Merge branches 'acpi-misc' and 'acpi-video'
[linux-2.6-block.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 "../internal.h" /* ugh */
82
83 #include <linux/uaccess.h>
84
85 /*
86  * There are five quota SMP locks:
87  * * dq_list_lock protects all lists with quotas and quota formats.
88  * * dquot->dq_dqb_lock protects data from dq_dqb
89  * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
90  *   consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
91  *   dquot_transfer() can stabilize amount it transfers
92  * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
93  *   pointers in the inode
94  * * dq_state_lock protects modifications of quota state (on quotaon and
95  *   quotaoff) and readers who care about latest values take it as well.
96  *
97  * The spinlock ordering is hence:
98  *   dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
99  *   dq_list_lock > dq_state_lock
100  *
101  * Note that some things (eg. sb pointer, type, id) doesn't change during
102  * the life of the dquot structure and so needn't to be protected by a lock
103  *
104  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
105  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
106  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
107  * inode and before dropping dquot references to avoid use of dquots after
108  * they are freed. dq_data_lock is used to serialize the pointer setting and
109  * clearing operations.
110  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
111  * inode is a quota file). Functions adding pointers from inode to dquots have
112  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
113  * have to do all pointer modifications before dropping dq_data_lock. This makes
114  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
115  * then drops all pointers to dquots from an inode.
116  *
117  * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
118  * memory (or space for it is being allocated) on the first dqget(), when it is
119  * being written out, and when it is being released on the last dqput(). The
120  * allocation and release operations are serialized by the dq_lock and by
121  * checking the use count in dquot_release().
122  *
123  * Lock ordering (including related VFS locks) is the following:
124  *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
125  */
126
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
128 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
129 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
130 EXPORT_SYMBOL(dq_data_lock);
131 DEFINE_STATIC_SRCU(dquot_srcu);
132
133 static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
134
135 void __quota_error(struct super_block *sb, const char *func,
136                    const char *fmt, ...)
137 {
138         if (printk_ratelimit()) {
139                 va_list args;
140                 struct va_format vaf;
141
142                 va_start(args, fmt);
143
144                 vaf.fmt = fmt;
145                 vaf.va = &args;
146
147                 printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
148                        sb->s_id, func, &vaf);
149
150                 va_end(args);
151         }
152 }
153 EXPORT_SYMBOL(__quota_error);
154
155 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
156 static char *quotatypes[] = INITQFNAMES;
157 #endif
158 static struct quota_format_type *quota_formats; /* List of registered formats */
159 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
160
161 /* SLAB cache for dquot structures */
162 static struct kmem_cache *dquot_cachep;
163
164 int register_quota_format(struct quota_format_type *fmt)
165 {
166         spin_lock(&dq_list_lock);
167         fmt->qf_next = quota_formats;
168         quota_formats = fmt;
169         spin_unlock(&dq_list_lock);
170         return 0;
171 }
172 EXPORT_SYMBOL(register_quota_format);
173
174 void unregister_quota_format(struct quota_format_type *fmt)
175 {
176         struct quota_format_type **actqf;
177
178         spin_lock(&dq_list_lock);
179         for (actqf = &quota_formats; *actqf && *actqf != fmt;
180              actqf = &(*actqf)->qf_next)
181                 ;
182         if (*actqf)
183                 *actqf = (*actqf)->qf_next;
184         spin_unlock(&dq_list_lock);
185 }
186 EXPORT_SYMBOL(unregister_quota_format);
187
188 static struct quota_format_type *find_quota_format(int id)
189 {
190         struct quota_format_type *actqf;
191
192         spin_lock(&dq_list_lock);
193         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
194              actqf = actqf->qf_next)
195                 ;
196         if (!actqf || !try_module_get(actqf->qf_owner)) {
197                 int qm;
198
199                 spin_unlock(&dq_list_lock);
200
201                 for (qm = 0; module_names[qm].qm_fmt_id &&
202                              module_names[qm].qm_fmt_id != id; qm++)
203                         ;
204                 if (!module_names[qm].qm_fmt_id ||
205                     request_module(module_names[qm].qm_mod_name))
206                         return NULL;
207
208                 spin_lock(&dq_list_lock);
209                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
210                      actqf = actqf->qf_next)
211                         ;
212                 if (actqf && !try_module_get(actqf->qf_owner))
213                         actqf = NULL;
214         }
215         spin_unlock(&dq_list_lock);
216         return actqf;
217 }
218
219 static void put_quota_format(struct quota_format_type *fmt)
220 {
221         module_put(fmt->qf_owner);
222 }
223
224 /*
225  * Dquot List Management:
226  * The quota code uses four lists for dquot management: the inuse_list,
227  * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
228  * structure may be on some of those lists, depending on its current state.
229  *
230  * All dquots are placed to the end of inuse_list when first created, and this
231  * list is used for invalidate operation, which must look at every dquot.
232  *
233  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
234  * and this list is searched whenever we need an available dquot.  Dquots are
235  * removed from the list as soon as they are used again, and
236  * dqstats.free_dquots gives the number of dquots on the list. When
237  * dquot is invalidated it's completely released from memory.
238  *
239  * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
240  * dirtied, and this list is searched when writing dirty dquots back to
241  * quota file. Note that some filesystems do dirty dquot tracking on their
242  * own (e.g. in a journal) and thus don't use dqi_dirty_list.
243  *
244  * Dquots with a specific identity (device, type and id) are placed on
245  * one of the dquot_hash[] hash chains. The provides an efficient search
246  * mechanism to locate a specific dquot.
247  */
248
249 static LIST_HEAD(inuse_list);
250 static LIST_HEAD(free_dquots);
251 static unsigned int dq_hash_bits, dq_hash_mask;
252 static struct hlist_head *dquot_hash;
253
254 struct dqstats dqstats;
255 EXPORT_SYMBOL(dqstats);
256
257 static qsize_t inode_get_rsv_space(struct inode *inode);
258 static qsize_t __inode_get_rsv_space(struct inode *inode);
259 static int __dquot_initialize(struct inode *inode, int type);
260
261 static inline unsigned int
262 hashfn(const struct super_block *sb, struct kqid qid)
263 {
264         unsigned int id = from_kqid(&init_user_ns, qid);
265         int type = qid.type;
266         unsigned long tmp;
267
268         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
269         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
270 }
271
272 /*
273  * Following list functions expect dq_list_lock to be held
274  */
275 static inline void insert_dquot_hash(struct dquot *dquot)
276 {
277         struct hlist_head *head;
278         head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
279         hlist_add_head(&dquot->dq_hash, head);
280 }
281
282 static inline void remove_dquot_hash(struct dquot *dquot)
283 {
284         hlist_del_init(&dquot->dq_hash);
285 }
286
287 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
288                                 struct kqid qid)
289 {
290         struct hlist_node *node;
291         struct dquot *dquot;
292
293         hlist_for_each (node, dquot_hash+hashent) {
294                 dquot = hlist_entry(node, struct dquot, dq_hash);
295                 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
296                         return dquot;
297         }
298         return NULL;
299 }
300
301 /* Add a dquot to the tail of the free list */
302 static inline void put_dquot_last(struct dquot *dquot)
303 {
304         list_add_tail(&dquot->dq_free, &free_dquots);
305         dqstats_inc(DQST_FREE_DQUOTS);
306 }
307
308 static inline void remove_free_dquot(struct dquot *dquot)
309 {
310         if (list_empty(&dquot->dq_free))
311                 return;
312         list_del_init(&dquot->dq_free);
313         dqstats_dec(DQST_FREE_DQUOTS);
314 }
315
316 static inline void put_inuse(struct dquot *dquot)
317 {
318         /* We add to the back of inuse list so we don't have to restart
319          * when traversing this list and we block */
320         list_add_tail(&dquot->dq_inuse, &inuse_list);
321         dqstats_inc(DQST_ALLOC_DQUOTS);
322 }
323
324 static inline void remove_inuse(struct dquot *dquot)
325 {
326         dqstats_dec(DQST_ALLOC_DQUOTS);
327         list_del(&dquot->dq_inuse);
328 }
329 /*
330  * End of list functions needing dq_list_lock
331  */
332
333 static void wait_on_dquot(struct dquot *dquot)
334 {
335         mutex_lock(&dquot->dq_lock);
336         mutex_unlock(&dquot->dq_lock);
337 }
338
339 static inline int dquot_dirty(struct dquot *dquot)
340 {
341         return test_bit(DQ_MOD_B, &dquot->dq_flags);
342 }
343
344 static inline int mark_dquot_dirty(struct dquot *dquot)
345 {
346         return dquot->dq_sb->dq_op->mark_dirty(dquot);
347 }
348
349 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
350 int dquot_mark_dquot_dirty(struct dquot *dquot)
351 {
352         int ret = 1;
353
354         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
355                 return 0;
356
357         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
358                 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
359
360         /* If quota is dirty already, we don't have to acquire dq_list_lock */
361         if (test_bit(DQ_MOD_B, &dquot->dq_flags))
362                 return 1;
363
364         spin_lock(&dq_list_lock);
365         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
366                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
367                                 info[dquot->dq_id.type].dqi_dirty_list);
368                 ret = 0;
369         }
370         spin_unlock(&dq_list_lock);
371         return ret;
372 }
373 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
374
375 /* Dirtify all the dquots - this can block when journalling */
376 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
377 {
378         int ret, err, cnt;
379
380         ret = err = 0;
381         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
382                 if (dquot[cnt])
383                         /* Even in case of error we have to continue */
384                         ret = mark_dquot_dirty(dquot[cnt]);
385                 if (!err)
386                         err = ret;
387         }
388         return err;
389 }
390
391 static inline void dqput_all(struct dquot **dquot)
392 {
393         unsigned int cnt;
394
395         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
396                 dqput(dquot[cnt]);
397 }
398
399 static inline int clear_dquot_dirty(struct dquot *dquot)
400 {
401         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
402                 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
403
404         spin_lock(&dq_list_lock);
405         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
406                 spin_unlock(&dq_list_lock);
407                 return 0;
408         }
409         list_del_init(&dquot->dq_dirty);
410         spin_unlock(&dq_list_lock);
411         return 1;
412 }
413
414 void mark_info_dirty(struct super_block *sb, int type)
415 {
416         spin_lock(&dq_data_lock);
417         sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
418         spin_unlock(&dq_data_lock);
419 }
420 EXPORT_SYMBOL(mark_info_dirty);
421
422 /*
423  *      Read dquot from disk and alloc space for it
424  */
425
426 int dquot_acquire(struct dquot *dquot)
427 {
428         int ret = 0, ret2 = 0;
429         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
430
431         mutex_lock(&dquot->dq_lock);
432         if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
433                 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
434                 if (ret < 0)
435                         goto out_iolock;
436         }
437         /* Make sure flags update is visible after dquot has been filled */
438         smp_mb__before_atomic();
439         set_bit(DQ_READ_B, &dquot->dq_flags);
440         /* Instantiate dquot if needed */
441         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
442                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
443                 /* Write the info if needed */
444                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
445                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
446                                         dquot->dq_sb, dquot->dq_id.type);
447                 }
448                 if (ret < 0)
449                         goto out_iolock;
450                 if (ret2 < 0) {
451                         ret = ret2;
452                         goto out_iolock;
453                 }
454         }
455         /*
456          * Make sure flags update is visible after on-disk struct has been
457          * allocated. Paired with smp_rmb() in dqget().
