (!sb_has_quota_active(sb, cnt))
1481                         continue;
1482                 /* We could race with quotaon or dqget() could have failed */
1483                 if (!got[cnt])
1484                         continue;
1485                 if (!dquots[cnt]) {
1486                         dquots[cnt] = got[cnt];
1487                         got[cnt] = NULL;
1488                         /*
1489                          * Make quota reservation system happy if someone
1490                          * did a write before quota was turned on
1491                          */
1492                         rsv = inode_get_rsv_space(inode);
1493                         if (unlikely(rsv))
1494                                 dquot_resv_space(dquots[cnt], rsv);
1495                 }
1496         }
1497 out_lock:
1498         spin_unlock(&dq_data_lock);
1499 out_put:
1500         /* Drop unused references */
1501         dqput_all(got);
1502
1503         return ret;
1504 }
1505
1506 int dquot_initialize(struct inode *inode)
1507 {
1508         return __dquot_initialize(inode, -1);
1509 }
1510 EXPORT_SYMBOL(dquot_initialize);
1511
1512 bool dquot_initialize_needed(struct inode *inode)
1513 {
1514         struct dquot **dquots;
1515         int i;
1516
1517         if (!dquot_active(inode))
1518                 return false;
1519
1520         dquots = i_dquot(inode);
1521         for (i = 0; i < MAXQUOTAS; i++)
1522                 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1523                         return true;
1524         return false;
1525 }
1526 EXPORT_SYMBOL(dquot_initialize_needed);
1527
1528 /*
1529  * Release all quotas referenced by inode.
1530  *
1531  * This function only be called on inode free or converting
1532  * a file to quota file, no other users for the i_dquot in
1533  * both cases, so we needn't call synchronize_srcu() after
1534  * clearing i_dquot.
1535  */
1536 static void __dquot_drop(struct inode *inode)
1537 {
1538         int cnt;
1539         struct dquot **dquots = i_dquot(inode);
1540         struct dquot *put[MAXQUOTAS];
1541
1542         spin_lock(&dq_data_lock);
1543         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1544                 put[cnt] = dquots[cnt];
1545                 dquots[cnt] = NULL;
1546         }
1547         spin_unlock(&dq_data_lock);
1548         dqput_all(put);
1549 }
1550
1551 void dquot_drop(struct inode *inode)
1552 {
1553         struct dquot * const *dquots;
1554         int cnt;
1555
1556         if (IS_NOQUOTA(inode))
1557                 return;
1558
1559         /*
1560          * Test before calling to rule out calls from proc and such
1561          * where we are not allowed to block. Note that this is
1562          * actually reliable test even without the lock - the caller
1563          * must assure that nobody can come after the DQUOT_DROP and
1564          * add quota pointers back anyway.
1565          */
1566         dquots = i_dquot(inode);
1567         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1568                 if (dquots[cnt])
1569                         break;
1570         }
1571
1572         if (cnt < MAXQUOTAS)
1573                 __dquot_drop(inode);
1574 }
1575 EXPORT_SYMBOL(dquot_drop);
1576
1577 /*
1578  * inode_reserved_space is managed internally by quota, and protected by
1579  * i_lock similar to i_blocks+i_bytes.
1580  */
1581 static qsize_t *inode_reserved_space(struct inode * inode)
1582 {
1583         /* Filesystem must explicitly define it's own method in order to use
1584          * quota reservation interface */
1585         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1586         return inode->i_sb->dq_op->get_reserved_space(inode);
1587 }
1588
1589 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1590 {
1591         spin_lock(&inode->i_lock);
1592         *inode_reserved_space(inode) += number;
1593         spin_unlock(&inode->i_lock);
1594 }
1595 EXPORT_SYMBOL(inode_add_rsv_space);
1596
1597 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1598 {
1599         spin_lock(&inode->i_lock);
1600         *inode_reserved_space(inode) -= number;
1601         __inode_add_bytes(inode, number);
1602         spin_unlock(&inode->i_lock);
1603 }
1604 EXPORT_SYMBOL(inode_claim_rsv_space);
1605
1606 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1607 {
1608         spin_lock(&inode->i_lock);
1609         *inode_reserved_space(inode) += number;
1610         __inode_sub_bytes(inode, number);
1611         spin_unlock(&inode->i_lock);
1612 }
1613 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1614
1615 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1616 {
1617         spin_lock(&inode->i_lock);
1618         *inode_reserved_space(inode) -= number;
1619         spin_unlock(&inode->i_lock);
1620 }
1621 EXPORT_SYMBOL(inode_sub_rsv_space);
1622
1623 static qsize_t inode_get_rsv_space(struct inode *inode)
1624 {
1625         qsize_t ret;
1626
1627         if (!inode->i_sb->dq_op->get_reserved_space)
1628                 return 0;
1629         spin_lock(&inode->i_lock);
1630         ret = *inode_reserved_space(inode);
1631         spin_unlock(&inode->i_lock);
1632         return ret;
1633 }
1634
1635 static void inode_incr_space(struct inode *inode, qsize_t number,
1636                                 int reserve)
1637 {
1638         if (reserve)
1639                 inode_add_rsv_space(inode, number);
1640         else
1641                 inode_add_bytes(inode, number);
1642 }
1643
1644 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1645 {
1646         if (reserve)
1647                 inode_sub_rsv_space(inode, number);
1648         else
1649                 inode_sub_bytes(inode, number);
1650 }
1651
1652 /*
1653  * This functions updates i_blocks+i_bytes fields and quota information
1654  * (together with appropriate checks).
1655  *
1656  * NOTE: We absolutely rely on the fact that caller dirties the inode
1657  * (usually helpers in quotaops.h care about this) and holds a handle for
1658  * the current transaction so that dquot write and inode write go into the
1659  * same transaction.
