xref: /openbmc/linux/fs/quota/dquot.c (revision afb46f79)
1 /*
2  * Implementation of the diskquota system for the LINUX operating system. QUOTA
3  * is implemented using the BSD system call interface as the means of
4  * communication with the user level. This file contains the generic routines
5  * called by the different filesystems on allocation of an inode or block.
6  * These routines take care of the administration needed to have a consistent
7  * diskquota tracking system. The ideas of both user and group quotas are based
8  * on the Melbourne quota system as used on BSD derived systems. The internal
9  * implementation is based on one of the several variants of the LINUX
10  * inode-subsystem with added complexity of the diskquota system.
11  *
12  * Author:	Marco van Wieringen <mvw@planets.elm.net>
13  *
14  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
15  *
16  *		Revised list management to avoid races
17  *		-- Bill Hawes, <whawes@star.net>, 9/98
18  *
19  *		Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
20  *		As the consequence the locking was moved from dquot_decr_...(),
21  *		dquot_incr_...() to calling functions.
22  *		invalidate_dquots() now writes modified dquots.
23  *		Serialized quota_off() and quota_on() for mount point.
24  *		Fixed a few bugs in grow_dquots().
25  *		Fixed deadlock in write_dquot() - we no longer account quotas on
26  *		quota files
27  *		remove_dquot_ref() moved to inode.c - it now traverses through inodes
28  *		add_dquot_ref() restarts after blocking
29  *		Added check for bogus uid and fixed check for group in quotactl.
30  *		Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
31  *
32  *		Used struct list_head instead of own list struct
33  *		Invalidation of referenced dquots is no longer possible
34  *		Improved free_dquots list management
35  *		Quota and i_blocks are now updated in one place to avoid races
36  *		Warnings are now delayed so we won't block in critical section
37  *		Write updated not to require dquot lock
38  *		Jan Kara, <jack@suse.cz>, 9/2000
39  *
40  *		Added dynamic quota structure allocation
41  *		Jan Kara <jack@suse.cz> 12/2000
42  *
43  *		Rewritten quota interface. Implemented new quota format and
44  *		formats registering.
45  *		Jan Kara, <jack@suse.cz>, 2001,2002
46  *
47  *		New SMP locking.
48  *		Jan Kara, <jack@suse.cz>, 10/2002
49  *
50  *		Added journalled quota support, fix lock inversion problems
51  *		Jan Kara, <jack@suse.cz>, 2003,2004
52  *
53  * (C) Copyright 1994 - 1997 Marco van Wieringen
54  */
55 
56 #include <linux/errno.h>
57 #include <linux/kernel.h>
58 #include <linux/fs.h>
59 #include <linux/mount.h>
60 #include <linux/mm.h>
61 #include <linux/time.h>
62 #include <linux/types.h>
63 #include <linux/string.h>
64 #include <linux/fcntl.h>
65 #include <linux/stat.h>
66 #include <linux/tty.h>
67 #include <linux/file.h>
68 #include <linux/slab.h>
69 #include <linux/sysctl.h>
70 #include <linux/init.h>
71 #include <linux/module.h>
72 #include <linux/proc_fs.h>
73 #include <linux/security.h>
74 #include <linux/sched.h>
75 #include <linux/kmod.h>
76 #include <linux/namei.h>
77 #include <linux/capability.h>
78 #include <linux/quotaops.h>
79 #include "../internal.h" /* ugh */
80 
81 #include <linux/uaccess.h>
82 
83 /*
84  * There are three quota SMP locks. dq_list_lock protects all lists with quotas
85  * and quota formats.
86  * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
87  * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
88  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
89  * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
90  * modifications of quota state (on quotaon and quotaoff) and readers who care
91  * about latest values take it as well.
92  *
93  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
94  *   dq_list_lock > dq_state_lock
95  *
96  * Note that some things (eg. sb pointer, type, id) doesn't change during
97  * the life of the dquot structure and so needn't to be protected by a lock
98  *
99  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
100  * operation is just reading pointers from inode (or not using them at all) the
101  * read lock is enough. If pointers are altered function must hold write lock.
102  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
103  * inode is a quota file). Functions adding pointers from inode to dquots have
104  * to check this flag under dqptr_sem and then (if S_NOQUOTA is not set) they
105  * have to do all pointer modifications before dropping dqptr_sem. This makes
106  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
107  * then drops all pointers to dquots from an inode.
108  *
109  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
110  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
111  * Currently dquot is locked only when it is being read to memory (or space for
112  * it is being allocated) on the first dqget() and when it is being released on
113  * the last dqput(). The allocation and release oparations are serialized by
114  * the dq_lock and by checking the use count in dquot_release().  Write
115  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
116  * spinlock to internal buffers before writing.
117  *
118  * Lock ordering (including related VFS locks) is the following:
119  *   dqonoff_mutex > i_mutex > journal_lock > dqptr_sem > dquot->dq_lock >
120  *   dqio_mutex
121  * dqonoff_mutex > i_mutex comes from dquot_quota_sync, dquot_enable, etc.
122  * The lock ordering of dqptr_sem imposed by quota code is only dqonoff_sem >
123  * dqptr_sem. But filesystem has to count with the fact that functions such as
124  * dquot_alloc_space() acquire dqptr_sem and they usually have to be called
125  * from inside a transaction to keep filesystem consistency after a crash. Also
126  * filesystems usually want to do some IO on dquot from ->mark_dirty which is
127  * called with dqptr_sem held.
128  */
129 
130 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
131 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
132 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
133 EXPORT_SYMBOL(dq_data_lock);
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 void __dquot_initialize(struct inode *inode, int type);
254 
255 static inline unsigned int
256 hashfn(const struct super_block *sb, struct kqid qid)
257 {
258 	unsigned int id = from_kqid(&init_user_ns, qid);
259 	int type = qid.type;
260 	unsigned long tmp;
261 
262 	tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
263 	return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
264 }
265 
266 /*
267  * Following list functions expect dq_list_lock to be held
268  */
269 static inline void insert_dquot_hash(struct dquot *dquot)
270 {
271 	struct hlist_head *head;
272 	head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
273 	hlist_add_head(&dquot->dq_hash, head);
274 }
275 
276 static inline void remove_dquot_hash(struct dquot *dquot)
277 {
278 	hlist_del_init(&dquot->dq_hash);
279 }
280 
281 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
282 				struct kqid qid)
283 {
284 	struct hlist_node *node;
285 	struct dquot *dquot;
286 
287 	hlist_for_each (node, dquot_hash+hashent) {
288 		dquot = hlist_entry(node, struct dquot, dq_hash);
289 		if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
290 			return dquot;
291 	}
292 	return NULL;
293 }
294 
295 /* Add a dquot to the tail of the free list */
296 static inline void put_dquot_last(struct dquot *dquot)
297 {
298 	list_add_tail(&dquot->dq_free, &free_dquots);
299 	dqstats_inc(DQST_FREE_DQUOTS);
300 }
301 
302 static inline void remove_free_dquot(struct dquot *dquot)
303 {
304 	if (list_empty(&dquot->dq_free))
305 		return;
306 	list_del_init(&dquot->dq_free);
307 	dqstats_dec(DQST_FREE_DQUOTS);
308 }
309 
310 static inline void put_inuse(struct dquot *dquot)
311 {
312 	/* We add to the back of inuse list so we don't have to restart
313 	 * when traversing this list and we block */
314 	list_add_tail(&dquot->dq_inuse, &inuse_list);
315 	dqstats_inc(DQST_ALLOC_DQUOTS);
316 }
317 
318 static inline void remove_inuse(struct dquot *dquot)
319 {
320 	dqstats_dec(DQST_ALLOC_DQUOTS);
321 	list_del(&dquot->dq_inuse);
322 }
323 /*
324  * End of list functions needing dq_list_lock
325  */
326 
327 static void wait_on_dquot(struct dquot *dquot)
328 {
329 	mutex_lock(&dquot->dq_lock);
330 	mutex_unlock(&dquot->dq_lock);
331 }
332 
333 static inline int dquot_dirty(struct dquot *dquot)
334 {
335 	return test_bit(DQ_MOD_B, &dquot->dq_flags);
336 }
337 
338 static inline int mark_dquot_dirty(struct dquot *dquot)
339 {
340 	return dquot->dq_sb->dq_op->mark_dirty(dquot);
341 }
342 
343 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
344 int dquot_mark_dquot_dirty(struct dquot *dquot)
345 {
346 	int ret = 1;
347 
348 	/* If quota is dirty already, we don't have to acquire dq_list_lock */
349 	if (test_bit(DQ_MOD_B, &dquot->dq_flags))
350 		return 1;
351 
352 	spin_lock(&dq_list_lock);
353 	if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
354 		list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
355 				info[dquot->dq_id.type].dqi_dirty_list);
356 		ret = 0;
357 	}
358 	spin_unlock(&dq_list_lock);
359 	return ret;
360 }
361 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
362 
363 /* Dirtify all the dquots - this can block when journalling */
364 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
365 {
366 	int ret, err, cnt;
367 
368 	ret = err = 0;
369 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
370 		if (dquot[cnt])
371 			/* Even in case of error we have to continue */
372 			ret = mark_dquot_dirty(dquot[cnt]);
373 		if (!err)
374 			err = ret;
375 	}
376 	return err;
377 }
378 
379 static inline void dqput_all(struct dquot **dquot)
380 {
381 	unsigned int cnt;
382 
383 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
384 		dqput(dquot[cnt]);
385 }
386 
387 /* This function needs dq_list_lock */
388 static inline int clear_dquot_dirty(struct dquot *dquot)
389 {
390 	if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
391 		return 0;
392 	list_del_init(&dquot->dq_dirty);
393 	return 1;
394 }
395 
396 void mark_info_dirty(struct super_block *sb, int type)
397 {
398 	set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
399 }
400 EXPORT_SYMBOL(mark_info_dirty);
401 
402 /*
403  *	Read dquot from disk and alloc space for it
404  */
405 
406 int dquot_acquire(struct dquot *dquot)
407 {
408 	int ret = 0, ret2 = 0;
409 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
410 
411 	mutex_lock(&dquot->dq_lock);
412 	mutex_lock(&dqopt->dqio_mutex);
413 	if (!test_bit(DQ_READ_B, &dquot->dq_flags))
414 		ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
415 	if (ret < 0)
416 		goto out_iolock;
417 	set_bit(DQ_READ_B, &dquot->dq_flags);
418 	/* Instantiate dquot if needed */
419 	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
420 		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
421 		/* Write the info if needed */
422 		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
423 			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
424 					dquot->dq_sb, dquot->dq_id.type);
425 		}
426 		if (ret < 0)
427 			goto out_iolock;
428 		if (ret2 < 0) {
429 			ret = ret2;
430 			goto out_iolock;
431 		}
432 	}
433 	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
434 out_iolock:
435 	mutex_unlock(&dqopt->dqio_mutex);
436 	mutex_unlock(&dquot->dq_lock);
437 	return ret;
438 }
439 EXPORT_SYMBOL(dquot_acquire);
440 
441 /*
442  *	Write dquot to disk
443  */
444 int dquot_commit(struct dquot *dquot)
445 {
446 	int ret = 0;
447 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
448 
449 	mutex_lock(&dqopt->dqio_mutex);
450 	spin_lock(&dq_list_lock);
451 	if (!clear_dquot_dirty(dquot)) {
452 		spin_unlock(&dq_list_lock);
453 		goto out_sem;
454 	}
455 	spin_unlock(&dq_list_lock);
456 	/* Inactive dquot can be only if there was error during read/init
457 	 * => we have better not writing it */
458 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
459 		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
460 	else
461 		ret = -EIO;
462 out_sem:
463 	mutex_unlock(&dqopt->dqio_mutex);
464 	return ret;
465 }
466 EXPORT_SYMBOL(dquot_commit);
467 
468 /*
469  *	Release dquot
470  */
471 int dquot_release(struct dquot *dquot)
472 {
473 	int ret = 0, ret2 = 0;
474 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
475 
476 	mutex_lock(&dquot->dq_lock);
477 	/* Check whether we are not racing with some other dqget() */
478 	if (atomic_read(&dquot->dq_count) > 1)
479 		goto out_dqlock;
480 	mutex_lock(&dqopt->dqio_mutex);
481 	if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
482 		ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
483 		/* Write the info */
484 		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
485 			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
486 						dquot->dq_sb, dquot->dq_id.type);
487 		}
488 		if (ret >= 0)
489 			ret = ret2;
490 	}
491 	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
492 	mutex_unlock(&dqopt->dqio_mutex);
493 out_dqlock:
494 	mutex_unlock(&dquot->dq_lock);
495 	return ret;
496 }
497 EXPORT_SYMBOL(dquot_release);
498 
499 void dquot_destroy(struct dquot *dquot)
500 {
501 	kmem_cache_free(dquot_cachep, dquot);
502 }
503 EXPORT_SYMBOL(dquot_destroy);
504 
505 static inline void do_destroy_dquot(struct dquot *dquot)
506 {
507 	dquot->dq_sb->dq_op->destroy_dquot(dquot);
508 }
509 
510 /* Invalidate all dquots on the list. Note that this function is called after
511  * quota is disabled and pointers from inodes removed so there cannot be new
512  * quota users. There can still be some users of quotas due to inodes being
513  * just deleted or pruned by prune_icache() (those are not attached to any
514  * list) or parallel quotactl call. We have to wait for such users.
