]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/quota/dquot.c
62c7502022d6fa1e8e18c776744514dd3df574d5
[mirror_ubuntu-bionic-kernel.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 three lists for dquot management: the inuse_list,
227 * free_dquots, and dquot_hash[] array. A single dquot structure may be
228 * on all three 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 * Dquots with a specific identity (device, type and id) are placed on
240 * one of the dquot_hash[] hash chains. The provides an efficient search
241 * mechanism to locate a specific dquot.
242 */
243
244 static LIST_HEAD(inuse_list);
245 static LIST_HEAD(free_dquots);
246 static unsigned int dq_hash_bits, dq_hash_mask;
247 static struct hlist_head *dquot_hash;
248
249 struct dqstats dqstats;
250 EXPORT_SYMBOL(dqstats);
251
252 static qsize_t inode_get_rsv_space(struct inode *inode);
253 static qsize_t __inode_get_rsv_space(struct inode *inode);
254 static int __dquot_initialize(struct inode *inode, int type);
255
256 static inline unsigned int
257 hashfn(const struct super_block *sb, struct kqid qid)
258 {
259 unsigned int id = from_kqid(&init_user_ns, qid);
260 int type = qid.type;
261 unsigned long tmp;
262
263 tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
264 return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
265 }
266
267 /*
268 * Following list functions expect dq_list_lock to be held
269 */
270 static inline void insert_dquot_hash(struct dquot *dquot)
271 {
272 struct hlist_head *head;
273 head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
274 hlist_add_head(&dquot->dq_hash, head);
275 }
276
277 static inline void remove_dquot_hash(struct dquot *dquot)
278 {
279 hlist_del_init(&dquot->dq_hash);
280 }
281
282 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
283 struct kqid qid)
284 {
285 struct hlist_node *node;
286 struct dquot *dquot;
287
288 hlist_for_each (node, dquot_hash+hashent) {
289 dquot = hlist_entry(node, struct dquot, dq_hash);
290 if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
291 return dquot;
292 }
293 return NULL;
294 }
295
296 /* Add a dquot to the tail of the free list */
297 static inline void put_dquot_last(struct dquot *dquot)
298 {
299 list_add_tail(&dquot->dq_free, &free_dquots);
300 dqstats_inc(DQST_FREE_DQUOTS);
301 }
302
303 static inline void remove_free_dquot(struct dquot *dquot)
304 {
305 if (list_empty(&dquot->dq_free))
306 return;
307 list_del_init(&dquot->dq_free);
308 dqstats_dec(DQST_FREE_DQUOTS);
309 }
310
311 static inline void put_inuse(struct dquot *dquot)
312 {
313 /* We add to the back of inuse list so we don't have to restart
314 * when traversing this list and we block */
315 list_add_tail(&dquot->dq_inuse, &inuse_list);
316 dqstats_inc(DQST_ALLOC_DQUOTS);
317 }
318
319 static inline void remove_inuse(struct dquot *dquot)
320 {
321 dqstats_dec(DQST_ALLOC_DQUOTS);
322 list_del(&dquot->dq_inuse);
323 }
324 /*
325 * End of list functions needing dq_list_lock
326 */
327
328 static void wait_on_dquot(struct dquot *dquot)
329 {
330 mutex_lock(&dquot->dq_lock);
331 mutex_unlock(&dquot->dq_lock);
332 }
333
334 static inline int dquot_dirty(struct dquot *dquot)
335 {
336 return test_bit(DQ_MOD_B, &dquot->dq_flags);
337 }
338
339 static inline int mark_dquot_dirty(struct dquot *dquot)
340 {
341 return dquot->dq_sb->dq_op->mark_dirty(dquot);
342 }
343
344 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
345 int dquot_mark_dquot_dirty(struct dquot *dquot)
346 {
347 int ret = 1;
348
349 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
350 return 0;
351
352 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
353 return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
354
355 /* If quota is dirty already, we don't have to acquire dq_list_lock */
356 if (test_bit(DQ_MOD_B, &dquot->dq_flags))
357 return 1;
358
359 spin_lock(&dq_list_lock);
360 if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
361 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
362 info[dquot->dq_id.type].dqi_dirty_list);
363 ret = 0;
364 }
365 spin_unlock(&dq_list_lock);
366 return ret;
367 }
368 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
369
370 /* Dirtify all the dquots - this can block when journalling */
371 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
372 {
373 int ret, err, cnt;
374
375 ret = err = 0;
376 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
377 if (dquot[cnt])
378 /* Even in case of error we have to continue */
379 ret = mark_dquot_dirty(dquot[cnt]);
380 if (!err)
381 err = ret;
382 }
383 return err;
384 }
385
386 static inline void dqput_all(struct dquot **dquot)
387 {
388 unsigned int cnt;
389
390 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
391 dqput(dquot[cnt]);
392 }
393
394 static inline int clear_dquot_dirty(struct dquot *dquot)
395 {
396 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
397 return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
398
399 spin_lock(&dq_list_lock);
400 if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
401 spin_unlock(&dq_list_lock);
402 return 0;
403 }
404 list_del_init(&dquot->dq_dirty);
405 spin_unlock(&dq_list_lock);
406 return 1;
407 }
408
409 void mark_info_dirty(struct super_block *sb, int type)
410 {
411 spin_lock(&dq_data_lock);
412 sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
413 spin_unlock(&dq_data_lock);
414 }
415 EXPORT_SYMBOL(mark_info_dirty);
416
417 /*
418 * Read dquot from disk and alloc space for it
419 */
420
421 int dquot_acquire(struct dquot *dquot)
422 {
423 int ret = 0, ret2 = 0;
424 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
425
426 mutex_lock(&dquot->dq_lock);
427 if (!test_bit(DQ_READ_B, &dquot->dq_flags))
428 ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
429 if (ret < 0)
430 goto out_iolock;
431 /* Make sure flags update is visible after dquot has been filled */
432 smp_mb__before_atomic();
433 set_bit(DQ_READ_B, &dquot->dq_flags);
434 /* Instantiate dquot if needed */
435 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
436 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
437 /* Write the info if needed */
438 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
439 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
440 dquot->dq_sb, dquot->dq_id.type);
441 }
442 if (ret < 0)
443 goto out_iolock;
444 if (ret2 < 0) {
445 ret = ret2;
446 goto out_iolock;
447 }
448 }
449 /*
450 * Make sure flags update is visible after on-disk struct has been
451 * allocated. Paired with smp_rmb() in dqget().
452 */
453 smp_mb__before_atomic();
454 set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
455 out_iolock:
456 mutex_unlock(&dquot->dq_lock);
457 return ret;
458 }
459 EXPORT_SYMBOL(dquot_acquire);
460
461 /*
462 * Write dquot to disk
463 */
464 int dquot_commit(struct dquot *dquot)
465 {
466 int ret = 0;
467 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
468
469 mutex_lock(&dquot->dq_lock);
470 if (!clear_dquot_dirty(dquot))
471 goto out_lock;
472 /* Inactive dquot can be only if there was error during read/init
473 * => we have better not writing it */
474 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
475 ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
476 else
477 ret = -EIO;
478 out_lock:
479 mutex_unlock(&dquot->dq_lock);
480 return ret;
481 }
482 EXPORT_SYMBOL(dquot_commit);
483
484 /*
485 * Release dquot
486 */
487 int dquot_release(struct dquot *dquot)
488 {
489 int ret = 0, ret2 = 0;
490 struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
491
492 mutex_lock(&dquot->dq_lock);
493 /* Check whether we are not racing with some other dqget() */
494 if (dquot_is_busy(dquot))
495 goto out_dqlock;
496 if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
497 ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
498 /* Write the info */
499 if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
500 ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
501 dquot->dq_sb, dquot->dq_id.type);
502 }
503 if (ret >= 0)
504 ret = ret2;
505 }
506 clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
507 out_dqlock:
508 mutex_unlock(&dquot->dq_lock);
509 return ret;
510 }
511 EXPORT_SYMBOL(dquot_release);
512
513 void dquot_destroy(struct dquot *dquot)
514 {
515 kmem_cache_free(dquot_cachep, dquot);
516 }
517 EXPORT_SYMBOL(dquot_destroy);
518
519 static inline void do_destroy_dquot(struct dquot *dquot)
520 {
521 dquot->dq_sb->dq_op->destroy_dquot(dquot);
522 }
523
524 /* Invalidate all dquots on the list. Note that this function is called after
525 * quota is disabled and pointers from inodes removed so there cannot be new
526 * quota users. There can still be some users of quotas due to inodes being
527 * just deleted or pruned by prune_icache() (those are not attached to any
528 * list) or parallel quotactl call. We have to wait for such users.
529 */
530 static void invalidate_dquots(struct super_block *sb, int type)
531 {
532 struct dquot *dquot, *tmp;
533
534 restart:
535 spin_lock(&dq_list_lock);
536 list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
537 if (dquot->dq_sb != sb)
538 continue;
539 if (dquot->dq_id.type != type)
540 continue;
541 /* Wait for dquot users */
542 if (atomic_read(&dquot->dq_count)) {
543 dqgrab(dquot);
544 spin_unlock(&dq_list_lock);
545 /*
546 * Once dqput() wakes us up, we know it's time to free
547 * the dquot.
548 * IMPORTANT: we rely on the fact that there is always
549 * at most one process waiting for dquot to free.
550 * Otherwise dq_count would be > 1 and we would never
551 * wake up.
552 */
553 wait_event(dquot_ref_wq,
554 atomic_read(&dquot->dq_count) == 1);
555 dqput(dquot);
556 /* At this moment dquot() need not exist (it could be
557 * reclaimed by prune_dqcache(). Hence we must
558 * restart. */
559 goto restart;
560 }
561 /*
562 * Quota now has no users and it has been written on last
563 * dqput()
564 */
565 remove_dquot_hash(dquot);
566 remove_free_dquot(dquot);
567 remove_inuse(dquot);
568 do_destroy_dquot(dquot);
569 }
570 spin_unlock(&dq_list_lock);
571 }
572
573 /* Call callback for every active dquot on given filesystem */
574 int dquot_scan_active(struct super_block *sb,
575 int (*fn)(struct dquot *dquot, unsigned long priv),
576 unsigned long priv)
577 {
578 struct dquot *dquot, *old_dquot = NULL;
579 int ret = 0;
580
581 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
582
583 spin_lock(&dq_list_lock);
584 list_for_each_entry(dquot, &inuse_list, dq_inuse) {
585 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
586 continue;
587 if (dquot->dq_sb != sb)
588 continue;
589 /* Now we have active dquot so we can just increase use count */
590 atomic_inc(&dquot->dq_count);
591 spin_unlock(&dq_list_lock);
592 dqstats_inc(DQST_LOOKUPS);
593 dqput(old_dquot);
594 old_dquot = dquot;
595 /*
596 * ->release_dquot() can be racing with us. Our reference
597 * protects us from new calls to it so just wait for any
598 * outstanding call and recheck the DQ_ACTIVE_B after that.