1660  */
1661
1662 /*
1663  * This operation can block, but only after everything is updated
1664  */
1665 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1666 {
1667         int cnt, ret = 0, index;
1668         struct dquot_warn warn[MAXQUOTAS];
1669         int reserve = flags & DQUOT_SPACE_RESERVE;
1670         struct dquot **dquots;
1671
1672         if (!dquot_active(inode)) {
1673                 inode_incr_space(inode, number, reserve);
1674                 goto out;
1675         }
1676
1677         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1678                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1679
1680         dquots = i_dquot(inode);
1681         index = srcu_read_lock(&dquot_srcu);
1682         spin_lock(&dq_data_lock);
1683         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1684                 if (!dquots[cnt])
1685                         continue;
1686                 ret = check_bdq(dquots[cnt], number,
1687                                 !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1688                 if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1689                         spin_unlock(&dq_data_lock);
1690                         goto out_flush_warn;
1691                 }
1692         }
1693         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1694                 if (!dquots[cnt])
1695                         continue;
1696                 if (reserve)
1697                         dquot_resv_space(dquots[cnt], number);
1698                 else
1699                         dquot_incr_space(dquots[cnt], number);
1700         }
1701         inode_incr_space(inode, number, reserve);
1702         spin_unlock(&dq_data_lock);
1703
1704         if (reserve)
1705                 goto out_flush_warn;
1706         mark_all_dquot_dirty(dquots);
1707 out_flush_warn:
1708         srcu_read_unlock(&dquot_srcu, index);
1709         flush_warnings(warn);
1710 out:
1711         return ret;
1712 }
1713 EXPORT_SYMBOL(__dquot_alloc_space);
1714
1715 /*
1716  * This operation can block, but only after everything is updated
1717  */
1718 int dquot_alloc_inode(struct inode *inode)
1719 {
1720         int cnt, ret = 0, index;
1721         struct dquot_warn warn[MAXQUOTAS];
1722         struct dquot * const *dquots;
1723
1724         if (!dquot_active(inode))
1725                 return 0;
1726         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1727                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1728
1729         dquots = i_dquot(inode);
1730         index = srcu_read_lock(&dquot_srcu);
1731         spin_lock(&dq_data_lock);
1732         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1733                 if (!dquots[cnt])
1734                         continue;
1735                 ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1736                 if (ret)
1737                         goto warn_put_all;
1738         }
1739
1740         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1741                 if (!dquots[cnt])
1742                         continue;
1743                 dquot_incr_inodes(dquots[cnt], 1);
1744         }
1745
1746 warn_put_all:
1747         spin_unlock(&dq_data_lock);
1748         if (ret == 0)
1749                 mark_all_dquot_dirty(dquots);
1750         srcu_read_unlock(&dquot_srcu, index);
1751         flush_warnings(warn);
1752         return ret;
1753 }
1754 EXPORT_SYMBOL(dquot_alloc_inode);
1755
1756 /*
1757  * Convert in-memory reserved quotas to real consumed quotas
1758  */
1759 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1760 {
1761         struct dquot **dquots;
1762         int cnt, index;
1763
1764         if (!dquot_active(inode)) {
1765                 inode_claim_rsv_space(inode, number);
1766                 return 0;
1767         }
1768
1769         dquots = i_dquot(inode);
1770         index = srcu_read_lock(&dquot_srcu);
1771         spin_lock(&dq_data_lock);
1772         /* Claim reserved quotas to allocated quotas */
1773         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1774                 if (dquots[cnt])
1775                         dquot_claim_reserved_space(dquots[cnt], number);
1776         }
1777         /* Update inode bytes */
1778         inode_claim_rsv_space(inode, number);
1779         spin_unlock(&dq_data_lock);
1780         mark_all_dquot_dirty(dquots);
1781         srcu_read_unlock(&dquot_srcu, index);
1782         return 0;
1783 }
1784 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1785
1786 /*
1787  * Convert allocated space back to in-memory reserved quotas
1788  */
1789 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1790 {
1791         struct dquot **dquots;
1792         int cnt, index;
1793
1794         if (!dquot_active(inode)) {
1795                 inode_reclaim_rsv_space(inode, number);
1796                 return;
1797         }
1798
1799         dquots = i_dquot(inode);
1800         index = srcu_read_lock(&dquot_srcu);
1801         spin_lock(&dq_data_lock);
1802         /* Claim reserved quotas to allocated quotas */
1803         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1804                 if (dquots[cnt])
1805                         dquot_reclaim_reserved_space(dquots[cnt], number);
1806         }
1807         /* Update inode bytes */
1808         inode_reclaim_rsv_space(inode, number);
1809         spin_unlock(&dq_data_lock);
1810         mark_all_dquot_dirty(dquots);
1811         srcu_read_unlock(&dquot_srcu, index);
1812         return;
1813 }
1814 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1815
1816 /*
1817  * This operation can block, but only after everything is updated
1818  */
1819 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1820 {
1821         unsigned int cnt;
1822         struct dquot_warn warn[MAXQUOTAS];
1823         struct dquot **dquots;
1824         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1825
1826         if (!dquot_active(inode)) {
1827                 inode_decr_space(inode, number, reserve);
1828                 return;
1829         }
1830
1831         dquots = i_dquot(inode);
1832         index = srcu_read_lock(&dquot_srcu);
1833         spin_lock(&dq_data_lock);
1834         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1835                 int wtype;
1836
1837                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1838                 if (!dquots[cnt])
1839                         continue;
1840                 wtype = info_bdq_free(dquots[cnt], number);
1841                 if (wtype != QUOTA_NL_NOWARN)
1842                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1843                 if (reserve)
1844                         dquot_free_reserved_space(dquots[cnt], number);
1845                 else
1846                         dquot_decr_space(dquots[cnt], number);
1847         }
1848         inode_decr_space(inode, number, reserve);
1849         spin_unlock(&dq_data_lock);
1850
1851         if (reserve)
1852                 goto out_unlock;
1853         mark_all_dquot_dirty(dquots);
1854 out_unlock:
1855         srcu_read_unlock(&dquot_srcu, index);
1856         flush_warnings(warn);
1857 }
1858 EXPORT_SYMBOL(__dquot_free_space);
1859
1860 /*
1861  * This operation can block, but only after everything is updated
1862  */
1863 void dquot_free_inode(struct inode *inode)
1864 {
1865         unsigned int cnt;
1866         struct dquot_warn warn[MAXQUOTAS];
1867         struct dquot * const *dquots;
1868         int index;
1869
1870         if (!dquot_active(inode))
1871                 return;
1872
1873         dquots = i_dquot(inode);
1874         index = srcu_read_lock(&dquot_srcu);
1875         spin_lock(&dq_data_lock);
1876         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1877                 int wtype;
1878
1879                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1880                 if (!dquots[cnt])
1881                         continue;
1882                 wtype = info_idq_free(dquots[cnt], 1);
1883                 if (wtype != QUOTA_NL_NOWARN)
1884                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1885                 dquot_decr_inodes(dquots[cnt], 1);
1886         }
1887         spin_unlock(&dq_data_lock);
1888         mark_all_dquot_dirty(dquots);
1889         srcu_read_unlock(&dquot_srcu, index);
1890         flush_warnings(warn);
1891 }
1892 EXPORT_SYMBOL(dquot_free_inode);
1893
1894 /*
1895  * Transfer the number of inode and blocks from one diskquota to an other.