515  */
516 static void invalidate_dquots(struct super_block *sb, int type)
517 {
518 	struct dquot *dquot, *tmp;
519 
520 restart:
521 	spin_lock(&dq_list_lock);
522 	list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
523 		if (dquot->dq_sb != sb)
524 			continue;
525 		if (dquot->dq_id.type != type)
526 			continue;
527 		/* Wait for dquot users */
528 		if (atomic_read(&dquot->dq_count)) {
529 			DEFINE_WAIT(wait);
530 
531 			dqgrab(dquot);
532 			prepare_to_wait(&dquot->dq_wait_unused, &wait,
533 					TASK_UNINTERRUPTIBLE);
534 			spin_unlock(&dq_list_lock);
535 			/* Once dqput() wakes us up, we know it's time to free
536 			 * the dquot.
537 			 * IMPORTANT: we rely on the fact that there is always
538 			 * at most one process waiting for dquot to free.
539 			 * Otherwise dq_count would be > 1 and we would never
540 			 * wake up.
541 			 */
542 			if (atomic_read(&dquot->dq_count) > 1)
543 				schedule();
544 			finish_wait(&dquot->dq_wait_unused, &wait);
545 			dqput(dquot);
546 			/* At this moment dquot() need not exist (it could be
547 			 * reclaimed by prune_dqcache(). Hence we must
548 			 * restart. */
549 			goto restart;
550 		}
551 		/*
552 		 * Quota now has no users and it has been written on last
553 		 * dqput()
554 		 */
555 		remove_dquot_hash(dquot);
556 		remove_free_dquot(dquot);
557 		remove_inuse(dquot);
558 		do_destroy_dquot(dquot);
559 	}
560 	spin_unlock(&dq_list_lock);
561 }
562 
563 /* Call callback for every active dquot on given filesystem */
564 int dquot_scan_active(struct super_block *sb,
565 		      int (*fn)(struct dquot *dquot, unsigned long priv),
566 		      unsigned long priv)
567 {
568 	struct dquot *dquot, *old_dquot = NULL;
569 	int ret = 0;
570 
571 	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
572 	spin_lock(&dq_list_lock);
573 	list_for_each_entry(dquot, &inuse_list, dq_inuse) {
574 		if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
575 			continue;
576 		if (dquot->dq_sb != sb)
577 			continue;
578 		/* Now we have active dquot so we can just increase use count */
579 		atomic_inc(&dquot->dq_count);
580 		spin_unlock(&dq_list_lock);
581 		dqstats_inc(DQST_LOOKUPS);
582 		dqput(old_dquot);
583 		old_dquot = dquot;
584 		/*
585 		 * ->release_dquot() can be racing with us. Our reference
586 		 * protects us from new calls to it so just wait for any
587 		 * outstanding call and recheck the DQ_ACTIVE_B after that.
588 		 */
589 		wait_on_dquot(dquot);
590 		if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
591 			ret = fn(dquot, priv);
592 			if (ret < 0)
593 				goto out;
594 		}
595 		spin_lock(&dq_list_lock);
596 		/* We are safe to continue now because our dquot could not
597 		 * be moved out of the inuse list while we hold the reference */
598 	}
599 	spin_unlock(&dq_list_lock);
600 out:
601 	dqput(old_dquot);
602 	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
603 	return ret;
604 }
605 EXPORT_SYMBOL(dquot_scan_active);
606 
607 /* Write all dquot structures to quota files */
608 int dquot_writeback_dquots(struct super_block *sb, int type)
609 {
610 	struct list_head *dirty;
611 	struct dquot *dquot;
612 	struct quota_info *dqopt = sb_dqopt(sb);
613 	int cnt;
614 	int err, ret = 0;
615 
616 	mutex_lock(&dqopt->dqonoff_mutex);
617 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
618 		if (type != -1 && cnt != type)
619 			continue;
620 		if (!sb_has_quota_active(sb, cnt))
621 			continue;
622 		spin_lock(&dq_list_lock);
623 		dirty = &dqopt->info[cnt].dqi_dirty_list;
624 		while (!list_empty(dirty)) {
625 			dquot = list_first_entry(dirty, struct dquot,
626 						 dq_dirty);
627 			/* Dirty and inactive can be only bad dquot... */
628 			if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
629 				clear_dquot_dirty(dquot);
630 				continue;
631 			}
632 			/* Now we have active dquot from which someone is
633  			 * holding reference so we can safely just increase
634 			 * use count */
635 			dqgrab(dquot);
636 			spin_unlock(&dq_list_lock);
637 			dqstats_inc(DQST_LOOKUPS);
638 			err = sb->dq_op->write_dquot(dquot);
639 			if (!ret && err)
640 				err = ret;
641 			dqput(dquot);
642 			spin_lock(&dq_list_lock);
643 		}
644 		spin_unlock(&dq_list_lock);
645 	}
646 
647 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
648 		if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
649 		    && info_dirty(&dqopt->info[cnt]))
650 			sb->dq_op->write_info(sb, cnt);
651 	dqstats_inc(DQST_SYNCS);
652 	mutex_unlock(&dqopt->dqonoff_mutex);
653 
654 	return ret;
655 }
656 EXPORT_SYMBOL(dquot_writeback_dquots);
657 
658 /* Write all dquot structures to disk and make them visible from userspace */
659 int dquot_quota_sync(struct super_block *sb, int type)
660 {
661 	struct quota_info *dqopt = sb_dqopt(sb);
662 	int cnt;
663 	int ret;
664 
665 	ret = dquot_writeback_dquots(sb, type);
666 	if (ret)
667 		return ret;
668 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
669 		return 0;
670 
671 	/* This is not very clever (and fast) but currently I don't know about
672 	 * any other simple way of getting quota data to disk and we must get
673 	 * them there for userspace to be visible... */
674 	if (sb->s_op->sync_fs)
675 		sb->s_op->sync_fs(sb, 1);
676 	sync_blockdev(sb->s_bdev);
677 
678 	/*
679 	 * Now when everything is written we can discard the pagecache so
680 	 * that userspace sees the changes.
681 	 */
682 	mutex_lock(&dqopt->dqonoff_mutex);
683 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
684 		if (type != -1 && cnt != type)
685 			continue;
686 		if (!sb_has_quota_active(sb, cnt))
687 			continue;
688 		mutex_lock(&dqopt->files[cnt]->i_mutex);
689 		truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
690 		mutex_unlock(&dqopt->files[cnt]->i_mutex);
691 	}
692 	mutex_unlock(&dqopt->dqonoff_mutex);
693 
694 	return 0;
695 }
696 EXPORT_SYMBOL(dquot_quota_sync);
697 
698 static unsigned long
699 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
700 {
701 	struct list_head *head;
702 	struct dquot *dquot;
703 	unsigned long freed = 0;
704 
705 	head = free_dquots.prev;
706 	while (head != &free_dquots && sc->nr_to_scan) {
707 		dquot = list_entry(head, struct dquot, dq_free);
708 		remove_dquot_hash(dquot);
709 		remove_free_dquot(dquot);
710 		remove_inuse(dquot);
711 		do_destroy_dquot(dquot);
712 		sc->nr_to_scan--;
713 		freed++;
714 		head = free_dquots.prev;
715 	}
716 	return freed;
717 }
718 
719 static unsigned long
720 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
721 {
722 	return vfs_pressure_ratio(
723 	percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
724 }
725 
726 static struct shrinker dqcache_shrinker = {
727 	.count_objects = dqcache_shrink_count,
728 	.scan_objects = dqcache_shrink_scan,
729 	.seeks = DEFAULT_SEEKS,
730 };
731 
732 /*
733  * Put reference to dquot
734  * NOTE: If you change this function please check whether dqput_blocks() works right...