599 */
600 wait_on_dquot(dquot);
601 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
602 ret = fn(dquot, priv);
603 if (ret < 0)
604 goto out;
605 }
606 spin_lock(&dq_list_lock);
607 /* We are safe to continue now because our dquot could not
608 * be moved out of the inuse list while we hold the reference */
609 }
610 spin_unlock(&dq_list_lock);
611 out:
612 dqput(old_dquot);
613 return ret;
614 }
615 EXPORT_SYMBOL(dquot_scan_active);
616
617 /* Write all dquot structures to quota files */
618 int dquot_writeback_dquots(struct super_block *sb, int type)
619 {
620 struct list_head dirty;
621 struct dquot *dquot;
622 struct quota_info *dqopt = sb_dqopt(sb);
623 int cnt;
624 int err, ret = 0;
625
626 WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
627
628 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
629 if (type != -1 && cnt != type)
630 continue;
631 if (!sb_has_quota_active(sb, cnt))
632 continue;
633 spin_lock(&dq_list_lock);
634 /* Move list away to avoid livelock. */
635 list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
636 while (!list_empty(&dirty)) {
637 dquot = list_first_entry(&dirty, struct dquot,
638 dq_dirty);
639
640 WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
641
642 /* Now we have active dquot from which someone is
643 * holding reference so we can safely just increase
644 * use count */
645 dqgrab(dquot);
646 spin_unlock(&dq_list_lock);
647 dqstats_inc(DQST_LOOKUPS);
648 err = sb->dq_op->write_dquot(dquot);
649 if (err) {
650 /*
651 * Clear dirty bit anyway to avoid infinite
652 * loop here.
653 */
654 clear_dquot_dirty(dquot);
655 if (!ret)
656 ret = err;
657 }
658 dqput(dquot);
659 spin_lock(&dq_list_lock);
660 }
661 spin_unlock(&dq_list_lock);
662 }
663
664 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
665 if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
666 && info_dirty(&dqopt->info[cnt]))
667 sb->dq_op->write_info(sb, cnt);
668 dqstats_inc(DQST_SYNCS);
669
670 return ret;
671 }
672 EXPORT_SYMBOL(dquot_writeback_dquots);
673
674 /* Write all dquot structures to disk and make them visible from userspace */
675 int dquot_quota_sync(struct super_block *sb, int type)
676 {
677 struct quota_info *dqopt = sb_dqopt(sb);
678 int cnt;
679 int ret;
680
681 ret = dquot_writeback_dquots(sb, type);
682 if (ret)
683 return ret;
684 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
685 return 0;
686
687 /* This is not very clever (and fast) but currently I don't know about
688 * any other simple way of getting quota data to disk and we must get
689 * them there for userspace to be visible... */
690 if (sb->s_op->sync_fs)
691 sb->s_op->sync_fs(sb, 1);
692 sync_blockdev(sb->s_bdev);
693
694 /*
695 * Now when everything is written we can discard the pagecache so
696 * that userspace sees the changes.
697 */
698 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
699 if (type != -1 && cnt != type)
700 continue;
701 if (!sb_has_quota_active(sb, cnt))
702 continue;
703 inode_lock(dqopt->files[cnt]);
704 truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
705 inode_unlock(dqopt->files[cnt]);
706 }
707
708 return 0;
709 }
710 EXPORT_SYMBOL(dquot_quota_sync);
711
712 static unsigned long
713 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
714 {
715 struct list_head *head;
716 struct dquot *dquot;
717 unsigned long freed = 0;
718
719 spin_lock(&dq_list_lock);
720 head = free_dquots.prev;
721 while (head != &free_dquots && sc->nr_to_scan) {
722 dquot = list_entry(head, struct dquot, dq_free);
723 remove_dquot_hash(dquot);
724 remove_free_dquot(dquot);
725 remove_inuse(dquot);
726 do_destroy_dquot(dquot);
727 sc->nr_to_scan--;
728 freed++;
729 head = free_dquots.prev;
730 }
731 spin_unlock(&dq_list_lock);
732 return freed;
733 }
734
735 static unsigned long
736 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
737 {
738 return vfs_pressure_ratio(
739 percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
740 }
741
742 static struct shrinker dqcache_shrinker = {
743 .count_objects = dqcache_shrink_count,
744 .scan_objects = dqcache_shrink_scan,
745 .seeks = DEFAULT_SEEKS,
746 };
747
748 /*
749 * Put reference to dquot
750 */
751 void dqput(struct dquot *dquot)
752 {
753 int ret;
754
755 if (!dquot)
756 return;
757 #ifdef CONFIG_QUOTA_DEBUG
758 if (!atomic_read(&dquot->dq_count)) {
759 quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
760 quotatypes[dquot->dq_id.type],
761 from_kqid(&init_user_ns, dquot->dq_id));
762 BUG();
763 }
764 #endif
765 dqstats_inc(DQST_DROPS);
766 we_slept:
767 spin_lock(&dq_list_lock);
768 if (atomic_read(&dquot->dq_count) > 1) {
769 /* We have more than one user... nothing to do */
770 atomic_dec(&dquot->dq_count);
771 /* Releasing dquot during quotaoff phase? */
772 if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
773 atomic_read(&dquot->dq_count) == 1)
774 wake_up(&dquot_ref_wq);
775 spin_unlock(&dq_list_lock);
776 return;
777 }
778 /* Need to release dquot? */
779 if (dquot_dirty(dquot)) {
780 spin_unlock(&dq_list_lock);
781 /* Commit dquot before releasing */
782 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
783 if (ret < 0) {
784 quota_error(dquot->dq_sb, "Can't write quota structure"
785 " (error %d). Quota may get out of sync!",
786 ret);
787 /*
788 * We clear dirty bit anyway, so that we avoid
789 * infinite loop here
790 */
791 clear_dquot_dirty(dquot);
792 }
793 goto we_slept;
794 }
795 if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
796 spin_unlock(&dq_list_lock);
797 dquot->dq_sb->dq_op->release_dquot(dquot);
798 goto we_slept;
799 }
800 atomic_dec(&dquot->dq_count);
801 #ifdef CONFIG_QUOTA_DEBUG
802 /* sanity check */
803 BUG_ON(!list_empty(&dquot->dq_free));
804 #endif
805 put_dquot_last(dquot);
806 spin_unlock(&dq_list_lock);
807 }
808 EXPORT_SYMBOL(dqput);
809
810 struct dquot *dquot_alloc(struct super_block *sb, int type)
811 {
812 return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
813 }
814 EXPORT_SYMBOL(dquot_alloc);
815
816 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
817 {
818 struct dquot *dquot;
819
820 dquot = sb->dq_op->alloc_dquot(sb, type);
821 if(!dquot)
822 return NULL;
823
824 mutex_init(&dquot->dq_lock);
825 INIT_LIST_HEAD(&dquot->dq_free);
826 INIT_LIST_HEAD(&dquot->dq_inuse);
827 INIT_HLIST_NODE(&dquot->dq_hash);
828 INIT_LIST_HEAD(&dquot->dq_dirty);
829 dquot->dq_sb = sb;
830 dquot->dq_id = make_kqid_invalid(type);
831 atomic_set(&dquot->dq_count, 1);
832 spin_lock_init(&dquot->dq_dqb_lock);
833
834 return dquot;
835 }
836
837 /*
838 * Get reference to dquot
839 *
840 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
841 * destroying our dquot by:
842 * a) checking for quota flags under dq_list_lock and
843 * b) getting a reference to dquot before we release dq_list_lock
844 */
845 struct dquot *dqget(struct super_block *sb, struct kqid qid)
846 {
847 unsigned int hashent = hashfn(sb, qid);
848 struct dquot *dquot, *empty = NULL;
849
850 if (!qid_has_mapping(sb->s_user_ns, qid))
851 return ERR_PTR(-EINVAL);
852
853 if (!sb_has_quota_active(sb, qid.type))
854 return ERR_PTR(-ESRCH);
855 we_slept:
856 spin_lock(&dq_list_lock);
857 spin_lock(&dq_state_lock);
858 if (!sb_has_quota_active(sb, qid.type)) {
859 spin_unlock(&dq_state_lock);
860 spin_unlock(&dq_list_lock);
861 dquot = ERR_PTR(-ESRCH);
862 goto out;
863 }
864 spin_unlock(&dq_state_lock);
865
866 dquot = find_dquot(hashent, sb, qid);
867 if (!dquot) {
868 if (!empty) {
869 spin_unlock(&dq_list_lock);
870 empty = get_empty_dquot(sb, qid.type);
871 if (!empty)
872 schedule(); /* Try to wait for a moment... */
873 goto we_slept;
874 }
875 dquot = empty;
876 empty = NULL;
877 dquot->dq_id = qid;
878 /* all dquots go on the inuse_list */
879 put_inuse(dquot);
880 /* hash it first so it can be found */
881 insert_dquot_hash(dquot);
882 spin_unlock(&dq_list_lock);
883 dqstats_inc(DQST_LOOKUPS);
884 } else {
885 if (!atomic_read(&dquot->dq_count))
886 remove_free_dquot(dquot);
887 atomic_inc(&dquot->dq_count);
888 spin_unlock(&dq_list_lock);
889 dqstats_inc(DQST_CACHE_HITS);
890 dqstats_inc(DQST_LOOKUPS);
891 }
892 /* Wait for dq_lock - after this we know that either dquot_release() is
893 * already finished or it will be canceled due to dq_count > 1 test */
894 wait_on_dquot(dquot);
895 /* Read the dquot / allocate space in quota file */
896 if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
897 int err;
898
899 err = sb->dq_op->acquire_dquot(dquot);
900 if (err < 0) {
901 dqput(dquot);
902 dquot = ERR_PTR(err);
903 goto out;
904 }
905 }
906 /*
907 * Make sure following reads see filled structure - paired with
908 * smp_mb__before_atomic() in dquot_acquire().
909 */
910 smp_rmb();
911 #ifdef CONFIG_QUOTA_DEBUG
912 BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
913 #endif
914 out:
915 if (empty)
916 do_destroy_dquot(empty);
917
918 return dquot;
919 }
920 EXPORT_SYMBOL(dqget);
921
922 static inline struct dquot **i_dquot(struct inode *inode)
923 {
924 return inode->i_sb->s_op->get_dquots(inode);
925 }
926
927 static int dqinit_needed(struct inode *inode, int type)
928 {
929 struct dquot * const *dquots;
930 int cnt;
931
932 if (IS_NOQUOTA(inode))
933 return 0;
934
935 dquots = i_dquot(inode);
936 if (type != -1)
937 return !dquots[type];
938 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
939 if (!dquots[cnt])
940 return 1;
941 return 0;
942 }
943
944 /* This routine is guarded by s_umount semaphore */
945 static int add_dquot_ref(struct super_block *sb, int type)
946 {
947 struct inode *inode, *old_inode = NULL;
948 #ifdef CONFIG_QUOTA_DEBUG
949 int reserved = 0;
950 #endif
951 int err = 0;
952
953 spin_lock(&sb->s_inode_list_lock);
954 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
955 spin_lock(&inode->i_lock);
956 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
957 !atomic_read(&inode->i_writecount) ||
958 !dqinit_needed(inode, type)) {
959 spin_unlock(&inode->i_lock);
960 continue;
961 }
962 __iget(inode);
963 spin_unlock(&inode->i_lock);
964 spin_unlock(&sb->s_inode_list_lock);
965
966 #ifdef CONFIG_QUOTA_DEBUG
967 if (unlikely(inode_get_rsv_space(inode) > 0))
968 reserved = 1;
969 #endif
970 iput(old_inode);
971 err = __dquot_initialize(inode, type);
972 if (err) {
973 iput(inode);
974 goto out;
975 }
976
977 /*
978 * We hold a reference to 'inode' so it couldn't have been
979 * removed from s_inodes list while we dropped the
980 * s_inode_list_lock. We cannot iput the inode now as we can be
981 * holding the last reference and we cannot iput it under
982 * s_inode_list_lock. So we keep the reference and iput it
983 * later.
