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/kmod.h>
75#include <linux/namei.h>
76#include <linux/buffer_head.h>
77#include <linux/capability.h>
78#include <linux/quotaops.h>
79#include <linux/writeback.h> /* for inode_lock, oddly enough.. */
80
81#include <asm/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 *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
120 *   dqio_mutex
121 * The lock ordering of dqptr_sem imposed by quota code is only dqonoff_sem >
122 * dqptr_sem. But filesystem has to count with the fact that functions such as
123 * dquot_alloc_space() acquire dqptr_sem and they usually have to be called
124 * from inside a transaction to keep filesystem consistency after a crash. Also
125 * filesystems usually want to do some IO on dquot from ->mark_dirty which is
126 * called with dqptr_sem held.
127 * i_mutex on quota files is special (it's below dqio_mutex)
128 */
129
130static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
131static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
132__cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
133EXPORT_SYMBOL(dq_data_lock);
134
135void __quota_error(struct super_block *sb, const char *func,
136		  const char *fmt, ...)
137{
138	va_list args;
139
140	if (printk_ratelimit()) {
141		va_start(args, fmt);
142		printk(KERN_ERR "Quota error (device %s): %s: ",
143		       sb->s_id, func);
144		vprintk(fmt, args);
145		printk("\n");
146		va_end(args);
147	}
148}
149EXPORT_SYMBOL(__quota_error);
150
151#if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
152static char *quotatypes[] = INITQFNAMES;
153#endif
154static struct quota_format_type *quota_formats;	/* List of registered formats */
155static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
156
157/* SLAB cache for dquot structures */
158static struct kmem_cache *dquot_cachep;
159
160int register_quota_format(struct quota_format_type *fmt)
161{
162	spin_lock(&dq_list_lock);
163	fmt->qf_next = quota_formats;
164	quota_formats = fmt;
165	spin_unlock(&dq_list_lock);
166	return 0;
167}
168EXPORT_SYMBOL(register_quota_format);
169
170void unregister_quota_format(struct quota_format_type *fmt)
171{
172	struct quota_format_type **actqf;
173
174	spin_lock(&dq_list_lock);
175	for (actqf = &quota_formats; *actqf && *actqf != fmt;
176	     actqf = &(*actqf)->qf_next)
177		;
178	if (*actqf)
179		*actqf = (*actqf)->qf_next;
180	spin_unlock(&dq_list_lock);
181}
182EXPORT_SYMBOL(unregister_quota_format);
183
184static struct quota_format_type *find_quota_format(int id)
185{
186	struct quota_format_type *actqf;
187
188	spin_lock(&dq_list_lock);
189	for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
190	     actqf = actqf->qf_next)
191		;
192	if (!actqf || !try_module_get(actqf->qf_owner)) {
193		int qm;
194
195		spin_unlock(&dq_list_lock);
196
197		for (qm = 0; module_names[qm].qm_fmt_id &&
198			     module_names[qm].qm_fmt_id != id; qm++)
199			;
200		if (!module_names[qm].qm_fmt_id ||
201		    request_module(module_names[qm].qm_mod_name))
202			return NULL;
203
204		spin_lock(&dq_list_lock);
205		for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
206		     actqf = actqf->qf_next)
207			;
208		if (actqf && !try_module_get(actqf->qf_owner))
209			actqf = NULL;
210	}
211	spin_unlock(&dq_list_lock);
212	return actqf;
213}
214
215static void put_quota_format(struct quota_format_type *fmt)
216{
217	module_put(fmt->qf_owner);
218}
219
220/*
221 * Dquot List Management:
222 * The quota code uses three lists for dquot management: the inuse_list,
223 * free_dquots, and dquot_hash[] array. A single dquot structure may be
224 * on all three lists, depending on its current state.
225 *
226 * All dquots are placed to the end of inuse_list when first created, and this
227 * list is used for invalidate operation, which must look at every dquot.
228 *
229 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
230 * and this list is searched whenever we need an available dquot.  Dquots are
231 * removed from the list as soon as they are used again, and
232 * dqstats.free_dquots gives the number of dquots on the list. When
233 * dquot is invalidated it's completely released from memory.
234 *
235 * Dquots with a specific identity (device, type and id) are placed on
236 * one of the dquot_hash[] hash chains. The provides an efficient search
237 * mechanism to locate a specific dquot.
238 */
239
240static LIST_HEAD(inuse_list);
241static LIST_HEAD(free_dquots);
242static unsigned int dq_hash_bits, dq_hash_mask;
243static struct hlist_head *dquot_hash;
244
245struct dqstats dqstats;
246EXPORT_SYMBOL(dqstats);
247
248static qsize_t inode_get_rsv_space(struct inode *inode);
249static void __dquot_initialize(struct inode *inode, int type);
250
251static inline unsigned int
252hashfn(const struct super_block *sb, unsigned int id, int type)
253{
254	unsigned long tmp;
255
256	tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
257	return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
258}
259
260/*
261 * Following list functions expect dq_list_lock to be held
262 */
263static inline void insert_dquot_hash(struct dquot *dquot)
264{
265	struct hlist_head *head;
266	head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
267	hlist_add_head(&dquot->dq_hash, head);
268}
269
270static inline void remove_dquot_hash(struct dquot *dquot)
271{
272	hlist_del_init(&dquot->dq_hash);
273}
274
275static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
276				unsigned int id, int type)
277{
278	struct hlist_node *node;
279	struct dquot *dquot;
280
281	hlist_for_each (node, dquot_hash+hashent) {
282		dquot = hlist_entry(node, struct dquot, dq_hash);
283		if (dquot->dq_sb == sb && dquot->dq_id == id &&
284		    dquot->dq_type == type)
285			return dquot;
286	}
287	return NULL;
288}
289
290/* Add a dquot to the tail of the free list */
291static inline void put_dquot_last(struct dquot *dquot)
292{
293	list_add_tail(&dquot->dq_free, &free_dquots);
294	dqstats_inc(DQST_FREE_DQUOTS);
295}
296
297static inline void remove_free_dquot(struct dquot *dquot)
298{
299	if (list_empty(&dquot->dq_free))
300		return;
301	list_del_init(&dquot->dq_free);
302	dqstats_dec(DQST_FREE_DQUOTS);
303}
304
305static inline void put_inuse(struct dquot *dquot)
306{
307	/* We add to the back of inuse list so we don't have to restart
308	 * when traversing this list and we block */
309	list_add_tail(&dquot->dq_inuse, &inuse_list);
310	dqstats_inc(DQST_ALLOC_DQUOTS);
311}
312
313static inline void remove_inuse(struct dquot *dquot)
314{
315	dqstats_dec(DQST_ALLOC_DQUOTS);
316	list_del(&dquot->dq_inuse);
317}
318/*
319 * End of list functions needing dq_list_lock
320 */
321
322static void wait_on_dquot(struct dquot *dquot)
323{
324	mutex_lock(&dquot->dq_lock);
325	mutex_unlock(&dquot->dq_lock);
326}
327
328static inline int dquot_dirty(struct dquot *dquot)
329{
330	return test_bit(DQ_MOD_B, &dquot->dq_flags);
331}
332
333static inline int mark_dquot_dirty(struct dquot *dquot)
334{
335	return dquot->dq_sb->dq_op->mark_dirty(dquot);
336}
337
338/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
339int dquot_mark_dquot_dirty(struct dquot *dquot)
340{
341	int ret = 1;
342
343	/* If quota is dirty already, we don't have to acquire dq_list_lock */
344	if (test_bit(DQ_MOD_B, &dquot->dq_flags))
345		return 1;
346
347	spin_lock(&dq_list_lock);
348	if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
349		list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
350				info[dquot->dq_type].dqi_dirty_list);
351		ret = 0;
352	}
353	spin_unlock(&dq_list_lock);
354	return ret;
355}
356EXPORT_SYMBOL(dquot_mark_dquot_dirty);
357
358/* Dirtify all the dquots - this can block when journalling */
359static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
360{
361	int ret, err, cnt;
362
363	ret = err = 0;
364	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
365		if (dquot[cnt])
366			/* Even in case of error we have to continue */
367			ret = mark_dquot_dirty(dquot[cnt]);
368		if (!err)
369			err = ret;
370	}
371	return err;
372}
373
374static inline void dqput_all(struct dquot **dquot)
375{
376	unsigned int cnt;
377
378	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
379		dqput(dquot[cnt]);
380}
381
382/* This function needs dq_list_lock */
383static inline int clear_dquot_dirty(struct dquot *dquot)
384{
385	if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
386		return 0;
387	list_del_init(&dquot->dq_dirty);
388	return 1;
389}
390
391void mark_info_dirty(struct super_block *sb, int type)
392{
393	set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
394}
395EXPORT_SYMBOL(mark_info_dirty);
396
397/*
398 *	Read dquot from disk and alloc space for it
399 */
400
401int dquot_acquire(struct dquot *dquot)
402{
403	int ret = 0, ret2 = 0;
404	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
405
406	mutex_lock(&dquot->dq_lock);
407	mutex_lock(&dqopt->dqio_mutex);
408	if (!test_bit(DQ_READ_B, &dquot->dq_flags))
409		ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
410	if (ret < 0)
411		goto out_iolock;
412	set_bit(DQ_READ_B, &dquot->dq_flags);
413	/* Instantiate dquot if needed */
414	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
415		ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
416		/* Write the info if needed */
417		if (info_dirty(&dqopt->info[dquot->dq_type])) {
418			ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
419						dquot->dq_sb, dquot->dq_type);
420		}
421		if (ret < 0)
422			goto out_iolock;
423		if (ret2 < 0) {
424			ret = ret2;
425			goto out_iolock;
426		}
427	}
428	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
429out_iolock:
430	mutex_unlock(&dqopt->dqio_mutex);
431	mutex_unlock(&dquot->dq_lock);
432	return ret;
433}
434EXPORT_SYMBOL(dquot_acquire);
435
436/*
437 *	Write dquot to disk
438 */
439int dquot_commit(struct dquot *dquot)
440{
441	int ret = 0, ret2 = 0;
442	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
443
444	mutex_lock(&dqopt->dqio_mutex);
445	spin_lock(&dq_list_lock);
446	if (!clear_dquot_dirty(dquot)) {
447		spin_unlock(&dq_list_lock);
448		goto out_sem;
449	}
450	spin_unlock(&dq_list_lock);
451	/* Inactive dquot can be only if there was error during read/init
452	 * => we have better not writing it */
453	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
454		ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
455		if (info_dirty(&dqopt->info[dquot->dq_type])) {
456			ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
457						dquot->dq_sb, dquot->dq_type);
458		}
459		if (ret >= 0)
460			ret = ret2;
461	}
462out_sem:
463	mutex_unlock(&dqopt->dqio_mutex);
464	return ret;
465}
466EXPORT_SYMBOL(dquot_commit);
467
468/*
469 *	Release dquot
470 */
471int 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_type]->release_dqblk) {
482		ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
