1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Implementation of the diskquota system for the LINUX operating system. QUOTA
4 * is implemented using the BSD system call interface as the means of
5 * communication with the user level. This file contains the generic routines
6 * called by the different filesystems on allocation of an inode or block.
7 * These routines take care of the administration needed to have a consistent
8 * diskquota tracking system. The ideas of both user and group quotas are based
9 * on the Melbourne quota system as used on BSD derived systems. The internal
10 * implementation is based on one of the several variants of the LINUX
11 * inode-subsystem with added complexity of the diskquota system.
12 *
13 * Author:	Marco van Wieringen <mvw@planets.elm.net>
14 *
15 * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
16 *
17 *		Revised list management to avoid races
18 *		-- Bill Hawes, <whawes@star.net>, 9/98
19 *
20 *		Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
21 *		As the consequence the locking was moved from dquot_decr_...(),
22 *		dquot_incr_...() to calling functions.
23 *		invalidate_dquots() now writes modified dquots.
24 *		Serialized quota_off() and quota_on() for mount point.
25 *		Fixed a few bugs in grow_dquots().
26 *		Fixed deadlock in write_dquot() - we no longer account quotas on
27 *		quota files
28 *		remove_dquot_ref() moved to inode.c - it now traverses through inodes
29 *		add_dquot_ref() restarts after blocking
30 *		Added check for bogus uid and fixed check for group in quotactl.
31 *		Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
32 *
33 *		Used struct list_head instead of own list struct
34 *		Invalidation of referenced dquots is no longer possible
35 *		Improved free_dquots list management
36 *		Quota and i_blocks are now updated in one place to avoid races
37 *		Warnings are now delayed so we won't block in critical section
38 *		Write updated not to require dquot lock
39 *		Jan Kara, <jack@suse.cz>, 9/2000
40 *
41 *		Added dynamic quota structure allocation
42 *		Jan Kara <jack@suse.cz> 12/2000
43 *
44 *		Rewritten quota interface. Implemented new quota format and
45 *		formats registering.
46 *		Jan Kara, <jack@suse.cz>, 2001,2002
47 *
48 *		New SMP locking.
49 *		Jan Kara, <jack@suse.cz>, 10/2002
50 *
51 *		Added journalled quota support, fix lock inversion problems
52 *		Jan Kara, <jack@suse.cz>, 2003,2004
53 *
54 * (C) Copyright 1994 - 1997 Marco van Wieringen
55 */
56
57#include <linux/errno.h>
58#include <linux/kernel.h>
59#include <linux/fs.h>
60#include <linux/mount.h>
61#include <linux/mm.h>
62#include <linux/time.h>
63#include <linux/types.h>
64#include <linux/string.h>
65#include <linux/fcntl.h>
66#include <linux/stat.h>
67#include <linux/tty.h>
68#include <linux/file.h>
69#include <linux/slab.h>
70#include <linux/sysctl.h>
71#include <linux/init.h>
72#include <linux/module.h>
73#include <linux/proc_fs.h>
74#include <linux/security.h>
75#include <linux/sched.h>
76#include <linux/cred.h>
77#include <linux/kmod.h>
78#include <linux/namei.h>
79#include <linux/capability.h>
80#include <linux/quotaops.h>
81#include <linux/blkdev.h>
82#include <linux/sched/mm.h>
83#include "../internal.h" /* ugh */
84
85#include <linux/uaccess.h>
86
87/*
88 * There are five quota SMP locks:
89 * * dq_list_lock protects all lists with quotas and quota formats.
90 * * dquot->dq_dqb_lock protects data from dq_dqb
91 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
92 *   consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
93 *   dquot_transfer() can stabilize amount it transfers
94 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
95 *   pointers in the inode
96 * * dq_state_lock protects modifications of quota state (on quotaon and
97 *   quotaoff) and readers who care about latest values take it as well.
98 *
99 * The spinlock ordering is hence:
100 *   dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
101 *   dq_list_lock > dq_state_lock
102 *
103 * Note that some things (eg. sb pointer, type, id) doesn't change during
104 * the life of the dquot structure and so needn't to be protected by a lock
105 *
106 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
107 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
108 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
109 * inode and before dropping dquot references to avoid use of dquots after
110 * they are freed. dq_data_lock is used to serialize the pointer setting and
111 * clearing operations.
112 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
113 * inode is a quota file). Functions adding pointers from inode to dquots have
114 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
115 * have to do all pointer modifications before dropping dq_data_lock. This makes
116 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
117 * then drops all pointers to dquots from an inode.
118 *
119 * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
120 * memory (or space for it is being allocated) on the first dqget(), when it is
121 * being written out, and when it is being released on the last dqput(). The
122 * allocation and release operations are serialized by the dq_lock and by
123 * checking the use count in dquot_release().
124 *
125 * Lock ordering (including related VFS locks) is the following:
126 *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
127 */
128
129static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
130static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
131__cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
132EXPORT_SYMBOL(dq_data_lock);
133DEFINE_STATIC_SRCU(dquot_srcu);
134
135static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
136
137void __quota_error(struct super_block *sb, const char *func,
138		   const char *fmt, ...)
139{
140	if (printk_ratelimit()) {
141		va_list args;
142		struct va_format vaf;
143
144		va_start(args, fmt);
145
146		vaf.fmt = fmt;
147		vaf.va = &args;
148
149		printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
150		       sb->s_id, func, &vaf);
151
152		va_end(args);
153	}
154}
155EXPORT_SYMBOL(__quota_error);
156
157#if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
158static char *quotatypes[] = INITQFNAMES;
159#endif
160static struct quota_format_type *quota_formats;	/* List of registered formats */
161static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
162
163/* SLAB cache for dquot structures */
164static struct kmem_cache *dquot_cachep;
165
166int register_quota_format(struct quota_format_type *fmt)
167{
168	spin_lock(&dq_list_lock);
169	fmt->qf_next = quota_formats;
170	quota_formats = fmt;
171	spin_unlock(&dq_list_lock);
172	return 0;
173}
174EXPORT_SYMBOL(register_quota_format);
175
176void unregister_quota_format(struct quota_format_type *fmt)
177{
178	struct quota_format_type **actqf;
179
180	spin_lock(&dq_list_lock);
181	for (actqf = &quota_formats; *actqf && *actqf != fmt;
182	     actqf = &(*actqf)->qf_next)
183		;
184	if (*actqf)
185		*actqf = (*actqf)->qf_next;
186	spin_unlock(&dq_list_lock);
187}
188EXPORT_SYMBOL(unregister_quota_format);
189
190static struct quota_format_type *find_quota_format(int id)
191{
192	struct quota_format_type *actqf;
193
194	spin_lock(&dq_list_lock);
195	for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
196	     actqf = actqf->qf_next)
197		;
198	if (!actqf || !try_module_get(actqf->qf_owner)) {
199		int qm;
200
201		spin_unlock(&dq_list_lock);
202
203		for (qm = 0; module_names[qm].qm_fmt_id &&
204			     module_names[qm].qm_fmt_id != id; qm++)
205			;
206		if (!module_names[qm].qm_fmt_id ||
207		    request_module(module_names[qm].qm_mod_name))
208			return NULL;
209
210		spin_lock(&dq_list_lock);
211		for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
212		     actqf = actqf->qf_next)
213			;
214		if (actqf && !try_module_get(actqf->qf_owner))
215			actqf = NULL;
216	}
217	spin_unlock(&dq_list_lock);
218	return actqf;
219}
220
221static void put_quota_format(struct quota_format_type *fmt)
222{
223	module_put(fmt->qf_owner);
224}
225
226/*
227 * Dquot List Management:
228 * The quota code uses five lists for dquot management: the inuse_list,
229 * releasing_dquots, free_dquots, dqi_dirty_list, and dquot_hash[] array.
230 * A single dquot structure may be on some of those lists, depending on
231 * its current state.
232 *
233 * All dquots are placed to the end of inuse_list when first created, and this
234 * list is used for invalidate operation, which must look at every dquot.
235 *
236 * When the last reference of a dquot is dropped, the dquot is added to
237 * releasing_dquots. We'll then queue work item which will call
238 * synchronize_srcu() and after that perform the final cleanup of all the
239 * dquots on the list. Each cleaned up dquot is moved to free_dquots list.
240 * Both releasing_dquots and free_dquots use the dq_free list_head in the dquot
241 * struct.
242 *
243 * Unused and cleaned up dquots are in the free_dquots list and this list is
244 * searched whenever we need an available dquot. Dquots are removed from the
245 * list as soon as they are used again and dqstats.free_dquots gives the number
246 * of dquots on the list. When dquot is invalidated it's completely released
247 * from memory.
248 *
249 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
250 * dirtied, and this list is searched when writing dirty dquots back to
251 * quota file. Note that some filesystems do dirty dquot tracking on their
252 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
253 *
254 * Dquots with a specific identity (device, type and id) are placed on
255 * one of the dquot_hash[] hash chains. The provides an efficient search
256 * mechanism to locate a specific dquot.