984 */
985 old_inode = inode;
986 spin_lock(&sb->s_inode_list_lock);
987 }
988 spin_unlock(&sb->s_inode_list_lock);
989 iput(old_inode);
990 out:
991 #ifdef CONFIG_QUOTA_DEBUG
992 if (reserved) {
993 quota_error(sb, "Writes happened before quota was turned on "
994 "thus quota information is probably inconsistent. "
995 "Please run quotacheck(8)");
996 }
997 #endif
998 return err;
999 }
1000
1001 /*
1002 * Remove references to dquots from inode and add dquot to list for freeing
1003 * if we have the last reference to dquot
1004 */
1005 static void remove_inode_dquot_ref(struct inode *inode, int type,
1006 struct list_head *tofree_head)
1007 {
1008 struct dquot **dquots = i_dquot(inode);
1009 struct dquot *dquot = dquots[type];
1010
1011 if (!dquot)
1012 return;
1013
1014 dquots[type] = NULL;
1015 if (list_empty(&dquot->dq_free)) {
1016 /*
1017 * The inode still has reference to dquot so it can't be in the
1018 * free list
1019 */
1020 spin_lock(&dq_list_lock);
1021 list_add(&dquot->dq_free, tofree_head);
1022 spin_unlock(&dq_list_lock);
1023 } else {
1024 /*
1025 * Dquot is already in a list to put so we won't drop the last
1026 * reference here.
1027 */
1028 dqput(dquot);
1029 }
1030 }
1031
1032 /*
1033 * Free list of dquots
1034 * Dquots are removed from inodes and no new references can be got so we are
1035 * the only ones holding reference
1036 */
1037 static void put_dquot_list(struct list_head *tofree_head)
1038 {
1039 struct list_head *act_head;
1040 struct dquot *dquot;
1041
1042 act_head = tofree_head->next;
1043 while (act_head != tofree_head) {
1044 dquot = list_entry(act_head, struct dquot, dq_free);
1045 act_head = act_head->next;
1046 /* Remove dquot from the list so we won't have problems... */
1047 list_del_init(&dquot->dq_free);
1048 dqput(dquot);
1049 }
1050 }
1051
1052 static void remove_dquot_ref(struct super_block *sb, int type,
1053 struct list_head *tofree_head)
1054 {
1055 struct inode *inode;
1056 int reserved = 0;
1057
1058 spin_lock(&sb->s_inode_list_lock);
1059 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1060 /*
1061 * We have to scan also I_NEW inodes because they can already
1062 * have quota pointer initialized. Luckily, we need to touch
1063 * only quota pointers and these have separate locking
1064 * (dq_data_lock).
1065 */
1066 spin_lock(&dq_data_lock);
1067 if (!IS_NOQUOTA(inode)) {
1068 if (unlikely(inode_get_rsv_space(inode) > 0))
1069 reserved = 1;
1070 remove_inode_dquot_ref(inode, type, tofree_head);
1071 }
1072 spin_unlock(&dq_data_lock);
1073 }
1074 spin_unlock(&sb->s_inode_list_lock);
1075 #ifdef CONFIG_QUOTA_DEBUG
1076 if (reserved) {
1077 printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1078 " was disabled thus quota information is probably "
1079 "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1080 }
1081 #endif
1082 }
1083
1084 /* Gather all references from inodes and drop them */
1085 static void drop_dquot_ref(struct super_block *sb, int type)
1086 {
1087 LIST_HEAD(tofree_head);
1088
1089 if (sb->dq_op) {
1090 remove_dquot_ref(sb, type, &tofree_head);
1091 synchronize_srcu(&dquot_srcu);
1092 put_dquot_list(&tofree_head);
1093 }
1094 }
1095
1096 static inline
1097 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1098 {
1099 if (dquot->dq_dqb.dqb_rsvspace >= number)
1100 dquot->dq_dqb.dqb_rsvspace -= number;
1101 else {
1102 WARN_ON_ONCE(1);
1103 dquot->dq_dqb.dqb_rsvspace = 0;
1104 }
1105 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1106 dquot->dq_dqb.dqb_bsoftlimit)
1107 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1108 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1109 }
1110
1111 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1112 {
1113 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1114 dquot->dq_dqb.dqb_curinodes >= number)
1115 dquot->dq_dqb.dqb_curinodes -= number;
1116 else
1117 dquot->dq_dqb.dqb_curinodes = 0;
1118 if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1119 dquot->dq_dqb.dqb_itime = (time64_t) 0;
1120 clear_bit(DQ_INODES_B, &dquot->dq_flags);
1121 }
1122
1123 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1124 {
1125 if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1126 dquot->dq_dqb.dqb_curspace >= number)
1127 dquot->dq_dqb.dqb_curspace -= number;
1128 else
1129 dquot->dq_dqb.dqb_curspace = 0;
1130 if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1131 dquot->dq_dqb.dqb_bsoftlimit)
1132 dquot->dq_dqb.dqb_btime = (time64_t) 0;
1133 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1134 }
1135
1136 struct dquot_warn {
1137 struct super_block *w_sb;
1138 struct kqid w_dq_id;
1139 short w_type;
1140 };
1141
1142 static int warning_issued(struct dquot *dquot, const int warntype)
1143 {
1144 int flag = (warntype == QUOTA_NL_BHARDWARN ||
1145 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1146 ((warntype == QUOTA_NL_IHARDWARN ||
1147 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1148
1149 if (!flag)
1150 return 0;
1151 return test_and_set_bit(flag, &dquot->dq_flags);
1152 }
1153
1154 #ifdef CONFIG_PRINT_QUOTA_WARNING
1155 static int flag_print_warnings = 1;
1156
1157 static int need_print_warning(struct dquot_warn *warn)
1158 {
1159 if (!flag_print_warnings)
1160 return 0;
1161
1162 switch (warn->w_dq_id.type) {
1163 case USRQUOTA:
1164 return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1165 case GRPQUOTA:
1166 return in_group_p(warn->w_dq_id.gid);
1167 case PRJQUOTA:
1168 return 1;
1169 }
1170 return 0;
1171 }
1172
1173 /* Print warning to user which exceeded quota */
1174 static void print_warning(struct dquot_warn *warn)
1175 {
1176 char *msg = NULL;
1177 struct tty_struct *tty;
1178 int warntype = warn->w_type;
1179
1180 if (warntype == QUOTA_NL_IHARDBELOW ||
1181 warntype == QUOTA_NL_ISOFTBELOW ||
1182 warntype == QUOTA_NL_BHARDBELOW ||
1183 warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1184 return;
1185
1186 tty = get_current_tty();
1187 if (!tty)
1188 return;
1189 tty_write_message(tty, warn->w_sb->s_id);
1190 if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1191 tty_write_message(tty, ": warning, ");
1192 else
1193 tty_write_message(tty, ": write failed, ");
1194 tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1195 switch (warntype) {
1196 case QUOTA_NL_IHARDWARN:
1197 msg = " file limit reached.\r\n";
1198 break;
1199 case QUOTA_NL_ISOFTLONGWARN:
1200 msg = " file quota exceeded too long.\r\n";
1201 break;
1202 case QUOTA_NL_ISOFTWARN:
1203 msg = " file quota exceeded.\r\n";
1204 break;
1205 case QUOTA_NL_BHARDWARN:
1206 msg = " block limit reached.\r\n";
1207 break;
1208 case QUOTA_NL_BSOFTLONGWARN:
1209 msg = " block quota exceeded too long.\r\n";
1210 break;
1211 case QUOTA_NL_BSOFTWARN:
1212 msg = " block quota exceeded.\r\n";
1213 break;
1214 }
1215 tty_write_message(tty, msg);
1216 tty_kref_put(tty);
1217 }
1218 #endif
1219
1220 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1221 int warntype)
1222 {
1223 if (warning_issued(dquot, warntype))
1224 return;
1225 warn->w_type = warntype;
1226 warn->w_sb = dquot->dq_sb;
1227 warn->w_dq_id = dquot->dq_id;
1228 }
1229
1230 /*
1231 * Write warnings to the console and send warning messages over netlink.
1232 *
1233 * Note that this function can call into tty and networking code.
1234 */
1235 static void flush_warnings(struct dquot_warn *warn)
1236 {
1237 int i;
1238
1239 for (i = 0; i < MAXQUOTAS; i++) {
1240 if (warn[i].w_type == QUOTA_NL_NOWARN)
1241 continue;
1242 #ifdef CONFIG_PRINT_QUOTA_WARNING
1243 print_warning(&warn[i]);
1244 #endif
1245 quota_send_warning(warn[i].w_dq_id,
1246 warn[i].w_sb->s_dev, warn[i].w_type);
1247 }
1248 }
1249
1250 static int ignore_hardlimit(struct dquot *dquot)
1251 {
1252 struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1253
1254 return capable(CAP_SYS_RESOURCE) &&
1255 (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1256 !(info->dqi_flags & DQF_ROOT_SQUASH));
1257 }
1258
1259 static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1260 struct dquot_warn *warn)
1261 {
1262 qsize_t newinodes;
1263 int ret = 0;
1264
1265 spin_lock(&dquot->dq_dqb_lock);
1266 newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1267 if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1268 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1269 goto add;
1270
1271 if (dquot->dq_dqb.dqb_ihardlimit &&
1272 newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1273 !ignore_hardlimit(dquot)) {
1274 prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1275 ret = -EDQUOT;
1276 goto out;
1277 }
1278
1279 if (dquot->dq_dqb.dqb_isoftlimit &&
1280 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1281 dquot->dq_dqb.dqb_itime &&
1282 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1283 !ignore_hardlimit(dquot)) {
1284 prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1285 ret = -EDQUOT;
1286 goto out;
1287 }
1288
1289 if (dquot->dq_dqb.dqb_isoftlimit &&
1290 newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1291 dquot->dq_dqb.dqb_itime == 0) {
1292 prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1293 dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1294 sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1295 }
1296 add:
1297 dquot->dq_dqb.dqb_curinodes = newinodes;
1298
1299 out:
1300 spin_unlock(&dquot->dq_dqb_lock);
1301 return ret;
1302 }
1303
1304 static int dquot_add_space(struct dquot *dquot, qsize_t space,
1305 qsize_t rsv_space, unsigned int flags,
1306 struct dquot_warn *warn)
1307 {
1308 qsize_t tspace;
1309 struct super_block *sb = dquot->dq_sb;
1310 int ret = 0;
1311
1312 spin_lock(&dquot->dq_dqb_lock);
1313 if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1314 test_bit(DQ_FAKE_B, &dquot->dq_flags))
1315 goto finish;
1316
1317 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1318 + space + rsv_space;
1319
1320 if (dquot->dq_dqb.dqb_bhardlimit &&
1321 tspace > dquot->dq_dqb.dqb_bhardlimit &&
1322 !ignore_hardlimit(dquot)) {
1323 if (flags & DQUOT_SPACE_WARN)
1324 prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1325 ret = -EDQUOT;
1326 goto finish;
1327 }
1328
1329 if (dquot->dq_dqb.dqb_bsoftlimit &&
1330 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1331 dquot->dq_dqb.dqb_btime &&
1332 ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1333 !ignore_hardlimit(dquot)) {
1334 if (flags & DQUOT_SPACE_WARN)
1335 prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1336 ret = -EDQUOT;
1337 goto finish;
1338 }
1339
1340 if (dquot->dq_dqb.dqb_bsoftlimit &&
1341 tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1342 dquot->dq_dqb.dqb_btime == 0) {
1343 if (flags & DQUOT_SPACE_WARN) {
1344 prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1345 dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1346 sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1347 } else {
1348 /*
1349 * We don't allow preallocation to exceed softlimit so exceeding will
1350 * be always printed
1351 */
1352 ret = -EDQUOT;
1353 goto finish;
1354 }
1355 }
1356 finish:
1357 /*
1358 * We have to be careful and go through warning generation & grace time
1359 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1360 * only here...