1896  * On success, dquot references in transfer_to are consumed and references
1897  * to original dquots that need to be released are placed there. On failure,
1898  * references are kept untouched.
1899  *
1900  * This operation can block, but only after everything is updated
1901  * A transaction must be started when entering this function.
1902  *
1903  * We are holding reference on transfer_from & transfer_to, no need to
1904  * protect them by srcu_read_lock().
1905  */
1906 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1907 {
1908         qsize_t space, cur_space;
1909         qsize_t rsv_space = 0;
1910         qsize_t inode_usage = 1;
1911         struct dquot *transfer_from[MAXQUOTAS] = {};
1912         int cnt, ret = 0;
1913         char is_valid[MAXQUOTAS] = {};
1914         struct dquot_warn warn_to[MAXQUOTAS];
1915         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1916         struct dquot_warn warn_from_space[MAXQUOTAS];
1917
1918         if (IS_NOQUOTA(inode))
1919                 return 0;
1920
1921         if (inode->i_sb->dq_op->get_inode_usage) {
1922                 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1923                 if (ret)
1924                         return ret;
1925         }
1926
1927         /* Initialize the arrays */
1928         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1929                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1930                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1931                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1932         }
1933
1934         spin_lock(&dq_data_lock);
1935         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1936                 spin_unlock(&dq_data_lock);
1937                 return 0;
1938         }
1939         cur_space = inode_get_bytes(inode);
1940         rsv_space = inode_get_rsv_space(inode);
1941         space = cur_space + rsv_space;
1942         /* Build the transfer_from list and check the limits */
1943         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1944                 /*
1945                  * Skip changes for same uid or gid or for turned off quota-type.
1946                  */
1947                 if (!transfer_to[cnt])
1948                         continue;
1949                 /* Avoid races with quotaoff() */
1950                 if (!sb_has_quota_active(inode->i_sb, cnt))
1951                         continue;
1952                 is_valid[cnt] = 1;
1953                 transfer_from[cnt] = i_dquot(inode)[cnt];
1954                 ret = check_idq(transfer_to[cnt], inode_usage, &warn_to[cnt]);
1955                 if (ret)
1956                         goto over_quota;
1957                 ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1958                 if (ret)
1959                         goto over_quota;
1960         }
1961
1962         /*
1963          * Finally perform the needed transfer from transfer_from to transfer_to
1964          */
1965         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1966                 if (!is_valid[cnt])
1967                         continue;
1968                 /* Due to IO error we might not have transfer_from[] structure */
1969                 if (transfer_from[cnt]) {
1970                         int wtype;
1971                         wtype = info_idq_free(transfer_from[cnt], inode_usage);
1972                         if (wtype != QUOTA_NL_NOWARN)
1973                                 prepare_warning(&warn_from_inodes[cnt],
1974                                                 transfer_from[cnt], wtype);
1975                         wtype = info_bdq_free(transfer_from[cnt], space);
1976                         if (wtype != QUOTA_NL_NOWARN)
1977                                 prepare_warning(&warn_from_space[cnt],
1978                                                 transfer_from[cnt], wtype);
1979                         dquot_decr_inodes(transfer_from[cnt], inode_usage);
1980                         dquot_decr_space(transfer_from[cnt], cur_space);
1981                         dquot_free_reserved_space(transfer_from[cnt],
1982                                                   rsv_space);
1983                 }
1984
1985                 dquot_incr_inodes(transfer_to[cnt], inode_usage);
1986                 dquot_incr_space(transfer_to[cnt], cur_space);
1987                 dquot_resv_space(transfer_to[cnt], rsv_space);
1988
1989                 i_dquot(inode)[cnt] = transfer_to[cnt];
1990         }
1991         spin_unlock(&dq_data_lock);
1992
1993         mark_all_dquot_dirty(transfer_from);
1994         mark_all_dquot_dirty(transfer_to);
1995         flush_warnings(warn_to);
1996         flush_warnings(warn_from_inodes);
1997         flush_warnings(warn_from_space);
1998         /* Pass back references to put */
1999         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2000                 if (is_valid[cnt])
2001                         transfer_to[cnt] = transfer_from[cnt];
2002         return 0;
2003 over_quota:
2004         spin_unlock(&dq_data_lock);
2005         flush_warnings(warn_to);
2006         return ret;
2007 }
2008 EXPORT_SYMBOL(__dquot_transfer);
2009
2010 /* Wrapper for transferring ownership of an inode for uid/gid only
2011  * Called from FSXXX_setattr()
2012  */
2013 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2014 {
2015         struct dquot *transfer_to[MAXQUOTAS] = {};
2016         struct dquot *dquot;
2017         struct super_block *sb = inode->i_sb;
2018         int ret;
2019
2020         if (!dquot_active(inode))
2021                 return 0;
2022
2023         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2024                 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2025                 if (IS_ERR(dquot)) {
2026                         if (PTR_ERR(dquot) != -ESRCH) {
2027                                 ret = PTR_ERR(dquot);
2028                                 goto out_put;
2029                         }
2030                         dquot = NULL;
2031                 }
2032                 transfer_to[USRQUOTA] = dquot;
2033         }
2034         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2035                 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2036                 if (IS_ERR(dquot)) {
2037                         if (PTR_ERR(dquot) != -ESRCH) {
2038                                 ret = PTR_ERR(dquot);
2039                                 goto out_put;
2040                         }
2041                         dquot = NULL;
2042                 }
2043                 transfer_to[GRPQUOTA] = dquot;
2044         }
2045         ret = __dquot_transfer(inode, transfer_to);
2046 out_put:
2047         dqput_all(transfer_to);
2048         return ret;
2049 }
2050 EXPORT_SYMBOL(dquot_transfer);
2051
2052 /*
2053  * Write info of quota file to disk
2054  */
2055 int dquot_commit_info(struct super_block *sb, int type)
2056 {
2057         int ret;
2058         struct quota_info *dqopt = sb_dqopt(sb);
2059
2060         down_write(&dqopt->dqio_sem);
2061         ret = dqopt->ops[type]->write_file_info(sb, type);
2062         up_write(&dqopt->dqio_sem);
2063         return ret;
2064 }
2065 EXPORT_SYMBOL(dquot_commit_info);
2066
2067 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2068 {
2069         struct quota_info *dqopt = sb_dqopt(sb);
2070         int err;
2071
2072         if (!sb_has_quota_active(sb, qid->type))
2073                 return -ESRCH;
2074         if (!dqopt->ops[qid->type]->get_next_id)
2075                 return -ENOSYS;
2076         down_read(&dqopt->dqio_sem);
2077         err = dqopt->ops[qid->type]->get_next_id(sb, qid);
2078         up_read(&dqopt->dqio_sem);
2079         return err;
2080 }
2081 EXPORT_SYMBOL(dquot_get_next_id);
2082
2083 /*
2084  * Definitions of diskquota operations.