257 */
258
259static LIST_HEAD(inuse_list);
260static LIST_HEAD(free_dquots);
261static LIST_HEAD(releasing_dquots);
262static unsigned int dq_hash_bits, dq_hash_mask;
263static struct hlist_head *dquot_hash;
264
265struct dqstats dqstats;
266EXPORT_SYMBOL(dqstats);
267
268static qsize_t inode_get_rsv_space(struct inode *inode);
269static qsize_t __inode_get_rsv_space(struct inode *inode);
270static int __dquot_initialize(struct inode *inode, int type);
271
272static void quota_release_workfn(struct work_struct *work);
273static DECLARE_DELAYED_WORK(quota_release_work, quota_release_workfn);
274
275static inline unsigned int
276hashfn(const struct super_block *sb, struct kqid qid)
277{
278	unsigned int id = from_kqid(&init_user_ns, qid);
279	int type = qid.type;
280	unsigned long tmp;
281
282	tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
283	return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
284}
285
286/*
287 * Following list functions expect dq_list_lock to be held
288 */
289static inline void insert_dquot_hash(struct dquot *dquot)
290{
291	struct hlist_head *head;
292	head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
293	hlist_add_head(&dquot->dq_hash, head);
294}
295
296static inline void remove_dquot_hash(struct dquot *dquot)
297{
298	hlist_del_init(&dquot->dq_hash);
299}
300
301static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
302				struct kqid qid)
303{
304	struct dquot *dquot;
305
306	hlist_for_each_entry(dquot, dquot_hash+hashent, dq_hash)
307		if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
308			return dquot;
309
310	return NULL;
311}
312
313/* Add a dquot to the tail of the free list */
314static inline void put_dquot_last(struct dquot *dquot)
315{
316	list_add_tail(&dquot->dq_free, &free_dquots);
317	dqstats_inc(DQST_FREE_DQUOTS);
318}
319
320static inline void put_releasing_dquots(struct dquot *dquot)
321{
322	list_add_tail(&dquot->dq_free, &releasing_dquots);
323	set_bit(DQ_RELEASING_B, &dquot->dq_flags);
324}
325
326static inline void remove_free_dquot(struct dquot *dquot)
327{
328	if (list_empty(&dquot->dq_free))
329		return;
330	list_del_init(&dquot->dq_free);
331	if (!test_bit(DQ_RELEASING_B, &dquot->dq_flags))
332		dqstats_dec(DQST_FREE_DQUOTS);
333	else
334		clear_bit(DQ_RELEASING_B, &dquot->dq_flags);
335}
336
337static inline void put_inuse(struct dquot *dquot)
338{
339	/* We add to the back of inuse list so we don't have to restart
340	 * when traversing this list and we block */
341	list_add_tail(&dquot->dq_inuse, &inuse_list);
342	dqstats_inc(DQST_ALLOC_DQUOTS);
343}
344
345static inline void remove_inuse(struct dquot *dquot)
346{
347	dqstats_dec(DQST_ALLOC_DQUOTS);
348	list_del(&dquot->dq_inuse);
349}
350/*
351 * End of list functions needing dq_list_lock
352 */
353
354static void wait_on_dquot(struct dquot *dquot)
355{
356	mutex_lock(&dquot->dq_lock);
357	mutex_unlock(&dquot->dq_lock);
358}
359
360static inline int dquot_active(struct dquot *dquot)
361{
362	return test_bit(DQ_ACTIVE_B, &dquot->dq_flags);
363}
364
365static inline int dquot_dirty(struct dquot *dquot)
366{
367	return test_bit(DQ_MOD_B, &dquot->dq_flags);
368}
369
370static inline int mark_dquot_dirty(struct dquot *dquot)
371{
372	return dquot->dq_sb->dq_op->mark_dirty(dquot);
373}
374
375/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
376int dquot_mark_dquot_dirty(struct dquot *dquot)
377{
378	int ret = 1;
379
380	if (!dquot_active(dquot))
381		return 0;
382
383	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
384		return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
385
386	/* If quota is dirty already, we don't have to acquire dq_list_lock */
387	if (dquot_dirty(dquot))
388		return 1;
389
390	spin_lock(&dq_list_lock);
391	if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
392		list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
393				info[dquot->dq_id.type].dqi_dirty_list);
394		ret = 0;
395	}
396	spin_unlock(&dq_list_lock);
397	return ret;
398}
399EXPORT_SYMBOL(dquot_mark_dquot_dirty);
400
401/* Dirtify all the dquots - this can block when journalling */
402static inline int mark_all_dquot_dirty(struct dquot __rcu * const *dquots)
403{
404	int ret, err, cnt;
405	struct dquot *dquot;
406
407	ret = err = 0;
408	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
409		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
410		if (dquot)
411			/* Even in case of error we have to continue */
412			ret = mark_dquot_dirty(dquot);
413		if (!err && ret < 0)
414			err = ret;
415	}
416	return err;
417}
418
419static inline void dqput_all(struct dquot **dquot)
420{
421	unsigned int cnt;
422
423	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
424		dqput(dquot[cnt]);
425}
426
427static inline int clear_dquot_dirty(struct dquot *dquot)
428{
429	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
430		return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
431
432	spin_lock(&dq_list_lock);
433	if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
434		spin_unlock(&dq_list_lock);
435		return 0;
436	}
437	list_del_init(&dquot->dq_dirty);
438	spin_unlock(&dq_list_lock);
439	return 1;
440}
441
442void mark_info_dirty(struct super_block *sb, int type)
443{
444	spin_lock(&dq_data_lock);
445	sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
446	spin_unlock(&dq_data_lock);
447}
448EXPORT_SYMBOL(mark_info_dirty);
449
450/*
451 *	Read dquot from disk and alloc space for it
452 */
453
454int dquot_acquire(struct dquot *dquot)
455{
456	int ret = 0, ret2 = 0;
457	unsigned int memalloc;
458	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
459
460	mutex_lock(&dquot->dq_lock);
461	memalloc = memalloc_nofs_save();
462	if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
463		ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
464		if (ret < 0)
465			goto out_iolock;
466	}
467	/* Make sure flags update is visible after dquot has been filled */
468	smp_mb__before_atomic();
469	set_bit(DQ_READ_B, &dquot->dq_flags);
470	/* Instantiate dquot if needed */
471	if (!dquot_active(dquot) && !dquot->dq_off) {
472		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
473		/* Write the info if needed */
474		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
475			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
476					dquot->dq_sb, dquot->dq_id.type);
477		}
478		if (ret < 0)
479			goto out_iolock;
480		if (ret2 < 0) {
481			ret = ret2;
482			goto out_iolock;
483		}
484	}
485	/*
486	 * Make sure flags update is visible after on-disk struct has been
487	 * allocated. Paired with smp_rmb() in dqget().
488	 */
489	smp_mb__before_atomic();
490	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
491out_iolock:
492	memalloc_nofs_restore(memalloc);
493	mutex_unlock(&dquot->dq_lock);
494	return ret;
495}
496EXPORT_SYMBOL(dquot_acquire);
497
498/*
499 *	Write dquot to disk
500 */
501int dquot_commit(struct dquot *dquot)
502{
503	int ret = 0;
504	unsigned int memalloc;
505	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
506
507	mutex_lock(&dquot->dq_lock);
508	memalloc = memalloc_nofs_save();
509	if (!clear_dquot_dirty(dquot))
510		goto out_lock;
511	/* Inactive dquot can be only if there was error during read/init
512	 * => we have better not writing it */
513	if (dquot_active(dquot))
514		ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
515	else
516		ret = -EIO;
517out_lock:
518	memalloc_nofs_restore(memalloc);
519	mutex_unlock(&dquot->dq_lock);
520	return ret;
521}
522EXPORT_SYMBOL(dquot_commit);
523
524/*
525 *	Release dquot
526 */
527int dquot_release(struct dquot *dquot)
528{
529	int ret = 0, ret2 = 0;
530	unsigned int memalloc;
531	struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
532
533	mutex_lock(&dquot->dq_lock);
534	memalloc = memalloc_nofs_save();
535	/* Check whether we are not racing with some other dqget() */
536	if (dquot_is_busy(dquot))
537		goto out_dqlock;
538	if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
539		ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
540		/* Write the info */
541		if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
542			ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
543						dquot->dq_sb, dquot->dq_id.type);
544		}
545		if (ret >= 0)
546			ret = ret2;
547	}
548	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
549out_dqlock:
550	memalloc_nofs_restore(memalloc);
551	mutex_unlock(&dquot->dq_lock);
552	return ret;
553}
554EXPORT_SYMBOL(dquot_release);
555
556void dquot_destroy(struct dquot *dquot)
557{
558	kmem_cache_free(dquot_cachep, dquot);
559}
560EXPORT_SYMBOL(dquot_destroy);
561
562static inline void do_destroy_dquot(struct dquot *dquot)
563{
564	dquot->dq_sb->dq_op->destroy_dquot(dquot);
565}
566
567/* Invalidate all dquots on the list. Note that this function is called after
568 * quota is disabled and pointers from inodes removed so there cannot be new
569 * quota users. There can still be some users of quotas due to inodes being
570 * just deleted or pruned by prune_icache() (those are not attached to any
571 * list) or parallel quotactl call. We have to wait for such users.
572 */
573static void invalidate_dquots(struct super_block *sb, int type)
574{
575	struct dquot *dquot, *tmp;
576
577restart:
578	flush_delayed_work(&quota_release_work);
579
580	spin_lock(&dq_list_lock);
581	list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
582		if (dquot->dq_sb != sb)
583			continue;
584		if (dquot->dq_id.type != type)
585			continue;
586		/* Wait for dquot users */
587		if (atomic_read(&dquot->dq_count)) {
588			atomic_inc(&dquot->dq_count);
589			spin_unlock(&dq_list_lock);
590			/*
591			 * Once dqput() wakes us up, we know it's time to free
592			 * the dquot.
593			 * IMPORTANT: we rely on the fact that there is always
594			 * at most one process waiting for dquot to free.
595			 * Otherwise dq_count would be > 1 and we would never
596			 * wake up.
