GNU Linux-libre 5.10.153-gnu1
[releases.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 <linux/sched/mm.h>
83 #include "../internal.h" /* ugh */
84
85 #include <linux/uaccess.h>
86
87 /*
88  * There are five quota SMP locks:
89  * * dq_list_lock protects all lists with quotas and quota formats.
90  * * dquot->dq_dqb_lock protects data from dq_dqb
91  * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
92  *   consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
93  *   dquot_transfer() can stabilize amount it transfers
94  * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
95  *   pointers in the inode
96  * * dq_state_lock protects modifications of quota state (on quotaon and
97  *   quotaoff) and readers who care about latest values take it as well.
98  *
99  * The spinlock ordering is hence:
100  *   dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
101  *   dq_list_lock > dq_state_lock
102  *
103  * Note that some things (eg. sb pointer, type, id) doesn't change during
104  * the life of the dquot structure and so needn't to be protected by a lock
105  *
106  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
107  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
108  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
109  * inode and before dropping dquot references to avoid use of dquots after
110  * they are freed. dq_data_lock is used to serialize the pointer setting and
111  * clearing operations.
112  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
113  * inode is a quota file). Functions adding pointers from inode to dquots have
114  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
115  * have to do all pointer modifications before dropping dq_data_lock. This makes
116  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
117  * then drops all pointers to dquots from an inode.
118  *
119  * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
120  * memory (or space for it is being allocated) on the first dqget(), when it is
121  * being written out, and when it is being released on the last dqput(). The
122  * allocation and release operations are serialized by the dq_lock and by
123  * checking the use count in dquot_release().
124  *
125  * Lock ordering (including related VFS locks) is the following:
126  *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
127  */
128
129 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
130 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
131 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
132 EXPORT_SYMBOL(dq_data_lock);
133 DEFINE_STATIC_SRCU(dquot_srcu);
134
135 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
136
137 void __quota_error(struct super_block *sb, const char *func,
138                    const char *fmt, ...)
139 {
140         if (printk_ratelimit()) {
141                 va_list args;
142                 struct va_format vaf;
143
144                 va_start(args, fmt);
145
146                 vaf.fmt = fmt;
147                 vaf.va = &args;
148
149                 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
150                        sb->s_id, func, &vaf);
151
152                 va_end(args);
153         }
154 }
155 EXPORT_SYMBOL(__quota_error);
156
157 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
158 static char *quotatypes[] = INITQFNAMES;
159 #endif
160 static struct quota_format_type *quota_formats; /* List of registered formats */
161 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
162
163 /* SLAB cache for dquot structures */
164 static struct kmem_cache *dquot_cachep;
165
166 int register_quota_format(struct quota_format_type *fmt)
167 {
168         spin_lock(&dq_list_lock);
169         fmt->qf_next = quota_formats;
170         quota_formats = fmt;
171         spin_unlock(&dq_list_lock);
172         return 0;
173 }
174 EXPORT_SYMBOL(register_quota_format);
175
176 void unregister_quota_format(struct quota_format_type *fmt)
177 {
178         struct quota_format_type **actqf;
179
180         spin_lock(&dq_list_lock);
181         for (actqf = &quota_formats; *actqf && *actqf != fmt;
182              actqf = &(*actqf)->qf_next)
183                 ;
184         if (*actqf)
185                 *actqf = (*actqf)->qf_next;
186         spin_unlock(&dq_list_lock);
187 }
188 EXPORT_SYMBOL(unregister_quota_format);
189
190 static struct quota_format_type *find_quota_format(int id)
191 {
192         struct quota_format_type *actqf;
193
194         spin_lock(&dq_list_lock);
195         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
196              actqf = actqf->qf_next)
197                 ;
198         if (!actqf || !try_module_get(actqf->qf_owner)) {
199                 int qm;
200
201                 spin_unlock(&dq_list_lock);
202
203                 for (qm = 0; module_names[qm].qm_fmt_id &&
204                              module_names[qm].qm_fmt_id != id; qm++)
205                         ;
206                 if (!module_names[qm].qm_fmt_id ||
207                     request_module(module_names[qm].qm_mod_name))
208                         return NULL;
209
210                 spin_lock(&dq_list_lock);
211                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
212                      actqf = actqf->qf_next)
213                         ;
214                 if (actqf && !try_module_get(actqf->qf_owner))
215                         actqf = NULL;
216         }
217         spin_unlock(&dq_list_lock);
218         return actqf;
219 }
220
221 static void put_quota_format(struct quota_format_type *fmt)
222 {
223         module_put(fmt->qf_owner);
224 }
225
226 /*
227  * Dquot List Management:
228  * The quota code uses four lists for dquot management: the inuse_list,
229  * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
230  * structure may be on some of those lists, depending on its current state.
231  *
232  * All dquots are placed to the end of inuse_list when first created, and this
233  * list is used for invalidate operation, which must look at every dquot.
234  *
235  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
236  * and this list is searched whenever we need an available dquot.  Dquots are
237  * removed from the list as soon as they are used again, and
238  * dqstats.free_dquots gives the number of dquots on the list. When
239  * dquot is invalidated it's completely released from memory.
240  *
241  * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
242  * dirtied, and this list is searched when writing dirty dquots back to
243  * quota file. Note that some filesystems do dirty dquot tracking on their
244  * own (e.g. in a journal) and thus don't use dqi_dirty_list.
245  *
246  * Dquots with a specific identity (device, type and id) are placed on
247  * one of the dquot_hash[] hash chains. The provides an efficient search
248  * mechanism to locate a specific dquot.
249  */
250
251 static LIST_HEAD(inuse_list);
252 static LIST_HEAD(free_dquots);
253 static unsigned int dq_hash_bits, dq_hash_mask;
254 static struct hlist_head *dquot_hash;
255
256 struct dqstats dqstats;
257 EXPORT_SYMBOL(dqstats);
258
259 static qsize_t inode_get_rsv_space(struct inode *inode);
260 static qsize_t __inode_get_rsv_space(struct inode *inode);
261 static int __dquot_initialize(struct inode *inode, int type);
262
263 static inline unsigned int
264 hashfn(const struct super_block *sb, struct kqid qid)
265 {
266         unsigned int id = from_kqid(&init_user_ns, qid);
267         int type = qid.type;
268         unsigned long tmp;
269
270         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
271         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
272 }
273
274 /*
275  * Following list functions expect dq_list_lock to be held
276  */
277 static inline void insert_dquot_hash(struct dquot *dquot)
278 {
279         struct hlist_head *head;
280         head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
281         hlist_add_head(&dquot->dq_hash, head);
282 }
283
284 static inline void remove_dquot_hash(struct dquot *dquot)
285 {
286         hlist_del_init(&dquot->dq_hash);
287 }
288
289 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
290                                 struct kqid qid)
291 {
292         struct hlist_node *node;
293         struct dquot *dquot;
294
295         hlist_for_each (node, dquot_hash+hashent) {
296                 dquot = hlist_entry(node, struct dquot, dq_hash);
297                 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
298                         return dquot;
299         }
300         return NULL;
301 }
302
303 /* Add a dquot to the tail of the free list */
304 static inline void put_dquot_last(struct dquot *dquot)
305 {
306         list_add_tail(&dquot->dq_free, &free_dquots);
307         dqstats_inc(DQST_FREE_DQUOTS);
308 }
309
310 static inline void remove_free_dquot(struct dquot *dquot)
311 {
312         if (list_empty(&dquot->dq_free))
313                 return;
314         list_del_init(&dquot->dq_free);
315         dqstats_dec(DQST_FREE_DQUOTS);
316 }
317
318 static inline void put_inuse(struct dquot *dquot)
319 {
320         /* We add to the back of inuse list so we don't have to restart
321          * when traversing this list and we block */
322         list_add_tail(&dquot->dq_inuse, &inuse_list);
323         dqstats_inc(DQST_ALLOC_DQUOTS);
324 }
325
326 static inline void remove_inuse(struct dquot *dquot)
327 {
328         dqstats_dec(DQST_ALLOC_DQUOTS);
329         list_del(&dquot->dq_inuse);
330 }
331 /*
332  * End of list functions needing dq_list_lock
333  */
334
335 static void wait_on_dquot(struct dquot *dquot)
336 {
337         mutex_lock(&dquot->dq_lock);
338         mutex_unlock(&dquot->dq_lock);
339 }
340
341 static inline int dquot_dirty(struct dquot *dquot)
342 {
343         return test_bit(DQ_MOD_B, &dquot->dq_flags);
344 }
345
346 static inline int mark_dquot_dirty(struct dquot *dquot)
347 {
348         return dquot->dq_sb->dq_op->mark_dirty(dquot);
349 }
350
351 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
352 int dquot_mark_dquot_dirty(struct dquot *dquot)
353 {
354         int ret = 1;
355
356         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
357                 return 0;
358
359         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
360                 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
361
362         /* If quota is dirty already, we don't have to acquire dq_list_lock */
363         if (test_bit(DQ_MOD_B, &dquot->dq_flags))
364                 return 1;
365
366         spin_lock(&dq_list_lock);
367         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
368                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
369                                 info[dquot->dq_id.type].dqi_dirty_list);
370                 ret = 0;
371         }
372         spin_unlock(&dq_list_lock);
373         return ret;
374 }
375 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
376
377 /* Dirtify all the dquots - this can block when journalling */
378 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
379 {
380         int ret, err, cnt;
381
382         ret = err = 0;
383         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
384                 if (dquot[cnt])
385                         /* Even in case of error we have to continue */
386                         ret = mark_dquot_dirty(dquot[cnt]);
387                 if (!err)
388                         err = ret;
389         }
390         return err;
391 }
392
393 static inline void dqput_all(struct dquot **dquot)
394 {
395         unsigned int cnt;
396
397         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
398                 dqput(dquot[cnt]);
399 }
400
401 static inline int clear_dquot_dirty(struct dquot *dquot)
402 {
403         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
404                 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
405
406         spin_lock(&dq_list_lock);
407         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
408                 spin_unlock(&dq_list_lock);
409                 return 0;
410         }
411         list_del_init(&dquot->dq_dirty);
412         spin_unlock(&dq_list_lock);
413         return 1;
414 }
415
416 void mark_info_dirty(struct super_block *sb, int type)
417 {
418         spin_lock(&dq_data_lock);
419         sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
420         spin_unlock(&dq_data_lock);
421 }
422 EXPORT_SYMBOL(mark_info_dirty);
423
424 /*
425  *      Read dquot from disk and alloc space for it
426  */
427
428 int dquot_acquire(struct dquot *dquot)
429 {
430         int ret = 0, ret2 = 0;
431         unsigned int memalloc;
432         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
433
434         mutex_lock(&dquot->dq_lock);
435         memalloc = memalloc_nofs_save();
436         if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
437                 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
438                 if (ret < 0)
439                         goto out_iolock;
440         }
441         /* Make sure flags update is visible after dquot has been filled */
442         smp_mb__before_atomic();
443         set_bit(DQ_READ_B, &dquot->dq_flags);
444         /* Instantiate dquot if needed */
445         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
446                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
447                 /* Write the info if needed */
448                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
449                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
450                                         dquot->dq_sb, dquot->dq_id.type);
451                 }
452                 if (ret < 0)
453                         goto out_iolock;
454                 if (ret2 < 0) {
455                         ret = ret2;
456                         goto out_iolock;
457                 }
458         }
459         /*
460          * Make sure flags update is visible after on-disk struct has been
461          * allocated. Paired with smp_rmb() in dqget().