1361 */
1362 if (flags & DQUOT_SPACE_NOFAIL)
1363 ret = 0;
1364 if (!ret) {
1365 dquot->dq_dqb.dqb_rsvspace += rsv_space;
1366 dquot->dq_dqb.dqb_curspace += space;
1367 }
1368 spin_unlock(&dquot->dq_dqb_lock);
1369 return ret;
1370 }
1371
1372 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1373 {
1374 qsize_t newinodes;
1375
1376 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1377 dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1378 !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1379 return QUOTA_NL_NOWARN;
1380
1381 newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1382 if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1383 return QUOTA_NL_ISOFTBELOW;
1384 if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1385 newinodes < dquot->dq_dqb.dqb_ihardlimit)
1386 return QUOTA_NL_IHARDBELOW;
1387 return QUOTA_NL_NOWARN;
1388 }
1389
1390 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1391 {
1392 qsize_t tspace;
1393
1394 tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1395
1396 if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1397 tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1398 return QUOTA_NL_NOWARN;
1399
1400 if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1401 return QUOTA_NL_BSOFTBELOW;
1402 if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1403 tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1404 return QUOTA_NL_BHARDBELOW;
1405 return QUOTA_NL_NOWARN;
1406 }
1407
1408 static int dquot_active(const struct inode *inode)
1409 {
1410 struct super_block *sb = inode->i_sb;
1411
1412 if (IS_NOQUOTA(inode))
1413 return 0;
1414 return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1415 }
1416
1417 /*
1418 * Initialize quota pointers in inode
1419 *
1420 * It is better to call this function outside of any transaction as it
1421 * might need a lot of space in journal for dquot structure allocation.
1422 */
1423 static int __dquot_initialize(struct inode *inode, int type)
1424 {
1425 int cnt, init_needed = 0;
1426 struct dquot **dquots, *got[MAXQUOTAS] = {};
1427 struct super_block *sb = inode->i_sb;
1428 qsize_t rsv;
1429 int ret = 0;
1430
1431 if (!dquot_active(inode))
1432 return 0;
1433
1434 dquots = i_dquot(inode);
1435
1436 /* First get references to structures we might need. */
1437 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1438 struct kqid qid;
1439 kprojid_t projid;
1440 int rc;
1441 struct dquot *dquot;
1442
1443 if (type != -1 && cnt != type)
1444 continue;
1445 /*
1446 * The i_dquot should have been initialized in most cases,
1447 * we check it without locking here to avoid unnecessary
1448 * dqget()/dqput() calls.
1449 */
1450 if (dquots[cnt])
1451 continue;
1452
1453 if (!sb_has_quota_active(sb, cnt))
1454 continue;
1455
1456 init_needed = 1;
1457
1458 switch (cnt) {
1459 case USRQUOTA:
1460 qid = make_kqid_uid(inode->i_uid);
1461 break;
1462 case GRPQUOTA:
1463 qid = make_kqid_gid(inode->i_gid);
1464 break;
1465 case PRJQUOTA:
1466 rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1467 if (rc)
1468 continue;
1469 qid = make_kqid_projid(projid);
1470 break;
1471 }
1472 dquot = dqget(sb, qid);
1473 if (IS_ERR(dquot)) {
1474 /* We raced with somebody turning quotas off... */
1475 if (PTR_ERR(dquot) != -ESRCH) {
1476 ret = PTR_ERR(dquot);
1477 goto out_put;
1478 }
1479 dquot = NULL;
1480 }
1481 got[cnt] = dquot;
1482 }
1483
1484 /* All required i_dquot has been initialized */
1485 if (!init_needed)
1486 return 0;
1487
1488 spin_lock(&dq_data_lock);
1489 if (IS_NOQUOTA(inode))
1490 goto out_lock;
1491 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1492 if (type != -1 && cnt != type)
1493 continue;
1494 /* Avoid races with quotaoff() */
1495 if (!sb_has_quota_active(sb, cnt))
1496 continue;
1497 /* We could race with quotaon or dqget() could have failed */
1498 if (!got[cnt])
1499 continue;
1500 if (!dquots[cnt]) {
1501 dquots[cnt] = got[cnt];
1502 got[cnt] = NULL;
1503 /*
1504 * Make quota reservation system happy if someone
1505 * did a write before quota was turned on
1506 */
1507 rsv = inode_get_rsv_space(inode);
1508 if (unlikely(rsv)) {
1509 spin_lock(&inode->i_lock);
1510 /* Get reservation again under proper lock */
1511 rsv = __inode_get_rsv_space(inode);
1512 spin_lock(&dquots[cnt]->dq_dqb_lock);
1513 dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1514 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1515 spin_unlock(&inode->i_lock);
1516 }
1517 }
1518 }
1519 out_lock:
1520 spin_unlock(&dq_data_lock);
1521 out_put:
1522 /* Drop unused references */
1523 dqput_all(got);
1524
1525 return ret;
1526 }
1527
1528 int dquot_initialize(struct inode *inode)
1529 {
1530 return __dquot_initialize(inode, -1);
1531 }
1532 EXPORT_SYMBOL(dquot_initialize);
1533
1534 bool dquot_initialize_needed(struct inode *inode)
1535 {
1536 struct dquot **dquots;
1537 int i;
1538
1539 if (!dquot_active(inode))
1540 return false;
1541
1542 dquots = i_dquot(inode);
1543 for (i = 0; i < MAXQUOTAS; i++)
1544 if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1545 return true;
1546 return false;
1547 }
1548 EXPORT_SYMBOL(dquot_initialize_needed);
1549
1550 /*
1551 * Release all quotas referenced by inode.
1552 *
1553 * This function only be called on inode free or converting
1554 * a file to quota file, no other users for the i_dquot in
1555 * both cases, so we needn't call synchronize_srcu() after
1556 * clearing i_dquot.
1557 */
1558 static void __dquot_drop(struct inode *inode)
1559 {
1560 int cnt;
1561 struct dquot **dquots = i_dquot(inode);
1562 struct dquot *put[MAXQUOTAS];
1563
1564 spin_lock(&dq_data_lock);
1565 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1566 put[cnt] = dquots[cnt];
1567 dquots[cnt] = NULL;
1568 }
1569 spin_unlock(&dq_data_lock);
1570 dqput_all(put);
1571 }
1572
1573 void dquot_drop(struct inode *inode)
1574 {
1575 struct dquot * const *dquots;
1576 int cnt;
1577
1578 if (IS_NOQUOTA(inode))
1579 return;
1580
1581 /*
1582 * Test before calling to rule out calls from proc and such
1583 * where we are not allowed to block. Note that this is
1584 * actually reliable test even without the lock - the caller
1585 * must assure that nobody can come after the DQUOT_DROP and
1586 * add quota pointers back anyway.
1587 */
1588 dquots = i_dquot(inode);
1589 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1590 if (dquots[cnt])
1591 break;
1592 }
1593
1594 if (cnt < MAXQUOTAS)
1595 __dquot_drop(inode);
1596 }
1597 EXPORT_SYMBOL(dquot_drop);
1598
1599 /*
1600 * inode_reserved_space is managed internally by quota, and protected by
1601 * i_lock similar to i_blocks+i_bytes.
1602 */
1603 static qsize_t *inode_reserved_space(struct inode * inode)
1604 {
1605 /* Filesystem must explicitly define it's own method in order to use
1606 * quota reservation interface */
1607 BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1608 return inode->i_sb->dq_op->get_reserved_space(inode);
1609 }
1610
1611 static qsize_t __inode_get_rsv_space(struct inode *inode)
1612 {
1613 if (!inode->i_sb->dq_op->get_reserved_space)
1614 return 0;
1615 return *inode_reserved_space(inode);
1616 }
1617
1618 static qsize_t inode_get_rsv_space(struct inode *inode)
1619 {
1620 qsize_t ret;
1621
1622 if (!inode->i_sb->dq_op->get_reserved_space)
1623 return 0;
1624 spin_lock(&inode->i_lock);
1625 ret = __inode_get_rsv_space(inode);
1626 spin_unlock(&inode->i_lock);
1627 return ret;
1628 }
1629
1630 /*
1631 * This functions updates i_blocks+i_bytes fields and quota information
1632 * (together with appropriate checks).
1633 *
1634 * NOTE: We absolutely rely on the fact that caller dirties the inode
1635 * (usually helpers in quotaops.h care about this) and holds a handle for
1636 * the current transaction so that dquot write and inode write go into the
1637 * same transaction.