483		/* Write the info */
484		if (info_dirty(&dqopt->info[dquot->dq_type])) {
485			ret2 = dqopt->ops[dquot->dq_type]->write_file_info(
486						dquot->dq_sb, dquot->dq_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);
493out_dqlock:
494	mutex_unlock(&dquot->dq_lock);
495	return ret;
496}
497EXPORT_SYMBOL(dquot_release);
498
499void dquot_destroy(struct dquot *dquot)
500{
501	kmem_cache_free(dquot_cachep, dquot);
502}
503EXPORT_SYMBOL(dquot_destroy);
504
505static 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 */
516static void invalidate_dquots(struct super_block *sb, int type)
517{
518	struct dquot *dquot, *tmp;
519
520restart:
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_type != type)
526			continue;
527		/* Wait for dquot users */
528		if (atomic_read(&dquot->dq_count)) {
529			DEFINE_WAIT(wait);
530
531			atomic_inc(&dquot->dq_count);
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 */
564int 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		ret = fn(dquot, priv);
585		if (ret < 0)
586			goto out;
587		spin_lock(&dq_list_lock);
588		/* We are safe to continue now because our dquot could not
589		 * be moved out of the inuse list while we hold the reference */
590	}
591	spin_unlock(&dq_list_lock);
592out:
593	dqput(old_dquot);
594	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
595	return ret;
596}
597EXPORT_SYMBOL(dquot_scan_active);
598
599int dquot_quota_sync(struct super_block *sb, int type, int wait)
600{
601	struct list_head *dirty;
602	struct dquot *dquot;
603	struct quota_info *dqopt = sb_dqopt(sb);
604	int cnt;
605
606	mutex_lock(&dqopt->dqonoff_mutex);
607	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
608		if (type != -1 && cnt != type)
609			continue;
610		if (!sb_has_quota_active(sb, cnt))
611			continue;
612		spin_lock(&dq_list_lock);
613		dirty = &dqopt->info[cnt].dqi_dirty_list;
614		while (!list_empty(dirty)) {
615			dquot = list_first_entry(dirty, struct dquot,
616						 dq_dirty);
617			/* Dirty and inactive can be only bad dquot... */
618			if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
619				clear_dquot_dirty(dquot);
620				continue;
621			}
622			/* Now we have active dquot from which someone is
623 			 * holding reference so we can safely just increase
624			 * use count */
625			atomic_inc(&dquot->dq_count);
626			spin_unlock(&dq_list_lock);
627			dqstats_inc(DQST_LOOKUPS);
628			sb->dq_op->write_dquot(dquot);
629			dqput(dquot);
630			spin_lock(&dq_list_lock);
631		}
632		spin_unlock(&dq_list_lock);
633	}
634
635	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
636		if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
637		    && info_dirty(&dqopt->info[cnt]))
638			sb->dq_op->write_info(sb, cnt);
639	dqstats_inc(DQST_SYNCS);
640	mutex_unlock(&dqopt->dqonoff_mutex);
641
642	if (!wait || (sb_dqopt(sb)->flags & DQUOT_QUOTA_SYS_FILE))
643		return 0;
644
645	/* This is not very clever (and fast) but currently I don't know about
646	 * any other simple way of getting quota data to disk and we must get
647	 * them there for userspace to be visible... */
648	if (sb->s_op->sync_fs)
649		sb->s_op->sync_fs(sb, 1);
650	sync_blockdev(sb->s_bdev);
651
652	/*
653	 * Now when everything is written we can discard the pagecache so
654	 * that userspace sees the changes.
655	 */
656	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
657	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
658		if (type != -1 && cnt != type)
659			continue;
660		if (!sb_has_quota_active(sb, cnt))
661			continue;
662		mutex_lock_nested(&sb_dqopt(sb)->files[cnt]->i_mutex,
663				  I_MUTEX_QUOTA);
664		truncate_inode_pages(&sb_dqopt(sb)->files[cnt]->i_data, 0);
665		mutex_unlock(&sb_dqopt(sb)->files[cnt]->i_mutex);
666	}
667	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
668
669	return 0;
670}
671EXPORT_SYMBOL(dquot_quota_sync);
672
673/* Free unused dquots from cache */
674static void prune_dqcache(int count)
675{
676	struct list_head *head;
677	struct dquot *dquot;
678
679	head = free_dquots.prev;
680	while (head != &free_dquots && count) {
681		dquot = list_entry(head, struct dquot, dq_free);
682		remove_dquot_hash(dquot);
683		remove_free_dquot(dquot);
684		remove_inuse(dquot);
685		do_destroy_dquot(dquot);
686		count--;
687		head = free_dquots.prev;
688	}
689}
690
691/*
692 * This is called from kswapd when we think we need some
693 * more memory
694 */
695static int shrink_dqcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
696{
697	if (nr) {
698		spin_lock(&dq_list_lock);
699		prune_dqcache(nr);
700		spin_unlock(&dq_list_lock);
701	}
702	return ((unsigned)
703		percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS])
704		/100) * sysctl_vfs_cache_pressure;
705}
706
707static struct shrinker dqcache_shrinker = {
708	.shrink = shrink_dqcache_memory,
709	.seeks = DEFAULT_SEEKS,
710};
711
712/*
713 * Put reference to dquot
714 * NOTE: If you change this function please check whether dqput_blocks() works right...
715 */
716void dqput(struct dquot *dquot)
717{
718	int ret;
719
720	if (!dquot)
721		return;
722#ifdef CONFIG_QUOTA_DEBUG
723	if (!atomic_read(&dquot->dq_count)) {
724		quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
725			    quotatypes[dquot->dq_type], dquot->dq_id);
726		BUG();
727	}
728#endif
729	dqstats_inc(DQST_DROPS);
730we_slept:
731	spin_lock(&dq_list_lock);
732	if (atomic_read(&dquot->dq_count) > 1) {
733		/* We have more than one user... nothing to do */
734		atomic_dec(&dquot->dq_count);
735		/* Releasing dquot during quotaoff phase? */
736		if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_type) &&
737		    atomic_read(&dquot->dq_count) == 1)
738			wake_up(&dquot->dq_wait_unused);
739		spin_unlock(&dq_list_lock);
740		return;
741	}
742	/* Need to release dquot? */
743	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
744		spin_unlock(&dq_list_lock);
745		/* Commit dquot before releasing */
746		ret = dquot->dq_sb->dq_op->write_dquot(dquot);
747		if (ret < 0) {
748			quota_error(dquot->dq_sb, "Can't write quota structure"
749				    " (error %d). Quota may get out of sync!",
750				    ret);
751			/*
752			 * We clear dirty bit anyway, so that we avoid
753			 * infinite loop here
754			 */
755			spin_lock(&dq_list_lock);
756			clear_dquot_dirty(dquot);
757			spin_unlock(&dq_list_lock);
758		}
759		goto we_slept;
760	}
761	/* Clear flag in case dquot was inactive (something bad happened) */
762	clear_dquot_dirty(dquot);
763	if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
764		spin_unlock(&dq_list_lock);
765		dquot->dq_sb->dq_op->release_dquot(dquot);
766		goto we_slept;
767	}
768	atomic_dec(&dquot->dq_count);
769#ifdef CONFIG_QUOTA_DEBUG
770	/* sanity check */
771	BUG_ON(!list_empty(&dquot->dq_free));
772#endif
773	put_dquot_last(dquot);
774	spin_unlock(&dq_list_lock);
775}
776EXPORT_SYMBOL(dqput);
777
778struct dquot *dquot_alloc(struct super_block *sb, int type)
779{
780	return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
781}
782EXPORT_SYMBOL(dquot_alloc);
783
784static struct dquot *get_empty_dquot(struct super_block *sb, int type)
785{
786	struct dquot *dquot;
787
788	dquot = sb->dq_op->alloc_dquot(sb, type);
789	if(!dquot)
790		return NULL;
791
792	mutex_init(&dquot->dq_lock);
793	INIT_LIST_HEAD(&dquot->dq_free);
794	INIT_LIST_HEAD(&dquot->dq_inuse);
795	INIT_HLIST_NODE(&dquot->dq_hash);
796	INIT_LIST_HEAD(&dquot->dq_dirty);
797	init_waitqueue_head(&dquot->dq_wait_unused);
798	dquot->dq_sb = sb;
799	dquot->dq_type = type;
800	atomic_set(&dquot->dq_count, 1);
801
802	return dquot;
803}
804
805/*
806 * Get reference to dquot
807 *
808 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
809 * destroying our dquot by:
810 *   a) checking for quota flags under dq_list_lock and
811 *   b) getting a reference to dquot before we release dq_list_lock
812 */
813struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
814{
815	unsigned int hashent = hashfn(sb, id, type);
816	struct dquot *dquot = NULL, *empty = NULL;
817
818        if (!sb_has_quota_active(sb, type))
819		return NULL;
820we_slept:
821	spin_lock(&dq_list_lock);
822	spin_lock(&dq_state_lock);
823	if (!sb_has_quota_active(sb, type)) {
824		spin_unlock(&dq_state_lock);
825		spin_unlock(&dq_list_lock);
826		goto out;
827	}
828	spin_unlock(&dq_state_lock);
829
830	dquot = find_dquot(hashent, sb, id, type);
831	if (!dquot) {
832		if (!empty) {
833			spin_unlock(&dq_list_lock);
834			empty = get_empty_dquot(sb, type);
835			if (!empty)
836				schedule();	/* Try to wait for a moment... */
837			goto we_slept;
838		}
839		dquot = empty;
840		empty = NULL;
841		dquot->dq_id = id;
842		/* all dquots go on the inuse_list */
843		put_inuse(dquot);
844		/* hash it first so it can be found */
845		insert_dquot_hash(dquot);
846		spin_unlock(&dq_list_lock);
847		dqstats_inc(DQST_LOOKUPS);
848	} else {
849		if (!atomic_read(&dquot->dq_count))
850			remove_free_dquot(dquot);
851		atomic_inc(&dquot->dq_count);
852		spin_unlock(&dq_list_lock);
853		dqstats_inc(DQST_CACHE_HITS);
854		dqstats_inc(DQST_LOOKUPS);
855	}
856	/* Wait for dq_lock - after this we know that either dquot_release() is
857	 * already finished or it will be canceled due to dq_count > 1 test */
858	wait_on_dquot(dquot);
859	/* Read the dquot / allocate space in quota file */
860	if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) &&
861	    sb->dq_op->acquire_dquot(dquot) < 0) {
862		dqput(dquot);
863		dquot = NULL;
864		goto out;
865	}
866#ifdef CONFIG_QUOTA_DEBUG
867	BUG_ON(!dquot->dq_sb);	/* Has somebody invalidated entry under us? */
868#endif
869out:
870	if (empty)
871		do_destroy_dquot(empty);
872
873	return dquot;
874}
875EXPORT_SYMBOL(dqget);
876
877static int dqinit_needed(struct inode *inode, int type)
878{
879	int cnt;
880
881	if (IS_NOQUOTA(inode))
882		return 0;
883	if (type != -1)
884		return !inode->i_dquot[type];
885	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
886		if (!inode->i_dquot[cnt])
887			return 1;
888	return 0;
889}
890
891/* This routine is guarded by dqonoff_mutex mutex */
892static void add_dquot_ref(struct super_block *sb, int type)
893{
894	struct inode *inode, *old_inode = NULL;
895#ifdef CONFIG_QUOTA_DEBUG
896	int reserved = 0;
897#endif
898
899	spin_lock(&inode_lock);
900	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
901		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
902			continue;
903#ifdef CONFIG_QUOTA_DEBUG
904		if (unlikely(inode_get_rsv_space(inode) > 0))
905			reserved = 1;
906#endif
907		if (!atomic_read(&inode->i_writecount))
908			continue;
909		if (!dqinit_needed(inode, type))
910			continue;
911
912		__iget(inode);
913		spin_unlock(&inode_lock);
914
915		iput(old_inode);
916		__dquot_initialize(inode, type);
917		/* We hold a reference to 'inode' so it couldn't have been
918		 * removed from s_inodes list while we dropped the inode_lock.