462          */
463         smp_mb__before_atomic();
464         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
465 out_iolock:
466         memalloc_nofs_restore(memalloc);
467         mutex_unlock(&dquot->dq_lock);
468         return ret;
469 }
470 EXPORT_SYMBOL(dquot_acquire);
471
472 /*
473  *      Write dquot to disk
474  */
475 int dquot_commit(struct dquot *dquot)
476 {
477         int ret = 0;
478         unsigned int memalloc;
479         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
480
481         mutex_lock(&dquot->dq_lock);
482         memalloc = memalloc_nofs_save();
483         if (!clear_dquot_dirty(dquot))
484                 goto out_lock;
485         /* Inactive dquot can be only if there was error during read/init
486          * => we have better not writing it */
487         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
488                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
489         else
490                 ret = -EIO;
491 out_lock:
492         memalloc_nofs_restore(memalloc);
493         mutex_unlock(&dquot->dq_lock);
494         return ret;
495 }
496 EXPORT_SYMBOL(dquot_commit);
497
498 /*
499  *      Release dquot
500  */
501 int dquot_release(struct dquot *dquot)
502 {
503         int ret = 0, ret2 = 0;
504         unsigned int memalloc;
505         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
506
507         mutex_lock(&dquot->dq_lock);
508         memalloc = memalloc_nofs_save();
509         /* Check whether we are not racing with some other dqget() */
510         if (dquot_is_busy(dquot))
511                 goto out_dqlock;
512         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
513                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
514                 /* Write the info */
515                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
516                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
517                                                 dquot->dq_sb, dquot->dq_id.type);
518                 }
519                 if (ret >= 0)
520                         ret = ret2;
521         }
522         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
523 out_dqlock:
524         memalloc_nofs_restore(memalloc);
525         mutex_unlock(&dquot->dq_lock);
526         return ret;
527 }
528 EXPORT_SYMBOL(dquot_release);
529
530 void dquot_destroy(struct dquot *dquot)
531 {
532         kmem_cache_free(dquot_cachep, dquot);
533 }
534 EXPORT_SYMBOL(dquot_destroy);
535
536 static inline void do_destroy_dquot(struct dquot *dquot)
537 {
538         dquot->dq_sb->dq_op->destroy_dquot(dquot);
539 }
540
541 /* Invalidate all dquots on the list. Note that this function is called after
542  * quota is disabled and pointers from inodes removed so there cannot be new
543  * quota users. There can still be some users of quotas due to inodes being
544  * just deleted or pruned by prune_icache() (those are not attached to any
545  * list) or parallel quotactl call. We have to wait for such users.
546  */
547 static void invalidate_dquots(struct super_block *sb, int type)
548 {
549         struct dquot *dquot, *tmp;
550
551 restart:
552         spin_lock(&dq_list_lock);
553         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
554                 if (dquot->dq_sb != sb)
555                         continue;
556                 if (dquot->dq_id.type != type)
557                         continue;
558                 /* Wait for dquot users */
559                 if (atomic_read(&dquot->dq_count)) {
560                         dqgrab(dquot);
561                         spin_unlock(&dq_list_lock);
562                         /*
563                          * Once dqput() wakes us up, we know it's time to free
564                          * the dquot.
565                          * IMPORTANT: we rely on the fact that there is always
566                          * at most one process waiting for dquot to free.
567                          * Otherwise dq_count would be > 1 and we would never
568                          * wake up.
569                          */
570                         wait_event(dquot_ref_wq,
571                                    atomic_read(&dquot->dq_count) == 1);
572                         dqput(dquot);
573                         /* At this moment dquot() need not exist (it could be
574                          * reclaimed by prune_dqcache(). Hence we must
575                          * restart. */
576                         goto restart;
577                 }
578                 /*
579                  * Quota now has no users and it has been written on last
580                  * dqput()
581                  */
582                 remove_dquot_hash(dquot);
583                 remove_free_dquot(dquot);
584                 remove_inuse(dquot);
585                 do_destroy_dquot(dquot);
586         }
587         spin_unlock(&dq_list_lock);
588 }
589
590 /* Call callback for every active dquot on given filesystem */
591 int dquot_scan_active(struct super_block *sb,
592                       int (*fn)(struct dquot *dquot, unsigned long priv),
593                       unsigned long priv)
594 {
595         struct dquot *dquot, *old_dquot = NULL;
596         int ret = 0;
597
598         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
599
600         spin_lock(&dq_list_lock);
601         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
602                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
603                         continue;
604                 if (dquot->dq_sb != sb)
605                         continue;
606                 /* Now we have active dquot so we can just increase use count */
607                 atomic_inc(&dquot->dq_count);
608                 spin_unlock(&dq_list_lock);
609                 dqput(old_dquot);
610                 old_dquot = dquot;
611                 /*
612                  * ->release_dquot() can be racing with us. Our reference
613                  * protects us from new calls to it so just wait for any
614                  * outstanding call and recheck the DQ_ACTIVE_B after that.
615                  */
616                 wait_on_dquot(dquot);
617                 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
618                         ret = fn(dquot, priv);
619                         if (ret < 0)
620                                 goto out;
621                 }
622                 spin_lock(&dq_list_lock);
623                 /* We are safe to continue now because our dquot could not
624                  * be moved out of the inuse list while we hold the reference */
625         }
626         spin_unlock(&dq_list_lock);
627 out:
628         dqput(old_dquot);
629         return ret;
630 }
631 EXPORT_SYMBOL(dquot_scan_active);
632
633 /* Write all dquot structures to quota files */
634 int dquot_writeback_dquots(struct super_block *sb, int type)
635 {
636         struct list_head dirty;
637         struct dquot *dquot;
638         struct quota_info *dqopt = sb_dqopt(sb);
639         int cnt;
640         int err, ret = 0;
641
642         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
643
644         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
645                 if (type != -1 && cnt != type)
646                         continue;
647                 if (!sb_has_quota_active(sb, cnt))
648                         continue;
649                 spin_lock(&dq_list_lock);
650                 /* Move list away to avoid livelock. */
651                 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
652                 while (!list_empty(&dirty)) {
653                         dquot = list_first_entry(&dirty, struct dquot,
654                                                  dq_dirty);
655
656                         WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
657
658                         /* Now we have active dquot from which someone is
659                          * holding reference so we can safely just increase
660                          * use count */
661                         dqgrab(dquot);
662                         spin_unlock(&dq_list_lock);
663                         err = sb->dq_op->write_dquot(dquot);
664                         if (err) {
665                                 /*
666                                  * Clear dirty bit anyway to avoid infinite
667                                  * loop here.
668                                  */
669                                 clear_dquot_dirty(dquot);
670                                 if (!ret)
671                                         ret = err;
672                         }
673                         dqput(dquot);
674                         spin_lock(&dq_list_lock);
675                 }
676                 spin_unlock(&dq_list_lock);
677         }
678
679         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
680                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
681                     && info_dirty(&dqopt->info[cnt]))
682                         sb->dq_op->write_info(sb, cnt);
683         dqstats_inc(DQST_SYNCS);
684
685         return ret;
686 }
687 EXPORT_SYMBOL(dquot_writeback_dquots);
688
689 /* Write all dquot structures to disk and make them visible from userspace */
690 int dquot_quota_sync(struct super_block *sb, int type)
691 {
692         struct quota_info *dqopt = sb_dqopt(sb);
693         int cnt;
694         int ret;
695
696         ret = dquot_writeback_dquots(sb, type);
697         if (ret)
698                 return ret;
699         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
700                 return 0;
701
702         /* This is not very clever (and fast) but currently I don't know about
703          * any other simple way of getting quota data to disk and we must get
704          * them there for userspace to be visible... */
705         if (sb->s_op->sync_fs) {
706                 ret = sb->s_op->sync_fs(sb, 1);
707                 if (ret)
708                         return ret;
709         }
710         ret = sync_blockdev(sb->s_bdev);
711         if (ret)
712                 return ret;
713
714         /*
715          * Now when everything is written we can discard the pagecache so
716          * that userspace sees the changes.
717          */
718         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
719                 if (type != -1 && cnt != type)
720                         continue;
721                 if (!sb_has_quota_active(sb, cnt))
722                         continue;
723                 inode_lock(dqopt->files[cnt]);
724                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
725                 inode_unlock(dqopt->files[cnt]);
726         }
727
728         return 0;
729 }
730 EXPORT_SYMBOL(dquot_quota_sync);
731
732 static unsigned long
733 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
734 {
735         struct dquot *dquot;
736         unsigned long freed = 0;
737
738         spin_lock(&dq_list_lock);
739         while (!list_empty(&free_dquots) && sc->nr_to_scan) {
740                 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
741                 remove_dquot_hash(dquot);
742                 remove_free_dquot(dquot);
743                 remove_inuse(dquot);
744                 do_destroy_dquot(dquot);
745                 sc->nr_to_scan--;
746                 freed++;
747         }
748         spin_unlock(&dq_list_lock);
749         return freed;
750 }
751
752 static unsigned long
753 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
754 {
755         return vfs_pressure_ratio(
756         percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
757 }
758
759 static struct shrinker dqcache_shrinker = {
760         .count_objects = dqcache_shrink_count,
761         .scan_objects = dqcache_shrink_scan,
762         .seeks = DEFAULT_SEEKS,
763 };
764
765 /*
766  * Put reference to dquot
767  */
768 void dqput(struct dquot *dquot)
769 {
770         int ret;
771
772         if (!dquot)
773                 return;
774 #ifdef CONFIG_QUOTA_DEBUG
775         if (!atomic_read(&dquot->dq_count)) {
776                 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
777                             quotatypes[dquot->dq_id.type],
778                             from_kqid(&init_user_ns, dquot->dq_id));
779                 BUG();
780         }
781 #endif
782         dqstats_inc(DQST_DROPS);
783 we_slept:
784         spin_lock(&dq_list_lock);
785         if (atomic_read(&dquot->dq_count) > 1) {
786                 /* We have more than one user... nothing to do */
787                 atomic_dec(&dquot->dq_count);
788                 /* Releasing dquot during quotaoff phase? */
789                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
790                     atomic_read(&dquot->dq_count) == 1)
791                         wake_up(&dquot_ref_wq);
792                 spin_unlock(&dq_list_lock);
793                 return;
794         }
795         /* Need to release dquot? */
796         if (dquot_dirty(dquot)) {
797                 spin_unlock(&dq_list_lock);
798                 /* Commit dquot before releasing */
799                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
800                 if (ret < 0) {
801                         quota_error(dquot->dq_sb, "Can't write quota structure"
802                                     " (error %d). Quota may get out of sync!",
803                                     ret);
804                         /*
805                          * We clear dirty bit anyway, so that we avoid
806                          * infinite loop here
807                          */
808                         clear_dquot_dirty(dquot);
809                 }
810                 goto we_slept;
811         }
812         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
813                 spin_unlock(&dq_list_lock);
814                 dquot->dq_sb->dq_op->release_dquot(dquot);
815                 goto we_slept;
816         }
817         atomic_dec(&dquot->dq_count);
818 #ifdef CONFIG_QUOTA_DEBUG
819         /* sanity check */
820         BUG_ON(!list_empty(&dquot->dq_free));
821 #endif
822         put_dquot_last(dquot);
823         spin_unlock(&dq_list_lock);
824 }
825 EXPORT_SYMBOL(dqput);
826
827 struct dquot *dquot_alloc(struct super_block *sb, int type)
828 {
829         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
830 }
831 EXPORT_SYMBOL(dquot_alloc);
832
833 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
834 {
835         struct dquot *dquot;
836
837         dquot = sb->dq_op->alloc_dquot(sb, type);
838         if(!dquot)
839                 return NULL;
840
841         mutex_init(&dquot->dq_lock);
842         INIT_LIST_HEAD(&dquot->dq_free);
843         INIT_LIST_HEAD(&dquot->dq_inuse);
844         INIT_HLIST_NODE(&dquot->dq_hash);
845         INIT_LIST_HEAD(&dquot->dq_dirty);
846         dquot->dq_sb = sb;
847         dquot->dq_id = make_kqid_invalid(type);
848         atomic_set(&dquot->dq_count, 1);
849         spin_lock_init(&dquot->dq_dqb_lock);
850
851         return dquot;
852 }
853
854 /*
855  * Get reference to dquot
856  *
857  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
858  * destroying our dquot by:
859  *   a) checking for quota flags under dq_list_lock and
860  *   b) getting a reference to dquot before we release dq_list_lock
861  */
862 struct dquot *dqget(struct super_block *sb, struct kqid qid)
863 {
864         unsigned int hashent = hashfn(sb, qid);
865         struct dquot *dquot, *empty = NULL;
866
867         if (!qid_has_mapping(sb->s_user_ns, qid))
868                 return ERR_PTR(-EINVAL);
869
870         if (!sb_has_quota_active(sb, qid.type))
871                 return ERR_PTR(-ESRCH);
872 we_slept:
873         spin_lock(&dq_list_lock);
874         spin_lock(&dq_state_lock);
875         if (!sb_has_quota_active(sb, qid.type)) {
876                 spin_unlock(&dq_state_lock);
877                 spin_unlock(&dq_list_lock);
878                 dquot = ERR_PTR(-ESRCH);
879                 goto out;
880         }
881         spin_unlock(&dq_state_lock);
882
883         dquot = find_dquot(hashent, sb, qid);
884         if (!dquot) {
885                 if (!empty) {
886                         spin_unlock(&dq_list_lock);
887                         empty = get_empty_dquot(sb, qid.type);
888                         if (!empty)
889                                 schedule();     /* Try to wait for a moment... */
890                         goto we_slept;
891                 }
892                 dquot = empty;
893                 empty = NULL;
894                 dquot->dq_id = qid;
895                 /* all dquots go on the inuse_list */
896                 put_inuse(dquot);
897                 /* hash it first so it can be found */
898                 insert_dquot_hash(dquot);
899                 spin_unlock(&dq_list_lock);
900                 dqstats_inc(DQST_LOOKUPS);
901         } else {
902                 if (!atomic_read(&dquot->dq_count))
903                         remove_free_dquot(dquot);
904                 atomic_inc(&dquot->dq_count);
905                 spin_unlock(&dq_list_lock);
906                 dqstats_inc(DQST_CACHE_HITS);
907                 dqstats_inc(DQST_LOOKUPS);
908         }
909         /* Wait for dq_lock - after this we know that either dquot_release() is
910          * already finished or it will be canceled due to dq_count > 1 test */
911         wait_on_dquot(dquot);
912         /* Read the dquot / allocate space in quota file */
913         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
914                 int err;
915
916                 err = sb->dq_op->acquire_dquot(dquot);
917                 if (err < 0) {
918                         dqput(dquot);
919                         dquot = ERR_PTR(err);
920                         goto out;
921                 }
922         }
923         /*
924          * Make sure following reads see filled structure - paired with
925          * smp_mb__before_atomic() in dquot_acquire().