458          */
459         smp_mb__before_atomic();
460         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
461 out_iolock:
462         mutex_unlock(&dquot->dq_lock);
463         return ret;
464 }
465 EXPORT_SYMBOL(dquot_acquire);
466
467 /*
468  *      Write dquot to disk
469  */
470 int dquot_commit(struct dquot *dquot)
471 {
472         int ret = 0;
473         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
474
475         mutex_lock(&dquot->dq_lock);
476         if (!clear_dquot_dirty(dquot))
477                 goto out_lock;
478         /* Inactive dquot can be only if there was error during read/init
479          * => we have better not writing it */
480         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
481                 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
482         else
483                 ret = -EIO;
484 out_lock:
485         mutex_unlock(&dquot->dq_lock);
486         return ret;
487 }
488 EXPORT_SYMBOL(dquot_commit);
489
490 /*
491  *      Release dquot
492  */
493 int dquot_release(struct dquot *dquot)
494 {
495         int ret = 0, ret2 = 0;
496         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
497
498         mutex_lock(&dquot->dq_lock);
499         /* Check whether we are not racing with some other dqget() */
500         if (atomic_read(&dquot->dq_count) > 1)
501                 goto out_dqlock;
502         if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
503                 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
504                 /* Write the info */
505                 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
506                         ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
507                                                 dquot->dq_sb, dquot->dq_id.type);
508                 }
509                 if (ret >= 0)
510                         ret = ret2;
511         }
512         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
513 out_dqlock:
514         mutex_unlock(&dquot->dq_lock);
515         return ret;
516 }
517 EXPORT_SYMBOL(dquot_release);
518
519 void dquot_destroy(struct dquot *dquot)
520 {
521         kmem_cache_free(dquot_cachep, dquot);
522 }
523 EXPORT_SYMBOL(dquot_destroy);
524
525 static inline void do_destroy_dquot(struct dquot *dquot)
526 {
527         dquot->dq_sb->dq_op->destroy_dquot(dquot);
528 }
529
530 /* Invalidate all dquots on the list. Note that this function is called after
531  * quota is disabled and pointers from inodes removed so there cannot be new
532  * quota users. There can still be some users of quotas due to inodes being
533  * just deleted or pruned by prune_icache() (those are not attached to any
534  * list) or parallel quotactl call. We have to wait for such users.
535  */
536 static void invalidate_dquots(struct super_block *sb, int type)
537 {
538         struct dquot *dquot, *tmp;
539
540 restart:
541         spin_lock(&dq_list_lock);
542         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
543                 if (dquot->dq_sb != sb)
544                         continue;
545                 if (dquot->dq_id.type != type)
546                         continue;
547                 /* Wait for dquot users */
548                 if (atomic_read(&dquot->dq_count)) {
549                         dqgrab(dquot);
550                         spin_unlock(&dq_list_lock);
551                         /*
552                          * Once dqput() wakes us up, we know it's time to free
553                          * the dquot.
554                          * IMPORTANT: we rely on the fact that there is always
555                          * at most one process waiting for dquot to free.
556                          * Otherwise dq_count would be > 1 and we would never
557                          * wake up.
558                          */
559                         wait_event(dquot_ref_wq,
560                                    atomic_read(&dquot->dq_count) == 1);
561                         dqput(dquot);
562                         /* At this moment dquot() need not exist (it could be
563                          * reclaimed by prune_dqcache(). Hence we must
564                          * restart. */
565                         goto restart;
566                 }
567                 /*
568                  * Quota now has no users and it has been written on last
569                  * dqput()
570                  */
571                 remove_dquot_hash(dquot);
572                 remove_free_dquot(dquot);
573                 remove_inuse(dquot);
574                 do_destroy_dquot(dquot);
575         }
576         spin_unlock(&dq_list_lock);
577 }
578
579 /* Call callback for every active dquot on given filesystem */
580 int dquot_scan_active(struct super_block *sb,
581                       int (*fn)(struct dquot *dquot, unsigned long priv),
582                       unsigned long priv)
583 {
584         struct dquot *dquot, *old_dquot = NULL;
585         int ret = 0;
586
587         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
588
589         spin_lock(&dq_list_lock);
590         list_for_each_entry(dquot, &inuse_list, dq_inuse) {
591                 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
592                         continue;
593                 if (dquot->dq_sb != sb)
594                         continue;
595                 /* Now we have active dquot so we can just increase use count */
596                 atomic_inc(&dquot->dq_count);
597                 spin_unlock(&dq_list_lock);
598                 dqstats_inc(DQST_LOOKUPS);
599                 dqput(old_dquot);
600                 old_dquot = dquot;
601                 /*
602                  * ->release_dquot() can be racing with us. Our reference
603                  * protects us from new calls to it so just wait for any
604                  * outstanding call and recheck the DQ_ACTIVE_B after that.
605                  */
606                 wait_on_dquot(dquot);
607                 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
608                         ret = fn(dquot, priv);
609                         if (ret < 0)
610                                 goto out;
611                 }
612                 spin_lock(&dq_list_lock);
613                 /* We are safe to continue now because our dquot could not
614                  * be moved out of the inuse list while we hold the reference */
615         }
616         spin_unlock(&dq_list_lock);
617 out:
618         dqput(old_dquot);
619         return ret;
620 }
621 EXPORT_SYMBOL(dquot_scan_active);
622
623 /* Write all dquot structures to quota files */
624 int dquot_writeback_dquots(struct super_block *sb, int type)
625 {
626         struct list_head *dirty;
627         struct dquot *dquot;
628         struct quota_info *dqopt = sb_dqopt(sb);
629         int cnt;
630         int err, ret = 0;
631
632         WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
633
634         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
635                 if (type != -1 && cnt != type)
636                         continue;
637                 if (!sb_has_quota_active(sb, cnt))
638                         continue;
639                 spin_lock(&dq_list_lock);
640                 dirty = &dqopt->info[cnt].dqi_dirty_list;
641                 while (!list_empty(dirty)) {
642                         dquot = list_first_entry(dirty, struct dquot,
643                                                  dq_dirty);
644
645                         WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
646
647                         /* Now we have active dquot from which someone is
648                          * holding reference so we can safely just increase
649                          * use count */
650                         dqgrab(dquot);
651                         spin_unlock(&dq_list_lock);
652                         dqstats_inc(DQST_LOOKUPS);
653                         err = sb->dq_op->write_dquot(dquot);
654                         if (err) {
655                                 /*
656                                  * Clear dirty bit anyway to avoid infinite
657                                  * loop here.
658                                  */
659                                 clear_dquot_dirty(dquot);
660                                 if (!ret)
661                                         ret = err;
662                         }
663                         dqput(dquot);
664                         spin_lock(&dq_list_lock);
665                 }
666                 spin_unlock(&dq_list_lock);
667         }
668
669         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
670                 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
671                     && info_dirty(&dqopt->info[cnt]))
672                         sb->dq_op->write_info(sb, cnt);
673         dqstats_inc(DQST_SYNCS);
674
675         return ret;
676 }
677 EXPORT_SYMBOL(dquot_writeback_dquots);
678
679 /* Write all dquot structures to disk and make them visible from userspace */
680 int dquot_quota_sync(struct super_block *sb, int type)
681 {
682         struct quota_info *dqopt = sb_dqopt(sb);
683         int cnt;
684         int ret;
685
686         ret = dquot_writeback_dquots(sb, type);
687         if (ret)
688                 return ret;
689         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
690                 return 0;
691
692         /* This is not very clever (and fast) but currently I don't know about
693          * any other simple way of getting quota data to disk and we must get
694          * them there for userspace to be visible... */
695         if (sb->s_op->sync_fs)
696                 sb->s_op->sync_fs(sb, 1);
697         sync_blockdev(sb->s_bdev);
698
699         /*
700          * Now when everything is written we can discard the pagecache so
701          * that userspace sees the changes.
702          */
703         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
704                 if (type != -1 && cnt != type)
705                         continue;
706                 if (!sb_has_quota_active(sb, cnt))
707                         continue;
708                 inode_lock(dqopt->files[cnt]);
709                 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
710                 inode_unlock(dqopt->files[cnt]);
711         }
712
713         return 0;
714 }
715 EXPORT_SYMBOL(dquot_quota_sync);
716
717 static unsigned long
718 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
719 {
720         struct dquot *dquot;
721         unsigned long freed = 0;
722
723         spin_lock(&dq_list_lock);
724         while (!list_empty(&free_dquots) && sc->nr_to_scan) {
725                 dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
726                 remove_dquot_hash(dquot);
727                 remove_free_dquot(dquot);
728                 remove_inuse(dquot);
729                 do_destroy_dquot(dquot);
730                 sc->nr_to_scan--;
731                 freed++;
732         }
733         spin_unlock(&dq_list_lock);
734         return freed;
735 }
736
737 static unsigned long
738 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
739 {
740         return vfs_pressure_ratio(
741         percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
742 }
743
744 static struct shrinker dqcache_shrinker = {
745         .count_objects = dqcache_shrink_count,
746         .scan_objects = dqcache_shrink_scan,
747         .seeks = DEFAULT_SEEKS,
748 };
749
750 /*
751  * Put reference to dquot
752  */
753 void dqput(struct dquot *dquot)
754 {
755         int ret;
756
757         if (!dquot)
758                 return;
759 #ifdef CONFIG_QUOTA_DEBUG
760         if (!atomic_read(&dquot->dq_count)) {
761                 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
762                             quotatypes[dquot->dq_id.type],
763                             from_kqid(&init_user_ns, dquot->dq_id));
764                 BUG();
765         }
766 #endif
767         dqstats_inc(DQST_DROPS);
768 we_slept:
769         spin_lock(&dq_list_lock);
770         if (atomic_read(&dquot->dq_count) > 1) {
771                 /* We have more than one user... nothing to do */
772                 atomic_dec(&dquot->dq_count);
773                 /* Releasing dquot during quotaoff phase? */
774                 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
775                     atomic_read(&dquot->dq_count) == 1)
776                         wake_up(&dquot_ref_wq);
777                 spin_unlock(&dq_list_lock);
778                 return;
779         }
780         /* Need to release dquot? */
781         if (dquot_dirty(dquot)) {
782                 spin_unlock(&dq_list_lock);
783                 /* Commit dquot before releasing */
784                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
785                 if (ret < 0) {
786                         quota_error(dquot->dq_sb, "Can't write quota structure"
787                                     " (error %d). Quota may get out of sync!",
788                                     ret);
789                         /*
790                          * We clear dirty bit anyway, so that we avoid
791                          * infinite loop here
792                          */
793                         clear_dquot_dirty(dquot);
794                 }
795                 goto we_slept;
796         }
797         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
798                 spin_unlock(&dq_list_lock);
799                 dquot->dq_sb->dq_op->release_dquot(dquot);
800                 goto we_slept;
801         }
802         atomic_dec(&dquot->dq_count);
803 #ifdef CONFIG_QUOTA_DEBUG
804         /* sanity check */
805         BUG_ON(!list_empty(&dquot->dq_free));
806 #endif
807         put_dquot_last(dquot);
808         spin_unlock(&dq_list_lock);
809 }
810 EXPORT_SYMBOL(dqput);
811
812 struct dquot *dquot_alloc(struct super_block *sb, int type)
813 {
814         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
815 }
816 EXPORT_SYMBOL(dquot_alloc);
817
818 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
819 {
820         struct dquot *dquot;
821
822         dquot = sb->dq_op->alloc_dquot(sb, type);
823         if(!dquot)
824                 return NULL;
825
826         mutex_init(&dquot->dq_lock);
827         INIT_LIST_HEAD(&dquot->dq_free);
828         INIT_LIST_HEAD(&dquot->dq_inuse);
829         INIT_HLIST_NODE(&dquot->dq_hash);
830         INIT_LIST_HEAD(&dquot->dq_dirty);
831         dquot->dq_sb = sb;
832         dquot->dq_id = make_kqid_invalid(type);
833         atomic_set(&dquot->dq_count, 1);
834         spin_lock_init(&dquot->dq_dqb_lock);
835
836         return dquot;
837 }
838
839 /*
840  * Get reference to dquot
841  *
842  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
843  * destroying our dquot by:
844  *   a) checking for quota flags under dq_list_lock and
845  *   b) getting a reference to dquot before we release dq_list_lock
846  */
847 struct dquot *dqget(struct super_block *sb, struct kqid qid)
848 {
849         unsigned int hashent = hashfn(sb, qid);
850         struct dquot *dquot, *empty = NULL;
851
852         if (!qid_has_mapping(sb->s_user_ns, qid))
853                 return ERR_PTR(-EINVAL);
854
855         if (!sb_has_quota_active(sb, qid.type))
856                 return ERR_PTR(-ESRCH);
857 we_slept:
858         spin_lock(&dq_list_lock);
859         spin_lock(&dq_state_lock);
860         if (!sb_has_quota_active(sb, qid.type)) {
861                 spin_unlock(&dq_state_lock);
862                 spin_unlock(&dq_list_lock);
863                 dquot = ERR_PTR(-ESRCH);
864                 goto out;
865         }
866         spin_unlock(&dq_state_lock);
867
868         dquot = find_dquot(hashent, sb, qid);
869         if (!dquot) {
870                 if (!empty) {
871                         spin_unlock(&dq_list_lock);
872                         empty = get_empty_dquot(sb, qid.type);
873                         if (!empty)
874                                 schedule();     /* Try to wait for a moment... */
875                         goto we_slept;
876                 }
877                 dquot = empty;
878                 empty = NULL;
879                 dquot->dq_id = qid;
880                 /* all dquots go on the inuse_list */
881                 put_inuse(dquot);
882                 /* hash it first so it can be found */
883                 insert_dquot_hash(dquot);
884                 spin_unlock(&dq_list_lock);
885                 dqstats_inc(DQST_LOOKUPS);
886         } else {
887                 if (!atomic_read(&dquot->dq_count))
888                         remove_free_dquot(dquot);
889                 atomic_inc(&dquot->dq_count);
890                 spin_unlock(&dq_list_lock);
891                 dqstats_inc(DQST_CACHE_HITS);
892                 dqstats_inc(DQST_LOOKUPS);
893         }
894         /* Wait for dq_lock - after this we know that either dquot_release() is
895          * already finished or it will be canceled due to dq_count > 1 test */
896         wait_on_dquot(dquot);
897         /* Read the dquot / allocate space in quota file */
898         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
899                 int err;
900
901                 err = sb->dq_op->acquire_dquot(dquot);
902                 if (err < 0) {
903                         dqput(dquot);
904                         dquot = ERR_PTR(err);
905                         goto out;
906                 }
907         }
908         /*
909          * Make sure following reads see filled structure - paired with
910          * smp_mb__before_atomic() in dquot_acquire().
