Merge branch 'topic/cs35l41' into for-next
[linux-2.6-microblaze.git] / fs / quota / dquot.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Implementation of the diskquota system for the LINUX operating system. QUOTA
4  * is implemented using the BSD system call interface as the means of
5  * communication with the user level. This file contains the generic routines
6  * called by the different filesystems on allocation of an inode or block.
7  * These routines take care of the administration needed to have a consistent
8  * diskquota tracking system. The ideas of both user and group quotas are based
9  * on the Melbourne quota system as used on BSD derived systems. The internal
10  * implementation is based on one of the several variants of the LINUX
11  * inode-subsystem with added complexity of the diskquota system.
12  *
13  * Author:      Marco van Wieringen <mvw@planets.elm.net>
14  *
15  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16  *
17  *              Revised list management to avoid races
18  *              -- Bill Hawes, <whawes@star.net>, 9/98
19  *
20  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
21  *              As the consequence the locking was moved from dquot_decr_...(),
22  *              dquot_incr_...() to calling functions.
23  *              invalidate_dquots() now writes modified dquots.
24  *              Serialized quota_off() and quota_on() for mount point.
25  *              Fixed a few bugs in grow_dquots().
26  *              Fixed deadlock in write_dquot() - we no longer account quotas on
27  *              quota files
28  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
29  *              add_dquot_ref() restarts after blocking
30  *              Added check for bogus uid and fixed check for group in quotactl.
31  *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32  *
33  *              Used struct list_head instead of own list struct
34  *              Invalidation of referenced dquots is no longer possible
35  *              Improved free_dquots list management
36  *              Quota and i_blocks are now updated in one place to avoid races
37  *              Warnings are now delayed so we won't block in critical section
38  *              Write updated not to require dquot lock
39  *              Jan Kara, <jack@suse.cz>, 9/2000
40  *
41  *              Added dynamic quota structure allocation
42  *              Jan Kara <jack@suse.cz> 12/2000
43  *
44  *              Rewritten quota interface. Implemented new quota format and
45  *              formats registering.
46  *              Jan Kara, <jack@suse.cz>, 2001,2002
47  *
48  *              New SMP locking.
49  *              Jan Kara, <jack@suse.cz>, 10/2002
50  *
51  *              Added journalled quota support, fix lock inversion problems
52  *              Jan Kara, <jack@suse.cz>, 2003,2004
53  *
54  * (C) Copyright 1994 - 1997 Marco van Wieringen
55  */
56
57 #include <linux/errno.h>
58 #include <linux/kernel.h>
59 #include <linux/fs.h>
60 #include <linux/mount.h>
61 #include <linux/mm.h>
62 #include <linux/time.h>
63 #include <linux/types.h>
64 #include <linux/string.h>
65 #include <linux/fcntl.h>
66 #include <linux/stat.h>
67 #include <linux/tty.h>
68 #include <linux/file.h>
69 #include <linux/slab.h>
70 #include <linux/sysctl.h>
71 #include <linux/init.h>
72 #include <linux/module.h>
73 #include <linux/proc_fs.h>
74 #include <linux/security.h>
75 #include <linux/sched.h>
76 #include <linux/cred.h>
77 #include <linux/kmod.h>
78 #include <linux/namei.h>
79 #include <linux/capability.h>
80 #include <linux/quotaops.h>
81 #include <linux/blkdev.h>
82 #include "../internal.h" /* ugh */
83
84 #include <linux/uaccess.h>
85
86 /*
87  * There are five quota SMP locks:
88  * * dq_list_lock protects all lists with quotas and quota formats.
89  * * dquot->dq_dqb_lock protects data from dq_dqb
90  * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
91  *   consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
92  *   dquot_transfer() can stabilize amount it transfers
93  * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
94  *   pointers in the inode
95  * * dq_state_lock protects modifications of quota state (on quotaon and
96  *   quotaoff) and readers who care about latest values take it as well.
97  *
98  * The spinlock ordering is hence:
99  *   dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
100  *   dq_list_lock > dq_state_lock
101  *
102  * Note that some things (eg. sb pointer, type, id) doesn't change during
103  * the life of the dquot structure and so needn't to be protected by a lock
104  *
105  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
106  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
107  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
108  * inode and before dropping dquot references to avoid use of dquots after
109  * they are freed. dq_data_lock is used to serialize the pointer setting and
110  * clearing operations.
111  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
112  * inode is a quota file). Functions adding pointers from inode to dquots have
113  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
114  * have to do all pointer modifications before dropping dq_data_lock. This makes
115  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
116  * then drops all pointers to dquots from an inode.
117  *
118  * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
119  * memory (or space for it is being allocated) on the first dqget(), when it is
120  * being written out, and when it is being released on the last dqput(). The
121  * allocation and release operations are serialized by the dq_lock and by
122  * checking the use count in dquot_release().
123  *
124  * Lock ordering (including related VFS locks) is the following:
125  *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
126  */
127
128 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
129 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
130 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
131 EXPORT_SYMBOL(dq_data_lock);
132 DEFINE_STATIC_SRCU(dquot_srcu);
133
134 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
135
136 void __quota_error(struct super_block *sb, const char *func,
137                    const char *fmt, ...)
138 {
139         if (printk_ratelimit()) {
140                 va_list args;
141                 struct va_format vaf;
142
143                 va_start(args, fmt);
144
145                 vaf.fmt = fmt;
146                 vaf.va = &args;
147
148                 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
149                        sb->s_id, func, &vaf);
150
151                 va_end(args);
152         }
153 }
154 EXPORT_SYMBOL(__quota_error);
155
156 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
157 static char *quotatypes[] = INITQFNAMES;
158 #endif
159 static struct quota_format_type *quota_formats; /* List of registered formats */
160 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
161
162 /* SLAB cache for dquot structures */
163 static struct kmem_cache *dquot_cachep;
164
165 int register_quota_format(struct quota_format_type *fmt)
166 {
167         spin_lock(&dq_list_lock);
168         fmt->qf_next = quota_formats;
169         quota_formats = fmt;
170         spin_unlock(&dq_list_lock);
171         return 0;
172 }
173 EXPORT_SYMBOL(register_quota_format);
174
175 void unregister_quota_format(struct quota_format_type *fmt)
176 {
177         struct quota_format_type **actqf;
178
179         spin_lock(&dq_list_lock);
180         for (actqf = &quota_formats; *actqf && *actqf != fmt;
181              actqf = &(*actqf)->qf_next)
182                 ;
183         if (*actqf)
184                 *actqf = (*actqf)->qf_next;
185         spin_unlock(&dq_list_lock);
186 }
187 EXPORT_SYMBOL(unregister_quota_format);
188
189 static struct quota_format_type *find_quota_format(int id)
190 {
191         struct quota_format_type *actqf;
192
193         spin_lock(&dq_list_lock);
194         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
195              actqf = actqf->qf_next)
196                 ;
197         if (!actqf || !try_module_get(actqf->qf_owner)) {
198                 int qm;
199
200                 spin_unlock(&dq_list_lock);
201
202                 for (qm = 0; module_names[qm].qm_fmt_id &&
203                              module_names[qm].qm_fmt_id != id; qm++)
204                         ;
205                 if (!module_names[qm].qm_fmt_id ||
206                     request_module(module_names[qm].qm_mod_name))
207                         return NULL;
208
209                 spin_lock(&dq_list_lock);
210                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
211                      actqf = actqf->qf_next)
212                         ;
213                 if (actqf && !try_module_get(actqf->qf_owner))
214                         actqf = NULL;
215         }
216         spin_unlock(&dq_list_lock);
217         return actqf;
218 }
219
220 static void put_quota_format(struct quota_format_type *fmt)
221 {
222         module_put(fmt->qf_owner);
223 }
224
225 /*
226  * Dquot List Management:
227  * The quota code uses four lists for dquot management: the inuse_list,
228  * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
229  * structure may be on some of those lists, depending on its current state.
230  *
231  * All dquots are placed to the end of inuse_list when first created, and this
232  * list is used for invalidate operation, which must look at every dquot.
233  *
234  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
235  * and this list is searched whenever we need an available dquot.  Dquots are
236  * removed from the list as soon as they are used again, and
237  * dqstats.free_dquots gives the number of dquots on the list. When
238  * dquot is invalidated it's completely released from memory.
239  *
240  * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
241  * dirtied, and this list is searched when writing dirty dquots back to
242  * quota file. Note that some filesystems do dirty dquot tracking on their
243  * own (e.g. in a journal) and thus don't use dqi_dirty_list.
244  *
245  * Dquots with a specific identity (device, type and id) are placed on
246  * one of the dquot_hash[] hash chains. The provides an efficient search
247  * mechanism to locate a specific dquot.
248  */
249
250 static LIST_HEAD(inuse_list);
251 static LIST_HEAD(free_dquots);
252 static unsigned int dq_hash_bits, dq_hash_mask;
253 static struct hlist_head *dquot_hash;
254
255 struct dqstats dqstats;
256 EXPORT_SYMBOL(dqstats);
257
258 static qsize_t inode_get_rsv_space(struct inode *inode);
259 static qsize_t __inode_get_rsv_space(struct inode *inode);
260 static int __dquot_initialize(struct inode *inode, int type);
261
262 static inline unsigned int
263 hashfn(const struct super_block *sb, struct kqid qid)
264 {
265         unsigned int id = from_kqid(&init_user_ns, qid);
266         int type = qid.type;
267         unsigned long tmp;
268
269         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
270         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
271 }
272
273 /*
274  * Following list functions expect dq_list_lock to be held
275  */
276 static inline void insert_dquot_hash(struct dquot *dquot)
277 {
278         struct hlist_head *head;
279         head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
280         hlist_add_head(&dquot->dq_hash, head);
281 }
282
283 static inline void remove_dquot_hash(struct dquot *dquot)
284 {
285         hlist_del_init(&dquot->dq_hash);
286 }
287
288 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
289                                 struct kqid qid)
290 {
291         struct dquot *dquot;
292
293         hlist_for_each_entry(dquot, dquot_hash+hashent, dq_hash)
294                 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
295                         return dquot;
296
297         return NULL;
298 }
299
300 /* Add a dquot to the tail of the free list */
301 static inline void put_dquot_last(struct dquot *dquot)
302 {
303         list_add_tail(&dquot->dq_free, &free_dquots);
304         dqstats_inc(DQST_FREE_DQUOTS);
305 }
306
307 static inline void remove_free_dquot(struct dquot *dquot)
308 {
309         if (list_empty(&dquot->dq_free))
310                 return;
311         list_del_init(&dquot->dq_free);
312         dqstats_dec(DQST_FREE_DQUOTS);
313 }
314
315 static inline void put_inuse(struct dquot *dquot)
316 {
317         /* We add to the back of inuse list so we don't have to restart
318          * when traversing this list and we block */
319         list_add_tail(&dquot->dq_inuse, &inuse_list);
320         dqstats_inc(DQST_ALLOC_DQUOTS);
321 }
322
323 static inline void remove_inuse(struct dquot *dquot)
324 {
325         dqstats_dec(DQST_ALLOC_DQUOTS);
326         list_del(&dquot->dq_inuse);
327 }
328 /*
329  * End of list functions needing dq_list_lock
330  */
331
332 static void wait_on_dquot(struct dquot *dquot)
333 {
334         mutex_lock(&dquot->dq_lock);
335         mutex_unlock(&dquot->dq_lock);
336 }
337
338 static inline int dquot_dirty(struct dquot *dquot)
339 {
340         return test_bit(DQ_MOD_B, &dquot->dq_flags);
341 }
342
343 static inline int mark_dquot_dirty(struct dquot *dquot)
344 {
345         return dquot->dq_sb->dq_op->mark_dirty(dquot);
346 }
347
348 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
349 int dquot_mark_dquot_dirty(struct dquot *dquot)
350 {
351         int ret = 1;
352
353         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
354                 return 0;
355
356         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
357                 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
358
359         /* If quota is dirty already, we don't have to acquire dq_list_lock */
360         if (test_bit(DQ_MOD_B, &dquot->dq_flags))
361                 return 1;
362
363         spin_lock(&dq_list_lock);
364         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
365                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
366                                 info[dquot->dq_id.type].dqi_dirty_list);
367                 ret = 0;
368         }
369         spin_unlock(&dq_list_lock);
370         return ret;
371 }
372 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
373
374 /* Dirtify all the dquots - this can block when journalling */
375 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
376 {
377         int ret, err, cnt;
378
379         ret = err = 0;
380         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
381                 if (dquot[cnt])
382                         /* Even in case of error we have to continue */
383                         ret = mark_dquot_dirty(dquot[cnt]);
384                 if (!err)
385                         err = ret;
386         }
387         return err;
388 }
389
390 static inline void dqput_all(struct dquot **dquot)
391 {
392         unsigned int cnt;
393
394         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
395                 dqput(dquot[cnt]);
396 }
397
398 static inline int clear_dquot_dirty(struct dquot *dquot)
399 {
400         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
401                 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
402
403         spin_lock(&dq_list_lock);
404         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
405                 spin_unlock(&dq_list_lock);
406                 return 0;
407         }
408         list_del_init(&dquot->dq_dirty);
409         spin_unlock(&dq_list_lock);
410         return 1;
411 }
412
413 void mark_info_dirty(struct super_block *sb, int type)
414 {
415         spin_lock(&dq_data_lock);
416         sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
417         spin_unlock(&dq_data_lock);
418 }
419 EXPORT_SYMBOL(mark_info_dirty);
420
421 /*
422  *      Read dquot from disk and alloc space for it
423  */
424
425 int dquot_acquire(struct dquot *dquot)
426 {
427         int ret = 0, ret2 = 0;
428         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
429
430         mutex_lock(&dquot->dq_lock);
431         if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
432                 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
433                 if (ret < 0)
434                         goto out_iolock;
435         }
436         /* Make sure flags update is visible after dquot has been filled */
437         smp_mb__before_atomic();
438         set_bit(DQ_READ_B, &dquot->dq_flags);
439         /* Instantiate dquot if needed */
440         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
441                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
442                 /* Write the info if needed */
443                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
444                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
445                                         dquot->dq_sb, dquot->dq_id.type);
446                 }
447                 if (ret < 0)
448                         goto out_iolock;
449                 if (ret2 < 0) {
450                         ret = ret2;
451                         goto out_iolock;
452                 }
453         }
454         /*
455          * Make sure flags update is visible after on-disk struct has been
456          * allocated. Paired with smp_rmb() in dqget().