926          */
927         smp_rmb();
928 #ifdef CONFIG_QUOTA_DEBUG
929         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
930 #endif
931 out:
932         if (empty)
933                 do_destroy_dquot(empty);
934
935         return dquot;
936 }
937 EXPORT_SYMBOL(dqget);
938
939 static inline struct dquot **i_dquot(struct inode *inode)
940 {
941         return inode->i_sb->s_op->get_dquots(inode);
942 }
943
944 static int dqinit_needed(struct inode *inode, int type)
945 {
946         struct dquot * const *dquots;
947         int cnt;
948
949         if (IS_NOQUOTA(inode))
950                 return 0;
951
952         dquots = i_dquot(inode);
953         if (type != -1)
954                 return !dquots[type];
955         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
956                 if (!dquots[cnt])
957                         return 1;
958         return 0;
959 }
960
961 /* This routine is guarded by s_umount semaphore */
962 static int add_dquot_ref(struct super_block *sb, int type)
963 {
964         struct inode *inode, *old_inode = NULL;
965 #ifdef CONFIG_QUOTA_DEBUG
966         int reserved = 0;
967 #endif
968         int err = 0;
969
970         spin_lock(&sb->s_inode_list_lock);
971         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
972                 spin_lock(&inode->i_lock);
973                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
974                     !atomic_read(&inode->i_writecount) ||
975                     !dqinit_needed(inode, type)) {
976                         spin_unlock(&inode->i_lock);
977                         continue;
978                 }
979                 __iget(inode);
980                 spin_unlock(&inode->i_lock);
981                 spin_unlock(&sb->s_inode_list_lock);
982
983 #ifdef CONFIG_QUOTA_DEBUG
984                 if (unlikely(inode_get_rsv_space(inode) > 0))
985                         reserved = 1;
986 #endif
987                 iput(old_inode);
988                 err = __dquot_initialize(inode, type);
989                 if (err) {
990                         iput(inode);
991                         goto out;
992                 }
993
994                 /*
995                  * We hold a reference to 'inode' so it couldn't have been
996                  * removed from s_inodes list while we dropped the
997                  * s_inode_list_lock. We cannot iput the inode now as we can be
998                  * holding the last reference and we cannot iput it under
999                  * s_inode_list_lock. So we keep the reference and iput it
1000                  * later.
1001                  */
1002                 old_inode = inode;
1003                 cond_resched();
1004                 spin_lock(&sb->s_inode_list_lock);
1005         }
1006         spin_unlock(&sb->s_inode_list_lock);
1007         iput(old_inode);
1008 out:
1009 #ifdef CONFIG_QUOTA_DEBUG
1010         if (reserved) {
1011                 quota_error(sb, "Writes happened before quota was turned on "
1012                         "thus quota information is probably inconsistent. "
1013                         "Please run quotacheck(8)");
1014         }
1015 #endif
1016         return err;
1017 }
1018
1019 /*
1020  * Remove references to dquots from inode and add dquot to list for freeing
1021  * if we have the last reference to dquot
1022  */
1023 static void remove_inode_dquot_ref(struct inode *inode, int type,
1024                                    struct list_head *tofree_head)
1025 {
1026         struct dquot **dquots = i_dquot(inode);
1027         struct dquot *dquot = dquots[type];
1028
1029         if (!dquot)
1030                 return;
1031
1032         dquots[type] = NULL;
1033         if (list_empty(&dquot->dq_free)) {
1034                 /*
1035                  * The inode still has reference to dquot so it can't be in the
1036                  * free list
1037                  */
1038                 spin_lock(&dq_list_lock);
1039                 list_add(&dquot->dq_free, tofree_head);
1040                 spin_unlock(&dq_list_lock);
1041         } else {
1042                 /*
1043                  * Dquot is already in a list to put so we won't drop the last
1044                  * reference here.
1045                  */
1046                 dqput(dquot);
1047         }
1048 }
1049
1050 /*
1051  * Free list of dquots
1052  * Dquots are removed from inodes and no new references can be got so we are
1053  * the only ones holding reference
1054  */
1055 static void put_dquot_list(struct list_head *tofree_head)
1056 {
1057         struct list_head *act_head;
1058         struct dquot *dquot;
1059
1060         act_head = tofree_head->next;
1061         while (act_head != tofree_head) {
1062                 dquot = list_entry(act_head, struct dquot, dq_free);
1063                 act_head = act_head->next;
1064                 /* Remove dquot from the list so we won't have problems... */
1065                 list_del_init(&dquot->dq_free);
1066                 dqput(dquot);
1067         }
1068 }
1069
1070 static void remove_dquot_ref(struct super_block *sb, int type,
1071                 struct list_head *tofree_head)
1072 {
1073         struct inode *inode;
1074 #ifdef CONFIG_QUOTA_DEBUG
1075         int reserved = 0;
1076 #endif
1077
1078         spin_lock(&sb->s_inode_list_lock);
1079         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1080                 /*
1081                  *  We have to scan also I_NEW inodes because they can already
1082                  *  have quota pointer initialized. Luckily, we need to touch
1083                  *  only quota pointers and these have separate locking
1084                  *  (dq_data_lock).
1085                  */
1086                 spin_lock(&dq_data_lock);
1087                 if (!IS_NOQUOTA(inode)) {
1088 #ifdef CONFIG_QUOTA_DEBUG
1089                         if (unlikely(inode_get_rsv_space(inode) > 0))
1090                                 reserved = 1;
1091 #endif
1092                         remove_inode_dquot_ref(inode, type, tofree_head);
1093                 }
1094                 spin_unlock(&dq_data_lock);
1095         }
1096         spin_unlock(&sb->s_inode_list_lock);
1097 #ifdef CONFIG_QUOTA_DEBUG
1098         if (reserved) {
1099                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1100                         " was disabled thus quota information is probably "
1101                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1102         }
1103 #endif
1104 }
1105
1106 /* Gather all references from inodes and drop them */
1107 static void drop_dquot_ref(struct super_block *sb, int type)
1108 {
1109         LIST_HEAD(tofree_head);
1110
1111         if (sb->dq_op) {
1112                 remove_dquot_ref(sb, type, &tofree_head);
1113                 synchronize_srcu(&dquot_srcu);
1114                 put_dquot_list(&tofree_head);
1115         }
1116 }
1117
1118 static inline
1119 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1120 {
1121         if (dquot->dq_dqb.dqb_rsvspace >= number)
1122                 dquot->dq_dqb.dqb_rsvspace -= number;
1123         else {
1124                 WARN_ON_ONCE(1);
1125                 dquot->dq_dqb.dqb_rsvspace = 0;
1126         }
1127         if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1128             dquot->dq_dqb.dqb_bsoftlimit)
1129                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1130         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1131 }
1132
1133 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1134 {
1135         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1136             dquot->dq_dqb.dqb_curinodes >= number)
1137                 dquot->dq_dqb.dqb_curinodes -= number;
1138         else
1139                 dquot->dq_dqb.dqb_curinodes = 0;
1140         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1141                 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1142         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1143 }
1144
1145 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1146 {
1147         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1148             dquot->dq_dqb.dqb_curspace >= number)
1149                 dquot->dq_dqb.dqb_curspace -= number;
1150         else
1151                 dquot->dq_dqb.dqb_curspace = 0;
1152         if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1153             dquot->dq_dqb.dqb_bsoftlimit)
1154                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1155         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1156 }
1157
1158 struct dquot_warn {
1159         struct super_block *w_sb;
1160         struct kqid w_dq_id;
1161         short w_type;
1162 };
1163
1164 static int warning_issued(struct dquot *dquot, const int warntype)
1165 {
1166         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1167                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1168                 ((warntype == QUOTA_NL_IHARDWARN ||
1169                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1170
1171         if (!flag)
1172                 return 0;
1173         return test_and_set_bit(flag, &dquot->dq_flags);
1174 }
1175
1176 #ifdef CONFIG_PRINT_QUOTA_WARNING
1177 static int flag_print_warnings = 1;
1178
1179 static int need_print_warning(struct dquot_warn *warn)
1180 {
1181         if (!flag_print_warnings)
1182                 return 0;
1183
1184         switch (warn->w_dq_id.type) {
1185                 case USRQUOTA:
1186                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1187                 case GRPQUOTA:
1188                         return in_group_p(warn->w_dq_id.gid);
1189                 case PRJQUOTA:
1190                         return 1;
1191         }
1192         return 0;
1193 }
1194
1195 /* Print warning to user which exceeded quota */
1196 static void print_warning(struct dquot_warn *warn)
1197 {
1198         char *msg = NULL;
1199         struct tty_struct *tty;
1200         int warntype = warn->w_type;
1201
1202         if (warntype == QUOTA_NL_IHARDBELOW ||
1203             warntype == QUOTA_NL_ISOFTBELOW ||
1204             warntype == QUOTA_NL_BHARDBELOW ||
1205             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1206                 return;
1207
1208         tty = get_current_tty();
1209         if (!tty)
1210                 return;
1211         tty_write_message(tty, warn->w_sb->s_id);
1212         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1213                 tty_write_message(tty, ": warning, ");
1214         else
1215                 tty_write_message(tty, ": write failed, ");
1216         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1217         switch (warntype) {
1218                 case QUOTA_NL_IHARDWARN:
1219                         msg = " file limit reached.\r\n";
1220                         break;
1221                 case QUOTA_NL_ISOFTLONGWARN:
1222                         msg = " file quota exceeded too long.\r\n";
1223                         break;
1224                 case QUOTA_NL_ISOFTWARN:
1225                         msg = " file quota exceeded.\r\n";
1226                         break;
1227                 case QUOTA_NL_BHARDWARN:
1228                         msg = " block limit reached.\r\n";
1229                         break;
1230                 case QUOTA_NL_BSOFTLONGWARN:
1231                         msg = " block quota exceeded too long.\r\n";
1232                         break;
1233                 case QUOTA_NL_BSOFTWARN:
1234                         msg = " block quota exceeded.\r\n";
1235                         break;
1236         }
1237         tty_write_message(tty, msg);
1238         tty_kref_put(tty);
1239 }
1240 #endif
1241
1242 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1243                             int warntype)
1244 {
1245         if (warning_issued(dquot, warntype))
1246                 return;
1247         warn->w_type = warntype;
1248         warn->w_sb = dquot->dq_sb;
1249         warn->w_dq_id = dquot->dq_id;
1250 }
1251
1252 /*
1253  * Write warnings to the console and send warning messages over netlink.