735  */
736 void dqput(struct dquot *dquot)
737 {
738 	int ret;
739 
740 	if (!dquot)
741 		return;
742 #ifdef CONFIG_QUOTA_DEBUG
743 	if (!atomic_read(&dquot->dq_count)) {
744 		quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
745 			    quotatypes[dquot->dq_id.type],
746 			    from_kqid(&init_user_ns, dquot->dq_id));
747 		BUG();
748 	}
749 #endif
750 	dqstats_inc(DQST_DROPS);
751 we_slept:
752 	spin_lock(&dq_list_lock);
753 	if (atomic_read(&dquot->dq_count) > 1) {
754 		/* We have more than one user... nothing to do */
755 		atomic_dec(&dquot->dq_count);
756 		/* Releasing dquot during quotaoff phase? */
757 		if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
758 		    atomic_read(&dquot->dq_count) == 1)
759 			wake_up(&dquot->dq_wait_unused);
760 		spin_unlock(&dq_list_lock);
761 		return;
762 	}
763 	/* Need to release dquot? */
764 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
765 		spin_unlock(&dq_list_lock);
766 		/* Commit dquot before releasing */
767 		ret = dquot->dq_sb->dq_op->write_dquot(dquot);
768 		if (ret < 0) {
769 			quota_error(dquot->dq_sb, "Can't write quota structure"
770 				    " (error %d). Quota may get out of sync!",
771 				    ret);
772 			/*
773 			 * We clear dirty bit anyway, so that we avoid
774 			 * infinite loop here
775 			 */
776 			spin_lock(&dq_list_lock);
777 			clear_dquot_dirty(dquot);
778 			spin_unlock(&dq_list_lock);
779 		}
780 		goto we_slept;
781 	}
782 	/* Clear flag in case dquot was inactive (something bad happened) */
783 	clear_dquot_dirty(dquot);
784 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
785 		spin_unlock(&dq_list_lock);
786 		dquot->dq_sb->dq_op->release_dquot(dquot);
787 		goto we_slept;
788 	}
789 	atomic_dec(&dquot->dq_count);
790 #ifdef CONFIG_QUOTA_DEBUG
791 	/* sanity check */
792 	BUG_ON(!list_empty(&dquot->dq_free));
793 #endif
794 	put_dquot_last(dquot);
795 	spin_unlock(&dq_list_lock);
796 }
797 EXPORT_SYMBOL(dqput);
798 
799 struct dquot *dquot_alloc(struct super_block *sb, int type)
800 {
801 	return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
802 }
803 EXPORT_SYMBOL(dquot_alloc);
804 
805 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
806 {
807 	struct dquot *dquot;
808 
809 	dquot = sb->dq_op->alloc_dquot(sb, type);
810 	if(!dquot)
811 		return NULL;
812 
813 	mutex_init(&dquot->dq_lock);
814 	INIT_LIST_HEAD(&dquot->dq_free);
815 	INIT_LIST_HEAD(&dquot->dq_inuse);
816 	INIT_HLIST_NODE(&dquot->dq_hash);
817 	INIT_LIST_HEAD(&dquot->dq_dirty);
818 	init_waitqueue_head(&dquot->dq_wait_unused);
819 	dquot->dq_sb = sb;
820 	dquot->dq_id = make_kqid_invalid(type);
821 	atomic_set(&dquot->dq_count, 1);
822 
823 	return dquot;
824 }
825 
826 /*
827  * Get reference to dquot
828  *
829  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
830  * destroying our dquot by:
831  *   a) checking for quota flags under dq_list_lock and
832  *   b) getting a reference to dquot before we release dq_list_lock
833  */
834 struct dquot *dqget(struct super_block *sb, struct kqid qid)
835 {
836 	unsigned int hashent = hashfn(sb, qid);
837 	struct dquot *dquot = NULL, *empty = NULL;
838 
839         if (!sb_has_quota_active(sb, qid.type))
840 		return NULL;
841 we_slept:
842 	spin_lock(&dq_list_lock);
843 	spin_lock(&dq_state_lock);
844 	if (!sb_has_quota_active(sb, qid.type)) {
845 		spin_unlock(&dq_state_lock);
846 		spin_unlock(&dq_list_lock);
847 		goto out;
848 	}
849 	spin_unlock(&dq_state_lock);
850 
851 	dquot = find_dquot(hashent, sb, qid);
852 	if (!dquot) {
853 		if (!empty) {
854 			spin_unlock(&dq_list_lock);
855 			empty = get_empty_dquot(sb, qid.type);
856 			if (!empty)
857 				schedule();	/* Try to wait for a moment... */
858 			goto we_slept;
859 		}
860 		dquot = empty;
861 		empty = NULL;
862 		dquot->dq_id = qid;
863 		/* all dquots go on the inuse_list */
864 		put_inuse(dquot);
865 		/* hash it first so it can be found */
866 		insert_dquot_hash(dquot);
867 		spin_unlock(&dq_list_lock);
868 		dqstats_inc(DQST_LOOKUPS);
869 	} else {
870 		if (!atomic_read(&dquot->dq_count))
871 			remove_free_dquot(dquot);
872 		atomic_inc(&dquot->dq_count);
873 		spin_unlock(&dq_list_lock);
874 		dqstats_inc(DQST_CACHE_HITS);
875 		dqstats_inc(DQST_LOOKUPS);
876 	}
877 	/* Wait for dq_lock - after this we know that either dquot_release() is
878 	 * already finished or it will be canceled due to dq_count > 1 test */
879 	wait_on_dquot(dquot);
880 	/* Read the dquot / allocate space in quota file */
881 	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
882 	    sb->dq_op->acquire_dquot(dquot) < 0) {
883 		dqput(dquot);
884 		dquot = NULL;
885 		goto out;
886 	}
887 #ifdef CONFIG_QUOTA_DEBUG
888 	BUG_ON(!dquot->dq_sb);	/* Has somebody invalidated entry under us? */
889 #endif
890 out:
891 	if (empty)
892 		do_destroy_dquot(empty);
893 
894 	return dquot;
895 }
896 EXPORT_SYMBOL(dqget);
897 
898 static int dqinit_needed(struct inode *inode, int type)
899 {
900 	int cnt;
901 
902 	if (IS_NOQUOTA(inode))
903 		return 0;
904 	if (type != -1)
905 		return !inode->i_dquot[type];
906 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
907 		if (!inode->i_dquot[cnt])
908 			return 1;
909 	return 0;
910 }
911 
912 /* This routine is guarded by dqonoff_mutex mutex */
913 static void add_dquot_ref(struct super_block *sb, int type)
914 {
915 	struct inode *inode, *old_inode = NULL;
916 #ifdef CONFIG_QUOTA_DEBUG
917 	int reserved = 0;
918 #endif
919 
920 	spin_lock(&inode_sb_list_lock);
921 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
922 		spin_lock(&inode->i_lock);
923 		if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
924 		    !atomic_read(&inode->i_writecount) ||
925 		    !dqinit_needed(inode, type)) {
926 			spin_unlock(&inode->i_lock);
927 			continue;
928 		}
929 		__iget(inode);
930 		spin_unlock(&inode->i_lock);
931 		spin_unlock(&inode_sb_list_lock);
932 
933 #ifdef CONFIG_QUOTA_DEBUG
934 		if (unlikely(inode_get_rsv_space(inode) > 0))
935 			reserved = 1;
936 #endif
937 		iput(old_inode);
938 		__dquot_initialize(inode, type);
939 
940 		/*
941 		 * We hold a reference to 'inode' so it couldn't have been
942 		 * removed from s_inodes list while we dropped the
943 		 * inode_sb_list_lock We cannot iput the inode now as we can be
944 		 * holding the last reference and we cannot iput it under
945 		 * inode_sb_list_lock. So we keep the reference and iput it
946 		 * later.
947 		 */
948 		old_inode = inode;
949 		spin_lock(&inode_sb_list_lock);
950 	}
951 	spin_unlock(&inode_sb_list_lock);
952 	iput(old_inode);
953 
954 #ifdef CONFIG_QUOTA_DEBUG
955 	if (reserved) {
956 		quota_error(sb, "Writes happened before quota was turned on "
957 			"thus quota information is probably inconsistent. "
958 			"Please run quotacheck(8)");
959 	}
960 #endif
961 }
962 
963 /*
964  * Return 0 if dqput() won't block.
965  * (note that 1 doesn't necessarily mean blocking)
966  */
967 static inline int dqput_blocks(struct dquot *dquot)
968 {
969 	if (atomic_read(&dquot->dq_count) <= 1)
970 		return 1;
971 	return 0;
972 }
973 
974 /*
975  * Remove references to dquots from inode and add dquot to list for freeing
976  * if we have the last reference to dquot
977  * We can't race with anybody because we hold dqptr_sem for writing...
978  */
979 static int remove_inode_dquot_ref(struct inode *inode, int type,
980 				  struct list_head *tofree_head)
981 {
982 	struct dquot *dquot = inode->i_dquot[type];
983 
984 	inode->i_dquot[type] = NULL;
985 	if (dquot) {
986 		if (dqput_blocks(dquot)) {
987 #ifdef CONFIG_QUOTA_DEBUG
988 			if (atomic_read(&dquot->dq_count) != 1)
989 				quota_error(inode->i_sb, "Adding dquot with "
990 					    "dq_count %d to dispose list",
991 					    atomic_read(&dquot->dq_count));
992 #endif
993 			spin_lock(&dq_list_lock);
994 			/* As dquot must have currently users it can't be on
995 			 * the free list... */
996 			list_add(&dquot->dq_free, tofree_head);
997 			spin_unlock(&dq_list_lock);
998 			return 1;
999 		}
1000 		else
1001 			dqput(dquot);   /* We have guaranteed we won't block */
1002 	}
1003 	return 0;
1004 }
1005 
1006 /*
1007  * Free list of dquots
1008  * Dquots are removed from inodes and no new references can be got so we are
1009  * the only ones holding reference
1010  */
1011 static void put_dquot_list(struct list_head *tofree_head)
1012 {
1013 	struct list_head *act_head;
1014 	struct dquot *dquot;
1015 
1016 	act_head = tofree_head->next;
1017 	while (act_head != tofree_head) {
1018 		dquot = list_entry(act_head, struct dquot, dq_free);
1019 		act_head = act_head->next;
1020 		/* Remove dquot from the list so we won't have problems... */
1021 		list_del_init(&dquot->dq_free);
1022 		dqput(dquot);
1023 	}
1024 }
1025 
1026 static void remove_dquot_ref(struct super_block *sb, int type,
1027 		struct list_head *tofree_head)
1028 {
1029 	struct inode *inode;
1030 	int reserved = 0;
1031 
1032 	spin_lock(&inode_sb_list_lock);
1033 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1034 		/*
1035 		 *  We have to scan also I_NEW inodes because they can already
1036 		 *  have quota pointer initialized. Luckily, we need to touch
1037 		 *  only quota pointers and these have separate locking
1038 		 *  (dqptr_sem).
