xref: /openbmc/linux/fs/quota/dquot.c (revision 5bcd3b6f)
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  * Operation accessing dquots via inode pointers are protected by dquot_srcu.
100  * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
101  * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
102  * inode and before dropping dquot references to avoid use of dquots after
103  * they are freed. dq_data_lock is used to serialize the pointer setting and
104  * clearing operations.
105  * Special care needs to be taken about S_NOQUOTA inode flag (marking that
106  * inode is a quota file). Functions adding pointers from inode to dquots have
107  * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
108  * have to do all pointer modifications before dropping dq_data_lock. This makes
109  * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
110  * then drops all pointers to dquots from an inode.
111  *
112  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
113  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
114  * Currently dquot is locked only when it is being read to memory (or space for
115  * it is being allocated) on the first dqget() and when it is being released on
116  * the last dqput(). The allocation and release oparations are serialized by
117  * the dq_lock and by checking the use count in dquot_release().  Write
118  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
119  * spinlock to internal buffers before writing.
120  *
121  * Lock ordering (including related VFS locks) is the following:
122  *   dqonoff_mutex > i_mutex > journal_lock > dquot->dq_lock > dqio_mutex
123  * dqonoff_mutex > i_mutex comes from dquot_quota_sync, dquot_enable, etc.
124  */
125 
126 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
127 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
128 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
129 EXPORT_SYMBOL(dq_data_lock);
130 DEFINE_STATIC_SRCU(dquot_srcu);
131 
132 void __quota_error(struct super_block *sb, const char *func,
133 		   const char *fmt, ...)
134 {
135 	if (printk_ratelimit()) {
136 		va_list args;
137 		struct va_format vaf;
138 
139 		va_start(args, fmt);
140 
141 		vaf.fmt = fmt;
142 		vaf.va = &args;
143 
144 		printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
145 		       sb->s_id, func, &vaf);
146 
147 		va_end(args);
148 	}
149 }
150 EXPORT_SYMBOL(__quota_error);
151 
152 #if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
153 static char *quotatypes[] = INITQFNAMES;
154 #endif
155 static struct quota_format_type *quota_formats;	/* List of registered formats */
156 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
157 
158 /* SLAB cache for dquot structures */
159 static struct kmem_cache *dquot_cachep;
160 
161 int register_quota_format(struct quota_format_type *fmt)
162 {
163 	spin_lock(&dq_list_lock);
164 	fmt->qf_next = quota_formats;
165 	quota_formats = fmt;
166 	spin_unlock(&dq_list_lock);
167 	return 0;
168 }
169 EXPORT_SYMBOL(register_quota_format);
170 
171 void unregister_quota_format(struct quota_format_type *fmt)
172 {
173 	struct quota_format_type **actqf;
174 
175 	spin_lock(&dq_list_lock);
176 	for (actqf = &quota_formats; *actqf && *actqf != fmt;
177 	     actqf = &(*actqf)->qf_next)
178 		;
179 	if (*actqf)
180 		*actqf = (*actqf)->qf_next;
181 	spin_unlock(&dq_list_lock);
182 }
183 EXPORT_SYMBOL(unregister_quota_format);
184 
185 static struct quota_format_type *find_quota_format(int id)
186 {
187 	struct quota_format_type *actqf;
188 
189 	spin_lock(&dq_list_lock);
190 	for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
191 	     actqf = actqf->qf_next)
192 		;
193 	if (!actqf || !try_module_get(actqf->qf_owner)) {
194 		int qm;
195 
196 		spin_unlock(&dq_list_lock);
197 
198 		for (qm = 0; module_names[qm].qm_fmt_id &&
199 			     module_names[qm].qm_fmt_id != id; qm++)
200 			;
201 		if (!module_names[qm].qm_fmt_id ||
202 		    request_module(module_names[qm].qm_mod_name))
203 			return NULL;
204 
205 		spin_lock(&dq_list_lock);
206 		for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
207 		     actqf = actqf->qf_next)
208 			;
209 		if (actqf && !try_module_get(actqf->qf_owner))
210 			actqf = NULL;
211 	}
212 	spin_unlock(&dq_list_lock);
213 	return actqf;
214 }
215 
216 static void put_quota_format(struct quota_format_type *fmt)
217 {
218 	module_put(fmt->qf_owner);
219 }
220 
221 /*
222  * Dquot List Management:
223  * The quota code uses three lists for dquot management: the inuse_list,
224  * free_dquots, and dquot_hash[] array. A single dquot structure may be
225  * on all three lists, depending on its current state.
226  *
227  * All dquots are placed to the end of inuse_list when first created, and this
228  * list is used for invalidate operation, which must look at every dquot.
229  *
230  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
231  * and this list is searched whenever we need an available dquot.  Dquots are
232  * removed from the list as soon as they are used again, and
233  * dqstats.free_dquots gives the number of dquots on the list. When
234  * dquot is invalidated it's completely released from memory.
235  *
236  * Dquots with a specific identity (device, type and id) are placed on
237  * one of the dquot_hash[] hash chains. The provides an efficient search
238  * mechanism to locate a specific dquot.
239  */
240 
241 static LIST_HEAD(inuse_list);
242 static LIST_HEAD(free_dquots);
243 static unsigned int dq_hash_bits, dq_hash_mask;
244 static struct hlist_head *dquot_hash;
245 
246 struct dqstats dqstats;
247 EXPORT_SYMBOL(dqstats);
248 
249 static qsize_t inode_get_rsv_space(struct inode *inode);
250 static void __dquot_initialize(struct inode *inode, int type);
251 
252 static inline unsigned int
253 hashfn(const struct super_block *sb, struct kqid qid)
254 {
255 	unsigned int id = from_kqid(&init_user_ns, qid);
256 	int type = qid.type;
257 	unsigned long tmp;
258 
259 	tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
260 	return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
261 }
262 
263 /*
264  * Following list functions expect dq_list_lock to be held
265  */
266 static inline void insert_dquot_hash(struct dquot *dquot)
267 {
268 	struct hlist_head *head;
269 	head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
270 	hlist_add_head(&dquot->dq_hash, head);
271 }
272 
273 static inline void remove_dquot_hash(struct dquot *dquot)
274 {
275 	hlist_del_init(&dquot->dq_hash);
276 }
277 
278 static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
279 				struct kqid qid)
280 {
281 	struct hlist_node *node;
282 	struct dquot *dquot;
283 
284 	hlist_for_each (node, dquot_hash+hashent) {
285 		dquot = hlist_entry(node, struct dquot, dq_hash);
286 		if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
287 			return dquot;
288 	}
289 	return NULL;
290 }
291 
292 /* Add a dquot to the tail of the free list */
293 static inline void put_dquot_last(struct dquot *dquot)
294 {
295 	list_add_tail(&dquot->dq_free, &free_dquots);
296 	dqstats_inc(DQST_FREE_DQUOTS);
297 }
298 
299 static inline void remove_free_dquot(struct dquot *dquot)
300 {
301 	if (list_empty(&dquot->dq_free))
302 		return;
303 	list_del_init(&dquot->dq_free);
304 	dqstats_dec(DQST_FREE_DQUOTS);
305 }
306 
307 static inline void put_inuse(struct dquot *dquot)
308 {
309 	/* We add to the back of inuse list so we don't have to restart
310 	 * when traversing this list and we block */
311 	list_add_tail(&dquot->dq_inuse, &inuse_list);
312 	dqstats_inc(DQST_ALLOC_DQUOTS);
313 }
314 
315 static inline void remove_inuse(struct dquot *dquot)
316 {
317 	dqstats_dec(DQST_ALLOC_DQUOTS);
318 	list_del(&dquot->dq_inuse);
319 }
320 /*
321  * End of list functions needing dq_list_lock
322  */
323 
324 static void wait_on_dquot(struct dquot *dquot)
325 {
326 	mutex_lock(&dquot->dq_lock);
327 	mutex_unlock(&dquot->dq_lock);
328 }
329 
330 static inline int dquot_dirty(struct dquot *dquot)
331 {
332 	return test_bit(DQ_MOD_B, &dquot->dq_flags);
333 }
334 
335 static inline int mark_dquot_dirty(struct dquot *dquot)
336 {
337 	return dquot->dq_sb->dq_op->mark_dirty(dquot);
338 }
339 
340 /* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
341 int dquot_mark_dquot_dirty(struct dquot *dquot)
342 {
343 	int ret = 1;
344 
345 	/* If quota is dirty already, we don't have to acquire dq_list_lock */
346 	if (test_bit(DQ_MOD_B, &dquot->dq_flags))
347 		return 1;
348 
349 	spin_lock(&dq_list_lock);
350 	if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
351 		list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
352 				info[dquot->dq_id.type].dqi_dirty_list);
353 		ret = 0;
354 	}
355 	spin_unlock(&dq_list_lock);
356 	return ret;
357 }
358 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
359 
360 /* Dirtify all the dquots - this can block when journalling */
361 static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
362 {
363 	int ret, err, cnt;
364 
365 	ret = err = 0;
366 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
367 		if (dquot[cnt])
368 			/* Even in case of error we have to continue */
369 			ret = mark_dquot_dirty(dquot[cnt]);
370 		if (!err)
371 			err = ret;
372 	}
373 	return err;
374 }
375 
376 static inline void dqput_all(struct dquot **dquot)
377 {
378 	unsigned int cnt;
379 
380 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
381 		dqput(dquot[cnt]);
382 }
383 
384 /* This function needs dq_list_lock */
385 static inline int clear_dquot_dirty(struct dquot *dquot)
386 {
387 	if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
388 		return 0;
389 	list_del_init(&dquot->dq_dirty);
390 	return 1;
391 }
392 
393 void mark_info_dirty(struct super_block *sb, int type)
394 {
395 	set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
396 }
397 EXPORT_SYMBOL(mark_info_dirty);
398 
399 /*
400  *	Read dquot from disk and alloc space for it
401  */
402 
403 int dquot_acquire(struct dquot *dquot)
404 {
405 	int ret = 0, ret2 = 0;
406 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
407 
408 	mutex_lock(&dquot->dq_lock);
409 	mutex_lock(&dqopt->dqio_mutex);
410 	if (!test_bit(DQ_READ_B, &dquot->dq_flags))
411 		ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
412 	if (ret < 0)
413 		goto out_iolock;
414 	set_bit(DQ_READ_B, &dquot->dq_flags);
415 	/* Instantiate dquot if needed */
416 	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
417 		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
418 		/* Write the info if needed */
419 		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
420 			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
421 					dquot->dq_sb, dquot->dq_id.type);
422 		}
423 		if (ret < 0)
424 			goto out_iolock;
425 		if (ret2 < 0) {
426 			ret = ret2;
427 			goto out_iolock;
428 		}
429 	}
430 	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
431 out_iolock:
432 	mutex_unlock(&dqopt->dqio_mutex);
433 	mutex_unlock(&dquot->dq_lock);
434 	return ret;
435 }
436 EXPORT_SYMBOL(dquot_acquire);
437 
438 /*
439  *	Write dquot to disk
440  */
441 int dquot_commit(struct dquot *dquot)
442 {
443 	int ret = 0;
444 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
445 
446 	mutex_lock(&dqopt->dqio_mutex);
447 	spin_lock(&dq_list_lock);
448 	if (!clear_dquot_dirty(dquot)) {
449 		spin_unlock(&dq_list_lock);
450 		goto out_sem;
451 	}
452 	spin_unlock(&dq_list_lock);
453 	/* Inactive dquot can be only if there was error during read/init
454 	 * => we have better not writing it */
455 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
456 		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
457 	else
458 		ret = -EIO;
459 out_sem:
460 	mutex_unlock(&dqopt->dqio_mutex);
461 	return ret;
462 }
463 EXPORT_SYMBOL(dquot_commit);
464 
465 /*
466  *	Release dquot
467  */
468 int dquot_release(struct dquot *dquot)
469 {
470 	int ret = 0, ret2 = 0;
471 	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
472 
473 	mutex_lock(&dquot->dq_lock);
474 	/* Check whether we are not racing with some other dqget() */
475 	if (atomic_read(&dquot->dq_count) > 1)
476 		goto out_dqlock;
477 	mutex_lock(&dqopt->dqio_mutex);
478 	if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
479 		ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
480 		/* Write the info */
481 		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
482 			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
483 						dquot->dq_sb, dquot->dq_id.type);
484 		}
485 		if (ret >= 0)
486 			ret = ret2;
487 	}
488 	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
489 	mutex_unlock(&dqopt->dqio_mutex);
490 out_dqlock:
491 	mutex_unlock(&dquot->dq_lock);
492 	return ret;
493 }
494 EXPORT_SYMBOL(dquot_release);
495 
496 void dquot_destroy(struct dquot *dquot)
497 {
498 	kmem_cache_free(dquot_cachep, dquot);
499 }
500 EXPORT_SYMBOL(dquot_destroy);
501 
502 static inline void do_destroy_dquot(struct dquot *dquot)
503 {
504 	dquot->dq_sb->dq_op->destroy_dquot(dquot);
505 }
506 
507 /* Invalidate all dquots on the list. Note that this function is called after
508  * quota is disabled and pointers from inodes removed so there cannot be new
509  * quota users. There can still be some users of quotas due to inodes being
510  * just deleted or pruned by prune_icache() (those are not attached to any
511  * list) or parallel quotactl call. We have to wait for such users.