919		 * We cannot iput the inode now as we can be holding the last
920		 * reference and we cannot iput it under inode_lock. So we
921		 * keep the reference and iput it later. */
922		old_inode = inode;
923		spin_lock(&inode_lock);
924	}
925	spin_unlock(&inode_lock);
926	iput(old_inode);
927
928#ifdef CONFIG_QUOTA_DEBUG
929	if (reserved) {
930		quota_error(sb, "Writes happened before quota was turned on "
931			"thus quota information is probably inconsistent. "
932			"Please run quotacheck(8)");
933	}
934#endif
935}
936
937/*
938 * Return 0 if dqput() won't block.
939 * (note that 1 doesn't necessarily mean blocking)
940 */
941static inline int dqput_blocks(struct dquot *dquot)
942{
943	if (atomic_read(&dquot->dq_count) <= 1)
944		return 1;
945	return 0;
946}
947
948/*
949 * Remove references to dquots from inode and add dquot to list for freeing
950 * if we have the last referece to dquot
951 * We can't race with anybody because we hold dqptr_sem for writing...
952 */
953static int remove_inode_dquot_ref(struct inode *inode, int type,
954				  struct list_head *tofree_head)
955{
956	struct dquot *dquot = inode->i_dquot[type];
957
958	inode->i_dquot[type] = NULL;
959	if (dquot) {
960		if (dqput_blocks(dquot)) {
961#ifdef CONFIG_QUOTA_DEBUG
962			if (atomic_read(&dquot->dq_count) != 1)
963				quota_error(inode->i_sb, "Adding dquot with "
964					    "dq_count %d to dispose list",
965					    atomic_read(&dquot->dq_count));
966#endif
967			spin_lock(&dq_list_lock);
968			/* As dquot must have currently users it can't be on
969			 * the free list... */
970			list_add(&dquot->dq_free, tofree_head);
971			spin_unlock(&dq_list_lock);
972			return 1;
973		}
974		else
975			dqput(dquot);   /* We have guaranteed we won't block */
976	}
977	return 0;
978}
979
980/*
981 * Free list of dquots
982 * Dquots are removed from inodes and no new references can be got so we are
983 * the only ones holding reference
984 */
985static void put_dquot_list(struct list_head *tofree_head)
986{
987	struct list_head *act_head;
988	struct dquot *dquot;
989
990	act_head = tofree_head->next;
991	while (act_head != tofree_head) {
992		dquot = list_entry(act_head, struct dquot, dq_free);
993		act_head = act_head->next;
994		/* Remove dquot from the list so we won't have problems... */
995		list_del_init(&dquot->dq_free);
996		dqput(dquot);
997	}
998}
999
1000static void remove_dquot_ref(struct super_block *sb, int type,
1001		struct list_head *tofree_head)
1002{
1003	struct inode *inode;
1004	int reserved = 0;
1005
1006	spin_lock(&inode_lock);
1007	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1008		/*
1009		 *  We have to scan also I_NEW inodes because they can already
1010		 *  have quota pointer initialized. Luckily, we need to touch
1011		 *  only quota pointers and these have separate locking
1012		 *  (dqptr_sem).
1013		 */
1014		if (!IS_NOQUOTA(inode)) {
1015			if (unlikely(inode_get_rsv_space(inode) > 0))
1016				reserved = 1;
1017			remove_inode_dquot_ref(inode, type, tofree_head);
1018		}
1019	}
1020	spin_unlock(&inode_lock);
1021#ifdef CONFIG_QUOTA_DEBUG
1022	if (reserved) {
1023		printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1024			" was disabled thus quota information is probably "
1025			"inconsistent. Please run quotacheck(8).\n", sb->s_id);
1026	}
1027#endif
1028}
1029
1030/* Gather all references from inodes and drop them */
1031static void drop_dquot_ref(struct super_block *sb, int type)
1032{
1033	LIST_HEAD(tofree_head);
1034
1035	if (sb->dq_op) {
1036		down_write(&sb_dqopt(sb)->dqptr_sem);
1037		remove_dquot_ref(sb, type, &tofree_head);
1038		up_write(&sb_dqopt(sb)->dqptr_sem);
1039		put_dquot_list(&tofree_head);
1040	}
1041}
1042
1043static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1044{
1045	dquot->dq_dqb.dqb_curinodes += number;
1046}
1047
1048static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1049{
1050	dquot->dq_dqb.dqb_curspace += number;
1051}
1052
1053static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1054{
1055	dquot->dq_dqb.dqb_rsvspace += number;
1056}
1057
1058/*
1059 * Claim reserved quota space
1060 */
1061static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1062{
1063	if (dquot->dq_dqb.dqb_rsvspace < number) {
1064		WARN_ON_ONCE(1);
1065		number = dquot->dq_dqb.dqb_rsvspace;
1066	}
1067	dquot->dq_dqb.dqb_curspace += number;
1068	dquot->dq_dqb.dqb_rsvspace -= number;
1069}
1070
1071static inline
1072void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1073{
1074	if (dquot->dq_dqb.dqb_rsvspace >= number)
1075		dquot->dq_dqb.dqb_rsvspace -= number;
1076	else {
1077		WARN_ON_ONCE(1);
1078		dquot->dq_dqb.dqb_rsvspace = 0;
1079	}
1080}
1081
1082static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1083{
1084	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1085	    dquot->dq_dqb.dqb_curinodes >= number)
1086		dquot->dq_dqb.dqb_curinodes -= number;
1087	else
1088		dquot->dq_dqb.dqb_curinodes = 0;
1089	if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1090		dquot->dq_dqb.dqb_itime = (time_t) 0;
1091	clear_bit(DQ_INODES_B, &dquot->dq_flags);
1092}
1093
1094static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1095{
1096	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1097	    dquot->dq_dqb.dqb_curspace >= number)
1098		dquot->dq_dqb.dqb_curspace -= number;
1099	else
1100		dquot->dq_dqb.dqb_curspace = 0;
1101	if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1102		dquot->dq_dqb.dqb_btime = (time_t) 0;
1103	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1104}
1105
1106static int warning_issued(struct dquot *dquot, const int warntype)
1107{
1108	int flag = (warntype == QUOTA_NL_BHARDWARN ||
1109		warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1110		((warntype == QUOTA_NL_IHARDWARN ||
1111		warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1112
1113	if (!flag)
1114		return 0;
1115	return test_and_set_bit(flag, &dquot->dq_flags);
1116}
1117
1118#ifdef CONFIG_PRINT_QUOTA_WARNING
1119static int flag_print_warnings = 1;
1120
1121static int need_print_warning(struct dquot *dquot)
1122{
1123	if (!flag_print_warnings)
1124		return 0;
1125
1126	switch (dquot->dq_type) {
1127		case USRQUOTA:
1128			return current_fsuid() == dquot->dq_id;
1129		case GRPQUOTA:
1130			return in_group_p(dquot->dq_id);
1131	}
1132	return 0;
1133}
1134
1135/* Print warning to user which exceeded quota */
1136static void print_warning(struct dquot *dquot, const int warntype)
1137{
1138	char *msg = NULL;
1139	struct tty_struct *tty;
1140
1141	if (warntype == QUOTA_NL_IHARDBELOW ||
1142	    warntype == QUOTA_NL_ISOFTBELOW ||
1143	    warntype == QUOTA_NL_BHARDBELOW ||
1144	    warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
1145		return;
1146
1147	tty = get_current_tty();
1148	if (!tty)
1149		return;
1150	tty_write_message(tty, dquot->dq_sb->s_id);
1151	if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1152		tty_write_message(tty, ": warning, ");
1153	else
1154		tty_write_message(tty, ": write failed, ");
1155	tty_write_message(tty, quotatypes[dquot->dq_type]);
1156	switch (warntype) {
1157		case QUOTA_NL_IHARDWARN:
1158			msg = " file limit reached.\r\n";
1159			break;
1160		case QUOTA_NL_ISOFTLONGWARN:
1161			msg = " file quota exceeded too long.\r\n";
1162			break;
1163		case QUOTA_NL_ISOFTWARN:
1164			msg = " file quota exceeded.\r\n";
1165			break;
1166		case QUOTA_NL_BHARDWARN:
1167			msg = " block limit reached.\r\n";
1168			break;
1169		case QUOTA_NL_BSOFTLONGWARN:
1170			msg = " block quota exceeded too long.\r\n";
1171			break;
1172		case QUOTA_NL_BSOFTWARN:
1173			msg = " block quota exceeded.\r\n";
1174			break;
1175	}
1176	tty_write_message(tty, msg);
1177	tty_kref_put(tty);
1178}
1179#endif
1180
1181/*
1182 * Write warnings to the console and send warning messages over netlink.