1039 		 */
1040 		if (!IS_NOQUOTA(inode)) {
1041 			if (unlikely(inode_get_rsv_space(inode) > 0))
1042 				reserved = 1;
1043 			remove_inode_dquot_ref(inode, type, tofree_head);
1044 		}
1045 	}
1046 	spin_unlock(&inode_sb_list_lock);
1047 #ifdef CONFIG_QUOTA_DEBUG
1048 	if (reserved) {
1049 		printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1050 			" was disabled thus quota information is probably "
1051 			"inconsistent. Please run quotacheck(8).\n", sb->s_id);
1052 	}
1053 #endif
1054 }
1055 
1056 /* Gather all references from inodes and drop them */
1057 static void drop_dquot_ref(struct super_block *sb, int type)
1058 {
1059 	LIST_HEAD(tofree_head);
1060 
1061 	if (sb->dq_op) {
1062 		down_write(&sb_dqopt(sb)->dqptr_sem);
1063 		remove_dquot_ref(sb, type, &tofree_head);
1064 		up_write(&sb_dqopt(sb)->dqptr_sem);
1065 		put_dquot_list(&tofree_head);
1066 	}
1067 }
1068 
1069 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1070 {
1071 	dquot->dq_dqb.dqb_curinodes += number;
1072 }
1073 
1074 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1075 {
1076 	dquot->dq_dqb.dqb_curspace += number;
1077 }
1078 
1079 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1080 {
1081 	dquot->dq_dqb.dqb_rsvspace += number;
1082 }
1083 
1084 /*
1085  * Claim reserved quota space
1086  */
1087 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1088 {
1089 	if (dquot->dq_dqb.dqb_rsvspace < number) {
1090 		WARN_ON_ONCE(1);
1091 		number = dquot->dq_dqb.dqb_rsvspace;
1092 	}
1093 	dquot->dq_dqb.dqb_curspace += number;
1094 	dquot->dq_dqb.dqb_rsvspace -= number;
1095 }
1096 
1097 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1098 {
1099 	if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1100 		number = dquot->dq_dqb.dqb_curspace;
1101 	dquot->dq_dqb.dqb_rsvspace += number;
1102 	dquot->dq_dqb.dqb_curspace -= number;
1103 }
1104 
1105 static inline
1106 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1107 {
1108 	if (dquot->dq_dqb.dqb_rsvspace >= number)
1109 		dquot->dq_dqb.dqb_rsvspace -= number;
1110 	else {
1111 		WARN_ON_ONCE(1);
1112 		dquot->dq_dqb.dqb_rsvspace = 0;
1113 	}
1114 }
1115 
1116 static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1117 {
1118 	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1119 	    dquot->dq_dqb.dqb_curinodes >= number)
1120 		dquot->dq_dqb.dqb_curinodes -= number;
1121 	else
1122 		dquot->dq_dqb.dqb_curinodes = 0;
1123 	if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1124 		dquot->dq_dqb.dqb_itime = (time_t) 0;
1125 	clear_bit(DQ_INODES_B, &dquot->dq_flags);
1126 }
1127 
1128 static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1129 {
1130 	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1131 	    dquot->dq_dqb.dqb_curspace >= number)
1132 		dquot->dq_dqb.dqb_curspace -= number;
1133 	else
1134 		dquot->dq_dqb.dqb_curspace = 0;
1135 	if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1136 		dquot->dq_dqb.dqb_btime = (time_t) 0;
1137 	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1138 }
1139 
1140 struct dquot_warn {
1141 	struct super_block *w_sb;
1142 	struct kqid w_dq_id;
1143 	short w_type;
1144 };
1145 
1146 static int warning_issued(struct dquot *dquot, const int warntype)
1147 {
1148 	int flag = (warntype == QUOTA_NL_BHARDWARN ||
1149 		warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1150 		((warntype == QUOTA_NL_IHARDWARN ||
1151 		warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1152 
1153 	if (!flag)
1154 		return 0;
1155 	return test_and_set_bit(flag, &dquot->dq_flags);
1156 }
1157 
1158 #ifdef CONFIG_PRINT_QUOTA_WARNING
1159 static int flag_print_warnings = 1;
1160 
1161 static int need_print_warning(struct dquot_warn *warn)
1162 {
1163 	if (!flag_print_warnings)
1164 		return 0;
1165 
1166 	switch (warn->w_dq_id.type) {
1167 		case USRQUOTA:
1168 			return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1169 		case GRPQUOTA:
1170 			return in_group_p(warn->w_dq_id.gid);
1171 		case PRJQUOTA:	/* Never taken... Just make gcc happy */
1172 			return 0;
1173 	}
1174 	return 0;
1175 }
1176 
1177 /* Print warning to user which exceeded quota */
1178 static void print_warning(struct dquot_warn *warn)
1179 {
1180 	char *msg = NULL;
1181 	struct tty_struct *tty;
1182 	int warntype = warn->w_type;
1183 
1184 	if (warntype == QUOTA_NL_IHARDBELOW ||
1185 	    warntype == QUOTA_NL_ISOFTBELOW ||
1186 	    warntype == QUOTA_NL_BHARDBELOW ||
1187 	    warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1188 		return;
1189 
1190 	tty = get_current_tty();
1191 	if (!tty)
1192 		return;
1193 	tty_write_message(tty, warn->w_sb->s_id);
1194 	if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1195 		tty_write_message(tty, ": warning, ");
1196 	else
1197 		tty_write_message(tty, ": write failed, ");
1198 	tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1199 	switch (warntype) {
1200 		case QUOTA_NL_IHARDWARN:
1201 			msg = " file limit reached.\r\n";
1202 			break;
1203 		case QUOTA_NL_ISOFTLONGWARN:
1204 			msg = " file quota exceeded too long.\r\n";
1205 			break;
1206 		case QUOTA_NL_ISOFTWARN:
1207 			msg = " file quota exceeded.\r\n";
1208 			break;
1209 		case QUOTA_NL_BHARDWARN:
1210 			msg = " block limit reached.\r\n";
1211 			break;
1212 		case QUOTA_NL_BSOFTLONGWARN:
1213 			msg = " block quota exceeded too long.\r\n";
1214 			break;
1215 		case QUOTA_NL_BSOFTWARN:
1216 			msg = " block quota exceeded.\r\n";
1217 			break;
1218 	}
1219 	tty_write_message(tty, msg);
1220 	tty_kref_put(tty);
1221 }
1222 #endif
1223 
1224 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1225 			    int warntype)
1226 {
1227 	if (warning_issued(dquot, warntype))
1228 		return;
1229 	warn->w_type = warntype;
1230 	warn->w_sb = dquot->dq_sb;
1231 	warn->w_dq_id = dquot->dq_id;
1232 }
1233 
1234 /*
1235  * Write warnings to the console and send warning messages over netlink.
1236  *
1237  * Note that this function can call into tty and networking code.
1238  */
1239 static void flush_warnings(struct dquot_warn *warn)
1240 {
1241 	int i;
1242 
1243 	for (i = 0; i < MAXQUOTAS; i++) {
1244 		if (warn[i].w_type == QUOTA_NL_NOWARN)
1245 			continue;
1246 #ifdef CONFIG_PRINT_QUOTA_WARNING
1247 		print_warning(&warn[i]);
1248 #endif
1249 		quota_send_warning(warn[i].w_dq_id,
1250 				   warn[i].w_sb->s_dev, warn[i].w_type);
1251 	}
1252 }
1253 
1254 static int ignore_hardlimit(struct dquot *dquot)
1255 {
1256 	struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1257 
1258 	return capable(CAP_SYS_RESOURCE) &&
1259 	       (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1260 		!(info->dqi_flags & V1_DQF_RSQUASH));
1261 }
1262 
1263 /* needs dq_data_lock */
1264 static int check_idq(struct dquot *dquot, qsize_t inodes,
1265 		     struct dquot_warn *warn)
1266 {
1267 	qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1268 
1269 	if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1270 	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1271 		return 0;
1272 
1273 	if (dquot->dq_dqb.dqb_ihardlimit &&
1274 	    newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1275             !ignore_hardlimit(dquot)) {
1276 		prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1277 		return -EDQUOT;
1278 	}
1279 
1280 	if (dquot->dq_dqb.dqb_isoftlimit &&
1281 	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1282 	    dquot->dq_dqb.dqb_itime &&
1283 	    get_seconds() >= dquot->dq_dqb.dqb_itime &&
1284             !ignore_hardlimit(dquot)) {
1285 		prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1286 		return -EDQUOT;
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 = get_seconds() +
1294 		    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1295 	}
1296 
1297 	return 0;
1298 }
1299 
1300 /* needs dq_data_lock */
1301 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1302 		     struct dquot_warn *warn)
1303 {
1304 	qsize_t tspace;
1305 	struct super_block *sb = dquot->dq_sb;
1306 
1307 	if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1308 	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1309 		return 0;
1310 
1311 	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1312 		+ space;
1313 
1314 	if (dquot->dq_dqb.dqb_bhardlimit &&
1315 	    tspace > dquot->dq_dqb.dqb_bhardlimit &&
1316             !ignore_hardlimit(dquot)) {
1317 		if (!prealloc)
1318 			prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1319 		return -EDQUOT;
1320 	}
1321 
1322 	if (dquot->dq_dqb.dqb_bsoftlimit &&
1323 	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1324 	    dquot->dq_dqb.dqb_btime &&
1325 	    get_seconds() >= dquot->dq_dqb.dqb_btime &&
1326             !ignore_hardlimit(dquot)) {
1327 		if (!prealloc)
1328 			prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1329 		return -EDQUOT;
1330 	}
1331 
1332 	if (dquot->dq_dqb.dqb_bsoftlimit &&
1333 	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1334 	    dquot->dq_dqb.dqb_btime == 0) {
1335 		if (!prealloc) {
1336 			prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1337 			dquot->dq_dqb.dqb_btime = get_seconds() +
1338 			    sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1339 		}
1340 		else
1341 			/*
1342 			 * We don't allow preallocation to exceed softlimit so exceeding will
1343 			 * be always printed
1344 			 */
1345 			return -EDQUOT;
1346 	}
1347 
1348 	return 0;
1349 }
1350 
1351 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1352 {
1353 	qsize_t newinodes;
1354 
1355 	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1356 	    dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1357 	    !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1358 		return QUOTA_NL_NOWARN;
1359 
1360 	newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1361 	if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1362 		return QUOTA_NL_ISOFTBELOW;
1363 	if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1364 	    newinodes < dquot->dq_dqb.dqb_ihardlimit)
1365 		return QUOTA_NL_IHARDBELOW;
1366 	return QUOTA_NL_NOWARN;
1367 }
1368 
1369 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1370 {
1371 	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1372 	    dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1373 		return QUOTA_NL_NOWARN;
1374 
1375 	if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1376 		return QUOTA_NL_BSOFTBELOW;
1377 	if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1378 	    dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1379 		return QUOTA_NL_BHARDBELOW;
1380 	return QUOTA_NL_NOWARN;
1381 }
1382 
1383 static int dquot_active(const struct inode *inode)
1384 {
1385 	struct super_block *sb = inode->i_sb;
1386 
1387 	if (IS_NOQUOTA(inode))
1388 		return 0;
1389 	return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1390 }
1391 
1392 /*
1393  * Initialize quota pointers in inode
1394  *
1395  * We do things in a bit complicated way but by that we avoid calling
1396  * dqget() and thus filesystem callbacks under dqptr_sem.
1397  *
1398  * It is better to call this function outside of any transaction as it
1399  * might need a lot of space in journal for dquot structure allocation.
1400  */
1401 static void __dquot_initialize(struct inode *inode, int type)
1402 {
1403 	int cnt;
1404 	struct dquot *got[MAXQUOTAS];
1405 	struct super_block *sb = inode->i_sb;
1406 	qsize_t rsv;
1407 
1408 	/* First test before acquiring mutex - solves deadlocks when we
1409          * re-enter the quota code and are already holding the mutex */
1410 	if (!dquot_active(inode))
1411 		return;
1412 
1413 	/* First get references to structures we might need. */
1414 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1415 		struct kqid qid;
1416 		got[cnt] = NULL;
1417 		if (type != -1 && cnt != type)
1418 			continue;
1419 		switch (cnt) {
1420 		case USRQUOTA:
1421 			qid = make_kqid_uid(inode->i_uid);
1422 			break;
1423 		case GRPQUOTA:
1424 			qid = make_kqid_gid(inode->i_gid);
1425 			break;
1426 		}
1427 		got[cnt] = dqget(sb, qid);
1428 	}
1429 
1430 	down_write(&sb_dqopt(sb)->dqptr_sem);
1431 	if (IS_NOQUOTA(inode))
1432 		goto out_err;
1433 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1434 		if (type != -1 && cnt != type)
1435 			continue;
1436 		/* Avoid races with quotaoff() */
1437 		if (!sb_has_quota_active(sb, cnt))
1438 			continue;
1439 		/* We could race with quotaon or dqget() could have failed */
1440 		if (!got[cnt])
1441 			continue;
1442 		if (!inode->i_dquot[cnt]) {
1443 			inode->i_dquot[cnt] = got[cnt];
1444 			got[cnt] = NULL;
1445 			/*
1446 			 * Make quota reservation system happy if someone
1447 			 * did a write before quota was turned on
1448 			 */
1449 			rsv = inode_get_rsv_space(inode);
1450 			if (unlikely(rsv)) {
1451 				spin_lock(&dq_data_lock);
1452 				dquot_resv_space(inode->i_dquot[cnt], rsv);
1453 				spin_unlock(&dq_data_lock);
1454 			}
1455 		}
1456 	}
1457 out_err:
1458 	up_write(&sb_dqopt(sb)->dqptr_sem);
1459 	/* Drop unused references */
1460 	dqput_all(got);
1461 }
1462 
1463 void dquot_initialize(struct inode *inode)
1464 {
1465 	__dquot_initialize(inode, -1);
1466 }
1467 EXPORT_SYMBOL(dquot_initialize);
1468 
1469 /*
1470  * 	Release all quotas referenced by inode
1471  */
1472 static void __dquot_drop(struct inode *inode)
1473 {
1474 	int cnt;
1475 	struct dquot *put[MAXQUOTAS];
1476 
1477 	down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1478 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1479 		put[cnt] = inode->i_dquot[cnt];
1480 		inode->i_dquot[cnt] = NULL;
1481 	}
1482 	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1483 	dqput_all(put);
1484 }
1485 
1486 void dquot_drop(struct inode *inode)
1487 {
1488 	int cnt;
1489 
1490 	if (IS_NOQUOTA(inode))
1491 		return;
1492 
1493 	/*
1494 	 * Test before calling to rule out calls from proc and such
1495 	 * where we are not allowed to block. Note that this is
1496 	 * actually reliable test even without the lock - the caller
1497 	 * must assure that nobody can come after the DQUOT_DROP and
1498 	 * add quota pointers back anyway.