2085  */
2086 const struct dquot_operations dquot_operations = {
2087         .write_dquot    = dquot_commit,
2088         .acquire_dquot  = dquot_acquire,
2089         .release_dquot  = dquot_release,
2090         .mark_dirty     = dquot_mark_dquot_dirty,
2091         .write_info     = dquot_commit_info,
2092         .alloc_dquot    = dquot_alloc,
2093         .destroy_dquot  = dquot_destroy,
2094         .get_next_id    = dquot_get_next_id,
2095 };
2096 EXPORT_SYMBOL(dquot_operations);
2097
2098 /*
2099  * Generic helper for ->open on filesystems supporting disk quotas.
2100  */
2101 int dquot_file_open(struct inode *inode, struct file *file)
2102 {
2103         int error;
2104
2105         error = generic_file_open(inode, file);
2106         if (!error && (file->f_mode & FMODE_WRITE))
2107                 dquot_initialize(inode);
2108         return error;
2109 }
2110 EXPORT_SYMBOL(dquot_file_open);
2111
2112 /*
2113  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2114  */
2115 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2116 {
2117         int cnt, ret = 0;
2118         struct quota_info *dqopt = sb_dqopt(sb);
2119         struct inode *toputinode[MAXQUOTAS];
2120
2121         /* s_umount should be held in exclusive mode */
2122         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2123                 up_read(&sb->s_umount);
2124
2125         /* Cannot turn off usage accounting without turning off limits, or
2126          * suspend quotas and simultaneously turn quotas off. */
2127         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2128             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2129             DQUOT_USAGE_ENABLED)))
2130                 return -EINVAL;
2131
2132         /*
2133          * Skip everything if there's nothing to do. We have to do this because
2134          * sometimes we are called when fill_super() failed and calling
2135          * sync_fs() in such cases does no good.
2136          */
2137         if (!sb_any_quota_loaded(sb))
2138                 return 0;
2139
2140         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2141                 toputinode[cnt] = NULL;
2142                 if (type != -1 && cnt != type)
2143                         continue;
2144                 if (!sb_has_quota_loaded(sb, cnt))
2145                         continue;
2146
2147                 if (flags & DQUOT_SUSPENDED) {
2148                         spin_lock(&dq_state_lock);
2149                         dqopt->flags |=
2150                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2151                         spin_unlock(&dq_state_lock);
2152                 } else {
2153                         spin_lock(&dq_state_lock);
2154                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2155                         /* Turning off suspended quotas? */
2156                         if (!sb_has_quota_loaded(sb, cnt) &&
2157                             sb_has_quota_suspended(sb, cnt)) {
2158                                 dqopt->flags &= ~dquot_state_flag(
2159                                                         DQUOT_SUSPENDED, cnt);
2160                                 spin_unlock(&dq_state_lock);
2161                                 iput(dqopt->files[cnt]);
2162                                 dqopt->files[cnt] = NULL;
2163                                 continue;
2164                         }
2165                         spin_unlock(&dq_state_lock);
2166                 }
2167
2168                 /* We still have to keep quota loaded? */
2169                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2170                         continue;
2171
2172                 /* Note: these are blocking operations */
2173                 drop_dquot_ref(sb, cnt);
2174                 invalidate_dquots(sb, cnt);
2175                 /*
2176                  * Now all dquots should be invalidated, all writes done so we
2177                  * should be only users of the info. No locks needed.