911          */
912         smp_rmb();
913 #ifdef CONFIG_QUOTA_DEBUG
914         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
915 #endif
916 out:
917         if (empty)
918                 do_destroy_dquot(empty);
919
920         return dquot;
921 }
922 EXPORT_SYMBOL(dqget);
923
924 static inline struct dquot **i_dquot(struct inode *inode)
925 {
926         return inode->i_sb->s_op->get_dquots(inode);
927 }
928
929 static int dqinit_needed(struct inode *inode, int type)
930 {
931         struct dquot * const *dquots;
932         int cnt;
933
934         if (IS_NOQUOTA(inode))
935                 return 0;
936
937         dquots = i_dquot(inode);
938         if (type != -1)
939                 return !dquots[type];
940         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
941                 if (!dquots[cnt])
942                         return 1;
943         return 0;
944 }
945
946 /* This routine is guarded by s_umount semaphore */
947 static int add_dquot_ref(struct super_block *sb, int type)
948 {
949         struct inode *inode, *old_inode = NULL;
950 #ifdef CONFIG_QUOTA_DEBUG
951         int reserved = 0;
952 #endif
953         int err = 0;
954
955         spin_lock(&sb->s_inode_list_lock);
956         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
957                 spin_lock(&inode->i_lock);
958                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
959                     !atomic_read(&inode->i_writecount) ||
960                     !dqinit_needed(inode, type)) {
961                         spin_unlock(&inode->i_lock);
962                         continue;
963                 }
964                 __iget(inode);
965                 spin_unlock(&inode->i_lock);
966                 spin_unlock(&sb->s_inode_list_lock);
967
968 #ifdef CONFIG_QUOTA_DEBUG
969                 if (unlikely(inode_get_rsv_space(inode) > 0))
970                         reserved = 1;
971 #endif
972                 iput(old_inode);
973                 err = __dquot_initialize(inode, type);
974                 if (err) {
975                         iput(inode);
976                         goto out;
977                 }
978
979                 /*
980                  * We hold a reference to 'inode' so it couldn't have been
981                  * removed from s_inodes list while we dropped the
982                  * s_inode_list_lock. We cannot iput the inode now as we can be
983                  * holding the last reference and we cannot iput it under
984                  * s_inode_list_lock. So we keep the reference and iput it
985                  * later.
986                  */
987                 old_inode = inode;
988                 spin_lock(&sb->s_inode_list_lock);
989         }
990         spin_unlock(&sb->s_inode_list_lock);
991         iput(old_inode);
992 out:
993 #ifdef CONFIG_QUOTA_DEBUG
994         if (reserved) {
995                 quota_error(sb, "Writes happened before quota was turned on "
996                         "thus quota information is probably inconsistent. "
997                         "Please run quotacheck(8)");
998         }
999 #endif
1000         return err;
1001 }
1002
1003 /*
1004  * Remove references to dquots from inode and add dquot to list for freeing
1005  * if we have the last reference to dquot
1006  */
1007 static void remove_inode_dquot_ref(struct inode *inode, int type,
1008                                    struct list_head *tofree_head)
1009 {
1010         struct dquot **dquots = i_dquot(inode);
1011         struct dquot *dquot = dquots[type];
1012
1013         if (!dquot)
1014                 return;
1015
1016         dquots[type] = NULL;
1017         if (list_empty(&dquot->dq_free)) {
1018                 /*
1019                  * The inode still has reference to dquot so it can't be in the
1020                  * free list
1021                  */
1022                 spin_lock(&dq_list_lock);
1023                 list_add(&dquot->dq_free, tofree_head);
1024                 spin_unlock(&dq_list_lock);
1025         } else {
1026                 /*
1027                  * Dquot is already in a list to put so we won't drop the last
1028                  * reference here.
1029                  */
1030                 dqput(dquot);
1031         }
1032 }
1033
1034 /*
1035  * Free list of dquots
1036  * Dquots are removed from inodes and no new references can be got so we are
1037  * the only ones holding reference
1038  */
1039 static void put_dquot_list(struct list_head *tofree_head)
1040 {
1041         struct list_head *act_head;
1042         struct dquot *dquot;
1043
1044         act_head = tofree_head->next;
1045         while (act_head != tofree_head) {
1046                 dquot = list_entry(act_head, struct dquot, dq_free);
1047                 act_head = act_head->next;
1048                 /* Remove dquot from the list so we won't have problems... */
1049                 list_del_init(&dquot->dq_free);
1050                 dqput(dquot);
1051         }
1052 }
1053
1054 static void remove_dquot_ref(struct super_block *sb, int type,
1055                 struct list_head *tofree_head)
1056 {
1057         struct inode *inode;
1058 #ifdef CONFIG_QUOTA_DEBUG
1059         int reserved = 0;
1060 #endif
1061
1062         spin_lock(&sb->s_inode_list_lock);
1063         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1064                 /*
1065                  *  We have to scan also I_NEW inodes because they can already
1066                  *  have quota pointer initialized. Luckily, we need to touch
1067                  *  only quota pointers and these have separate locking
1068                  *  (dq_data_lock).
1069                  */
1070                 spin_lock(&dq_data_lock);
1071                 if (!IS_NOQUOTA(inode)) {
1072 #ifdef CONFIG_QUOTA_DEBUG
1073                         if (unlikely(inode_get_rsv_space(inode) > 0))
1074                                 reserved = 1;
1075 #endif
1076                         remove_inode_dquot_ref(inode, type, tofree_head);
1077                 }
1078                 spin_unlock(&dq_data_lock);
1079         }
1080         spin_unlock(&sb->s_inode_list_lock);
1081 #ifdef CONFIG_QUOTA_DEBUG
1082         if (reserved) {
1083                 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1084                         " was disabled thus quota information is probably "
1085                         "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1086         }
1087 #endif
1088 }
1089
1090 /* Gather all references from inodes and drop them */
1091 static void drop_dquot_ref(struct super_block *sb, int type)
1092 {
1093         LIST_HEAD(tofree_head);
1094
1095         if (sb->dq_op) {
1096                 remove_dquot_ref(sb, type, &tofree_head);
1097                 synchronize_srcu(&dquot_srcu);
1098                 put_dquot_list(&tofree_head);
1099         }
1100 }
1101
1102 static inline
1103 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1104 {
1105         if (dquot->dq_dqb.dqb_rsvspace >= number)
1106                 dquot->dq_dqb.dqb_rsvspace -= number;
1107         else {
1108                 WARN_ON_ONCE(1);
1109                 dquot->dq_dqb.dqb_rsvspace = 0;
1110         }
1111         if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1112             dquot->dq_dqb.dqb_bsoftlimit)
1113                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1114         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1115 }
1116
1117 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1118 {
1119         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1120             dquot->dq_dqb.dqb_curinodes >= number)
1121                 dquot->dq_dqb.dqb_curinodes -= number;
1122         else
1123                 dquot->dq_dqb.dqb_curinodes = 0;
1124         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1125                 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1126         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1127 }
1128
1129 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1130 {
1131         if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1132             dquot->dq_dqb.dqb_curspace >= number)
1133                 dquot->dq_dqb.dqb_curspace -= number;
1134         else
1135                 dquot->dq_dqb.dqb_curspace = 0;
1136         if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1137             dquot->dq_dqb.dqb_bsoftlimit)
1138                 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1139         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1140 }
1141
1142 struct dquot_warn {
1143         struct super_block *w_sb;
1144         struct kqid w_dq_id;
1145         short w_type;
1146 };
1147
1148 static int warning_issued(struct dquot *dquot, const int warntype)
1149 {
1150         int flag = (warntype == QUOTA_NL_BHARDWARN ||
1151                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1152                 ((warntype == QUOTA_NL_IHARDWARN ||
1153                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1154
1155         if (!flag)
1156                 return 0;
1157         return test_and_set_bit(flag, &dquot->dq_flags);
1158 }
1159
1160 #ifdef CONFIG_PRINT_QUOTA_WARNING
1161 static int flag_print_warnings = 1;
1162
1163 static int need_print_warning(struct dquot_warn *warn)
1164 {
1165         if (!flag_print_warnings)
1166                 return 0;
1167
1168         switch (warn->w_dq_id.type) {
1169                 case USRQUOTA:
1170                         return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1171                 case GRPQUOTA:
1172                         return in_group_p(warn->w_dq_id.gid);
1173                 case PRJQUOTA:
1174                         return 1;
1175         }
1176         return 0;
1177 }
1178
1179 /* Print warning to user which exceeded quota */
1180 static void print_warning(struct dquot_warn *warn)
1181 {
1182         char *msg = NULL;
1183         struct tty_struct *tty;
1184         int warntype = warn->w_type;
1185
1186         if (warntype == QUOTA_NL_IHARDBELOW ||
1187             warntype == QUOTA_NL_ISOFTBELOW ||
1188             warntype == QUOTA_NL_BHARDBELOW ||
1189             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1190                 return;
1191
1192         tty = get_current_tty();
1193         if (!tty)
1194                 return;
1195         tty_write_message(tty, warn->w_sb->s_id);
1196         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1197                 tty_write_message(tty, ": warning, ");
1198         else
1199                 tty_write_message(tty, ": write failed, ");
1200         tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1201         switch (warntype) {
1202                 case QUOTA_NL_IHARDWARN:
1203                         msg = " file limit reached.\r\n";
1204                         break;
1205                 case QUOTA_NL_ISOFTLONGWARN:
1206                         msg = " file quota exceeded too long.\r\n";
1207                         break;
1208                 case QUOTA_NL_ISOFTWARN:
1209                         msg = " file quota exceeded.\r\n";
1210                         break;
1211                 case QUOTA_NL_BHARDWARN:
1212                         msg = " block limit reached.\r\n";
1213                         break;
1214                 case QUOTA_NL_BSOFTLONGWARN:
1215                         msg = " block quota exceeded too long.\r\n";
1216                         break;
1217                 case QUOTA_NL_BSOFTWARN:
1218                         msg = " block quota exceeded.\r\n";
1219                         break;
1220         }
1221         tty_write_message(tty, msg);
1222         tty_kref_put(tty);
1223 }
1224 #endif
1225
1226 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1227                             int warntype)
1228 {
1229         if (warning_issued(dquot, warntype))
1230                 return;
1231         warn->w_type = warntype;
1232         warn->w_sb = dquot->dq_sb;
1233         warn->w_dq_id = dquot->dq_id;
1234 }
1235
1236 /*
1237  * Write warnings to the console and send warning messages over netlink.