457          */
458         smp_mb__before_atomic();
459         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
460 out_iolock:
461         mutex_unlock(&dquot->dq_lock);
462         return ret;
463 }
464 EXPORT_SYMBOL(dquot_acquire);
465
466 /*
467  *      Write dquot to disk
468  */
469 int dquot_commit(struct dquot *dquot)
470 {
471         int ret = 0;
472         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
473
474         mutex_lock(&dquot->dq_lock);
475         if (!clear_dquot_dirty(dquot))
476                 goto out_lock;
477         /* Inactive dquot can be only if there was error during read/init
478          * => we have better not writing it */
479         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
480                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
481         else
482                 ret = -EIO;
483 out_lock:
484         mutex_unlock(&dquot->dq_lock);
485         return ret;
486 }
487 EXPORT_SYMBOL(dquot_commit);
488
489 /*
490  *      Release dquot
491  */
492 int dquot_release(struct dquot *dquot)
493 {
494         int ret = 0, ret2 = 0;
495         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
496
497         mutex_lock(&dquot->dq_lock);
498         /* Check whether we are not racing with some other dqget() */
499         if (dquot_is_busy(dquot))
500                 goto out_dqlock;
501         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
502                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
503                 /* Write the info */
504                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
505                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
506                                                 dquot->dq_sb, dquot->dq_id.type);
507                 }
508                 if (ret >= 0)
509                         ret = ret2;
510         }
511         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
512 out_dqlock:
513         mutex_unlock(&dquot->dq_lock);
514         return ret;
515 }
516 EXPORT_SYMBOL(dquot_release);
517
518 void dquot_destroy(struct dquot *dquot)
519 {
520         kmem_cache_free(dquot_cachep, dquot);
521 }
522 EXPORT_SYMBOL(dquot_destroy);
523
524 static inline void do_destroy_dquot(struct dquot *dquot)
525 {
526         dquot->dq_sb->dq_op->destroy_dquot(dquot);
527 }
528
529 /* Invalidate all dquots on the list. Note that this function is called after
530  * quota is disabled and pointers from inodes removed so there cannot be new
531  * quota users. There can still be some users of quotas due to inodes being
532  * just deleted or pruned by prune_icache() (those are not attached to any
533  * list) or parallel quotactl call. We have to wait for such users.
534  */
535 static void invalidate_dquots(struct super_block *sb, int type)
536 {
537         struct dquot *dquot, *tmp;
538
539 restart:
540         spin_lock(&dq_list_lock);
541         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
542                 if (dquot->dq_sb != sb)
543                         continue;
544                 if (dquot->dq_id.type != type)
545                         continue;
546                 /* Wait for dquot users */
547                 if (atomic_read(&dquot->dq_count)) {
548                         dqgrab(dquot);
549                         spin_unlock(&dq_list_lock);
550                         /*
551                          * Once dqput() wakes us up, we know it's time to free
552                          * the dquot.
553                          * IMPORTANT: we rely on the fact that there is always
554                          * at most one process waiting for dquot to free.
555                          * Otherwise dq_count would be > 1 and we would never
556                          * wake up.
557                          */
558                         wait_event(dquot_ref_wq,
559                                    atomic_read(&dquot->dq_count) == 1);
560                         dqput(dquot);
561                         /* At this moment dquot() need not exist (it could be
562                          * reclaimed by prune_dqcache(). Hence we must
563                          * restart. */
564                         goto restart;
565                 }
566                 /*
567                  * Quota now has no users and it has been written on last
568                  * dqput()
569                  */
570                 remove_dquot_hash(dquot);
571                 remove_free_dquot(dquot);
572                 remove_inuse(dquot);
573                 do_destroy_dquot(dquot);
574         }
575         spin_unlock(&dq_list_lock);
576 }
577
578 /* Call callback for every active dquot on given filesystem */
579 int dquot_scan_active(struct super_block *sb,
580                       int (*fn)(struct dquot *dquot, unsigned long priv),
581                       unsigned long priv)
582 {
583         struct dquot *dquot, *old_dquot = NULL;
584         int ret = 0;
585
586         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
587
588         spin_lock(&dq_list_lock);
589         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
590                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
591                         continue;
592                 if (dquot->dq_sb != sb)
593                         continue;
594                 /* Now we have active dquot so we can just increase use count */
595                 atomic_inc(&dquot->dq_count);
596                 spin_unlock(&dq_list_lock);
597                 dqput(old_dquot);
598                 old_dquot = dquot;
599                 /*
600                  * ->release_dquot() can be racing with us. Our reference
601                  * protects us from new calls to it so just wait for any
602                  * outstanding call and recheck the DQ_ACTIVE_B after that.
603                  */
604                 wait_on_dquot(dquot);
605                 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
606                         ret = fn(dquot, priv);
607                         if (ret < 0)
608                                 goto out;
609                 }
610                 spin_lock(&dq_list_lock);
611                 /* We are safe to continue now because our dquot could not
612                  * be moved out of the inuse list while we hold the reference */
613         }
614         spin_unlock(&dq_list_lock);
615 out:
616         dqput(old_dquot);
617         return ret;
618 }
619 EXPORT_SYMBOL(dquot_scan_active);
620
621 /* Write all dquot structures to quota files */
622 int dquot_writeback_dquots(struct super_block *sb, int type)
623 {
624         struct list_head dirty;
625         struct dquot *dquot;
626         struct quota_info *dqopt = sb_dqopt(sb);
627         int cnt;
628         int err, ret = 0;
629
630         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
631
632         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
633                 if (type != -1 && cnt != type)
634                         continue;
635                 if (!sb_has_quota_active(sb, cnt))
636                         continue;
637                 spin_lock(&dq_list_lock);
638                 /* Move list away to avoid livelock. */
639                 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
640                 while (!list_empty(&dirty)) {
641                         dquot = list_first_entry(&dirty, struct dquot,
642                                                  dq_dirty);
643
644                         WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
645
646                         /* Now we have active dquot from which someone is
647                          * holding reference so we can safely just increase
648                          * use count */
649                         dqgrab(dquot);
650                         spin_unlock(&dq_list_lock);
651                         err = sb->dq_op->write_dquot(dquot);
652                         if (err) {
653                                 /*
654                                  * Clear dirty bit anyway to avoid infinite
655                                  * loop here.
656                                  */
657                                 clear_dquot_dirty(dquot);
658                                 if (!ret)
659                                         ret = err;
660                         }
661                         dqput(dquot);
662                         spin_lock(&dq_list_lock);
663                 }
664                 spin_unlock(&dq_list_lock);
665         }
666
667         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
668                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
669                     && info_dirty(&dqopt->info[cnt]))
670                         sb->dq_op->write_info(sb, cnt);
671         dqstats_inc(DQST_SYNCS);
672
673         return ret;
674 }
675 EXPORT_SYMBOL(dquot_writeback_dquots);
676
677 /* Write all dquot structures to disk and make them visible from userspace */
678 int dquot_quota_sync(struct super_block *sb, int type)
679 {
680         struct quota_info *dqopt = sb_dqopt(sb);
681         int cnt;
682         int ret;
683
684         ret = dquot_writeback_dquots(sb, type);
685         if (ret)
686                 return ret;
687         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
688                 return 0;
689
690         /* This is not very clever (and fast) but currently I don't know about
691          * any other simple way of getting quota data to disk and we must get
692          * them there for userspace to be visible... */
693         if (sb->s_op->sync_fs) {
694                 ret = sb->s_op->sync_fs(sb, 1);
695                 if (ret)
696                         return ret;
697         }
698         ret = sync_blockdev(sb->s_bdev);
699         if (ret)
700                 return ret;
701
702         /*
703          * Now when everything is written we can discard the pagecache so
704          * that userspace sees the changes.