1183 *
1184 * Note that this function can sleep.
1185 */
1186static void flush_warnings(struct dquot *const *dquots, char *warntype)
1187{
1188	struct dquot *dq;
1189	int i;
1190
1191	for (i = 0; i < MAXQUOTAS; i++) {
1192		dq = dquots[i];
1193		if (dq && warntype[i] != QUOTA_NL_NOWARN &&
1194		    !warning_issued(dq, warntype[i])) {
1195#ifdef CONFIG_PRINT_QUOTA_WARNING
1196			print_warning(dq, warntype[i]);
1197#endif
1198			quota_send_warning(dq->dq_type, dq->dq_id,
1199					   dq->dq_sb->s_dev, warntype[i]);
1200		}
1201	}
1202}
1203
1204static int ignore_hardlimit(struct dquot *dquot)
1205{
1206	struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1207
1208	return capable(CAP_SYS_RESOURCE) &&
1209	       (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1210		!(info->dqi_flags & V1_DQF_RSQUASH));
1211}
1212
1213/* needs dq_data_lock */
1214static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1215{
1216	qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1217
1218	*warntype = QUOTA_NL_NOWARN;
1219	if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type) ||
1220	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1221		return 0;
1222
1223	if (dquot->dq_dqb.dqb_ihardlimit &&
1224	    newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1225            !ignore_hardlimit(dquot)) {
1226		*warntype = QUOTA_NL_IHARDWARN;
1227		return -EDQUOT;
1228	}
1229
1230	if (dquot->dq_dqb.dqb_isoftlimit &&
1231	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1232	    dquot->dq_dqb.dqb_itime &&
1233	    get_seconds() >= dquot->dq_dqb.dqb_itime &&
1234            !ignore_hardlimit(dquot)) {
1235		*warntype = QUOTA_NL_ISOFTLONGWARN;
1236		return -EDQUOT;
1237	}
1238
1239	if (dquot->dq_dqb.dqb_isoftlimit &&
1240	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1241	    dquot->dq_dqb.dqb_itime == 0) {
1242		*warntype = QUOTA_NL_ISOFTWARN;
1243		dquot->dq_dqb.dqb_itime = get_seconds() +
1244		    sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1245	}
1246
1247	return 0;
1248}
1249
1250/* needs dq_data_lock */
1251static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1252{
1253	qsize_t tspace;
1254	struct super_block *sb = dquot->dq_sb;
1255
1256	*warntype = QUOTA_NL_NOWARN;
1257	if (!sb_has_quota_limits_enabled(sb, dquot->dq_type) ||
1258	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1259		return 0;
1260
1261	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1262		+ space;
1263
1264	if (dquot->dq_dqb.dqb_bhardlimit &&
1265	    tspace > dquot->dq_dqb.dqb_bhardlimit &&
1266            !ignore_hardlimit(dquot)) {
1267		if (!prealloc)
1268			*warntype = QUOTA_NL_BHARDWARN;
1269		return -EDQUOT;
1270	}
1271
1272	if (dquot->dq_dqb.dqb_bsoftlimit &&
1273	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1274	    dquot->dq_dqb.dqb_btime &&
1275	    get_seconds() >= dquot->dq_dqb.dqb_btime &&
1276            !ignore_hardlimit(dquot)) {
1277		if (!prealloc)
1278			*warntype = QUOTA_NL_BSOFTLONGWARN;
1279		return -EDQUOT;
1280	}
1281
1282	if (dquot->dq_dqb.dqb_bsoftlimit &&
1283	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1284	    dquot->dq_dqb.dqb_btime == 0) {
1285		if (!prealloc) {
1286			*warntype = QUOTA_NL_BSOFTWARN;
1287			dquot->dq_dqb.dqb_btime = get_seconds() +
1288			    sb_dqopt(sb)->info[dquot->dq_type].dqi_bgrace;
1289		}
1290		else
1291			/*
1292			 * We don't allow preallocation to exceed softlimit so exceeding will
1293			 * be always printed
1294			 */
1295			return -EDQUOT;
1296	}
1297
1298	return 0;
1299}
1300
1301static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1302{
1303	qsize_t newinodes;
1304
1305	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1306	    dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1307	    !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_type))
1308		return QUOTA_NL_NOWARN;
1309
1310	newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1311	if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1312		return QUOTA_NL_ISOFTBELOW;
1313	if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1314	    newinodes < dquot->dq_dqb.dqb_ihardlimit)
1315		return QUOTA_NL_IHARDBELOW;
1316	return QUOTA_NL_NOWARN;
1317}
1318
1319static int info_bdq_free(struct dquot *dquot, qsize_t space)
1320{
1321	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1322	    dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1323		return QUOTA_NL_NOWARN;
1324
1325	if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1326		return QUOTA_NL_BSOFTBELOW;
1327	if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1328	    dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1329		return QUOTA_NL_BHARDBELOW;
1330	return QUOTA_NL_NOWARN;
1331}
1332
1333static int dquot_active(const struct inode *inode)
1334{
1335	struct super_block *sb = inode->i_sb;
1336
1337	if (IS_NOQUOTA(inode))
1338		return 0;
1339	return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1340}
1341
1342/*
1343 * Initialize quota pointers in inode
1344 *
1345 * We do things in a bit complicated way but by that we avoid calling
1346 * dqget() and thus filesystem callbacks under dqptr_sem.
1347 *
1348 * It is better to call this function outside of any transaction as it
1349 * might need a lot of space in journal for dquot structure allocation.
1350 */
1351static void __dquot_initialize(struct inode *inode, int type)
1352{
1353	unsigned int id = 0;
1354	int cnt;
1355	struct dquot *got[MAXQUOTAS];
1356	struct super_block *sb = inode->i_sb;
1357	qsize_t rsv;
1358
1359	/* First test before acquiring mutex - solves deadlocks when we
1360         * re-enter the quota code and are already holding the mutex */
1361	if (!dquot_active(inode))
1362		return;
1363
1364	/* First get references to structures we might need. */
1365	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1366		got[cnt] = NULL;
1367		if (type != -1 && cnt != type)
1368			continue;
1369		switch (cnt) {
1370		case USRQUOTA:
1371			id = inode->i_uid;
1372			break;
1373		case GRPQUOTA:
1374			id = inode->i_gid;
1375			break;
1376		}
1377		got[cnt] = dqget(sb, id, cnt);
1378	}
1379
1380	down_write(&sb_dqopt(sb)->dqptr_sem);
1381	if (IS_NOQUOTA(inode))
1382		goto out_err;
1383	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1384		if (type != -1 && cnt != type)
1385			continue;
1386		/* Avoid races with quotaoff() */
1387		if (!sb_has_quota_active(sb, cnt))
1388			continue;
1389		if (!inode->i_dquot[cnt]) {
1390			inode->i_dquot[cnt] = got[cnt];
1391			got[cnt] = NULL;
1392			/*
1393			 * Make quota reservation system happy if someone
1394			 * did a write before quota was turned on
1395			 */
1396			rsv = inode_get_rsv_space(inode);
1397			if (unlikely(rsv))
1398				dquot_resv_space(inode->i_dquot[cnt], rsv);
1399		}
1400	}
1401out_err:
1402	up_write(&sb_dqopt(sb)->dqptr_sem);
1403	/* Drop unused references */
1404	dqput_all(got);
1405}
1406
1407void dquot_initialize(struct inode *inode)
1408{
1409	__dquot_initialize(inode, -1);
1410}
1411EXPORT_SYMBOL(dquot_initialize);
1412
1413/*
1414 * 	Release all quotas referenced by inode
1415 */
1416static void __dquot_drop(struct inode *inode)
1417{
1418	int cnt;
1419	struct dquot *put[MAXQUOTAS];
1420
1421	down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1422	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1423		put[cnt] = inode->i_dquot[cnt];
1424		inode->i_dquot[cnt] = NULL;
1425	}
1426	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1427	dqput_all(put);
1428}
1429
1430void dquot_drop(struct inode *inode)
1431{
1432	int cnt;
1433
1434	if (IS_NOQUOTA(inode))
1435		return;
1436
1437	/*
1438	 * Test before calling to rule out calls from proc and such
1439	 * where we are not allowed to block. Note that this is
1440	 * actually reliable test even without the lock - the caller
1441	 * must assure that nobody can come after the DQUOT_DROP and
1442	 * add quota pointers back anyway.
1443	 */
1444	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1445		if (inode->i_dquot[cnt])
1446			break;
1447	}
1448
1449	if (cnt < MAXQUOTAS)
1450		__dquot_drop(inode);
1451}
1452EXPORT_SYMBOL(dquot_drop);
1453
1454/*
1455 * inode_reserved_space is managed internally by quota, and protected by
1456 * i_lock similar to i_blocks+i_bytes.
1457 */
1458static qsize_t *inode_reserved_space(struct inode * inode)
1459{
1460	/* Filesystem must explicitly define it's own method in order to use
1461	 * quota reservation interface */
1462	BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1463	return inode->i_sb->dq_op->get_reserved_space(inode);
1464}
1465
1466void inode_add_rsv_space(struct inode *inode, qsize_t number)
1467{
1468	spin_lock(&inode->i_lock);
1469	*inode_reserved_space(inode) += number;
1470	spin_unlock(&inode->i_lock);
1471}
1472EXPORT_SYMBOL(inode_add_rsv_space);
1473
1474void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1475{
1476	spin_lock(&inode->i_lock);
1477	*inode_reserved_space(inode) -= number;
1478	__inode_add_bytes(inode, number);
1479	spin_unlock(&inode->i_lock);
1480}
1481EXPORT_SYMBOL(inode_claim_rsv_space);
1482
1483void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1484{
1485	spin_lock(&inode->i_lock);
1486	*inode_reserved_space(inode) -= number;
1487	spin_unlock(&inode->i_lock);
1488}
1489EXPORT_SYMBOL(inode_sub_rsv_space);
1490
1491static qsize_t inode_get_rsv_space(struct inode *inode)
1492{
1493	qsize_t ret;
1494
1495	if (!inode->i_sb->dq_op->get_reserved_space)
1496		return 0;
1497	spin_lock(&inode->i_lock);
1498	ret = *inode_reserved_space(inode);
1499	spin_unlock(&inode->i_lock);
1500	return ret;
1501}
1502
1503static void inode_incr_space(struct inode *inode, qsize_t number,
1504				int reserve)
1505{
1506	if (reserve)
1507		inode_add_rsv_space(inode, number);
1508	else
1509		inode_add_bytes(inode, number);
1510}
1511
1512static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1513{
1514	if (reserve)
1515		inode_sub_rsv_space(inode, number);
1516	else
1517		inode_sub_bytes(inode, number);
1518}
1519
1520/*
1521 * This functions updates i_blocks+i_bytes fields and quota information
1522 * (together with appropriate checks).