1254  *
1255  * Note that this function can call into tty and networking code.
1256  */
1257 static void flush_warnings(struct dquot_warn *warn)
1258 {
1259         int i;
1260
1261         for (i = 0; i < MAXQUOTAS; i++) {
1262                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1263                         continue;
1264 #ifdef CONFIG_PRINT_QUOTA_WARNING
1265                 print_warning(&warn[i]);
1266 #endif
1267                 quota_send_warning(warn[i].w_dq_id,
1268                                    warn[i].w_sb->s_dev, warn[i].w_type);
1269         }
1270 }
1271
1272 static int ignore_hardlimit(struct dquot *dquot)
1273 {
1274         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1275
1276         return capable(CAP_SYS_RESOURCE) &&
1277                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1278                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1279 }
1280
1281 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1282                             struct dquot_warn *warn)
1283 {
1284         qsize_t newinodes;
1285         int ret = 0;
1286
1287         spin_lock(&dquot->dq_dqb_lock);
1288         newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1289         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1290             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1291                 goto add;
1292
1293         if (dquot->dq_dqb.dqb_ihardlimit &&
1294             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1295             !ignore_hardlimit(dquot)) {
1296                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1297                 ret = -EDQUOT;
1298                 goto out;
1299         }
1300
1301         if (dquot->dq_dqb.dqb_isoftlimit &&
1302             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1303             dquot->dq_dqb.dqb_itime &&
1304             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1305             !ignore_hardlimit(dquot)) {
1306                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1307                 ret = -EDQUOT;
1308                 goto out;
1309         }
1310
1311         if (dquot->dq_dqb.dqb_isoftlimit &&
1312             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1313             dquot->dq_dqb.dqb_itime == 0) {
1314                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1315                 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1316                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1317         }
1318 add:
1319         dquot->dq_dqb.dqb_curinodes = newinodes;
1320
1321 out:
1322         spin_unlock(&dquot->dq_dqb_lock);
1323         return ret;
1324 }
1325
1326 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1327                            qsize_t rsv_space, unsigned int flags,
1328                            struct dquot_warn *warn)
1329 {
1330         qsize_t tspace;
1331         struct super_block *sb = dquot->dq_sb;
1332         int ret = 0;
1333
1334         spin_lock(&dquot->dq_dqb_lock);
1335         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1336             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1337                 goto finish;
1338
1339         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1340                 + space + rsv_space;
1341
1342         if (dquot->dq_dqb.dqb_bhardlimit &&
1343             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1344             !ignore_hardlimit(dquot)) {
1345                 if (flags & DQUOT_SPACE_WARN)
1346                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1347                 ret = -EDQUOT;
1348                 goto finish;
1349         }
1350
1351         if (dquot->dq_dqb.dqb_bsoftlimit &&
1352             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1353             dquot->dq_dqb.dqb_btime &&
1354             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1355             !ignore_hardlimit(dquot)) {
1356                 if (flags & DQUOT_SPACE_WARN)
1357                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1358                 ret = -EDQUOT;
1359                 goto finish;
1360         }
1361
1362         if (dquot->dq_dqb.dqb_bsoftlimit &&
1363             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1364             dquot->dq_dqb.dqb_btime == 0) {
1365                 if (flags & DQUOT_SPACE_WARN) {
1366                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1367                         dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1368                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1369                 } else {
1370                         /*
1371                          * We don't allow preallocation to exceed softlimit so exceeding will
1372                          * be always printed
1373                          */
1374                         ret = -EDQUOT;
1375                         goto finish;
1376                 }
1377         }
1378 finish:
1379         /*
1380          * We have to be careful and go through warning generation & grace time
1381          * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1382          * only here...
1383          */
1384         if (flags & DQUOT_SPACE_NOFAIL)
1385                 ret = 0;
1386         if (!ret) {
1387                 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1388                 dquot->dq_dqb.dqb_curspace += space;
1389         }
1390         spin_unlock(&dquot->dq_dqb_lock);
1391         return ret;
1392 }
1393
1394 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1395 {
1396         qsize_t newinodes;
1397
1398         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1399             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1400             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1401                 return QUOTA_NL_NOWARN;
1402
1403         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1404         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1405                 return QUOTA_NL_ISOFTBELOW;
1406         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1407             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1408                 return QUOTA_NL_IHARDBELOW;
1409         return QUOTA_NL_NOWARN;
1410 }
1411
1412 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1413 {
1414         qsize_t tspace;
1415
1416         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1417
1418         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1419             tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1420                 return QUOTA_NL_NOWARN;
1421
1422         if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1423                 return QUOTA_NL_BSOFTBELOW;
1424         if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1425             tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1426                 return QUOTA_NL_BHARDBELOW;
1427         return QUOTA_NL_NOWARN;
1428 }
1429
1430 static int dquot_active(const struct inode *inode)
1431 {
1432         struct super_block *sb = inode->i_sb;
1433
1434         if (IS_NOQUOTA(inode))
1435                 return 0;
1436         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1437 }
1438
1439 /*
1440  * Initialize quota pointers in inode
1441  *
1442  * It is better to call this function outside of any transaction as it
1443  * might need a lot of space in journal for dquot structure allocation.
1444  */
1445 static int __dquot_initialize(struct inode *inode, int type)
1446 {
1447         int cnt, init_needed = 0;
1448         struct dquot **dquots, *got[MAXQUOTAS] = {};
1449         struct super_block *sb = inode->i_sb;
1450         qsize_t rsv;
1451         int ret = 0;
1452
1453         if (!dquot_active(inode))
1454                 return 0;
1455
1456         dquots = i_dquot(inode);
1457
1458         /* First get references to structures we might need. */
1459         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1460                 struct kqid qid;
1461                 kprojid_t projid;
1462                 int rc;
1463                 struct dquot *dquot;
1464
1465                 if (type != -1 && cnt != type)
1466                         continue;
1467                 /*
1468                  * The i_dquot should have been initialized in most cases,
1469                  * we check it without locking here to avoid unnecessary
1470                  * dqget()/dqput() calls.
1471                  */
1472                 if (dquots[cnt])
1473                         continue;
1474
1475                 if (!sb_has_quota_active(sb, cnt))
1476                         continue;
1477
1478                 init_needed = 1;
1479
1480                 switch (cnt) {
1481                 case USRQUOTA:
1482                         qid = make_kqid_uid(inode->i_uid);
1483                         break;
1484                 case GRPQUOTA:
1485                         qid = make_kqid_gid(inode->i_gid);
1486                         break;
1487                 case PRJQUOTA:
1488                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1489                         if (rc)
1490                                 continue;
1491                         qid = make_kqid_projid(projid);
1492                         break;
1493                 }
1494                 dquot = dqget(sb, qid);
1495                 if (IS_ERR(dquot)) {
1496                         /* We raced with somebody turning quotas off... */
1497                         if (PTR_ERR(dquot) != -ESRCH) {
1498                                 ret = PTR_ERR(dquot);
1499                                 goto out_put;
1500                         }
1501                         dquot = NULL;
1502                 }
1503                 got[cnt] = dquot;
1504         }
1505
1506         /* All required i_dquot has been initialized */
1507         if (!init_needed)
1508                 return 0;
1509
1510         spin_lock(&dq_data_lock);
1511         if (IS_NOQUOTA(inode))
1512                 goto out_lock;
1513         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1514                 if (type != -1 && cnt != type)
1515                         continue;
1516                 /* Avoid races with quotaoff() */
1517                 if (!sb_has_quota_active(sb, cnt))
1518                         continue;
1519                 /* We could race with quotaon or dqget() could have failed */
1520                 if (!got[cnt])
1521                         continue;
1522                 if (!dquots[cnt]) {
1523                         dquots[cnt] = got[cnt];
1524                         got[cnt] = NULL;
1525                         /*
1526                          * Make quota reservation system happy if someone
1527                          * did a write before quota was turned on
1528                          */
1529                         rsv = inode_get_rsv_space(inode);
1530                         if (unlikely(rsv)) {
1531                                 spin_lock(&inode->i_lock);
1532                                 /* Get reservation again under proper lock */
1533                                 rsv = __inode_get_rsv_space(inode);
1534                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1535                                 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1536                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1537                                 spin_unlock(&inode->i_lock);
1538                         }
1539                 }
1540         }
1541 out_lock:
1542         spin_unlock(&dq_data_lock);
1543 out_put:
1544         /* Drop unused references */
1545         dqput_all(got);
1546
1547         return ret;
1548 }
1549
1550 int dquot_initialize(struct inode *inode)
1551 {
1552         return __dquot_initialize(inode, -1);
1553 }
1554 EXPORT_SYMBOL(dquot_initialize);
1555
1556 bool dquot_initialize_needed(struct inode *inode)
1557 {
1558         struct dquot **dquots;
1559         int i;
1560
1561         if (!dquot_active(inode))
1562                 return false;
1563
1564         dquots = i_dquot(inode);
1565         for (i = 0; i < MAXQUOTAS; i++)
1566                 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1567                         return true;
1568         return false;
1569 }
1570 EXPORT_SYMBOL(dquot_initialize_needed);
1571
1572 /*
1573  * Release all quotas referenced by inode.
1574  *
1575  * This function only be called on inode free or converting
1576  * a file to quota file, no other users for the i_dquot in
1577  * both cases, so we needn't call synchronize_srcu() after
1578  * clearing i_dquot.
1579  */
1580 static void __dquot_drop(struct inode *inode)
1581 {
1582         int cnt;
1583         struct dquot **dquots = i_dquot(inode);
1584         struct dquot *put[MAXQUOTAS];
1585
1586         spin_lock(&dq_data_lock);
1587         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1588                 put[cnt] = dquots[cnt];
1589                 dquots[cnt] = NULL;
1590         }
1591         spin_unlock(&dq_data_lock);
1592         dqput_all(put);
1593 }
1594
1595 void dquot_drop(struct inode *inode)
1596 {
1597         struct dquot * const *dquots;
1598         int cnt;
1599
1600         if (IS_NOQUOTA(inode))
1601                 return;
1602
1603         /*
1604          * Test before calling to rule out calls from proc and such
1605          * where we are not allowed to block. Note that this is
1606          * actually reliable test even without the lock - the caller
1607          * must assure that nobody can come after the DQUOT_DROP and
1608          * add quota pointers back anyway.