597			 */
598			wait_event(dquot_ref_wq,
599				   atomic_read(&dquot->dq_count) == 1);
600			dqput(dquot);
601			/* At this moment dquot() need not exist (it could be
602			 * reclaimed by prune_dqcache(). Hence we must
603			 * restart. */
604			goto restart;
605		}
606		/*
607		 * The last user already dropped its reference but dquot didn't
608		 * get fully cleaned up yet. Restart the scan which flushes the
609		 * work cleaning up released dquots.
610		 */
611		if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
612			spin_unlock(&dq_list_lock);
613			goto restart;
614		}
615		/*
616		 * Quota now has no users and it has been written on last
617		 * dqput()
618		 */
619		remove_dquot_hash(dquot);
620		remove_free_dquot(dquot);
621		remove_inuse(dquot);
622		do_destroy_dquot(dquot);
623	}
624	spin_unlock(&dq_list_lock);
625}
626
627/* Call callback for every active dquot on given filesystem */
628int dquot_scan_active(struct super_block *sb,
629		      int (*fn)(struct dquot *dquot, unsigned long priv),
630		      unsigned long priv)
631{
632	struct dquot *dquot, *old_dquot = NULL;
633	int ret = 0;
634
635	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
636
637	spin_lock(&dq_list_lock);
638	list_for_each_entry(dquot, &inuse_list, dq_inuse) {
639		if (!dquot_active(dquot))
640			continue;
641		if (dquot->dq_sb != sb)
642			continue;
643		/* Now we have active dquot so we can just increase use count */
644		atomic_inc(&dquot->dq_count);
645		spin_unlock(&dq_list_lock);
646		dqput(old_dquot);
647		old_dquot = dquot;
648		/*
649		 * ->release_dquot() can be racing with us. Our reference
650		 * protects us from new calls to it so just wait for any
651		 * outstanding call and recheck the DQ_ACTIVE_B after that.
652		 */
653		wait_on_dquot(dquot);
654		if (dquot_active(dquot)) {
655			ret = fn(dquot, priv);
656			if (ret < 0)
657				goto out;
658		}
659		spin_lock(&dq_list_lock);
660		/* We are safe to continue now because our dquot could not
661		 * be moved out of the inuse list while we hold the reference */
662	}
663	spin_unlock(&dq_list_lock);
664out:
665	dqput(old_dquot);
666	return ret;
667}
668EXPORT_SYMBOL(dquot_scan_active);
669
670static inline int dquot_write_dquot(struct dquot *dquot)
671{
672	int ret = dquot->dq_sb->dq_op->write_dquot(dquot);
673	if (ret < 0) {
674		quota_error(dquot->dq_sb, "Can't write quota structure "
675			    "(error %d). Quota may get out of sync!", ret);
676		/* Clear dirty bit anyway to avoid infinite loop. */
677		clear_dquot_dirty(dquot);
678	}
679	return ret;
680}
681
682/* Write all dquot structures to quota files */
683int dquot_writeback_dquots(struct super_block *sb, int type)
684{
685	struct list_head dirty;
686	struct dquot *dquot;
687	struct quota_info *dqopt = sb_dqopt(sb);
688	int cnt;
689	int err, ret = 0;
690
691	WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
692
693	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
694		if (type != -1 && cnt != type)
695			continue;
696		if (!sb_has_quota_active(sb, cnt))
697			continue;
698		spin_lock(&dq_list_lock);
699		/* Move list away to avoid livelock. */
700		list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
701		while (!list_empty(&dirty)) {
702			dquot = list_first_entry(&dirty, struct dquot,
703						 dq_dirty);
704
705			WARN_ON(!dquot_active(dquot));
706			/* If the dquot is releasing we should not touch it */
707			if (test_bit(DQ_RELEASING_B, &dquot->dq_flags)) {
708				spin_unlock(&dq_list_lock);
709				flush_delayed_work(&quota_release_work);
710				spin_lock(&dq_list_lock);
711				continue;
712			}
713
714			/* Now we have active dquot from which someone is
715 			 * holding reference so we can safely just increase
716			 * use count */
717			dqgrab(dquot);
718			spin_unlock(&dq_list_lock);
719			err = dquot_write_dquot(dquot);
720			if (err && !ret)
721				ret = err;
722			dqput(dquot);
723			spin_lock(&dq_list_lock);
724		}
725		spin_unlock(&dq_list_lock);
726	}
727
728	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
729		if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
730		    && info_dirty(&dqopt->info[cnt]))
731			sb->dq_op->write_info(sb, cnt);
732	dqstats_inc(DQST_SYNCS);
733
734	return ret;
735}
736EXPORT_SYMBOL(dquot_writeback_dquots);
737
738/* Write all dquot structures to disk and make them visible from userspace */
739int dquot_quota_sync(struct super_block *sb, int type)
740{
741	struct quota_info *dqopt = sb_dqopt(sb);
742	int cnt;
743	int ret;
744
745	ret = dquot_writeback_dquots(sb, type);
746	if (ret)
747		return ret;
748	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
749		return 0;
750
751	/* This is not very clever (and fast) but currently I don't know about
752	 * any other simple way of getting quota data to disk and we must get
753	 * them there for userspace to be visible... */
754	if (sb->s_op->sync_fs) {
755		ret = sb->s_op->sync_fs(sb, 1);
756		if (ret)
757			return ret;
758	}
759	ret = sync_blockdev(sb->s_bdev);
760	if (ret)
761		return ret;
762
763	/*
764	 * Now when everything is written we can discard the pagecache so
765	 * that userspace sees the changes.
766	 */
767	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
768		if (type != -1 && cnt != type)
769			continue;
770		if (!sb_has_quota_active(sb, cnt))
771			continue;
772		inode_lock(dqopt->files[cnt]);
773		truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
774		inode_unlock(dqopt->files[cnt]);
775	}
776
777	return 0;
778}
779EXPORT_SYMBOL(dquot_quota_sync);
780
781static unsigned long
782dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
783{
784	struct dquot *dquot;
785	unsigned long freed = 0;
786
787	spin_lock(&dq_list_lock);
788	while (!list_empty(&free_dquots) && sc->nr_to_scan) {
789		dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
790		remove_dquot_hash(dquot);
791		remove_free_dquot(dquot);
792		remove_inuse(dquot);
793		do_destroy_dquot(dquot);
794		sc->nr_to_scan--;
795		freed++;
796	}
797	spin_unlock(&dq_list_lock);
798	return freed;
799}
800
801static unsigned long
802dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
803{
804	return vfs_pressure_ratio(
805	percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
806}
807
808/*
809 * Safely release dquot and put reference to dquot.
810 */
811static void quota_release_workfn(struct work_struct *work)
812{
813	struct dquot *dquot;
814	struct list_head rls_head;
815
816	spin_lock(&dq_list_lock);
817	/* Exchange the list head to avoid livelock. */
818	list_replace_init(&releasing_dquots, &rls_head);
819	spin_unlock(&dq_list_lock);
820	synchronize_srcu(&dquot_srcu);
821
822restart:
823	spin_lock(&dq_list_lock);
824	while (!list_empty(&rls_head)) {
825		dquot = list_first_entry(&rls_head, struct dquot, dq_free);
826		WARN_ON_ONCE(atomic_read(&dquot->dq_count));
827		/*
828		 * Note that DQ_RELEASING_B protects us from racing with
829		 * invalidate_dquots() calls so we are safe to work with the
830		 * dquot even after we drop dq_list_lock.