1638 */
1639
1640 /*
1641 * This operation can block, but only after everything is updated
1642 */
1643 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1644 {
1645 int cnt, ret = 0, index;
1646 struct dquot_warn warn[MAXQUOTAS];
1647 int reserve = flags & DQUOT_SPACE_RESERVE;
1648 struct dquot **dquots;
1649
1650 if (!dquot_active(inode)) {
1651 if (reserve) {
1652 spin_lock(&inode->i_lock);
1653 *inode_reserved_space(inode) += number;
1654 spin_unlock(&inode->i_lock);
1655 } else {
1656 inode_add_bytes(inode, number);
1657 }
1658 goto out;
1659 }
1660
1661 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1662 warn[cnt].w_type = QUOTA_NL_NOWARN;
1663
1664 dquots = i_dquot(inode);
1665 index = srcu_read_lock(&dquot_srcu);
1666 spin_lock(&inode->i_lock);
1667 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1668 if (!dquots[cnt])
1669 continue;
1670 if (flags & DQUOT_SPACE_RESERVE) {
1671 ret = dquot_add_space(dquots[cnt], 0, number, flags,
1672 &warn[cnt]);
1673 } else {
1674 ret = dquot_add_space(dquots[cnt], number, 0, flags,
1675 &warn[cnt]);
1676 }
1677 if (ret) {
1678 /* Back out changes we already did */
1679 for (cnt--; cnt >= 0; cnt--) {
1680 if (!dquots[cnt])
1681 continue;
1682 spin_lock(&dquots[cnt]->dq_dqb_lock);
1683 if (flags & DQUOT_SPACE_RESERVE) {
1684 dquots[cnt]->dq_dqb.dqb_rsvspace -=
1685 number;
1686 } else {
1687 dquots[cnt]->dq_dqb.dqb_curspace -=
1688 number;
1689 }
1690 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1691 }
1692 spin_unlock(&inode->i_lock);
1693 goto out_flush_warn;
1694 }
1695 }
1696 if (reserve)
1697 *inode_reserved_space(inode) += number;
1698 else
1699 __inode_add_bytes(inode, number);
1700 spin_unlock(&inode->i_lock);
1701
1702 if (reserve)
1703 goto out_flush_warn;
1704 mark_all_dquot_dirty(dquots);
1705 out_flush_warn:
1706 srcu_read_unlock(&dquot_srcu, index);
1707 flush_warnings(warn);
1708 out:
1709 return ret;
1710 }
1711 EXPORT_SYMBOL(__dquot_alloc_space);
1712
1713 /*
1714 * This operation can block, but only after everything is updated
1715 */
1716 int dquot_alloc_inode(struct inode *inode)
1717 {
1718 int cnt, ret = 0, index;
1719 struct dquot_warn warn[MAXQUOTAS];
1720 struct dquot * const *dquots;
1721
1722 if (!dquot_active(inode))
1723 return 0;
1724 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1725 warn[cnt].w_type = QUOTA_NL_NOWARN;
1726
1727 dquots = i_dquot(inode);
1728 index = srcu_read_lock(&dquot_srcu);
1729 spin_lock(&inode->i_lock);
1730 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1731 if (!dquots[cnt])
1732 continue;
1733 ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1734 if (ret) {
1735 for (cnt--; cnt >= 0; cnt--) {
1736 if (!dquots[cnt])
1737 continue;
1738 /* Back out changes we already did */
1739 spin_lock(&dquots[cnt]->dq_dqb_lock);
1740 dquots[cnt]->dq_dqb.dqb_curinodes--;
1741 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1742 }
1743 goto warn_put_all;
1744 }
1745 }
1746
1747 warn_put_all:
1748 spin_unlock(&inode->i_lock);
1749 if (ret == 0)
1750 mark_all_dquot_dirty(dquots);
1751 srcu_read_unlock(&dquot_srcu, index);
1752 flush_warnings(warn);
1753 return ret;
1754 }
1755 EXPORT_SYMBOL(dquot_alloc_inode);
1756
1757 /*
1758 * Convert in-memory reserved quotas to real consumed quotas
1759 */
1760 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1761 {
1762 struct dquot **dquots;
1763 int cnt, index;
1764
1765 if (!dquot_active(inode)) {
1766 spin_lock(&inode->i_lock);
1767 *inode_reserved_space(inode) -= number;
1768 __inode_add_bytes(inode, number);
1769 spin_unlock(&inode->i_lock);
1770 return 0;
1771 }
1772
1773 dquots = i_dquot(inode);
1774 index = srcu_read_lock(&dquot_srcu);
1775 spin_lock(&inode->i_lock);
1776 /* Claim reserved quotas to allocated quotas */
1777 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1778 if (dquots[cnt]) {
1779 struct dquot *dquot = dquots[cnt];
1780
1781 spin_lock(&dquot->dq_dqb_lock);
1782 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1783 number = dquot->dq_dqb.dqb_rsvspace;
1784 dquot->dq_dqb.dqb_curspace += number;
1785 dquot->dq_dqb.dqb_rsvspace -= number;
1786 spin_unlock(&dquot->dq_dqb_lock);
1787 }
1788 }
1789 /* Update inode bytes */
1790 *inode_reserved_space(inode) -= number;
1791 __inode_add_bytes(inode, number);
1792 spin_unlock(&inode->i_lock);
1793 mark_all_dquot_dirty(dquots);
1794 srcu_read_unlock(&dquot_srcu, index);
1795 return 0;
1796 }
1797 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1798
1799 /*
1800 * Convert allocated space back to in-memory reserved quotas
1801 */
1802 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1803 {
1804 struct dquot **dquots;
1805 int cnt, index;
1806
1807 if (!dquot_active(inode)) {
1808 spin_lock(&inode->i_lock);
1809 *inode_reserved_space(inode) += number;
1810 __inode_sub_bytes(inode, number);
1811 spin_unlock(&inode->i_lock);
1812 return;
1813 }
1814
1815 dquots = i_dquot(inode);
1816 index = srcu_read_lock(&dquot_srcu);
1817 spin_lock(&inode->i_lock);
1818 /* Claim reserved quotas to allocated quotas */
1819 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1820 if (dquots[cnt]) {
1821 struct dquot *dquot = dquots[cnt];
1822
1823 spin_lock(&dquot->dq_dqb_lock);
1824 if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1825 number = dquot->dq_dqb.dqb_curspace;
1826 dquot->dq_dqb.dqb_rsvspace += number;
1827 dquot->dq_dqb.dqb_curspace -= number;
1828 spin_unlock(&dquot->dq_dqb_lock);
1829 }
1830 }
1831 /* Update inode bytes */
1832 *inode_reserved_space(inode) += number;
1833 __inode_sub_bytes(inode, number);
1834 spin_unlock(&inode->i_lock);
1835 mark_all_dquot_dirty(dquots);
1836 srcu_read_unlock(&dquot_srcu, index);
1837 return;
1838 }
1839 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1840
1841 /*
1842 * This operation can block, but only after everything is updated
1843 */
1844 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1845 {
1846 unsigned int cnt;
1847 struct dquot_warn warn[MAXQUOTAS];
1848 struct dquot **dquots;
1849 int reserve = flags & DQUOT_SPACE_RESERVE, index;
1850
1851 if (!dquot_active(inode)) {
1852 if (reserve) {
1853 spin_lock(&inode->i_lock);
1854 *inode_reserved_space(inode) -= number;
1855 spin_unlock(&inode->i_lock);
1856 } else {
1857 inode_sub_bytes(inode, number);
1858 }
1859 return;
1860 }
1861
1862 dquots = i_dquot(inode);
1863 index = srcu_read_lock(&dquot_srcu);
1864 spin_lock(&inode->i_lock);
1865 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1866 int wtype;
1867
1868 warn[cnt].w_type = QUOTA_NL_NOWARN;
1869 if (!dquots[cnt])
1870 continue;
1871 spin_lock(&dquots[cnt]->dq_dqb_lock);
1872 wtype = info_bdq_free(dquots[cnt], number);
1873 if (wtype != QUOTA_NL_NOWARN)
1874 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1875 if (reserve)
1876 dquot_free_reserved_space(dquots[cnt], number);
1877 else
1878 dquot_decr_space(dquots[cnt], number);
1879 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1880 }
1881 if (reserve)
1882 *inode_reserved_space(inode) -= number;
1883 else
1884 __inode_sub_bytes(inode, number);
1885 spin_unlock(&inode->i_lock);
1886
1887 if (reserve)
1888 goto out_unlock;
1889 mark_all_dquot_dirty(dquots);
1890 out_unlock:
1891 srcu_read_unlock(&dquot_srcu, index);
1892 flush_warnings(warn);
1893 }
1894 EXPORT_SYMBOL(__dquot_free_space);
1895
1896 /*
1897 * This operation can block, but only after everything is updated
1898 */
1899 void dquot_free_inode(struct inode *inode)
1900 {
1901 unsigned int cnt;
1902 struct dquot_warn warn[MAXQUOTAS];
1903 struct dquot * const *dquots;
1904 int index;
1905
1906 if (!dquot_active(inode))
1907 return;
1908
1909 dquots = i_dquot(inode);
1910 index = srcu_read_lock(&dquot_srcu);
1911 spin_lock(&inode->i_lock);
1912 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1913 int wtype;
1914
1915 warn[cnt].w_type = QUOTA_NL_NOWARN;
1916 if (!dquots[cnt])
1917 continue;
1918 spin_lock(&dquots[cnt]->dq_dqb_lock);
1919 wtype = info_idq_free(dquots[cnt], 1);
1920 if (wtype != QUOTA_NL_NOWARN)
1921 prepare_warning(&warn[cnt], dquots[cnt], wtype);
1922 dquot_decr_inodes(dquots[cnt], 1);
1923 spin_unlock(&dquots[cnt]->dq_dqb_lock);
1924 }
1925 spin_unlock(&inode->i_lock);
1926 mark_all_dquot_dirty(dquots);
1927 srcu_read_unlock(&dquot_srcu, index);
1928 flush_warnings(warn);
1929 }
1930 EXPORT_SYMBOL(dquot_free_inode);
1931
1932 /*
1933 * Transfer the number of inode and blocks from one diskquota to an other.
1934 * On success, dquot references in transfer_to are consumed and references
1935 * to original dquots that need to be released are placed there. On failure,
1936 * references are kept untouched.
1937 *
1938 * This operation can block, but only after everything is updated
1939 * A transaction must be started when entering this function.
1940 *
1941 * We are holding reference on transfer_from & transfer_to, no need to
1942 * protect them by srcu_read_lock().
1943 */
1944 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1945 {
1946 qsize_t cur_space;
1947 qsize_t rsv_space = 0;
1948 qsize_t inode_usage = 1;
1949 struct dquot *transfer_from[MAXQUOTAS] = {};
1950 int cnt, ret = 0;
1951 char is_valid[MAXQUOTAS] = {};
1952 struct dquot_warn warn_to[MAXQUOTAS];
1953 struct dquot_warn warn_from_inodes[MAXQUOTAS];
1954 struct dquot_warn warn_from_space[MAXQUOTAS];
1955
1956 if (IS_NOQUOTA(inode))
1957 return 0;
1958
1959 if (inode->i_sb->dq_op->get_inode_usage) {
1960 ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1961 if (ret)
1962 return ret;
1963 }
1964
1965 /* Initialize the arrays */
1966 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1967 warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1968 warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1969 warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1970 }
1971
1972 spin_lock(&dq_data_lock);
1973 spin_lock(&inode->i_lock);
1974 if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
1975 spin_unlock(&inode->i_lock);
1976 spin_unlock(&dq_data_lock);
1977 return 0;
1978 }
1979 cur_space = __inode_get_bytes(inode);
1980 rsv_space = __inode_get_rsv_space(inode);
1981 /*
1982 * Build the transfer_from list, check limits, and update usage in
1983 * the target structures.
1984 */
1985 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1986 /*
1987 * Skip changes for same uid or gid or for turned off quota-type.