1499 	 */
1500 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1501 		if (inode->i_dquot[cnt])
1502 			break;
1503 	}
1504 
1505 	if (cnt < MAXQUOTAS)
1506 		__dquot_drop(inode);
1507 }
1508 EXPORT_SYMBOL(dquot_drop);
1509 
1510 /*
1511  * inode_reserved_space is managed internally by quota, and protected by
1512  * i_lock similar to i_blocks+i_bytes.
1513  */
1514 static qsize_t *inode_reserved_space(struct inode * inode)
1515 {
1516 	/* Filesystem must explicitly define it's own method in order to use
1517 	 * quota reservation interface */
1518 	BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1519 	return inode->i_sb->dq_op->get_reserved_space(inode);
1520 }
1521 
1522 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1523 {
1524 	spin_lock(&inode->i_lock);
1525 	*inode_reserved_space(inode) += number;
1526 	spin_unlock(&inode->i_lock);
1527 }
1528 EXPORT_SYMBOL(inode_add_rsv_space);
1529 
1530 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1531 {
1532 	spin_lock(&inode->i_lock);
1533 	*inode_reserved_space(inode) -= number;
1534 	__inode_add_bytes(inode, number);
1535 	spin_unlock(&inode->i_lock);
1536 }
1537 EXPORT_SYMBOL(inode_claim_rsv_space);
1538 
1539 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1540 {
1541 	spin_lock(&inode->i_lock);
1542 	*inode_reserved_space(inode) += number;
1543 	__inode_sub_bytes(inode, number);
1544 	spin_unlock(&inode->i_lock);
1545 }
1546 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1547 
1548 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1549 {
1550 	spin_lock(&inode->i_lock);
1551 	*inode_reserved_space(inode) -= number;
1552 	spin_unlock(&inode->i_lock);
1553 }
1554 EXPORT_SYMBOL(inode_sub_rsv_space);
1555 
1556 static qsize_t inode_get_rsv_space(struct inode *inode)
1557 {
1558 	qsize_t ret;
1559 
1560 	if (!inode->i_sb->dq_op->get_reserved_space)
1561 		return 0;
1562 	spin_lock(&inode->i_lock);
1563 	ret = *inode_reserved_space(inode);
1564 	spin_unlock(&inode->i_lock);
1565 	return ret;
1566 }
1567 
1568 static void inode_incr_space(struct inode *inode, qsize_t number,
1569 				int reserve)
1570 {
1571 	if (reserve)
1572 		inode_add_rsv_space(inode, number);
1573 	else
1574 		inode_add_bytes(inode, number);
1575 }
1576 
1577 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1578 {
1579 	if (reserve)
1580 		inode_sub_rsv_space(inode, number);
1581 	else
1582 		inode_sub_bytes(inode, number);
1583 }
1584 
1585 /*
1586  * This functions updates i_blocks+i_bytes fields and quota information
1587  * (together with appropriate checks).
1588  *
1589  * NOTE: We absolutely rely on the fact that caller dirties the inode
1590  * (usually helpers in quotaops.h care about this) and holds a handle for
1591  * the current transaction so that dquot write and inode write go into the
1592  * same transaction.
1593  */
1594 
1595 /*
1596  * This operation can block, but only after everything is updated
1597  */
1598 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1599 {
1600 	int cnt, ret = 0;
1601 	struct dquot_warn warn[MAXQUOTAS];
1602 	struct dquot **dquots = inode->i_dquot;
1603 	int reserve = flags & DQUOT_SPACE_RESERVE;
1604 
1605 	/*
1606 	 * First test before acquiring mutex - solves deadlocks when we
1607 	 * re-enter the quota code and are already holding the mutex
1608 	 */
1609 	if (!dquot_active(inode)) {
1610 		inode_incr_space(inode, number, reserve);
1611 		goto out;
1612 	}
1613 
1614 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1615 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1616 
1617 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1618 	spin_lock(&dq_data_lock);
1619 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1620 		if (!dquots[cnt])
1621 			continue;
1622 		ret = check_bdq(dquots[cnt], number,
1623 				!(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1624 		if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1625 			spin_unlock(&dq_data_lock);
1626 			goto out_flush_warn;
1627 		}
1628 	}
1629 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1630 		if (!dquots[cnt])
1631 			continue;
1632 		if (reserve)
1633 			dquot_resv_space(dquots[cnt], number);
1634 		else
1635 			dquot_incr_space(dquots[cnt], number);
1636 	}
1637 	inode_incr_space(inode, number, reserve);
1638 	spin_unlock(&dq_data_lock);
1639 
1640 	if (reserve)
1641 		goto out_flush_warn;
1642 	mark_all_dquot_dirty(dquots);
1643 out_flush_warn:
1644 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1645 	flush_warnings(warn);
1646 out:
1647 	return ret;
1648 }
1649 EXPORT_SYMBOL(__dquot_alloc_space);
1650 
1651 /*
1652  * This operation can block, but only after everything is updated
1653  */
1654 int dquot_alloc_inode(const struct inode *inode)
1655 {
1656 	int cnt, ret = 0;
1657 	struct dquot_warn warn[MAXQUOTAS];
1658 	struct dquot * const *dquots = inode->i_dquot;
1659 
1660 	/* First test before acquiring mutex - solves deadlocks when we
1661          * re-enter the quota code and are already holding the mutex */
1662 	if (!dquot_active(inode))
1663 		return 0;
1664 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1665 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1666 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1667 	spin_lock(&dq_data_lock);
1668 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1669 		if (!dquots[cnt])
1670 			continue;
1671 		ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1672 		if (ret)
1673 			goto warn_put_all;
1674 	}
1675 
1676 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1677 		if (!dquots[cnt])
1678 			continue;
1679 		dquot_incr_inodes(dquots[cnt], 1);
1680 	}
1681 
1682 warn_put_all:
1683 	spin_unlock(&dq_data_lock);
1684 	if (ret == 0)
1685 		mark_all_dquot_dirty(dquots);
1686 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1687 	flush_warnings(warn);
1688 	return ret;
1689 }
1690 EXPORT_SYMBOL(dquot_alloc_inode);
1691 
1692 /*
1693  * Convert in-memory reserved quotas to real consumed quotas
1694  */
1695 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1696 {
1697 	int cnt;
1698 
1699 	if (!dquot_active(inode)) {
1700 		inode_claim_rsv_space(inode, number);
1701 		return 0;
1702 	}
1703 
1704 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1705 	spin_lock(&dq_data_lock);
1706 	/* Claim reserved quotas to allocated quotas */
1707 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1708 		if (inode->i_dquot[cnt])
1709 			dquot_claim_reserved_space(inode->i_dquot[cnt],
1710 							number);
1711 	}
1712 	/* Update inode bytes */
1713 	inode_claim_rsv_space(inode, number);
1714 	spin_unlock(&dq_data_lock);
1715 	mark_all_dquot_dirty(inode->i_dquot);
1716 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1717 	return 0;
1718 }
1719 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1720 
1721 /*
1722  * Convert allocated space back to in-memory reserved quotas
1723  */
1724 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1725 {
1726 	int cnt;
1727 
1728 	if (!dquot_active(inode)) {
1729 		inode_reclaim_rsv_space(inode, number);
1730 		return;
1731 	}
1732 
1733 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1734 	spin_lock(&dq_data_lock);
1735 	/* Claim reserved quotas to allocated quotas */
1736 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1737 		if (inode->i_dquot[cnt])
1738 			dquot_reclaim_reserved_space(inode->i_dquot[cnt],
1739 						     number);
1740 	}
1741 	/* Update inode bytes */
1742 	inode_reclaim_rsv_space(inode, number);
1743 	spin_unlock(&dq_data_lock);
1744 	mark_all_dquot_dirty(inode->i_dquot);
1745 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1746 	return;
1747 }
1748 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1749 
1750 /*
1751  * This operation can block, but only after everything is updated
1752  */
1753 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1754 {
1755 	unsigned int cnt;
1756 	struct dquot_warn warn[MAXQUOTAS];
1757 	struct dquot **dquots = inode->i_dquot;
1758 	int reserve = flags & DQUOT_SPACE_RESERVE;
1759 
1760 	/* First test before acquiring mutex - solves deadlocks when we
1761          * re-enter the quota code and are already holding the mutex */
1762 	if (!dquot_active(inode)) {
1763 		inode_decr_space(inode, number, reserve);
1764 		return;
1765 	}
1766 
1767 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1768 	spin_lock(&dq_data_lock);
1769 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1770 		int wtype;
1771 
1772 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1773 		if (!dquots[cnt])
1774 			continue;
1775 		wtype = info_bdq_free(dquots[cnt], number);
1776 		if (wtype != QUOTA_NL_NOWARN)
1777 			prepare_warning(&warn[cnt], dquots[cnt], wtype);
1778 		if (reserve)
1779 			dquot_free_reserved_space(dquots[cnt], number);
1780 		else
1781 			dquot_decr_space(dquots[cnt], number);
1782 	}
1783 	inode_decr_space(inode, number, reserve);
1784 	spin_unlock(&dq_data_lock);
1785 
1786 	if (reserve)
1787 		goto out_unlock;
1788 	mark_all_dquot_dirty(dquots);
1789 out_unlock:
1790 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1791 	flush_warnings(warn);
1792 }
1793 EXPORT_SYMBOL(__dquot_free_space);
1794 
1795 /*
1796  * This operation can block, but only after everything is updated
1797  */
1798 void dquot_free_inode(const struct inode *inode)
1799 {
1800 	unsigned int cnt;
1801 	struct dquot_warn warn[MAXQUOTAS];
1802 	struct dquot * const *dquots = inode->i_dquot;
1803 
1804 	/* First test before acquiring mutex - solves deadlocks when we
1805          * re-enter the quota code and are already holding the mutex */
1806 	if (!dquot_active(inode))
1807 		return;
1808 
1809 	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1810 	spin_lock(&dq_data_lock);
1811 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1812 		int wtype;
1813 
1814 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1815 		if (!dquots[cnt])
1816 			continue;
1817 		wtype = info_idq_free(dquots[cnt], 1);
1818 		if (wtype != QUOTA_NL_NOWARN)
1819 			prepare_warning(&warn[cnt], dquots[cnt], wtype);
1820 		dquot_decr_inodes(dquots[cnt], 1);
1821 	}
1822 	spin_unlock(&dq_data_lock);
1823 	mark_all_dquot_dirty(dquots);
1824 	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1825 	flush_warnings(warn);
1826 }
1827 EXPORT_SYMBOL(dquot_free_inode);
1828 
1829 /*
1830  * Transfer the number of inode and blocks from one diskquota to an other.
1831  * On success, dquot references in transfer_to are consumed and references
1832  * to original dquots that need to be released are placed there. On failure,
1833  * references are kept untouched.
1834  *
1835  * This operation can block, but only after everything is updated
1836  * A transaction must be started when entering this function.
1837  *
1838  */
1839 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1840 {
1841 	qsize_t space, cur_space;
1842 	qsize_t rsv_space = 0;
1843 	struct dquot *transfer_from[MAXQUOTAS] = {};
1844 	int cnt, ret = 0;
1845 	char is_valid[MAXQUOTAS] = {};
1846 	struct dquot_warn warn_to[MAXQUOTAS];
1847 	struct dquot_warn warn_from_inodes[MAXQUOTAS];
1848 	struct dquot_warn warn_from_space[MAXQUOTAS];
1849 
1850 	/* First test before acquiring mutex - solves deadlocks when we
1851          * re-enter the quota code and are already holding the mutex */
1852 	if (IS_NOQUOTA(inode))
1853 		return 0;
1854 	/* Initialize the arrays */
1855 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1856 		warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1857 		warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1858 		warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1859 	}
1860 	down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1861 	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
1862 		up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1863 		return 0;
1864 	}
1865 	spin_lock(&dq_data_lock);
1866 	cur_space = inode_get_bytes(inode);
1867 	rsv_space = inode_get_rsv_space(inode);
1868 	space = cur_space + rsv_space;
1869 	/* Build the transfer_from list and check the limits */
1870 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1871 		/*
1872 		 * Skip changes for same uid or gid or for turned off quota-type.