831		 */
832		if (dquot_dirty(dquot)) {
833			spin_unlock(&dq_list_lock);
834			/* Commit dquot before releasing */
835			dquot_write_dquot(dquot);
836			goto restart;
837		}
838		if (dquot_active(dquot)) {
839			spin_unlock(&dq_list_lock);
840			dquot->dq_sb->dq_op->release_dquot(dquot);
841			goto restart;
842		}
843		/* Dquot is inactive and clean, now move it to free list */
844		remove_free_dquot(dquot);
845		put_dquot_last(dquot);
846	}
847	spin_unlock(&dq_list_lock);
848}
849
850/*
851 * Put reference to dquot
852 */
853void dqput(struct dquot *dquot)
854{
855	if (!dquot)
856		return;
857#ifdef CONFIG_QUOTA_DEBUG
858	if (!atomic_read(&dquot->dq_count)) {
859		quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
860			    quotatypes[dquot->dq_id.type],
861			    from_kqid(&init_user_ns, dquot->dq_id));
862		BUG();
863	}
864#endif
865	dqstats_inc(DQST_DROPS);
866
867	spin_lock(&dq_list_lock);
868	if (atomic_read(&dquot->dq_count) > 1) {
869		/* We have more than one user... nothing to do */
870		atomic_dec(&dquot->dq_count);
871		/* Releasing dquot during quotaoff phase? */
872		if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
873		    atomic_read(&dquot->dq_count) == 1)
874			wake_up(&dquot_ref_wq);
875		spin_unlock(&dq_list_lock);
876		return;
877	}
878
879	/* Need to release dquot? */
880	WARN_ON_ONCE(!list_empty(&dquot->dq_free));
881	put_releasing_dquots(dquot);
882	atomic_dec(&dquot->dq_count);
883	spin_unlock(&dq_list_lock);
884	queue_delayed_work(system_unbound_wq, &quota_release_work, 1);
885}
886EXPORT_SYMBOL(dqput);
887
888struct dquot *dquot_alloc(struct super_block *sb, int type)
889{
890	return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
891}
892EXPORT_SYMBOL(dquot_alloc);
893
894static struct dquot *get_empty_dquot(struct super_block *sb, int type)
895{
896	struct dquot *dquot;
897
898	dquot = sb->dq_op->alloc_dquot(sb, type);
899	if(!dquot)
900		return NULL;
901
902	mutex_init(&dquot->dq_lock);
903	INIT_LIST_HEAD(&dquot->dq_free);
904	INIT_LIST_HEAD(&dquot->dq_inuse);
905	INIT_HLIST_NODE(&dquot->dq_hash);
906	INIT_LIST_HEAD(&dquot->dq_dirty);
907	dquot->dq_sb = sb;
908	dquot->dq_id = make_kqid_invalid(type);
909	atomic_set(&dquot->dq_count, 1);
910	spin_lock_init(&dquot->dq_dqb_lock);
911
912	return dquot;
913}
914
915/*
916 * Get reference to dquot
917 *
918 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
919 * destroying our dquot by:
920 *   a) checking for quota flags under dq_list_lock and
921 *   b) getting a reference to dquot before we release dq_list_lock
922 */
923struct dquot *dqget(struct super_block *sb, struct kqid qid)
924{
925	unsigned int hashent = hashfn(sb, qid);
926	struct dquot *dquot, *empty = NULL;
927
928	if (!qid_has_mapping(sb->s_user_ns, qid))
929		return ERR_PTR(-EINVAL);
930
931        if (!sb_has_quota_active(sb, qid.type))
932		return ERR_PTR(-ESRCH);
933we_slept:
934	spin_lock(&dq_list_lock);
935	spin_lock(&dq_state_lock);
936	if (!sb_has_quota_active(sb, qid.type)) {
937		spin_unlock(&dq_state_lock);
938		spin_unlock(&dq_list_lock);
939		dquot = ERR_PTR(-ESRCH);
940		goto out;
941	}
942	spin_unlock(&dq_state_lock);
943
944	dquot = find_dquot(hashent, sb, qid);
945	if (!dquot) {
946		if (!empty) {
947			spin_unlock(&dq_list_lock);
948			empty = get_empty_dquot(sb, qid.type);
949			if (!empty)
950				schedule();	/* Try to wait for a moment... */
951			goto we_slept;
952		}
953		dquot = empty;
954		empty = NULL;
955		dquot->dq_id = qid;
956		/* all dquots go on the inuse_list */
957		put_inuse(dquot);
958		/* hash it first so it can be found */
959		insert_dquot_hash(dquot);
960		spin_unlock(&dq_list_lock);
961		dqstats_inc(DQST_LOOKUPS);
962	} else {
963		if (!atomic_read(&dquot->dq_count))
964			remove_free_dquot(dquot);
965		atomic_inc(&dquot->dq_count);
966		spin_unlock(&dq_list_lock);
967		dqstats_inc(DQST_CACHE_HITS);
968		dqstats_inc(DQST_LOOKUPS);
969	}
970	/* Wait for dq_lock - after this we know that either dquot_release() is
971	 * already finished or it will be canceled due to dq_count > 0 test */
972	wait_on_dquot(dquot);
973	/* Read the dquot / allocate space in quota file */
974	if (!dquot_active(dquot)) {
975		int err;
976
977		err = sb->dq_op->acquire_dquot(dquot);
978		if (err < 0) {
979			dqput(dquot);
980			dquot = ERR_PTR(err);
981			goto out;
982		}
983	}
984	/*
985	 * Make sure following reads see filled structure - paired with
986	 * smp_mb__before_atomic() in dquot_acquire().
987	 */
988	smp_rmb();
989	/* Has somebody invalidated entry under us? */
990	WARN_ON_ONCE(hlist_unhashed(&dquot->dq_hash));
991out:
992	if (empty)
993		do_destroy_dquot(empty);
994
995	return dquot;
996}
997EXPORT_SYMBOL(dqget);
998
999static inline struct dquot __rcu **i_dquot(struct inode *inode)
1000{
1001	return inode->i_sb->s_op->get_dquots(inode);
1002}
1003
1004static int dqinit_needed(struct inode *inode, int type)
1005{
1006	struct dquot __rcu * const *dquots;
1007	int cnt;
1008
1009	if (IS_NOQUOTA(inode))
1010		return 0;
1011
1012	dquots = i_dquot(inode);
1013	if (type != -1)
1014		return !dquots[type];
1015	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1016		if (!dquots[cnt])
1017			return 1;
1018	return 0;
1019}
1020
1021/* This routine is guarded by s_umount semaphore */
1022static int add_dquot_ref(struct super_block *sb, int type)
1023{
1024	struct inode *inode, *old_inode = NULL;
1025#ifdef CONFIG_QUOTA_DEBUG
1026	int reserved = 0;
1027#endif
1028	int err = 0;
1029
1030	spin_lock(&sb->s_inode_list_lock);
1031	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1032		spin_lock(&inode->i_lock);
1033		if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1034		    !atomic_read(&inode->i_writecount) ||
1035		    !dqinit_needed(inode, type)) {
1036			spin_unlock(&inode->i_lock);
1037			continue;
1038		}
1039		__iget(inode);
1040		spin_unlock(&inode->i_lock);
1041		spin_unlock(&sb->s_inode_list_lock);
1042
1043#ifdef CONFIG_QUOTA_DEBUG
1044		if (unlikely(inode_get_rsv_space(inode) > 0))
1045			reserved = 1;
1046#endif
1047		iput(old_inode);
1048		err = __dquot_initialize(inode, type);
1049		if (err) {
1050			iput(inode);
1051			goto out;
1052		}
1053
1054		/*
1055		 * We hold a reference to 'inode' so it couldn't have been
1056		 * removed from s_inodes list while we dropped the
1057		 * s_inode_list_lock. We cannot iput the inode now as we can be
1058		 * holding the last reference and we cannot iput it under
1059		 * s_inode_list_lock. So we keep the reference and iput it
1060		 * later.
1061		 */
1062		old_inode = inode;
1063		cond_resched();
1064		spin_lock(&sb->s_inode_list_lock);
1065	}
1066	spin_unlock(&sb->s_inode_list_lock);
1067	iput(old_inode);
1068out:
1069#ifdef CONFIG_QUOTA_DEBUG
1070	if (reserved) {
1071		quota_error(sb, "Writes happened before quota was turned on "
1072			"thus quota information is probably inconsistent. "
1073			"Please run quotacheck(8)");
1074	}
1075#endif
1076	return err;
1077}
1078
1079static void remove_dquot_ref(struct super_block *sb, int type)
1080{
1081	struct inode *inode;
1082#ifdef CONFIG_QUOTA_DEBUG
1083	int reserved = 0;
1084#endif
1085
1086	spin_lock(&sb->s_inode_list_lock);
1087	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1088		/*
1089		 *  We have to scan also I_NEW inodes because they can already
1090		 *  have quota pointer initialized. Luckily, we need to touch
1091		 *  only quota pointers and these have separate locking
1092		 *  (dq_data_lock).
1093		 */
1094		spin_lock(&dq_data_lock);
1095		if (!IS_NOQUOTA(inode)) {
1096			struct dquot __rcu **dquots = i_dquot(inode);
1097			struct dquot *dquot = srcu_dereference_check(
1098				dquots[type], &dquot_srcu,
1099				lockdep_is_held(&dq_data_lock));
1100
1101#ifdef CONFIG_QUOTA_DEBUG
1102			if (unlikely(inode_get_rsv_space(inode) > 0))
1103				reserved = 1;
1104#endif
1105			rcu_assign_pointer(dquots[type], NULL);
1106			if (dquot)
1107				dqput(dquot);
1108		}
1109		spin_unlock(&dq_data_lock);
1110	}
1111	spin_unlock(&sb->s_inode_list_lock);
1112#ifdef CONFIG_QUOTA_DEBUG
1113	if (reserved) {
1114		printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1115			" was disabled thus quota information is probably "
1116			"inconsistent. Please run quotacheck(8).\n", sb->s_id);
1117	}
1118#endif
1119}
1120
1121/* Gather all references from inodes and drop them */
1122static void drop_dquot_ref(struct super_block *sb, int type)
1123{
1124	if (sb->dq_op)
1125		remove_dquot_ref(sb, type);
1126}
1127
1128static inline
1129void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1130{
1131	if (dquot->dq_dqb.dqb_rsvspace >= number)
1132		dquot->dq_dqb.dqb_rsvspace -= number;
1133	else {
1134		WARN_ON_ONCE(1);
1135		dquot->dq_dqb.dqb_rsvspace = 0;
1136	}
1137	if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1138	    dquot->dq_dqb.dqb_bsoftlimit)
1139		dquot->dq_dqb.dqb_btime = (time64_t) 0;
1140	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1141}
1142
1143static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1144{
1145	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1146	    dquot->dq_dqb.dqb_curinodes >= number)
1147		dquot->dq_dqb.dqb_curinodes -= number;
1148	else
1149		dquot->dq_dqb.dqb_curinodes = 0;
1150	if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1151		dquot->dq_dqb.dqb_itime = (time64_t) 0;
1152	clear_bit(DQ_INODES_B, &dquot->dq_flags);
1153}
1154
1155static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1156{
1157	if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1158	    dquot->dq_dqb.dqb_curspace >= number)
1159		dquot->dq_dqb.dqb_curspace -= number;
1160	else
1161		dquot->dq_dqb.dqb_curspace = 0;
1162	if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1163	    dquot->dq_dqb.dqb_bsoftlimit)
1164		dquot->dq_dqb.dqb_btime = (time64_t) 0;
1165	clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1166}
1167
1168struct dquot_warn {
1169	struct super_block *w_sb;
1170	struct kqid w_dq_id;
1171	short w_type;
1172};
1173
1174static int warning_issued(struct dquot *dquot, const int warntype)
1175{
1176	int flag = (warntype == QUOTA_NL_BHARDWARN ||
1177		warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1178		((warntype == QUOTA_NL_IHARDWARN ||
1179		warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1180
1181	if (!flag)
1182		return 0;
1183	return test_and_set_bit(flag, &dquot->dq_flags);
1184}
1185
1186#ifdef CONFIG_PRINT_QUOTA_WARNING
1187static int flag_print_warnings = 1;
1188
1189static int need_print_warning(struct dquot_warn *warn)
1190{
1191	if (!flag_print_warnings)
1192		return 0;
1193
1194	switch (warn->w_dq_id.type) {
1195		case USRQUOTA:
1196			return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1197		case GRPQUOTA:
1198			return in_group_p(warn->w_dq_id.gid);
1199		case PRJQUOTA:
1200			return 1;
1201	}
1202	return 0;
1203}
1204
1205/* Print warning to user which exceeded quota */
1206static void print_warning(struct dquot_warn *warn)
1207{
1208	char *msg = NULL;
1209	struct tty_struct *tty;
1210	int warntype = warn->w_type;
1211
1212	if (warntype == QUOTA_NL_IHARDBELOW ||
1213	    warntype == QUOTA_NL_ISOFTBELOW ||
1214	    warntype == QUOTA_NL_BHARDBELOW ||
1215	    warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1216		return;
1217
1218	tty = get_current_tty();
1219	if (!tty)
1220		return;
1221	tty_write_message(tty, warn->w_sb->s_id);
1222	if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1223		tty_write_message(tty, ": warning, ");
1224	else
1225		tty_write_message(tty, ": write failed, ");
1226	tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1227	switch (warntype) {
1228		case QUOTA_NL_IHARDWARN:
1229			msg = " file limit reached.\r\n";
1230			break;
1231		case QUOTA_NL_ISOFTLONGWARN:
1232			msg = " file quota exceeded too long.\r\n";
1233			break;
1234		case QUOTA_NL_ISOFTWARN:
1235			msg = " file quota exceeded.\r\n";
1236			break;
1237		case QUOTA_NL_BHARDWARN:
1238			msg = " block limit reached.\r\n";
1239			break;
1240		case QUOTA_NL_BSOFTLONGWARN:
1241			msg = " block quota exceeded too long.\r\n";
1242			break;
1243		case QUOTA_NL_BSOFTWARN:
1244			msg = " block quota exceeded.\r\n";
1245			break;
1246	}
1247	tty_write_message(tty, msg);
1248	tty_kref_put(tty);
1249}
1250#endif
1251
1252static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1253			    int warntype)
1254{
1255	if (warning_issued(dquot, warntype))
1256		return;
1257	warn->w_type = warntype;
1258	warn->w_sb = dquot->dq_sb;
1259	warn->w_dq_id = dquot->dq_id;
1260}
1261
1262/*
1263 * Write warnings to the console and send warning messages over netlink.