1523 *
1524 * NOTE: We absolutely rely on the fact that caller dirties the inode
1525 * (usually helpers in quotaops.h care about this) and holds a handle for
1526 * the current transaction so that dquot write and inode write go into the
1527 * same transaction.
1528 */
1529
1530/*
1531 * This operation can block, but only after everything is updated
1532 */
1533int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1534{
1535	int cnt, ret = 0;
1536	char warntype[MAXQUOTAS];
1537	int warn = flags & DQUOT_SPACE_WARN;
1538	int reserve = flags & DQUOT_SPACE_RESERVE;
1539	int nofail = flags & DQUOT_SPACE_NOFAIL;
1540
1541	/*
1542	 * First test before acquiring mutex - solves deadlocks when we
1543	 * re-enter the quota code and are already holding the mutex
1544	 */
1545	if (!dquot_active(inode)) {
1546		inode_incr_space(inode, number, reserve);
1547		goto out;
1548	}
1549
1550	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1551	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1552		warntype[cnt] = QUOTA_NL_NOWARN;
1553
1554	spin_lock(&dq_data_lock);
1555	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1556		if (!inode->i_dquot[cnt])
1557			continue;
1558		ret = check_bdq(inode->i_dquot[cnt], number, !warn,
1559				warntype+cnt);
1560		if (ret && !nofail) {
1561			spin_unlock(&dq_data_lock);
1562			goto out_flush_warn;
1563		}
1564	}
1565	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1566		if (!inode->i_dquot[cnt])
1567			continue;
1568		if (reserve)
1569			dquot_resv_space(inode->i_dquot[cnt], number);
1570		else
1571			dquot_incr_space(inode->i_dquot[cnt], number);
1572	}
1573	inode_incr_space(inode, number, reserve);
1574	spin_unlock(&dq_data_lock);
1575
1576	if (reserve)
1577		goto out_flush_warn;
1578	mark_all_dquot_dirty(inode->i_dquot);
1579out_flush_warn:
1580	flush_warnings(inode->i_dquot, warntype);
1581	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1582out:
1583	return ret;
1584}
1585EXPORT_SYMBOL(__dquot_alloc_space);
1586
1587/*
1588 * This operation can block, but only after everything is updated
1589 */
1590int dquot_alloc_inode(const struct inode *inode)
1591{
1592	int cnt, ret = 0;
1593	char warntype[MAXQUOTAS];
1594
1595	/* First test before acquiring mutex - solves deadlocks when we
1596         * re-enter the quota code and are already holding the mutex */
1597	if (!dquot_active(inode))
1598		return 0;
1599	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1600		warntype[cnt] = QUOTA_NL_NOWARN;
1601	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1602	spin_lock(&dq_data_lock);
1603	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1604		if (!inode->i_dquot[cnt])
1605			continue;
1606		ret = check_idq(inode->i_dquot[cnt], 1, warntype + cnt);
1607		if (ret)
1608			goto warn_put_all;
1609	}
1610
1611	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1612		if (!inode->i_dquot[cnt])
1613			continue;
1614		dquot_incr_inodes(inode->i_dquot[cnt], 1);
1615	}
1616
1617warn_put_all:
1618	spin_unlock(&dq_data_lock);
1619	if (ret == 0)
1620		mark_all_dquot_dirty(inode->i_dquot);
1621	flush_warnings(inode->i_dquot, warntype);
1622	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1623	return ret;
1624}
1625EXPORT_SYMBOL(dquot_alloc_inode);
1626
1627/*
1628 * Convert in-memory reserved quotas to real consumed quotas
1629 */
1630int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1631{
1632	int cnt;
1633
1634	if (!dquot_active(inode)) {
1635		inode_claim_rsv_space(inode, number);
1636		return 0;
1637	}
1638
1639	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1640	spin_lock(&dq_data_lock);
1641	/* Claim reserved quotas to allocated quotas */
1642	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1643		if (inode->i_dquot[cnt])
1644			dquot_claim_reserved_space(inode->i_dquot[cnt],
1645							number);
1646	}
1647	/* Update inode bytes */
1648	inode_claim_rsv_space(inode, number);
1649	spin_unlock(&dq_data_lock);
1650	mark_all_dquot_dirty(inode->i_dquot);
1651	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1652	return 0;
1653}
1654EXPORT_SYMBOL(dquot_claim_space_nodirty);
1655
1656/*
1657 * This operation can block, but only after everything is updated
1658 */
1659void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1660{
1661	unsigned int cnt;
1662	char warntype[MAXQUOTAS];
1663	int reserve = flags & DQUOT_SPACE_RESERVE;
1664
1665	/* First test before acquiring mutex - solves deadlocks when we
1666         * re-enter the quota code and are already holding the mutex */
1667	if (!dquot_active(inode)) {
1668		inode_decr_space(inode, number, reserve);
1669		return;
1670	}
1671
1672	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1673	spin_lock(&dq_data_lock);
1674	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1675		if (!inode->i_dquot[cnt])
1676			continue;
1677		warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1678		if (reserve)
1679			dquot_free_reserved_space(inode->i_dquot[cnt], number);
1680		else
1681			dquot_decr_space(inode->i_dquot[cnt], number);
1682	}
1683	inode_decr_space(inode, number, reserve);
1684	spin_unlock(&dq_data_lock);
1685
1686	if (reserve)
1687		goto out_unlock;
1688	mark_all_dquot_dirty(inode->i_dquot);
1689out_unlock:
1690	flush_warnings(inode->i_dquot, warntype);
1691	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1692}
1693EXPORT_SYMBOL(__dquot_free_space);
1694
1695/*
1696 * This operation can block, but only after everything is updated
1697 */
1698void dquot_free_inode(const struct inode *inode)
1699{
1700	unsigned int cnt;
1701	char warntype[MAXQUOTAS];
1702
1703	/* First test before acquiring mutex - solves deadlocks when we
1704         * re-enter the quota code and are already holding the mutex */
1705	if (!dquot_active(inode))
1706		return;
1707
1708	down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1709	spin_lock(&dq_data_lock);
1710	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1711		if (!inode->i_dquot[cnt])
1712			continue;
1713		warntype[cnt] = info_idq_free(inode->i_dquot[cnt], 1);
1714		dquot_decr_inodes(inode->i_dquot[cnt], 1);
1715	}
1716	spin_unlock(&dq_data_lock);
1717	mark_all_dquot_dirty(inode->i_dquot);
1718	flush_warnings(inode->i_dquot, warntype);
1719	up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1720}
1721EXPORT_SYMBOL(dquot_free_inode);
1722
1723/*
1724 * Transfer the number of inode and blocks from one diskquota to an other.
1725 * On success, dquot references in transfer_to are consumed and references
1726 * to original dquots that need to be released are placed there. On failure,
1727 * references are kept untouched.
1728 *
1729 * This operation can block, but only after everything is updated
1730 * A transaction must be started when entering this function.
1731 *
1732 */
1733int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1734{
1735	qsize_t space, cur_space;
1736	qsize_t rsv_space = 0;
1737	struct dquot *transfer_from[MAXQUOTAS] = {};
1738	int cnt, ret = 0;
1739	char warntype_to[MAXQUOTAS];
1740	char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1741
1742	/* First test before acquiring mutex - solves deadlocks when we
1743         * re-enter the quota code and are already holding the mutex */
1744	if (IS_NOQUOTA(inode))
1745		return 0;
1746	/* Initialize the arrays */
1747	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1748		warntype_to[cnt] = QUOTA_NL_NOWARN;
1749	down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1750	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
1751		up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1752		return 0;
1753	}
1754	spin_lock(&dq_data_lock);
1755	cur_space = inode_get_bytes(inode);
1756	rsv_space = inode_get_rsv_space(inode);
1757	space = cur_space + rsv_space;
1758	/* Build the transfer_from list and check the limits */
1759	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1760		if (!transfer_to[cnt])
1761			continue;
1762		transfer_from[cnt] = inode->i_dquot[cnt];
1763		ret = check_idq(transfer_to[cnt], 1, warntype_to + cnt);
1764		if (ret)
1765			goto over_quota;
1766		ret = check_bdq(transfer_to[cnt], space, 0, warntype_to + cnt);
1767		if (ret)
1768			goto over_quota;
1769	}
1770
1771	/*
1772	 * Finally perform the needed transfer from transfer_from to transfer_to
1773	 */
1774	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1775		/*
1776		 * Skip changes for same uid or gid or for turned off quota-type.