2178                  */
2179                 if (info_dirty(&dqopt->info[cnt]))
2180                         sb->dq_op->write_info(sb, cnt);
2181                 if (dqopt->ops[cnt]->free_file_info)
2182                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2183                 put_quota_format(dqopt->info[cnt].dqi_format);
2184
2185                 toputinode[cnt] = dqopt->files[cnt];
2186                 if (!sb_has_quota_loaded(sb, cnt))
2187                         dqopt->files[cnt] = NULL;
2188                 dqopt->info[cnt].dqi_flags = 0;
2189                 dqopt->info[cnt].dqi_igrace = 0;
2190                 dqopt->info[cnt].dqi_bgrace = 0;
2191                 dqopt->ops[cnt] = NULL;
2192         }
2193
2194         /* Skip syncing and setting flags if quota files are hidden */
2195         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2196                 goto put_inodes;
2197
2198         /* Sync the superblock so that buffers with quota data are written to
2199          * disk (and so userspace sees correct data afterwards). */
2200         if (sb->s_op->sync_fs)
2201                 sb->s_op->sync_fs(sb, 1);
2202         sync_blockdev(sb->s_bdev);
2203         /* Now the quota files are just ordinary files and we can set the
2204          * inode flags back. Moreover we discard the pagecache so that
2205          * userspace sees the writes we did bypassing the pagecache. We
2206          * must also discard the blockdev buffers so that we see the
2207          * changes done by userspace on the next quotaon() */
2208         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2209                 /* This can happen when suspending quotas on remount-ro... */
2210                 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2211                         inode_lock(toputinode[cnt]);
2212                         toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2213                         truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2214                         inode_unlock(toputinode[cnt]);
2215                         mark_inode_dirty_sync(toputinode[cnt]);
2216                 }
2217         if (sb->s_bdev)
2218                 invalidate_bdev(sb->s_bdev);
2219 put_inodes:
2220         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2221                 if (toputinode[cnt]) {
2222                         /* On remount RO, we keep the inode pointer so that we
2223                          * can reenable quota on the subsequent remount RW. We
2224                          * have to check 'flags' variable and not use sb_has_
2225                          * function because another quotaon / quotaoff could
2226                          * change global state before we got here. We refuse
2227                          * to suspend quotas when there is pending delete on
2228                          * the quota file... */
2229                         if (!(flags & DQUOT_SUSPENDED))
2230                                 iput(toputinode[cnt]);
2231                         else if (!toputinode[cnt]->i_nlink)
2232                                 ret = -EBUSY;
2233                 }
2234         return ret;
2235 }
2236 EXPORT_SYMBOL(dquot_disable);
2237
2238 int dquot_quota_off(struct super_block *sb, int type)
2239 {
2240         return dquot_disable(sb, type,
2241                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2242 }
2243 EXPORT_SYMBOL(dquot_quota_off);
2244
2245 /*
2246  *      Turn quotas on on a device
2247  */
2248
2249 /*
2250  * Helper function to turn quotas on when we already have the inode of
2251  * quota file and no quota information is loaded.
2252  */
2253 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2254         unsigned int flags)
2255 {
2256         struct quota_format_type *fmt = find_quota_format(format_id);
2257         struct super_block *sb = inode->i_sb;
2258         struct quota_info *dqopt = sb_dqopt(sb);
2259         int error;
2260
2261         if (!fmt)
2262                 return -ESRCH;
2263         if (!S_ISREG(inode->i_mode)) {
2264                 error = -EACCES;
2265                 goto out_fmt;
2266         }
2267         if (IS_RDONLY(inode)) {
2268                 error = -EROFS;
2269                 goto out_fmt;
2270         }
2271         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2272             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2273                 error = -EINVAL;
2274                 goto out_fmt;
2275         }
2276         /* Filesystems outside of init_user_ns not yet supported */
2277         if (sb->s_user_ns != &init_user_ns) {
2278                 error = -EINVAL;
2279                 goto out_fmt;
2280         }
2281         /* Usage always has to be set... */
2282         if (!(flags & DQUOT_USAGE_ENABLED)) {
2283                 error = -EINVAL;
2284                 goto out_fmt;
2285         }
2286         if (sb_has_quota_loaded(sb, type)) {
2287                 error = -EBUSY;
2288                 goto out_fmt;
2289         }
2290
2291         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2292                 /* As we bypass the pagecache we must now flush all the
2293                  * dirty data and invalidate caches so that kernel sees
2294                  * changes from userspace. It is not enough to just flush
2295                  * the quota file since if blocksize < pagesize, invalidation
2296                  * of the cache could fail because of other unrelated dirty
2297                  * data */
2298                 sync_filesystem(sb);
2299                 invalidate_bdev(sb->s_bdev);
2300         }
2301
2302         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2303                 /* We don't want quota and atime on quota files (deadlocks
2304                  * possible) Also nobody should write to the file - we use
2305                  * special IO operations which ignore the immutable bit. */
2306                 inode_lock(inode);
2307                 inode->i_flags |= S_NOQUOTA;
2308                 inode_unlock(inode);
2309                 /*
2310                  * When S_NOQUOTA is set, remove dquot references as no more
2311                  * references can be added
2312                  */
2313                 __dquot_drop(inode);
2314         }
2315
2316         error = -EIO;
2317         dqopt->files[type] = igrab(inode);
2318         if (!dqopt->files[type])
2319                 goto out_file_flags;
2320         error = -EINVAL;
2321         if (!fmt->qf_ops->check_quota_file(sb, type))
2322                 goto out_file_init;
2323
2324         dqopt->ops[type] = fmt->qf_ops;
2325         dqopt->info[type].dqi_format = fmt;
2326         dqopt->info[type].dqi_fmt_id = format_id;
2327         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2328         down_read(&dqopt->dqio_sem);
2329         error = dqopt->ops[type]->read_file_info(sb, type);
2330         if (error < 0) {
2331                 up_read(&dqopt->dqio_sem);
2332                 goto out_file_init;
2333         }