1264 *
1265 * Note that this function can call into tty and networking code.
1266 */
1267static void flush_warnings(struct dquot_warn *warn)
1268{
1269	int i;
1270
1271	for (i = 0; i < MAXQUOTAS; i++) {
1272		if (warn[i].w_type == QUOTA_NL_NOWARN)
1273			continue;
1274#ifdef CONFIG_PRINT_QUOTA_WARNING
1275		print_warning(&warn[i]);
1276#endif
1277		quota_send_warning(warn[i].w_dq_id,
1278				   warn[i].w_sb->s_dev, warn[i].w_type);
1279	}
1280}
1281
1282static int ignore_hardlimit(struct dquot *dquot)
1283{
1284	struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1285
1286	return capable(CAP_SYS_RESOURCE) &&
1287	       (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1288		!(info->dqi_flags & DQF_ROOT_SQUASH));
1289}
1290
1291static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1292			    struct dquot_warn *warn)
1293{
1294	qsize_t newinodes;
1295	int ret = 0;
1296
1297	spin_lock(&dquot->dq_dqb_lock);
1298	newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1299	if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1300	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1301		goto add;
1302
1303	if (dquot->dq_dqb.dqb_ihardlimit &&
1304	    newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1305            !ignore_hardlimit(dquot)) {
1306		prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1307		ret = -EDQUOT;
1308		goto out;
1309	}
1310
1311	if (dquot->dq_dqb.dqb_isoftlimit &&
1312	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1313	    dquot->dq_dqb.dqb_itime &&
1314	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1315            !ignore_hardlimit(dquot)) {
1316		prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1317		ret = -EDQUOT;
1318		goto out;
1319	}
1320
1321	if (dquot->dq_dqb.dqb_isoftlimit &&
1322	    newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1323	    dquot->dq_dqb.dqb_itime == 0) {
1324		prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1325		dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1326		    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1327	}
1328add:
1329	dquot->dq_dqb.dqb_curinodes = newinodes;
1330
1331out:
1332	spin_unlock(&dquot->dq_dqb_lock);
1333	return ret;
1334}
1335
1336static int dquot_add_space(struct dquot *dquot, qsize_t space,
1337			   qsize_t rsv_space, unsigned int flags,
1338			   struct dquot_warn *warn)
1339{
1340	qsize_t tspace;
1341	struct super_block *sb = dquot->dq_sb;
1342	int ret = 0;
1343
1344	spin_lock(&dquot->dq_dqb_lock);
1345	if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1346	    test_bit(DQ_FAKE_B, &dquot->dq_flags))
1347		goto finish;
1348
1349	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1350		+ space + rsv_space;
1351
1352	if (dquot->dq_dqb.dqb_bhardlimit &&
1353	    tspace > dquot->dq_dqb.dqb_bhardlimit &&
1354            !ignore_hardlimit(dquot)) {
1355		if (flags & DQUOT_SPACE_WARN)
1356			prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1357		ret = -EDQUOT;
1358		goto finish;
1359	}
1360
1361	if (dquot->dq_dqb.dqb_bsoftlimit &&
1362	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1363	    dquot->dq_dqb.dqb_btime &&
1364	    ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1365            !ignore_hardlimit(dquot)) {
1366		if (flags & DQUOT_SPACE_WARN)
1367			prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1368		ret = -EDQUOT;
1369		goto finish;
1370	}
1371
1372	if (dquot->dq_dqb.dqb_bsoftlimit &&
1373	    tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1374	    dquot->dq_dqb.dqb_btime == 0) {
1375		if (flags & DQUOT_SPACE_WARN) {
1376			prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1377			dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1378			    sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1379		} else {
1380			/*
1381			 * We don't allow preallocation to exceed softlimit so exceeding will
1382			 * be always printed
1383			 */
1384			ret = -EDQUOT;
1385			goto finish;
1386		}
1387	}
1388finish:
1389	/*
1390	 * We have to be careful and go through warning generation & grace time
1391	 * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1392	 * only here...
1393	 */
1394	if (flags & DQUOT_SPACE_NOFAIL)
1395		ret = 0;
1396	if (!ret) {
1397		dquot->dq_dqb.dqb_rsvspace += rsv_space;
1398		dquot->dq_dqb.dqb_curspace += space;
1399	}
1400	spin_unlock(&dquot->dq_dqb_lock);
1401	return ret;
1402}
1403
1404static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1405{
1406	qsize_t newinodes;
1407
1408	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1409	    dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1410	    !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1411		return QUOTA_NL_NOWARN;
1412
1413	newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1414	if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1415		return QUOTA_NL_ISOFTBELOW;
1416	if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1417	    newinodes < dquot->dq_dqb.dqb_ihardlimit)
1418		return QUOTA_NL_IHARDBELOW;
1419	return QUOTA_NL_NOWARN;
1420}
1421
1422static int info_bdq_free(struct dquot *dquot, qsize_t space)
1423{
1424	qsize_t tspace;
1425
1426	tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1427
1428	if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1429	    tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1430		return QUOTA_NL_NOWARN;
1431
1432	if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1433		return QUOTA_NL_BSOFTBELOW;
1434	if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1435	    tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1436		return QUOTA_NL_BHARDBELOW;
1437	return QUOTA_NL_NOWARN;
1438}
1439
1440static int inode_quota_active(const struct inode *inode)
1441{
1442	struct super_block *sb = inode->i_sb;
1443
1444	if (IS_NOQUOTA(inode))
1445		return 0;
1446	return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1447}
1448
1449/*
1450 * Initialize quota pointers in inode
1451 *
1452 * It is better to call this function outside of any transaction as it
1453 * might need a lot of space in journal for dquot structure allocation.
1454 */
1455static int __dquot_initialize(struct inode *inode, int type)
1456{
1457	int cnt, init_needed = 0;
1458	struct dquot __rcu **dquots;
1459	struct dquot *got[MAXQUOTAS] = {};
1460	struct super_block *sb = inode->i_sb;
1461	qsize_t rsv;
1462	int ret = 0;
1463
1464	if (!inode_quota_active(inode))
1465		return 0;
1466
1467	dquots = i_dquot(inode);
1468
1469	/* First get references to structures we might need. */
1470	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1471		struct kqid qid;
1472		kprojid_t projid;
1473		int rc;
1474		struct dquot *dquot;
1475
1476		if (type != -1 && cnt != type)
1477			continue;
1478		/*
1479		 * The i_dquot should have been initialized in most cases,
1480		 * we check it without locking here to avoid unnecessary
1481		 * dqget()/dqput() calls.