1238  *
1239  * Note that this function can call into tty and networking code.
1240  */
1241 static void flush_warnings(struct dquot_warn *warn)
1242 {
1243         int i;
1244
1245         for (i = 0; i < MAXQUOTAS; i++) {
1246                 if (warn[i].w_type == QUOTA_NL_NOWARN)
1247                         continue;
1248 #ifdef CONFIG_PRINT_QUOTA_WARNING
1249                 print_warning(&warn[i]);
1250 #endif
1251                 quota_send_warning(warn[i].w_dq_id,
1252                                    warn[i].w_sb->s_dev, warn[i].w_type);
1253         }
1254 }
1255
1256 static int ignore_hardlimit(struct dquot *dquot)
1257 {
1258         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1259
1260         return capable(CAP_SYS_RESOURCE) &&
1261                (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1262                 !(info->dqi_flags & DQF_ROOT_SQUASH));
1263 }
1264
1265 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1266                             struct dquot_warn *warn)
1267 {
1268         qsize_t newinodes;
1269         int ret = 0;
1270
1271         spin_lock(&dquot->dq_dqb_lock);
1272         newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1273         if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1274             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1275                 goto add;
1276
1277         if (dquot->dq_dqb.dqb_ihardlimit &&
1278             newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1279             !ignore_hardlimit(dquot)) {
1280                 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1281                 ret = -EDQUOT;
1282                 goto out;
1283         }
1284
1285         if (dquot->dq_dqb.dqb_isoftlimit &&
1286             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1287             dquot->dq_dqb.dqb_itime &&
1288             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1289             !ignore_hardlimit(dquot)) {
1290                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1291                 ret = -EDQUOT;
1292                 goto out;
1293         }
1294
1295         if (dquot->dq_dqb.dqb_isoftlimit &&
1296             newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1297             dquot->dq_dqb.dqb_itime == 0) {
1298                 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1299                 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1300                     sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1301         }
1302 add:
1303         dquot->dq_dqb.dqb_curinodes = newinodes;
1304
1305 out:
1306         spin_unlock(&dquot->dq_dqb_lock);
1307         return ret;
1308 }
1309
1310 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1311                            qsize_t rsv_space, unsigned int flags,
1312                            struct dquot_warn *warn)
1313 {
1314         qsize_t tspace;
1315         struct super_block *sb = dquot->dq_sb;
1316         int ret = 0;
1317
1318         spin_lock(&dquot->dq_dqb_lock);
1319         if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1320             test_bit(DQ_FAKE_B, &dquot->dq_flags))
1321                 goto finish;
1322
1323         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1324                 + space + rsv_space;
1325
1326         if (dquot->dq_dqb.dqb_bhardlimit &&
1327             tspace > dquot->dq_dqb.dqb_bhardlimit &&
1328             !ignore_hardlimit(dquot)) {
1329                 if (flags & DQUOT_SPACE_WARN)
1330                         prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1331                 ret = -EDQUOT;
1332                 goto finish;
1333         }
1334
1335         if (dquot->dq_dqb.dqb_bsoftlimit &&
1336             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1337             dquot->dq_dqb.dqb_btime &&
1338             ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1339             !ignore_hardlimit(dquot)) {
1340                 if (flags & DQUOT_SPACE_WARN)
1341                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1342                 ret = -EDQUOT;
1343                 goto finish;
1344         }
1345
1346         if (dquot->dq_dqb.dqb_bsoftlimit &&
1347             tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1348             dquot->dq_dqb.dqb_btime == 0) {
1349                 if (flags & DQUOT_SPACE_WARN) {
1350                         prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1351                         dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1352                             sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1353                 } else {
1354                         /*
1355                          * We don't allow preallocation to exceed softlimit so exceeding will
1356                          * be always printed
1357                          */
1358                         ret = -EDQUOT;
1359                         goto finish;
1360                 }
1361         }
1362 finish:
1363         /*
1364          * We have to be careful and go through warning generation & grace time
1365          * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1366          * only here...
1367          */
1368         if (flags & DQUOT_SPACE_NOFAIL)
1369                 ret = 0;
1370         if (!ret) {
1371                 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1372                 dquot->dq_dqb.dqb_curspace += space;
1373         }
1374         spin_unlock(&dquot->dq_dqb_lock);
1375         return ret;
1376 }
1377
1378 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1379 {
1380         qsize_t newinodes;
1381
1382         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1383             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1384             !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1385                 return QUOTA_NL_NOWARN;
1386
1387         newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1388         if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1389                 return QUOTA_NL_ISOFTBELOW;
1390         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1391             newinodes < dquot->dq_dqb.dqb_ihardlimit)
1392                 return QUOTA_NL_IHARDBELOW;
1393         return QUOTA_NL_NOWARN;
1394 }
1395
1396 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1397 {
1398         qsize_t tspace;
1399
1400         tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1401
1402         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1403             tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1404                 return QUOTA_NL_NOWARN;
1405
1406         if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1407                 return QUOTA_NL_BSOFTBELOW;
1408         if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1409             tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1410                 return QUOTA_NL_BHARDBELOW;
1411         return QUOTA_NL_NOWARN;
1412 }
1413
1414 static int dquot_active(const struct inode *inode)
1415 {
1416         struct super_block *sb = inode->i_sb;
1417
1418         if (IS_NOQUOTA(inode))
1419                 return 0;
1420         return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1421 }
1422
1423 /*
1424  * Initialize quota pointers in inode
1425  *
1426  * It is better to call this function outside of any transaction as it
1427  * might need a lot of space in journal for dquot structure allocation.
1428  */
1429 static int __dquot_initialize(struct inode *inode, int type)
1430 {
1431         int cnt, init_needed = 0;
1432         struct dquot **dquots, *got[MAXQUOTAS] = {};
1433         struct super_block *sb = inode->i_sb;
1434         qsize_t rsv;
1435         int ret = 0;
1436
1437         if (!dquot_active(inode))
1438                 return 0;
1439
1440         dquots = i_dquot(inode);
1441
1442         /* First get references to structures we might need. */
1443         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1444                 struct kqid qid;
1445                 kprojid_t projid;
1446                 int rc;
1447                 struct dquot *dquot;
1448
1449                 if (type != -1 && cnt != type)
1450                         continue;
1451                 /*
1452                  * The i_dquot should have been initialized in most cases,
1453                  * we check it without locking here to avoid unnecessary
1454                  * dqget()/dqput() calls.
1455                  */
1456                 if (dquots[cnt])
1457                         continue;
1458
1459                 if (!sb_has_quota_active(sb, cnt))
1460                         continue;
1461
1462                 init_needed = 1;
1463
1464                 switch (cnt) {
1465                 case USRQUOTA:
1466                         qid = make_kqid_uid(inode->i_uid);
1467                         break;
1468                 case GRPQUOTA:
1469                         qid = make_kqid_gid(inode->i_gid);
1470                         break;
1471                 case PRJQUOTA:
1472                         rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1473                         if (rc)
1474                                 continue;
1475                         qid = make_kqid_projid(projid);
1476                         break;
1477                 }
1478                 dquot = dqget(sb, qid);
1479                 if (IS_ERR(dquot)) {
1480                         /* We raced with somebody turning quotas off... */
1481                         if (PTR_ERR(dquot) != -ESRCH) {
1482                                 ret = PTR_ERR(dquot);
1483                                 goto out_put;
1484                         }
1485                         dquot = NULL;
1486                 }
1487                 got[cnt] = dquot;
1488         }
1489
1490         /* All required i_dquot has been initialized */
1491         if (!init_needed)
1492                 return 0;
1493
1494         spin_lock(&dq_data_lock);
1495         if (IS_NOQUOTA(inode))
1496                 goto out_lock;
1497         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1498                 if (type != -1 && cnt != type)
1499                         continue;
1500                 /* Avoid races with quotaoff() */
1501                 if (!sb_has_quota_active(sb, cnt))
1502                         continue;
1503                 /* We could race with quotaon or dqget() could have failed */
1504                 if (!got[cnt])
1505                         continue;
1506                 if (!dquots[cnt]) {
1507                         dquots[cnt] = got[cnt];
1508                         got[cnt] = NULL;
1509                         /*
1510                          * Make quota reservation system happy if someone
1511                          * did a write before quota was turned on
1512                          */
1513                         rsv = inode_get_rsv_space(inode);
1514                         if (unlikely(rsv)) {
1515                                 spin_lock(&inode->i_lock);
1516                                 /* Get reservation again under proper lock */
1517                                 rsv = __inode_get_rsv_space(inode);
1518                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1519                                 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1520                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1521                                 spin_unlock(&inode->i_lock);
1522                         }
1523                 }
1524         }
1525 out_lock:
1526         spin_unlock(&dq_data_lock);
1527 out_put:
1528         /* Drop unused references */
1529         dqput_all(got);
1530
1531         return ret;
1532 }
1533
1534 int dquot_initialize(struct inode *inode)
1535 {
1536         return __dquot_initialize(inode, -1);
1537 }
1538 EXPORT_SYMBOL(dquot_initialize);
1539
1540 bool dquot_initialize_needed(struct inode *inode)
1541 {
1542         struct dquot **dquots;
1543         int i;
1544
1545         if (!dquot_active(inode))
1546                 return false;
1547
1548         dquots = i_dquot(inode);
1549         for (i = 0; i < MAXQUOTAS; i++)
1550                 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1551                         return true;
1552         return false;
1553 }
1554 EXPORT_SYMBOL(dquot_initialize_needed);
1555
1556 /*
1557  * Release all quotas referenced by inode.
1558  *
1559  * This function only be called on inode free or converting
1560  * a file to quota file, no other users for the i_dquot in
1561  * both cases, so we needn't call synchronize_srcu() after
1562  * clearing i_dquot.
1563  */
1564 static void __dquot_drop(struct inode *inode)
1565 {
1566         int cnt;
1567         struct dquot **dquots = i_dquot(inode);
1568         struct dquot *put[MAXQUOTAS];
1569
1570         spin_lock(&dq_data_lock);
1571         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1572                 put[cnt] = dquots[cnt];
1573                 dquots[cnt] = NULL;
1574         }
1575         spin_unlock(&dq_data_lock);
1576         dqput_all(put);
1577 }
1578
1579 void dquot_drop(struct inode *inode)
1580 {
1581         struct dquot * const *dquots;
1582         int cnt;
1583
1584         if (IS_NOQUOTA(inode))
1585                 return;
1586
1587         /*
1588          * Test before calling to rule out calls from proc and such
1589          * where we are not allowed to block. Note that this is
1590          * actually reliable test even without the lock - the caller
1591          * must assure that nobody can come after the DQUOT_DROP and
1592          * add quota pointers back anyway.
1593          */
1594         dquots = i_dquot(inode);
1595         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1596                 if (dquots[cnt])
1597                         break;
1598         }
1599
1600         if (cnt < MAXQUOTAS)
1601                 __dquot_drop(inode);
1602 }
1603 EXPORT_SYMBOL(dquot_drop);
1604
1605 /*
1606  * inode_reserved_space is managed internally by quota, and protected by
1607  * i_lock similar to i_blocks+i_bytes.