2334         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2335                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2336         up_read(&dqopt->dqio_sem);
2337         spin_lock(&dq_state_lock);
2338         dqopt->flags |= dquot_state_flag(flags, type);
2339         spin_unlock(&dq_state_lock);
2340
2341         add_dquot_ref(sb, type);
2342
2343         return 0;
2344
2345 out_file_init:
2346         dqopt->files[type] = NULL;
2347         iput(inode);
2348 out_file_flags:
2349         inode_lock(inode);
2350         inode->i_flags &= ~S_NOQUOTA;
2351         inode_unlock(inode);
2352 out_fmt:
2353         put_quota_format(fmt);
2354
2355         return error; 
2356 }
2357
2358 /* Reenable quotas on remount RW */
2359 int dquot_resume(struct super_block *sb, int type)
2360 {
2361         struct quota_info *dqopt = sb_dqopt(sb);
2362         struct inode *inode;
2363         int ret = 0, cnt;
2364         unsigned int flags;
2365
2366         /* s_umount should be held in exclusive mode */
2367         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2368                 up_read(&sb->s_umount);
2369
2370         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2371                 if (type != -1 && cnt != type)
2372                         continue;
2373                 if (!sb_has_quota_suspended(sb, cnt))
2374                         continue;
2375
2376                 inode = dqopt->files[cnt];
2377                 dqopt->files[cnt] = NULL;
2378                 spin_lock(&dq_state_lock);
2379                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2380                                                         DQUOT_LIMITS_ENABLED,
2381                                                         cnt);
2382                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2383                 spin_unlock(&dq_state_lock);
2384
2385                 flags = dquot_generic_flag(flags, cnt);
2386                 ret = vfs_load_quota_inode(inode, cnt,
2387                                 dqopt->info[cnt].dqi_fmt_id, flags);
2388                 iput(inode);
2389         }
2390
2391         return ret;
2392 }
2393 EXPORT_SYMBOL(dquot_resume);
2394
2395 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2396                    const struct path *path)
2397 {
2398         int error = security_quota_on(path->dentry);
2399         if (error)
2400                 return error;
2401         /* Quota file not on the same filesystem? */
2402         if (path->dentry->d_sb != sb)
2403                 error = -EXDEV;
2404         else
2405                 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2406                                              format_id, DQUOT_USAGE_ENABLED |
2407                                              DQUOT_LIMITS_ENABLED);
2408         return error;
2409 }
2410 EXPORT_SYMBOL(dquot_quota_on);
2411
2412 /*
2413  * More powerful function for turning on quotas allowing setting
2414  * of individual quota flags
2415  */
2416 int dquot_enable(struct inode *inode, int type, int format_id,
2417                  unsigned int flags)
2418 {
2419         struct super_block *sb = inode->i_sb;
2420
2421         /* Just unsuspend quotas? */
2422         BUG_ON(flags & DQUOT_SUSPENDED);
2423         /* s_umount should be held in exclusive mode */
2424         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2425                 up_read(&sb->s_umount);
2426
2427         if (!flags)
2428                 return 0;
2429         /* Just updating flags needed? */
2430         if (sb_has_quota_loaded(sb, type)) {
2431                 if (flags & DQUOT_USAGE_ENABLED &&
2432                     sb_has_quota_usage_enabled(sb, type))
2433                         return -EBUSY;
2434                 if (flags & DQUOT_LIMITS_ENABLED &&
2435                     sb_has_quota_limits_enabled(sb, type))
2436                         return -EBUSY;
2437                 spin_lock(&dq_state_lock);
2438                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2439                 spin_unlock(&dq_state_lock);
2440                 return 0;
2441         }
2442
2443         return vfs_load_quota_inode(inode, type, format_id, flags);
2444 }
2445 EXPORT_SYMBOL(dquot_enable);
2446
2447 /*
2448  * This function is used when filesystem needs to initialize quotas
2449  * during mount time.
2450  */
2451 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2452                 int format_id, int type)
2453 {
2454         struct dentry *dentry;
2455         int error;
2456
2457         dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2458         if (IS_ERR(dentry))
2459                 return PTR_ERR(dentry);
2460
2461         if (d_really_is_negative(dentry)) {
2462                 error = -ENOENT;
2463                 goto out;
2464         }
2465
2466         error = security_quota_on(dentry);
2467         if (!error)
2468                 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2469                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2470
2471 out:
2472         dput(dentry);
2473         return error;
2474 }
2475 EXPORT_SYMBOL(dquot_quota_on_mount);
2476
2477 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2478 {
2479         int ret;
2480         int type;
2481         struct quota_info *dqopt = sb_dqopt(sb);
2482
2483         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2484                 return -ENOSYS;
2485         /* Accounting cannot be turned on while fs is mounted */
2486         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2487         if (!flags)
2488                 return -EINVAL;
2489         for (type = 0; type < MAXQUOTAS; type++) {
2490                 if (!(flags & qtype_enforce_flag(type)))
2491                         continue;
2492                 /* Can't enforce without accounting */
2493                 if (!sb_has_quota_usage_enabled(sb, type))
2494                         return -EINVAL;
2495                 ret = dquot_enable(dqopt->files[type], type,
2496                                    dqopt->info[type].dqi_fmt_id,
2497                                    DQUOT_LIMITS_ENABLED);
2498                 if (ret < 0)
2499                         goto out_err;
2500         }
2501         return 0;
2502 out_err:
2503         /* Backout enforcement enablement we already did */
2504         for (type--; type >= 0; type--)  {
2505                 if (flags & qtype_enforce_flag(type))
2506                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2507         }
2508         /* Error code translation for better compatibility with XFS */
2509         if (ret == -EBUSY)
2510                 ret = -EEXIST;
2511         return ret;
2512 }
2513
2514 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2515 {
2516         int ret;
2517         int type;
2518         struct quota_info *dqopt = sb_dqopt(sb);
2519
2520         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2521                 return -ENOSYS;
2522         /*
2523          * We don't support turning off accounting via quotactl. In principle
2524          * quota infrastructure can do this but filesystems don't expect
2525          * userspace to be able to do it.