1482		 */
1483		if (dquots[cnt])
1484			continue;
1485
1486		if (!sb_has_quota_active(sb, cnt))
1487			continue;
1488
1489		init_needed = 1;
1490
1491		switch (cnt) {
1492		case USRQUOTA:
1493			qid = make_kqid_uid(inode->i_uid);
1494			break;
1495		case GRPQUOTA:
1496			qid = make_kqid_gid(inode->i_gid);
1497			break;
1498		case PRJQUOTA:
1499			rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1500			if (rc)
1501				continue;
1502			qid = make_kqid_projid(projid);
1503			break;
1504		}
1505		dquot = dqget(sb, qid);
1506		if (IS_ERR(dquot)) {
1507			/* We raced with somebody turning quotas off... */
1508			if (PTR_ERR(dquot) != -ESRCH) {
1509				ret = PTR_ERR(dquot);
1510				goto out_put;
1511			}
1512			dquot = NULL;
1513		}
1514		got[cnt] = dquot;
1515	}
1516
1517	/* All required i_dquot has been initialized */
1518	if (!init_needed)
1519		return 0;
1520
1521	spin_lock(&dq_data_lock);
1522	if (IS_NOQUOTA(inode))
1523		goto out_lock;
1524	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1525		if (type != -1 && cnt != type)
1526			continue;
1527		/* Avoid races with quotaoff() */
1528		if (!sb_has_quota_active(sb, cnt))
1529			continue;
1530		/* We could race with quotaon or dqget() could have failed */
1531		if (!got[cnt])
1532			continue;
1533		if (!dquots[cnt]) {
1534			rcu_assign_pointer(dquots[cnt], got[cnt]);
1535			got[cnt] = NULL;
1536			/*
1537			 * Make quota reservation system happy if someone
1538			 * did a write before quota was turned on
1539			 */
1540			rsv = inode_get_rsv_space(inode);
1541			if (unlikely(rsv)) {
1542				struct dquot *dquot = srcu_dereference_check(
1543					dquots[cnt], &dquot_srcu,
1544					lockdep_is_held(&dq_data_lock));
1545
1546				spin_lock(&inode->i_lock);
1547				/* Get reservation again under proper lock */
1548				rsv = __inode_get_rsv_space(inode);
1549				spin_lock(&dquot->dq_dqb_lock);
1550				dquot->dq_dqb.dqb_rsvspace += rsv;
1551				spin_unlock(&dquot->dq_dqb_lock);
1552				spin_unlock(&inode->i_lock);
1553			}
1554		}
1555	}
1556out_lock:
1557	spin_unlock(&dq_data_lock);
1558out_put:
1559	/* Drop unused references */
1560	dqput_all(got);
1561
1562	return ret;
1563}
1564
1565int dquot_initialize(struct inode *inode)
1566{
1567	return __dquot_initialize(inode, -1);
1568}
1569EXPORT_SYMBOL(dquot_initialize);
1570
1571bool dquot_initialize_needed(struct inode *inode)
1572{
1573	struct dquot __rcu **dquots;
1574	int i;
1575
1576	if (!inode_quota_active(inode))
1577		return false;
1578
1579	dquots = i_dquot(inode);
1580	for (i = 0; i < MAXQUOTAS; i++)
1581		if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1582			return true;
1583	return false;
1584}
1585EXPORT_SYMBOL(dquot_initialize_needed);
1586
1587/*
1588 * Release all quotas referenced by inode.
1589 *
1590 * This function only be called on inode free or converting
1591 * a file to quota file, no other users for the i_dquot in
1592 * both cases, so we needn't call synchronize_srcu() after
1593 * clearing i_dquot.
1594 */
1595static void __dquot_drop(struct inode *inode)
1596{
1597	int cnt;
1598	struct dquot __rcu **dquots = i_dquot(inode);
1599	struct dquot *put[MAXQUOTAS];
1600
1601	spin_lock(&dq_data_lock);
1602	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1603		put[cnt] = srcu_dereference_check(dquots[cnt], &dquot_srcu,
1604					lockdep_is_held(&dq_data_lock));
1605		rcu_assign_pointer(dquots[cnt], NULL);
1606	}
1607	spin_unlock(&dq_data_lock);
1608	dqput_all(put);
1609}
1610
1611void dquot_drop(struct inode *inode)
1612{
1613	struct dquot __rcu * const *dquots;
1614	int cnt;
1615
1616	if (IS_NOQUOTA(inode))
1617		return;
1618
1619	/*
1620	 * Test before calling to rule out calls from proc and such
1621	 * where we are not allowed to block. Note that this is
1622	 * actually reliable test even without the lock - the caller
1623	 * must assure that nobody can come after the DQUOT_DROP and
1624	 * add quota pointers back anyway.
1625	 */
1626	dquots = i_dquot(inode);
1627	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1628		if (dquots[cnt])
1629			break;
1630	}
1631
1632	if (cnt < MAXQUOTAS)
1633		__dquot_drop(inode);
1634}
1635EXPORT_SYMBOL(dquot_drop);
1636
1637/*
1638 * inode_reserved_space is managed internally by quota, and protected by
1639 * i_lock similar to i_blocks+i_bytes.
1640 */
1641static qsize_t *inode_reserved_space(struct inode * inode)
1642{
1643	/* Filesystem must explicitly define it's own method in order to use
1644	 * quota reservation interface */
1645	BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1646	return inode->i_sb->dq_op->get_reserved_space(inode);
1647}
1648
1649static qsize_t __inode_get_rsv_space(struct inode *inode)
1650{
1651	if (!inode->i_sb->dq_op->get_reserved_space)
1652		return 0;
1653	return *inode_reserved_space(inode);
1654}
1655
1656static qsize_t inode_get_rsv_space(struct inode *inode)
1657{
1658	qsize_t ret;
1659
1660	if (!inode->i_sb->dq_op->get_reserved_space)
1661		return 0;
1662	spin_lock(&inode->i_lock);
1663	ret = __inode_get_rsv_space(inode);
1664	spin_unlock(&inode->i_lock);
1665	return ret;
1666}
1667
1668/*
1669 * This functions updates i_blocks+i_bytes fields and quota information
1670 * (together with appropriate checks).
1671 *
1672 * NOTE: We absolutely rely on the fact that caller dirties the inode
1673 * (usually helpers in quotaops.h care about this) and holds a handle for
1674 * the current transaction so that dquot write and inode write go into the
1675 * same transaction.
1676 */
1677
1678/*
1679 * This operation can block, but only after everything is updated
1680 */
1681int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1682{
1683	int cnt, ret = 0, index;
1684	struct dquot_warn warn[MAXQUOTAS];
1685	int reserve = flags & DQUOT_SPACE_RESERVE;
1686	struct dquot __rcu **dquots;
1687	struct dquot *dquot;
1688
1689	if (!inode_quota_active(inode)) {
1690		if (reserve) {
1691			spin_lock(&inode->i_lock);
1692			*inode_reserved_space(inode) += number;
1693			spin_unlock(&inode->i_lock);
1694		} else {
1695			inode_add_bytes(inode, number);
1696		}
1697		goto out;
1698	}
1699
1700	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1701		warn[cnt].w_type = QUOTA_NL_NOWARN;
1702
1703	dquots = i_dquot(inode);
1704	index = srcu_read_lock(&dquot_srcu);
1705	spin_lock(&inode->i_lock);
1706	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1707		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1708		if (!dquot)
1709			continue;
1710		if (reserve) {
1711			ret = dquot_add_space(dquot, 0, number, flags, &warn[cnt]);
1712		} else {
1713			ret = dquot_add_space(dquot, number, 0, flags, &warn[cnt]);
1714		}
1715		if (ret) {
1716			/* Back out changes we already did */
1717			for (cnt--; cnt >= 0; cnt--) {
1718				dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1719				if (!dquot)
1720					continue;
1721				spin_lock(&dquot->dq_dqb_lock);
1722				if (reserve)
1723					dquot_free_reserved_space(dquot, number);
1724				else
1725					dquot_decr_space(dquot, number);
1726				spin_unlock(&dquot->dq_dqb_lock);
1727			}
1728			spin_unlock(&inode->i_lock);
1729			goto out_flush_warn;
1730		}
1731	}
1732	if (reserve)
1733		*inode_reserved_space(inode) += number;
1734	else
1735		__inode_add_bytes(inode, number);
1736	spin_unlock(&inode->i_lock);
1737
1738	if (reserve)
1739		goto out_flush_warn;
1740	ret = mark_all_dquot_dirty(dquots);
1741out_flush_warn:
1742	srcu_read_unlock(&dquot_srcu, index);
1743	flush_warnings(warn);
1744out:
1745	return ret;
1746}
1747EXPORT_SYMBOL(__dquot_alloc_space);
1748
1749/*
1750 * This operation can block, but only after everything is updated
1751 */
1752int dquot_alloc_inode(struct inode *inode)
1753{
1754	int cnt, ret = 0, index;
1755	struct dquot_warn warn[MAXQUOTAS];
1756	struct dquot __rcu * const *dquots;
1757	struct dquot *dquot;
1758
1759	if (!inode_quota_active(inode))
1760		return 0;
1761	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1762		warn[cnt].w_type = QUOTA_NL_NOWARN;
1763
1764	dquots = i_dquot(inode);
1765	index = srcu_read_lock(&dquot_srcu);
1766	spin_lock(&inode->i_lock);
1767	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1768		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1769		if (!dquot)