705          */
706         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
707                 if (type != -1 && cnt != type)
708                         continue;
709                 if (!sb_has_quota_active(sb, cnt))
710                         continue;
711                 inode_lock(dqopt->files[cnt]);
712                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
713                 inode_unlock(dqopt->files[cnt]);
714         }
715
716         return 0;
717 }
718 EXPORT_SYMBOL(dquot_quota_sync);
719
720 static unsigned long
721 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
722 {
723         struct dquot *dquot;
724         unsigned long freed = 0;
725
726         spin_lock(&dq_list_lock);
727         while (!list_empty(&free_dquots) && sc->nr_to_scan) {
728                 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
729                 remove_dquot_hash(dquot);
730                 remove_free_dquot(dquot);
731                 remove_inuse(dquot);
732                 do_destroy_dquot(dquot);
733                 sc->nr_to_scan--;
734                 freed++;
735         }
736         spin_unlock(&dq_list_lock);
737         return freed;
738 }
739
740 static unsigned long
741 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
742 {
743         return vfs_pressure_ratio(
744         percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
745 }
746
747 static struct shrinker dqcache_shrinker = {
748         .count_objects = dqcache_shrink_count,
749         .scan_objects = dqcache_shrink_scan,
750         .seeks = DEFAULT_SEEKS,
751 };
752
753 /*
754  * Put reference to dquot
755  */
756 void dqput(struct dquot *dquot)
757 {
758         int ret;
759
760         if (!dquot)
761                 return;
762 #ifdef CONFIG_QUOTA_DEBUG
763         if (!atomic_read(&dquot->dq_count)) {
764                 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
765                             quotatypes[dquot->dq_id.type],
766                             from_kqid(&init_user_ns, dquot->dq_id));
767                 BUG();
768         }
769 #endif
770         dqstats_inc(DQST_DROPS);
771 we_slept:
772         spin_lock(&dq_list_lock);
773         if (atomic_read(&dquot->dq_count) > 1) {
774                 /* We have more than one user... nothing to do */
775                 atomic_dec(&dquot->dq_count);
776                 /* Releasing dquot during quotaoff phase? */
777                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
778                     atomic_read(&dquot->dq_count) == 1)
779                         wake_up(&dquot_ref_wq);
780                 spin_unlock(&dq_list_lock);
781                 return;
782         }
783         /* Need to release dquot? */
784         if (dquot_dirty(dquot)) {
785                 spin_unlock(&dq_list_lock);
786                 /* Commit dquot before releasing */
787                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
788                 if (ret < 0) {
789                         quota_error(dquot->dq_sb, "Can't write quota structure"
790                                     " (error %d). Quota may get out of sync!",
791                                     ret);
792                         /*
793                          * We clear dirty bit anyway, so that we avoid
794                          * infinite loop here
795                          */
796                         clear_dquot_dirty(dquot);
797                 }
798                 goto we_slept;
799         }
800         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
801                 spin_unlock(&dq_list_lock);
802                 dquot->dq_sb->dq_op->release_dquot(dquot);
803                 goto we_slept;
804         }
805         atomic_dec(&dquot->dq_count);
806 #ifdef CONFIG_QUOTA_DEBUG
807         /* sanity check */
808         BUG_ON(!list_empty(&dquot->dq_free));
809 #endif
810         put_dquot_last(dquot);
811         spin_unlock(&dq_list_lock);
812 }
813 EXPORT_SYMBOL(dqput);
814
815 struct dquot *dquot_alloc(struct super_block *sb, int type)
816 {
817         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
818 }
819 EXPORT_SYMBOL(dquot_alloc);
820
821 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
822 {
823         struct dquot *dquot;
824
825         dquot = sb->dq_op->alloc_dquot(sb, type);
826         if(!dquot)
827                 return NULL;
828
829         mutex_init(&dquot->dq_lock);
830         INIT_LIST_HEAD(&dquot->dq_free);
831         INIT_LIST_HEAD(&dquot->dq_inuse);
832         INIT_HLIST_NODE(&dquot->dq_hash);
833         INIT_LIST_HEAD(&dquot->dq_dirty);
834         dquot->dq_sb = sb;
835         dquot->dq_id = make_kqid_invalid(type);
836         atomic_set(&dquot->dq_count, 1);
837         spin_lock_init(&dquot->dq_dqb_lock);
838
839         return dquot;
840 }
841
842 /*
843  * Get reference to dquot
844  *
845  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
846  * destroying our dquot by:
847  *   a) checking for quota flags under dq_list_lock and
848  *   b) getting a reference to dquot before we release dq_list_lock
849  */
850 struct dquot *dqget(struct super_block *sb, struct kqid qid)
851 {
852         unsigned int hashent = hashfn(sb, qid);
853         struct dquot *dquot, *empty = NULL;
854
855         if (!qid_has_mapping(sb->s_user_ns, qid))
856                 return ERR_PTR(-EINVAL);
857
858         if (!sb_has_quota_active(sb, qid.type))
859                 return ERR_PTR(-ESRCH);
860 we_slept:
861         spin_lock(&dq_list_lock);
862         spin_lock(&dq_state_lock);
863         if (!sb_has_quota_active(sb, qid.type)) {
864                 spin_unlock(&dq_state_lock);
865                 spin_unlock(&dq_list_lock);
866                 dquot = ERR_PTR(-ESRCH);
867                 goto out;
868         }
869         spin_unlock(&dq_state_lock);
870
871         dquot = find_dquot(hashent, sb, qid);
872         if (!dquot) {
873                 if (!empty) {
874                         spin_unlock(&dq_list_lock);
875                         empty = get_empty_dquot(sb, qid.type);
876                         if (!empty)
877                                 schedule();     /* Try to wait for a moment... */
878                         goto we_slept;
879                 }
880                 dquot = empty;
881                 empty = NULL;
882                 dquot->dq_id = qid;
883                 /* all dquots go on the inuse_list */
884                 put_inuse(dquot);
885                 /* hash it first so it can be found */
886                 insert_dquot_hash(dquot);
887                 spin_unlock(&dq_list_lock);
888                 dqstats_inc(DQST_LOOKUPS);
889         } else {
890                 if (!atomic_read(&dquot->dq_count))
891                         remove_free_dquot(dquot);
892                 atomic_inc(&dquot->dq_count);
893                 spin_unlock(&dq_list_lock);
894                 dqstats_inc(DQST_CACHE_HITS);
895                 dqstats_inc(DQST_LOOKUPS);
896         }
897         /* Wait for dq_lock - after this we know that either dquot_release() is
898          * already finished or it will be canceled due to dq_count > 1 test */
899         wait_on_dquot(dquot);
900         /* Read the dquot / allocate space in quota file */
901         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
902                 int err;
903
904                 err = sb->dq_op->acquire_dquot(dquot);
905                 if (err < 0) {
906                         dqput(dquot);
907                         dquot = ERR_PTR(err);
908                         goto out;
909                 }
910         }
911         /*
912          * Make sure following reads see filled structure - paired with
913          * smp_mb__before_atomic() in dquot_acquire().
914          */
915         smp_rmb();
916 #ifdef CONFIG_QUOTA_DEBUG
917         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
918 #endif
919 out:
920         if (empty)
921                 do_destroy_dquot(empty);
922
923         return dquot;
924 }
925 EXPORT_SYMBOL(dqget);
926
927 static inline struct dquot **i_dquot(struct inode *inode)
928 {
929         return inode->i_sb->s_op->get_dquots(inode);
930 }
931
932 static int dqinit_needed(struct inode *inode, int type)
933 {
934         struct dquot * const *dquots;
935         int cnt;
936
937         if (IS_NOQUOTA(inode))
938                 return 0;
939
940         dquots = i_dquot(inode);
941         if (type != -1)
942                 return !dquots[type];
943         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
944                 if (!dquots[cnt])
945                         return 1;
946         return 0;
947 }
948
949 /* This routine is guarded by s_umount semaphore */
950 static int add_dquot_ref(struct super_block *sb, int type)
951 {
952         struct inode *inode, *old_inode = NULL;
953 #ifdef CONFIG_QUOTA_DEBUG
954         int reserved = 0;
955 #endif
956         int err = 0;
957
958         spin_lock(&sb->s_inode_list_lock);
959         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
960                 spin_lock(&inode->i_lock);
961                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
962                     !atomic_read(&inode->i_writecount) ||
963                     !dqinit_needed(inode, type)) {
964                         spin_unlock(&inode->i_lock);
965                         continue;
966                 }
967                 __iget(inode);
968                 spin_unlock(&inode->i_lock);
969                 spin_unlock(&sb->s_inode_list_lock);
970
971 #ifdef CONFIG_QUOTA_DEBUG
972                 if (unlikely(inode_get_rsv_space(inode) > 0))
973                         reserved = 1;
974 #endif
975                 iput(old_inode);
976                 err = __dquot_initialize(inode, type);
977                 if (err) {
978                         iput(inode);
979                         goto out;
980                 }
981
982                 /*
983                  * We hold a reference to 'inode' so it couldn't have been
984                  * removed from s_inodes list while we dropped the
985                  * s_inode_list_lock. We cannot iput the inode now as we can be
986                  * holding the last reference and we cannot iput it under
987                  * s_inode_list_lock. So we keep the reference and iput it
988                  * later.
989                  */
990                 old_inode = inode;
991                 cond_resched();
992                 spin_lock(&sb->s_inode_list_lock);
993         }
994         spin_unlock(&sb->s_inode_list_lock);
995         iput(old_inode);
996 out:
997 #ifdef CONFIG_QUOTA_DEBUG
998         if (reserved) {
999                 quota_error(sb, "Writes happened before quota was turned on "
1000                         "thus quota information is probably inconsistent. "
1001                         "Please run quotacheck(8)");
1002         }
1003 #endif
1004         return err;
1005 }
1006
1007 /*
1008  * Remove references to dquots from inode and add dquot to list for freeing
1009  * if we have the last reference to dquot
1010  */
1011 static void remove_inode_dquot_ref(struct inode *inode, int type,
1012                                    struct list_head *tofree_head)
1013 {
1014         struct dquot **dquots = i_dquot(inode);
1015         struct dquot *dquot = dquots[type];
1016
1017         if (!dquot)
1018                 return;
1019
1020         dquots[type] = NULL;
1021         if (list_empty(&dquot->dq_free)) {
1022                 /*
1023                  * The inode still has reference to dquot so it can't be in the
1024                  * free list
1025                  */
1026                 spin_lock(&dq_list_lock);
1027                 list_add(&dquot->dq_free, tofree_head);
1028                 spin_unlock(&dq_list_lock);
1029         } else {
1030                 /*
1031                  * Dquot is already in a list to put so we won't drop the last
1032                  * reference here.
1033                  */
1034                 dqput(dquot);
1035         }
1036 }
1037
1038 /*
1039  * Free list of dquots
1040  * Dquots are removed from inodes and no new references can be got so we are
1041  * the only ones holding reference
1042  */
1043 static void put_dquot_list(struct list_head *tofree_head)
1044 {
1045         struct list_head *act_head;
1046         struct dquot *dquot;
1047
1048         act_head = tofree_head->next;
1049         while (act_head != tofree_head) {
1050                 dquot = list_entry(act_head, struct dquot, dq_free);
1051                 act_head = act_head->next;
1052                 /* Remove dquot from the list so we won't have problems... */
1053                 list_del_init(&dquot->dq_free);
1054                 dqput(dquot);
1055         }
1056 }
1057
1058 static void remove_dquot_ref(struct super_block *sb, int type,
1059                 struct list_head *tofree_head)
1060 {
1061         struct inode *inode;
1062 #ifdef CONFIG_QUOTA_DEBUG
1063         int reserved = 0;
1064 #endif
1065
1066         spin_lock(&sb->s_inode_list_lock);
1067         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1068                 /*
1069                  *  We have to scan also I_NEW inodes because they can already
1070                  *  have quota pointer initialized. Luckily, we need to touch
1071                  *  only quota pointers and these have separate locking
1072                  *  (dq_data_lock).