1777		 */
1778		if (!transfer_to[cnt])
1779			continue;
1780
1781		/* Due to IO error we might not have transfer_from[] structure */
1782		if (transfer_from[cnt]) {
1783			warntype_from_inodes[cnt] =
1784				info_idq_free(transfer_from[cnt], 1);
1785			warntype_from_space[cnt] =
1786				info_bdq_free(transfer_from[cnt], space);
1787			dquot_decr_inodes(transfer_from[cnt], 1);
1788			dquot_decr_space(transfer_from[cnt], cur_space);
1789			dquot_free_reserved_space(transfer_from[cnt],
1790						  rsv_space);
1791		}
1792
1793		dquot_incr_inodes(transfer_to[cnt], 1);
1794		dquot_incr_space(transfer_to[cnt], cur_space);
1795		dquot_resv_space(transfer_to[cnt], rsv_space);
1796
1797		inode->i_dquot[cnt] = transfer_to[cnt];
1798	}
1799	spin_unlock(&dq_data_lock);
1800	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1801
1802	mark_all_dquot_dirty(transfer_from);
1803	mark_all_dquot_dirty(transfer_to);
1804	/* Pass back references to put */
1805	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1806		transfer_to[cnt] = transfer_from[cnt];
1807warn:
1808	flush_warnings(transfer_to, warntype_to);
1809	flush_warnings(transfer_from, warntype_from_inodes);
1810	flush_warnings(transfer_from, warntype_from_space);
1811	return ret;
1812over_quota:
1813	spin_unlock(&dq_data_lock);
1814	up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1815	goto warn;
1816}
1817EXPORT_SYMBOL(__dquot_transfer);
1818
1819/* Wrapper for transferring ownership of an inode for uid/gid only
1820 * Called from FSXXX_setattr()
1821 */
1822int dquot_transfer(struct inode *inode, struct iattr *iattr)
1823{
1824	struct dquot *transfer_to[MAXQUOTAS] = {};
1825	struct super_block *sb = inode->i_sb;
1826	int ret;
1827
1828	if (!dquot_active(inode))
1829		return 0;
1830
1831	if (iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid)
1832		transfer_to[USRQUOTA] = dqget(sb, iattr->ia_uid, USRQUOTA);
1833	if (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)
1834		transfer_to[GRPQUOTA] = dqget(sb, iattr->ia_gid, GRPQUOTA);
1835
1836	ret = __dquot_transfer(inode, transfer_to);
1837	dqput_all(transfer_to);
1838	return ret;
1839}
1840EXPORT_SYMBOL(dquot_transfer);
1841
1842/*
1843 * Write info of quota file to disk
1844 */
1845int dquot_commit_info(struct super_block *sb, int type)
1846{
1847	int ret;
1848	struct quota_info *dqopt = sb_dqopt(sb);
1849
1850	mutex_lock(&dqopt->dqio_mutex);
1851	ret = dqopt->ops[type]->write_file_info(sb, type);
1852	mutex_unlock(&dqopt->dqio_mutex);
1853	return ret;
1854}
1855EXPORT_SYMBOL(dquot_commit_info);
1856
1857/*
1858 * Definitions of diskquota operations.
1859 */
1860const struct dquot_operations dquot_operations = {
1861	.write_dquot	= dquot_commit,
1862	.acquire_dquot	= dquot_acquire,
1863	.release_dquot	= dquot_release,
1864	.mark_dirty	= dquot_mark_dquot_dirty,
1865	.write_info	= dquot_commit_info,
1866	.alloc_dquot	= dquot_alloc,
1867	.destroy_dquot	= dquot_destroy,
1868};
1869EXPORT_SYMBOL(dquot_operations);
1870
1871/*
1872 * Generic helper for ->open on filesystems supporting disk quotas.
1873 */
1874int dquot_file_open(struct inode *inode, struct file *file)
1875{
1876	int error;
1877
1878	error = generic_file_open(inode, file);
1879	if (!error && (file->f_mode & FMODE_WRITE))
1880		dquot_initialize(inode);
1881	return error;
1882}
1883EXPORT_SYMBOL(dquot_file_open);
1884
1885/*
1886 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1887 */
1888int dquot_disable(struct super_block *sb, int type, unsigned int flags)
1889{
1890	int cnt, ret = 0;
1891	struct quota_info *dqopt = sb_dqopt(sb);
1892	struct inode *toputinode[MAXQUOTAS];
1893
1894	/* Cannot turn off usage accounting without turning off limits, or
1895	 * suspend quotas and simultaneously turn quotas off. */
1896	if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
1897	    || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
1898	    DQUOT_USAGE_ENABLED)))
1899		return -EINVAL;
1900
1901	/* We need to serialize quota_off() for device */
1902	mutex_lock(&dqopt->dqonoff_mutex);
1903
1904	/*
1905	 * Skip everything if there's nothing to do. We have to do this because
1906	 * sometimes we are called when fill_super() failed and calling
1907	 * sync_fs() in such cases does no good.
1908	 */
1909	if (!sb_any_quota_loaded(sb)) {
1910		mutex_unlock(&dqopt->dqonoff_mutex);
1911		return 0;
1912	}
1913	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1914		toputinode[cnt] = NULL;
1915		if (type != -1 && cnt != type)
1916			continue;
1917		if (!sb_has_quota_loaded(sb, cnt))
1918			continue;
1919
1920		if (flags & DQUOT_SUSPENDED) {
1921			spin_lock(&dq_state_lock);
1922			dqopt->flags |=
1923				dquot_state_flag(DQUOT_SUSPENDED, cnt);
1924			spin_unlock(&dq_state_lock);
1925		} else {
1926			spin_lock(&dq_state_lock);
1927			dqopt->flags &= ~dquot_state_flag(flags, cnt);
1928			/* Turning off suspended quotas? */
1929			if (!sb_has_quota_loaded(sb, cnt) &&
1930			    sb_has_quota_suspended(sb, cnt)) {
1931				dqopt->flags &=	~dquot_state_flag(
1932							DQUOT_SUSPENDED, cnt);
1933				spin_unlock(&dq_state_lock);
1934				iput(dqopt->files[cnt]);
1935				dqopt->files[cnt] = NULL;
1936				continue;
1937			}
1938			spin_unlock(&dq_state_lock);
1939		}
1940
1941		/* We still have to keep quota loaded? */
1942		if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
1943			continue;
1944
1945		/* Note: these are blocking operations */
1946		drop_dquot_ref(sb, cnt);
1947		invalidate_dquots(sb, cnt);
1948		/*
1949		 * Now all dquots should be invalidated, all writes done so we
1950		 * should be only users of the info. No locks needed.
1951		 */
1952		if (info_dirty(&dqopt->info[cnt]))
1953			sb->dq_op->write_info(sb, cnt);
1954		if (dqopt->ops[cnt]->free_file_info)
1955			dqopt->ops[cnt]->free_file_info(sb, cnt);
1956		put_quota_format(dqopt->info[cnt].dqi_format);
1957
1958		toputinode[cnt] = dqopt->files[cnt];
1959		if (!sb_has_quota_loaded(sb, cnt))
1960			dqopt->files[cnt] = NULL;
1961		dqopt->info[cnt].dqi_flags = 0;
1962		dqopt->info[cnt].dqi_igrace = 0;
1963		dqopt->info[cnt].dqi_bgrace = 0;
1964		dqopt->ops[cnt] = NULL;
1965	}
1966	mutex_unlock(&dqopt->dqonoff_mutex);
1967
1968	/* Skip syncing and setting flags if quota files are hidden */
1969	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
1970		goto put_inodes;
1971
1972	/* Sync the superblock so that buffers with quota data are written to
1973	 * disk (and so userspace sees correct data afterwards). */
1974	if (sb->s_op->sync_fs)
1975		sb->s_op->sync_fs(sb, 1);
1976	sync_blockdev(sb->s_bdev);
1977	/* Now the quota files are just ordinary files and we can set the
1978	 * inode flags back. Moreover we discard the pagecache so that
1979	 * userspace sees the writes we did bypassing the pagecache. We
1980	 * must also discard the blockdev buffers so that we see the
1981	 * changes done by userspace on the next quotaon() */
1982	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1983		if (toputinode[cnt]) {
1984			mutex_lock(&dqopt->dqonoff_mutex);
1985			/* If quota was reenabled in the meantime, we have
1986			 * nothing to do */
1987			if (!sb_has_quota_loaded(sb, cnt)) {
1988				mutex_lock_nested(&toputinode[cnt]->i_mutex,
1989						  I_MUTEX_QUOTA);
1990				toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1991				  S_NOATIME | S_NOQUOTA);
1992				truncate_inode_pages(&toputinode[cnt]->i_data,
1993						     0);
1994				mutex_unlock(&toputinode[cnt]->i_mutex);
1995				mark_inode_dirty_sync(toputinode[cnt]);
1996			}
1997			mutex_unlock(&dqopt->dqonoff_mutex);
1998		}
1999	if (sb->s_bdev)
2000		invalidate_bdev(sb->s_bdev);
2001put_inodes:
2002	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2003		if (toputinode[cnt]) {
2004			/* On remount RO, we keep the inode pointer so that we
2005			 * can reenable quota on the subsequent remount RW. We
2006			 * have to check 'flags' variable and not use sb_has_
2007			 * function because another quotaon / quotaoff could
2008			 * change global state before we got here. We refuse
2009			 * to suspend quotas when there is pending delete on
2010			 * the quota file... */
2011			if (!(flags & DQUOT_SUSPENDED))
2012				iput(toputinode[cnt]);
2013			else if (!toputinode[cnt]->i_nlink)
2014				ret = -EBUSY;
2015		}
2016	return ret;
2017}
2018EXPORT_SYMBOL(dquot_disable);
2019
2020int dquot_quota_off(struct super_block *sb, int type)
2021{
2022	return dquot_disable(sb, type,
2023			     DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2024}
2025EXPORT_SYMBOL(dquot_quota_off);
2026
2027/*
2028 *	Turn quotas on on a device
2029 */
2030
2031/*
2032 * Helper function to turn quotas on when we already have the inode of
2033 * quota file and no quota information is loaded.