512  */
513 static void invalidate_dquots(struct super_block *sb, int type)
514 {
515 	struct dquot *dquot, *tmp;
516 
517 restart:
518 	spin_lock(&dq_list_lock);
519 	list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
520 		if (dquot->dq_sb != sb)
521 			continue;
522 		if (dquot->dq_id.type != type)
523 			continue;
524 		/* Wait for dquot users */
525 		if (atomic_read(&dquot->dq_count)) {
526 			DEFINE_WAIT(wait);
527 
528 			dqgrab(dquot);
529 			prepare_to_wait(&dquot->dq_wait_unused, &wait,
530 					TASK_UNINTERRUPTIBLE);
531 			spin_unlock(&dq_list_lock);
532 			/* Once dqput() wakes us up, we know it's time to free
533 			 * the dquot.
534 			 * IMPORTANT: we rely on the fact that there is always
535 			 * at most one process waiting for dquot to free.
536 			 * Otherwise dq_count would be > 1 and we would never
537 			 * wake up.
538 			 */
539 			if (atomic_read(&dquot->dq_count) > 1)
540 				schedule();
541 			finish_wait(&dquot->dq_wait_unused, &wait);
542 			dqput(dquot);
543 			/* At this moment dquot() need not exist (it could be
544 			 * reclaimed by prune_dqcache(). Hence we must
545 			 * restart. */
546 			goto restart;
547 		}
548 		/*
549 		 * Quota now has no users and it has been written on last
550 		 * dqput()
551 		 */
552 		remove_dquot_hash(dquot);
553 		remove_free_dquot(dquot);
554 		remove_inuse(dquot);
555 		do_destroy_dquot(dquot);
556 	}
557 	spin_unlock(&dq_list_lock);
558 }
559 
560 /* Call callback for every active dquot on given filesystem */
561 int dquot_scan_active(struct super_block *sb,
562 		      int (*fn)(struct dquot *dquot, unsigned long priv),
563 		      unsigned long priv)
564 {
565 	struct dquot *dquot, *old_dquot = NULL;
566 	int ret = 0;
567 
568 	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
569 	spin_lock(&dq_list_lock);
570 	list_for_each_entry(dquot, &inuse_list, dq_inuse) {
571 		if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
572 			continue;
573 		if (dquot->dq_sb != sb)
574 			continue;
575 		/* Now we have active dquot so we can just increase use count */
576 		atomic_inc(&dquot->dq_count);
577 		spin_unlock(&dq_list_lock);
578 		dqstats_inc(DQST_LOOKUPS);
579 		dqput(old_dquot);
580 		old_dquot = dquot;
581 		/*
582 		 * ->release_dquot() can be racing with us. Our reference
583 		 * protects us from new calls to it so just wait for any
584 		 * outstanding call and recheck the DQ_ACTIVE_B after that.
585 		 */
586 		wait_on_dquot(dquot);
587 		if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
588 			ret = fn(dquot, priv);
589 			if (ret < 0)
590 				goto out;
591 		}
592 		spin_lock(&dq_list_lock);
593 		/* We are safe to continue now because our dquot could not
594 		 * be moved out of the inuse list while we hold the reference */
595 	}
596 	spin_unlock(&dq_list_lock);
597 out:
598 	dqput(old_dquot);
599 	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
600 	return ret;
601 }
602 EXPORT_SYMBOL(dquot_scan_active);
603 
604 /* Write all dquot structures to quota files */
605 int dquot_writeback_dquots(struct super_block *sb, int type)
606 {
607 	struct list_head *dirty;
608 	struct dquot *dquot;
609 	struct quota_info *dqopt = sb_dqopt(sb);
610 	int cnt;
611 	int err, ret = 0;
612 
613 	mutex_lock(&dqopt->dqonoff_mutex);
614 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
615 		if (type != -1 && cnt != type)
616 			continue;
617 		if (!sb_has_quota_active(sb, cnt))
618 			continue;
619 		spin_lock(&dq_list_lock);
620 		dirty = &dqopt->info[cnt].dqi_dirty_list;
621 		while (!list_empty(dirty)) {
622 			dquot = list_first_entry(dirty, struct dquot,
623 						 dq_dirty);
624 			/* Dirty and inactive can be only bad dquot... */
625 			if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
626 				clear_dquot_dirty(dquot);
627 				continue;
628 			}
629 			/* Now we have active dquot from which someone is
630  			 * holding reference so we can safely just increase
631 			 * use count */
632 			dqgrab(dquot);
633 			spin_unlock(&dq_list_lock);
634 			dqstats_inc(DQST_LOOKUPS);
635 			err = sb->dq_op->write_dquot(dquot);
636 			if (!ret && err)
637 				ret = err;
638 			dqput(dquot);
639 			spin_lock(&dq_list_lock);
640 		}
641 		spin_unlock(&dq_list_lock);
642 	}
643 
644 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
645 		if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
646 		    && info_dirty(&dqopt->info[cnt]))
647 			sb->dq_op->write_info(sb, cnt);
648 	dqstats_inc(DQST_SYNCS);
649 	mutex_unlock(&dqopt->dqonoff_mutex);
650 
651 	return ret;
652 }
653 EXPORT_SYMBOL(dquot_writeback_dquots);
654 
655 /* Write all dquot structures to disk and make them visible from userspace */
656 int dquot_quota_sync(struct super_block *sb, int type)
657 {
658 	struct quota_info *dqopt = sb_dqopt(sb);
659 	int cnt;
660 	int ret;
661 
662 	ret = dquot_writeback_dquots(sb, type);
663 	if (ret)
664 		return ret;
665 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
666 		return 0;
667 
668 	/* This is not very clever (and fast) but currently I don't know about
669 	 * any other simple way of getting quota data to disk and we must get
670 	 * them there for userspace to be visible... */
671 	if (sb->s_op->sync_fs)
672 		sb->s_op->sync_fs(sb, 1);
673 	sync_blockdev(sb->s_bdev);
674 
675 	/*
676 	 * Now when everything is written we can discard the pagecache so
677 	 * that userspace sees the changes.
678 	 */
679 	mutex_lock(&dqopt->dqonoff_mutex);
680 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
681 		if (type != -1 && cnt != type)
682 			continue;
683 		if (!sb_has_quota_active(sb, cnt))
684 			continue;
685 		mutex_lock(&dqopt->files[cnt]->i_mutex);
686 		truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
687 		mutex_unlock(&dqopt->files[cnt]->i_mutex);
688 	}
689 	mutex_unlock(&dqopt->dqonoff_mutex);
690 
691 	return 0;
692 }
693 EXPORT_SYMBOL(dquot_quota_sync);
694 
695 static unsigned long
696 dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
697 {
698 	struct list_head *head;
699 	struct dquot *dquot;
700 	unsigned long freed = 0;
701 
702 	spin_lock(&dq_list_lock);
703 	head = free_dquots.prev;
704 	while (head != &free_dquots && sc->nr_to_scan) {
705 		dquot = list_entry(head, struct dquot, dq_free);
706 		remove_dquot_hash(dquot);
707 		remove_free_dquot(dquot);
708 		remove_inuse(dquot);
709 		do_destroy_dquot(dquot);
710 		sc->nr_to_scan--;
711 		freed++;
712 		head = free_dquots.prev;
713 	}
714 	spin_unlock(&dq_list_lock);
715 	return freed;
716 }
717 
718 static unsigned long
719 dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
720 {
721 	return vfs_pressure_ratio(
722 	percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
723 }
724 
725 static struct shrinker dqcache_shrinker = {
726 	.count_objects = dqcache_shrink_count,
727 	.scan_objects = dqcache_shrink_scan,
728 	.seeks = DEFAULT_SEEKS,
729 };
730 
731 /*
732  * Put reference to dquot
733  */
734 void dqput(struct dquot *dquot)
735 {
736 	int ret;
737 
738 	if (!dquot)
739 		return;
740 #ifdef CONFIG_QUOTA_DEBUG
741 	if (!atomic_read(&dquot->dq_count)) {
742 		quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
743 			    quotatypes[dquot->dq_id.type],
744 			    from_kqid(&init_user_ns, dquot->dq_id));
745 		BUG();
746 	}
747 #endif
748 	dqstats_inc(DQST_DROPS);
749 we_slept:
750 	spin_lock(&dq_list_lock);
751 	if (atomic_read(&dquot->dq_count) > 1) {
752 		/* We have more than one user... nothing to do */
753 		atomic_dec(&dquot->dq_count);
754 		/* Releasing dquot during quotaoff phase? */
755 		if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
756 		    atomic_read(&dquot->dq_count) == 1)
757 			wake_up(&dquot->dq_wait_unused);
758 		spin_unlock(&dq_list_lock);
759 		return;
760 	}
761 	/* Need to release dquot? */
762 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
763 		spin_unlock(&dq_list_lock);
764 		/* Commit dquot before releasing */
765 		ret = dquot->dq_sb->dq_op->write_dquot(dquot);
766 		if (ret < 0) {
767 			quota_error(dquot->dq_sb, "Can't write quota structure"
768 				    " (error %d). Quota may get out of sync!",
769 				    ret);
770 			/*
771 			 * We clear dirty bit anyway, so that we avoid
772 			 * infinite loop here
773 			 */
774 			spin_lock(&dq_list_lock);
775 			clear_dquot_dirty(dquot);
776 			spin_unlock(&dq_list_lock);
777 		}
778 		goto we_slept;
779 	}
780 	/* Clear flag in case dquot was inactive (something bad happened) */
781 	clear_dquot_dirty(dquot);
782 	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
783 		spin_unlock(&dq_list_lock);
784 		dquot->dq_sb->dq_op->release_dquot(dquot);
785 		goto we_slept;
786 	}
787 	atomic_dec(&dquot->dq_count);
788 #ifdef CONFIG_QUOTA_DEBUG
789 	/* sanity check */
790 	BUG_ON(!list_empty(&dquot->dq_free));
791 #endif
792 	put_dquot_last(dquot);
793 	spin_unlock(&dq_list_lock);
794 }
795 EXPORT_SYMBOL(dqput);
796 
797 struct dquot *dquot_alloc(struct super_block *sb, int type)
798 {
799 	return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
800 }
801 EXPORT_SYMBOL(dquot_alloc);
802 
803 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
804 {
805 	struct dquot *dquot;
806 
807 	dquot = sb->dq_op->alloc_dquot(sb, type);
808 	if(!dquot)
809 		return NULL;
810 
811 	mutex_init(&dquot->dq_lock);
812 	INIT_LIST_HEAD(&dquot->dq_free);
813 	INIT_LIST_HEAD(&dquot->dq_inuse);
814 	INIT_HLIST_NODE(&dquot->dq_hash);
815 	INIT_LIST_HEAD(&dquot->dq_dirty);
816 	init_waitqueue_head(&dquot->dq_wait_unused);
817 	dquot->dq_sb = sb;
818 	dquot->dq_id = make_kqid_invalid(type);
819 	atomic_set(&dquot->dq_count, 1);
820 
821 	return dquot;
822 }
823 
824 /*
825  * Get reference to dquot
826  *
827  * Locking is slightly tricky here. We are guarded from parallel quotaoff()
828  * destroying our dquot by:
829  *   a) checking for quota flags under dq_list_lock and
830  *   b) getting a reference to dquot before we release dq_list_lock
831  */
832 struct dquot *dqget(struct super_block *sb, struct kqid qid)
833 {
834 	unsigned int hashent = hashfn(sb, qid);
835 	struct dquot *dquot = NULL, *empty = NULL;
836 
837         if (!sb_has_quota_active(sb, qid.type))
838 		return NULL;
839 we_slept:
840 	spin_lock(&dq_list_lock);
841 	spin_lock(&dq_state_lock);
842 	if (!sb_has_quota_active(sb, qid.type)) {
843 		spin_unlock(&dq_state_lock);
844 		spin_unlock(&dq_list_lock);
845 		goto out;
846 	}
847 	spin_unlock(&dq_state_lock);
848 
849 	dquot = find_dquot(hashent, sb, qid);
850 	if (!dquot) {
851 		if (!empty) {
852 			spin_unlock(&dq_list_lock);
853 			empty = get_empty_dquot(sb, qid.type);
854 			if (!empty)