1770			continue;
1771		ret = dquot_add_inodes(dquot, 1, &warn[cnt]);
1772		if (ret) {
1773			for (cnt--; cnt >= 0; cnt--) {
1774				dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1775				if (!dquot)
1776					continue;
1777				/* Back out changes we already did */
1778				spin_lock(&dquot->dq_dqb_lock);
1779				dquot_decr_inodes(dquot, 1);
1780				spin_unlock(&dquot->dq_dqb_lock);
1781			}
1782			goto warn_put_all;
1783		}
1784	}
1785
1786warn_put_all:
1787	spin_unlock(&inode->i_lock);
1788	if (ret == 0)
1789		ret = mark_all_dquot_dirty(dquots);
1790	srcu_read_unlock(&dquot_srcu, index);
1791	flush_warnings(warn);
1792	return ret;
1793}
1794EXPORT_SYMBOL(dquot_alloc_inode);
1795
1796/*
1797 * Convert in-memory reserved quotas to real consumed quotas
1798 */
1799void dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1800{
1801	struct dquot __rcu **dquots;
1802	struct dquot *dquot;
1803	int cnt, index;
1804
1805	if (!inode_quota_active(inode)) {
1806		spin_lock(&inode->i_lock);
1807		*inode_reserved_space(inode) -= number;
1808		__inode_add_bytes(inode, number);
1809		spin_unlock(&inode->i_lock);
1810		return;
1811	}
1812
1813	dquots = i_dquot(inode);
1814	index = srcu_read_lock(&dquot_srcu);
1815	spin_lock(&inode->i_lock);
1816	/* Claim reserved quotas to allocated quotas */
1817	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1818		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1819		if (dquot) {
1820			spin_lock(&dquot->dq_dqb_lock);
1821			if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1822				number = dquot->dq_dqb.dqb_rsvspace;
1823			dquot->dq_dqb.dqb_curspace += number;
1824			dquot->dq_dqb.dqb_rsvspace -= number;
1825			spin_unlock(&dquot->dq_dqb_lock);
1826		}
1827	}
1828	/* Update inode bytes */
1829	*inode_reserved_space(inode) -= number;
1830	__inode_add_bytes(inode, number);
1831	spin_unlock(&inode->i_lock);
1832	mark_all_dquot_dirty(dquots);
1833	srcu_read_unlock(&dquot_srcu, index);
1834	return;
1835}
1836EXPORT_SYMBOL(dquot_claim_space_nodirty);
1837
1838/*
1839 * Convert allocated space back to in-memory reserved quotas
1840 */
1841void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1842{
1843	struct dquot __rcu **dquots;
1844	struct dquot *dquot;
1845	int cnt, index;
1846
1847	if (!inode_quota_active(inode)) {
1848		spin_lock(&inode->i_lock);
1849		*inode_reserved_space(inode) += number;
1850		__inode_sub_bytes(inode, number);
1851		spin_unlock(&inode->i_lock);
1852		return;
1853	}
1854
1855	dquots = i_dquot(inode);
1856	index = srcu_read_lock(&dquot_srcu);
1857	spin_lock(&inode->i_lock);
1858	/* Claim reserved quotas to allocated quotas */
1859	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1860		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1861		if (dquot) {
1862			spin_lock(&dquot->dq_dqb_lock);
1863			if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1864				number = dquot->dq_dqb.dqb_curspace;
1865			dquot->dq_dqb.dqb_rsvspace += number;
1866			dquot->dq_dqb.dqb_curspace -= number;
1867			spin_unlock(&dquot->dq_dqb_lock);
1868		}
1869	}
1870	/* Update inode bytes */
1871	*inode_reserved_space(inode) += number;
1872	__inode_sub_bytes(inode, number);
1873	spin_unlock(&inode->i_lock);
1874	mark_all_dquot_dirty(dquots);
1875	srcu_read_unlock(&dquot_srcu, index);
1876	return;
1877}
1878EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1879
1880/*
1881 * This operation can block, but only after everything is updated
1882 */
1883void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1884{
1885	unsigned int cnt;
1886	struct dquot_warn warn[MAXQUOTAS];
1887	struct dquot __rcu **dquots;
1888	struct dquot *dquot;
1889	int reserve = flags & DQUOT_SPACE_RESERVE, index;
1890
1891	if (!inode_quota_active(inode)) {
1892		if (reserve) {
1893			spin_lock(&inode->i_lock);
1894			*inode_reserved_space(inode) -= number;
1895			spin_unlock(&inode->i_lock);
1896		} else {
1897			inode_sub_bytes(inode, number);
1898		}
1899		return;
1900	}
1901
1902	dquots = i_dquot(inode);
1903	index = srcu_read_lock(&dquot_srcu);
1904	spin_lock(&inode->i_lock);
1905	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1906		int wtype;
1907
1908		warn[cnt].w_type = QUOTA_NL_NOWARN;
1909		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1910		if (!dquot)
1911			continue;
1912		spin_lock(&dquot->dq_dqb_lock);
1913		wtype = info_bdq_free(dquot, number);
1914		if (wtype != QUOTA_NL_NOWARN)
1915			prepare_warning(&warn[cnt], dquot, wtype);
1916		if (reserve)
1917			dquot_free_reserved_space(dquot, number);
1918		else
1919			dquot_decr_space(dquot, number);
1920		spin_unlock(&dquot->dq_dqb_lock);
1921	}
1922	if (reserve)
1923		*inode_reserved_space(inode) -= number;
1924	else
1925		__inode_sub_bytes(inode, number);
1926	spin_unlock(&inode->i_lock);
1927
1928	if (reserve)
1929		goto out_unlock;
1930	mark_all_dquot_dirty(dquots);
1931out_unlock:
1932	srcu_read_unlock(&dquot_srcu, index);
1933	flush_warnings(warn);
1934}
1935EXPORT_SYMBOL(__dquot_free_space);
1936
1937/*
1938 * This operation can block, but only after everything is updated
1939 */
1940void dquot_free_inode(struct inode *inode)
1941{
1942	unsigned int cnt;
1943	struct dquot_warn warn[MAXQUOTAS];
1944	struct dquot __rcu * const *dquots;
1945	struct dquot *dquot;
1946	int index;
1947
1948	if (!inode_quota_active(inode))
1949		return;
1950
1951	dquots = i_dquot(inode);
1952	index = srcu_read_lock(&dquot_srcu);
1953	spin_lock(&inode->i_lock);
1954	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1955		int wtype;
1956		warn[cnt].w_type = QUOTA_NL_NOWARN;
1957		dquot = srcu_dereference(dquots[cnt], &dquot_srcu);
1958		if (!dquot)
1959			continue;
1960		spin_lock(&dquot->dq_dqb_lock);
1961		wtype = info_idq_free(dquot, 1);
1962		if (wtype != QUOTA_NL_NOWARN)
1963			prepare_warning(&warn[cnt], dquot, wtype);
1964		dquot_decr_inodes(dquot, 1);
1965		spin_unlock(&dquot->dq_dqb_lock);
1966	}
1967	spin_unlock(&inode->i_lock);
1968	mark_all_dquot_dirty(dquots);
1969	srcu_read_unlock(&dquot_srcu, index);
1970	flush_warnings(warn);
1971}
1972EXPORT_SYMBOL(dquot_free_inode);
1973
1974/*
1975 * Transfer the number of inode and blocks from one diskquota to an other.
1976 * On success, dquot references in transfer_to are consumed and references
1977 * to original dquots that need to be released are placed there. On failure,
1978 * references are kept untouched.
1979 *
1980 * This operation can block, but only after everything is updated
1981 * A transaction must be started when entering this function.
1982 *
1983 * We are holding reference on transfer_from & transfer_to, no need to
1984 * protect them by srcu_read_lock().
1985 */
1986int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1987{
1988	qsize_t cur_space;
1989	qsize_t rsv_space = 0;
1990	qsize_t inode_usage = 1;
1991	struct dquot __rcu **dquots;
1992	struct dquot *transfer_from[MAXQUOTAS] = {};
1993	int cnt, index, ret = 0, err;
1994	char is_valid[MAXQUOTAS] = {};
1995	struct dquot_warn warn_to[MAXQUOTAS];
1996	struct dquot_warn warn_from_inodes[MAXQUOTAS];
1997	struct dquot_warn warn_from_space[MAXQUOTAS];
1998
1999	if (IS_NOQUOTA(inode))
2000		return 0;
2001
2002	if (inode->i_sb->dq_op->get_inode_usage) {
2003		ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
2004		if (ret)
2005			return ret;
2006	}
2007
2008	/* Initialize the arrays */
2009	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2010		warn_to[cnt].w_type = QUOTA_NL_NOWARN;
2011		warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
2012		warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
2013	}
2014
2015	spin_lock(&dq_data_lock);
2016	spin_lock(&inode->i_lock);
2017	if (IS_NOQUOTA(inode)) {	/* File without quota accounting? */
2018		spin_unlock(&inode->i_lock);
2019		spin_unlock(&dq_data_lock);
2020		return 0;
2021	}
2022	cur_space = __inode_get_bytes(inode);
2023	rsv_space = __inode_get_rsv_space(inode);
2024	dquots = i_dquot(inode);
2025	/*
2026	 * Build the transfer_from list, check limits, and update usage in
2027	 * the target structures.
2028	 */
2029	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2030		/*
2031		 * Skip changes for same uid or gid or for turned off quota-type.