1608  */
1609 static qsize_t *inode_reserved_space(struct inode * inode)
1610 {
1611         /* Filesystem must explicitly define it's own method in order to use
1612          * quota reservation interface */
1613         BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1614         return inode->i_sb->dq_op->get_reserved_space(inode);
1615 }
1616
1617 static qsize_t __inode_get_rsv_space(struct inode *inode)
1618 {
1619         if (!inode->i_sb->dq_op->get_reserved_space)
1620                 return 0;
1621         return *inode_reserved_space(inode);
1622 }
1623
1624 static qsize_t inode_get_rsv_space(struct inode *inode)
1625 {
1626         qsize_t ret;
1627
1628         if (!inode->i_sb->dq_op->get_reserved_space)
1629                 return 0;
1630         spin_lock(&inode->i_lock);
1631         ret = __inode_get_rsv_space(inode);
1632         spin_unlock(&inode->i_lock);
1633         return ret;
1634 }
1635
1636 /*
1637  * This functions updates i_blocks+i_bytes fields and quota information
1638  * (together with appropriate checks).
1639  *
1640  * NOTE: We absolutely rely on the fact that caller dirties the inode
1641  * (usually helpers in quotaops.h care about this) and holds a handle for
1642  * the current transaction so that dquot write and inode write go into the
1643  * same transaction.
1644  */
1645
1646 /*
1647  * This operation can block, but only after everything is updated
1648  */
1649 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1650 {
1651         int cnt, ret = 0, index;
1652         struct dquot_warn warn[MAXQUOTAS];
1653         int reserve = flags & DQUOT_SPACE_RESERVE;
1654         struct dquot **dquots;
1655
1656         if (!dquot_active(inode)) {
1657                 if (reserve) {
1658                         spin_lock(&inode->i_lock);
1659                         *inode_reserved_space(inode) += number;
1660                         spin_unlock(&inode->i_lock);
1661                 } else {
1662                         inode_add_bytes(inode, number);
1663                 }
1664                 goto out;
1665         }
1666
1667         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1668                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1669
1670         dquots = i_dquot(inode);
1671         index = srcu_read_lock(&dquot_srcu);
1672         spin_lock(&inode->i_lock);
1673         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1674                 if (!dquots[cnt])
1675                         continue;
1676                 if (reserve) {
1677                         ret = dquot_add_space(dquots[cnt], 0, number, flags,
1678                                               &warn[cnt]);
1679                 } else {
1680                         ret = dquot_add_space(dquots[cnt], number, 0, flags,
1681                                               &warn[cnt]);
1682                 }
1683                 if (ret) {
1684                         /* Back out changes we already did */
1685                         for (cnt--; cnt >= 0; cnt--) {
1686                                 if (!dquots[cnt])
1687                                         continue;
1688                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1689                                 if (reserve)
1690                                         dquot_free_reserved_space(dquots[cnt],
1691                                                                   number);
1692                                 else
1693                                         dquot_decr_space(dquots[cnt], number);
1694                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1695                         }
1696                         spin_unlock(&inode->i_lock);
1697                         goto out_flush_warn;
1698                 }
1699         }
1700         if (reserve)
1701                 *inode_reserved_space(inode) += number;
1702         else
1703                 __inode_add_bytes(inode, number);
1704         spin_unlock(&inode->i_lock);
1705
1706         if (reserve)
1707                 goto out_flush_warn;
1708         mark_all_dquot_dirty(dquots);
1709 out_flush_warn:
1710         srcu_read_unlock(&dquot_srcu, index);
1711         flush_warnings(warn);
1712 out:
1713         return ret;
1714 }
1715 EXPORT_SYMBOL(__dquot_alloc_space);
1716
1717 /*
1718  * This operation can block, but only after everything is updated
1719  */
1720 int dquot_alloc_inode(struct inode *inode)
1721 {
1722         int cnt, ret = 0, index;
1723         struct dquot_warn warn[MAXQUOTAS];
1724         struct dquot * const *dquots;
1725
1726         if (!dquot_active(inode))
1727                 return 0;
1728         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1729                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1730
1731         dquots = i_dquot(inode);
1732         index = srcu_read_lock(&dquot_srcu);
1733         spin_lock(&inode->i_lock);
1734         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1735                 if (!dquots[cnt])
1736                         continue;
1737                 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1738                 if (ret) {
1739                         for (cnt--; cnt >= 0; cnt--) {
1740                                 if (!dquots[cnt])
1741                                         continue;
1742                                 /* Back out changes we already did */
1743                                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1744                                 dquot_decr_inodes(dquots[cnt], 1);
1745                                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1746                         }
1747                         goto warn_put_all;
1748                 }
1749         }
1750
1751 warn_put_all:
1752         spin_unlock(&inode->i_lock);
1753         if (ret == 0)
1754                 mark_all_dquot_dirty(dquots);
1755         srcu_read_unlock(&dquot_srcu, index);
1756         flush_warnings(warn);
1757         return ret;
1758 }
1759 EXPORT_SYMBOL(dquot_alloc_inode);
1760
1761 /*
1762  * Convert in-memory reserved quotas to real consumed quotas
1763  */
1764 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1765 {
1766         struct dquot **dquots;
1767         int cnt, index;
1768
1769         if (!dquot_active(inode)) {
1770                 spin_lock(&inode->i_lock);
1771                 *inode_reserved_space(inode) -= number;
1772                 __inode_add_bytes(inode, number);
1773                 spin_unlock(&inode->i_lock);
1774                 return 0;
1775         }
1776
1777         dquots = i_dquot(inode);
1778         index = srcu_read_lock(&dquot_srcu);
1779         spin_lock(&inode->i_lock);
1780         /* Claim reserved quotas to allocated quotas */
1781         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1782                 if (dquots[cnt]) {
1783                         struct dquot *dquot = dquots[cnt];
1784
1785                         spin_lock(&dquot->dq_dqb_lock);
1786                         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1787                                 number = dquot->dq_dqb.dqb_rsvspace;
1788                         dquot->dq_dqb.dqb_curspace += number;
1789                         dquot->dq_dqb.dqb_rsvspace -= number;
1790                         spin_unlock(&dquot->dq_dqb_lock);
1791                 }
1792         }
1793         /* Update inode bytes */
1794         *inode_reserved_space(inode) -= number;
1795         __inode_add_bytes(inode, number);
1796         spin_unlock(&inode->i_lock);
1797         mark_all_dquot_dirty(dquots);
1798         srcu_read_unlock(&dquot_srcu, index);
1799         return 0;
1800 }
1801 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1802
1803 /*
1804  * Convert allocated space back to in-memory reserved quotas
1805  */
1806 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1807 {
1808         struct dquot **dquots;
1809         int cnt, index;
1810
1811         if (!dquot_active(inode)) {
1812                 spin_lock(&inode->i_lock);
1813                 *inode_reserved_space(inode) += number;
1814                 __inode_sub_bytes(inode, number);
1815                 spin_unlock(&inode->i_lock);
1816                 return;
1817         }
1818
1819         dquots = i_dquot(inode);
1820         index = srcu_read_lock(&dquot_srcu);
1821         spin_lock(&inode->i_lock);
1822         /* Claim reserved quotas to allocated quotas */
1823         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1824                 if (dquots[cnt]) {
1825                         struct dquot *dquot = dquots[cnt];
1826
1827                         spin_lock(&dquot->dq_dqb_lock);
1828                         if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1829                                 number = dquot->dq_dqb.dqb_curspace;
1830                         dquot->dq_dqb.dqb_rsvspace += number;
1831                         dquot->dq_dqb.dqb_curspace -= number;
1832                         spin_unlock(&dquot->dq_dqb_lock);
1833                 }
1834         }
1835         /* Update inode bytes */
1836         *inode_reserved_space(inode) += number;
1837         __inode_sub_bytes(inode, number);
1838         spin_unlock(&inode->i_lock);
1839         mark_all_dquot_dirty(dquots);
1840         srcu_read_unlock(&dquot_srcu, index);
1841         return;
1842 }
1843 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1844
1845 /*
1846  * This operation can block, but only after everything is updated
1847  */
1848 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1849 {
1850         unsigned int cnt;
1851         struct dquot_warn warn[MAXQUOTAS];
1852         struct dquot **dquots;
1853         int reserve = flags & DQUOT_SPACE_RESERVE, index;
1854
1855         if (!dquot_active(inode)) {
1856                 if (reserve) {
1857                         spin_lock(&inode->i_lock);
1858                         *inode_reserved_space(inode) -= number;
1859                         spin_unlock(&inode->i_lock);
1860                 } else {
1861                         inode_sub_bytes(inode, number);
1862                 }
1863                 return;
1864         }
1865
1866         dquots = i_dquot(inode);
1867         index = srcu_read_lock(&dquot_srcu);
1868         spin_lock(&inode->i_lock);
1869         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1870                 int wtype;
1871
1872                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1873                 if (!dquots[cnt])
1874                         continue;
1875                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1876                 wtype = info_bdq_free(dquots[cnt], number);
1877                 if (wtype != QUOTA_NL_NOWARN)
1878                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1879                 if (reserve)
1880                         dquot_free_reserved_space(dquots[cnt], number);
1881                 else
1882                         dquot_decr_space(dquots[cnt], number);
1883                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1884         }
1885         if (reserve)
1886                 *inode_reserved_space(inode) -= number;
1887         else
1888                 __inode_sub_bytes(inode, number);
1889         spin_unlock(&inode->i_lock);
1890
1891         if (reserve)
1892                 goto out_unlock;
1893         mark_all_dquot_dirty(dquots);
1894 out_unlock:
1895         srcu_read_unlock(&dquot_srcu, index);
1896         flush_warnings(warn);
1897 }
1898 EXPORT_SYMBOL(__dquot_free_space);
1899
1900 /*
1901  * This operation can block, but only after everything is updated
1902  */
1903 void dquot_free_inode(struct inode *inode)
1904 {
1905         unsigned int cnt;
1906         struct dquot_warn warn[MAXQUOTAS];
1907         struct dquot * const *dquots;
1908         int index;
1909
1910         if (!dquot_active(inode))
1911                 return;
1912
1913         dquots = i_dquot(inode);
1914         index = srcu_read_lock(&dquot_srcu);
1915         spin_lock(&inode->i_lock);
1916         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1917                 int wtype;
1918
1919                 warn[cnt].w_type = QUOTA_NL_NOWARN;
1920                 if (!dquots[cnt])
1921                         continue;
1922                 spin_lock(&dquots[cnt]->dq_dqb_lock);
1923                 wtype = info_idq_free(dquots[cnt], 1);
1924                 if (wtype != QUOTA_NL_NOWARN)
1925                         prepare_warning(&warn[cnt], dquots[cnt], wtype);
1926                 dquot_decr_inodes(dquots[cnt], 1);
1927                 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1928         }
1929         spin_unlock(&inode->i_lock);
1930         mark_all_dquot_dirty(dquots);
1931         srcu_read_unlock(&dquot_srcu, index);
1932         flush_warnings(warn);
1933 }
1934 EXPORT_SYMBOL(dquot_free_inode);
1935
1936 /*
1937  * Transfer the number of inode and blocks from one diskquota to an other.
1938  * On success, dquot references in transfer_to are consumed and references
1939  * to original dquots that need to be released are placed there. On failure,
1940  * references are kept untouched.
1941  *
1942  * This operation can block, but only after everything is updated
1943  * A transaction must be started when entering this function.
1944  *
1945  * We are holding reference on transfer_from & transfer_to, no need to
1946  * protect them by srcu_read_lock().