855 				schedule();	/* Try to wait for a moment... */
856 			goto we_slept;
857 		}
858 		dquot = empty;
859 		empty = NULL;
860 		dquot->dq_id = qid;
861 		/* all dquots go on the inuse_list */
862 		put_inuse(dquot);
863 		/* hash it first so it can be found */
864 		insert_dquot_hash(dquot);
865 		spin_unlock(&dq_list_lock);
866 		dqstats_inc(DQST_LOOKUPS);
867 	} else {
868 		if (!atomic_read(&dquot->dq_count))
869 			remove_free_dquot(dquot);
870 		atomic_inc(&dquot->dq_count);
871 		spin_unlock(&dq_list_lock);
872 		dqstats_inc(DQST_CACHE_HITS);
873 		dqstats_inc(DQST_LOOKUPS);
874 	}
875 	/* Wait for dq_lock - after this we know that either dquot_release() is
876 	 * already finished or it will be canceled due to dq_count > 1 test */
877 	wait_on_dquot(dquot);
878 	/* Read the dquot / allocate space in quota file */
879 	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
880 	    sb->dq_op->acquire_dquot(dquot) < 0) {
881 		dqput(dquot);
882 		dquot = NULL;
883 		goto out;
884 	}
885 #ifdef CONFIG_QUOTA_DEBUG
886 	BUG_ON(!dquot->dq_sb);	/* Has somebody invalidated entry under us? */
887 #endif
888 out:
889 	if (empty)
890 		do_destroy_dquot(empty);
891 
892 	return dquot;
893 }
894 EXPORT_SYMBOL(dqget);
895 
896 static inline struct dquot **i_dquot(struct inode *inode)
897 {
898 	return inode->i_sb->s_op->get_dquots(inode);
899 }
900 
901 static int dqinit_needed(struct inode *inode, int type)
902 {
903 	struct dquot * const *dquots;
904 	int cnt;
905 
906 	if (IS_NOQUOTA(inode))
907 		return 0;
908 
909 	dquots = i_dquot(inode);
910 	if (type != -1)
911 		return !dquots[type];
912 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
913 		if (!dquots[cnt])
914 			return 1;
915 	return 0;
916 }
917 
918 /* This routine is guarded by dqonoff_mutex mutex */
919 static void add_dquot_ref(struct super_block *sb, int type)
920 {
921 	struct inode *inode, *old_inode = NULL;
922 #ifdef CONFIG_QUOTA_DEBUG
923 	int reserved = 0;
924 #endif
925 
926 	spin_lock(&inode_sb_list_lock);
927 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
928 		spin_lock(&inode->i_lock);
929 		if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
930 		    !atomic_read(&inode->i_writecount) ||
931 		    !dqinit_needed(inode, type)) {
932 			spin_unlock(&inode->i_lock);
933 			continue;
934 		}
935 		__iget(inode);
936 		spin_unlock(&inode->i_lock);
937 		spin_unlock(&inode_sb_list_lock);
938 
939 #ifdef CONFIG_QUOTA_DEBUG
940 		if (unlikely(inode_get_rsv_space(inode) > 0))
941 			reserved = 1;
942 #endif
943 		iput(old_inode);
944 		__dquot_initialize(inode, type);
945 
946 		/*
947 		 * We hold a reference to 'inode' so it couldn't have been
948 		 * removed from s_inodes list while we dropped the
949 		 * inode_sb_list_lock We cannot iput the inode now as we can be
950 		 * holding the last reference and we cannot iput it under
951 		 * inode_sb_list_lock. So we keep the reference and iput it
952 		 * later.
953 		 */
954 		old_inode = inode;
955 		spin_lock(&inode_sb_list_lock);
956 	}
957 	spin_unlock(&inode_sb_list_lock);
958 	iput(old_inode);
959 
960 #ifdef CONFIG_QUOTA_DEBUG
961 	if (reserved) {
962 		quota_error(sb, "Writes happened before quota was turned on "
963 			"thus quota information is probably inconsistent. "
964 			"Please run quotacheck(8)");
965 	}
966 #endif
967 }
968 
969 /*
970  * Remove references to dquots from inode and add dquot to list for freeing
971  * if we have the last reference to dquot
972  */
973 static void remove_inode_dquot_ref(struct inode *inode, int type,
974 				   struct list_head *tofree_head)
975 {
976 	struct dquot **dquots = i_dquot(inode);
977 	struct dquot *dquot = dquots[type];
978 
979 	if (!dquot)
980 		return;
981 
982 	dquots[type] = NULL;
983 	if (list_empty(&dquot->dq_free)) {
984 		/*
985 		 * The inode still has reference to dquot so it can't be in the
986 		 * free list
987 		 */
988 		spin_lock(&dq_list_lock);
989 		list_add(&dquot->dq_free, tofree_head);
990 		spin_unlock(&dq_list_lock);
991 	} else {
992 		/*
993 		 * Dquot is already in a list to put so we won't drop the last
994 		 * reference here.
995 		 */
996 		dqput(dquot);
997 	}
998 }
999 
1000 /*
1001  * Free list of dquots
1002  * Dquots are removed from inodes and no new references can be got so we are
1003  * the only ones holding reference
1004  */
1005 static void put_dquot_list(struct list_head *tofree_head)
1006 {
1007 	struct list_head *act_head;
1008 	struct dquot *dquot;
1009 
1010 	act_head = tofree_head->next;
1011 	while (act_head != tofree_head) {
1012 		dquot = list_entry(act_head, struct dquot, dq_free);
1013 		act_head = act_head->next;
1014 		/* Remove dquot from the list so we won't have problems... */
1015 		list_del_init(&dquot->dq_free);
1016 		dqput(dquot);
1017 	}
1018 }
1019 
1020 static void remove_dquot_ref(struct super_block *sb, int type,
1021 		struct list_head *tofree_head)
1022 {
1023 	struct inode *inode;
1024 	int reserved = 0;
1025 
1026 	spin_lock(&inode_sb_list_lock);
1027 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1028 		/*
1029 		 *  We have to scan also I_NEW inodes because they can already
1030 		 *  have quota pointer initialized. Luckily, we need to touch
1031 		 *  only quota pointers and these have separate locking
1032 		 *  (dq_data_lock).
1033 		 */
1034 		spin_lock(&dq_data_lock);
1035 		if (!IS_NOQUOTA(inode)) {
1036 			if (unlikely(inode_get_rsv_space(inode) > 0))
1037 				reserved = 1;
1038 			remove_inode_dquot_ref(inode, type, tofree_head);
1039 		}
1040 		spin_unlock(&dq_data_lock);
1041 	}
1042 	spin_unlock(&inode_sb_list_lock);
1043 #ifdef CONFIG_QUOTA_DEBUG
1044 	if (reserved) {
1045 		printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1046 			" was disabled thus quota information is probably "
1047 			"inconsistent. Please run quotacheck(8).\n", sb->s_id);
1048 	}
1049 #endif
1050 }
1051 
1052 /* Gather all references from inodes and drop them */
1053 static void drop_dquot_ref(struct super_block *sb, int type)
1054 {
1055 	LIST_HEAD(tofree_head);
1056 
1057 	if (sb->dq_op) {
1058 		remove_dquot_ref(sb, type, &tofree_head);
1059 		synchronize_srcu(&dquot_srcu);
1060 		put_dquot_list(&tofree_head);
1061 	}
1062 }
1063 
1064 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1065 {
1066 	dquot->dq_dqb.dqb_curinodes += number;
1067 }
1068 
1069 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1070 {
1071 	dquot->dq_dqb.dqb_curspace += number;
1072 }
1073 
1074 static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1075 {
1076 	dquot->dq_dqb.dqb_rsvspace += number;
1077 }
1078 
1079 /*
1080  * Claim reserved quota space
1081  */
1082 static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1083 {
1084 	if (dquot->dq_dqb.dqb_rsvspace < number) {
1085 		WARN_ON_ONCE(1);
1086 		number = dquot->dq_dqb.dqb_rsvspace;
1087 	}
1088 	dquot->dq_dqb.dqb_curspace += number;
1089 	dquot->dq_dqb.dqb_rsvspace -= number;
1090 }
1091 
1092 static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1093 {
1094 	if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1095 		number = dquot->dq_dqb.dqb_curspace;
1096 	dquot->dq_dqb.dqb_rsvspace += number;
1097 	dquot->dq_dqb.dqb_curspace -= number;
1098 }
1099 
1100 static inline
1101 void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1102 {
1103 	if (dquot->dq_dqb.dqb_rsvspace >= number)
1104 		dquot->dq_dqb.dqb_rsvspace -= number;
1105 	else {
1106 		WARN_ON_ONCE(1);
1107 		dquot->dq_dqb.dqb_rsvspace = 0;
1108 	}
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 = (time_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_bsoftlimit)
1131 		dquot->dq_dqb.dqb_btime = (time_t) 0;
1132 	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1133 }
1134 
1135 struct dquot_warn {
1136 	struct super_block *w_sb;
1137 	struct kqid w_dq_id;
1138 	short w_type;
1139 };
1140 
1141 static int warning_issued(struct dquot *dquot, const int warntype)
1142 {
1143 	int flag = (warntype == QUOTA_NL_BHARDWARN ||
1144 		warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1145 		((warntype == QUOTA_NL_IHARDWARN ||
1146 		warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1147 
1148 	if (!flag)
1149 		return 0;
1150 	return test_and_set_bit(flag, &dquot->dq_flags);
1151 }
1152 
1153 #ifdef CONFIG_PRINT_QUOTA_WARNING
1154 static int flag_print_warnings = 1;
1155 
1156 static int need_print_warning(struct dquot_warn *warn)
1157 {
1158 	if (!flag_print_warnings)
1159 		return 0;
1160 
1161 	switch (warn->w_dq_id.type) {
1162 		case USRQUOTA:
1163 			return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1164 		case GRPQUOTA:
1165 			return in_group_p(warn->w_dq_id.gid);
1166 		case PRJQUOTA:	/* Never taken... Just make gcc happy */
1167 			return 0;
1168 	}
1169 	return 0;
1170 }
1171 
1172 /* Print warning to user which exceeded quota */
1173 static void print_warning(struct dquot_warn *warn)
1174 {
1175 	char *msg = NULL;
1176 	struct tty_struct *tty;
1177 	int warntype = warn->w_type;
1178 
1179 	if (warntype == QUOTA_NL_IHARDBELOW ||
1180 	    warntype == QUOTA_NL_ISOFTBELOW ||
1181 	    warntype == QUOTA_NL_BHARDBELOW ||
1182 	    warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1183 		return;
1184 
1185 	tty = get_current_tty();
1186 	if (!tty)
1187 		return;
1188 	tty_write_message(tty, warn->w_sb->s_id);
1189 	if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1190 		tty_write_message(tty, ": warning, ");
1191 	else
1192 		tty_write_message(tty, ": write failed, ");
1193 	tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1194 	switch (warntype) {
1195 		case QUOTA_NL_IHARDWARN:
1196 			msg = " file limit reached.\r\n";
1197 			break;
1198 		case QUOTA_NL_ISOFTLONGWARN:
1199 			msg = " file quota exceeded too long.\r\n";
1200 			break;
1201 		case QUOTA_NL_ISOFTWARN:
1202 			msg = " file quota exceeded.\r\n";
1203 			break;
1204 		case QUOTA_NL_BHARDWARN:
1205 			msg = " block limit reached.\r\n";
1206 			break;
1207 		case QUOTA_NL_BSOFTLONGWARN:
1208 			msg = " block quota exceeded too long.\r\n";
1209 			break;
1210 		case QUOTA_NL_BSOFTWARN:
1211 			msg = " block quota exceeded.\r\n";
1212 			break;
1213 	}
1214 	tty_write_message(tty, msg);
1215 	tty_kref_put(tty);
1216 }
1217 #endif
1218 
1219 static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1220 			    int warntype)
1221 {
1222 	if (warning_issued(dquot, warntype))
1223 		return;
1224 	warn->w_type = warntype;
1225 	warn->w_sb = dquot->dq_sb;
1226 	warn->w_dq_id = dquot->dq_id;
1227 }
1228 
1229 /*
1230  * Write warnings to the console and send warning messages over netlink.