1873 		 */
1874 		if (!transfer_to[cnt])
1875 			continue;
1876 		/* Avoid races with quotaoff() */
1877 		if (!sb_has_quota_active(inode->i_sb, cnt))
1878 			continue;
1879 		is_valid[cnt] = 1;
1880 		transfer_from[cnt] = inode->i_dquot[cnt];
1881 		ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1882 		if (ret)
1883 			goto over_quota;
1884 		ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1885 		if (ret)
1886 			goto over_quota;
1887 	}
1888 
1889 	/*
1890 	 * Finally perform the needed transfer from transfer_from to transfer_to
1891 	 */
1892 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1893 		if (!is_valid[cnt])
1894 			continue;
1895 		/* Due to IO error we might not have transfer_from[] structure */
1896 		if (transfer_from[cnt]) {
1897 			int wtype;
1898 			wtype = info_idq_free(transfer_from[cnt], 1);
1899 			if (wtype != QUOTA_NL_NOWARN)
1900 				prepare_warning(&warn_from_inodes[cnt],
1901 						transfer_from[cnt], wtype);
1902 			wtype = info_bdq_free(transfer_from[cnt], space);
1903 			if (wtype != QUOTA_NL_NOWARN)
1904 				prepare_warning(&warn_from_space[cnt],
1905 						transfer_from[cnt], wtype);
1906 			dquot_decr_inodes(transfer_from[cnt], 1);
1907 			dquot_decr_space(transfer_from[cnt], cur_space);
1908 			dquot_free_reserved_space(transfer_from[cnt],
1909 						  rsv_space);
1910 		}
1911 
1912 		dquot_incr_inodes(transfer_to[cnt], 1);
1913 		dquot_incr_space(transfer_to[cnt], cur_space);
1914 		dquot_resv_space(transfer_to[cnt], rsv_space);
1915 
1916 		inode->i_dquot[cnt] = transfer_to[cnt];
1917 	}
1918 	spin_unlock(&dq_data_lock);
1919 	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1920 
1921 	mark_all_dquot_dirty(transfer_from);
1922 	mark_all_dquot_dirty(transfer_to);
1923 	flush_warnings(warn_to);
1924 	flush_warnings(warn_from_inodes);
1925 	flush_warnings(warn_from_space);
1926 	/* Pass back references to put */
1927 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1928 		if (is_valid[cnt])
1929 			transfer_to[cnt] = transfer_from[cnt];
1930 	return 0;
1931 over_quota:
1932 	spin_unlock(&dq_data_lock);
1933 	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1934 	flush_warnings(warn_to);
1935 	return ret;
1936 }
1937 EXPORT_SYMBOL(__dquot_transfer);
1938 
1939 /* Wrapper for transferring ownership of an inode for uid/gid only
1940  * Called from FSXXX_setattr()
1941  */
1942 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1943 {
1944 	struct dquot *transfer_to[MAXQUOTAS] = {};
1945 	struct super_block *sb = inode->i_sb;
1946 	int ret;
1947 
1948 	if (!dquot_active(inode))
1949 		return 0;
1950 
1951 	if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid))
1952 		transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(iattr->ia_uid));
1953 	if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))
1954 		transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(iattr->ia_gid));
1955 
1956 	ret = __dquot_transfer(inode, transfer_to);
1957 	dqput_all(transfer_to);
1958 	return ret;
1959 }
1960 EXPORT_SYMBOL(dquot_transfer);
1961 
1962 /*
1963  * Write info of quota file to disk
1964  */
1965 int dquot_commit_info(struct super_block *sb, int type)
1966 {
1967 	int ret;
1968 	struct quota_info *dqopt = sb_dqopt(sb);
1969 
1970 	mutex_lock(&dqopt->dqio_mutex);
1971 	ret = dqopt->ops[type]->write_file_info(sb, type);
1972 	mutex_unlock(&dqopt->dqio_mutex);
1973 	return ret;
1974 }
1975 EXPORT_SYMBOL(dquot_commit_info);
1976 
1977 /*
1978  * Definitions of diskquota operations.
1979  */
1980 const struct dquot_operations dquot_operations = {
1981 	.write_dquot	= dquot_commit,
1982 	.acquire_dquot	= dquot_acquire,
1983 	.release_dquot	= dquot_release,
1984 	.mark_dirty	= dquot_mark_dquot_dirty,
1985 	.write_info	= dquot_commit_info,
1986 	.alloc_dquot	= dquot_alloc,
1987 	.destroy_dquot	= dquot_destroy,
1988 };
1989 EXPORT_SYMBOL(dquot_operations);
1990 
1991 /*
1992  * Generic helper for ->open on filesystems supporting disk quotas.
1993  */
1994 int dquot_file_open(struct inode *inode, struct file *file)
1995 {
1996 	int error;
1997 
1998 	error = generic_file_open(inode, file);
1999 	if (!error && (file->f_mode & FMODE_WRITE))
2000 		dquot_initialize(inode);
2001 	return error;
2002 }
2003 EXPORT_SYMBOL(dquot_file_open);
2004 
2005 /*
2006  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2007  */
2008 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2009 {
2010 	int cnt, ret = 0;
2011 	struct quota_info *dqopt = sb_dqopt(sb);
2012 	struct inode *toputinode[MAXQUOTAS];
2013 
2014 	/* Cannot turn off usage accounting without turning off limits, or
2015 	 * suspend quotas and simultaneously turn quotas off. */
2016 	if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2017 	    || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2018 	    DQUOT_USAGE_ENABLED)))
2019 		return -EINVAL;
2020 
2021 	/* We need to serialize quota_off() for device */
2022 	mutex_lock(&dqopt->dqonoff_mutex);
2023 
2024 	/*
2025 	 * Skip everything if there's nothing to do. We have to do this because
2026 	 * sometimes we are called when fill_super() failed and calling
2027 	 * sync_fs() in such cases does no good.
2028 	 */
2029 	if (!sb_any_quota_loaded(sb)) {
2030 		mutex_unlock(&dqopt->dqonoff_mutex);
2031 		return 0;
2032 	}
2033 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2034 		toputinode[cnt] = NULL;
2035 		if (type != -1 && cnt != type)
2036 			continue;
2037 		if (!sb_has_quota_loaded(sb, cnt))
2038 			continue;
2039 
2040 		if (flags & DQUOT_SUSPENDED) {
2041 			spin_lock(&dq_state_lock);
2042 			dqopt->flags |=
2043 				dquot_state_flag(DQUOT_SUSPENDED, cnt);
2044 			spin_unlock(&dq_state_lock);
2045 		} else {
2046 			spin_lock(&dq_state_lock);
2047 			dqopt->flags &= ~dquot_state_flag(flags, cnt);
2048 			/* Turning off suspended quotas? */
2049 			if (!sb_has_quota_loaded(sb, cnt) &&
2050 			    sb_has_quota_suspended(sb, cnt)) {
2051 				dqopt->flags &=	~dquot_state_flag(
2052 							DQUOT_SUSPENDED, cnt);
2053 				spin_unlock(&dq_state_lock);
2054 				iput(dqopt->files[cnt]);
2055 				dqopt->files[cnt] = NULL;
2056 				continue;
2057 			}
2058 			spin_unlock(&dq_state_lock);
2059 		}
2060 
2061 		/* We still have to keep quota loaded? */
2062 		if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2063 			continue;
2064 
2065 		/* Note: these are blocking operations */
2066 		drop_dquot_ref(sb, cnt);
2067 		invalidate_dquots(sb, cnt);
2068 		/*
2069 		 * Now all dquots should be invalidated, all writes done so we
2070 		 * should be only users of the info. No locks needed.
2071 		 */
2072 		if (info_dirty(&dqopt->info[cnt]))
2073 			sb->dq_op->write_info(sb, cnt);
2074 		if (dqopt->ops[cnt]->free_file_info)
2075 			dqopt->ops[cnt]->free_file_info(sb, cnt);
2076 		put_quota_format(dqopt->info[cnt].dqi_format);
2077 
2078 		toputinode[cnt] = dqopt->files[cnt];
2079 		if (!sb_has_quota_loaded(sb, cnt))
2080 			dqopt->files[cnt] = NULL;
2081 		dqopt->info[cnt].dqi_flags = 0;
2082 		dqopt->info[cnt].dqi_igrace = 0;
2083 		dqopt->info[cnt].dqi_bgrace = 0;
2084 		dqopt->ops[cnt] = NULL;
2085 	}
2086 	mutex_unlock(&dqopt->dqonoff_mutex);
2087 
2088 	/* Skip syncing and setting flags if quota files are hidden */
2089 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2090 		goto put_inodes;
2091 
2092 	/* Sync the superblock so that buffers with quota data are written to
2093 	 * disk (and so userspace sees correct data afterwards). */
2094 	if (sb->s_op->sync_fs)
2095 		sb->s_op->sync_fs(sb, 1);
2096 	sync_blockdev(sb->s_bdev);
2097 	/* Now the quota files are just ordinary files and we can set the
2098 	 * inode flags back. Moreover we discard the pagecache so that
2099 	 * userspace sees the writes we did bypassing the pagecache. We
2100 	 * must also discard the blockdev buffers so that we see the
2101 	 * changes done by userspace on the next quotaon() */
2102 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2103 		if (toputinode[cnt]) {
2104 			mutex_lock(&dqopt->dqonoff_mutex);
2105 			/* If quota was reenabled in the meantime, we have
2106 			 * nothing to do */
2107 			if (!sb_has_quota_loaded(sb, cnt)) {
2108 				mutex_lock(&toputinode[cnt]->i_mutex);
2109 				toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2110 				  S_NOATIME | S_NOQUOTA);
2111 				truncate_inode_pages(&toputinode[cnt]->i_data,
2112 						     0);
2113 				mutex_unlock(&toputinode[cnt]->i_mutex);
2114 				mark_inode_dirty_sync(toputinode[cnt]);
2115 			}
2116 			mutex_unlock(&dqopt->dqonoff_mutex);
2117 		}
2118 	if (sb->s_bdev)
2119 		invalidate_bdev(sb->s_bdev);
2120 put_inodes:
2121 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2122 		if (toputinode[cnt]) {
2123 			/* On remount RO, we keep the inode pointer so that we
2124 			 * can reenable quota on the subsequent remount RW. We
2125 			 * have to check 'flags' variable and not use sb_has_
2126 			 * function because another quotaon / quotaoff could
2127 			 * change global state before we got here. We refuse
2128 			 * to suspend quotas when there is pending delete on
2129 			 * the quota file... */
2130 			if (!(flags & DQUOT_SUSPENDED))
2131 				iput(toputinode[cnt]);
2132 			else if (!toputinode[cnt]->i_nlink)
2133 				ret = -EBUSY;
2134 		}
2135 	return ret;
2136 }
2137 EXPORT_SYMBOL(dquot_disable);
2138 
2139 int dquot_quota_off(struct super_block *sb, int type)
2140 {
2141 	return dquot_disable(sb, type,
2142 			     DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2143 }
2144 EXPORT_SYMBOL(dquot_quota_off);
2145 
2146 /*
2147  *	Turn quotas on on a device
2148  */
2149 
2150 /*
2151  * Helper function to turn quotas on when we already have the inode of
2152  * quota file and no quota information is loaded.