1609          */
1610         dquots = i_dquot(inode);
1611         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1612                 if (dquots[cnt])
1613                         break;
1614         }
1615
1616         if (cnt < MAXQUOTAS)
1617                 __dquot_drop(inode);
1618 }
1619 EXPORT_SYMBOL(dquot_drop);
1620
1621 /*
1622  * inode_reserved_space is managed internally by quota, and protected by
1623  * i_lock similar to i_blocks+i_bytes.
1624  */
1625 static qsize_t *inode_reserved_space(struct inode * inode)
1626 {
1627         /* Filesystem must explicitly define it's own method in order to use
1628          * quota reservation interface */
1629         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1630         return inode->i_sb->dq_op->get_reserved_space(inode);
1631 }
1632
1633 static qsize_t __inode_get_rsv_space(struct inode *inode)
1634 {
1635         if (!inode->i_sb->dq_op->get_reserved_space)
1636                 return 0;
1637         return *inode_reserved_space(inode);
1638 }
1639
1640 static qsize_t inode_get_rsv_space(struct inode *inode)
1641 {
1642         qsize_t ret;
1643
1644         if (!inode->i_sb->dq_op->get_reserved_space)
1645                 return 0;
1646         spin_lock(&inode->i_lock);
1647         ret = __inode_get_rsv_space(inode);
1648         spin_unlock(&inode->i_lock);
1649         return ret;
1650 }
1651
1652 /*
1653  * This functions updates i_blocks+i_bytes fields and quota information
1654  * (together with appropriate checks).
1655  *
1656  * NOTE: We absolutely rely on the fact that caller dirties the inode
1657  * (usually helpers in quotaops.h care about this) and holds a handle for
1658  * the current transaction so that dquot write and inode write go into the
1659  * same transaction.
1660  */
1661
1662 /*
1663  * This operation can block, but only after everything is updated
1664  */
1665 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1666 {
1667         int cnt, ret = 0, index;
1668         struct dquot_warn warn[MAXQUOTAS];
1669         int reserve = flags & DQUOT_SPACE_RESERVE;
1670         struct dquot **dquots;
1671
1672         if (!dquot_active(inode)) {
1673                 if (reserve) {
1674                         spin_lock(&inode->i_lock);
1675                         *inode_reserved_space(inode) += number;
1676                         spin_unlock(&inode->i_lock);
1677                 } else {
1678                         inode_add_bytes(inode, number);
1679                 }
1680                 goto out;
1681         }
1682
1683         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1684                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1685
1686         dquots = i_dquot(inode);
1687         index = srcu_read_lock(&dquot_srcu);
1688         spin_lock(&inode->i_lock);
1689         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1690                 if (!dquots[cnt])
1691                         continue;
1692                 if (reserve) {
1693                         ret = dquot_add_space(dquots[cnt], 0, number, flags,
1694                                               &warn[cnt]);
1695                 } else {
1696                         ret = dquot_add_space(dquots[cnt], number, 0, flags,
1697                                               &warn[cnt]);
1698                 }
1699                 if (ret) {
1700                         /* Back out changes we already did */
1701                         for (cnt--; cnt >= 0; cnt--) {
1702                                 if (!dquots[cnt])
1703                                         continue;
1704                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1705                                 if (reserve)
1706                                         dquot_free_reserved_space(dquots[cnt],
1707                                                                   number);
1708                                 else
1709                                         dquot_decr_space(dquots[cnt], number);
1710                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1711                         }
1712                         spin_unlock(&inode->i_lock);
1713                         goto out_flush_warn;
1714                 }
1715         }
1716         if (reserve)
1717                 *inode_reserved_space(inode) += number;
1718         else
1719                 __inode_add_bytes(inode, number);
1720         spin_unlock(&inode->i_lock);
1721
1722         if (reserve)
1723                 goto out_flush_warn;
1724         mark_all_dquot_dirty(dquots);
1725 out_flush_warn:
1726         srcu_read_unlock(&dquot_srcu, index);
1727         flush_warnings(warn);
1728 out:
1729         return ret;
1730 }
1731 EXPORT_SYMBOL(__dquot_alloc_space);
1732
1733 /*
1734  * This operation can block, but only after everything is updated
1735  */
1736 int dquot_alloc_inode(struct inode *inode)
1737 {
1738         int cnt, ret = 0, index;
1739         struct dquot_warn warn[MAXQUOTAS];
1740         struct dquot * const *dquots;
1741
1742         if (!dquot_active(inode))
1743                 return 0;
1744         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1745                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1746
1747         dquots = i_dquot(inode);
1748         index = srcu_read_lock(&dquot_srcu);
1749         spin_lock(&inode->i_lock);
1750         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1751                 if (!dquots[cnt])
1752                         continue;
1753                 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1754                 if (ret) {
1755                         for (cnt--; cnt >= 0; cnt--) {
1756                                 if (!dquots[cnt])
1757                                         continue;
1758                                 /* Back out changes we already did */
1759                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1760                                 dquot_decr_inodes(dquots[cnt], 1);
1761                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1762                         }
1763                         goto warn_put_all;
1764                 }
1765         }
1766
1767 warn_put_all:
1768         spin_unlock(&inode->i_lock);
1769         if (ret == 0)
1770                 mark_all_dquot_dirty(dquots);
1771         srcu_read_unlock(&dquot_srcu, index);
1772         flush_warnings(warn);
1773         return ret;
1774 }
1775 EXPORT_SYMBOL(dquot_alloc_inode);
1776
1777 /*
1778  * Convert in-memory reserved quotas to real consumed quotas
1779  */
1780 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1781 {
1782         struct dquot **dquots;
1783         int cnt, index;
1784
1785         if (!dquot_active(inode)) {
1786                 spin_lock(&inode->i_lock);
1787                 *inode_reserved_space(inode) -= number;
1788                 __inode_add_bytes(inode, number);
1789                 spin_unlock(&inode->i_lock);
1790                 return 0;
1791         }
1792
1793         dquots = i_dquot(inode);
1794         index = srcu_read_lock(&dquot_srcu);
1795         spin_lock(&inode->i_lock);
1796         /* Claim reserved quotas to allocated quotas */
1797         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1798                 if (dquots[cnt]) {
1799                         struct dquot *dquot = dquots[cnt];
1800
1801                         spin_lock(&dquot->dq_dqb_lock);
1802                         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1803                                 number = dquot->dq_dqb.dqb_rsvspace;
1804                         dquot->dq_dqb.dqb_curspace += number;
1805                         dquot->dq_dqb.dqb_rsvspace -= number;
1806                         spin_unlock(&dquot->dq_dqb_lock);
1807                 }
1808         }
1809         /* Update inode bytes */
1810         *inode_reserved_space(inode) -= number;
1811         __inode_add_bytes(inode, number);
1812         spin_unlock(&inode->i_lock);
1813         mark_all_dquot_dirty(dquots);
1814         srcu_read_unlock(&dquot_srcu, index);
1815         return 0;
1816 }
1817 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1818
1819 /*
1820  * Convert allocated space back to in-memory reserved quotas
1821  */
1822 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1823 {
1824         struct dquot **dquots;
1825         int cnt, index;
1826
1827         if (!dquot_active(inode)) {
1828                 spin_lock(&inode->i_lock);
1829                 *inode_reserved_space(inode) += number;
1830                 __inode_sub_bytes(inode, number);
1831                 spin_unlock(&inode->i_lock);
1832                 return;
1833         }
1834
1835         dquots = i_dquot(inode);
1836         index = srcu_read_lock(&dquot_srcu);
1837         spin_lock(&inode->i_lock);
1838         /* Claim reserved quotas to allocated quotas */
1839         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1840                 if (dquots[cnt]) {
1841                         struct dquot *dquot = dquots[cnt];
1842
1843                         spin_lock(&dquot->dq_dqb_lock);
1844                         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1845                                 number = dquot->dq_dqb.dqb_curspace;
1846                         dquot->dq_dqb.dqb_rsvspace += number;
1847                         dquot->dq_dqb.dqb_curspace -= number;
1848                         spin_unlock(&dquot->dq_dqb_lock);
1849                 }
1850         }
1851         /* Update inode bytes */
1852         *inode_reserved_space(inode) += number;
1853         __inode_sub_bytes(inode, number);
1854         spin_unlock(&inode->i_lock);
1855         mark_all_dquot_dirty(dquots);
1856         srcu_read_unlock(&dquot_srcu, index);
1857         return;
1858 }
1859 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1860
1861 /*
1862  * This operation can block, but only after everything is updated
1863  */
1864 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1865 {
1866         unsigned int cnt;
1867         struct dquot_warn warn[MAXQUOTAS];
1868         struct dquot **dquots;
1869         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1870
1871         if (!dquot_active(inode)) {
1872                 if (reserve) {
1873                         spin_lock(&inode->i_lock);
1874                         *inode_reserved_space(inode) -= number;
1875                         spin_unlock(&inode->i_lock);
1876                 } else {
1877                         inode_sub_bytes(inode, number);
1878                 }
1879                 return;
1880         }
1881
1882         dquots = i_dquot(inode);
1883         index = srcu_read_lock(&dquot_srcu);
1884         spin_lock(&inode->i_lock);
1885         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1886                 int wtype;
1887
1888                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1889                 if (!dquots[cnt])
1890                         continue;
1891                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1892                 wtype = info_bdq_free(dquots[cnt], number);
1893                 if (wtype != QUOTA_NL_NOWARN)
1894                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1895                 if (reserve)
1896                         dquot_free_reserved_space(dquots[cnt], number);
1897                 else
1898                         dquot_decr_space(dquots[cnt], number);
1899                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1900         }
1901         if (reserve)
1902                 *inode_reserved_space(inode) -= number;
1903         else
1904                 __inode_sub_bytes(inode, number);
1905         spin_unlock(&inode->i_lock);
1906
1907         if (reserve)
1908                 goto out_unlock;
1909         mark_all_dquot_dirty(dquots);
1910 out_unlock:
1911         srcu_read_unlock(&dquot_srcu, index);
1912         flush_warnings(warn);
1913 }
1914 EXPORT_SYMBOL(__dquot_free_space);
1915
1916 /*
1917  * This operation can block, but only after everything is updated
1918  */
1919 void dquot_free_inode(struct inode *inode)
1920 {
1921         unsigned int cnt;
1922         struct dquot_warn warn[MAXQUOTAS];
1923         struct dquot * const *dquots;
1924         int index;
1925
1926         if (!dquot_active(inode))
1927                 return;
1928
1929         dquots = i_dquot(inode);
1930         index = srcu_read_lock(&dquot_srcu);
1931         spin_lock(&inode->i_lock);
1932         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1933                 int wtype;
1934
1935                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1936                 if (!dquots[cnt])
1937                         continue;
1938                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1939                 wtype = info_idq_free(dquots[cnt], 1);
1940                 if (wtype != QUOTA_NL_NOWARN)
1941                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1942                 dquot_decr_inodes(dquots[cnt], 1);
1943                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1944         }
1945         spin_unlock(&inode->i_lock);
1946         mark_all_dquot_dirty(dquots);
1947         srcu_read_unlock(&dquot_srcu, index);
1948         flush_warnings(warn);
1949 }
1950 EXPORT_SYMBOL(dquot_free_inode);
1951
1952 /*
1953  * Transfer the number of inode and blocks from one diskquota to an other.