1231  *
1232  * Note that this function can call into tty and networking code.
1233  */
1234 static void flush_warnings(struct dquot_warn *warn)
1235 {
1236 	int i;
1237 
1238 	for (i = 0; i < MAXQUOTAS; i++) {
1239 		if (warn[i].w_type == QUOTA_NL_NOWARN)
1240 			continue;
1241 #ifdef CONFIG_PRINT_QUOTA_WARNING
1242 		print_warning(&warn[i]);
1243 #endif
1244 		quota_send_warning(warn[i].w_dq_id,
1245 				   warn[i].w_sb->s_dev, warn[i].w_type);
1246 	}
1247 }
1248 
1249 static int ignore_hardlimit(struct dquot *dquot)
1250 {
1251 	struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1252 
1253 	return capable(CAP_SYS_RESOURCE) &&
1254 	       (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1255 		!(info->dqi_flags & DQF_ROOT_SQUASH));
1256 }
1257 
1258 /* needs dq_data_lock */
1259 static int check_idq(struct dquot *dquot, qsize_t inodes,
1260 		     struct dquot_warn *warn)
1261 {
1262 	qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1263 
1264 	if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1265 	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1266 		return 0;
1267 
1268 	if (dquot->dq_dqb.dqb_ihardlimit &&
1269 	    newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1270             !ignore_hardlimit(dquot)) {
1271 		prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1272 		return -EDQUOT;
1273 	}
1274 
1275 	if (dquot->dq_dqb.dqb_isoftlimit &&
1276 	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1277 	    dquot->dq_dqb.dqb_itime &&
1278 	    get_seconds() >= dquot->dq_dqb.dqb_itime &&
1279             !ignore_hardlimit(dquot)) {
1280 		prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1281 		return -EDQUOT;
1282 	}
1283 
1284 	if (dquot->dq_dqb.dqb_isoftlimit &&
1285 	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1286 	    dquot->dq_dqb.dqb_itime == 0) {
1287 		prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1288 		dquot->dq_dqb.dqb_itime = get_seconds() +
1289 		    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1290 	}
1291 
1292 	return 0;
1293 }
1294 
1295 /* needs dq_data_lock */
1296 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1297 		     struct dquot_warn *warn)
1298 {
1299 	qsize_t tspace;
1300 	struct super_block *sb = dquot->dq_sb;
1301 
1302 	if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1303 	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1304 		return 0;
1305 
1306 	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1307 		+ space;
1308 
1309 	if (dquot->dq_dqb.dqb_bhardlimit &&
1310 	    tspace > dquot->dq_dqb.dqb_bhardlimit &&
1311             !ignore_hardlimit(dquot)) {
1312 		if (!prealloc)
1313 			prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1314 		return -EDQUOT;
1315 	}
1316 
1317 	if (dquot->dq_dqb.dqb_bsoftlimit &&
1318 	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1319 	    dquot->dq_dqb.dqb_btime &&
1320 	    get_seconds() >= dquot->dq_dqb.dqb_btime &&
1321             !ignore_hardlimit(dquot)) {
1322 		if (!prealloc)
1323 			prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1324 		return -EDQUOT;
1325 	}
1326 
1327 	if (dquot->dq_dqb.dqb_bsoftlimit &&
1328 	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1329 	    dquot->dq_dqb.dqb_btime == 0) {
1330 		if (!prealloc) {
1331 			prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1332 			dquot->dq_dqb.dqb_btime = get_seconds() +
1333 			    sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1334 		}
1335 		else
1336 			/*
1337 			 * We don't allow preallocation to exceed softlimit so exceeding will
1338 			 * be always printed
1339 			 */
1340 			return -EDQUOT;
1341 	}
1342 
1343 	return 0;
1344 }
1345 
1346 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1347 {
1348 	qsize_t newinodes;
1349 
1350 	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1351 	    dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1352 	    !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1353 		return QUOTA_NL_NOWARN;
1354 
1355 	newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1356 	if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1357 		return QUOTA_NL_ISOFTBELOW;
1358 	if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1359 	    newinodes < dquot->dq_dqb.dqb_ihardlimit)
1360 		return QUOTA_NL_IHARDBELOW;
1361 	return QUOTA_NL_NOWARN;
1362 }
1363 
1364 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1365 {
1366 	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1367 	    dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1368 		return QUOTA_NL_NOWARN;
1369 
1370 	if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1371 		return QUOTA_NL_BSOFTBELOW;
1372 	if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1373 	    dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1374 		return QUOTA_NL_BHARDBELOW;
1375 	return QUOTA_NL_NOWARN;
1376 }
1377 
1378 static int dquot_active(const struct inode *inode)
1379 {
1380 	struct super_block *sb = inode->i_sb;
1381 
1382 	if (IS_NOQUOTA(inode))
1383 		return 0;
1384 	return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1385 }
1386 
1387 /*
1388  * Initialize quota pointers in inode
1389  *
1390  * It is better to call this function outside of any transaction as it
1391  * might need a lot of space in journal for dquot structure allocation.
1392  */
1393 static void __dquot_initialize(struct inode *inode, int type)
1394 {
1395 	int cnt, init_needed = 0;
1396 	struct dquot **dquots, *got[MAXQUOTAS];
1397 	struct super_block *sb = inode->i_sb;
1398 	qsize_t rsv;
1399 
1400 	if (!dquot_active(inode))
1401 		return;
1402 
1403 	dquots = i_dquot(inode);
1404 
1405 	/* First get references to structures we might need. */
1406 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1407 		struct kqid qid;
1408 		got[cnt] = NULL;
1409 		if (type != -1 && cnt != type)
1410 			continue;
1411 		/*
1412 		 * The i_dquot should have been initialized in most cases,
1413 		 * we check it without locking here to avoid unnecessary
1414 		 * dqget()/dqput() calls.
1415 		 */
1416 		if (dquots[cnt])
1417 			continue;
1418 		init_needed = 1;
1419 
1420 		switch (cnt) {
1421 		case USRQUOTA:
1422 			qid = make_kqid_uid(inode->i_uid);
1423 			break;
1424 		case GRPQUOTA:
1425 			qid = make_kqid_gid(inode->i_gid);
1426 			break;
1427 		}
1428 		got[cnt] = dqget(sb, qid);
1429 	}
1430 
1431 	/* All required i_dquot has been initialized */
1432 	if (!init_needed)
1433 		return;
1434 
1435 	spin_lock(&dq_data_lock);
1436 	if (IS_NOQUOTA(inode))
1437 		goto out_err;
1438 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1439 		if (type != -1 && cnt != type)
1440 			continue;
1441 		/* Avoid races with quotaoff() */
1442 		if (!sb_has_quota_active(sb, cnt))
1443 			continue;
1444 		/* We could race with quotaon or dqget() could have failed */
1445 		if (!got[cnt])
1446 			continue;
1447 		if (!dquots[cnt]) {
1448 			dquots[cnt] = got[cnt];
1449 			got[cnt] = NULL;
1450 			/*
1451 			 * Make quota reservation system happy if someone
1452 			 * did a write before quota was turned on
1453 			 */
1454 			rsv = inode_get_rsv_space(inode);
1455 			if (unlikely(rsv))
1456 				dquot_resv_space(dquots[cnt], rsv);
1457 		}
1458 	}
1459 out_err:
1460 	spin_unlock(&dq_data_lock);
1461 	/* Drop unused references */
1462 	dqput_all(got);
1463 }
1464 
1465 void dquot_initialize(struct inode *inode)
1466 {
1467 	__dquot_initialize(inode, -1);
1468 }
1469 EXPORT_SYMBOL(dquot_initialize);
1470 
1471 /*
1472  * Release all quotas referenced by inode.
1473  *
1474  * This function only be called on inode free or converting
1475  * a file to quota file, no other users for the i_dquot in
1476  * both cases, so we needn't call synchronize_srcu() after
1477  * clearing i_dquot.
1478  */
1479 static void __dquot_drop(struct inode *inode)
1480 {
1481 	int cnt;
1482 	struct dquot **dquots = i_dquot(inode);
1483 	struct dquot *put[MAXQUOTAS];
1484 
1485 	spin_lock(&dq_data_lock);
1486 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1487 		put[cnt] = dquots[cnt];
1488 		dquots[cnt] = NULL;
1489 	}
1490 	spin_unlock(&dq_data_lock);
1491 	dqput_all(put);
1492 }
1493 
1494 void dquot_drop(struct inode *inode)
1495 {
1496 	struct dquot * const *dquots;
1497 	int cnt;
1498 
1499 	if (IS_NOQUOTA(inode))
1500 		return;
1501 
1502 	/*
1503 	 * Test before calling to rule out calls from proc and such
1504 	 * where we are not allowed to block. Note that this is
1505 	 * actually reliable test even without the lock - the caller
1506 	 * must assure that nobody can come after the DQUOT_DROP and
1507 	 * add quota pointers back anyway.
1508 	 */
1509 	dquots = i_dquot(inode);
1510 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1511 		if (dquots[cnt])
1512 			break;
1513 	}
1514 
1515 	if (cnt < MAXQUOTAS)
1516 		__dquot_drop(inode);
1517 }
1518 EXPORT_SYMBOL(dquot_drop);
1519 
1520 /*
1521  * inode_reserved_space is managed internally by quota, and protected by
1522  * i_lock similar to i_blocks+i_bytes.