1947  */
1948 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1949 {
1950         qsize_t cur_space;
1951         qsize_t rsv_space = 0;
1952         qsize_t inode_usage = 1;
1953         struct dquot *transfer_from[MAXQUOTAS] = {};
1954         int cnt, ret = 0;
1955         char is_valid[MAXQUOTAS] = {};
1956         struct dquot_warn warn_to[MAXQUOTAS];
1957         struct dquot_warn warn_from_inodes[MAXQUOTAS];
1958         struct dquot_warn warn_from_space[MAXQUOTAS];
1959
1960         if (IS_NOQUOTA(inode))
1961                 return 0;
1962
1963         if (inode->i_sb->dq_op->get_inode_usage) {
1964                 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1965                 if (ret)
1966                         return ret;
1967         }
1968
1969         /* Initialize the arrays */
1970         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1971                 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1972                 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1973                 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1974         }
1975
1976         spin_lock(&dq_data_lock);
1977         spin_lock(&inode->i_lock);
1978         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1979                 spin_unlock(&inode->i_lock);
1980                 spin_unlock(&dq_data_lock);
1981                 return 0;
1982         }
1983         cur_space = __inode_get_bytes(inode);
1984         rsv_space = __inode_get_rsv_space(inode);
1985         /*
1986          * Build the transfer_from list, check limits, and update usage in
1987          * the target structures.
1988          */
1989         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1990                 /*
1991                  * Skip changes for same uid or gid or for turned off quota-type.
1992                  */
1993                 if (!transfer_to[cnt])
1994                         continue;
1995                 /* Avoid races with quotaoff() */
1996                 if (!sb_has_quota_active(inode->i_sb, cnt))
1997                         continue;
1998                 is_valid[cnt] = 1;
1999                 transfer_from[cnt] = i_dquot(inode)[cnt];
2000                 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2001                                        &warn_to[cnt]);
2002                 if (ret)
2003                         goto over_quota;
2004                 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2005                                       DQUOT_SPACE_WARN, &warn_to[cnt]);
2006                 if (ret) {
2007                         spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2008                         dquot_decr_inodes(transfer_to[cnt], inode_usage);
2009                         spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2010                         goto over_quota;
2011                 }
2012         }
2013
2014         /* Decrease usage for source structures and update quota pointers */
2015         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2016                 if (!is_valid[cnt])
2017                         continue;
2018                 /* Due to IO error we might not have transfer_from[] structure */
2019                 if (transfer_from[cnt]) {
2020                         int wtype;
2021
2022                         spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2023                         wtype = info_idq_free(transfer_from[cnt], inode_usage);
2024                         if (wtype != QUOTA_NL_NOWARN)
2025                                 prepare_warning(&warn_from_inodes[cnt],
2026                                                 transfer_from[cnt], wtype);
2027                         wtype = info_bdq_free(transfer_from[cnt],
2028                                               cur_space + rsv_space);
2029                         if (wtype != QUOTA_NL_NOWARN)
2030                                 prepare_warning(&warn_from_space[cnt],
2031                                                 transfer_from[cnt], wtype);
2032                         dquot_decr_inodes(transfer_from[cnt], inode_usage);
2033                         dquot_decr_space(transfer_from[cnt], cur_space);
2034                         dquot_free_reserved_space(transfer_from[cnt],
2035                                                   rsv_space);
2036                         spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2037                 }
2038                 i_dquot(inode)[cnt] = transfer_to[cnt];
2039         }
2040         spin_unlock(&inode->i_lock);
2041         spin_unlock(&dq_data_lock);
2042
2043         mark_all_dquot_dirty(transfer_from);
2044         mark_all_dquot_dirty(transfer_to);
2045         flush_warnings(warn_to);
2046         flush_warnings(warn_from_inodes);
2047         flush_warnings(warn_from_space);
2048         /* Pass back references to put */
2049         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2050                 if (is_valid[cnt])
2051                         transfer_to[cnt] = transfer_from[cnt];
2052         return 0;
2053 over_quota:
2054         /* Back out changes we already did */
2055         for (cnt--; cnt >= 0; cnt--) {
2056                 if (!is_valid[cnt])
2057                         continue;
2058                 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2059                 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2060                 dquot_decr_space(transfer_to[cnt], cur_space);
2061                 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2062                 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2063         }
2064         spin_unlock(&inode->i_lock);
2065         spin_unlock(&dq_data_lock);
2066         flush_warnings(warn_to);
2067         return ret;
2068 }
2069 EXPORT_SYMBOL(__dquot_transfer);
2070
2071 /* Wrapper for transferring ownership of an inode for uid/gid only
2072  * Called from FSXXX_setattr()
2073  */
2074 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2075 {
2076         struct dquot *transfer_to[MAXQUOTAS] = {};
2077         struct dquot *dquot;
2078         struct super_block *sb = inode->i_sb;
2079         int ret;
2080
2081         if (!dquot_active(inode))
2082                 return 0;
2083
2084         if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2085                 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2086                 if (IS_ERR(dquot)) {
2087                         if (PTR_ERR(dquot) != -ESRCH) {
2088                                 ret = PTR_ERR(dquot);
2089                                 goto out_put;
2090                         }
2091                         dquot = NULL;
2092                 }
2093                 transfer_to[USRQUOTA] = dquot;
2094         }
2095         if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2096                 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2097                 if (IS_ERR(dquot)) {
2098                         if (PTR_ERR(dquot) != -ESRCH) {
2099                                 ret = PTR_ERR(dquot);
2100                                 goto out_put;
2101                         }
2102                         dquot = NULL;
2103                 }
2104                 transfer_to[GRPQUOTA] = dquot;
2105         }
2106         ret = __dquot_transfer(inode, transfer_to);
2107 out_put:
2108         dqput_all(transfer_to);
2109         return ret;
2110 }
2111 EXPORT_SYMBOL(dquot_transfer);
2112
2113 /*
2114  * Write info of quota file to disk
2115  */
2116 int dquot_commit_info(struct super_block *sb, int type)
2117 {
2118         struct quota_info *dqopt = sb_dqopt(sb);
2119
2120         return dqopt->ops[type]->write_file_info(sb, type);
2121 }
2122 EXPORT_SYMBOL(dquot_commit_info);
2123
2124 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2125 {
2126         struct quota_info *dqopt = sb_dqopt(sb);
2127
2128         if (!sb_has_quota_active(sb, qid->type))
2129                 return -ESRCH;
2130         if (!dqopt->ops[qid->type]->get_next_id)
2131                 return -ENOSYS;
2132         return dqopt->ops[qid->type]->get_next_id(sb, qid);
2133 }
2134 EXPORT_SYMBOL(dquot_get_next_id);
2135
2136 /*
2137  * Definitions of diskquota operations.
2138  */
2139 const struct dquot_operations dquot_operations = {
2140         .write_dquot    = dquot_commit,
2141         .acquire_dquot  = dquot_acquire,
2142         .release_dquot  = dquot_release,
2143         .mark_dirty     = dquot_mark_dquot_dirty,
2144         .write_info     = dquot_commit_info,
2145         .alloc_dquot    = dquot_alloc,
2146         .destroy_dquot  = dquot_destroy,
2147         .get_next_id    = dquot_get_next_id,
2148 };
2149 EXPORT_SYMBOL(dquot_operations);
2150
2151 /*
2152  * Generic helper for ->open on filesystems supporting disk quotas.
2153  */
2154 int dquot_file_open(struct inode *inode, struct file *file)
2155 {
2156         int error;
2157
2158         error = generic_file_open(inode, file);
2159         if (!error && (file->f_mode & FMODE_WRITE))
2160                 error = dquot_initialize(inode);
2161         return error;
2162 }
2163 EXPORT_SYMBOL(dquot_file_open);
2164
2165 /*
2166  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2167  */
2168 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2169 {
2170         int cnt, ret = 0;
2171         struct quota_info *dqopt = sb_dqopt(sb);
2172         struct inode *toputinode[MAXQUOTAS];
2173
2174         /* s_umount should be held in exclusive mode */
2175         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2176                 up_read(&sb->s_umount);
2177
2178         /* Cannot turn off usage accounting without turning off limits, or
2179          * suspend quotas and simultaneously turn quotas off. */
2180         if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2181             || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2182             DQUOT_USAGE_ENABLED)))
2183                 return -EINVAL;
2184
2185         /*
2186          * Skip everything if there's nothing to do. We have to do this because
2187          * sometimes we are called when fill_super() failed and calling
2188          * sync_fs() in such cases does no good.
2189          */
2190         if (!sb_any_quota_loaded(sb))
2191                 return 0;
2192
2193         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2194                 toputinode[cnt] = NULL;
2195                 if (type != -1 && cnt != type)
2196                         continue;
2197                 if (!sb_has_quota_loaded(sb, cnt))
2198                         continue;
2199
2200                 if (flags & DQUOT_SUSPENDED) {
2201                         spin_lock(&dq_state_lock);
2202                         dqopt->flags |=
2203                                 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2204                         spin_unlock(&dq_state_lock);
2205                 } else {
2206                         spin_lock(&dq_state_lock);
2207                         dqopt->flags &= ~dquot_state_flag(flags, cnt);
2208                         /* Turning off suspended quotas? */
2209                         if (!sb_has_quota_loaded(sb, cnt) &&
2210                             sb_has_quota_suspended(sb, cnt)) {
2211                                 dqopt->flags &= ~dquot_state_flag(
2212                                                         DQUOT_SUSPENDED, cnt);
2213                                 spin_unlock(&dq_state_lock);
2214                                 iput(dqopt->files[cnt]);
2215                                 dqopt->files[cnt] = NULL;
2216                                 continue;
2217                         }
2218                         spin_unlock(&dq_state_lock);
2219                 }
2220
2221                 /* We still have to keep quota loaded? */
2222                 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2223                         continue;
2224
2225                 /* Note: these are blocking operations */
2226                 drop_dquot_ref(sb, cnt);
2227                 invalidate_dquots(sb, cnt);
2228                 /*
2229                  * Now all dquots should be invalidated, all writes done so we
2230                  * should be only users of the info. No locks needed.
2231                  */
2232                 if (info_dirty(&dqopt->info[cnt]))
2233                         sb->dq_op->write_info(sb, cnt);
2234                 if (dqopt->ops[cnt]->free_file_info)
2235                         dqopt->ops[cnt]->free_file_info(sb, cnt);
2236                 put_quota_format(dqopt->info[cnt].dqi_format);
2237
2238                 toputinode[cnt] = dqopt->files[cnt];
2239                 if (!sb_has_quota_loaded(sb, cnt))
2240                         dqopt->files[cnt] = NULL;
2241                 dqopt->info[cnt].dqi_flags = 0;
2242                 dqopt->info[cnt].dqi_igrace = 0;
2243                 dqopt->info[cnt].dqi_bgrace = 0;
2244                 dqopt->ops[cnt] = NULL;
2245         }
2246
2247         /* Skip syncing and setting flags if quota files are hidden */
2248         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2249                 goto put_inodes;
2250
2251         /* Sync the superblock so that buffers with quota data are written to
2252          * disk (and so userspace sees correct data afterwards). */
2253         if (sb->s_op->sync_fs)
2254                 sb->s_op->sync_fs(sb, 1);
2255         sync_blockdev(sb->s_bdev);
2256         /* Now the quota files are just ordinary files and we can set the
2257          * inode flags back. Moreover we discard the pagecache so that
2258          * userspace sees the writes we did bypassing the pagecache. We
2259          * must also discard the blockdev buffers so that we see the
2260          * changes done by userspace on the next quotaon() */
2261         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2262                 /* This can happen when suspending quotas on remount-ro... */
2263                 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2264                         inode_lock(toputinode[cnt]);
2265                         toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2266                         truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2267                         inode_unlock(toputinode[cnt]);
2268                         mark_inode_dirty_sync(toputinode[cnt]);
2269                 }
2270         if (sb->s_bdev)
2271                 invalidate_bdev(sb->s_bdev);
2272 put_inodes:
2273         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2274                 if (toputinode[cnt]) {
2275                         /* On remount RO, we keep the inode pointer so that we
2276                          * can reenable quota on the subsequent remount RW. We
2277                          * have to check 'flags' variable and not use sb_has_
2278                          * function because another quotaon / quotaoff could
2279                          * change global state before we got here. We refuse
2280                          * to suspend quotas when there is pending delete on
2281                          * the quota file... */
2282                         if (!(flags & DQUOT_SUSPENDED))
2283                                 iput(toputinode[cnt]);
2284                         else if (!toputinode[cnt]->i_nlink)
2285                                 ret = -EBUSY;
2286                 }
2287         return ret;
2288 }
2289 EXPORT_SYMBOL(dquot_disable);
2290
2291 int dquot_quota_off(struct super_block *sb, int type)
2292 {
2293         return dquot_disable(sb, type,
2294                              DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2295 }
2296 EXPORT_SYMBOL(dquot_quota_off);
2297
2298 /*
2299  *      Turn quotas on on a device
2300  */
2301
2302 /*
2303  * Helper function to turn quotas on when we already have the inode of
2304  * quota file and no quota information is loaded.