1954  * On success, dquot references in transfer_to are consumed and references
1955  * to original dquots that need to be released are placed there. On failure,
1956  * references are kept untouched.
1957  *
1958  * This operation can block, but only after everything is updated
1959  * A transaction must be started when entering this function.
1960  *
1961  * We are holding reference on transfer_from & transfer_to, no need to
1962  * protect them by srcu_read_lock().
1963  */
1964 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1965 {
1966         qsize_t cur_space;
1967         qsize_t rsv_space = 0;
1968         qsize_t inode_usage = 1;
1969         struct dquot *transfer_from[MAXQUOTAS] = {};
1970         int cnt, ret = 0;
1971         char is_valid[MAXQUOTAS] = {};
1972         struct dquot_warn warn_to[MAXQUOTAS];
1973         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1974         struct dquot_warn warn_from_space[MAXQUOTAS];
1975
1976         if (IS_NOQUOTA(inode))
1977                 return 0;
1978
1979         if (inode->i_sb->dq_op->get_inode_usage) {
1980                 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1981                 if (ret)
1982                         return ret;
1983         }
1984
1985         /* Initialize the arrays */
1986         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1987                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1988                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1989                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1990         }
1991
1992         spin_lock(&dq_data_lock);
1993         spin_lock(&inode->i_lock);
1994         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1995                 spin_unlock(&inode->i_lock);
1996                 spin_unlock(&dq_data_lock);
1997                 return 0;
1998         }
1999         cur_space = __inode_get_bytes(inode);
2000         rsv_space = __inode_get_rsv_space(inode);
2001         /*
2002          * Build the transfer_from list, check limits, and update usage in
2003          * the target structures.
2004          */
2005         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2006                 /*
2007                  * Skip changes for same uid or gid or for turned off quota-type.
2008                  */
2009                 if (!transfer_to[cnt])
2010                         continue;
2011                 /* Avoid races with quotaoff() */
2012                 if (!sb_has_quota_active(inode->i_sb, cnt))
2013                         continue;
2014                 is_valid[cnt] = 1;
2015                 transfer_from[cnt] = i_dquot(inode)[cnt];
2016                 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2017                                        &warn_to[cnt]);
2018                 if (ret)
2019                         goto over_quota;
2020                 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2021                                       DQUOT_SPACE_WARN, &warn_to[cnt]);
2022                 if (ret) {
2023                         spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2024                         dquot_decr_inodes(transfer_to[cnt], inode_usage);
2025                         spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2026                         goto over_quota;
2027                 }
2028         }
2029
2030         /* Decrease usage for source structures and update quota pointers */
2031         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2032                 if (!is_valid[cnt])
2033                         continue;
2034                 /* Due to IO error we might not have transfer_from[] structure */
2035                 if (transfer_from[cnt]) {
2036                         int wtype;
2037
2038                         spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2039                         wtype = info_idq_free(transfer_from[cnt], inode_usage);
2040                         if (wtype != QUOTA_NL_NOWARN)
2041                                 prepare_warning(&warn_from_inodes[cnt],
2042                                                 transfer_from[cnt], wtype);
2043                         wtype = info_bdq_free(transfer_from[cnt],
2044                                               cur_space + rsv_space);
2045                         if (wtype != QUOTA_NL_NOWARN)
2046                                 prepare_warning(&warn_from_space[cnt],
2047                                                 transfer_from[cnt], wtype);
2048                         dquot_decr_inodes(transfer_from[cnt], inode_usage);
2049                         dquot_decr_space(transfer_from[cnt], cur_space);
2050                         dquot_free_reserved_space(transfer_from[cnt],
2051                                                   rsv_space);
2052                         spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2053                 }
2054                 i_dquot(inode)[cnt] = transfer_to[cnt];
2055         }
2056         spin_unlock(&inode->i_lock);
2057         spin_unlock(&dq_data_lock);
2058
2059         mark_all_dquot_dirty(transfer_from);
2060         mark_all_dquot_dirty(transfer_to);
2061         flush_warnings(warn_to);
2062         flush_warnings(warn_from_inodes);
2063         flush_warnings(warn_from_space);
2064         /* Pass back references to put */
2065         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2066                 if (is_valid[cnt])
2067                         transfer_to[cnt] = transfer_from[cnt];
2068         return 0;
2069 over_quota:
2070         /* Back out changes we already did */
2071         for (cnt--; cnt >= 0; cnt--) {
2072                 if (!is_valid[cnt])
2073                         continue;
2074                 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2075                 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2076                 dquot_decr_space(transfer_to[cnt], cur_space);
2077                 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2078                 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2079         }
2080         spin_unlock(&inode->i_lock);
2081         spin_unlock(&dq_data_lock);
2082         flush_warnings(warn_to);
2083         return ret;
2084 }
2085 EXPORT_SYMBOL(__dquot_transfer);
2086
2087 /* Wrapper for transferring ownership of an inode for uid/gid only
2088  * Called from FSXXX_setattr()
2089  */
2090 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2091 {
2092         struct dquot *transfer_to[MAXQUOTAS] = {};
2093         struct dquot *dquot;
2094         struct super_block *sb = inode->i_sb;
2095         int ret;
2096
2097         if (!dquot_active(inode))
2098                 return 0;
2099
2100         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2101                 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2102                 if (IS_ERR(dquot)) {
2103                         if (PTR_ERR(dquot) != -ESRCH) {
2104                                 ret = PTR_ERR(dquot);
2105                                 goto out_put;
2106                         }
2107                         dquot = NULL;
2108                 }
2109                 transfer_to[USRQUOTA] = dquot;
2110         }
2111         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2112                 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2113                 if (IS_ERR(dquot)) {
2114                         if (PTR_ERR(dquot) != -ESRCH) {
2115                                 ret = PTR_ERR(dquot);
2116                                 goto out_put;
2117                         }
2118                         dquot = NULL;
2119                 }
2120                 transfer_to[GRPQUOTA] = dquot;
2121         }
2122         ret = __dquot_transfer(inode, transfer_to);
2123 out_put:
2124         dqput_all(transfer_to);
2125         return ret;
2126 }
2127 EXPORT_SYMBOL(dquot_transfer);
2128
2129 /*
2130  * Write info of quota file to disk
2131  */
2132 int dquot_commit_info(struct super_block *sb, int type)
2133 {
2134         struct quota_info *dqopt = sb_dqopt(sb);
2135
2136         return dqopt->ops[type]->write_file_info(sb, type);
2137 }
2138 EXPORT_SYMBOL(dquot_commit_info);
2139
2140 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2141 {
2142         struct quota_info *dqopt = sb_dqopt(sb);
2143
2144         if (!sb_has_quota_active(sb, qid->type))
2145                 return -ESRCH;
2146         if (!dqopt->ops[qid->type]->get_next_id)
2147                 return -ENOSYS;
2148         return dqopt->ops[qid->type]->get_next_id(sb, qid);
2149 }
2150 EXPORT_SYMBOL(dquot_get_next_id);
2151
2152 /*
2153  * Definitions of diskquota operations.
2154  */
2155 const struct dquot_operations dquot_operations = {
2156         .write_dquot    = dquot_commit,
2157         .acquire_dquot  = dquot_acquire,
2158         .release_dquot  = dquot_release,
2159         .mark_dirty     = dquot_mark_dquot_dirty,
2160         .write_info     = dquot_commit_info,
2161         .alloc_dquot    = dquot_alloc,
2162         .destroy_dquot  = dquot_destroy,
2163         .get_next_id    = dquot_get_next_id,
2164 };
2165 EXPORT_SYMBOL(dquot_operations);
2166
2167 /*
2168  * Generic helper for ->open on filesystems supporting disk quotas.
2169  */
2170 int dquot_file_open(struct inode *inode, struct file *file)
2171 {
2172         int error;
2173
2174         error = generic_file_open(inode, file);
2175         if (!error && (file->f_mode & FMODE_WRITE))
2176                 error = dquot_initialize(inode);
2177         return error;
2178 }
2179 EXPORT_SYMBOL(dquot_file_open);
2180
2181 static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2182 {
2183         struct quota_info *dqopt = sb_dqopt(sb);
2184         struct inode *inode = dqopt->files[type];
2185
2186         if (!inode)
2187                 return;
2188         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2189                 inode_lock(inode);
2190                 inode->i_flags &= ~S_NOQUOTA;
2191                 inode_unlock(inode);
2192         }
2193         dqopt->files[type] = NULL;
2194         iput(inode);
2195 }
2196
2197 /*
2198  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2199  */
2200 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2201 {
2202         int cnt;
2203         struct quota_info *dqopt = sb_dqopt(sb);
2204
2205         /* s_umount should be held in exclusive mode */
2206         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2207                 up_read(&sb->s_umount);
2208
2209         /* Cannot turn off usage accounting without turning off limits, or
2210          * suspend quotas and simultaneously turn quotas off. */
2211         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2212             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2213             DQUOT_USAGE_ENABLED)))
2214                 return -EINVAL;
2215
2216         /*
2217          * Skip everything if there's nothing to do. We have to do this because
2218          * sometimes we are called when fill_super() failed and calling
2219          * sync_fs() in such cases does no good.
2220          */
2221         if (!sb_any_quota_loaded(sb))
2222                 return 0;
2223
2224         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2225                 if (type != -1 && cnt != type)
2226                         continue;
2227                 if (!sb_has_quota_loaded(sb, cnt))
2228                         continue;
2229
2230                 if (flags & DQUOT_SUSPENDED) {
2231                         spin_lock(&dq_state_lock);
2232                         dqopt->flags |=
2233                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2234                         spin_unlock(&dq_state_lock);
2235                 } else {
2236                         spin_lock(&dq_state_lock);
2237                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2238                         /* Turning off suspended quotas? */
2239                         if (!sb_has_quota_loaded(sb, cnt) &&
2240                             sb_has_quota_suspended(sb, cnt)) {
2241                                 dqopt->flags &= ~dquot_state_flag(
2242                                                         DQUOT_SUSPENDED, cnt);
2243                                 spin_unlock(&dq_state_lock);
2244                                 vfs_cleanup_quota_inode(sb, cnt);
2245                                 continue;
2246                         }
2247                         spin_unlock(&dq_state_lock);
2248                 }
2249
2250                 /* We still have to keep quota loaded? */
2251                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2252                         continue;
2253
2254                 /* Note: these are blocking operations */
2255                 drop_dquot_ref(sb, cnt);
2256                 invalidate_dquots(sb, cnt);
2257                 /*
2258                  * Now all dquots should be invalidated, all writes done so we
2259                  * should be only users of the info. No locks needed.