2034 */
2035static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2036	unsigned int flags)
2037{
2038	struct quota_format_type *fmt = find_quota_format(format_id);
2039	struct super_block *sb = inode->i_sb;
2040	struct quota_info *dqopt = sb_dqopt(sb);
2041	int error;
2042	int oldflags = -1;
2043
2044	if (!fmt)
2045		return -ESRCH;
2046	if (!S_ISREG(inode->i_mode)) {
2047		error = -EACCES;
2048		goto out_fmt;
2049	}
2050	if (IS_RDONLY(inode)) {
2051		error = -EROFS;
2052		goto out_fmt;
2053	}
2054	if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
2055		error = -EINVAL;
2056		goto out_fmt;
2057	}
2058	/* Usage always has to be set... */
2059	if (!(flags & DQUOT_USAGE_ENABLED)) {
2060		error = -EINVAL;
2061		goto out_fmt;
2062	}
2063
2064	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2065		/* As we bypass the pagecache we must now flush all the
2066		 * dirty data and invalidate caches so that kernel sees
2067		 * changes from userspace. It is not enough to just flush
2068		 * the quota file since if blocksize < pagesize, invalidation
2069		 * of the cache could fail because of other unrelated dirty
2070		 * data */
2071		sync_filesystem(sb);
2072		invalidate_bdev(sb->s_bdev);
2073	}
2074	mutex_lock(&dqopt->dqonoff_mutex);
2075	if (sb_has_quota_loaded(sb, type)) {
2076		error = -EBUSY;
2077		goto out_lock;
2078	}
2079
2080	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2081		/* We don't want quota and atime on quota files (deadlocks
2082		 * possible) Also nobody should write to the file - we use
2083		 * special IO operations which ignore the immutable bit. */
2084		mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2085		oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE |
2086					     S_NOQUOTA);
2087		inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
2088		mutex_unlock(&inode->i_mutex);
2089		/*
2090		 * When S_NOQUOTA is set, remove dquot references as no more
2091		 * references can be added
2092		 */
2093		__dquot_drop(inode);
2094	}
2095
2096	error = -EIO;
2097	dqopt->files[type] = igrab(inode);
2098	if (!dqopt->files[type])
2099		goto out_lock;
2100	error = -EINVAL;
2101	if (!fmt->qf_ops->check_quota_file(sb, type))
2102		goto out_file_init;
2103
2104	dqopt->ops[type] = fmt->qf_ops;
2105	dqopt->info[type].dqi_format = fmt;
2106	dqopt->info[type].dqi_fmt_id = format_id;
2107	INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2108	mutex_lock(&dqopt->dqio_mutex);
2109	error = dqopt->ops[type]->read_file_info(sb, type);
2110	if (error < 0) {
2111		mutex_unlock(&dqopt->dqio_mutex);
2112		goto out_file_init;
2113	}
2114	mutex_unlock(&dqopt->dqio_mutex);
2115	spin_lock(&dq_state_lock);
2116	dqopt->flags |= dquot_state_flag(flags, type);
2117	spin_unlock(&dq_state_lock);
2118
2119	add_dquot_ref(sb, type);
2120	mutex_unlock(&dqopt->dqonoff_mutex);
2121
2122	return 0;
2123
2124out_file_init:
2125	dqopt->files[type] = NULL;
2126	iput(inode);
2127out_lock:
2128	if (oldflags != -1) {
2129		mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2130		/* Set the flags back (in the case of accidental quotaon()
2131		 * on a wrong file we don't want to mess up the flags) */
2132		inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
2133		inode->i_flags |= oldflags;
2134		mutex_unlock(&inode->i_mutex);
2135	}
2136	mutex_unlock(&dqopt->dqonoff_mutex);
2137out_fmt:
2138	put_quota_format(fmt);
2139
2140	return error;
2141}
2142
2143/* Reenable quotas on remount RW */
2144int dquot_resume(struct super_block *sb, int type)
2145{
2146	struct quota_info *dqopt = sb_dqopt(sb);
2147	struct inode *inode;
2148	int ret = 0, cnt;
2149	unsigned int flags;
2150
2151	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2152		if (type != -1 && cnt != type)
2153			continue;
2154
2155		mutex_lock(&dqopt->dqonoff_mutex);
2156		if (!sb_has_quota_suspended(sb, cnt)) {
2157			mutex_unlock(&dqopt->dqonoff_mutex);
2158			continue;
2159		}
2160		inode = dqopt->files[cnt];
2161		dqopt->files[cnt] = NULL;
2162		spin_lock(&dq_state_lock);
2163		flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2164							DQUOT_LIMITS_ENABLED,
2165							cnt);
2166		dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2167		spin_unlock(&dq_state_lock);
2168		mutex_unlock(&dqopt->dqonoff_mutex);
2169
2170		flags = dquot_generic_flag(flags, cnt);
2171		ret = vfs_load_quota_inode(inode, cnt,
2172				dqopt->info[cnt].dqi_fmt_id, flags);
2173		iput(inode);
2174	}
2175
2176	return ret;
2177}
2178EXPORT_SYMBOL(dquot_resume);
2179
2180int dquot_quota_on_path(struct super_block *sb, int type, int format_id,
2181		      struct path *path)
2182{
2183	int error = security_quota_on(path->dentry);
2184	if (error)
2185		return error;
2186	/* Quota file not on the same filesystem? */
2187	if (path->mnt->mnt_sb != sb)
2188		error = -EXDEV;
2189	else
2190		error = vfs_load_quota_inode(path->dentry->d_inode, type,
2191					     format_id, DQUOT_USAGE_ENABLED |
2192					     DQUOT_LIMITS_ENABLED);
2193	return error;
2194}
2195EXPORT_SYMBOL(dquot_quota_on_path);
2196
2197int dquot_quota_on(struct super_block *sb, int type, int format_id, char *name)
2198{
2199	struct path path;
2200	int error;
2201
2202	error = kern_path(name, LOOKUP_FOLLOW, &path);
2203	if (!error) {
2204		error = dquot_quota_on_path(sb, type, format_id, &path);
2205		path_put(&path);
2206	}
2207	return error;
2208}
2209EXPORT_SYMBOL(dquot_quota_on);
2210
2211/*
2212 * More powerful function for turning on quotas allowing setting
2213 * of individual quota flags
2214 */
2215int dquot_enable(struct inode *inode, int type, int format_id,
2216		 unsigned int flags)
2217{
2218	int ret = 0;
2219	struct super_block *sb = inode->i_sb;
2220	struct quota_info *dqopt = sb_dqopt(sb);
2221
2222	/* Just unsuspend quotas? */
2223	BUG_ON(flags & DQUOT_SUSPENDED);
2224
2225	if (!flags)
2226		return 0;
2227	/* Just updating flags needed? */
2228	if (sb_has_quota_loaded(sb, type)) {
2229		mutex_lock(&dqopt->dqonoff_mutex);
2230		/* Now do a reliable test... */
2231		if (!sb_has_quota_loaded(sb, type)) {
2232			mutex_unlock(&dqopt->dqonoff_mutex);
2233			goto load_quota;
2234		}
2235		if (flags & DQUOT_USAGE_ENABLED &&
2236		    sb_has_quota_usage_enabled(sb, type)) {
2237			ret = -EBUSY;
2238			goto out_lock;
2239		}
2240		if (flags & DQUOT_LIMITS_ENABLED &&
2241		    sb_has_quota_limits_enabled(sb, type)) {
2242			ret = -EBUSY;
2243			goto out_lock;
2244		}
2245		spin_lock(&dq_state_lock);
2246		sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2247		spin_unlock(&dq_state_lock);
2248out_lock:
2249		mutex_unlock(&dqopt->dqonoff_mutex);
2250		return ret;
2251	}
2252
2253load_quota:
2254	return vfs_load_quota_inode(inode, type, format_id, flags);
2255}
2256EXPORT_SYMBOL(dquot_enable);
2257
2258/*
2259 * This function is used when filesystem needs to initialize quotas
2260 * during mount time.
2261 */
2262int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2263		int format_id, int type)
2264{
2265	struct dentry *dentry;
2266	int error;
2267
2268	mutex_lock(&sb->s_root->d_inode->i_mutex);
2269	dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
2270	mutex_unlock(&sb->s_root->d_inode->i_mutex);
2271	if (IS_ERR(dentry))
2272		return PTR_ERR(dentry);
2273
2274	if (!dentry->d_inode) {
2275		error = -ENOENT;
2276		goto out;
2277	}
2278
2279	error = security_quota_on(dentry);
2280	if (!error)
2281		error = vfs_load_quota_inode(dentry->d_inode, type, format_id,
2282				DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2283
2284out:
2285	dput(dentry);
2286	return error;
2287}
2288EXPORT_SYMBOL(dquot_quota_on_mount);
2289
2290static inline qsize_t qbtos(qsize_t blocks)
2291{
2292	return blocks << QIF_DQBLKSIZE_BITS;
2293}
2294
2295static inline qsize_t stoqb(qsize_t space)
2296{
2297	return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
2298}
2299
2300/* Generic routine for getting common part of quota structure */
2301static void do_get_dqblk(struct dquot *dquot, struct fs_disk_quota *di)
2302{
2303	struct mem_dqblk *dm = &dquot->dq_dqb;
2304
2305	memset(di, 0, sizeof(*di));
2306	di->d_version = FS_DQUOT_VERSION;
2307	di->d_flags = dquot->dq_type == USRQUOTA ?