1523  */
1524 static qsize_t *inode_reserved_space(struct inode * inode)
1525 {
1526 	/* Filesystem must explicitly define it's own method in order to use
1527 	 * quota reservation interface */
1528 	BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1529 	return inode->i_sb->dq_op->get_reserved_space(inode);
1530 }
1531 
1532 void inode_add_rsv_space(struct inode *inode, qsize_t number)
1533 {
1534 	spin_lock(&inode->i_lock);
1535 	*inode_reserved_space(inode) += number;
1536 	spin_unlock(&inode->i_lock);
1537 }
1538 EXPORT_SYMBOL(inode_add_rsv_space);
1539 
1540 void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1541 {
1542 	spin_lock(&inode->i_lock);
1543 	*inode_reserved_space(inode) -= number;
1544 	__inode_add_bytes(inode, number);
1545 	spin_unlock(&inode->i_lock);
1546 }
1547 EXPORT_SYMBOL(inode_claim_rsv_space);
1548 
1549 void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1550 {
1551 	spin_lock(&inode->i_lock);
1552 	*inode_reserved_space(inode) += number;
1553 	__inode_sub_bytes(inode, number);
1554 	spin_unlock(&inode->i_lock);
1555 }
1556 EXPORT_SYMBOL(inode_reclaim_rsv_space);
1557 
1558 void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1559 {
1560 	spin_lock(&inode->i_lock);
1561 	*inode_reserved_space(inode) -= number;
1562 	spin_unlock(&inode->i_lock);
1563 }
1564 EXPORT_SYMBOL(inode_sub_rsv_space);
1565 
1566 static qsize_t inode_get_rsv_space(struct inode *inode)
1567 {
1568 	qsize_t ret;
1569 
1570 	if (!inode->i_sb->dq_op->get_reserved_space)
1571 		return 0;
1572 	spin_lock(&inode->i_lock);
1573 	ret = *inode_reserved_space(inode);
1574 	spin_unlock(&inode->i_lock);
1575 	return ret;
1576 }
1577 
1578 static void inode_incr_space(struct inode *inode, qsize_t number,
1579 				int reserve)
1580 {
1581 	if (reserve)
1582 		inode_add_rsv_space(inode, number);
1583 	else
1584 		inode_add_bytes(inode, number);
1585 }
1586 
1587 static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1588 {
1589 	if (reserve)
1590 		inode_sub_rsv_space(inode, number);
1591 	else
1592 		inode_sub_bytes(inode, number);
1593 }
1594 
1595 /*
1596  * This functions updates i_blocks+i_bytes fields and quota information
1597  * (together with appropriate checks).
1598  *
1599  * NOTE: We absolutely rely on the fact that caller dirties the inode
1600  * (usually helpers in quotaops.h care about this) and holds a handle for
1601  * the current transaction so that dquot write and inode write go into the
1602  * same transaction.
1603  */
1604 
1605 /*
1606  * This operation can block, but only after everything is updated
1607  */
1608 int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1609 {
1610 	int cnt, ret = 0, index;
1611 	struct dquot_warn warn[MAXQUOTAS];
1612 	int reserve = flags & DQUOT_SPACE_RESERVE;
1613 	struct dquot **dquots;
1614 
1615 	if (!dquot_active(inode)) {
1616 		inode_incr_space(inode, number, reserve);
1617 		goto out;
1618 	}
1619 
1620 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1621 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1622 
1623 	dquots = i_dquot(inode);
1624 	index = srcu_read_lock(&dquot_srcu);
1625 	spin_lock(&dq_data_lock);
1626 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1627 		if (!dquots[cnt])
1628 			continue;
1629 		ret = check_bdq(dquots[cnt], number,
1630 				!(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1631 		if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1632 			spin_unlock(&dq_data_lock);
1633 			goto out_flush_warn;
1634 		}
1635 	}
1636 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1637 		if (!dquots[cnt])
1638 			continue;
1639 		if (reserve)
1640 			dquot_resv_space(dquots[cnt], number);
1641 		else
1642 			dquot_incr_space(dquots[cnt], number);
1643 	}
1644 	inode_incr_space(inode, number, reserve);
1645 	spin_unlock(&dq_data_lock);
1646 
1647 	if (reserve)
1648 		goto out_flush_warn;
1649 	mark_all_dquot_dirty(dquots);
1650 out_flush_warn:
1651 	srcu_read_unlock(&dquot_srcu, index);
1652 	flush_warnings(warn);
1653 out:
1654 	return ret;
1655 }
1656 EXPORT_SYMBOL(__dquot_alloc_space);
1657 
1658 /*
1659  * This operation can block, but only after everything is updated
1660  */
1661 int dquot_alloc_inode(struct inode *inode)
1662 {
1663 	int cnt, ret = 0, index;
1664 	struct dquot_warn warn[MAXQUOTAS];
1665 	struct dquot * const *dquots;
1666 
1667 	if (!dquot_active(inode))
1668 		return 0;
1669 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1670 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1671 
1672 	dquots = i_dquot(inode);
1673 	index = srcu_read_lock(&dquot_srcu);
1674 	spin_lock(&dq_data_lock);
1675 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1676 		if (!dquots[cnt])
1677 			continue;
1678 		ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1679 		if (ret)
1680 			goto warn_put_all;
1681 	}
1682 
1683 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1684 		if (!dquots[cnt])
1685 			continue;
1686 		dquot_incr_inodes(dquots[cnt], 1);
1687 	}
1688 
1689 warn_put_all:
1690 	spin_unlock(&dq_data_lock);
1691 	if (ret == 0)
1692 		mark_all_dquot_dirty(dquots);
1693 	srcu_read_unlock(&dquot_srcu, index);
1694 	flush_warnings(warn);
1695 	return ret;
1696 }
1697 EXPORT_SYMBOL(dquot_alloc_inode);
1698 
1699 /*
1700  * Convert in-memory reserved quotas to real consumed quotas
1701  */
1702 int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1703 {
1704 	struct dquot **dquots;
1705 	int cnt, index;
1706 
1707 	if (!dquot_active(inode)) {
1708 		inode_claim_rsv_space(inode, number);
1709 		return 0;
1710 	}
1711 
1712 	dquots = i_dquot(inode);
1713 	index = srcu_read_lock(&dquot_srcu);
1714 	spin_lock(&dq_data_lock);
1715 	/* Claim reserved quotas to allocated quotas */
1716 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1717 		if (dquots[cnt])
1718 			dquot_claim_reserved_space(dquots[cnt], number);
1719 	}
1720 	/* Update inode bytes */
1721 	inode_claim_rsv_space(inode, number);
1722 	spin_unlock(&dq_data_lock);
1723 	mark_all_dquot_dirty(dquots);
1724 	srcu_read_unlock(&dquot_srcu, index);
1725 	return 0;
1726 }
1727 EXPORT_SYMBOL(dquot_claim_space_nodirty);
1728 
1729 /*
1730  * Convert allocated space back to in-memory reserved quotas
1731  */
1732 void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1733 {
1734 	struct dquot **dquots;
1735 	int cnt, index;
1736 
1737 	if (!dquot_active(inode)) {
1738 		inode_reclaim_rsv_space(inode, number);
1739 		return;
1740 	}
1741 
1742 	dquots = i_dquot(inode);
1743 	index = srcu_read_lock(&dquot_srcu);
1744 	spin_lock(&dq_data_lock);
1745 	/* Claim reserved quotas to allocated quotas */
1746 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1747 		if (dquots[cnt])
1748 			dquot_reclaim_reserved_space(dquots[cnt], number);
1749 	}
1750 	/* Update inode bytes */
1751 	inode_reclaim_rsv_space(inode, number);
1752 	spin_unlock(&dq_data_lock);
1753 	mark_all_dquot_dirty(dquots);
1754 	srcu_read_unlock(&dquot_srcu, index);
1755 	return;
1756 }
1757 EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1758 
1759 /*
1760  * This operation can block, but only after everything is updated
1761  */
1762 void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1763 {
1764 	unsigned int cnt;
1765 	struct dquot_warn warn[MAXQUOTAS];
1766 	struct dquot **dquots;
1767 	int reserve = flags & DQUOT_SPACE_RESERVE, index;
1768 
1769 	if (!dquot_active(inode)) {
1770 		inode_decr_space(inode, number, reserve);
1771 		return;
1772 	}
1773 
1774 	dquots = i_dquot(inode);
1775 	index = srcu_read_lock(&dquot_srcu);
1776 	spin_lock(&dq_data_lock);
1777 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1778 		int wtype;
1779 
1780 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1781 		if (!dquots[cnt])
1782 			continue;
1783 		wtype = info_bdq_free(dquots[cnt], number);
1784 		if (wtype != QUOTA_NL_NOWARN)
1785 			prepare_warning(&warn[cnt], dquots[cnt], wtype);
1786 		if (reserve)
1787 			dquot_free_reserved_space(dquots[cnt], number);
1788 		else
1789 			dquot_decr_space(dquots[cnt], number);
1790 	}
1791 	inode_decr_space(inode, number, reserve);
1792 	spin_unlock(&dq_data_lock);
1793 
1794 	if (reserve)
1795 		goto out_unlock;
1796 	mark_all_dquot_dirty(dquots);
1797 out_unlock:
1798 	srcu_read_unlock(&dquot_srcu, index);
1799 	flush_warnings(warn);
1800 }
1801 EXPORT_SYMBOL(__dquot_free_space);
1802 
1803 /*
1804  * This operation can block, but only after everything is updated
1805  */
1806 void dquot_free_inode(struct inode *inode)
1807 {
1808 	unsigned int cnt;
1809 	struct dquot_warn warn[MAXQUOTAS];
1810 	struct dquot * const *dquots;
1811 	int index;
1812 
1813 	if (!dquot_active(inode))
1814 		return;
1815 
1816 	dquots = i_dquot(inode);
1817 	index = srcu_read_lock(&dquot_srcu);
1818 	spin_lock(&dq_data_lock);
1819 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1820 		int wtype;
1821 
1822 		warn[cnt].w_type = QUOTA_NL_NOWARN;
1823 		if (!dquots[cnt])
1824 			continue;
1825 		wtype = info_idq_free(dquots[cnt], 1);
1826 		if (wtype != QUOTA_NL_NOWARN)
1827 			prepare_warning(&warn[cnt], dquots[cnt], wtype);
1828 		dquot_decr_inodes(dquots[cnt], 1);
1829 	}
1830 	spin_unlock(&dq_data_lock);
1831 	mark_all_dquot_dirty(dquots);
1832 	srcu_read_unlock(&dquot_srcu, index);
1833 	flush_warnings(warn);
1834 }
1835 EXPORT_SYMBOL(dquot_free_inode);
1836 
1837 /*
1838  * Transfer the number of inode and blocks from one diskquota to an other.
1839  * On success, dquot references in transfer_to are consumed and references
1840  * to original dquots that need to be released are placed there. On failure,
1841  * references are kept untouched.
1842  *
1843  * This operation can block, but only after everything is updated
1844  * A transaction must be started when entering this function.
1845  *
1846  * We are holding reference on transfer_from & transfer_to, no need to
1847  * protect them by srcu_read_lock().
1848  */
1849 int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1850 {
1851 	qsize_t space, cur_space;
1852 	qsize_t rsv_space = 0;
1853 	struct dquot *transfer_from[MAXQUOTAS] = {};
1854 	int cnt, ret = 0;
1855 	char is_valid[MAXQUOTAS] = {};
1856 	struct dquot_warn warn_to[MAXQUOTAS];
1857 	struct dquot_warn warn_from_inodes[MAXQUOTAS];
1858 	struct dquot_warn warn_from_space[MAXQUOTAS];
1859 
1860 	if (IS_NOQUOTA(inode))
1861 		return 0;
1862 	/* Initialize the arrays */
1863 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1864 		warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1865 		warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1866 		warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1867 	}
1868 
1869 	spin_lock(&dq_data_lock);
1870 	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
1871 		spin_unlock(&dq_data_lock);
1872 		return 0;
1873 	}
1874 	cur_space = inode_get_bytes(inode);
1875 	rsv_space = inode_get_rsv_space(inode);
1876 	space = cur_space + rsv_space;
1877 	/* Build the transfer_from list and check the limits */
1878 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1879 		/*
1880 		 * Skip changes for same uid or gid or for turned off quota-type.