2260                  */
2261                 if (info_dirty(&dqopt->info[cnt]))
2262                         sb->dq_op->write_info(sb, cnt);
2263                 if (dqopt->ops[cnt]->free_file_info)
2264                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2265                 put_quota_format(dqopt->info[cnt].dqi_format);
2266                 dqopt->info[cnt].dqi_flags = 0;
2267                 dqopt->info[cnt].dqi_igrace = 0;
2268                 dqopt->info[cnt].dqi_bgrace = 0;
2269                 dqopt->ops[cnt] = NULL;
2270         }
2271
2272         /* Skip syncing and setting flags if quota files are hidden */
2273         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2274                 goto put_inodes;
2275
2276         /* Sync the superblock so that buffers with quota data are written to
2277          * disk (and so userspace sees correct data afterwards). */
2278         if (sb->s_op->sync_fs)
2279                 sb->s_op->sync_fs(sb, 1);
2280         sync_blockdev(sb->s_bdev);
2281         /* Now the quota files are just ordinary files and we can set the
2282          * inode flags back. Moreover we discard the pagecache so that
2283          * userspace sees the writes we did bypassing the pagecache. We
2284          * must also discard the blockdev buffers so that we see the
2285          * changes done by userspace on the next quotaon() */
2286         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2287                 if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2288                         inode_lock(dqopt->files[cnt]);
2289                         truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2290                         inode_unlock(dqopt->files[cnt]);
2291                 }
2292         if (sb->s_bdev)
2293                 invalidate_bdev(sb->s_bdev);
2294 put_inodes:
2295         /* We are done when suspending quotas */
2296         if (flags & DQUOT_SUSPENDED)
2297                 return 0;
2298
2299         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2300                 if (!sb_has_quota_loaded(sb, cnt))
2301                         vfs_cleanup_quota_inode(sb, cnt);
2302         return 0;
2303 }
2304 EXPORT_SYMBOL(dquot_disable);
2305
2306 int dquot_quota_off(struct super_block *sb, int type)
2307 {
2308         return dquot_disable(sb, type,
2309                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2310 }
2311 EXPORT_SYMBOL(dquot_quota_off);
2312
2313 /*
2314  *      Turn quotas on on a device
2315  */
2316
2317 static int vfs_setup_quota_inode(struct inode *inode, int type)
2318 {
2319         struct super_block *sb = inode->i_sb;
2320         struct quota_info *dqopt = sb_dqopt(sb);
2321
2322         if (!S_ISREG(inode->i_mode))
2323                 return -EACCES;
2324         if (IS_RDONLY(inode))
2325                 return -EROFS;
2326         if (sb_has_quota_loaded(sb, type))
2327                 return -EBUSY;
2328
2329         dqopt->files[type] = igrab(inode);
2330         if (!dqopt->files[type])
2331                 return -EIO;
2332         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2333                 /* We don't want quota and atime on quota files (deadlocks
2334                  * possible) Also nobody should write to the file - we use
2335                  * special IO operations which ignore the immutable bit. */
2336                 inode_lock(inode);
2337                 inode->i_flags |= S_NOQUOTA;
2338                 inode_unlock(inode);
2339                 /*
2340                  * When S_NOQUOTA is set, remove dquot references as no more
2341                  * references can be added
2342                  */
2343                 __dquot_drop(inode);
2344         }
2345         return 0;
2346 }
2347
2348 int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2349         unsigned int flags)
2350 {
2351         struct quota_format_type *fmt = find_quota_format(format_id);
2352         struct quota_info *dqopt = sb_dqopt(sb);
2353         int error;
2354
2355         /* Just unsuspend quotas? */
2356         BUG_ON(flags & DQUOT_SUSPENDED);
2357         /* s_umount should be held in exclusive mode */
2358         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2359                 up_read(&sb->s_umount);
2360
2361         if (!fmt)
2362                 return -ESRCH;
2363         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2364             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2365                 error = -EINVAL;
2366                 goto out_fmt;
2367         }
2368         /* Filesystems outside of init_user_ns not yet supported */
2369         if (sb->s_user_ns != &init_user_ns) {
2370                 error = -EINVAL;
2371                 goto out_fmt;
2372         }
2373         /* Usage always has to be set... */
2374         if (!(flags & DQUOT_USAGE_ENABLED)) {
2375                 error = -EINVAL;
2376                 goto out_fmt;
2377         }
2378         if (sb_has_quota_loaded(sb, type)) {
2379                 error = -EBUSY;
2380                 goto out_fmt;
2381         }
2382
2383         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2384                 /* As we bypass the pagecache we must now flush all the
2385                  * dirty data and invalidate caches so that kernel sees
2386                  * changes from userspace. It is not enough to just flush
2387                  * the quota file since if blocksize < pagesize, invalidation
2388                  * of the cache could fail because of other unrelated dirty
2389                  * data */
2390                 sync_filesystem(sb);
2391                 invalidate_bdev(sb->s_bdev);
2392         }
2393
2394         error = -EINVAL;
2395         if (!fmt->qf_ops->check_quota_file(sb, type))
2396                 goto out_fmt;
2397
2398         dqopt->ops[type] = fmt->qf_ops;
2399         dqopt->info[type].dqi_format = fmt;
2400         dqopt->info[type].dqi_fmt_id = format_id;
2401         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2402         error = dqopt->ops[type]->read_file_info(sb, type);
2403         if (error < 0)
2404                 goto out_fmt;
2405         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2406                 spin_lock(&dq_data_lock);
2407                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2408                 spin_unlock(&dq_data_lock);
2409         }
2410         spin_lock(&dq_state_lock);
2411         dqopt->flags |= dquot_state_flag(flags, type);
2412         spin_unlock(&dq_state_lock);
2413
2414         error = add_dquot_ref(sb, type);
2415         if (error)
2416                 dquot_disable(sb, type, flags);
2417
2418         return error;
2419 out_fmt:
2420         put_quota_format(fmt);
2421
2422         return error;
2423 }
2424 EXPORT_SYMBOL(dquot_load_quota_sb);
2425
2426 /*
2427  * More powerful function for turning on quotas on given quota inode allowing
2428  * setting of individual quota flags
2429  */
2430 int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2431         unsigned int flags)
2432 {
2433         int err;
2434
2435         err = vfs_setup_quota_inode(inode, type);
2436         if (err < 0)
2437                 return err;
2438         err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2439         if (err < 0)
2440                 vfs_cleanup_quota_inode(inode->i_sb, type);
2441         return err;
2442 }
2443 EXPORT_SYMBOL(dquot_load_quota_inode);
2444
2445 /* Reenable quotas on remount RW */
2446 int dquot_resume(struct super_block *sb, int type)
2447 {
2448         struct quota_info *dqopt = sb_dqopt(sb);
2449         int ret = 0, cnt;
2450         unsigned int flags;
2451
2452         /* s_umount should be held in exclusive mode */
2453         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2454                 up_read(&sb->s_umount);
2455
2456         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2457                 if (type != -1 && cnt != type)
2458                         continue;
2459                 if (!sb_has_quota_suspended(sb, cnt))
2460                         continue;
2461
2462                 spin_lock(&dq_state_lock);
2463                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2464                                                         DQUOT_LIMITS_ENABLED,
2465                                                         cnt);
2466                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2467                 spin_unlock(&dq_state_lock);
2468
2469                 flags = dquot_generic_flag(flags, cnt);
2470                 ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2471                                           flags);
2472                 if (ret < 0)
2473                         vfs_cleanup_quota_inode(sb, cnt);
2474         }
2475
2476         return ret;
2477 }
2478 EXPORT_SYMBOL(dquot_resume);
2479
2480 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2481                    const struct path *path)
2482 {
2483         int error = security_quota_on(path->dentry);
2484         if (error)
2485                 return error;
2486         /* Quota file not on the same filesystem? */
2487         if (path->dentry->d_sb != sb)
2488                 error = -EXDEV;
2489         else
2490                 error = dquot_load_quota_inode(d_inode(path->dentry), type,
2491                                              format_id, DQUOT_USAGE_ENABLED |
2492                                              DQUOT_LIMITS_ENABLED);
2493         return error;
2494 }
2495 EXPORT_SYMBOL(dquot_quota_on);
2496
2497 /*
2498  * This function is used when filesystem needs to initialize quotas
2499  * during mount time.
2500  */
2501 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2502                 int format_id, int type)
2503 {
2504         struct dentry *dentry;
2505         int error;
2506
2507         dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2508         if (IS_ERR(dentry))
2509                 return PTR_ERR(dentry);
2510
2511         error = security_quota_on(dentry);
2512         if (!error)
2513                 error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2514                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2515
2516         dput(dentry);
2517         return error;
2518 }
2519 EXPORT_SYMBOL(dquot_quota_on_mount);
2520
2521 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2522 {
2523         int ret;
2524         int type;
2525         struct quota_info *dqopt = sb_dqopt(sb);
2526
2527         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2528                 return -ENOSYS;
2529         /* Accounting cannot be turned on while fs is mounted */
2530         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2531         if (!flags)
2532                 return -EINVAL;
2533         for (type = 0; type < MAXQUOTAS; type++) {
2534                 if (!(flags & qtype_enforce_flag(type)))
2535                         continue;
2536                 /* Can't enforce without accounting */
2537                 if (!sb_has_quota_usage_enabled(sb, type)) {
2538                         ret = -EINVAL;
2539                         goto out_err;
2540                 }
2541                 if (sb_has_quota_limits_enabled(sb, type)) {
2542                         ret = -EBUSY;
2543                         goto out_err;
2544                 }
2545                 spin_lock(&dq_state_lock);
2546                 dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2547                 spin_unlock(&dq_state_lock);
2548         }
2549         return 0;
2550 out_err:
2551         /* Backout enforcement enablement we already did */
2552         for (type--; type >= 0; type--)  {
2553                 if (flags & qtype_enforce_flag(type))
2554                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2555         }
2556         /* Error code translation for better compatibility with XFS */
2557         if (ret == -EBUSY)
2558                 ret = -EEXIST;
2559         return ret;
2560 }
2561
2562 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2563 {
2564         int ret;
2565         int type;
2566         struct quota_info *dqopt = sb_dqopt(sb);
2567
2568         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2569                 return -ENOSYS;
2570         /*
2571          * We don't support turning off accounting via quotactl. In principle
2572          * quota infrastructure can do this but filesystems don't expect
2573          * userspace to be able to do it.