1988 */
1989 if (!transfer_to[cnt])
1990 continue;
1991 /* Avoid races with quotaoff() */
1992 if (!sb_has_quota_active(inode->i_sb, cnt))
1993 continue;
1994 is_valid[cnt] = 1;
1995 transfer_from[cnt] = i_dquot(inode)[cnt];
1996 ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1997 &warn_to[cnt]);
1998 if (ret)
1999 goto over_quota;
2000 ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2001 DQUOT_SPACE_WARN, &warn_to[cnt]);
2002 if (ret) {
2003 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2004 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2005 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2006 goto over_quota;
2007 }
2008 }
2009
2010 /* Decrease usage for source structures and update quota pointers */
2011 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2012 if (!is_valid[cnt])
2013 continue;
2014 /* Due to IO error we might not have transfer_from[] structure */
2015 if (transfer_from[cnt]) {
2016 int wtype;
2017
2018 spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2019 wtype = info_idq_free(transfer_from[cnt], inode_usage);
2020 if (wtype != QUOTA_NL_NOWARN)
2021 prepare_warning(&warn_from_inodes[cnt],
2022 transfer_from[cnt], wtype);
2023 wtype = info_bdq_free(transfer_from[cnt],
2024 cur_space + rsv_space);
2025 if (wtype != QUOTA_NL_NOWARN)
2026 prepare_warning(&warn_from_space[cnt],
2027 transfer_from[cnt], wtype);
2028 dquot_decr_inodes(transfer_from[cnt], inode_usage);
2029 dquot_decr_space(transfer_from[cnt], cur_space);
2030 dquot_free_reserved_space(transfer_from[cnt],
2031 rsv_space);
2032 spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2033 }
2034 i_dquot(inode)[cnt] = transfer_to[cnt];
2035 }
2036 spin_unlock(&inode->i_lock);
2037 spin_unlock(&dq_data_lock);
2038
2039 mark_all_dquot_dirty(transfer_from);
2040 mark_all_dquot_dirty(transfer_to);
2041 flush_warnings(warn_to);
2042 flush_warnings(warn_from_inodes);
2043 flush_warnings(warn_from_space);
2044 /* Pass back references to put */
2045 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2046 if (is_valid[cnt])
2047 transfer_to[cnt] = transfer_from[cnt];
2048 return 0;
2049 over_quota:
2050 /* Back out changes we already did */
2051 for (cnt--; cnt >= 0; cnt--) {
2052 if (!is_valid[cnt])
2053 continue;
2054 spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2055 dquot_decr_inodes(transfer_to[cnt], inode_usage);
2056 dquot_decr_space(transfer_to[cnt], cur_space);
2057 dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2058 spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2059 }
2060 spin_unlock(&inode->i_lock);
2061 spin_unlock(&dq_data_lock);
2062 flush_warnings(warn_to);
2063 return ret;
2064 }
2065 EXPORT_SYMBOL(__dquot_transfer);
2066
2067 /* Wrapper for transferring ownership of an inode for uid/gid only
2068 * Called from FSXXX_setattr()
2069 */
2070 int dquot_transfer(struct inode *inode, struct iattr *iattr)
2071 {
2072 struct dquot *transfer_to[MAXQUOTAS] = {};
2073 struct dquot *dquot;
2074 struct super_block *sb = inode->i_sb;
2075 int ret;
2076
2077 if (!dquot_active(inode))
2078 return 0;
2079
2080 if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2081 dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2082 if (IS_ERR(dquot)) {
2083 if (PTR_ERR(dquot) != -ESRCH) {
2084 ret = PTR_ERR(dquot);
2085 goto out_put;
2086 }
2087 dquot = NULL;
2088 }
2089 transfer_to[USRQUOTA] = dquot;
2090 }
2091 if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2092 dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2093 if (IS_ERR(dquot)) {
2094 if (PTR_ERR(dquot) != -ESRCH) {
2095 ret = PTR_ERR(dquot);
2096 goto out_put;
2097 }
2098 dquot = NULL;
2099 }
2100 transfer_to[GRPQUOTA] = dquot;
2101 }
2102 ret = __dquot_transfer(inode, transfer_to);
2103 out_put:
2104 dqput_all(transfer_to);
2105 return ret;
2106 }
2107 EXPORT_SYMBOL(dquot_transfer);
2108
2109 /*
2110 * Write info of quota file to disk
2111 */
2112 int dquot_commit_info(struct super_block *sb, int type)
2113 {
2114 struct quota_info *dqopt = sb_dqopt(sb);
2115
2116 return dqopt->ops[type]->write_file_info(sb, type);
2117 }
2118 EXPORT_SYMBOL(dquot_commit_info);
2119
2120 int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2121 {
2122 struct quota_info *dqopt = sb_dqopt(sb);
2123
2124 if (!sb_has_quota_active(sb, qid->type))
2125 return -ESRCH;
2126 if (!dqopt->ops[qid->type]->get_next_id)
2127 return -ENOSYS;
2128 return dqopt->ops[qid->type]->get_next_id(sb, qid);
2129 }
2130 EXPORT_SYMBOL(dquot_get_next_id);
2131
2132 /*
2133 * Definitions of diskquota operations.
2134 */
2135 const struct dquot_operations dquot_operations = {
2136 .write_dquot = dquot_commit,
2137 .acquire_dquot = dquot_acquire,
2138 .release_dquot = dquot_release,
2139 .mark_dirty = dquot_mark_dquot_dirty,
2140 .write_info = dquot_commit_info,
2141 .alloc_dquot = dquot_alloc,
2142 .destroy_dquot = dquot_destroy,
2143 .get_next_id = dquot_get_next_id,
2144 };
2145 EXPORT_SYMBOL(dquot_operations);
2146
2147 /*
2148 * Generic helper for ->open on filesystems supporting disk quotas.
2149 */
2150 int dquot_file_open(struct inode *inode, struct file *file)
2151 {
2152 int error;
2153
2154 error = generic_file_open(inode, file);
2155 if (!error && (file->f_mode & FMODE_WRITE))
2156 error = dquot_initialize(inode);
2157 return error;
2158 }
2159 EXPORT_SYMBOL(dquot_file_open);
2160
2161 /*
2162 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2163 */
2164 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2165 {
2166 int cnt, ret = 0;
2167 struct quota_info *dqopt = sb_dqopt(sb);
2168 struct inode *toputinode[MAXQUOTAS];
2169
2170 /* s_umount should be held in exclusive mode */
2171 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2172 up_read(&sb->s_umount);
2173
2174 /* Cannot turn off usage accounting without turning off limits, or
2175 * suspend quotas and simultaneously turn quotas off. */
2176 if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2177 || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2178 DQUOT_USAGE_ENABLED)))
2179 return -EINVAL;
2180
2181 /*
2182 * Skip everything if there's nothing to do. We have to do this because
2183 * sometimes we are called when fill_super() failed and calling
2184 * sync_fs() in such cases does no good.
2185 */
2186 if (!sb_any_quota_loaded(sb))
2187 return 0;
2188
2189 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2190 toputinode[cnt] = NULL;
2191 if (type != -1 && cnt != type)
2192 continue;
2193 if (!sb_has_quota_loaded(sb, cnt))
2194 continue;
2195
2196 if (flags & DQUOT_SUSPENDED) {
2197 spin_lock(&dq_state_lock);
2198 dqopt->flags |=
2199 dquot_state_flag(DQUOT_SUSPENDED, cnt);
2200 spin_unlock(&dq_state_lock);
2201 } else {
2202 spin_lock(&dq_state_lock);
2203 dqopt->flags &= ~dquot_state_flag(flags, cnt);
2204 /* Turning off suspended quotas? */
2205 if (!sb_has_quota_loaded(sb, cnt) &&
2206 sb_has_quota_suspended(sb, cnt)) {
2207 dqopt->flags &= ~dquot_state_flag(
2208 DQUOT_SUSPENDED, cnt);
2209 spin_unlock(&dq_state_lock);
2210 iput(dqopt->files[cnt]);
2211 dqopt->files[cnt] = NULL;
2212 continue;
2213 }
2214 spin_unlock(&dq_state_lock);
2215 }
2216
2217 /* We still have to keep quota loaded? */
2218 if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2219 continue;
2220
2221 /* Note: these are blocking operations */
2222 drop_dquot_ref(sb, cnt);
2223 invalidate_dquots(sb, cnt);
2224 /*
2225 * Now all dquots should be invalidated, all writes done so we
2226 * should be only users of the info. No locks needed.
2227 */
2228 if (info_dirty(&dqopt->info[cnt]))
2229 sb->dq_op->write_info(sb, cnt);
2230 if (dqopt->ops[cnt]->free_file_info)
2231 dqopt->ops[cnt]->free_file_info(sb, cnt);
2232 put_quota_format(dqopt->info[cnt].dqi_format);
2233
2234 toputinode[cnt] = dqopt->files[cnt];
2235 if (!sb_has_quota_loaded(sb, cnt))
2236 dqopt->files[cnt] = NULL;
2237 dqopt->info[cnt].dqi_flags = 0;
2238 dqopt->info[cnt].dqi_igrace = 0;
2239 dqopt->info[cnt].dqi_bgrace = 0;
2240 dqopt->ops[cnt] = NULL;
2241 }
2242
2243 /* Skip syncing and setting flags if quota files are hidden */
2244 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2245 goto put_inodes;
2246
2247 /* Sync the superblock so that buffers with quota data are written to
2248 * disk (and so userspace sees correct data afterwards). */
2249 if (sb->s_op->sync_fs)
2250 sb->s_op->sync_fs(sb, 1);
2251 sync_blockdev(sb->s_bdev);
2252 /* Now the quota files are just ordinary files and we can set the
2253 * inode flags back. Moreover we discard the pagecache so that
2254 * userspace sees the writes we did bypassing the pagecache. We
2255 * must also discard the blockdev buffers so that we see the
2256 * changes done by userspace on the next quotaon() */
2257 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2258 /* This can happen when suspending quotas on remount-ro... */
2259 if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2260 inode_lock(toputinode[cnt]);
2261 toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2262 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2263 inode_unlock(toputinode[cnt]);
2264 mark_inode_dirty_sync(toputinode[cnt]);
2265 }
2266 if (sb->s_bdev)
2267 invalidate_bdev(sb->s_bdev);
2268 put_inodes:
2269 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2270 if (toputinode[cnt]) {
2271 /* On remount RO, we keep the inode pointer so that we
2272 * can reenable quota on the subsequent remount RW. We
2273 * have to check 'flags' variable and not use sb_has_
2274 * function because another quotaon / quotaoff could
2275 * change global state before we got here. We refuse
2276 * to suspend quotas when there is pending delete on
2277 * the quota file... */
2278 if (!(flags & DQUOT_SUSPENDED))
2279 iput(toputinode[cnt]);
2280 else if (!toputinode[cnt]->i_nlink)
2281 ret = -EBUSY;
2282 }
2283 return ret;
2284 }
2285 EXPORT_SYMBOL(dquot_disable);
2286
2287 int dquot_quota_off(struct super_block *sb, int type)
2288 {
2289 return dquot_disable(sb, type,
2290 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2291 }
2292 EXPORT_SYMBOL(dquot_quota_off);
2293
2294 /*
2295 * Turn quotas on on a device
2296 */
2297
2298 /*
2299 * Helper function to turn quotas on when we already have the inode of
2300 * quota file and no quota information is loaded.