2153  */
2154 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2155 	unsigned int flags)
2156 {
2157 	struct quota_format_type *fmt = find_quota_format(format_id);
2158 	struct super_block *sb = inode->i_sb;
2159 	struct quota_info *dqopt = sb_dqopt(sb);
2160 	int error;
2161 	int oldflags = -1;
2162 
2163 	if (!fmt)
2164 		return -ESRCH;
2165 	if (!S_ISREG(inode->i_mode)) {
2166 		error = -EACCES;
2167 		goto out_fmt;
2168 	}
2169 	if (IS_RDONLY(inode)) {
2170 		error = -EROFS;
2171 		goto out_fmt;
2172 	}
2173 	if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
2174 		error = -EINVAL;
2175 		goto out_fmt;
2176 	}
2177 	/* Usage always has to be set... */
2178 	if (!(flags & DQUOT_USAGE_ENABLED)) {
2179 		error = -EINVAL;
2180 		goto out_fmt;
2181 	}
2182 
2183 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2184 		/* As we bypass the pagecache we must now flush all the
2185 		 * dirty data and invalidate caches so that kernel sees
2186 		 * changes from userspace. It is not enough to just flush
2187 		 * the quota file since if blocksize < pagesize, invalidation
2188 		 * of the cache could fail because of other unrelated dirty
2189 		 * data */
2190 		sync_filesystem(sb);
2191 		invalidate_bdev(sb->s_bdev);
2192 	}
2193 	mutex_lock(&dqopt->dqonoff_mutex);
2194 	if (sb_has_quota_loaded(sb, type)) {
2195 		error = -EBUSY;
2196 		goto out_lock;
2197 	}
2198 
2199 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2200 		/* We don't want quota and atime on quota files (deadlocks
2201 		 * possible) Also nobody should write to the file - we use
2202 		 * special IO operations which ignore the immutable bit. */
2203 		mutex_lock(&inode->i_mutex);
2204 		oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2205 					     S_NOQUOTA);
2206 		inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2207 		mutex_unlock(&inode->i_mutex);
2208 		/*
2209 		 * When S_NOQUOTA is set, remove dquot references as no more
2210 		 * references can be added
2211 		 */
2212 		__dquot_drop(inode);
2213 	}
2214 
2215 	error = -EIO;
2216 	dqopt->files[type] = igrab(inode);
2217 	if (!dqopt->files[type])
2218 		goto out_lock;
2219 	error = -EINVAL;
2220 	if (!fmt->qf_ops->check_quota_file(sb, type))
2221 		goto out_file_init;
2222 
2223 	dqopt->ops[type] = fmt->qf_ops;
2224 	dqopt->info[type].dqi_format = fmt;
2225 	dqopt->info[type].dqi_fmt_id = format_id;
2226 	INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2227 	mutex_lock(&dqopt->dqio_mutex);
2228 	error = dqopt->ops[type]->read_file_info(sb, type);
2229 	if (error < 0) {
2230 		mutex_unlock(&dqopt->dqio_mutex);
2231 		goto out_file_init;
2232 	}
2233 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2234 		dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2235 	mutex_unlock(&dqopt->dqio_mutex);
2236 	spin_lock(&dq_state_lock);
2237 	dqopt->flags |= dquot_state_flag(flags, type);
2238 	spin_unlock(&dq_state_lock);
2239 
2240 	add_dquot_ref(sb, type);
2241 	mutex_unlock(&dqopt->dqonoff_mutex);
2242 
2243 	return 0;
2244 
2245 out_file_init:
2246 	dqopt->files[type] = NULL;
2247 	iput(inode);
2248 out_lock:
2249 	if (oldflags != -1) {
2250 		mutex_lock(&inode->i_mutex);
2251 		/* Set the flags back (in the case of accidental quotaon()
2252 		 * on a wrong file we don't want to mess up the flags) */
2253 		inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2254 		inode->i_flags |= oldflags;
2255 		mutex_unlock(&inode->i_mutex);
2256 	}
2257 	mutex_unlock(&dqopt->dqonoff_mutex);
2258 out_fmt:
2259 	put_quota_format(fmt);
2260 
2261 	return error;
2262 }
2263 
2264 /* Reenable quotas on remount RW */
2265 int dquot_resume(struct super_block *sb, int type)
2266 {
2267 	struct quota_info *dqopt = sb_dqopt(sb);
2268 	struct inode *inode;
2269 	int ret = 0, cnt;
2270 	unsigned int flags;
2271 
2272 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2273 		if (type != -1 && cnt != type)
2274 			continue;
2275 
2276 		mutex_lock(&dqopt->dqonoff_mutex);
2277 		if (!sb_has_quota_suspended(sb, cnt)) {
2278 			mutex_unlock(&dqopt->dqonoff_mutex);
2279 			continue;
2280 		}
2281 		inode = dqopt->files[cnt];
2282 		dqopt->files[cnt] = NULL;
2283 		spin_lock(&dq_state_lock);
2284 		flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2285 							DQUOT_LIMITS_ENABLED,
2286 							cnt);
2287 		dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2288 		spin_unlock(&dq_state_lock);
2289 		mutex_unlock(&dqopt->dqonoff_mutex);
2290 
2291 		flags = dquot_generic_flag(flags, cnt);
2292 		ret = vfs_load_quota_inode(inode, cnt,
2293 				dqopt->info[cnt].dqi_fmt_id, flags);
2294 		iput(inode);
2295 	}
2296 
2297 	return ret;
2298 }
2299 EXPORT_SYMBOL(dquot_resume);
2300 
2301 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2302 		   struct path *path)
2303 {
2304 	int error = security_quota_on(path->dentry);
2305 	if (error)
2306 		return error;
2307 	/* Quota file not on the same filesystem? */
2308 	if (path->dentry->d_sb != sb)
2309 		error = -EXDEV;
2310 	else
2311 		error = vfs_load_quota_inode(path->dentry->d_inode, type,
2312 					     format_id, DQUOT_USAGE_ENABLED |
2313 					     DQUOT_LIMITS_ENABLED);
2314 	return error;
2315 }
2316 EXPORT_SYMBOL(dquot_quota_on);
2317 
2318 /*
2319  * More powerful function for turning on quotas allowing setting
2320  * of individual quota flags
2321  */
2322 int dquot_enable(struct inode *inode, int type, int format_id,
2323 		 unsigned int flags)
2324 {
2325 	int ret = 0;
2326 	struct super_block *sb = inode->i_sb;
2327 	struct quota_info *dqopt = sb_dqopt(sb);
2328 
2329 	/* Just unsuspend quotas? */
2330 	BUG_ON(flags & DQUOT_SUSPENDED);
2331 
2332 	if (!flags)
2333 		return 0;
2334 	/* Just updating flags needed? */
2335 	if (sb_has_quota_loaded(sb, type)) {
2336 		mutex_lock(&dqopt->dqonoff_mutex);
2337 		/* Now do a reliable test... */
2338 		if (!sb_has_quota_loaded(sb, type)) {
2339 			mutex_unlock(&dqopt->dqonoff_mutex);
2340 			goto load_quota;
2341 		}
2342 		if (flags & DQUOT_USAGE_ENABLED &&
2343 		    sb_has_quota_usage_enabled(sb, type)) {
2344 			ret = -EBUSY;
2345 			goto out_lock;
2346 		}
2347 		if (flags & DQUOT_LIMITS_ENABLED &&
2348 		    sb_has_quota_limits_enabled(sb, type)) {
2349 			ret = -EBUSY;
2350 			goto out_lock;
2351 		}
2352 		spin_lock(&dq_state_lock);
2353 		sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2354 		spin_unlock(&dq_state_lock);
2355 out_lock:
2356 		mutex_unlock(&dqopt->dqonoff_mutex);
2357 		return ret;
2358 	}
2359 
2360 load_quota:
2361 	return vfs_load_quota_inode(inode, type, format_id, flags);
2362 }
2363 EXPORT_SYMBOL(dquot_enable);
2364 
2365 /*
2366  * This function is used when filesystem needs to initialize quotas
2367  * during mount time.
2368  */
2369 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2370 		int format_id, int type)
2371 {
2372 	struct dentry *dentry;
2373 	int error;
2374 
2375 	mutex_lock(&sb->s_root->d_inode->i_mutex);
2376 	dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2377 	mutex_unlock(&sb->s_root->d_inode->i_mutex);
2378 	if (IS_ERR(dentry))
2379 		return PTR_ERR(dentry);
2380 
2381 	if (!dentry->d_inode) {
2382 		error = -ENOENT;
2383 		goto out;
2384 	}
2385 
2386 	error = security_quota_on(dentry);
2387 	if (!error)
2388 		error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
2389 				DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2390 
2391 out:
2392 	dput(dentry);
2393 	return error;
2394 }
2395 EXPORT_SYMBOL(dquot_quota_on_mount);
2396 
2397 static inline qsize_t qbtos(qsize_t blocks)
2398 {
2399 	return blocks << QIF_DQBLKSIZE_BITS;
2400 }
2401 
2402 static inline qsize_t stoqb(qsize_t space)
2403 {
2404 	return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
2405 }
2406 
2407 /* Generic routine for getting common part of quota structure */
2408 static void do_get_dqblk(struct dquot *dquot, struct fs_disk_quota *di)
2409 {
2410 	struct mem_dqblk *dm = &dquot->dq_dqb;
2411 
2412 	memset(di, 0, sizeof(*di));
2413 	di->d_version = FS_DQUOT_VERSION;
2414 	di->d_flags = dquot->dq_id.type == USRQUOTA ?