2526          */
2527         if (flags &
2528                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2529                 return -EOPNOTSUPP;
2530
2531         /* Filter out limits not enabled */
2532         for (type = 0; type < MAXQUOTAS; type++)
2533                 if (!sb_has_quota_limits_enabled(sb, type))
2534                         flags &= ~qtype_enforce_flag(type);
2535         /* Nothing left? */
2536         if (!flags)
2537                 return -EEXIST;
2538         for (type = 0; type < MAXQUOTAS; type++) {
2539                 if (flags & qtype_enforce_flag(type)) {
2540                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2541                         if (ret < 0)
2542                                 goto out_err;
2543                 }
2544         }
2545         return 0;
2546 out_err:
2547         /* Backout enforcement disabling we already did */
2548         for (type--; type >= 0; type--)  {
2549                 if (flags & qtype_enforce_flag(type))
2550                         dquot_enable(dqopt->files[type], type,
2551                                      dqopt->info[type].dqi_fmt_id,
2552                                      DQUOT_LIMITS_ENABLED);
2553         }
2554         return ret;
2555 }
2556
2557 /* Generic routine for getting common part of quota structure */
2558 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2559 {
2560         struct mem_dqblk *dm = &dquot->dq_dqb;
2561
2562         memset(di, 0, sizeof(*di));
2563         spin_lock(&dq_data_lock);
2564         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2565         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2566         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2567         di->d_ino_softlimit = dm->dqb_isoftlimit;
2568         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2569         di->d_ino_count = dm->dqb_curinodes;
2570         di->d_spc_timer = dm->dqb_btime;
2571         di->d_ino_timer = dm->dqb_itime;
2572         spin_unlock(&dq_data_lock);
2573 }
2574
2575 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2576                     struct qc_dqblk *di)
2577 {
2578         struct dquot *dquot;
2579
2580         dquot = dqget(sb, qid);
2581         if (IS_ERR(dquot))
2582                 return PTR_ERR(dquot);
2583         do_get_dqblk(dquot, di);
2584         dqput(dquot);
2585
2586         return 0;
2587 }
2588 EXPORT_SYMBOL(dquot_get_dqblk);
2589
2590 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2591                          struct qc_dqblk *di)
2592 {
2593         struct dquot *dquot;
2594         int err;
2595
2596         if (!sb->dq_op->get_next_id)
2597                 return -ENOSYS;
2598         err = sb->dq_op->get_next_id(sb, qid);
2599         if (err < 0)
2600                 return err;
2601         dquot = dqget(sb, *qid);
2602         if (IS_ERR(dquot))
2603                 return PTR_ERR(dquot);
2604         do_get_dqblk(dquot, di);
2605         dqput(dquot);
2606
2607         return 0;
2608 }
2609 EXPORT_SYMBOL(dquot_get_next_dqblk);
2610
2611 #define VFS_QC_MASK \
2612         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2613          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2614          QC_SPC_TIMER | QC_INO_TIMER)
2615
2616 /* Generic routine for setting common part of quota structure */
2617 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2618 {
2619         struct mem_dqblk *dm = &dquot->dq_dqb;
2620         int check_blim = 0, check_ilim = 0;
2621         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2622
2623         if (di->d_fieldmask & ~VFS_QC_MASK)
2624                 return -EINVAL;
2625
2626         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2627              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2628             ((di->d_fieldmask & QC_SPC_HARD) &&
2629              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2630             ((di->d_fieldmask & QC_INO_SOFT) &&
2631              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2632             ((di->d_fieldmask & QC_INO_HARD) &&
2633              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2634                 return -ERANGE;
2635
2636         spin_lock(&dq_data_lock);
2637         if (di->d_fieldmask & QC_SPACE) {
2638                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2639                 check_blim = 1;
2640                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2641         }
2642
2643         if (di->d_fieldmask & QC_SPC_SOFT)
2644                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2645         if (di->d_fieldmask & QC_SPC_HARD)
2646                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2647         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2648                 check_blim = 1;
2649                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2650         }
2651
2652         if (di->d_fieldmask & QC_INO_COUNT) {
2653                 dm->dqb_curinodes = di->d_ino_count;
2654                 check_ilim = 1;
2655                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2656         }
2657
2658         if (di->d_fieldmask & QC_INO_SOFT)
2659                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2660         if (di->d_fieldmask & QC_INO_HARD)
2661                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2662         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2663                 check_ilim = 1;
2664                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2665         }
2666
2667         if (di->d_fieldmask & QC_SPC_TIMER) {
2668                 dm->dqb_btime = di->d_spc_timer;
2669                 check_blim = 1;
2670                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2671         }
2672
2673         if (di->d_fieldmask & QC_INO_TIMER) {
2674                 dm->dqb_itime = di->d_ino_timer;
2675                 check_ilim = 1;
2676                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2677         }
2678
2679         if (check_blim) {
2680                 if (!dm->dqb_bsoftlimit ||
2681                     dm->dqb_curspace < dm->dqb_bsoftlimit) {
2682                         dm->dqb_btime = 0;
2683                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2684                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2685                         /* Set grace only if user hasn't provided his own... */
2686                         dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2687         }
2688         if (check_ilim) {
2689                 if (!dm->dqb_isoftlimit ||
2690                     dm->dqb_curinodes < dm->dqb_isoftlimit) {
2691                         dm->dqb_itime = 0;
2692                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2693                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2694                         /* Set grace only if user hasn't provided his own... */
2695                         dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2696         }
2697         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2698             dm->dqb_isoftlimit)
2699                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2700         else
2701                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2702         spin_unlock(&dq_data_lock);
2703         mark_dquot_dirty(dquot);
2704
2705         return 0;
2706 }
2707
2708 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2709                   struct qc_dqblk *di)
2710 {
2711         struct dquot *dquot;
2712         int rc;
2713
2714         dquot = dqget(sb, qid);
2715         if (IS_ERR(dquot)) {
2716                 rc = PTR_ERR(dquot);
2717                 goto out;
2718         }
2719         rc = do_set_dqblk(dquot, di);
2720         dqput(dquot);
2721 out:
2722         return rc;
2723 }
2724 EXPORT_SYMBOL(dquot_set_dqblk);
2725
2726 /* Generic routine for getting common part of quota file information */
2727 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2728 {
2729         struct mem_dqinfo *mi;
2730         struct qc_type_state *tstate;
2731         struct quota_info *dqopt = sb_dqopt(sb);
2732         int type;
2733   
2734         memset(state, 0, sizeof(*state));
2735         for (type = 0; type < MAXQUOTAS; type++) {
2736                 if (!sb_has_quota_active(sb, type))
2737                         continue;
2738                 tstate = state->s_state + type;
2739                 mi = sb_dqopt(sb)->info + type;