2301 */
2302 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2303 unsigned int flags)
2304 {
2305 struct quota_format_type *fmt = find_quota_format(format_id);
2306 struct super_block *sb = inode->i_sb;
2307 struct quota_info *dqopt = sb_dqopt(sb);
2308 int error;
2309
2310 if (!fmt)
2311 return -ESRCH;
2312 if (!S_ISREG(inode->i_mode)) {
2313 error = -EACCES;
2314 goto out_fmt;
2315 }
2316 if (IS_RDONLY(inode)) {
2317 error = -EROFS;
2318 goto out_fmt;
2319 }
2320 if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2321 (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2322 error = -EINVAL;
2323 goto out_fmt;
2324 }
2325 /* Filesystems outside of init_user_ns not yet supported */
2326 if (sb->s_user_ns != &init_user_ns) {
2327 error = -EINVAL;
2328 goto out_fmt;
2329 }
2330 /* Usage always has to be set... */
2331 if (!(flags & DQUOT_USAGE_ENABLED)) {
2332 error = -EINVAL;
2333 goto out_fmt;
2334 }
2335 if (sb_has_quota_loaded(sb, type)) {
2336 error = -EBUSY;
2337 goto out_fmt;
2338 }
2339
2340 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2341 /* As we bypass the pagecache we must now flush all the
2342 * dirty data and invalidate caches so that kernel sees
2343 * changes from userspace. It is not enough to just flush
2344 * the quota file since if blocksize < pagesize, invalidation
2345 * of the cache could fail because of other unrelated dirty
2346 * data */
2347 sync_filesystem(sb);
2348 invalidate_bdev(sb->s_bdev);
2349 }
2350
2351 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2352 /* We don't want quota and atime on quota files (deadlocks
2353 * possible) Also nobody should write to the file - we use
2354 * special IO operations which ignore the immutable bit. */
2355 inode_lock(inode);
2356 inode->i_flags |= S_NOQUOTA;
2357 inode_unlock(inode);
2358 /*
2359 * When S_NOQUOTA is set, remove dquot references as no more
2360 * references can be added
2361 */
2362 __dquot_drop(inode);
2363 }
2364
2365 error = -EIO;
2366 dqopt->files[type] = igrab(inode);
2367 if (!dqopt->files[type])
2368 goto out_file_flags;
2369 error = -EINVAL;
2370 if (!fmt->qf_ops->check_quota_file(sb, type))
2371 goto out_file_init;
2372
2373 dqopt->ops[type] = fmt->qf_ops;
2374 dqopt->info[type].dqi_format = fmt;
2375 dqopt->info[type].dqi_fmt_id = format_id;
2376 INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2377 error = dqopt->ops[type]->read_file_info(sb, type);
2378 if (error < 0)
2379 goto out_file_init;
2380 if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2381 spin_lock(&dq_data_lock);
2382 dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2383 spin_unlock(&dq_data_lock);
2384 }
2385 spin_lock(&dq_state_lock);
2386 dqopt->flags |= dquot_state_flag(flags, type);
2387 spin_unlock(&dq_state_lock);
2388
2389 error = add_dquot_ref(sb, type);
2390 if (error)
2391 dquot_disable(sb, type, flags);
2392
2393 return error;
2394 out_file_init:
2395 dqopt->files[type] = NULL;
2396 iput(inode);
2397 out_file_flags:
2398 inode_lock(inode);
2399 inode->i_flags &= ~S_NOQUOTA;
2400 inode_unlock(inode);
2401 out_fmt:
2402 put_quota_format(fmt);
2403
2404 return error;
2405 }
2406
2407 /* Reenable quotas on remount RW */
2408 int dquot_resume(struct super_block *sb, int type)
2409 {
2410 struct quota_info *dqopt = sb_dqopt(sb);
2411 struct inode *inode;
2412 int ret = 0, cnt;
2413 unsigned int flags;
2414
2415 /* s_umount should be held in exclusive mode */
2416 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2417 up_read(&sb->s_umount);
2418
2419 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2420 if (type != -1 && cnt != type)
2421 continue;
2422 if (!sb_has_quota_suspended(sb, cnt))
2423 continue;
2424
2425 inode = dqopt->files[cnt];
2426 dqopt->files[cnt] = NULL;
2427 spin_lock(&dq_state_lock);
2428 flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2429 DQUOT_LIMITS_ENABLED,
2430 cnt);
2431 dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2432 spin_unlock(&dq_state_lock);
2433
2434 flags = dquot_generic_flag(flags, cnt);
2435 ret = vfs_load_quota_inode(inode, cnt,
2436 dqopt->info[cnt].dqi_fmt_id, flags);
2437 iput(inode);
2438 }
2439
2440 return ret;
2441 }
2442 EXPORT_SYMBOL(dquot_resume);
2443
2444 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2445 const struct path *path)
2446 {
2447 int error = security_quota_on(path->dentry);
2448 if (error)
2449 return error;
2450 /* Quota file not on the same filesystem? */
2451 if (path->dentry->d_sb != sb)
2452 error = -EXDEV;
2453 else
2454 error = vfs_load_quota_inode(d_inode(path->dentry), type,
2455 format_id, DQUOT_USAGE_ENABLED |
2456 DQUOT_LIMITS_ENABLED);
2457 return error;
2458 }
2459 EXPORT_SYMBOL(dquot_quota_on);
2460
2461 /*
2462 * More powerful function for turning on quotas allowing setting
2463 * of individual quota flags
2464 */
2465 int dquot_enable(struct inode *inode, int type, int format_id,
2466 unsigned int flags)
2467 {
2468 struct super_block *sb = inode->i_sb;
2469
2470 /* Just unsuspend quotas? */
2471 BUG_ON(flags & DQUOT_SUSPENDED);
2472 /* s_umount should be held in exclusive mode */
2473 if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2474 up_read(&sb->s_umount);
2475
2476 if (!flags)
2477 return 0;
2478 /* Just updating flags needed? */
2479 if (sb_has_quota_loaded(sb, type)) {
2480 if (flags & DQUOT_USAGE_ENABLED &&
2481 sb_has_quota_usage_enabled(sb, type))
2482 return -EBUSY;
2483 if (flags & DQUOT_LIMITS_ENABLED &&
2484 sb_has_quota_limits_enabled(sb, type))
2485 return -EBUSY;
2486 spin_lock(&dq_state_lock);
2487 sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2488 spin_unlock(&dq_state_lock);
2489 return 0;
2490 }
2491
2492 return vfs_load_quota_inode(inode, type, format_id, flags);
2493 }
2494 EXPORT_SYMBOL(dquot_enable);
2495
2496 /*
2497 * This function is used when filesystem needs to initialize quotas
2498 * during mount time.
2499 */
2500 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2501 int format_id, int type)
2502 {
2503 struct dentry *dentry;
2504 int error;
2505
2506 dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2507 if (IS_ERR(dentry))
2508 return PTR_ERR(dentry);
2509
2510 if (d_really_is_negative(dentry)) {
2511 error = -ENOENT;
2512 goto out;
2513 }
2514
2515 error = security_quota_on(dentry);
2516 if (!error)
2517 error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2518 DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2519
2520 out:
2521 dput(dentry);
2522 return error;
2523 }
2524 EXPORT_SYMBOL(dquot_quota_on_mount);
2525
2526 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2527 {
2528 int ret;
2529 int type;
2530 struct quota_info *dqopt = sb_dqopt(sb);
2531
2532 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2533 return -ENOSYS;
2534 /* Accounting cannot be turned on while fs is mounted */
2535 flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2536 if (!flags)
2537 return -EINVAL;
2538 for (type = 0; type < MAXQUOTAS; type++) {
2539 if (!(flags & qtype_enforce_flag(type)))
2540 continue;
2541 /* Can't enforce without accounting */
2542 if (!sb_has_quota_usage_enabled(sb, type))
2543 return -EINVAL;
2544 ret = dquot_enable(dqopt->files[type], type,
2545 dqopt->info[type].dqi_fmt_id,
2546 DQUOT_LIMITS_ENABLED);
2547 if (ret < 0)
2548 goto out_err;
2549 }
2550 return 0;
2551 out_err:
2552 /* Backout enforcement enablement we already did */
2553 for (type--; type >= 0; type--) {
2554 if (flags & qtype_enforce_flag(type))
2555 dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2556 }
2557 /* Error code translation for better compatibility with XFS */
2558 if (ret == -EBUSY)
2559 ret = -EEXIST;
2560 return ret;
2561 }
2562
2563 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2564 {
2565 int ret;
2566 int type;
2567 struct quota_info *dqopt = sb_dqopt(sb);
2568
2569 if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2570 return -ENOSYS;
2571 /*
2572 * We don't support turning off accounting via quotactl. In principle
2573 * quota infrastructure can do this but filesystems don't expect
2574 * userspace to be able to do it.
2575 */
2576 if (flags &
2577 (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2578 return -EOPNOTSUPP;
2579
2580 /* Filter out limits not enabled */
2581 for (type = 0; type < MAXQUOTAS; type++)
2582 if (!sb_has_quota_limits_enabled(sb, type))
2583 flags &= ~qtype_enforce_flag(type);
2584 /* Nothing left? */
2585 if (!flags)
2586 return -EEXIST;
2587 for (type = 0; type < MAXQUOTAS; type++) {
2588 if (flags & qtype_enforce_flag(type)) {
2589 ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2590 if (ret < 0)
2591 goto out_err;
2592 }
2593 }
2594 return 0;
2595 out_err:
2596 /* Backout enforcement disabling we already did */
2597 for (type--; type >= 0; type--) {
2598 if (flags & qtype_enforce_flag(type))
2599 dquot_enable(dqopt->files[type], type,
2600 dqopt->info[type].dqi_fmt_id,
2601 DQUOT_LIMITS_ENABLED);
2602 }
2603 return ret;
2604 }
2605
2606 /* Generic routine for getting common part of quota structure */
2607 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2608 {
2609 struct mem_dqblk *dm = &dquot->dq_dqb;
2610
2611 memset(di, 0, sizeof(*di));
2612 spin_lock(&dquot->dq_dqb_lock);
2613 di->d_spc_hardlimit = dm->dqb_bhardlimit;
2614 di->d_spc_softlimit = dm->dqb_bsoftlimit;
2615 di->d_ino_hardlimit = dm->dqb_ihardlimit;
2616 di->d_ino_softlimit = dm->dqb_isoftlimit;
2617 di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2618 di->d_ino_count = dm->dqb_curinodes;
2619 di->d_spc_timer = dm->dqb_btime;
2620 di->d_ino_timer = dm->dqb_itime;
2621 spin_unlock(&dquot->dq_dqb_lock);
2622 }
2623
2624 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2625 struct qc_dqblk *di)
2626 {
2627 struct dquot *dquot;
2628
2629 dquot = dqget(sb, qid);
2630 if (IS_ERR(dquot))
2631 return PTR_ERR(dquot);
2632 do_get_dqblk(dquot, di);
2633 dqput(dquot);
2634
2635 return 0;
2636 }
2637 EXPORT_SYMBOL(dquot_get_dqblk);
2638
2639 int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2640 struct qc_dqblk *di)
2641 {
2642 struct dquot *dquot;
2643 int err;
2644
2645 if (!sb->dq_op->get_next_id)
2646 return -ENOSYS;
2647 err = sb->dq_op->get_next_id(sb, qid);
2648 if (err < 0)
2649 return err;
2650 dquot = dqget(sb, *qid);
2651 if (IS_ERR(dquot))
2652 return PTR_ERR(dquot);
2653 do_get_dqblk(dquot, di);
2654 dqput(dquot);
2655
2656 return 0;
2657 }