1881 		 */
1882 		if (!transfer_to[cnt])
1883 			continue;
1884 		/* Avoid races with quotaoff() */
1885 		if (!sb_has_quota_active(inode->i_sb, cnt))
1886 			continue;
1887 		is_valid[cnt] = 1;
1888 		transfer_from[cnt] = i_dquot(inode)[cnt];
1889 		ret = check_idq(transfer_to[cnt], 1, &warn_to[cnt]);
1890 		if (ret)
1891 			goto over_quota;
1892 		ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1893 		if (ret)
1894 			goto over_quota;
1895 	}
1896 
1897 	/*
1898 	 * Finally perform the needed transfer from transfer_from to transfer_to
1899 	 */
1900 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1901 		if (!is_valid[cnt])
1902 			continue;
1903 		/* Due to IO error we might not have transfer_from[] structure */
1904 		if (transfer_from[cnt]) {
1905 			int wtype;
1906 			wtype = info_idq_free(transfer_from[cnt], 1);
1907 			if (wtype != QUOTA_NL_NOWARN)
1908 				prepare_warning(&warn_from_inodes[cnt],
1909 						transfer_from[cnt], wtype);
1910 			wtype = info_bdq_free(transfer_from[cnt], space);
1911 			if (wtype != QUOTA_NL_NOWARN)
1912 				prepare_warning(&warn_from_space[cnt],
1913 						transfer_from[cnt], wtype);
1914 			dquot_decr_inodes(transfer_from[cnt], 1);
1915 			dquot_decr_space(transfer_from[cnt], cur_space);
1916 			dquot_free_reserved_space(transfer_from[cnt],
1917 						  rsv_space);
1918 		}
1919 
1920 		dquot_incr_inodes(transfer_to[cnt], 1);
1921 		dquot_incr_space(transfer_to[cnt], cur_space);
1922 		dquot_resv_space(transfer_to[cnt], rsv_space);
1923 
1924 		i_dquot(inode)[cnt] = transfer_to[cnt];
1925 	}
1926 	spin_unlock(&dq_data_lock);
1927 
1928 	mark_all_dquot_dirty(transfer_from);
1929 	mark_all_dquot_dirty(transfer_to);
1930 	flush_warnings(warn_to);
1931 	flush_warnings(warn_from_inodes);
1932 	flush_warnings(warn_from_space);
1933 	/* Pass back references to put */
1934 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1935 		if (is_valid[cnt])
1936 			transfer_to[cnt] = transfer_from[cnt];
1937 	return 0;
1938 over_quota:
1939 	spin_unlock(&dq_data_lock);
1940 	flush_warnings(warn_to);
1941 	return ret;
1942 }
1943 EXPORT_SYMBOL(__dquot_transfer);
1944 
1945 /* Wrapper for transferring ownership of an inode for uid/gid only
1946  * Called from FSXXX_setattr()
1947  */
1948 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1949 {
1950 	struct dquot *transfer_to[MAXQUOTAS] = {};
1951 	struct super_block *sb = inode->i_sb;
1952 	int ret;
1953 
1954 	if (!dquot_active(inode))
1955 		return 0;
1956 
1957 	if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid))
1958 		transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(iattr->ia_uid));
1959 	if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))
1960 		transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(iattr->ia_gid));
1961 
1962 	ret = __dquot_transfer(inode, transfer_to);
1963 	dqput_all(transfer_to);
1964 	return ret;
1965 }
1966 EXPORT_SYMBOL(dquot_transfer);
1967 
1968 /*
1969  * Write info of quota file to disk
1970  */
1971 int dquot_commit_info(struct super_block *sb, int type)
1972 {
1973 	int ret;
1974 	struct quota_info *dqopt = sb_dqopt(sb);
1975 
1976 	mutex_lock(&dqopt->dqio_mutex);
1977 	ret = dqopt->ops[type]->write_file_info(sb, type);
1978 	mutex_unlock(&dqopt->dqio_mutex);
1979 	return ret;
1980 }
1981 EXPORT_SYMBOL(dquot_commit_info);
1982 
1983 /*
1984  * Definitions of diskquota operations.
1985  */
1986 const struct dquot_operations dquot_operations = {
1987 	.write_dquot	= dquot_commit,
1988 	.acquire_dquot	= dquot_acquire,
1989 	.release_dquot	= dquot_release,
1990 	.mark_dirty	= dquot_mark_dquot_dirty,
1991 	.write_info	= dquot_commit_info,
1992 	.alloc_dquot	= dquot_alloc,
1993 	.destroy_dquot	= dquot_destroy,
1994 };
1995 EXPORT_SYMBOL(dquot_operations);
1996 
1997 /*
1998  * Generic helper for ->open on filesystems supporting disk quotas.
1999  */
2000 int dquot_file_open(struct inode *inode, struct file *file)
2001 {
2002 	int error;
2003 
2004 	error = generic_file_open(inode, file);
2005 	if (!error && (file->f_mode & FMODE_WRITE))
2006 		dquot_initialize(inode);
2007 	return error;
2008 }
2009 EXPORT_SYMBOL(dquot_file_open);
2010 
2011 /*
2012  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2013  */
2014 int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2015 {
2016 	int cnt, ret = 0;
2017 	struct quota_info *dqopt = sb_dqopt(sb);
2018 	struct inode *toputinode[MAXQUOTAS];
2019 
2020 	/* Cannot turn off usage accounting without turning off limits, or
2021 	 * suspend quotas and simultaneously turn quotas off. */
2022 	if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2023 	    || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2024 	    DQUOT_USAGE_ENABLED)))
2025 		return -EINVAL;
2026 
2027 	/* We need to serialize quota_off() for device */
2028 	mutex_lock(&dqopt->dqonoff_mutex);
2029 
2030 	/*
2031 	 * Skip everything if there's nothing to do. We have to do this because
2032 	 * sometimes we are called when fill_super() failed and calling
2033 	 * sync_fs() in such cases does no good.
2034 	 */
2035 	if (!sb_any_quota_loaded(sb)) {
2036 		mutex_unlock(&dqopt->dqonoff_mutex);
2037 		return 0;
2038 	}
2039 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2040 		toputinode[cnt] = NULL;
2041 		if (type != -1 && cnt != type)
2042 			continue;
2043 		if (!sb_has_quota_loaded(sb, cnt))
2044 			continue;
2045 
2046 		if (flags & DQUOT_SUSPENDED) {
2047 			spin_lock(&dq_state_lock);
2048 			dqopt->flags |=
2049 				dquot_state_flag(DQUOT_SUSPENDED, cnt);
2050 			spin_unlock(&dq_state_lock);
2051 		} else {
2052 			spin_lock(&dq_state_lock);
2053 			dqopt->flags &= ~dquot_state_flag(flags, cnt);
2054 			/* Turning off suspended quotas? */
2055 			if (!sb_has_quota_loaded(sb, cnt) &&
2056 			    sb_has_quota_suspended(sb, cnt)) {
2057 				dqopt->flags &=	~dquot_state_flag(
2058 							DQUOT_SUSPENDED, cnt);
2059 				spin_unlock(&dq_state_lock);
2060 				iput(dqopt->files[cnt]);
2061 				dqopt->files[cnt] = NULL;
2062 				continue;
2063 			}
2064 			spin_unlock(&dq_state_lock);
2065 		}
2066 
2067 		/* We still have to keep quota loaded? */
2068 		if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2069 			continue;
2070 
2071 		/* Note: these are blocking operations */
2072 		drop_dquot_ref(sb, cnt);
2073 		invalidate_dquots(sb, cnt);
2074 		/*
2075 		 * Now all dquots should be invalidated, all writes done so we
2076 		 * should be only users of the info. No locks needed.
2077 		 */
2078 		if (info_dirty(&dqopt->info[cnt]))
2079 			sb->dq_op->write_info(sb, cnt);
2080 		if (dqopt->ops[cnt]->free_file_info)
2081 			dqopt->ops[cnt]->free_file_info(sb, cnt);
2082 		put_quota_format(dqopt->info[cnt].dqi_format);
2083 
2084 		toputinode[cnt] = dqopt->files[cnt];
2085 		if (!sb_has_quota_loaded(sb, cnt))
2086 			dqopt->files[cnt] = NULL;
2087 		dqopt->info[cnt].dqi_flags = 0;
2088 		dqopt->info[cnt].dqi_igrace = 0;
2089 		dqopt->info[cnt].dqi_bgrace = 0;
2090 		dqopt->ops[cnt] = NULL;
2091 	}
2092 	mutex_unlock(&dqopt->dqonoff_mutex);
2093 
2094 	/* Skip syncing and setting flags if quota files are hidden */
2095 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2096 		goto put_inodes;
2097 
2098 	/* Sync the superblock so that buffers with quota data are written to
2099 	 * disk (and so userspace sees correct data afterwards). */
2100 	if (sb->s_op->sync_fs)
2101 		sb->s_op->sync_fs(sb, 1);
2102 	sync_blockdev(sb->s_bdev);
2103 	/* Now the quota files are just ordinary files and we can set the
2104 	 * inode flags back. Moreover we discard the pagecache so that
2105 	 * userspace sees the writes we did bypassing the pagecache. We
2106 	 * must also discard the blockdev buffers so that we see the
2107 	 * changes done by userspace on the next quotaon() */
2108 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2109 		if (toputinode[cnt]) {
2110 			mutex_lock(&dqopt->dqonoff_mutex);
2111 			/* If quota was reenabled in the meantime, we have
2112 			 * nothing to do */
2113 			if (!sb_has_quota_loaded(sb, cnt)) {
2114 				mutex_lock(&toputinode[cnt]->i_mutex);
2115 				toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
2116 				  S_NOATIME | S_NOQUOTA);
2117 				truncate_inode_pages(&toputinode[cnt]->i_data,
2118 						     0);
2119 				mutex_unlock(&toputinode[cnt]->i_mutex);
2120 				mark_inode_dirty_sync(toputinode[cnt]);
2121 			}
2122 			mutex_unlock(&dqopt->dqonoff_mutex);
2123 		}
2124 	if (sb->s_bdev)
2125 		invalidate_bdev(sb->s_bdev);
2126 put_inodes:
2127 	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2128 		if (toputinode[cnt]) {
2129 			/* On remount RO, we keep the inode pointer so that we
2130 			 * can reenable quota on the subsequent remount RW. We
2131 			 * have to check 'flags' variable and not use sb_has_
2132 			 * function because another quotaon / quotaoff could
2133 			 * change global state before we got here. We refuse
2134 			 * to suspend quotas when there is pending delete on
2135 			 * the quota file... */
2136 			if (!(flags & DQUOT_SUSPENDED))
2137 				iput(toputinode[cnt]);
2138 			else if (!toputinode[cnt]->i_nlink)
2139 				ret = -EBUSY;
2140 		}
2141 	return ret;
2142 }
2143 EXPORT_SYMBOL(dquot_disable);
2144 
2145 int dquot_quota_off(struct super_block *sb, int type)
2146 {
2147 	return dquot_disable(sb, type,
2148 			     DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2149 }
2150 EXPORT_SYMBOL(dquot_quota_off);
2151 
2152 /*
2153  *	Turn quotas on on a device
2154  */
2155 
2156 /*
2157  * Helper function to turn quotas on when we already have the inode of
2158  * quota file and no quota information is loaded.