2305  */
2306 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2307         unsigned int flags)
2308 {
2309         struct quota_format_type *fmt = find_quota_format(format_id);
2310         struct super_block *sb = inode->i_sb;
2311         struct quota_info *dqopt = sb_dqopt(sb);
2312         int error;
2313
2314         if (!fmt)
2315                 return -ESRCH;
2316         if (!S_ISREG(inode->i_mode)) {
2317                 error = -EACCES;
2318                 goto out_fmt;
2319         }
2320         if (IS_RDONLY(inode)) {
2321                 error = -EROFS;
2322                 goto out_fmt;
2323         }
2324         if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2325             (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2326                 error = -EINVAL;
2327                 goto out_fmt;
2328         }
2329         /* Filesystems outside of init_user_ns not yet supported */
2330         if (sb->s_user_ns != &init_user_ns) {
2331                 error = -EINVAL;
2332                 goto out_fmt;
2333         }
2334         /* Usage always has to be set... */
2335         if (!(flags & DQUOT_USAGE_ENABLED)) {
2336                 error = -EINVAL;
2337                 goto out_fmt;
2338         }
2339         if (sb_has_quota_loaded(sb, type)) {
2340                 error = -EBUSY;
2341                 goto out_fmt;
2342         }
2343
2344         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2345                 /* As we bypass the pagecache we must now flush all the
2346                  * dirty data and invalidate caches so that kernel sees
2347                  * changes from userspace. It is not enough to just flush
2348                  * the quota file since if blocksize < pagesize, invalidation
2349                  * of the cache could fail because of other unrelated dirty
2350                  * data */
2351                 sync_filesystem(sb);
2352                 invalidate_bdev(sb->s_bdev);
2353         }
2354
2355         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2356                 /* We don't want quota and atime on quota files (deadlocks
2357                  * possible) Also nobody should write to the file - we use
2358                  * special IO operations which ignore the immutable bit. */
2359                 inode_lock(inode);
2360                 inode->i_flags |= S_NOQUOTA;
2361                 inode_unlock(inode);
2362                 /*
2363                  * When S_NOQUOTA is set, remove dquot references as no more
2364                  * references can be added
2365                  */
2366                 __dquot_drop(inode);
2367         }
2368
2369         error = -EIO;
2370         dqopt->files[type] = igrab(inode);
2371         if (!dqopt->files[type])
2372                 goto out_file_flags;
2373         error = -EINVAL;
2374         if (!fmt->qf_ops->check_quota_file(sb, type))
2375                 goto out_file_init;
2376
2377         dqopt->ops[type] = fmt->qf_ops;
2378         dqopt->info[type].dqi_format = fmt;
2379         dqopt->info[type].dqi_fmt_id = format_id;
2380         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2381         error = dqopt->ops[type]->read_file_info(sb, type);
2382         if (error < 0)
2383                 goto out_file_init;
2384         if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2385                 spin_lock(&dq_data_lock);
2386                 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2387                 spin_unlock(&dq_data_lock);
2388         }
2389         spin_lock(&dq_state_lock);
2390         dqopt->flags |= dquot_state_flag(flags, type);
2391         spin_unlock(&dq_state_lock);
2392
2393         error = add_dquot_ref(sb, type);
2394         if (error)
2395                 dquot_disable(sb, type, flags);
2396
2397         return error;
2398 out_file_init:
2399         dqopt->files[type] = NULL;
2400         iput(inode);
2401 out_file_flags:
2402         inode_lock(inode);
2403         inode->i_flags &= ~S_NOQUOTA;
2404         inode_unlock(inode);
2405 out_fmt:
2406         put_quota_format(fmt);
2407
2408         return error;
2409 }
2410
2411 /* Reenable quotas on remount RW */
2412 int dquot_resume(struct super_block *sb, int type)
2413 {
2414         struct quota_info *dqopt = sb_dqopt(sb);
2415         struct inode *inode;
2416         int ret = 0, cnt;
2417         unsigned int flags;
2418
2419         /* s_umount should be held in exclusive mode */
2420         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2421                 up_read(&sb->s_umount);
2422
2423         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2424                 if (type != -1 && cnt != type)
2425                         continue;
2426                 if (!sb_has_quota_suspended(sb, cnt))
2427                         continue;
2428
2429                 inode = dqopt->files[cnt];
2430                 dqopt->files[cnt] = NULL;
2431                 spin_lock(&dq_state_lock);
2432                 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2433                                                         DQUOT_LIMITS_ENABLED,
2434                                                         cnt);
2435                 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2436                 spin_unlock(&dq_state_lock);
2437
2438                 flags = dquot_generic_flag(flags, cnt);
2439                 ret = vfs_load_quota_inode(inode, cnt,
2440                                 dqopt->info[cnt].dqi_fmt_id, flags);
2441                 iput(inode);
2442         }
2443
2444         return ret;
2445 }
2446 EXPORT_SYMBOL(dquot_resume);
2447
2448 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2449                    const struct path *path)
2450 {
2451         int error = security_quota_on(path->dentry);
2452         if (error)
2453                 return error;
2454         /* Quota file not on the same filesystem? */
2455         if (path->dentry->d_sb != sb)
2456                 error = -EXDEV;
2457         else
2458                 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2459                                              format_id, DQUOT_USAGE_ENABLED |
2460                                              DQUOT_LIMITS_ENABLED);
2461         return error;
2462 }
2463 EXPORT_SYMBOL(dquot_quota_on);
2464
2465 /*
2466  * More powerful function for turning on quotas allowing setting
2467  * of individual quota flags
2468  */
2469 int dquot_enable(struct inode *inode, int type, int format_id,
2470                  unsigned int flags)
2471 {
2472         struct super_block *sb = inode->i_sb;
2473
2474         /* Just unsuspend quotas? */
2475         BUG_ON(flags & DQUOT_SUSPENDED);
2476         /* s_umount should be held in exclusive mode */
2477         if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2478                 up_read(&sb->s_umount);
2479
2480         if (!flags)
2481                 return 0;
2482         /* Just updating flags needed? */
2483         if (sb_has_quota_loaded(sb, type)) {
2484                 if (flags & DQUOT_USAGE_ENABLED &&
2485                     sb_has_quota_usage_enabled(sb, type))
2486                         return -EBUSY;
2487                 if (flags & DQUOT_LIMITS_ENABLED &&
2488                     sb_has_quota_limits_enabled(sb, type))
2489                         return -EBUSY;
2490                 spin_lock(&dq_state_lock);
2491                 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2492                 spin_unlock(&dq_state_lock);
2493                 return 0;
2494         }
2495
2496         return vfs_load_quota_inode(inode, type, format_id, flags);
2497 }
2498 EXPORT_SYMBOL(dquot_enable);
2499
2500 /*
2501  * This function is used when filesystem needs to initialize quotas
2502  * during mount time.
2503  */
2504 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2505                 int format_id, int type)
2506 {
2507         struct dentry *dentry;
2508         int error;
2509
2510         dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2511         if (IS_ERR(dentry))
2512                 return PTR_ERR(dentry);
2513
2514         if (d_really_is_negative(dentry)) {
2515                 error = -ENOENT;
2516                 goto out;
2517         }
2518
2519         error = security_quota_on(dentry);
2520         if (!error)
2521                 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2522                                 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2523
2524 out:
2525         dput(dentry);
2526         return error;
2527 }
2528 EXPORT_SYMBOL(dquot_quota_on_mount);
2529
2530 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2531 {
2532         int ret;
2533         int type;
2534         struct quota_info *dqopt = sb_dqopt(sb);
2535
2536         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2537                 return -ENOSYS;
2538         /* Accounting cannot be turned on while fs is mounted */
2539         flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2540         if (!flags)
2541                 return -EINVAL;
2542         for (type = 0; type < MAXQUOTAS; type++) {
2543                 if (!(flags & qtype_enforce_flag(type)))
2544                         continue;
2545                 /* Can't enforce without accounting */
2546                 if (!sb_has_quota_usage_enabled(sb, type))
2547                         return -EINVAL;
2548                 ret = dquot_enable(dqopt->files[type], type,
2549                                    dqopt->info[type].dqi_fmt_id,
2550                                    DQUOT_LIMITS_ENABLED);
2551                 if (ret < 0)
2552                         goto out_err;
2553         }
2554         return 0;
2555 out_err:
2556         /* Backout enforcement enablement we already did */
2557         for (type--; type >= 0; type--)  {
2558                 if (flags & qtype_enforce_flag(type))
2559                         dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2560         }
2561         /* Error code translation for better compatibility with XFS */
2562         if (ret == -EBUSY)
2563                 ret = -EEXIST;
2564         return ret;
2565 }
2566
2567 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2568 {
2569         int ret;
2570         int type;
2571         struct quota_info *dqopt = sb_dqopt(sb);
2572
2573         if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2574                 return -ENOSYS;
2575         /*
2576          * We don't support turning off accounting via quotactl. In principle
2577          * quota infrastructure can do this but filesystems don't expect
2578          * userspace to be able to do it.