2415 			FS_USER_QUOTA : FS_GROUP_QUOTA;
2416 	di->d_id = from_kqid_munged(current_user_ns(), dquot->dq_id);
2417 
2418 	spin_lock(&dq_data_lock);
2419 	di->d_blk_hardlimit = stoqb(dm->dqb_bhardlimit);
2420 	di->d_blk_softlimit = stoqb(dm->dqb_bsoftlimit);
2421 	di->d_ino_hardlimit = dm->dqb_ihardlimit;
2422 	di->d_ino_softlimit = dm->dqb_isoftlimit;
2423 	di->d_bcount = dm->dqb_curspace + dm->dqb_rsvspace;
2424 	di->d_icount = dm->dqb_curinodes;
2425 	di->d_btimer = dm->dqb_btime;
2426 	di->d_itimer = dm->dqb_itime;
2427 	spin_unlock(&dq_data_lock);
2428 }
2429 
2430 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2431 		    struct fs_disk_quota *di)
2432 {
2433 	struct dquot *dquot;
2434 
2435 	dquot = dqget(sb, qid);
2436 	if (!dquot)
2437 		return -ESRCH;
2438 	do_get_dqblk(dquot, di);
2439 	dqput(dquot);
2440 
2441 	return 0;
2442 }
2443 EXPORT_SYMBOL(dquot_get_dqblk);
2444 
2445 #define VFS_FS_DQ_MASK \
2446 	(FS_DQ_BCOUNT | FS_DQ_BSOFT | FS_DQ_BHARD | \
2447 	 FS_DQ_ICOUNT | FS_DQ_ISOFT | FS_DQ_IHARD | \
2448 	 FS_DQ_BTIMER | FS_DQ_ITIMER)
2449 
2450 /* Generic routine for setting common part of quota structure */
2451 static int do_set_dqblk(struct dquot *dquot, struct fs_disk_quota *di)
2452 {
2453 	struct mem_dqblk *dm = &dquot->dq_dqb;
2454 	int check_blim = 0, check_ilim = 0;
2455 	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2456 
2457 	if (di->d_fieldmask & ~VFS_FS_DQ_MASK)
2458 		return -EINVAL;
2459 
2460 	if (((di->d_fieldmask & FS_DQ_BSOFT) &&
2461 	     (di->d_blk_softlimit > dqi->dqi_maxblimit)) ||
2462 	    ((di->d_fieldmask & FS_DQ_BHARD) &&
2463 	     (di->d_blk_hardlimit > dqi->dqi_maxblimit)) ||
2464 	    ((di->d_fieldmask & FS_DQ_ISOFT) &&
2465 	     (di->d_ino_softlimit > dqi->dqi_maxilimit)) ||
2466 	    ((di->d_fieldmask & FS_DQ_IHARD) &&
2467 	     (di->d_ino_hardlimit > dqi->dqi_maxilimit)))
2468 		return -ERANGE;
2469 
2470 	spin_lock(&dq_data_lock);
2471 	if (di->d_fieldmask & FS_DQ_BCOUNT) {
2472 		dm->dqb_curspace = di->d_bcount - dm->dqb_rsvspace;
2473 		check_blim = 1;
2474 		set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2475 	}
2476 
2477 	if (di->d_fieldmask & FS_DQ_BSOFT)
2478 		dm->dqb_bsoftlimit = qbtos(di->d_blk_softlimit);
2479 	if (di->d_fieldmask & FS_DQ_BHARD)
2480 		dm->dqb_bhardlimit = qbtos(di->d_blk_hardlimit);
2481 	if (di->d_fieldmask & (FS_DQ_BSOFT | FS_DQ_BHARD)) {
2482 		check_blim = 1;
2483 		set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2484 	}
2485 
2486 	if (di->d_fieldmask & FS_DQ_ICOUNT) {
2487 		dm->dqb_curinodes = di->d_icount;
2488 		check_ilim = 1;
2489 		set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2490 	}
2491 
2492 	if (di->d_fieldmask & FS_DQ_ISOFT)
2493 		dm->dqb_isoftlimit = di->d_ino_softlimit;
2494 	if (di->d_fieldmask & FS_DQ_IHARD)
2495 		dm->dqb_ihardlimit = di->d_ino_hardlimit;
2496 	if (di->d_fieldmask & (FS_DQ_ISOFT | FS_DQ_IHARD)) {
2497 		check_ilim = 1;
2498 		set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2499 	}
2500 
2501 	if (di->d_fieldmask & FS_DQ_BTIMER) {
2502 		dm->dqb_btime = di->d_btimer;
2503 		check_blim = 1;
2504 		set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2505 	}
2506 
2507 	if (di->d_fieldmask & FS_DQ_ITIMER) {
2508 		dm->dqb_itime = di->d_itimer;
2509 		check_ilim = 1;
2510 		set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2511 	}
2512 
2513 	if (check_blim) {
2514 		if (!dm->dqb_bsoftlimit ||
2515 		    dm->dqb_curspace < dm->dqb_bsoftlimit) {
2516 			dm->dqb_btime = 0;
2517 			clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2518 		} else if (!(di->d_fieldmask & FS_DQ_BTIMER))
2519 			/* Set grace only if user hasn't provided his own... */
2520 			dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2521 	}
2522 	if (check_ilim) {
2523 		if (!dm->dqb_isoftlimit ||
2524 		    dm->dqb_curinodes < dm->dqb_isoftlimit) {
2525 			dm->dqb_itime = 0;
2526 			clear_bit(DQ_INODES_B, &dquot->dq_flags);
2527 		} else if (!(di->d_fieldmask & FS_DQ_ITIMER))
2528 			/* Set grace only if user hasn't provided his own... */
2529 			dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2530 	}
2531 	if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2532 	    dm->dqb_isoftlimit)
2533 		clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2534 	else
2535 		set_bit(DQ_FAKE_B, &dquot->dq_flags);
2536 	spin_unlock(&dq_data_lock);
2537 	mark_dquot_dirty(dquot);
2538 
2539 	return 0;
2540 }
2541 
2542 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2543 		  struct fs_disk_quota *di)
2544 {
2545 	struct dquot *dquot;
2546 	int rc;
2547 
2548 	dquot = dqget(sb, qid);
2549 	if (!dquot) {
2550 		rc = -ESRCH;
2551 		goto out;
2552 	}
2553 	rc = do_set_dqblk(dquot, di);
2554 	dqput(dquot);
2555 out:
2556 	return rc;
2557 }
2558 EXPORT_SYMBOL(dquot_set_dqblk);
2559 
2560 /* Generic routine for getting common part of quota file information */
2561 int dquot_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2562 {
2563 	struct mem_dqinfo *mi;
2564 
2565 	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2566 	if (!sb_has_quota_active(sb, type)) {
2567 		mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2568 		return -ESRCH;
2569 	}
2570 	mi = sb_dqopt(sb)->info + type;
2571 	spin_lock(&dq_data_lock);
2572 	ii->dqi_bgrace = mi->dqi_bgrace;
2573 	ii->dqi_igrace = mi->dqi_igrace;
2574 	ii->dqi_flags = mi->dqi_flags & DQF_GETINFO_MASK;
2575 	ii->dqi_valid = IIF_ALL;
2576 	spin_unlock(&dq_data_lock);
2577 	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2578 	return 0;
2579 }
2580 EXPORT_SYMBOL(dquot_get_dqinfo);
2581 
2582 /* Generic routine for setting common part of quota file information */
2583 int dquot_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2584 {
2585 	struct mem_dqinfo *mi;
2586 	int err = 0;
2587 
2588 	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2589 	if (!sb_has_quota_active(sb, type)) {
2590 		err = -ESRCH;
2591 		goto out;
2592 	}
2593 	mi = sb_dqopt(sb)->info + type;
2594 	spin_lock(&dq_data_lock);
2595 	if (ii->dqi_valid & IIF_BGRACE)
2596 		mi->dqi_bgrace = ii->dqi_bgrace;
2597 	if (ii->dqi_valid & IIF_IGRACE)
2598 		mi->dqi_igrace = ii->dqi_igrace;
2599 	if (ii->dqi_valid & IIF_FLAGS)
2600 		mi->dqi_flags = (mi->dqi_flags & ~DQF_SETINFO_MASK) |
2601 				(ii->dqi_flags & DQF_SETINFO_MASK);
2602 	spin_unlock(&dq_data_lock);
2603 	mark_info_dirty(sb, type);
2604 	/* Force write to disk */
2605 	sb->dq_op->write_info(sb, type);
2606 out:
2607 	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2608 	return err;
2609 }
2610 EXPORT_SYMBOL(dquot_set_dqinfo);
2611 
2612 const struct quotactl_ops dquot_quotactl_ops = {
2613 	.quota_on	= dquot_quota_on,
2614 	.quota_off	= dquot_quota_off,
2615 	.quota_sync	= dquot_quota_sync,
2616 	.get_info	= dquot_get_dqinfo,
2617 	.set_info	= dquot_set_dqinfo,
2618 	.get_dqblk	= dquot_get_dqblk,
2619 	.set_dqblk	= dquot_set_dqblk
2620 };
2621 EXPORT_SYMBOL(dquot_quotactl_ops);
2622 
2623 static int do_proc_dqstats(struct ctl_table *table, int write,
2624 		     void __user *buffer, size_t *lenp, loff_t *ppos)
2625 {
2626 	unsigned int type = (int *)table->data - dqstats.stat;
2627 
2628 	/* Update global table */
2629 	dqstats.stat[type] =
2630 			percpu_counter_sum_positive(&dqstats.counter[type]);
2631 	return proc_dointvec(table, write, buffer, lenp, ppos);
2632 }
2633 
2634 static struct ctl_table fs_dqstats_table[] = {
2635 	{
2636 		.procname	= "lookups",
2637 		.data		= &dqstats.stat[DQST_LOOKUPS],
2638 		.maxlen		= sizeof(int),
2639 		.mode		= 0444,
2640 		.proc_handler	= do_proc_dqstats,
2641 	},
2642 	{
2643 		.procname	= "drops",
2644 		.data		= &dqstats.stat[DQST_DROPS],
2645 		.maxlen		= sizeof(int),
2646 		.mode		= 0444,
2647 		.proc_handler	= do_proc_dqstats,
2648 	},
2649 	{
2650 		.procname	= "reads",
2651 		.data		= &dqstats.stat[DQST_READS],
2652 		.maxlen		= sizeof(int),
2653 		.mode		= 0444,
2654 		.proc_handler	= do_proc_dqstats,
2655 	},
2656 	{
2657 		.procname	= "writes",
2658 		.data		= &dqstats.stat[DQST_WRITES],
2659 		.maxlen		= sizeof(int),
2660 		.mode		= 0444,
2661 		.proc_handler	= do_proc_dqstats,
2662 	},
2663 	{
2664 		.procname	= "cache_hits",
2665 		.data		= &dqstats.stat[DQST_CACHE_HITS],
2666 		.maxlen		= sizeof(int),
2667 		.mode		= 0444,
2668 		.proc_handler	= do_proc_dqstats,
2669 	},
2670 	{
2671 		.procname	= "allocated_dquots",
2672 		.data		= &dqstats.stat[DQST_ALLOC_DQUOTS],
2673 		.maxlen		= sizeof(int),
2674 		.mode		= 0444,
2675 		.proc_handler	= do_proc_dqstats,
2676 	},
2677 	{
2678 		.procname	= "free_dquots",
2679 		.data		= &dqstats.stat[DQST_FREE_DQUOTS],
2680 		.maxlen		= sizeof(int),
2681 		.mode		= 0444,
2682 		.proc_handler	= do_proc_dqstats,
2683 	},
2684 	{
2685 		.procname	= "syncs",
2686 		.data		= &dqstats.stat[DQST_SYNCS],
2687 		.maxlen		= sizeof(int),
2688 		.mode		= 0444,
2689 		.proc_handler	= do_proc_dqstats,
2690 	},
2691 #ifdef CONFIG_PRINT_QUOTA_WARNING
2692 	{
2693 		.procname	= "warnings",
2694 		.data		= &flag_print_warnings,
2695 		.maxlen		= sizeof(int),
2696 		.mode		= 0644,
2697 		.proc_handler	= proc_dointvec,
2698 	},
2699 #endif
2700 	{ },
2701 };
2702 
2703 static struct ctl_table fs_table[] = {
2704 	{
2705 		.procname	= "quota",
2706 		.mode		= 0555,
2707 		.child		= fs_dqstats_table,
2708 	},
2709 	{ },
2710 };
2711 
2712 static struct ctl_table sys_table[] = {
2713 	{
2714 		.procname	= "fs",
2715 		.mode		= 0555,
2716 		.child		= fs_table,
2717 	},
2718 	{ },
2719 };
2720 
2721 static int __init dquot_init(void)
2722 {
2723 	int i, ret;
2724 	unsigned long nr_hash, order;
2725 
2726 	printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2727 
2728 	register_sysctl_table(sys_table);
2729 
2730 	dquot_cachep = kmem_cache_create("dquot",
2731 			sizeof(struct dquot), sizeof(unsigned long) * 4,
2732 			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2733 				SLAB_MEM_SPREAD|SLAB_PANIC),
2734 			NULL);
2735 
2736 	order = 0;
2737 	dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2738 	if (!dquot_hash)
2739 		panic("Cannot create dquot hash table");
2740 
2741 	for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2742 		ret = percpu_counter_init(&dqstats.counter[i], 0);
2743 		if (ret)
2744 			panic("Cannot create dquot stat counters");
2745 	}
2746 
2747 	/* Find power-of-two hlist_heads which can fit into allocation */
2748 	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2749 	dq_hash_bits = 0;
2750 	do {
2751 		dq_hash_bits++;
2752 	} while (nr_hash >> dq_hash_bits);
2753 	dq_hash_bits--;
2754 
2755 	nr_hash = 1UL << dq_hash_bits;
2756 	dq_hash_mask = nr_hash - 1;
2757 	for (i = 0; i < nr_hash; i++)
2758 		INIT_HLIST_HEAD(dquot_hash + i);
2759 
2760 	printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2761 			nr_hash, order, (PAGE_SIZE << order));
2762 
2763 	register_shrinker(&dqcache_shrinker);
2764 
2765 	return 0;
2766 }
2767 module_init(dquot_init);
2768