2308			FS_USER_QUOTA : FS_GROUP_QUOTA;
2309	di->d_id = dquot->dq_id;
2310
2311	spin_lock(&dq_data_lock);
2312	di->d_blk_hardlimit = stoqb(dm->dqb_bhardlimit);
2313	di->d_blk_softlimit = stoqb(dm->dqb_bsoftlimit);
2314	di->d_ino_hardlimit = dm->dqb_ihardlimit;
2315	di->d_ino_softlimit = dm->dqb_isoftlimit;
2316	di->d_bcount = dm->dqb_curspace + dm->dqb_rsvspace;
2317	di->d_icount = dm->dqb_curinodes;
2318	di->d_btimer = dm->dqb_btime;
2319	di->d_itimer = dm->dqb_itime;
2320	spin_unlock(&dq_data_lock);
2321}
2322
2323int dquot_get_dqblk(struct super_block *sb, int type, qid_t id,
2324		    struct fs_disk_quota *di)
2325{
2326	struct dquot *dquot;
2327
2328	dquot = dqget(sb, id, type);
2329	if (!dquot)
2330		return -ESRCH;
2331	do_get_dqblk(dquot, di);
2332	dqput(dquot);
2333
2334	return 0;
2335}
2336EXPORT_SYMBOL(dquot_get_dqblk);
2337
2338#define VFS_FS_DQ_MASK \
2339	(FS_DQ_BCOUNT | FS_DQ_BSOFT | FS_DQ_BHARD | \
2340	 FS_DQ_ICOUNT | FS_DQ_ISOFT | FS_DQ_IHARD | \
2341	 FS_DQ_BTIMER | FS_DQ_ITIMER)
2342
2343/* Generic routine for setting common part of quota structure */
2344static int do_set_dqblk(struct dquot *dquot, struct fs_disk_quota *di)
2345{
2346	struct mem_dqblk *dm = &dquot->dq_dqb;
2347	int check_blim = 0, check_ilim = 0;
2348	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
2349
2350	if (di->d_fieldmask & ~VFS_FS_DQ_MASK)
2351		return -EINVAL;
2352
2353	if (((di->d_fieldmask & FS_DQ_BSOFT) &&
2354	     (di->d_blk_softlimit > dqi->dqi_maxblimit)) ||
2355	    ((di->d_fieldmask & FS_DQ_BHARD) &&
2356	     (di->d_blk_hardlimit > dqi->dqi_maxblimit)) ||
2357	    ((di->d_fieldmask & FS_DQ_ISOFT) &&
2358	     (di->d_ino_softlimit > dqi->dqi_maxilimit)) ||
2359	    ((di->d_fieldmask & FS_DQ_IHARD) &&
2360	     (di->d_ino_hardlimit > dqi->dqi_maxilimit)))
2361		return -ERANGE;
2362
2363	spin_lock(&dq_data_lock);
2364	if (di->d_fieldmask & FS_DQ_BCOUNT) {
2365		dm->dqb_curspace = di->d_bcount - dm->dqb_rsvspace;
2366		check_blim = 1;
2367		set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2368	}
2369
2370	if (di->d_fieldmask & FS_DQ_BSOFT)
2371		dm->dqb_bsoftlimit = qbtos(di->d_blk_softlimit);
2372	if (di->d_fieldmask & FS_DQ_BHARD)
2373		dm->dqb_bhardlimit = qbtos(di->d_blk_hardlimit);
2374	if (di->d_fieldmask & (FS_DQ_BSOFT | FS_DQ_BHARD)) {
2375		check_blim = 1;
2376		set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2377	}
2378
2379	if (di->d_fieldmask & FS_DQ_ICOUNT) {
2380		dm->dqb_curinodes = di->d_icount;
2381		check_ilim = 1;
2382		set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2383	}
2384
2385	if (di->d_fieldmask & FS_DQ_ISOFT)
2386		dm->dqb_isoftlimit = di->d_ino_softlimit;
2387	if (di->d_fieldmask & FS_DQ_IHARD)
2388		dm->dqb_ihardlimit = di->d_ino_hardlimit;
2389	if (di->d_fieldmask & (FS_DQ_ISOFT | FS_DQ_IHARD)) {
2390		check_ilim = 1;
2391		set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2392	}
2393
2394	if (di->d_fieldmask & FS_DQ_BTIMER) {
2395		dm->dqb_btime = di->d_btimer;
2396		check_blim = 1;
2397		set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2398	}
2399
2400	if (di->d_fieldmask & FS_DQ_ITIMER) {
2401		dm->dqb_itime = di->d_itimer;
2402		check_ilim = 1;
2403		set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2404	}
2405
2406	if (check_blim) {
2407		if (!dm->dqb_bsoftlimit ||
2408		    dm->dqb_curspace < dm->dqb_bsoftlimit) {
2409			dm->dqb_btime = 0;
2410			clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2411		} else if (!(di->d_fieldmask & FS_DQ_BTIMER))
2412			/* Set grace only if user hasn't provided his own... */
2413			dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
2414	}
2415	if (check_ilim) {
2416		if (!dm->dqb_isoftlimit ||
2417		    dm->dqb_curinodes < dm->dqb_isoftlimit) {
2418			dm->dqb_itime = 0;
2419			clear_bit(DQ_INODES_B, &dquot->dq_flags);
2420		} else if (!(di->d_fieldmask & FS_DQ_ITIMER))
2421			/* Set grace only if user hasn't provided his own... */
2422			dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
2423	}
2424	if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2425	    dm->dqb_isoftlimit)
2426		clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2427	else
2428		set_bit(DQ_FAKE_B, &dquot->dq_flags);
2429	spin_unlock(&dq_data_lock);
2430	mark_dquot_dirty(dquot);
2431
2432	return 0;
2433}
2434
2435int dquot_set_dqblk(struct super_block *sb, int type, qid_t id,
2436		  struct fs_disk_quota *di)
2437{
2438	struct dquot *dquot;
2439	int rc;
2440
2441	dquot = dqget(sb, id, type);
2442	if (!dquot) {
2443		rc = -ESRCH;
2444		goto out;
2445	}
2446	rc = do_set_dqblk(dquot, di);
2447	dqput(dquot);
2448out:
2449	return rc;
2450}
2451EXPORT_SYMBOL(dquot_set_dqblk);
2452
2453/* Generic routine for getting common part of quota file information */
2454int dquot_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2455{
2456	struct mem_dqinfo *mi;
2457
2458	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2459	if (!sb_has_quota_active(sb, type)) {
2460		mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2461		return -ESRCH;
2462	}
2463	mi = sb_dqopt(sb)->info + type;
2464	spin_lock(&dq_data_lock);
2465	ii->dqi_bgrace = mi->dqi_bgrace;
2466	ii->dqi_igrace = mi->dqi_igrace;
2467	ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2468	ii->dqi_valid = IIF_ALL;
2469	spin_unlock(&dq_data_lock);
2470	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2471	return 0;
2472}
2473EXPORT_SYMBOL(dquot_get_dqinfo);
2474
2475/* Generic routine for setting common part of quota file information */
2476int dquot_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2477{
2478	struct mem_dqinfo *mi;
2479	int err = 0;
2480
2481	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2482	if (!sb_has_quota_active(sb, type)) {
2483		err = -ESRCH;
2484		goto out;
2485	}
2486	mi = sb_dqopt(sb)->info + type;
2487	spin_lock(&dq_data_lock);
2488	if (ii->dqi_valid & IIF_BGRACE)
2489		mi->dqi_bgrace = ii->dqi_bgrace;
2490	if (ii->dqi_valid & IIF_IGRACE)
2491		mi->dqi_igrace = ii->dqi_igrace;
2492	if (ii->dqi_valid & IIF_FLAGS)
2493		mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) |
2494				(ii->dqi_flags & DQF_MASK);
2495	spin_unlock(&dq_data_lock);
2496	mark_info_dirty(sb, type);
2497	/* Force write to disk */
2498	sb->dq_op->write_info(sb, type);
2499out:
2500	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2501	return err;
2502}
2503EXPORT_SYMBOL(dquot_set_dqinfo);
2504
2505const struct quotactl_ops dquot_quotactl_ops = {
2506	.quota_on	= dquot_quota_on,
2507	.quota_off	= dquot_quota_off,
2508	.quota_sync	= dquot_quota_sync,
2509	.get_info	= dquot_get_dqinfo,
2510	.set_info	= dquot_set_dqinfo,
2511	.get_dqblk	= dquot_get_dqblk,
2512	.set_dqblk	= dquot_set_dqblk
2513};
2514EXPORT_SYMBOL(dquot_quotactl_ops);
2515
2516static int do_proc_dqstats(struct ctl_table *table, int write,
2517		     void __user *buffer, size_t *lenp, loff_t *ppos)
2518{
2519	unsigned int type = (int *)table->data - dqstats.stat;
2520
2521	/* Update global table */
2522	dqstats.stat[type] =
2523			percpu_counter_sum_positive(&dqstats.counter[type]);
2524	return proc_dointvec(table, write, buffer, lenp, ppos);
2525}
2526
2527static ctl_table fs_dqstats_table[] = {
2528	{
2529		.procname	= "lookups",
2530		.data		= &dqstats.stat[DQST_LOOKUPS],
2531		.maxlen		= sizeof(int),
2532		.mode		= 0444,
2533		.proc_handler	= do_proc_dqstats,
2534	},
2535	{
2536		.procname	= "drops",
2537		.data		= &dqstats.stat[DQST_DROPS],
2538		.maxlen		= sizeof(int),
2539		.mode		= 0444,
2540		.proc_handler	= do_proc_dqstats,
2541	},
2542	{
2543		.procname	= "reads",
2544		.data		= &dqstats.stat[DQST_READS],
2545		.maxlen		= sizeof(int),
2546		.mode		= 0444,
2547		.proc_handler	= do_proc_dqstats,
2548	},
2549	{
2550		.procname	= "writes",
2551		.data		= &dqstats.stat[DQST_WRITES],
2552		.maxlen		= sizeof(int),
2553		.mode		= 0444,
2554		.proc_handler	= do_proc_dqstats,
2555	},
2556	{
2557		.procname	= "cache_hits",
2558		.data		= &dqstats.stat[DQST_CACHE_HITS],
2559		.maxlen		= sizeof(int),
2560		.mode		= 0444,
2561		.proc_handler	= do_proc_dqstats,
2562	},
2563	{
2564		.procname	= "allocated_dquots",
2565		.data		= &dqstats.stat[DQST_ALLOC_DQUOTS],
2566		.maxlen		= sizeof(int),
2567		.mode		= 0444,
2568		.proc_handler	= do_proc_dqstats,
2569	},
2570	{
2571		.procname	= "free_dquots",
2572		.data		= &dqstats.stat[DQST_FREE_DQUOTS],
2573		.maxlen		= sizeof(int),
2574		.mode		= 0444,
2575		.proc_handler	= do_proc_dqstats,
2576	},
2577	{
2578		.procname	= "syncs",
2579		.data		= &dqstats.stat[DQST_SYNCS],
2580		.maxlen		= sizeof(int),
2581		.mode		= 0444,
2582		.proc_handler	= do_proc_dqstats,
2583	},
2584#ifdef CONFIG_PRINT_QUOTA_WARNING
2585	{
2586		.procname	= "warnings",
2587		.data		= &flag_print_warnings,
2588		.maxlen		= sizeof(int),
2589		.mode		= 0644,
2590		.proc_handler	= proc_dointvec,
2591	},
2592#endif
2593	{ },
2594};
2595
2596static ctl_table fs_table[] = {
2597	{
2598		.procname	= "quota",
2599		.mode		= 0555,
2600		.child		= fs_dqstats_table,
2601	},
2602	{ },
2603};
2604
2605static ctl_table sys_table[] = {
2606	{
2607		.procname	= "fs",
2608		.mode		= 0555,
2609		.child		= fs_table,
2610	},
2611	{ },
2612};
2613
2614static int __init dquot_init(void)
2615{
2616	int i, ret;
2617	unsigned long nr_hash, order;
2618
2619	printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2620
2621	register_sysctl_table(sys_table);
2622
2623	dquot_cachep = kmem_cache_create("dquot",
2624			sizeof(struct dquot), sizeof(unsigned long) * 4,
2625			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2626				SLAB_MEM_SPREAD|SLAB_PANIC),
2627			NULL);
2628
2629	order = 0;
2630	dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2631	if (!dquot_hash)
2632		panic("Cannot create dquot hash table");
2633
2634	for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2635		ret = percpu_counter_init(&dqstats.counter[i], 0);
2636		if (ret)
2637			panic("Cannot create dquot stat counters");
2638	}
2639
2640	/* Find power-of-two hlist_heads which can fit into allocation */
2641	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2642	dq_hash_bits = 0;
2643	do {
2644		dq_hash_bits++;
2645	} while (nr_hash >> dq_hash_bits);
2646	dq_hash_bits--;
2647
2648	nr_hash = 1UL << dq_hash_bits;
2649	dq_hash_mask = nr_hash - 1;
2650	for (i = 0; i < nr_hash; i++)
2651		INIT_HLIST_HEAD(dquot_hash + i);
2652
2653	printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2654			nr_hash, order, (PAGE_SIZE << order));
2655
2656	register_shrinker(&dqcache_shrinker);
2657
2658	return 0;
2659}
2660module_init(dquot_init);
2661