1073                  */
1074                 spin_lock(&dq_data_lock);
1075                 if (!IS_NOQUOTA(inode)) {
1076 #ifdef CONFIG_QUOTA_DEBUG
1077                         if (unlikely(inode_get_rsv_space(inode) > 0))
1078                                 reserved = 1;
1079 #endif
1080                         remove_inode_dquot_ref(inode, type, tofree_head);
1081                 }
1082                 spin_unlock(&dq_data_lock);
1083         }
1084         spin_unlock(&sb->s_inode_list_lock);
1085 #ifdef CONFIG_QUOTA_DEBUG
1086         if (reserved) {
1087                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1088                         " was disabled thus quota information is probably "
1089                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1090         }
1091 #endif
1092 }
1093
1094 /* Gather all references from inodes and drop them */
1095 static void drop_dquot_ref(struct super_block *sb, int type)
1096 {
1097         LIST_HEAD(tofree_head);
1098
1099         if (sb->dq_op) {
1100                 remove_dquot_ref(sb, type, &tofree_head);
1101                 synchronize_srcu(&dquot_srcu);
1102                 put_dquot_list(&tofree_head);
1103         }
1104 }
1105
1106 static inline
1107 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1108 {
1109         if (dquot->dq_dqb.dqb_rsvspace >= number)
1110                 dquot->dq_dqb.dqb_rsvspace -= number;
1111         else {
1112                 WARN_ON_ONCE(1);
1113                 dquot->dq_dqb.dqb_rsvspace = 0;
1114         }
1115         if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1116             dquot->dq_dqb.dqb_bsoftlimit)
1117                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1118         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1119 }
1120
1121 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1122 {
1123         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1124             dquot->dq_dqb.dqb_curinodes >= number)
1125                 dquot->dq_dqb.dqb_curinodes -= number;
1126         else
1127                 dquot->dq_dqb.dqb_curinodes = 0;
1128         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1129                 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1130         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1131 }
1132
1133 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1134 {
1135         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1136             dquot->dq_dqb.dqb_curspace >= number)
1137                 dquot->dq_dqb.dqb_curspace -= number;
1138         else
1139                 dquot->dq_dqb.dqb_curspace = 0;
1140         if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1141             dquot->dq_dqb.dqb_bsoftlimit)
1142                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1143         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1144 }
1145
1146 struct dquot_warn {
1147         struct super_block *w_sb;
1148         struct kqid w_dq_id;
1149         short w_type;
1150 };
1151
1152 static int warning_issued(struct dquot *dquot, const int warntype)
1153 {
1154         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1155                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1156                 ((warntype == QUOTA_NL_IHARDWARN ||
1157                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1158
1159         if (!flag)
1160                 return 0;
1161         return test_and_set_bit(flag, &dquot->dq_flags);
1162 }
1163
1164 #ifdef CONFIG_PRINT_QUOTA_WARNING
1165 static int flag_print_warnings = 1;
1166
1167 static int need_print_warning(struct dquot_warn *warn)
1168 {
1169         if (!flag_print_warnings)
1170                 return 0;
1171
1172         switch (warn->w_dq_id.type) {
1173                 case USRQUOTA:
1174                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1175                 case GRPQUOTA:
1176                         return in_group_p(warn->w_dq_id.gid);
1177                 case PRJQUOTA:
1178                         return 1;
1179         }
1180         return 0;
1181 }
1182
1183 /* Print warning to user which exceeded quota */
1184 static void print_warning(struct dquot_warn *warn)
1185 {
1186         char *msg = NULL;
1187         struct tty_struct *tty;
1188         int warntype = warn->w_type;
1189
1190         if (warntype == QUOTA_NL_IHARDBELOW ||
1191             warntype == QUOTA_NL_ISOFTBELOW ||
1192             warntype == QUOTA_NL_BHARDBELOW ||
1193             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1194                 return;
1195
1196         tty = get_current_tty();
1197         if (!tty)
1198                 return;
1199         tty_write_message(tty, warn->w_sb->s_id);
1200         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1201                 tty_write_message(tty, ": warning, ");
1202         else
1203                 tty_write_message(tty, ": write failed, ");
1204         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1205         switch (warntype) {
1206                 case QUOTA_NL_IHARDWARN:
1207                         msg = " file limit reached.\r\n";
1208                         break;
1209                 case QUOTA_NL_ISOFTLONGWARN:
1210                         msg = " file quota exceeded too long.\r\n";
1211                         break;
1212                 case QUOTA_NL_ISOFTWARN:
1213                         msg = " file quota exceeded.\r\n";
1214                         break;
1215                 case QUOTA_NL_BHARDWARN:
1216                         msg = " block limit reached.\r\n";
1217                         break;
1218                 case QUOTA_NL_BSOFTLONGWARN:
1219                         msg = " block quota exceeded too long.\r\n";
1220                         break;
1221                 case QUOTA_NL_BSOFTWARN:
1222                         msg = " block quota exceeded.\r\n";
1223                         break;
1224         }
1225         tty_write_message(tty, msg);
1226         tty_kref_put(tty);
1227 }
1228 #endif
1229
1230 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1231                             int warntype)
1232 {
1233         if (warning_issued(dquot, warntype))
1234                 return;
1235         warn->w_type = warntype;
1236         warn->w_sb = dquot->dq_sb;
1237         warn->w_dq_id = dquot->dq_id;
1238 }
1239
1240 /*
1241  * Write warnings to the console and send warning messages over netlink.
1242  *
1243  * Note that this function can call into tty and networking code.
1244  */
1245 static void flush_warnings(struct dquot_warn *warn)
1246 {
1247         int i;
1248
1249         for (i = 0; i < MAXQUOTAS; i++) {
1250                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1251                         continue;
1252 #ifdef CONFIG_PRINT_QUOTA_WARNING
1253                 print_warning(&warn[i]);
1254 #endif
1255                 quota_send_warning(warn[i].w_dq_id,
1256                                    warn[i].w_sb->s_dev, warn[i].w_type);
1257         }
1258 }
1259
1260 static int ignore_hardlimit(struct dquot *dquot)
1261 {
1262         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1263
1264         return capable(CAP_SYS_RESOURCE) &&
1265                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1266                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1267 }
1268
1269 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1270                             struct dquot_warn *warn)
1271 {
1272         qsize_t newinodes;
1273         int ret = 0;
1274
1275         spin_lock(&dquot->dq_dqb_lock);
1276         newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1277         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1278             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1279                 goto add;
1280
1281         if (dquot->dq_dqb.dqb_ihardlimit &&
1282             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1283             !ignore_hardlimit(dquot)) {
1284                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1285                 ret = -EDQUOT;
1286                 goto out;
1287         }
1288
1289         if (dquot->dq_dqb.dqb_isoftlimit &&
1290             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1291             dquot->dq_dqb.dqb_itime &&
1292             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1293             !ignore_hardlimit(dquot)) {
1294                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1295                 ret = -EDQUOT;
1296                 goto out;
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 add:
1307         dquot->dq_dqb.dqb_curinodes = newinodes;
1308
1309 out:
1310         spin_unlock(&dquot->dq_dqb_lock);
1311         return ret;
1312 }
1313
1314 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1315                            qsize_t rsv_space, unsigned int flags,
1316                            struct dquot_warn *warn)
1317 {
1318         qsize_t tspace;
1319         struct super_block *sb = dquot->dq_sb;
1320         int ret = 0;
1321
1322         spin_lock(&dquot->dq_dqb_lock);
1323         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1324             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1325                 goto finish;
1326
1327         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1328                 + space + rsv_space;
1329
1330         if (dquot->dq_dqb.dqb_bhardlimit &&
1331             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1332             !ignore_hardlimit(dquot)) {
1333                 if (flags & DQUOT_SPACE_WARN)
1334                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1335                 ret = -EDQUOT;
1336                 goto finish;
1337         }
1338
1339         if (dquot->dq_dqb.dqb_bsoftlimit &&
1340             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1341             dquot->dq_dqb.dqb_btime &&
1342             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1343             !ignore_hardlimit(dquot)) {
1344                 if (flags & DQUOT_SPACE_WARN)
1345                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1346                 ret = -EDQUOT;
1347                 goto finish;
1348         }
1349
1350         if (dquot->dq_dqb.dqb_bsoftlimit &&
1351             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1352             dquot->dq_dqb.dqb_btime == 0) {
1353                 if (flags & DQUOT_SPACE_WARN) {
1354                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1355                         dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1356                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1357                 } else {
1358                         /*
1359                          * We don't allow preallocation to exceed softlimit so exceeding will
1360                          * be always printed
1361                          */
1362                         ret = -EDQUOT;
1363                         goto finish;
1364                 }
1365         }
1366 finish:
1367         /*
1368          * We have to be careful and go through warning generation & grace time
1369          * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1370          * only here...
1371          */
1372         if (flags & DQUOT_SPACE_NOFAIL)
1373                 ret = 0;
1374         if (!ret) {
1375                 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1376                 dquot->dq_dqb.dqb_curspace += space;
1377         }
1378         spin_unlock(&dquot->dq_dqb_lock);
1379         return ret;
1380 }
1381
1382 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1383 {
1384         qsize_t newinodes;
1385
1386         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1387             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1388             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1389                 return QUOTA_NL_NOWARN;
1390
1391         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1392         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1393                 return QUOTA_NL_ISOFTBELOW;
1394         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1395             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1396                 return QUOTA_NL_IHARDBELOW;
1397         return QUOTA_NL_NOWARN;
1398 }
1399
1400 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1401 {
1402         qsize_t tspace;
1403
1404         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1405
1406         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1407             tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1408                 return QUOTA_NL_NOWARN;
1409
1410         if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1411                 return QUOTA_NL_BSOFTBELOW;
1412         if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1413             tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1414                 return QUOTA_NL_BHARDBELOW;
1415         return QUOTA_NL_NOWARN;
1416 }
1417
1418 static int dquot_active(const struct inode *inode)
1419 {
1420         struct super_block *sb = inode->i_sb;
1421
1422         if (IS_NOQUOTA(inode))
1423                 return 0;
1424         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1425 }
1426
1427 /*
1428  * Initialize quota pointers in inode
1429  *
1430  * It is better to call this function outside of any transaction as it
1431  * might need a lot of space in journal for dquot structure allocation.
1432  */
1433 static int __dquot_initialize(struct inode *inode, int type)
1434 {
1435         int cnt, init_needed = 0;
1436         struct dquot **dquots, *got[MAXQUOTAS] = {};
1437         struct super_block *sb = inode->i_sb;
1438         qsize_t rsv;
1439         int ret = 0;
1440
1441         if (!dquot_active(inode))
1442                 return 0;
1443
1444         dquots = i_dquot(inode);
1445
1446         /* First get references to structures we might need. */
1447         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1448                 struct kqid qid;
1449                 kprojid_t projid;
1450                 int rc;
1451                 struct dquot *dquot;
1452
1453                 if (type != -1 && cnt != type)
1454                         continue;
1455                 /*
1456                  * The i_dquot should have been initialized in most cases,
1457                  * we check it without locking here to avoid unnecessary
1458                  * dqget()/dqput() calls.
1459                  */
1460                 if (dquots[cnt])
1461                         continue;
1462
1463                 if (!sb_has_quota_active(sb, cnt))
1464                         continue;
1465
1466                 init_needed = 1;
1467
1468                 switch (cnt) {
1469                 case USRQUOTA:
1470                         qid = make_kqid_uid(inode->i_uid);
1471                         break;
1472                 case GRPQUOTA:
1473                         qid = make_kqid_gid(inode->i_gid);
1474                         break;
1475                 case PRJQUOTA:
1476                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1477                         if (rc)
1478                                 continue;
1479                         qid = make_kqid_projid(projid);
1480                         break;
1481                 }
1482                 dquot = dqget(sb, qid);
1483                 if (IS_ERR(dquot)) {
1484                         /* We raced with somebody turning quotas off... */
1485                         if (PTR_ERR(dquot) != -ESRCH) {
1486                                 ret = PTR_ERR(dquot);
1487                                 goto out_put;
1488                         }
1489                         dquot = NULL;
1490                 }
1491                 got[cnt] = dquot;
1492         }
1493
1494         /* All required i_dquot has been initialized */
1495         if (!init_needed)
1496                 return 0;
1497
1498         spin_lock(&dq_data_lock);
1499         if (IS_NOQUOTA(inode))
1500                 goto out_lock;
1501         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1502                 if (type != -1 && cnt != type)
1503                         continue;
1504                 /* Avoid races with quotaoff() */
1505                 if (!sb_has_quota_active(sb, cnt))
1506                         continue;
1507                 /* We could race with quotaon or dqget() could have failed */
1508                 if (!got[cnt])
1509                         continue;
1510                 if (!dquots[cnt]) {
1511                         dquots[cnt] = got[cnt];
1512                         got[cnt] = NULL;
1513                         /*
1514                          * Make quota reservation system happy if someone
1515                          * did a write before quota was turned on
1516                          */
1517                         rsv = inode_get_rsv_space(inode);
1518                         if (unlikely(rsv)) {
1519                                 spin_lock(&inode->i_lock);
1520                                 /* Get reservation again under proper lock */
1521                                 rsv = __inode_get_rsv_space(inode);
1522                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1523                                 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1524                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1525                                 spin_unlock(&inode->i_lock);
1526                         }
1527                 }
1528         }
1529 out_lock:
1530         spin_unlock(&dq_data_lock);
1531 out_put:
1532         /* Drop unused references */
1533         dqput_all(got);
1534
1535         return ret;
1536 }
1537
1538 int dquot_initialize(struct inode *inode)
1539 {
1540         return __dquot_initialize(inode, -1);
1541 }
1542 EXPORT_SYMBOL(dquot_initialize);
1543
1544 bool dquot_initialize_needed(struct inode *inode)
1545 {
1546         struct dquot **dquots;
1547         int i;
1548
1549         if (!dquot_active(inode))
1550                 return false;
1551
1552         dquots = i_dquot(inode);
1553         for (i = 0; i < MAXQUOTAS; i++)
1554                 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1555                         return true;
1556         return false;
1557 }
1558 EXPORT_SYMBOL(dquot_initialize_needed);
1559
1560 /*
1561  * Release all quotas referenced by inode.