2574          */
2575         if (flags &
2576                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2577                 return -EOPNOTSUPP;
2578
2579         /* Filter out limits not enabled */
2580         for (type = 0; type < MAXQUOTAS; type++)
2581                 if (!sb_has_quota_limits_enabled(sb, type))
2582                         flags &= ~qtype_enforce_flag(type);
2583         /* Nothing left? */
2584         if (!flags)
2585                 return -EEXIST;
2586         for (type = 0; type < MAXQUOTAS; type++) {
2587                 if (flags & qtype_enforce_flag(type)) {
2588                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2589                         if (ret < 0)
2590                                 goto out_err;
2591                 }
2592         }
2593         return 0;
2594 out_err:
2595         /* Backout enforcement disabling we already did */
2596         for (type--; type >= 0; type--)  {
2597                 if (flags & qtype_enforce_flag(type)) {
2598                         spin_lock(&dq_state_lock);
2599                         dqopt->flags |=
2600                                 dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2601                         spin_unlock(&dq_state_lock);
2602                 }
2603         }
2604         return ret;
2605 }
2606
2607 /* Generic routine for getting common part of quota structure */
2608 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2609 {
2610         struct mem_dqblk *dm = &dquot->dq_dqb;
2611
2612         memset(di, 0, sizeof(*di));
2613         spin_lock(&dquot->dq_dqb_lock);
2614         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2615         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2616         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2617         di->d_ino_softlimit = dm->dqb_isoftlimit;
2618         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2619         di->d_ino_count = dm->dqb_curinodes;
2620         di->d_spc_timer = dm->dqb_btime;
2621         di->d_ino_timer = dm->dqb_itime;
2622         spin_unlock(&dquot->dq_dqb_lock);
2623 }
2624
2625 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2626                     struct qc_dqblk *di)
2627 {
2628         struct dquot *dquot;
2629
2630         dquot = dqget(sb, qid);
2631         if (IS_ERR(dquot))
2632                 return PTR_ERR(dquot);
2633         do_get_dqblk(dquot, di);
2634         dqput(dquot);
2635
2636         return 0;
2637 }
2638 EXPORT_SYMBOL(dquot_get_dqblk);
2639
2640 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2641                          struct qc_dqblk *di)
2642 {
2643         struct dquot *dquot;
2644         int err;
2645
2646         if (!sb->dq_op->get_next_id)
2647                 return -ENOSYS;
2648         err = sb->dq_op->get_next_id(sb, qid);
2649         if (err < 0)
2650                 return err;
2651         dquot = dqget(sb, *qid);
2652         if (IS_ERR(dquot))
2653                 return PTR_ERR(dquot);
2654         do_get_dqblk(dquot, di);
2655         dqput(dquot);
2656
2657         return 0;
2658 }
2659 EXPORT_SYMBOL(dquot_get_next_dqblk);
2660
2661 #define VFS_QC_MASK \
2662         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2663          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2664          QC_SPC_TIMER | QC_INO_TIMER)
2665
2666 /* Generic routine for setting common part of quota structure */
2667 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2668 {
2669         struct mem_dqblk *dm = &dquot->dq_dqb;
2670         int check_blim = 0, check_ilim = 0;
2671         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2672
2673         if (di->d_fieldmask & ~VFS_QC_MASK)
2674                 return -EINVAL;
2675
2676         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2677              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2678             ((di->d_fieldmask & QC_SPC_HARD) &&
2679              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2680             ((di->d_fieldmask & QC_INO_SOFT) &&
2681              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2682             ((di->d_fieldmask & QC_INO_HARD) &&
2683              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2684                 return -ERANGE;
2685
2686         spin_lock(&dquot->dq_dqb_lock);
2687         if (di->d_fieldmask & QC_SPACE) {
2688                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2689                 check_blim = 1;
2690                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2691         }
2692
2693         if (di->d_fieldmask & QC_SPC_SOFT)
2694                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2695         if (di->d_fieldmask & QC_SPC_HARD)
2696                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2697         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2698                 check_blim = 1;
2699                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2700         }
2701
2702         if (di->d_fieldmask & QC_INO_COUNT) {
2703                 dm->dqb_curinodes = di->d_ino_count;
2704                 check_ilim = 1;
2705                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2706         }
2707
2708         if (di->d_fieldmask & QC_INO_SOFT)
2709                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2710         if (di->d_fieldmask & QC_INO_HARD)
2711                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2712         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2713                 check_ilim = 1;
2714                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2715         }
2716
2717         if (di->d_fieldmask & QC_SPC_TIMER) {
2718                 dm->dqb_btime = di->d_spc_timer;
2719                 check_blim = 1;
2720                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2721         }
2722
2723         if (di->d_fieldmask & QC_INO_TIMER) {
2724                 dm->dqb_itime = di->d_ino_timer;
2725                 check_ilim = 1;
2726                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2727         }
2728
2729         if (check_blim) {
2730                 if (!dm->dqb_bsoftlimit ||
2731                     dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2732                         dm->dqb_btime = 0;
2733                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2734                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2735                         /* Set grace only if user hasn't provided his own... */
2736                         dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2737         }
2738         if (check_ilim) {
2739                 if (!dm->dqb_isoftlimit ||
2740                     dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2741                         dm->dqb_itime = 0;
2742                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2743                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2744                         /* Set grace only if user hasn't provided his own... */
2745                         dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2746         }
2747         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2748             dm->dqb_isoftlimit)
2749                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2750         else
2751                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2752         spin_unlock(&dquot->dq_dqb_lock);
2753         mark_dquot_dirty(dquot);
2754
2755         return 0;
2756 }
2757
2758 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2759                   struct qc_dqblk *di)
2760 {
2761         struct dquot *dquot;
2762         int rc;
2763
2764         dquot = dqget(sb, qid);
2765         if (IS_ERR(dquot)) {
2766                 rc = PTR_ERR(dquot);
2767                 goto out;
2768         }
2769         rc = do_set_dqblk(dquot, di);
2770         dqput(dquot);
2771 out:
2772         return rc;
2773 }
2774 EXPORT_SYMBOL(dquot_set_dqblk);
2775
2776 /* Generic routine for getting common part of quota file information */
2777 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2778 {
2779         struct mem_dqinfo *mi;
2780         struct qc_type_state *tstate;
2781         struct quota_info *dqopt = sb_dqopt(sb);
2782         int type;
2783
2784         memset(state, 0, sizeof(*state));
2785         for (type = 0; type < MAXQUOTAS; type++) {
2786                 if (!sb_has_quota_active(sb, type))
2787                         continue;
2788                 tstate = state->s_state + type;
2789                 mi = sb_dqopt(sb)->info + type;
2790                 tstate->flags = QCI_ACCT_ENABLED;
2791                 spin_lock(&dq_data_lock);
2792                 if (mi->dqi_flags & DQF_SYS_FILE)
2793                         tstate->flags |= QCI_SYSFILE;
2794                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2795                         tstate->flags |= QCI_ROOT_SQUASH;
2796                 if (sb_has_quota_limits_enabled(sb, type))
2797                         tstate->flags |= QCI_LIMITS_ENFORCED;
2798                 tstate->spc_timelimit = mi->dqi_bgrace;
2799                 tstate->ino_timelimit = mi->dqi_igrace;
2800                 if (dqopt->files[type]) {
2801                         tstate->ino = dqopt->files[type]->i_ino;
2802                         tstate->blocks = dqopt->files[type]->i_blocks;
2803                 }
2804                 tstate->nextents = 1;   /* We don't know... */
2805                 spin_unlock(&dq_data_lock);
2806         }
2807         return 0;
2808 }
2809 EXPORT_SYMBOL(dquot_get_state);
2810
2811 /* Generic routine for setting common part of quota file information */
2812 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2813 {
2814         struct mem_dqinfo *mi;
2815         int err = 0;
2816
2817         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2818             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2819                 return -EINVAL;
2820         if (!sb_has_quota_active(sb, type))
2821                 return -ESRCH;
2822         mi = sb_dqopt(sb)->info + type;
2823         if (ii->i_fieldmask & QC_FLAGS) {
2824                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2825                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2826                         return -EINVAL;
2827         }
2828         spin_lock(&dq_data_lock);
2829         if (ii->i_fieldmask & QC_SPC_TIMER)
2830                 mi->dqi_bgrace = ii->i_spc_timelimit;
2831         if (ii->i_fieldmask & QC_INO_TIMER)
2832                 mi->dqi_igrace = ii->i_ino_timelimit;
2833         if (ii->i_fieldmask & QC_FLAGS) {
2834                 if (ii->i_flags & QCI_ROOT_SQUASH)
2835                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2836                 else
2837                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2838         }
2839         spin_unlock(&dq_data_lock);
2840         mark_info_dirty(sb, type);
2841         /* Force write to disk */
2842         sb->dq_op->write_info(sb, type);
2843         return err;
2844 }
2845 EXPORT_SYMBOL(dquot_set_dqinfo);
2846
2847 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2848         .quota_enable   = dquot_quota_enable,
2849         .quota_disable  = dquot_quota_disable,
2850         .quota_sync     = dquot_quota_sync,
2851         .get_state      = dquot_get_state,
2852         .set_info       = dquot_set_dqinfo,
2853         .get_dqblk      = dquot_get_dqblk,
2854         .get_nextdqblk  = dquot_get_next_dqblk,
2855         .set_dqblk      = dquot_set_dqblk
2856 };
2857 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2858
2859 static int do_proc_dqstats(struct ctl_table *table, int write,
2860                      void *buffer, size_t *lenp, loff_t *ppos)
2861 {
2862         unsigned int type = (unsigned long *)table->data - dqstats.stat;
2863         s64 value = percpu_counter_sum(&dqstats.counter[type]);
2864
2865         /* Filter negative values for non-monotonic counters */
2866         if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2867                           type == DQST_FREE_DQUOTS))
2868                 value = 0;
2869
2870         /* Update global table */
2871         dqstats.stat[type] = value;
2872         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2873 }
2874
2875 static struct ctl_table fs_dqstats_table[] = {
2876         {
2877                 .procname       = "lookups",
2878                 .data           = &dqstats.stat[DQST_LOOKUPS],
2879                 .maxlen         = sizeof(unsigned long),
2880                 .mode           = 0444,
2881                 .proc_handler   = do_proc_dqstats,
2882         },
2883         {
2884                 .procname       = "drops",
2885                 .data           = &dqstats.stat[DQST_DROPS],
2886                 .maxlen         = sizeof(unsigned long),
2887                 .mode           = 0444,
2888                 .proc_handler   = do_proc_dqstats,
2889         },
2890         {
2891                 .procname       = "reads",
2892                 .data           = &dqstats.stat[DQST_READS],
2893                 .maxlen         = sizeof(unsigned long),
2894                 .mode           = 0444,
2895                 .proc_handler   = do_proc_dqstats,
2896         },
2897         {
2898                 .procname       = "writes",
2899                 .data           = &dqstats.stat[DQST_WRITES],
2900                 .maxlen         = sizeof(unsigned long),
2901                 .mode           = 0444,
2902                 .proc_handler   = do_proc_dqstats,
2903         },
2904         {
2905                 .procname       = "cache_hits",
2906                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2907                 .maxlen         = sizeof(unsigned long),
2908                 .mode           = 0444,
2909                 .proc_handler   = do_proc_dqstats,
2910         },
2911         {
2912                 .procname       = "allocated_dquots",
2913                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2914                 .maxlen         = sizeof(unsigned long),
2915                 .mode           = 0444,
2916                 .proc_handler   = do_proc_dqstats,
2917         },
2918         {
2919                 .procname       = "free_dquots",
2920                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2921                 .maxlen         = sizeof(unsigned long),
2922                 .mode           = 0444,
2923                 .proc_handler   = do_proc_dqstats,
2924         },
2925         {
2926                 .procname       = "syncs",
2927                 .data           = &dqstats.stat[DQST_SYNCS],
2928                 .maxlen         = sizeof(unsigned long),
2929                 .mode           = 0444,
2930                 .proc_handler   = do_proc_dqstats,
2931         },
2932 #ifdef CONFIG_PRINT_QUOTA_WARNING
2933         {
2934                 .procname       = "warnings",
2935                 .data           = &flag_print_warnings,
2936                 .maxlen         = sizeof(int),
2937                 .mode           = 0644,
2938                 .proc_handler   = proc_dointvec,
2939         },
2940 #endif
2941         { },
2942 };
2943
2944 static struct ctl_table fs_table[] = {
2945         {
2946                 .procname       = "quota",
2947                 .mode           = 0555,
2948                 .child          = fs_dqstats_table,
2949         },
2950         { },
2951 };
2952
2953 static struct ctl_table sys_table[] = {
2954         {
2955                 .procname       = "fs",
2956                 .mode           = 0555,
2957                 .child          = fs_table,
2958         },
2959         { },
2960 };
2961
2962 static int __init dquot_init(void)
2963 {
2964         int i, ret;
2965         unsigned long nr_hash, order;
2966
2967         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2968
2969         register_sysctl_table(sys_table);
2970
2971         dquot_cachep = kmem_cache_create("dquot",
2972                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2973                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2974                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2975                         NULL);
2976
2977         order = 0;
2978         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2979         if (!dquot_hash)
2980                 panic("Cannot create dquot hash table");
2981
2982         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2983                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2984                 if (ret)
2985                         panic("Cannot create dquot stat counters");
2986         }
2987
2988         /* Find power-of-two hlist_heads which can fit into allocation */
2989         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2990         dq_hash_bits = ilog2(nr_hash);
2991
2992         nr_hash = 1UL << dq_hash_bits;
2993         dq_hash_mask = nr_hash - 1;
2994         for (i = 0; i < nr_hash; i++)
2995                 INIT_HLIST_HEAD(dquot_hash + i);
2996
2997         pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2998                 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2999
3000         if (register_shrinker(&dqcache_shrinker))
3001                 panic("Cannot register dquot shrinker");
3002
3003         return 0;
3004 }
3005 fs_initcall(dquot_init);