2740                 tstate->flags = QCI_ACCT_ENABLED;
2741                 spin_lock(&dq_data_lock);
2742                 if (mi->dqi_flags & DQF_SYS_FILE)
2743                         tstate->flags |= QCI_SYSFILE;
2744                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2745                         tstate->flags |= QCI_ROOT_SQUASH;
2746                 if (sb_has_quota_limits_enabled(sb, type))
2747                         tstate->flags |= QCI_LIMITS_ENFORCED;
2748                 tstate->spc_timelimit = mi->dqi_bgrace;
2749                 tstate->ino_timelimit = mi->dqi_igrace;
2750                 tstate->ino = dqopt->files[type]->i_ino;
2751                 tstate->blocks = dqopt->files[type]->i_blocks;
2752                 tstate->nextents = 1;   /* We don't know... */
2753                 spin_unlock(&dq_data_lock);
2754         }
2755         return 0;
2756 }
2757 EXPORT_SYMBOL(dquot_get_state);
2758
2759 /* Generic routine for setting common part of quota file information */
2760 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2761 {
2762         struct mem_dqinfo *mi;
2763         int err = 0;
2764
2765         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2766             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2767                 return -EINVAL;
2768         if (!sb_has_quota_active(sb, type))
2769                 return -ESRCH;
2770         mi = sb_dqopt(sb)->info + type;
2771         if (ii->i_fieldmask & QC_FLAGS) {
2772                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2773                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2774                         return -EINVAL;
2775         }
2776         spin_lock(&dq_data_lock);
2777         if (ii->i_fieldmask & QC_SPC_TIMER)
2778                 mi->dqi_bgrace = ii->i_spc_timelimit;
2779         if (ii->i_fieldmask & QC_INO_TIMER)
2780                 mi->dqi_igrace = ii->i_ino_timelimit;
2781         if (ii->i_fieldmask & QC_FLAGS) {
2782                 if (ii->i_flags & QCI_ROOT_SQUASH)
2783                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2784                 else
2785                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2786         }
2787         spin_unlock(&dq_data_lock);
2788         mark_info_dirty(sb, type);
2789         /* Force write to disk */
2790         sb->dq_op->write_info(sb, type);
2791         return err;
2792 }
2793 EXPORT_SYMBOL(dquot_set_dqinfo);
2794
2795 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2796         .quota_enable   = dquot_quota_enable,
2797         .quota_disable  = dquot_quota_disable,
2798         .quota_sync     = dquot_quota_sync,
2799         .get_state      = dquot_get_state,
2800         .set_info       = dquot_set_dqinfo,
2801         .get_dqblk      = dquot_get_dqblk,
2802         .get_nextdqblk  = dquot_get_next_dqblk,
2803         .set_dqblk      = dquot_set_dqblk
2804 };
2805 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2806
2807 static int do_proc_dqstats(struct ctl_table *table, int write,
2808                      void __user *buffer, size_t *lenp, loff_t *ppos)
2809 {
2810         unsigned int type = (int *)table->data - dqstats.stat;
2811
2812         /* Update global table */
2813         dqstats.stat[type] =
2814                         percpu_counter_sum_positive(&dqstats.counter[type]);
2815         return proc_dointvec(table, write, buffer, lenp, ppos);
2816 }
2817
2818 static struct ctl_table fs_dqstats_table[] = {
2819         {
2820                 .procname       = "lookups",
2821                 .data           = &dqstats.stat[DQST_LOOKUPS],
2822                 .maxlen         = sizeof(int),
2823                 .mode           = 0444,
2824                 .proc_handler   = do_proc_dqstats,
2825         },
2826         {
2827                 .procname       = "drops",
2828                 .data           = &dqstats.stat[DQST_DROPS],
2829                 .maxlen         = sizeof(int),
2830                 .mode           = 0444,
2831                 .proc_handler   = do_proc_dqstats,
2832         },
2833         {
2834                 .procname       = "reads",
2835                 .data           = &dqstats.stat[DQST_READS],
2836                 .maxlen         = sizeof(int),
2837                 .mode           = 0444,
2838                 .proc_handler   = do_proc_dqstats,
2839         },
2840         {
2841                 .procname       = "writes",
2842                 .data           = &dqstats.stat[DQST_WRITES],
2843                 .maxlen         = sizeof(int),
2844                 .mode           = 0444,
2845                 .proc_handler   = do_proc_dqstats,
2846         },
2847         {
2848                 .procname       = "cache_hits",
2849                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2850                 .maxlen         = sizeof(int),
2851                 .mode           = 0444,
2852                 .proc_handler   = do_proc_dqstats,
2853         },
2854         {
2855                 .procname       = "allocated_dquots",
2856                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2857                 .maxlen         = sizeof(int),
2858                 .mode           = 0444,
2859                 .proc_handler   = do_proc_dqstats,
2860         },
2861         {
2862                 .procname       = "free_dquots",
2863                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2864                 .maxlen         = sizeof(int),
2865                 .mode           = 0444,
2866                 .proc_handler   = do_proc_dqstats,
2867         },
2868         {
2869                 .procname       = "syncs",
2870                 .data           = &dqstats.stat[DQST_SYNCS],
2871                 .maxlen         = sizeof(int),
2872                 .mode           = 0444,
2873                 .proc_handler   = do_proc_dqstats,
2874         },
2875 #ifdef CONFIG_PRINT_QUOTA_WARNING
2876         {
2877                 .procname       = "warnings",
2878                 .data           = &flag_print_warnings,
2879                 .maxlen         = sizeof(int),
2880                 .mode           = 0644,
2881                 .proc_handler   = proc_dointvec,
2882         },
2883 #endif
2884         { },
2885 };
2886
2887 static struct ctl_table fs_table[] = {
2888         {
2889                 .procname       = "quota",
2890                 .mode           = 0555,
2891                 .child          = fs_dqstats_table,
2892         },
2893         { },
2894 };
2895
2896 static struct ctl_table sys_table[] = {
2897         {
2898                 .procname       = "fs",
2899                 .mode           = 0555,
2900                 .child          = fs_table,
2901         },
2902         { },
2903 };
2904
2905 static int __init dquot_init(void)
2906 {
2907         int i, ret;
2908         unsigned long nr_hash, order;
2909
2910         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2911
2912         register_sysctl_table(sys_table);
2913
2914         dquot_cachep = kmem_cache_create("dquot",
2915                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2916                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2917                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2918                         NULL);
2919
2920         order = 0;
2921         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2922         if (!dquot_hash)
2923                 panic("Cannot create dquot hash table");
2924
2925         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2926                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2927                 if (ret)
2928                         panic("Cannot create dquot stat counters");
2929         }
2930
2931         /* Find power-of-two hlist_heads which can fit into allocation */
2932         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2933         dq_hash_bits = 0;
2934         do {
2935                 dq_hash_bits++;
2936         } while (nr_hash >> dq_hash_bits);
2937         dq_hash_bits--;
2938
2939         nr_hash = 1UL << dq_hash_bits;
2940         dq_hash_mask = nr_hash - 1;
2941         for (i = 0; i < nr_hash; i++)
2942                 INIT_HLIST_HEAD(dquot_hash + i);
2943
2944         pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2945                 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2946
2947         register_shrinker(&dqcache_shrinker);
2948
2949         return 0;
2950 }
2951 fs_initcall(dquot_init);