1562  *
1563  * This function only be called on inode free or converting
1564  * a file to quota file, no other users for the i_dquot in
1565  * both cases, so we needn't call synchronize_srcu() after
1566  * clearing i_dquot.
1567  */
1568 static void __dquot_drop(struct inode *inode)
1569 {
1570         int cnt;
1571         struct dquot **dquots = i_dquot(inode);
1572         struct dquot *put[MAXQUOTAS];
1573
1574         spin_lock(&dq_data_lock);
1575         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1576                 put[cnt] = dquots[cnt];
1577                 dquots[cnt] = NULL;
1578         }
1579         spin_unlock(&dq_data_lock);
1580         dqput_all(put);
1581 }
1582
1583 void dquot_drop(struct inode *inode)
1584 {
1585         struct dquot * const *dquots;
1586         int cnt;
1587
1588         if (IS_NOQUOTA(inode))
1589                 return;
1590
1591         /*
1592          * Test before calling to rule out calls from proc and such
1593          * where we are not allowed to block. Note that this is
1594          * actually reliable test even without the lock - the caller
1595          * must assure that nobody can come after the DQUOT_DROP and
1596          * add quota pointers back anyway.
1597          */
1598         dquots = i_dquot(inode);
1599         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1600                 if (dquots[cnt])
1601                         break;
1602         }
1603
1604         if (cnt < MAXQUOTAS)
1605                 __dquot_drop(inode);
1606 }
1607 EXPORT_SYMBOL(dquot_drop);
1608
1609 /*
1610  * inode_reserved_space is managed internally by quota, and protected by
1611  * i_lock similar to i_blocks+i_bytes.
1612  */
1613 static qsize_t *inode_reserved_space(struct inode * inode)
1614 {
1615         /* Filesystem must explicitly define it's own method in order to use
1616          * quota reservation interface */
1617         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1618         return inode->i_sb->dq_op->get_reserved_space(inode);
1619 }
1620
1621 static qsize_t __inode_get_rsv_space(struct inode *inode)
1622 {
1623         if (!inode->i_sb->dq_op->get_reserved_space)
1624                 return 0;
1625         return *inode_reserved_space(inode);
1626 }
1627
1628 static qsize_t inode_get_rsv_space(struct inode *inode)
1629 {
1630         qsize_t ret;
1631
1632         if (!inode->i_sb->dq_op->get_reserved_space)
1633                 return 0;
1634         spin_lock(&inode->i_lock);
1635         ret = __inode_get_rsv_space(inode);
1636         spin_unlock(&inode->i_lock);
1637         return ret;
1638 }
1639
1640 /*
1641  * This functions updates i_blocks+i_bytes fields and quota information
1642  * (together with appropriate checks).
1643  *
1644  * NOTE: We absolutely rely on the fact that caller dirties the inode
1645  * (usually helpers in quotaops.h care about this) and holds a handle for
1646  * the current transaction so that dquot write and inode write go into the
1647  * same transaction.
1648  */
1649
1650 /*
1651  * This operation can block, but only after everything is updated
1652  */
1653 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1654 {
1655         int cnt, ret = 0, index;
1656         struct dquot_warn warn[MAXQUOTAS];
1657         int reserve = flags & DQUOT_SPACE_RESERVE;
1658         struct dquot **dquots;
1659
1660         if (!dquot_active(inode)) {
1661                 if (reserve) {
1662                         spin_lock(&inode->i_lock);
1663                         *inode_reserved_space(inode) += number;
1664                         spin_unlock(&inode->i_lock);
1665                 } else {
1666                         inode_add_bytes(inode, number);
1667                 }
1668                 goto out;
1669         }
1670
1671         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1672                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1673
1674         dquots = i_dquot(inode);
1675         index = srcu_read_lock(&dquot_srcu);
1676         spin_lock(&inode->i_lock);
1677         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1678                 if (!dquots[cnt])
1679                         continue;
1680                 if (reserve) {
1681                         ret = dquot_add_space(dquots[cnt], 0, number, flags,
1682                                               &warn[cnt]);
1683                 } else {
1684                         ret = dquot_add_space(dquots[cnt], number, 0, flags,
1685                                               &warn[cnt]);
1686                 }
1687                 if (ret) {
1688                         /* Back out changes we already did */
1689                         for (cnt--; cnt >= 0; cnt--) {
1690                                 if (!dquots[cnt])
1691                                         continue;
1692                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1693                                 if (reserve)
1694                                         dquot_free_reserved_space(dquots[cnt],
1695                                                                   number);
1696                                 else
1697                                         dquot_decr_space(dquots[cnt], number);
1698                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1699                         }
1700                         spin_unlock(&inode->i_lock);
1701                         goto out_flush_warn;
1702                 }
1703         }
1704         if (reserve)
1705                 *inode_reserved_space(inode) += number;
1706         else
1707                 __inode_add_bytes(inode, number);
1708         spin_unlock(&inode->i_lock);
1709
1710         if (reserve)
1711                 goto out_flush_warn;
1712         mark_all_dquot_dirty(dquots);
1713 out_flush_warn:
1714         srcu_read_unlock(&dquot_srcu, index);
1715         flush_warnings(warn);
1716 out:
1717         return ret;
1718 }
1719 EXPORT_SYMBOL(__dquot_alloc_space);
1720
1721 /*
1722  * This operation can block, but only after everything is updated
1723  */
1724 int dquot_alloc_inode(struct inode *inode)
1725 {
1726         int cnt, ret = 0, index;
1727         struct dquot_warn warn[MAXQUOTAS];
1728         struct dquot * const *dquots;
1729
1730         if (!dquot_active(inode))
1731                 return 0;
1732         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1733                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1734
1735         dquots = i_dquot(inode);
1736         index = srcu_read_lock(&dquot_srcu);
1737         spin_lock(&inode->i_lock);
1738         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1739                 if (!dquots[cnt])
1740                         continue;
1741                 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1742                 if (ret) {
1743                         for (cnt--; cnt >= 0; cnt--) {
1744                                 if (!dquots[cnt])
1745                                         continue;
1746                                 /* Back out changes we already did */
1747                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1748                                 dquot_decr_inodes(dquots[cnt], 1);
1749                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1750                         }
1751                         goto warn_put_all;
1752                 }
1753         }
1754
1755 warn_put_all:
1756         spin_unlock(&inode->i_lock);
1757         if (ret == 0)
1758                 mark_all_dquot_dirty(dquots);
1759         srcu_read_unlock(&dquot_srcu, index);
1760         flush_warnings(warn);
1761         return ret;
1762 }
1763 EXPORT_SYMBOL(dquot_alloc_inode);
1764
1765 /*
1766  * Convert in-memory reserved quotas to real consumed quotas
1767  */
1768 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1769 {
1770         struct dquot **dquots;
1771         int cnt, index;
1772
1773         if (!dquot_active(inode)) {
1774                 spin_lock(&inode->i_lock);
1775                 *inode_reserved_space(inode) -= number;
1776                 __inode_add_bytes(inode, number);
1777                 spin_unlock(&inode->i_lock);
1778                 return 0;
1779         }
1780
1781         dquots = i_dquot(inode);
1782         index = srcu_read_lock(&dquot_srcu);
1783         spin_lock(&inode->i_lock);
1784         /* Claim reserved quotas to allocated quotas */
1785         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1786                 if (dquots[cnt]) {
1787                         struct dquot *dquot = dquots[cnt];
1788
1789                         spin_lock(&dquot->dq_dqb_lock);
1790                         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1791                                 number = dquot->dq_dqb.dqb_rsvspace;
1792                         dquot->dq_dqb.dqb_curspace += number;
1793                         dquot->dq_dqb.dqb_rsvspace -= number;
1794                         spin_unlock(&dquot->dq_dqb_lock);
1795                 }
1796         }
1797         /* Update inode bytes */
1798         *inode_reserved_space(inode) -= number;
1799         __inode_add_bytes(inode, number);
1800         spin_unlock(&inode->i_lock);
1801         mark_all_dquot_dirty(dquots);
1802         srcu_read_unlock(&dquot_srcu, index);
1803         return 0;
1804 }
1805 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1806
1807 /*
1808  * Convert allocated space back to in-memory reserved quotas
1809  */
1810 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1811 {
1812         struct dquot **dquots;
1813         int cnt, index;
1814
1815         if (!dquot_active(inode)) {
1816                 spin_lock(&inode->i_lock);
1817                 *inode_reserved_space(inode) += number;
1818                 __inode_sub_bytes(inode, number);
1819                 spin_unlock(&inode->i_lock);
1820                 return;
1821         }
1822
1823         dquots = i_dquot(inode);
1824         index = srcu_read_lock(&dquot_srcu);
1825         spin_lock(&inode->i_lock);
1826         /* Claim reserved quotas to allocated quotas */
1827         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1828                 if (dquots[cnt]) {
1829                         struct dquot *dquot = dquots[cnt];
1830
1831                         spin_lock(&dquot->dq_dqb_lock);
1832                         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1833                                 number = dquot->dq_dqb.dqb_curspace;
1834                         dquot->dq_dqb.dqb_rsvspace += number;
1835                         dquot->dq_dqb.dqb_curspace -= number;
1836                         spin_unlock(&dquot->dq_dqb_lock);
1837                 }
1838         }
1839         /* Update inode bytes */
1840         *inode_reserved_space(inode) += number;
1841         __inode_sub_bytes(inode, number);
1842         spin_unlock(&inode->i_lock);
1843         mark_all_dquot_dirty(dquots);
1844         srcu_read_unlock(&dquot_srcu, index);
1845         return;
1846 }
1847 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1848
1849 /*
1850  * This operation can block, but only after everything is updated
1851  */
1852 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1853 {
1854         unsigned int cnt;
1855         struct dquot_warn warn[MAXQUOTAS];
1856         struct dquot **dquots;
1857         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1858
1859         if (!dquot_active(inode)) {
1860                 if (reserve) {
1861                         spin_lock(&inode->i_lock);
1862                         *inode_reserved_space(inode) -= number;
1863                         spin_unlock(&inode->i_lock);
1864                 } else {
1865                         inode_sub_bytes(inode, number);
1866                 }
1867                 return;
1868         }
1869
1870         dquots = i_dquot(inode);
1871         index = srcu_read_lock(&dquot_srcu);
1872         spin_lock(&inode->i_lock);
1873         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1874                 int wtype;
1875
1876                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1877                 if (!dquots[cnt])
1878                         continue;
1879                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1880                 wtype = info_bdq_free(dquots[cnt], number);
1881                 if (wtype != QUOTA_NL_NOWARN)
1882                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1883                 if (reserve)
1884                         dquot_free_reserved_space(dquots[cnt], number);
1885                 else
1886                         dquot_decr_space(dquots[cnt], number);
1887                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1888         }
1889         if (reserve)
1890                 *inode_reserved_space(inode) -= number;
1891         else
1892                 __inode_sub_bytes(inode, number);
1893         spin_unlock(&inode->i_lock);
1894
1895         if (reserve)
1896                 goto out_unlock;
1897         mark_all_dquot_dirty(dquots);
1898 out_unlock:
1899         srcu_read_unlock(&dquot_srcu, index);
1900         flush_warnings(warn);
1901 }
1902 EXPORT_SYMBOL(__dquot_free_space);
1903
1904 /*
1905  * This operation can block, but only after everything is updated
1906  */
1907 void dquot_free_inode(struct inode *inode)
1908 {
1909         unsigned int cnt;
1910         struct dquot_warn warn[MAXQUOTAS];
1911         struct dquot * const *dquots;
1912         int index;
1913
1914         if (!dquot_active(inode))
1915                 return;
1916
1917         dquots = i_dquot(inode);
1918         index = srcu_read_lock(&dquot_srcu);
1919         spin_lock(&inode->i_lock);
1920         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1921                 int wtype;
1922
1923                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1924                 if (!dquots[cnt])
1925                         continue;
1926                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1927                 wtype = info_idq_free(dquots[cnt], 1);
1928                 if (wtype != QUOTA_NL_NOWARN)
1929                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1930                 dquot_decr_inodes(dquots[cnt], 1);
1931                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1932         }
1933         spin_unlock(&inode->i_lock);
1934         mark_all_dquot_dirty(dquots);
1935         srcu_read_unlock(&dquot_srcu, index);
1936         flush_warnings(warn);
1937 }
1938 EXPORT_SYMBOL(dquot_free_inode);
1939
1940 /*
1941  * Transfer the number of inode and blocks from one diskquota to an other.