2159  */
2160 static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2161 	unsigned int flags)
2162 {
2163 	struct quota_format_type *fmt = find_quota_format(format_id);
2164 	struct super_block *sb = inode->i_sb;
2165 	struct quota_info *dqopt = sb_dqopt(sb);
2166 	int error;
2167 	int oldflags = -1;
2168 
2169 	if (!fmt)
2170 		return -ESRCH;
2171 	if (!S_ISREG(inode->i_mode)) {
2172 		error = -EACCES;
2173 		goto out_fmt;
2174 	}
2175 	if (IS_RDONLY(inode)) {
2176 		error = -EROFS;
2177 		goto out_fmt;
2178 	}
2179 	if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
2180 		error = -EINVAL;
2181 		goto out_fmt;
2182 	}
2183 	/* Usage always has to be set... */
2184 	if (!(flags & DQUOT_USAGE_ENABLED)) {
2185 		error = -EINVAL;
2186 		goto out_fmt;
2187 	}
2188 
2189 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2190 		/* As we bypass the pagecache we must now flush all the
2191 		 * dirty data and invalidate caches so that kernel sees
2192 		 * changes from userspace. It is not enough to just flush
2193 		 * the quota file since if blocksize < pagesize, invalidation
2194 		 * of the cache could fail because of other unrelated dirty
2195 		 * data */
2196 		sync_filesystem(sb);
2197 		invalidate_bdev(sb->s_bdev);
2198 	}
2199 	mutex_lock(&dqopt->dqonoff_mutex);
2200 	if (sb_has_quota_loaded(sb, type)) {
2201 		error = -EBUSY;
2202 		goto out_lock;
2203 	}
2204 
2205 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2206 		/* We don't want quota and atime on quota files (deadlocks
2207 		 * possible) Also nobody should write to the file - we use
2208 		 * special IO operations which ignore the immutable bit. */
2209 		mutex_lock(&inode->i_mutex);
2210 		oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2211 					     S_NOQUOTA);
2212 		inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2213 		mutex_unlock(&inode->i_mutex);
2214 		/*
2215 		 * When S_NOQUOTA is set, remove dquot references as no more
2216 		 * references can be added
2217 		 */
2218 		__dquot_drop(inode);
2219 	}
2220 
2221 	error = -EIO;
2222 	dqopt->files[type] = igrab(inode);
2223 	if (!dqopt->files[type])
2224 		goto out_lock;
2225 	error = -EINVAL;
2226 	if (!fmt->qf_ops->check_quota_file(sb, type))
2227 		goto out_file_init;
2228 
2229 	dqopt->ops[type] = fmt->qf_ops;
2230 	dqopt->info[type].dqi_format = fmt;
2231 	dqopt->info[type].dqi_fmt_id = format_id;
2232 	INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2233 	mutex_lock(&dqopt->dqio_mutex);
2234 	error = dqopt->ops[type]->read_file_info(sb, type);
2235 	if (error < 0) {
2236 		mutex_unlock(&dqopt->dqio_mutex);
2237 		goto out_file_init;
2238 	}
2239 	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2240 		dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2241 	mutex_unlock(&dqopt->dqio_mutex);
2242 	spin_lock(&dq_state_lock);
2243 	dqopt->flags |= dquot_state_flag(flags, type);
2244 	spin_unlock(&dq_state_lock);
2245 
2246 	add_dquot_ref(sb, type);
2247 	mutex_unlock(&dqopt->dqonoff_mutex);
2248 
2249 	return 0;
2250 
2251 out_file_init:
2252 	dqopt->files[type] = NULL;
2253 	iput(inode);
2254 out_lock:
2255 	if (oldflags != -1) {
2256 		mutex_lock(&inode->i_mutex);
2257 		/* Set the flags back (in the case of accidental quotaon()
2258 		 * on a wrong file we don't want to mess up the flags) */
2259 		inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2260 		inode->i_flags |= oldflags;
2261 		mutex_unlock(&inode->i_mutex);
2262 	}
2263 	mutex_unlock(&dqopt->dqonoff_mutex);
2264 out_fmt:
2265 	put_quota_format(fmt);
2266 
2267 	return error;
2268 }
2269 
2270 /* Reenable quotas on remount RW */
2271 int dquot_resume(struct super_block *sb, int type)
2272 {
2273 	struct quota_info *dqopt = sb_dqopt(sb);
2274 	struct inode *inode;
2275 	int ret = 0, cnt;
2276 	unsigned int flags;
2277 
2278 	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2279 		if (type != -1 && cnt != type)
2280 			continue;
2281 
2282 		mutex_lock(&dqopt->dqonoff_mutex);
2283 		if (!sb_has_quota_suspended(sb, cnt)) {
2284 			mutex_unlock(&dqopt->dqonoff_mutex);
2285 			continue;
2286 		}
2287 		inode = dqopt->files[cnt];
2288 		dqopt->files[cnt] = NULL;
2289 		spin_lock(&dq_state_lock);
2290 		flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2291 							DQUOT_LIMITS_ENABLED,
2292 							cnt);
2293 		dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2294 		spin_unlock(&dq_state_lock);
2295 		mutex_unlock(&dqopt->dqonoff_mutex);
2296 
2297 		flags = dquot_generic_flag(flags, cnt);
2298 		ret = vfs_load_quota_inode(inode, cnt,
2299 				dqopt->info[cnt].dqi_fmt_id, flags);
2300 		iput(inode);
2301 	}
2302 
2303 	return ret;
2304 }
2305 EXPORT_SYMBOL(dquot_resume);
2306 
2307 int dquot_quota_on(struct super_block *sb, int type, int format_id,
2308 		   struct path *path)
2309 {
2310 	int error = security_quota_on(path->dentry);
2311 	if (error)
2312 		return error;
2313 	/* Quota file not on the same filesystem? */
2314 	if (path->dentry->d_sb != sb)
2315 		error = -EXDEV;
2316 	else
2317 		error = vfs_load_quota_inode(path->dentry->d_inode, type,
2318 					     format_id, DQUOT_USAGE_ENABLED |
2319 					     DQUOT_LIMITS_ENABLED);
2320 	return error;
2321 }
2322 EXPORT_SYMBOL(dquot_quota_on);
2323 
2324 /*
2325  * More powerful function for turning on quotas allowing setting
2326  * of individual quota flags
2327  */
2328 int dquot_enable(struct inode *inode, int type, int format_id,
2329 		 unsigned int flags)
2330 {
2331 	int ret = 0;
2332 	struct super_block *sb = inode->i_sb;
2333 	struct quota_info *dqopt = sb_dqopt(sb);
2334 
2335 	/* Just unsuspend quotas? */
2336 	BUG_ON(flags & DQUOT_SUSPENDED);
2337 
2338 	if (!flags)
2339 		return 0;
2340 	/* Just updating flags needed? */
2341 	if (sb_has_quota_loaded(sb, type)) {
2342 		mutex_lock(&dqopt->dqonoff_mutex);
2343 		/* Now do a reliable test... */
2344 		if (!sb_has_quota_loaded(sb, type)) {
2345 			mutex_unlock(&dqopt->dqonoff_mutex);
2346 			goto load_quota;
2347 		}
2348 		if (flags & DQUOT_USAGE_ENABLED &&
2349 		    sb_has_quota_usage_enabled(sb, type)) {
2350 			ret = -EBUSY;
2351 			goto out_lock;
2352 		}
2353 		if (flags & DQUOT_LIMITS_ENABLED &&
2354 		    sb_has_quota_limits_enabled(sb, type)) {
2355 			ret = -EBUSY;
2356 			goto out_lock;
2357 		}
2358 		spin_lock(&dq_state_lock);
2359 		sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2360 		spin_unlock(&dq_state_lock);
2361 out_lock:
2362 		mutex_unlock(&dqopt->dqonoff_mutex);
2363 		return ret;
2364 	}
2365 
2366 load_quota:
2367 	return vfs_load_quota_inode(inode, type, format_id, flags);
2368 }
2369 EXPORT_SYMBOL(dquot_enable);
2370 
2371 /*
2372  * This function is used when filesystem needs to initialize quotas
2373  * during mount time.
2374  */
2375 int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2376 		int format_id, int type)
2377 {
2378 	struct dentry *dentry;
2379 	int error;
2380 
2381 	mutex_lock(&sb->s_root->d_inode->i_mutex);
2382 	dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2383 	mutex_unlock(&sb->s_root->d_inode->i_mutex);
2384 	if (IS_ERR(dentry))
2385 		return PTR_ERR(dentry);
2386 
2387 	if (!dentry->d_inode) {
2388 		error = -ENOENT;
2389 		goto out;
2390 	}
2391 
2392 	error = security_quota_on(dentry);
2393 	if (!error)
2394 		error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
2395 				DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2396 
2397 out:
2398 	dput(dentry);
2399 	return error;
2400 }
2401 EXPORT_SYMBOL(dquot_quota_on_mount);
2402 
2403 static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2404 {
2405 	int ret;
2406 	int type;
2407 	struct quota_info *dqopt = sb_dqopt(sb);
2408 
2409 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2410 		return -ENOSYS;
2411 	/* Accounting cannot be turned on while fs is mounted */
2412 	flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2413 	if (!flags)
2414 		return -EINVAL;
2415 	for (type = 0; type < MAXQUOTAS; type++) {
2416 		if (!(flags & qtype_enforce_flag(type)))
2417 			continue;
2418 		/* Can't enforce without accounting */
2419 		if (!sb_has_quota_usage_enabled(sb, type))
2420 			return -EINVAL;
2421 		ret = dquot_enable(dqopt->files[type], type,
2422 				   dqopt->info[type].dqi_fmt_id,
2423 				   DQUOT_LIMITS_ENABLED);
2424 		if (ret < 0)
2425 			goto out_err;
2426 	}
2427 	return 0;
2428 out_err:
2429 	/* Backout enforcement enablement we already did */
2430 	for (type--; type >= 0; type--)  {
2431 		if (flags & qtype_enforce_flag(type))
2432 			dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2433 	}
2434 	/* Error code translation for better compatibility with XFS */
2435 	if (ret == -EBUSY)
2436 		ret = -EEXIST;
2437 	return ret;
2438 }
2439 
2440 static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2441 {
2442 	int ret;
2443 	int type;
2444 	struct quota_info *dqopt = sb_dqopt(sb);
2445 
2446 	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2447 		return -ENOSYS;
2448 	/*
2449 	 * We don't support turning off accounting via quotactl. In principle
2450 	 * quota infrastructure can do this but filesystems don't expect
2451 	 * userspace to be able to do it.