2658 EXPORT_SYMBOL(dquot_get_next_dqblk);
2659
2660 #define VFS_QC_MASK \
2661 (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2662 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2663 QC_SPC_TIMER | QC_INO_TIMER)
2664
2665 /* Generic routine for setting common part of quota structure */
2666 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2667 {
2668 struct mem_dqblk *dm = &dquot->dq_dqb;
2669 int check_blim = 0, check_ilim = 0;
2670 struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2671
2672 if (di->d_fieldmask & ~VFS_QC_MASK)
2673 return -EINVAL;
2674
2675 if (((di->d_fieldmask & QC_SPC_SOFT) &&
2676 di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2677 ((di->d_fieldmask & QC_SPC_HARD) &&
2678 di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2679 ((di->d_fieldmask & QC_INO_SOFT) &&
2680 (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2681 ((di->d_fieldmask & QC_INO_HARD) &&
2682 (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2683 return -ERANGE;
2684
2685 spin_lock(&dquot->dq_dqb_lock);
2686 if (di->d_fieldmask & QC_SPACE) {
2687 dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2688 check_blim = 1;
2689 set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2690 }
2691
2692 if (di->d_fieldmask & QC_SPC_SOFT)
2693 dm->dqb_bsoftlimit = di->d_spc_softlimit;
2694 if (di->d_fieldmask & QC_SPC_HARD)
2695 dm->dqb_bhardlimit = di->d_spc_hardlimit;
2696 if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2697 check_blim = 1;
2698 set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2699 }
2700
2701 if (di->d_fieldmask & QC_INO_COUNT) {
2702 dm->dqb_curinodes = di->d_ino_count;
2703 check_ilim = 1;
2704 set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2705 }
2706
2707 if (di->d_fieldmask & QC_INO_SOFT)
2708 dm->dqb_isoftlimit = di->d_ino_softlimit;
2709 if (di->d_fieldmask & QC_INO_HARD)
2710 dm->dqb_ihardlimit = di->d_ino_hardlimit;
2711 if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2712 check_ilim = 1;
2713 set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2714 }
2715
2716 if (di->d_fieldmask & QC_SPC_TIMER) {
2717 dm->dqb_btime = di->d_spc_timer;
2718 check_blim = 1;
2719 set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2720 }
2721
2722 if (di->d_fieldmask & QC_INO_TIMER) {
2723 dm->dqb_itime = di->d_ino_timer;
2724 check_ilim = 1;
2725 set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2726 }
2727
2728 if (check_blim) {
2729 if (!dm->dqb_bsoftlimit ||
2730 dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2731 dm->dqb_btime = 0;
2732 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2733 } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2734 /* Set grace only if user hasn't provided his own... */
2735 dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2736 }
2737 if (check_ilim) {
2738 if (!dm->dqb_isoftlimit ||
2739 dm->dqb_curinodes < dm->dqb_isoftlimit) {
2740 dm->dqb_itime = 0;
2741 clear_bit(DQ_INODES_B, &dquot->dq_flags);
2742 } else if (!(di->d_fieldmask & QC_INO_TIMER))
2743 /* Set grace only if user hasn't provided his own... */
2744 dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2745 }
2746 if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2747 dm->dqb_isoftlimit)
2748 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2749 else
2750 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2751 spin_unlock(&dquot->dq_dqb_lock);
2752 mark_dquot_dirty(dquot);
2753
2754 return 0;
2755 }
2756
2757 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2758 struct qc_dqblk *di)
2759 {
2760 struct dquot *dquot;
2761 int rc;
2762
2763 dquot = dqget(sb, qid);
2764 if (IS_ERR(dquot)) {
2765 rc = PTR_ERR(dquot);
2766 goto out;
2767 }
2768 rc = do_set_dqblk(dquot, di);
2769 dqput(dquot);
2770 out:
2771 return rc;
2772 }
2773 EXPORT_SYMBOL(dquot_set_dqblk);
2774
2775 /* Generic routine for getting common part of quota file information */
2776 int dquot_get_state(struct super_block *sb, struct qc_state *state)
2777 {
2778 struct mem_dqinfo *mi;
2779 struct qc_type_state *tstate;
2780 struct quota_info *dqopt = sb_dqopt(sb);
2781 int type;
2782
2783 memset(state, 0, sizeof(*state));
2784 for (type = 0; type < MAXQUOTAS; type++) {
2785 if (!sb_has_quota_active(sb, type))
2786 continue;
2787 tstate = state->s_state + type;
2788 mi = sb_dqopt(sb)->info + type;
2789 tstate->flags = QCI_ACCT_ENABLED;
2790 spin_lock(&dq_data_lock);
2791 if (mi->dqi_flags & DQF_SYS_FILE)
2792 tstate->flags |= QCI_SYSFILE;
2793 if (mi->dqi_flags & DQF_ROOT_SQUASH)
2794 tstate->flags |= QCI_ROOT_SQUASH;
2795 if (sb_has_quota_limits_enabled(sb, type))
2796 tstate->flags |= QCI_LIMITS_ENFORCED;
2797 tstate->spc_timelimit = mi->dqi_bgrace;
2798 tstate->ino_timelimit = mi->dqi_igrace;
2799 tstate->ino = dqopt->files[type]->i_ino;
2800 tstate->blocks = dqopt->files[type]->i_blocks;
2801 tstate->nextents = 1; /* We don't know... */
2802 spin_unlock(&dq_data_lock);
2803 }
2804 return 0;
2805 }
2806 EXPORT_SYMBOL(dquot_get_state);
2807
2808 /* Generic routine for setting common part of quota file information */
2809 int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2810 {
2811 struct mem_dqinfo *mi;
2812 int err = 0;
2813
2814 if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2815 (ii->i_fieldmask & QC_RT_SPC_TIMER))
2816 return -EINVAL;
2817 if (!sb_has_quota_active(sb, type))
2818 return -ESRCH;
2819 mi = sb_dqopt(sb)->info + type;
2820 if (ii->i_fieldmask & QC_FLAGS) {
2821 if ((ii->i_flags & QCI_ROOT_SQUASH &&
2822 mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2823 return -EINVAL;
2824 }
2825 spin_lock(&dq_data_lock);
2826 if (ii->i_fieldmask & QC_SPC_TIMER)
2827 mi->dqi_bgrace = ii->i_spc_timelimit;
2828 if (ii->i_fieldmask & QC_INO_TIMER)
2829 mi->dqi_igrace = ii->i_ino_timelimit;
2830 if (ii->i_fieldmask & QC_FLAGS) {
2831 if (ii->i_flags & QCI_ROOT_SQUASH)
2832 mi->dqi_flags |= DQF_ROOT_SQUASH;
2833 else
2834 mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2835 }
2836 spin_unlock(&dq_data_lock);
2837 mark_info_dirty(sb, type);
2838 /* Force write to disk */
2839 sb->dq_op->write_info(sb, type);
2840 return err;
2841 }
2842 EXPORT_SYMBOL(dquot_set_dqinfo);
2843
2844 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2845 .quota_enable = dquot_quota_enable,
2846 .quota_disable = dquot_quota_disable,
2847 .quota_sync = dquot_quota_sync,
2848 .get_state = dquot_get_state,
2849 .set_info = dquot_set_dqinfo,
2850 .get_dqblk = dquot_get_dqblk,
2851 .get_nextdqblk = dquot_get_next_dqblk,
2852 .set_dqblk = dquot_set_dqblk
2853 };
2854 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2855
2856 static int do_proc_dqstats(struct ctl_table *table, int write,
2857 void __user *buffer, size_t *lenp, loff_t *ppos)
2858 {
2859 unsigned int type = (int *)table->data - dqstats.stat;
2860
2861 /* Update global table */
2862 dqstats.stat[type] =
2863 percpu_counter_sum_positive(&dqstats.counter[type]);
2864 return proc_dointvec(table, write, buffer, lenp, ppos);
2865 }
2866
2867 static struct ctl_table fs_dqstats_table[] = {
2868 {
2869 .procname = "lookups",
2870 .data = &dqstats.stat[DQST_LOOKUPS],
2871 .maxlen = sizeof(int),
2872 .mode = 0444,
2873 .proc_handler = do_proc_dqstats,
2874 },
2875 {
2876 .procname = "drops",
2877 .data = &dqstats.stat[DQST_DROPS],
2878 .maxlen = sizeof(int),
2879 .mode = 0444,
2880 .proc_handler = do_proc_dqstats,
2881 },
2882 {
2883 .procname = "reads",
2884 .data = &dqstats.stat[DQST_READS],
2885 .maxlen = sizeof(int),
2886 .mode = 0444,
2887 .proc_handler = do_proc_dqstats,
2888 },
2889 {
2890 .procname = "writes",
2891 .data = &dqstats.stat[DQST_WRITES],
2892 .maxlen = sizeof(int),
2893 .mode = 0444,
2894 .proc_handler = do_proc_dqstats,
2895 },
2896 {
2897 .procname = "cache_hits",
2898 .data = &dqstats.stat[DQST_CACHE_HITS],
2899 .maxlen = sizeof(int),
2900 .mode = 0444,
2901 .proc_handler = do_proc_dqstats,
2902 },
2903 {
2904 .procname = "allocated_dquots",
2905 .data = &dqstats.stat[DQST_ALLOC_DQUOTS],
2906 .maxlen = sizeof(int),
2907 .mode = 0444,
2908 .proc_handler = do_proc_dqstats,
2909 },
2910 {
2911 .procname = "free_dquots",
2912 .data = &dqstats.stat[DQST_FREE_DQUOTS],
2913 .maxlen = sizeof(int),
2914 .mode = 0444,
2915 .proc_handler = do_proc_dqstats,
2916 },
2917 {
2918 .procname = "syncs",
2919 .data = &dqstats.stat[DQST_SYNCS],
2920 .maxlen = sizeof(int),
2921 .mode = 0444,
2922 .proc_handler = do_proc_dqstats,
2923 },
2924 #ifdef CONFIG_PRINT_QUOTA_WARNING
2925 {
2926 .procname = "warnings",
2927 .data = &flag_print_warnings,
2928 .maxlen = sizeof(int),
2929 .mode = 0644,
2930 .proc_handler = proc_dointvec,
2931 },
2932 #endif
2933 { },
2934 };
2935
2936 static struct ctl_table fs_table[] = {
2937 {
2938 .procname = "quota",
2939 .mode = 0555,
2940 .child = fs_dqstats_table,
2941 },
2942 { },
2943 };
2944
2945 static struct ctl_table sys_table[] = {
2946 {
2947 .procname = "fs",
2948 .mode = 0555,
2949 .child = fs_table,
2950 },
2951 { },
2952 };
2953
2954 static int __init dquot_init(void)
2955 {
2956 int i, ret;
2957 unsigned long nr_hash, order;
2958
2959 printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2960
2961 register_sysctl_table(sys_table);
2962
2963 dquot_cachep = kmem_cache_create("dquot",
2964 sizeof(struct dquot), sizeof(unsigned long) * 4,
2965 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2966 SLAB_MEM_SPREAD|SLAB_PANIC),
2967 NULL);
2968
2969 order = 0;
2970 dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2971 if (!dquot_hash)
2972 panic("Cannot create dquot hash table");
2973
2974 for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2975 ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2976 if (ret)
2977 panic("Cannot create dquot stat counters");
2978 }
2979
2980 /* Find power-of-two hlist_heads which can fit into allocation */
2981 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2982 dq_hash_bits = 0;
2983 do {
2984 dq_hash_bits++;
2985 } while (nr_hash >> dq_hash_bits);
2986 dq_hash_bits--;
2987
2988 nr_hash = 1UL << dq_hash_bits;
2989 dq_hash_mask = nr_hash - 1;
2990 for (i = 0; i < nr_hash; i++)
2991 INIT_HLIST_HEAD(dquot_hash + i);
2992
2993 pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2994 " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2995
2996 if (register_shrinker(&dqcache_shrinker))
2997 panic("Cannot register dquot shrinker");
2998
2999 return 0;
3000 }
3001 fs_initcall(dquot_init);