1942  * On success, dquot references in transfer_to are consumed and references
1943  * to original dquots that need to be released are placed there. On failure,
1944  * references are kept untouched.
1945  *
1946  * This operation can block, but only after everything is updated
1947  * A transaction must be started when entering this function.
1948  *
1949  * We are holding reference on transfer_from & transfer_to, no need to
1950  * protect them by srcu_read_lock().
1951  */
1952 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1953 {
1954         qsize_t cur_space;
1955         qsize_t rsv_space = 0;
1956         qsize_t inode_usage = 1;
1957         struct dquot *transfer_from[MAXQUOTAS] = {};
1958         int cnt, ret = 0;
1959         char is_valid[MAXQUOTAS] = {};
1960         struct dquot_warn warn_to[MAXQUOTAS];
1961         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1962         struct dquot_warn warn_from_space[MAXQUOTAS];
1963
1964         if (IS_NOQUOTA(inode))
1965                 return 0;
1966
1967         if (inode->i_sb->dq_op->get_inode_usage) {
1968                 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1969                 if (ret)
1970                         return ret;
1971         }
1972
1973         /* Initialize the arrays */
1974         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1975                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1976                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1977                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1978         }
1979
1980         spin_lock(&dq_data_lock);
1981         spin_lock(&inode->i_lock);
1982         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1983                 spin_unlock(&inode->i_lock);
1984                 spin_unlock(&dq_data_lock);
1985                 return 0;
1986         }
1987         cur_space = __inode_get_bytes(inode);
1988         rsv_space = __inode_get_rsv_space(inode);
1989         /*
1990          * Build the transfer_from list, check limits, and update usage in
1991          * the target structures.
1992          */
1993         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1994                 /*
1995                  * Skip changes for same uid or gid or for turned off quota-type.
1996                  */
1997                 if (!transfer_to[cnt])
1998                         continue;
1999                 /* Avoid races with quotaoff() */
2000                 if (!sb_has_quota_active(inode->i_sb, cnt))
2001                         continue;
2002                 is_valid[cnt] = 1;
2003                 transfer_from[cnt] = i_dquot(inode)[cnt];
2004                 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2005                                        &warn_to[cnt]);
2006                 if (ret)
2007                         goto over_quota;
2008                 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2009                                       DQUOT_SPACE_WARN, &warn_to[cnt]);
2010                 if (ret) {
2011                         spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2012                         dquot_decr_inodes(transfer_to[cnt], inode_usage);
2013                         spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2014                         goto over_quota;
2015                 }
2016         }
2017
2018         /* Decrease usage for source structures and update quota pointers */
2019         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2020                 if (!is_valid[cnt])
2021                         continue;
2022                 /* Due to IO error we might not have transfer_from[] structure */
2023                 if (transfer_from[cnt]) {
2024                         int wtype;
2025
2026                         spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2027                         wtype = info_idq_free(transfer_from[cnt], inode_usage);
2028                         if (wtype != QUOTA_NL_NOWARN)
2029                                 prepare_warning(&warn_from_inodes[cnt],
2030                                                 transfer_from[cnt], wtype);
2031                         wtype = info_bdq_free(transfer_from[cnt],
2032                                               cur_space + rsv_space);
2033                         if (wtype != QUOTA_NL_NOWARN)
2034                                 prepare_warning(&warn_from_space[cnt],
2035                                                 transfer_from[cnt], wtype);
2036                         dquot_decr_inodes(transfer_from[cnt], inode_usage);
2037                         dquot_decr_space(transfer_from[cnt], cur_space);
2038                         dquot_free_reserved_space(transfer_from[cnt],
2039                                                   rsv_space);
2040                         spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2041                 }
2042                 i_dquot(inode)[cnt] = transfer_to[cnt];
2043         }
2044         spin_unlock(&inode->i_lock);
2045         spin_unlock(&dq_data_lock);
2046
2047         mark_all_dquot_dirty(transfer_from);
2048         mark_all_dquot_dirty(transfer_to);
2049         flush_warnings(warn_to);
2050         flush_warnings(warn_from_inodes);
2051         flush_warnings(warn_from_space);
2052         /* Pass back references to put */
2053         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2054                 if (is_valid[cnt])
2055                         transfer_to[cnt] = transfer_from[cnt];
2056         return 0;
2057 over_quota:
2058         /* Back out changes we already did */
2059         for (cnt--; cnt >= 0; cnt--) {
2060                 if (!is_valid[cnt])
2061                         continue;
2062                 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2063                 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2064                 dquot_decr_space(transfer_to[cnt], cur_space);
2065                 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2066                 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2067         }
2068         spin_unlock(&inode->i_lock);
2069         spin_unlock(&dq_data_lock);
2070         flush_warnings(warn_to);
2071         return ret;
2072 }
2073 EXPORT_SYMBOL(__dquot_transfer);
2074
2075 /* Wrapper for transferring ownership of an inode for uid/gid only
2076  * Called from FSXXX_setattr()
2077  */
2078 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2079 {
2080         struct dquot *transfer_to[MAXQUOTAS] = {};
2081         struct dquot *dquot;
2082         struct super_block *sb = inode->i_sb;
2083         int ret;
2084
2085         if (!dquot_active(inode))
2086                 return 0;
2087
2088         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2089                 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2090                 if (IS_ERR(dquot)) {
2091                         if (PTR_ERR(dquot) != -ESRCH) {
2092                                 ret = PTR_ERR(dquot);
2093                                 goto out_put;
2094                         }
2095                         dquot = NULL;
2096                 }
2097                 transfer_to[USRQUOTA] = dquot;
2098         }
2099         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2100                 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2101                 if (IS_ERR(dquot)) {
2102                         if (PTR_ERR(dquot) != -ESRCH) {
2103                                 ret = PTR_ERR(dquot);
2104                                 goto out_put;
2105                         }
2106                         dquot = NULL;
2107                 }
2108                 transfer_to[GRPQUOTA] = dquot;
2109         }
2110         ret = __dquot_transfer(inode, transfer_to);
2111 out_put:
2112         dqput_all(transfer_to);
2113         return ret;
2114 }
2115 EXPORT_SYMBOL(dquot_transfer);
2116
2117 /*
2118  * Write info of quota file to disk
2119  */
2120 int dquot_commit_info(struct super_block *sb, int type)
2121 {
2122         struct quota_info *dqopt = sb_dqopt(sb);
2123
2124         return dqopt->ops[type]->write_file_info(sb, type);
2125 }
2126 EXPORT_SYMBOL(dquot_commit_info);
2127
2128 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2129 {
2130         struct quota_info *dqopt = sb_dqopt(sb);
2131
2132         if (!sb_has_quota_active(sb, qid->type))
2133                 return -ESRCH;
2134         if (!dqopt->ops[qid->type]->get_next_id)
2135                 return -ENOSYS;
2136         return dqopt->ops[qid->type]->get_next_id(sb, qid);
2137 }
2138 EXPORT_SYMBOL(dquot_get_next_id);
2139
2140 /*
2141  * Definitions of diskquota operations.
2142  */
2143 const struct dquot_operations dquot_operations = {
2144         .write_dquot    = dquot_commit,
2145         .acquire_dquot  = dquot_acquire,
2146         .release_dquot  = dquot_release,
2147         .mark_dirty     = dquot_mark_dquot_dirty,
2148         .write_info     = dquot_commit_info,
2149         .alloc_dquot    = dquot_alloc,
2150         .destroy_dquot  = dquot_destroy,
2151         .get_next_id    = dquot_get_next_id,
2152 };
2153 EXPORT_SYMBOL(dquot_operations);
2154
2155 /*
2156  * Generic helper for ->open on filesystems supporting disk quotas.
2157  */
2158 int dquot_file_open(struct inode *inode, struct file *file)
2159 {
2160         int error;
2161
2162         error = generic_file_open(inode, file);
2163         if (!error && (file->f_mode & FMODE_WRITE))
2164                 error = dquot_initialize(inode);
2165         return error;
2166 }
2167 EXPORT_SYMBOL(dquot_file_open);
2168
2169 static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2170 {
2171         struct quota_info *dqopt = sb_dqopt(sb);
2172         struct inode *inode = dqopt->files[type];
2173
2174         if (!inode)
2175                 return;
2176         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2177                 inode_lock(inode);
2178                 inode->i_flags &= ~S_NOQUOTA;
2179                 inode_unlock(inode);
2180         }
2181         dqopt->files[type] = NULL;
2182         iput(inode);
2183 }
2184
2185 /*
2186  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2187  */
2188 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2189 {
2190         int cnt;
2191         struct quota_info *dqopt = sb_dqopt(sb);
2192
2193         /* s_umount should be held in exclusive mode */
2194         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2195                 up_read(&sb->s_umount);
2196
2197         /* Cannot turn off usage accounting without turning off limits, or
2198          * suspend quotas and simultaneously turn quotas off. */
2199         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2200             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2201             DQUOT_USAGE_ENABLED)))
2202                 return -EINVAL;
2203
2204         /*
2205          * Skip everything if there's nothing to do. We have to do this because
2206          * sometimes we are called when fill_super() failed and calling
2207          * sync_fs() in such cases does no good.
2208          */
2209         if (!sb_any_quota_loaded(sb))
2210                 return 0;
2211
2212         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2213                 if (type != -1 && cnt != type)
2214                         continue;
2215                 if (!sb_has_quota_loaded(sb, cnt))
2216                         continue;
2217
2218                 if (flags & DQUOT_SUSPENDED) {
2219                         spin_lock(&dq_state_lock);
2220                         dqopt->flags |=
2221                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2222                         spin_unlock(&dq_state_lock);
2223                 } else {
2224                         spin_lock(&dq_state_lock);
2225                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2226                         /* Turning off suspended quotas? */
2227                         if (!sb_has_quota_loaded(sb, cnt) &&
2228                             sb_has_quota_suspended(sb, cnt)) {
2229                                 dqopt->flags &= ~dquot_state_flag(
2230                                                         DQUOT_SUSPENDED, cnt);
2231                                 spin_unlock(&dq_state_lock);
2232                                 vfs_cleanup_quota_inode(sb, cnt);
2233                                 continue;
2234                         }
2235                         spin_unlock(&dq_state_lock);
2236                 }
2237
2238                 /* We still have to keep quota loaded? */
2239                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2240                         continue;
2241
2242                 /* Note: these are blocking operations */
2243                 drop_dquot_ref(sb, cnt);
2244                 invalidate_dquots(sb, cnt);
2245                 /*
2246                  * Now all dquots should be invalidated, all writes done so we
2247                  * should be only users of the info. No locks needed.