2579          */
2580         if (flags &
2581                   (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2582                 return -EOPNOTSUPP;
2583
2584         /* Filter out limits not enabled */
2585         for (type = 0; type < MAXQUOTAS; type++)
2586                 if (!sb_has_quota_limits_enabled(sb, type))
2587                         flags &= ~qtype_enforce_flag(type);
2588         /* Nothing left? */
2589         if (!flags)
2590                 return -EEXIST;
2591         for (type = 0; type < MAXQUOTAS; type++) {
2592                 if (flags & qtype_enforce_flag(type)) {
2593                         ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2594                         if (ret < 0)
2595                                 goto out_err;
2596                 }
2597         }
2598         return 0;
2599 out_err:
2600         /* Backout enforcement disabling we already did */
2601         for (type--; type >= 0; type--)  {
2602                 if (flags & qtype_enforce_flag(type))
2603                         dquot_enable(dqopt->files[type], type,
2604                                      dqopt->info[type].dqi_fmt_id,
2605                                      DQUOT_LIMITS_ENABLED);
2606         }
2607         return ret;
2608 }
2609
2610 /* Generic routine for getting common part of quota structure */
2611 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2612 {
2613         struct mem_dqblk *dm = &dquot->dq_dqb;
2614
2615         memset(di, 0, sizeof(*di));
2616         spin_lock(&dquot->dq_dqb_lock);
2617         di->d_spc_hardlimit = dm->dqb_bhardlimit;
2618         di->d_spc_softlimit = dm->dqb_bsoftlimit;
2619         di->d_ino_hardlimit = dm->dqb_ihardlimit;
2620         di->d_ino_softlimit = dm->dqb_isoftlimit;
2621         di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2622         di->d_ino_count = dm->dqb_curinodes;
2623         di->d_spc_timer = dm->dqb_btime;
2624         di->d_ino_timer = dm->dqb_itime;
2625         spin_unlock(&dquot->dq_dqb_lock);
2626 }
2627
2628 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2629                     struct qc_dqblk *di)
2630 {
2631         struct dquot *dquot;
2632
2633         dquot = dqget(sb, qid);
2634         if (IS_ERR(dquot))
2635                 return PTR_ERR(dquot);
2636         do_get_dqblk(dquot, di);
2637         dqput(dquot);
2638
2639         return 0;
2640 }
2641 EXPORT_SYMBOL(dquot_get_dqblk);
2642
2643 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2644                          struct qc_dqblk *di)
2645 {
2646         struct dquot *dquot;
2647         int err;
2648
2649         if (!sb->dq_op->get_next_id)
2650                 return -ENOSYS;
2651         err = sb->dq_op->get_next_id(sb, qid);
2652         if (err < 0)
2653                 return err;
2654         dquot = dqget(sb, *qid);
2655         if (IS_ERR(dquot))
2656                 return PTR_ERR(dquot);
2657         do_get_dqblk(dquot, di);
2658         dqput(dquot);
2659
2660         return 0;
2661 }
2662 EXPORT_SYMBOL(dquot_get_next_dqblk);
2663
2664 #define VFS_QC_MASK \
2665         (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2666          QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2667          QC_SPC_TIMER | QC_INO_TIMER)
2668
2669 /* Generic routine for setting common part of quota structure */
2670 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2671 {
2672         struct mem_dqblk *dm = &dquot->dq_dqb;
2673         int check_blim = 0, check_ilim = 0;
2674         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2675
2676         if (di->d_fieldmask & ~VFS_QC_MASK)
2677                 return -EINVAL;
2678
2679         if (((di->d_fieldmask & QC_SPC_SOFT) &&
2680              di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2681             ((di->d_fieldmask & QC_SPC_HARD) &&
2682              di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2683             ((di->d_fieldmask & QC_INO_SOFT) &&
2684              (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2685             ((di->d_fieldmask & QC_INO_HARD) &&
2686              (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2687                 return -ERANGE;
2688
2689         spin_lock(&dquot->dq_dqb_lock);
2690         if (di->d_fieldmask & QC_SPACE) {
2691                 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2692                 check_blim = 1;
2693                 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2694         }
2695
2696         if (di->d_fieldmask & QC_SPC_SOFT)
2697                 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2698         if (di->d_fieldmask & QC_SPC_HARD)
2699                 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2700         if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2701                 check_blim = 1;
2702                 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2703         }
2704
2705         if (di->d_fieldmask & QC_INO_COUNT) {
2706                 dm->dqb_curinodes = di->d_ino_count;
2707                 check_ilim = 1;
2708                 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2709         }
2710
2711         if (di->d_fieldmask & QC_INO_SOFT)
2712                 dm->dqb_isoftlimit = di->d_ino_softlimit;
2713         if (di->d_fieldmask & QC_INO_HARD)
2714                 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2715         if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2716                 check_ilim = 1;
2717                 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2718         }
2719
2720         if (di->d_fieldmask & QC_SPC_TIMER) {
2721                 dm->dqb_btime = di->d_spc_timer;
2722                 check_blim = 1;
2723                 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2724         }
2725
2726         if (di->d_fieldmask & QC_INO_TIMER) {
2727                 dm->dqb_itime = di->d_ino_timer;
2728                 check_ilim = 1;
2729                 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2730         }
2731
2732         if (check_blim) {
2733                 if (!dm->dqb_bsoftlimit ||
2734                     dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2735                         dm->dqb_btime = 0;
2736                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2737                 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2738                         /* Set grace only if user hasn't provided his own... */
2739                         dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2740         }
2741         if (check_ilim) {
2742                 if (!dm->dqb_isoftlimit ||
2743                     dm->dqb_curinodes < dm->dqb_isoftlimit) {
2744                         dm->dqb_itime = 0;
2745                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
2746                 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2747                         /* Set grace only if user hasn't provided his own... */
2748                         dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2749         }
2750         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2751             dm->dqb_isoftlimit)
2752                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2753         else
2754                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2755         spin_unlock(&dquot->dq_dqb_lock);
2756         mark_dquot_dirty(dquot);
2757
2758         return 0;
2759 }
2760
2761 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2762                   struct qc_dqblk *di)
2763 {
2764         struct dquot *dquot;
2765         int rc;
2766
2767         dquot = dqget(sb, qid);
2768         if (IS_ERR(dquot)) {
2769                 rc = PTR_ERR(dquot);
2770                 goto out;
2771         }
2772         rc = do_set_dqblk(dquot, di);
2773         dqput(dquot);
2774 out:
2775         return rc;
2776 }
2777 EXPORT_SYMBOL(dquot_set_dqblk);
2778
2779 /* Generic routine for getting common part of quota file information */
2780 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2781 {
2782         struct mem_dqinfo *mi;
2783         struct qc_type_state *tstate;
2784         struct quota_info *dqopt = sb_dqopt(sb);
2785         int type;
2786
2787         memset(state, 0, sizeof(*state));
2788         for (type = 0; type < MAXQUOTAS; type++) {
2789                 if (!sb_has_quota_active(sb, type))
2790                         continue;
2791                 tstate = state->s_state + type;
2792                 mi = sb_dqopt(sb)->info + type;
2793                 tstate->flags = QCI_ACCT_ENABLED;
2794                 spin_lock(&dq_data_lock);
2795                 if (mi->dqi_flags & DQF_SYS_FILE)
2796                         tstate->flags |= QCI_SYSFILE;
2797                 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2798                         tstate->flags |= QCI_ROOT_SQUASH;
2799                 if (sb_has_quota_limits_enabled(sb, type))
2800                         tstate->flags |= QCI_LIMITS_ENFORCED;
2801                 tstate->spc_timelimit = mi->dqi_bgrace;
2802                 tstate->ino_timelimit = mi->dqi_igrace;
2803                 tstate->ino = dqopt->files[type]->i_ino;
2804                 tstate->blocks = dqopt->files[type]->i_blocks;
2805                 tstate->nextents = 1;   /* We don't know... */
2806                 spin_unlock(&dq_data_lock);
2807         }
2808         return 0;
2809 }
2810 EXPORT_SYMBOL(dquot_get_state);
2811
2812 /* Generic routine for setting common part of quota file information */
2813 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2814 {
2815         struct mem_dqinfo *mi;
2816         int err = 0;
2817
2818         if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2819             (ii->i_fieldmask & QC_RT_SPC_TIMER))
2820                 return -EINVAL;
2821         if (!sb_has_quota_active(sb, type))
2822                 return -ESRCH;
2823         mi = sb_dqopt(sb)->info + type;
2824         if (ii->i_fieldmask & QC_FLAGS) {
2825                 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2826                      mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2827                         return -EINVAL;
2828         }
2829         spin_lock(&dq_data_lock);
2830         if (ii->i_fieldmask & QC_SPC_TIMER)
2831                 mi->dqi_bgrace = ii->i_spc_timelimit;
2832         if (ii->i_fieldmask & QC_INO_TIMER)
2833                 mi->dqi_igrace = ii->i_ino_timelimit;
2834         if (ii->i_fieldmask & QC_FLAGS) {
2835                 if (ii->i_flags & QCI_ROOT_SQUASH)
2836                         mi->dqi_flags |= DQF_ROOT_SQUASH;
2837                 else
2838                         mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2839         }
2840         spin_unlock(&dq_data_lock);
2841         mark_info_dirty(sb, type);
2842         /* Force write to disk */
2843         sb->dq_op->write_info(sb, type);
2844         return err;
2845 }
2846 EXPORT_SYMBOL(dquot_set_dqinfo);
2847
2848 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2849         .quota_enable   = dquot_quota_enable,
2850         .quota_disable  = dquot_quota_disable,
2851         .quota_sync     = dquot_quota_sync,
2852         .get_state      = dquot_get_state,
2853         .set_info       = dquot_set_dqinfo,
2854         .get_dqblk      = dquot_get_dqblk,
2855         .get_nextdqblk  = dquot_get_next_dqblk,
2856         .set_dqblk      = dquot_set_dqblk
2857 };
2858 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2859
2860 static int do_proc_dqstats(struct ctl_table *table, int write,
2861                      void __user *buffer, size_t *lenp, loff_t *ppos)
2862 {
2863         unsigned int type = (int *)table->data - dqstats.stat;
2864
2865         /* Update global table */
2866         dqstats.stat[type] =
2867                         percpu_counter_sum_positive(&dqstats.counter[type]);
2868         return proc_dointvec(table, write, buffer, lenp, ppos);
2869 }
2870
2871 static struct ctl_table fs_dqstats_table[] = {
2872         {
2873                 .procname       = "lookups",
2874                 .data           = &dqstats.stat[DQST_LOOKUPS],
2875                 .maxlen         = sizeof(int),
2876                 .mode           = 0444,
2877                 .proc_handler   = do_proc_dqstats,
2878         },
2879         {
2880                 .procname       = "drops",
2881                 .data           = &dqstats.stat[DQST_DROPS],
2882                 .maxlen         = sizeof(int),
2883                 .mode           = 0444,
2884                 .proc_handler   = do_proc_dqstats,
2885         },
2886         {
2887                 .procname       = "reads",
2888                 .data           = &dqstats.stat[DQST_READS],
2889                 .maxlen         = sizeof(int),
2890                 .mode           = 0444,
2891                 .proc_handler   = do_proc_dqstats,
2892         },
2893         {
2894                 .procname       = "writes",
2895                 .data           = &dqstats.stat[DQST_WRITES],
2896                 .maxlen         = sizeof(int),
2897                 .mode           = 0444,
2898                 .proc_handler   = do_proc_dqstats,
2899         },
2900         {
2901                 .procname       = "cache_hits",
2902                 .data           = &dqstats.stat[DQST_CACHE_HITS],
2903                 .maxlen         = sizeof(int),
2904                 .mode           = 0444,
2905                 .proc_handler   = do_proc_dqstats,
2906         },
2907         {
2908                 .procname       = "allocated_dquots",
2909                 .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2910                 .maxlen         = sizeof(int),
2911                 .mode           = 0444,
2912                 .proc_handler   = do_proc_dqstats,
2913         },
2914         {
2915                 .procname       = "free_dquots",
2916                 .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2917                 .maxlen         = sizeof(int),
2918                 .mode           = 0444,
2919                 .proc_handler   = do_proc_dqstats,
2920         },
2921         {
2922                 .procname       = "syncs",
2923                 .data           = &dqstats.stat[DQST_SYNCS],
2924                 .maxlen         = sizeof(int),
2925                 .mode           = 0444,
2926                 .proc_handler   = do_proc_dqstats,
2927         },
2928 #ifdef CONFIG_PRINT_QUOTA_WARNING
2929         {
2930                 .procname       = "warnings",
2931                 .data           = &flag_print_warnings,
2932                 .maxlen         = sizeof(int),
2933                 .mode           = 0644,
2934                 .proc_handler   = proc_dointvec,
2935         },
2936 #endif
2937         { },
2938 };
2939
2940 static struct ctl_table fs_table[] = {
2941         {
2942                 .procname       = "quota",
2943                 .mode           = 0555,
2944                 .child          = fs_dqstats_table,
2945         },
2946         { },
2947 };
2948
2949 static struct ctl_table sys_table[] = {
2950         {
2951                 .procname       = "fs",
2952                 .mode           = 0555,
2953                 .child          = fs_table,
2954         },
2955         { },
2956 };
2957
2958 static int __init dquot_init(void)
2959 {
2960         int i, ret;
2961         unsigned long nr_hash, order;
2962
2963         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2964
2965         register_sysctl_table(sys_table);
2966
2967         dquot_cachep = kmem_cache_create("dquot",
2968                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2969                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2970                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2971                         NULL);
2972
2973         order = 0;
2974         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2975         if (!dquot_hash)
2976                 panic("Cannot create dquot hash table");
2977
2978         for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2979                 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2980                 if (ret)
2981                         panic("Cannot create dquot stat counters");
2982         }
2983
2984         /* Find power-of-two hlist_heads which can fit into allocation */
2985         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2986         dq_hash_bits = 0;
2987         do {
2988                 dq_hash_bits++;
2989         } while (nr_hash >> dq_hash_bits);
2990         dq_hash_bits--;
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);