2452 	 */
2453 	if (flags &
2454 		  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2455 		return -EOPNOTSUPP;
2456 
2457 	/* Filter out limits not enabled */
2458 	for (type = 0; type < MAXQUOTAS; type++)
2459 		if (!sb_has_quota_limits_enabled(sb, type))
2460 			flags &= ~qtype_enforce_flag(type);
2461 	/* Nothing left? */
2462 	if (!flags)
2463 		return -EEXIST;
2464 	for (type = 0; type < MAXQUOTAS; type++) {
2465 		if (flags & qtype_enforce_flag(type)) {
2466 			ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2467 			if (ret < 0)
2468 				goto out_err;
2469 		}
2470 	}
2471 	return 0;
2472 out_err:
2473 	/* Backout enforcement disabling we already did */
2474 	for (type--; type >= 0; type--)  {
2475 		if (flags & qtype_enforce_flag(type))
2476 			dquot_enable(dqopt->files[type], type,
2477 				     dqopt->info[type].dqi_fmt_id,
2478 				     DQUOT_LIMITS_ENABLED);
2479 	}
2480 	return ret;
2481 }
2482 
2483 /* Generic routine for getting common part of quota structure */
2484 static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2485 {
2486 	struct mem_dqblk *dm = &dquot->dq_dqb;
2487 
2488 	memset(di, 0, sizeof(*di));
2489 	spin_lock(&dq_data_lock);
2490 	di->d_spc_hardlimit = dm->dqb_bhardlimit;
2491 	di->d_spc_softlimit = dm->dqb_bsoftlimit;
2492 	di->d_ino_hardlimit = dm->dqb_ihardlimit;
2493 	di->d_ino_softlimit = dm->dqb_isoftlimit;
2494 	di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2495 	di->d_ino_count = dm->dqb_curinodes;
2496 	di->d_spc_timer = dm->dqb_btime;
2497 	di->d_ino_timer = dm->dqb_itime;
2498 	spin_unlock(&dq_data_lock);
2499 }
2500 
2501 int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2502 		    struct qc_dqblk *di)
2503 {
2504 	struct dquot *dquot;
2505 
2506 	dquot = dqget(sb, qid);
2507 	if (!dquot)
2508 		return -ESRCH;
2509 	do_get_dqblk(dquot, di);
2510 	dqput(dquot);
2511 
2512 	return 0;
2513 }
2514 EXPORT_SYMBOL(dquot_get_dqblk);
2515 
2516 #define VFS_QC_MASK \
2517 	(QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2518 	 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2519 	 QC_SPC_TIMER | QC_INO_TIMER)
2520 
2521 /* Generic routine for setting common part of quota structure */
2522 static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2523 {
2524 	struct mem_dqblk *dm = &dquot->dq_dqb;
2525 	int check_blim = 0, check_ilim = 0;
2526 	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2527 
2528 	if (di->d_fieldmask & ~VFS_QC_MASK)
2529 		return -EINVAL;
2530 
2531 	if (((di->d_fieldmask & QC_SPC_SOFT) &&
2532 	     di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2533 	    ((di->d_fieldmask & QC_SPC_HARD) &&
2534 	     di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2535 	    ((di->d_fieldmask & QC_INO_SOFT) &&
2536 	     (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2537 	    ((di->d_fieldmask & QC_INO_HARD) &&
2538 	     (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2539 		return -ERANGE;
2540 
2541 	spin_lock(&dq_data_lock);
2542 	if (di->d_fieldmask & QC_SPACE) {
2543 		dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2544 		check_blim = 1;
2545 		set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2546 	}
2547 
2548 	if (di->d_fieldmask & QC_SPC_SOFT)
2549 		dm->dqb_bsoftlimit = di->d_spc_softlimit;
2550 	if (di->d_fieldmask & QC_SPC_HARD)
2551 		dm->dqb_bhardlimit = di->d_spc_hardlimit;
2552 	if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2553 		check_blim = 1;
2554 		set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2555 	}
2556 
2557 	if (di->d_fieldmask & QC_INO_COUNT) {
2558 		dm->dqb_curinodes = di->d_ino_count;
2559 		check_ilim = 1;
2560 		set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2561 	}
2562 
2563 	if (di->d_fieldmask & QC_INO_SOFT)
2564 		dm->dqb_isoftlimit = di->d_ino_softlimit;
2565 	if (di->d_fieldmask & QC_INO_HARD)
2566 		dm->dqb_ihardlimit = di->d_ino_hardlimit;
2567 	if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2568 		check_ilim = 1;
2569 		set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2570 	}
2571 
2572 	if (di->d_fieldmask & QC_SPC_TIMER) {
2573 		dm->dqb_btime = di->d_spc_timer;
2574 		check_blim = 1;
2575 		set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2576 	}
2577 
2578 	if (di->d_fieldmask & QC_INO_TIMER) {
2579 		dm->dqb_itime = di->d_ino_timer;
2580 		check_ilim = 1;
2581 		set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2582 	}
2583 
2584 	if (check_blim) {
2585 		if (!dm->dqb_bsoftlimit ||
2586 		    dm->dqb_curspace < dm->dqb_bsoftlimit) {
2587 			dm->dqb_btime = 0;
2588 			clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2589 		} else if (!(di->d_fieldmask & QC_SPC_TIMER))
2590 			/* Set grace only if user hasn't provided his own... */
2591 			dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2592 	}
2593 	if (check_ilim) {
2594 		if (!dm->dqb_isoftlimit ||
2595 		    dm->dqb_curinodes < dm->dqb_isoftlimit) {
2596 			dm->dqb_itime = 0;
2597 			clear_bit(DQ_INODES_B, &dquot->dq_flags);
2598 		} else if (!(di->d_fieldmask & QC_INO_TIMER))
2599 			/* Set grace only if user hasn't provided his own... */
2600 			dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2601 	}
2602 	if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2603 	    dm->dqb_isoftlimit)
2604 		clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2605 	else
2606 		set_bit(DQ_FAKE_B, &dquot->dq_flags);
2607 	spin_unlock(&dq_data_lock);
2608 	mark_dquot_dirty(dquot);
2609 
2610 	return 0;
2611 }
2612 
2613 int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2614 		  struct qc_dqblk *di)
2615 {
2616 	struct dquot *dquot;
2617 	int rc;
2618 
2619 	dquot = dqget(sb, qid);
2620 	if (!dquot) {
2621 		rc = -ESRCH;
2622 		goto out;
2623 	}
2624 	rc = do_set_dqblk(dquot, di);
2625 	dqput(dquot);
2626 out:
2627 	return rc;
2628 }
2629 EXPORT_SYMBOL(dquot_set_dqblk);
2630 
2631 /* Generic routine for getting common part of quota file information */
2632 int dquot_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2633 {
2634 	struct mem_dqinfo *mi;
2635 
2636 	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2637 	if (!sb_has_quota_active(sb, type)) {
2638 		mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2639 		return -ESRCH;
2640 	}
2641 	mi = sb_dqopt(sb)->info + type;
2642 	spin_lock(&dq_data_lock);
2643 	ii->dqi_bgrace = mi->dqi_bgrace;
2644 	ii->dqi_igrace = mi->dqi_igrace;
2645 	ii->dqi_flags = mi->dqi_flags & DQF_GETINFO_MASK;
2646 	ii->dqi_valid = IIF_ALL;
2647 	spin_unlock(&dq_data_lock);
2648 	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2649 	return 0;
2650 }
2651 EXPORT_SYMBOL(dquot_get_dqinfo);
2652 
2653 /* Generic routine for setting common part of quota file information */
2654 int dquot_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2655 {
2656 	struct mem_dqinfo *mi;
2657 	int err = 0;
2658 
2659 	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2660 	if (!sb_has_quota_active(sb, type)) {
2661 		err = -ESRCH;
2662 		goto out;
2663 	}
2664 	mi = sb_dqopt(sb)->info + type;
2665 	if (ii->dqi_valid & IIF_FLAGS) {
2666 		if (ii->dqi_flags & ~DQF_SETINFO_MASK ||
2667 		    (ii->dqi_flags & DQF_ROOT_SQUASH &&
2668 		     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD)) {
2669 			err = -EINVAL;
2670 			goto out;
2671 		}
2672 	}
2673 	spin_lock(&dq_data_lock);
2674 	if (ii->dqi_valid & IIF_BGRACE)
2675 		mi->dqi_bgrace = ii->dqi_bgrace;
2676 	if (ii->dqi_valid & IIF_IGRACE)
2677 		mi->dqi_igrace = ii->dqi_igrace;
2678 	if (ii->dqi_valid & IIF_FLAGS)
2679 		mi->dqi_flags = (mi->dqi_flags & ~DQF_SETINFO_MASK) |
2680 				(ii->dqi_flags & DQF_SETINFO_MASK);
2681 	spin_unlock(&dq_data_lock);
2682 	mark_info_dirty(sb, type);
2683 	/* Force write to disk */
2684 	sb->dq_op->write_info(sb, type);
2685 out:
2686 	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2687 	return err;
2688 }
2689 EXPORT_SYMBOL(dquot_set_dqinfo);
2690 
2691 const struct quotactl_ops dquot_quotactl_ops = {
2692 	.quota_on	= dquot_quota_on,
2693 	.quota_off	= dquot_quota_off,
2694 	.quota_sync	= dquot_quota_sync,
2695 	.get_info	= dquot_get_dqinfo,
2696 	.set_info	= dquot_set_dqinfo,
2697 	.get_dqblk	= dquot_get_dqblk,
2698 	.set_dqblk	= dquot_set_dqblk
2699 };
2700 EXPORT_SYMBOL(dquot_quotactl_ops);
2701 
2702 const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2703 	.quota_enable	= dquot_quota_enable,
2704 	.quota_disable	= dquot_quota_disable,
2705 	.quota_sync	= dquot_quota_sync,
2706 	.get_info	= dquot_get_dqinfo,
2707 	.set_info	= dquot_set_dqinfo,
2708 	.get_dqblk	= dquot_get_dqblk,
2709 	.set_dqblk	= dquot_set_dqblk
2710 };
2711 EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2712 
2713 static int do_proc_dqstats(struct ctl_table *table, int write,
2714 		     void __user *buffer, size_t *lenp, loff_t *ppos)
2715 {
2716 	unsigned int type = (int *)table->data - dqstats.stat;
2717 
2718 	/* Update global table */
2719 	dqstats.stat[type] =
2720 			percpu_counter_sum_positive(&dqstats.counter[type]);
2721 	return proc_dointvec(table, write, buffer, lenp, ppos);
2722 }
2723 
2724 static struct ctl_table fs_dqstats_table[] = {
2725 	{
2726 		.procname	= "lookups",
2727 		.data		= &dqstats.stat[DQST_LOOKUPS],
2728 		.maxlen		= sizeof(int),
2729 		.mode		= 0444,
2730 		.proc_handler	= do_proc_dqstats,
2731 	},
2732 	{
2733 		.procname	= "drops",
2734 		.data		= &dqstats.stat[DQST_DROPS],
2735 		.maxlen		= sizeof(int),
2736 		.mode		= 0444,
2737 		.proc_handler	= do_proc_dqstats,
2738 	},
2739 	{
2740 		.procname	= "reads",
2741 		.data		= &dqstats.stat[DQST_READS],
2742 		.maxlen		= sizeof(int),
2743 		.mode		= 0444,
2744 		.proc_handler	= do_proc_dqstats,
2745 	},
2746 	{
2747 		.procname	= "writes",
2748 		.data		= &dqstats.stat[DQST_WRITES],
2749 		.maxlen		= sizeof(int),
2750 		.mode		= 0444,
2751 		.proc_handler	= do_proc_dqstats,
2752 	},
2753 	{
2754 		.procname	= "cache_hits",
2755 		.data		= &dqstats.stat[DQST_CACHE_HITS],
2756 		.maxlen		= sizeof(int),
2757 		.mode		= 0444,
2758 		.proc_handler	= do_proc_dqstats,
2759 	},
2760 	{
2761 		.procname	= "allocated_dquots",
2762 		.data		= &dqstats.stat[DQST_ALLOC_DQUOTS],
2763 		.maxlen		= sizeof(int),
2764 		.mode		= 0444,
2765 		.proc_handler	= do_proc_dqstats,
2766 	},
2767 	{
2768 		.procname	= "free_dquots",
2769 		.data		= &dqstats.stat[DQST_FREE_DQUOTS],
2770 		.maxlen		= sizeof(int),
2771 		.mode		= 0444,
2772 		.proc_handler	= do_proc_dqstats,
2773 	},
2774 	{
2775 		.procname	= "syncs",
2776 		.data		= &dqstats.stat[DQST_SYNCS],
2777 		.maxlen		= sizeof(int),
2778 		.mode		= 0444,
2779 		.proc_handler	= do_proc_dqstats,
2780 	},
2781 #ifdef CONFIG_PRINT_QUOTA_WARNING
2782 	{
2783 		.procname	= "warnings",
2784 		.data		= &flag_print_warnings,
2785 		.maxlen		= sizeof(int),
2786 		.mode		= 0644,
2787 		.proc_handler	= proc_dointvec,
2788 	},
2789 #endif
2790 	{ },
2791 };
2792 
2793 static struct ctl_table fs_table[] = {
2794 	{
2795 		.procname	= "quota",
2796 		.mode		= 0555,
2797 		.child		= fs_dqstats_table,
2798 	},
2799 	{ },
2800 };
2801 
2802 static struct ctl_table sys_table[] = {
2803 	{
2804 		.procname	= "fs",
2805 		.mode		= 0555,
2806 		.child		= fs_table,
2807 	},
2808 	{ },
2809 };
2810 
2811 static int __init dquot_init(void)
2812 {
2813 	int i, ret;
2814 	unsigned long nr_hash, order;
2815 
2816 	printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2817 
2818 	register_sysctl_table(sys_table);
2819 
2820 	dquot_cachep = kmem_cache_create("dquot",
2821 			sizeof(struct dquot), sizeof(unsigned long) * 4,
2822 			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2823 				SLAB_MEM_SPREAD|SLAB_PANIC),
2824 			NULL);
2825 
2826 	order = 0;
2827 	dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2828 	if (!dquot_hash)
2829 		panic("Cannot create dquot hash table");
2830 
2831 	for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2832 		ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2833 		if (ret)
2834 			panic("Cannot create dquot stat counters");
2835 	}
2836 
2837 	/* Find power-of-two hlist_heads which can fit into allocation */
2838 	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2839 	dq_hash_bits = 0;
2840 	do {
2841 		dq_hash_bits++;
2842 	} while (nr_hash >> dq_hash_bits);
2843 	dq_hash_bits--;
2844 
2845 	nr_hash = 1UL << dq_hash_bits;
2846 	dq_hash_mask = nr_hash - 1;
2847 	for (i = 0; i < nr_hash; i++)
2848 		INIT_HLIST_HEAD(dquot_hash + i);
2849 
2850 	pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2851 		" %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2852 
2853 	register_shrinker(&dqcache_shrinker);
2854 
2855 	return 0;
2856 }
2857 module_init(dquot_init);
2858