2248                  */
2249                 if (info_dirty(&dqopt->info[cnt]))
2250                         sb->dq_op->write_info(sb, cnt);
2251                 if (dqopt->ops[cnt]->free_file_info)
2252                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2253                 put_quota_format(dqopt->info[cnt].dqi_format);
2254                 dqopt->info[cnt].dqi_flags = 0;
2255                 dqopt->info[cnt].dqi_igrace = 0;
2256                 dqopt->info[cnt].dqi_bgrace = 0;
2257                 dqopt->ops[cnt] = NULL;
2258         }
2259
2260         /* Skip syncing and setting flags if quota files are hidden */
2261         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2262                 goto put_inodes;
2263
2264         /* Sync the superblock so that buffers with quota data are written to
2265          * disk (and so userspace sees correct data afterwards). */
2266         if (sb->s_op->sync_fs)
2267                 sb->s_op->sync_fs(sb, 1);
2268         sync_blockdev(sb->s_bdev);
2269         /* Now the quota files are just ordinary files and we can set the
2270          * inode flags back. Moreover we discard the pagecache so that
2271          * userspace sees the writes we did bypassing the pagecache. We
2272          * must also discard the blockdev buffers so that we see the
2273          * changes done by userspace on the next quotaon() */
2274         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2275                 if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2276                         inode_lock(dqopt->files[cnt]);
2277                         truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2278                         inode_unlock(dqopt->files[cnt]);
2279                 }
2280         if (sb->s_bdev)
2281                 invalidate_bdev(sb->s_bdev);
2282 put_inodes:
2283         /* We are done when suspending quotas */
2284         if (flags & DQUOT_SUSPENDED)
2285                 return 0;
2286
2287         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2288                 if (!sb_has_quota_loaded(sb, cnt))
2289                         vfs_cleanup_quota_inode(sb, cnt);
2290         return 0;
2291 }
2292 EXPORT_SYMBOL(dquot_disable);
2293
2294 int dquot_quota_off(struct super_block *sb, int type)
2295 {
2296         return dquot_disable(sb, type,
2297                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2298 }
2299 EXPORT_SYMBOL(dquot_quota_off);
2300
2301 /*
2302  *      Turn quotas on on a device
2303  */
2304
2305 static int vfs_setup_quota_inode(struct inode *inode, int type)
2306 {
2307         struct super_block *sb = inode->i_sb;
2308         struct quota_info *dqopt = sb_dqopt(sb);
2309
2310         if (!S_ISREG(inode->i_mode))
2311                 return -EACCES;
2312         if (IS_RDONLY(inode))
2313                 return -EROFS;
2314         if (sb_has_quota_loaded(sb, type))
2315                 return -EBUSY;
2316
2317         dqopt->files[type] = igrab(inode);
2318         if (!dqopt->files[type])
2319                 return -EIO;
2320         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2321                 /* We don't want quota and atime on quota files (deadlocks
2322                  * possible) Also nobody should write to the file - we use
2323                  * special IO operations which ignore the immutable bit. */
2324                 inode_lock(inode);
2325                 inode->i_flags |= S_NOQUOTA;
2326                 inode_unlock(inode);
2327                 /*
2328                  * When S_NOQUOTA is set, remove dquot references as no more
2329                  * references can be added
2330                  */
2331                 __dquot_drop(inode);
2332         }
2333         return 0;
2334 }
2335
2336 int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2337         unsigned int flags)
2338 {
2339         struct quota_format_type *fmt = find_quota_format(format_id);
2340         struct quota_info *dqopt = sb_dqopt(sb);
2341         int error;
2342
2343         /* Just unsuspend quotas? */
2344         BUG_ON(flags & DQUOT_SUSPENDED);
2345         /* s_umount should be held in exclusive mode */
2346         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2347                 up_read(&sb->s_umount);
2348
2349         if (!fmt)
2350                 return -ESRCH;
2351         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2352             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2353                 error = -EINVAL;
2354                 goto out_fmt;
2355         }
2356         /* Filesystems outside of init_user_ns not yet supported */
2357         if (sb->s_user_ns != &init_user_ns) {
2358                 error = -EINVAL;
2359                 goto out_fmt;
2360         }
2361         /* Usage always has to be set... */
2362         if (!(flags & DQUOT_USAGE_ENABLED)) {
2363                 error = -EINVAL;
2364                 goto out_fmt;
2365         }
2366         if (sb_has_quota_loaded(sb, type)) {
2367                 error = -EBUSY;
2368                 goto out_fmt;
2369         }
2370
2371         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2372                 /* As we bypass the pagecache we must now flush all the
2373                  * dirty data and invalidate caches so that kernel sees
2374                  * changes from userspace. It is not enough to just flush
2375                  * the quota file since if blocksize < pagesize, invalidation
2376                  * of the cache could fail because of other unrelated dirty
2377                  * data */
2378                 sync_filesystem(sb);
2379                 invalidate_bdev(sb->s_bdev);
2380         }
2381
2382         error = -EINVAL;
2383         if (!fmt->qf_ops->check_quota_file(sb, type))
2384                 goto out_fmt;
2385
2386         dqopt->ops[type] = fmt->qf_ops;
2387         dqopt->info[type].dqi_format = fmt;
2388         dqopt->info[type].dqi_fmt_id = format_id;
2389         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2390         error = dqopt->ops[type]->read_file_info(sb, type);
2391         if (error < 0)
2392                 goto out_fmt;
2393         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2394                 spin_lock(&dq_data_lock);
2395                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2396                 spin_unlock(&dq_data_lock);
2397         }
2398         spin_lock(&dq_state_lock);
2399         dqopt->flags |= dquot_state_flag(flags, type);
2400         spin_unlock(&dq_state_lock);
2401
2402         error = add_dquot_ref(sb, type);
2403         if (error)
2404                 dquot_disable(sb, type, flags);
2405
2406         return error;
2407 out_fmt:
2408         put_quota_format(fmt);
2409
2410         return error;
2411 }
2412 EXPORT_SYMBOL(dquot_load_quota_sb);
2413
2414 /*
2415  * More powerful function for turning on quotas on given quota inode allowing
2416  * setting of individual quota flags
2417  */
2418 int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2419         unsigned int flags)
2420 {
2421         int err;
2422
2423         err = vfs_setup_quota_inode(inode, type);
2424         if (err < 0)
2425                 return err;
2426         err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2427         if (err < 0)
2428                 vfs_cleanup_quota_inode(inode->i_sb, type);
2429         return err;
2430 }
2431 EXPORT_SYMBOL(dquot_load_quota_inode);
2432
2433 /* Reenable quotas on remount RW */
2434 int dquot_resume(struct super_block *sb, int type)
2435 {
2436         struct quota_info *dqopt = sb_dqopt(sb);
2437         int ret = 0, cnt;
2438         unsigned int flags;
2439
2440         /* s_umount should be held in exclusive mode */
2441         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2442                 up_read(&sb->s_umount);
2443
2444         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2445                 if (type != -1 && cnt != type)
2446                         continue;
2447                 if (!sb_has_quota_suspended(sb, cnt))
2448                         continue;
2449
2450                 spin_lock(&dq_state_lock);
2451                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2452                                                         DQUOT_LIMITS_ENABLED,
2453                                                         cnt);
2454                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2455                 spin_unlock(&dq_state_lock);
2456
2457                 flags = dquot_generic_flag(flags, cnt);
2458                 ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2459                                           flags);
2460                 if (ret < 0)
2461                         vfs_cleanup_quota_inode(sb, cnt);
2462         }
2463
2464         return ret;
2465 }
2466 EXPORT_SYMBOL(dquot_resume);
2467
2468 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2469                    const struct path *path)
2470 {
2471         int error = security_quota_on(path->dentry);
2472         if (error)
2473                 return error;
2474         /* Quota file not on the same filesystem? */
2475         if (path->dentry->d_sb != sb)
2476                 error = -EXDEV;
2477         else
2478                 error = dquot_load_quota_inode(d_inode(path->dentry), type,
2479                                              format_id, DQUOT_USAGE_ENABLED |
2480                                              DQUOT_LIMITS_ENABLED);
2481         return error;
2482 }
2483 EXPORT_SYMBOL(dquot_quota_on);
2484
2485 /*
2486  * This function is used when filesystem needs to initialize quotas
2487  * during mount time.
2488  */
2489 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2490                 int format_id, int type)
2491 {
2492         struct dentry *dentry;
2493         int error;
2494
2495         dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2496         if (IS_ERR(dentry))
2497                 return PTR_ERR(dentry);
2498
2499         error = security_quota_on(dentry);
2500         if (!error)
2501                 error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2502                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2503
2504         dput(dentry);
2505         return error;
2506 }
2507 EXPORT_SYMBOL(dquot_quota_on_mount);
2508
2509 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2510 {
2511         int ret;
2512         int type;
2513         struct quota_info *dqopt = sb_dqopt(sb);
2514
2515         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2516                 return -ENOSYS;
2517         /* Accounting cannot be turned on while fs is mounted */
2518         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2519         if (!flags)
2520                 return -EINVAL;
2521         for (type = 0; type < MAXQUOTAS; type++) {
2522                 if (!(flags & qtype_enforce_flag(type)))
2523                         continue;
2524                 /* Can't enforce without accounting */
2525                 if (!sb_has_quota_usage_enabled(sb, type)) {
2526                         ret = -EINVAL;
2527                         goto out_err;
2528                 }
2529                 if (sb_has_quota_limits_enabled(sb, type)) {
2530                         ret = -EBUSY;
2531                         goto out_err;
2532                 }
2533                 spin_lock(&dq_state_lock);
2534                 dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2535                 spin_unlock(&dq_state_lock);
2536         }
2537         return 0;
2538 out_err:
2539         /* Backout enforcement enablement we already did */
2540         for (type--; type >= 0; type--)  {
2541                 if (flags & qtype_enforce_flag(type))
2542                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2543         }
2544         /* Error code translation for better compatibility with XFS */
2545         if (ret == -EBUSY)
2546                 ret = -EEXIST;
2547         return ret;
2548 }
2549
2550 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2551 {
2552         int ret;
2553         int type;
2554         struct quota_info *dqopt = sb_dqopt(sb);
2555
2556         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2557                 return -ENOSYS;
2558         /*
2559          * We don't support turning off accounting via quotactl. In principle
2560          * quota infrastructure can do this but filesystems don't expect
2561          * userspace to be able to do it.
2562          */
2563         if (flags &
2564                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2565                 return -EOPNOTSUPP;
2566
2567         /* Filter out limits not enabled */
2568         for (type = 0; type < MAXQUOTAS; type++)
2569                 if (!sb_has_quota_limits_enabled(sb, type))
2570                         flags &= ~qtype_enforce_flag(type);
2571         /* Nothing left? */
2572         if (!flags)
2573                 return -EEXIST;
2574         for (type = 0; type < MAXQUOTAS; type++) {
2575                 if (flags & qtype_enforce_flag(type)) {
2576                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2577                         if (ret < 0)
2578                                 goto out_err;
2579                 }
2580         }
2581         return 0;
2582 out_err:
2583         /* Backout enforcement disabling we already did */
2584         for (type--; type >= 0; type--)  {
2585                 if (flags & qtype_enforce_flag(type)) {
2586                         spin_lock(&dq_state_lock);
2587                         dqopt->flags |=
2588                                 dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2589                         spin_unlock(&dq_state_lock);
2590                 }
2591         }
2592         return ret;
2593 }
2594
2595 /* Generic routine for getting common part of quota structure */
2596 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2597 {
2598         struct mem_dqblk *dm = &dquot->dq_dqb;
2599
2600         memset(di, 0, sizeof(*di));
2601         spin_lock(&dquot->dq_dqb_lock);
2602         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2603         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2604         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2605         di->d_ino_softlimit = dm->dqb_isoftlimit;
2606         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2607         di->d_ino_count = dm->dqb_curinodes;
2608         di->d_spc_timer = dm->dqb_btime;
2609         di->d_ino_timer = dm->dqb_itime;
2610         spin_unlock(&dquot->dq_dqb_lock);
2611 }
2612
2613 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2614                     struct qc_dqblk *di)
2615 {
2616         struct dquot *dquot;
2617
2618         dquot = dqget(sb, qid);
2619         if (IS_ERR(dquot))
2620                 return PTR_ERR(dquot);
2621         do_get_dqblk(dquot, di);
2622         dqput(dquot);
2623
2624         return 0;
2625 }
2626 EXPORT_SYMBOL(dquot_get_dqblk);
2627
2628 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2629                          struct qc_dqblk *di)
2630 {
2631         struct dquot *dquot;
2632         int err;
2633
2634         if (!sb->dq_op->get_next_id)
2635                 return -ENOSYS;
2636         err = sb->dq_op->get_next_id(sb, qid);
2637         if (err < 0)
2638                 return err;
2639         dquot = dqget(sb, *qid);
2640         if (IS_ERR(dquot))
2641                 return PTR_ERR(dquot);
2642         do_get_dqblk(dquot, di);
2643         dqput(dquot);
2644
2645         return 0;
2646 }
2647 EXPORT_SYMBOL(dquot_get_next_dqblk);
2648
2649 #define VFS_QC_MASK \
2650         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2651          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2652          QC_SPC_TIMER | QC_INO_TIMER)
2653
2654 /* Generic routine for setting common part of quota structure */
2655 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2656 {
2657         struct mem_dqblk *dm = &dquot->dq_dqb;
2658         int check_blim = 0, check_ilim = 0;
2659         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2660
2661         if (di->d_fieldmask & ~VFS_QC_MASK)
2662                 return -EINVAL;
2663
2664         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2665              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2666             ((di->d_fieldmask & QC_SPC_HARD) &&
2667              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2668             ((di->d_fieldmask & QC_INO_SOFT) &&
2669              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2670             ((di->d_fieldmask & QC_INO_HARD) &&
2671              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2672                 return -ERANGE;
2673
2674         spin_lock(&dquot->dq_dqb_lock);
2675         if (di->d_fieldmask & QC_SPACE) {
2676                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2677                 check_blim = 1;
2678                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2679         }
2680
2681         if (di->d_fieldmask & QC_SPC_SOFT)
2682                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2683         if (di->d_fieldmask & QC_SPC_HARD)
2684                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2685         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2686                 check_blim = 1;
2687                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2688         }
2689
2690         if (di->d_fieldmask & QC_INO_COUNT) {
2691                 dm->dqb_curinodes = di->d_ino_count;
2692                 check_ilim = 1;
2693                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2694         }
2695
2696         if (di->d_fieldmask & QC_INO_SOFT)
2697                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2698         if (di->d_fieldmask & QC_INO_HARD)
2699                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2700         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2701                 check_ilim = 1;
2702                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2703         }
2704
2705         if (di->d_fieldmask & QC_SPC_TIMER) {
2706                 dm->dqb_btime = di->d_spc_timer;
2707                 check_blim = 1;
2708                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2709         }
2710
2711         if (di->d_fieldmask & QC_INO_TIMER) {
2712                 dm->dqb_itime = di->d_ino_timer;
2713                 check_ilim = 1;
2714                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2715         }
2716
2717         if (check_blim) {
2718                 if (!dm->dqb_bsoftlimit ||
2719                     dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2720                         dm->dqb_btime = 0;
2721                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2722                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2723                         /* Set grace only if user hasn't provided his own... */
2724                         dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2725         }
2726         if (check_ilim) {
2727                 if (!dm->dqb_isoftlimit ||
2728                     dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2729                         dm->dqb_itime = 0;
2730                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2731                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2732                         /* Set grace only if user hasn't provided his own... */
2733                         dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2734         }
2735         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2736             dm->dqb_isoftlimit)
2737                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2738         else
2739                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2740         spin_unlock(&dquot->dq_dqb_lock);
2741         mark_dquot_dirty(dquot);
2742
2743         return 0;
2744 }
2745
2746 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2747                   struct qc_dqblk *di)
2748 {
2749         struct dquot *dquot;
2750         int rc;
2751
2752         dquot = dqget(sb, qid);
2753         if (IS_ERR(dquot)) {
2754                 rc = PTR_ERR(dquot);
2755                 goto out;
2756         }
2757         rc = do_set_dqblk(dquot, di);
2758         dqput(dquot);
2759 out:
2760         return rc;
2761 }
2762 EXPORT_SYMBOL(dquot_set_dqblk);
2763
2764 /* Generic routine for getting common part of quota file information */
2765 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2766 {
2767         struct mem_dqinfo *mi;
2768         struct qc_type_state *tstate;
2769         struct quota_info *dqopt = sb_dqopt(sb);
2770         int type;
2771
2772         memset(state, 0, sizeof(*state));
2773         for (type = 0; type < MAXQUOTAS; type++) {
2774                 if (!sb_has_quota_active(sb, type))
2775                         continue;
2776                 tstate = state->s_state + type;
2777                 mi = sb_dqopt(sb)->info + type;
2778                 tstate->flags = QCI_ACCT_ENABLED;
2779                 spin_lock(&dq_data_lock);
2780                 if (mi->dqi_flags & DQF_SYS_FILE)
2781                         tstate->flags |= QCI_SYSFILE;
2782                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2783                         tstate->flags |= QCI_ROOT_SQUASH;
2784                 if (sb_has_quota_limits_enabled(sb, type))
2785                         tstate->flags |= QCI_LIMITS_ENFORCED;
2786                 tstate->spc_timelimit = mi->dqi_bgrace;
2787                 tstate->ino_timelimit = mi->dqi_igrace;
2788                 if (dqopt->files[type]) {
2789                         tstate->ino = dqopt->files[type]->i_ino;
2790                         tstate->blocks = dqopt->files[type]->i_blocks;
2791                 }
2792                 tstate->nextents = 1;   /* We don't know... */
2793                 spin_unlock(&dq_data_lock);
2794         }
2795         return 0;
2796 }
2797 EXPORT_SYMBOL(dquot_get_state);
2798
2799 /* Generic routine for setting common part of quota file information */
2800 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2801 {
2802         struct mem_dqinfo *mi;
2803         int err = 0;
2804
2805         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2806             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2807                 return -EINVAL;
2808         if (!sb_has_quota_active(sb, type))
2809                 return -ESRCH;
2810         mi = sb_dqopt(sb)->info + type;
2811         if (ii->i_fieldmask & QC_FLAGS) {
2812                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2813                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2814                         return -EINVAL;
2815         }
2816         spin_lock(&dq_data_lock);
2817         if (ii->i_fieldmask & QC_SPC_TIMER)
2818                 mi->dqi_bgrace = ii->i_spc_timelimit;
2819         if (ii->i_fieldmask & QC_INO_TIMER)
2820                 mi->dqi_igrace = ii->i_ino_timelimit;
2821         if (ii->i_fieldmask & QC_FLAGS) {
2822                 if (ii->i_flags & QCI_ROOT_SQUASH)
2823                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2824                 else
2825                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2826         }
2827         spin_unlock(&dq_data_lock);
2828         mark_info_dirty(sb, type);
2829         /* Force write to disk */
2830         sb->dq_op->write_info(sb, type);
2831         return err;
2832 }
2833 EXPORT_SYMBOL(dquot_set_dqinfo);
2834
2835 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2836         .quota_enable   = dquot_quota_enable,
2837         .quota_disable  = dquot_quota_disable,
2838         .quota_sync     = dquot_quota_sync,
2839         .get_state      = dquot_get_state,
2840         .set_info       = dquot_set_dqinfo,
2841         .get_dqblk      = dquot_get_dqblk,
2842         .get_nextdqblk  = dquot_get_next_dqblk,
2843         .set_dqblk      = dquot_set_dqblk
2844 };
2845 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2846
2847 static int do_proc_dqstats(struct ctl_table *table, int write,
2848                      void *buffer, size_t *lenp, loff_t *ppos)
2849 {
2850         unsigned int type = (unsigned long *)table->data - dqstats.stat;
2851         s64 value = percpu_counter_sum(&dqstats.counter[type]);
2852
2853         /* Filter negative values for non-monotonic counters */
2854         if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2855                           type == DQST_FREE_DQUOTS))
2856                 value = 0;
2857
2858         /* Update global table */
2859         dqstats.stat[type] = value;
2860         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2861 }
2862
2863 static struct ctl_table fs_dqstats_table[] = {
2864         {
2865                 .procname       = "lookups",
2866                 .data           = &dqstats.stat[DQST_LOOKUPS],
2867                 .maxlen         = sizeof(unsigned long),
2868                 .mode           = 0444,
2869                 .proc_handler   = do_proc_dqstats,
2870         },
2871         {
2872                 .procname       = "drops",
2873                 .data           = &dqstats.stat[DQST_DROPS],
2874                 .maxlen         = sizeof(unsigned long),
2875                 .mode           = 0444,
2876                 .proc_handler   = do_proc_dqstats,
2877         },
2878         {
2879                 .procname       = "reads",
2880                 .data           = &dqstats.stat[DQST_READS],
2881                 .maxlen         = sizeof(unsigned long),
2882                 .mode           = 0444,
2883                 .proc_handler   = do_proc_dqstats,
2884         },
2885         {
2886                 .procname       = "writes",
2887                 .data           = &dqstats.stat[DQST_WRITES],
2888                 .maxlen         = sizeof(unsigned long),
2889                 .mode           = 0444,
2890                 .proc_handler   = do_proc_dqstats,
2891         },
2892         {
2893                 .procname       = "cache_hits",
2894                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2895                 .maxlen         = sizeof(unsigned long),
2896                 .mode           = 0444,
2897                 .proc_handler   = do_proc_dqstats,
2898         },
2899         {
2900                 .procname       = "allocated_dquots",
2901                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2902                 .maxlen         = sizeof(unsigned long),
2903                 .mode           = 0444,
2904                 .proc_handler   = do_proc_dqstats,
2905         },
2906         {
2907                 .procname       = "free_dquots",
2908                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2909                 .maxlen         = sizeof(unsigned long),
2910                 .mode           = 0444,
2911                 .proc_handler   = do_proc_dqstats,
2912         },
2913         {
2914                 .procname       = "syncs",
2915                 .data           = &dqstats.stat[DQST_SYNCS],
2916                 .maxlen         = sizeof(unsigned long),
2917                 .mode           = 0444,
2918                 .proc_handler   = do_proc_dqstats,
2919         },
2920 #ifdef CONFIG_PRINT_QUOTA_WARNING
2921         {
2922                 .procname       = "warnings",
2923                 .data           = &flag_print_warnings,
2924                 .maxlen         = sizeof(int),
2925                 .mode           = 0644,
2926                 .proc_handler   = proc_dointvec,
2927         },
2928 #endif
2929         { },
2930 };
2931
2932 static struct ctl_table fs_table[] = {
2933         {
2934                 .procname       = "quota",
2935                 .mode           = 0555,
2936                 .child          = fs_dqstats_table,
2937         },
2938         { },
2939 };
2940
2941 static struct ctl_table sys_table[] = {
2942         {
2943                 .procname       = "fs",
2944                 .mode           = 0555,
2945                 .child          = fs_table,
2946         },
2947         { },
2948 };
2949
2950 static int __init dquot_init(void)
2951 {
2952         int i, ret;
2953         unsigned long nr_hash, order;
2954
2955         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2956
2957         register_sysctl_table(sys_table);
2958
2959         dquot_cachep = kmem_cache_create("dquot",
2960                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2961                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2962                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2963                         NULL);
2964
2965         order = 0;
2966         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2967         if (!dquot_hash)
2968                 panic("Cannot create dquot hash table");
2969
2970         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2971                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2972                 if (ret)
2973                         panic("Cannot create dquot stat counters");
2974         }
2975
2976         /* Find power-of-two hlist_heads which can fit into allocation */
2977         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2978         dq_hash_bits = ilog2(nr_hash);
2979
2980         nr_hash = 1UL << dq_hash_bits;
2981         dq_hash_mask = nr_hash - 1;
2982         for (i = 0; i < nr_hash; i++)
2983                 INIT_HLIST_HEAD(dquot_hash + i);
2984
2985         pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2986                 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2987
2988         if (register_shrinker(&dqcache_shrinker))
2989                 panic("Cannot register dquot shrinker");
2990
2991         return 0;
2992 }
2993 fs_initcall(dquot_init);