2032		 */
2033		if (!transfer_to[cnt])
2034			continue;
2035		/* Avoid races with quotaoff() */
2036		if (!sb_has_quota_active(inode->i_sb, cnt))
2037			continue;
2038		is_valid[cnt] = 1;
2039		transfer_from[cnt] = srcu_dereference_check(dquots[cnt],
2040				&dquot_srcu, lockdep_is_held(&dq_data_lock));
2041		ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2042				       &warn_to[cnt]);
2043		if (ret)
2044			goto over_quota;
2045		ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2046				      DQUOT_SPACE_WARN, &warn_to[cnt]);
2047		if (ret) {
2048			spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2049			dquot_decr_inodes(transfer_to[cnt], inode_usage);
2050			spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2051			goto over_quota;
2052		}
2053	}
2054
2055	/* Decrease usage for source structures and update quota pointers */
2056	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2057		if (!is_valid[cnt])
2058			continue;
2059		/* Due to IO error we might not have transfer_from[] structure */
2060		if (transfer_from[cnt]) {
2061			int wtype;
2062
2063			spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2064			wtype = info_idq_free(transfer_from[cnt], inode_usage);
2065			if (wtype != QUOTA_NL_NOWARN)
2066				prepare_warning(&warn_from_inodes[cnt],
2067						transfer_from[cnt], wtype);
2068			wtype = info_bdq_free(transfer_from[cnt],
2069					      cur_space + rsv_space);
2070			if (wtype != QUOTA_NL_NOWARN)
2071				prepare_warning(&warn_from_space[cnt],
2072						transfer_from[cnt], wtype);
2073			dquot_decr_inodes(transfer_from[cnt], inode_usage);
2074			dquot_decr_space(transfer_from[cnt], cur_space);
2075			dquot_free_reserved_space(transfer_from[cnt],
2076						  rsv_space);
2077			spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2078		}
2079		rcu_assign_pointer(dquots[cnt], transfer_to[cnt]);
2080	}
2081	spin_unlock(&inode->i_lock);
2082	spin_unlock(&dq_data_lock);
2083
2084	/*
2085	 * These arrays are local and we hold dquot references so we don't need
2086	 * the srcu protection but still take dquot_srcu to avoid warning in
2087	 * mark_all_dquot_dirty().
2088	 */
2089	index = srcu_read_lock(&dquot_srcu);
2090	err = mark_all_dquot_dirty((struct dquot __rcu **)transfer_from);
2091	if (err < 0)
2092		ret = err;
2093	err = mark_all_dquot_dirty((struct dquot __rcu **)transfer_to);
2094	if (err < 0)
2095		ret = err;
2096	srcu_read_unlock(&dquot_srcu, index);
2097
2098	flush_warnings(warn_to);
2099	flush_warnings(warn_from_inodes);
2100	flush_warnings(warn_from_space);
2101	/* Pass back references to put */
2102	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2103		if (is_valid[cnt])
2104			transfer_to[cnt] = transfer_from[cnt];
2105	return ret;
2106over_quota:
2107	/* Back out changes we already did */
2108	for (cnt--; cnt >= 0; cnt--) {
2109		if (!is_valid[cnt])
2110			continue;
2111		spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2112		dquot_decr_inodes(transfer_to[cnt], inode_usage);
2113		dquot_decr_space(transfer_to[cnt], cur_space);
2114		dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2115		spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2116	}
2117	spin_unlock(&inode->i_lock);
2118	spin_unlock(&dq_data_lock);
2119	flush_warnings(warn_to);
2120	return ret;
2121}
2122EXPORT_SYMBOL(__dquot_transfer);
2123
2124/* Wrapper for transferring ownership of an inode for uid/gid only
2125 * Called from FSXXX_setattr()
2126 */
2127int dquot_transfer(struct mnt_idmap *idmap, struct inode *inode,
2128		   struct iattr *iattr)
2129{
2130	struct dquot *transfer_to[MAXQUOTAS] = {};
2131	struct dquot *dquot;
2132	struct super_block *sb = inode->i_sb;
2133	int ret;
2134
2135	if (!inode_quota_active(inode))
2136		return 0;
2137
2138	if (i_uid_needs_update(idmap, iattr, inode)) {
2139		kuid_t kuid = from_vfsuid(idmap, i_user_ns(inode),
2140					  iattr->ia_vfsuid);
2141
2142		dquot = dqget(sb, make_kqid_uid(kuid));
2143		if (IS_ERR(dquot)) {
2144			if (PTR_ERR(dquot) != -ESRCH) {
2145				ret = PTR_ERR(dquot);
2146				goto out_put;
2147			}
2148			dquot = NULL;
2149		}
2150		transfer_to[USRQUOTA] = dquot;
2151	}
2152	if (i_gid_needs_update(idmap, iattr, inode)) {
2153		kgid_t kgid = from_vfsgid(idmap, i_user_ns(inode),
2154					  iattr->ia_vfsgid);
2155
2156		dquot = dqget(sb, make_kqid_gid(kgid));
2157		if (IS_ERR(dquot)) {
2158			if (PTR_ERR(dquot) != -ESRCH) {
2159				ret = PTR_ERR(dquot);
2160				goto out_put;
2161			}
2162			dquot = NULL;
2163		}
2164		transfer_to[GRPQUOTA] = dquot;
2165	}
2166	ret = __dquot_transfer(inode, transfer_to);
2167out_put:
2168	dqput_all(transfer_to);
2169	return ret;
2170}
2171EXPORT_SYMBOL(dquot_transfer);
2172
2173/*
2174 * Write info of quota file to disk
2175 */
2176int dquot_commit_info(struct super_block *sb, int type)
2177{
2178	struct quota_info *dqopt = sb_dqopt(sb);
2179
2180	return dqopt->ops[type]->write_file_info(sb, type);
2181}
2182EXPORT_SYMBOL(dquot_commit_info);
2183
2184int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2185{
2186	struct quota_info *dqopt = sb_dqopt(sb);
2187
2188	if (!sb_has_quota_active(sb, qid->type))
2189		return -ESRCH;
2190	if (!dqopt->ops[qid->type]->get_next_id)
2191		return -ENOSYS;
2192	return dqopt->ops[qid->type]->get_next_id(sb, qid);
2193}
2194EXPORT_SYMBOL(dquot_get_next_id);
2195
2196/*
2197 * Definitions of diskquota operations.
2198 */
2199const struct dquot_operations dquot_operations = {
2200	.write_dquot	= dquot_commit,
2201	.acquire_dquot	= dquot_acquire,
2202	.release_dquot	= dquot_release,
2203	.mark_dirty	= dquot_mark_dquot_dirty,
2204	.write_info	= dquot_commit_info,
2205	.alloc_dquot	= dquot_alloc,
2206	.destroy_dquot	= dquot_destroy,
2207	.get_next_id	= dquot_get_next_id,
2208};
2209EXPORT_SYMBOL(dquot_operations);
2210
2211/*
2212 * Generic helper for ->open on filesystems supporting disk quotas.
2213 */
2214int dquot_file_open(struct inode *inode, struct file *file)
2215{
2216	int error;
2217
2218	error = generic_file_open(inode, file);
2219	if (!error && (file->f_mode & FMODE_WRITE))
2220		error = dquot_initialize(inode);
2221	return error;
2222}
2223EXPORT_SYMBOL(dquot_file_open);
2224
2225static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2226{
2227	struct quota_info *dqopt = sb_dqopt(sb);
2228	struct inode *inode = dqopt->files[type];
2229
2230	if (!inode)
2231		return;
2232	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2233		inode_lock(inode);
2234		inode->i_flags &= ~S_NOQUOTA;
2235		inode_unlock(inode);
2236	}
2237	dqopt->files[type] = NULL;
2238	iput(inode);
2239}
2240
2241/*
2242 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2243 */
2244int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2245{
2246	int cnt;
2247	struct quota_info *dqopt = sb_dqopt(sb);
2248
2249	/* s_umount should be held in exclusive mode */
2250	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2251		up_read(&sb->s_umount);
2252
2253	/* Cannot turn off usage accounting without turning off limits, or
2254	 * suspend quotas and simultaneously turn quotas off. */
2255	if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2256	    || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2257	    DQUOT_USAGE_ENABLED)))
2258		return -EINVAL;
2259
2260	/*
2261	 * Skip everything if there's nothing to do. We have to do this because
2262	 * sometimes we are called when fill_super() failed and calling
2263	 * sync_fs() in such cases does no good.
2264	 */
2265	if (!sb_any_quota_loaded(sb))
2266		return 0;
2267
2268	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2269		if (type != -1 && cnt != type)
2270			continue;
2271		if (!sb_has_quota_loaded(sb, cnt))
2272			continue;
2273
2274		if (flags & DQUOT_SUSPENDED) {
2275			spin_lock(&dq_state_lock);
2276			dqopt->flags |=
2277				dquot_state_flag(DQUOT_SUSPENDED, cnt);
2278			spin_unlock(&dq_state_lock);
2279		} else {
2280			spin_lock(&dq_state_lock);
2281			dqopt->flags &= ~dquot_state_flag(flags, cnt);
2282			/* Turning off suspended quotas? */
2283			if (!sb_has_quota_loaded(sb, cnt) &&
2284			    sb_has_quota_suspended(sb, cnt)) {
2285				dqopt->flags &=	~dquot_state_flag(
2286							DQUOT_SUSPENDED, cnt);
2287				spin_unlock(&dq_state_lock);
2288				vfs_cleanup_quota_inode(sb, cnt);
2289				continue;
2290			}
2291			spin_unlock(&dq_state_lock);
2292		}
2293
2294		/* We still have to keep quota loaded? */
2295		if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2296			continue;
2297
2298		/* Note: these are blocking operations */
2299		drop_dquot_ref(sb, cnt);
2300		invalidate_dquots(sb, cnt);
2301		/*
2302		 * Now all dquots should be invalidated, all writes done so we
2303		 * should be only users of the info. No locks needed.
2304		 */
2305		if (info_dirty(&dqopt->info[cnt]))
2306			sb->dq_op->write_info(sb, cnt);
2307		if (dqopt->ops[cnt]->free_file_info)
2308			dqopt->ops[cnt]->free_file_info(sb, cnt);
2309		put_quota_format(dqopt->info[cnt].dqi_format);
2310		dqopt->info[cnt].dqi_flags = 0;
2311		dqopt->info[cnt].dqi_igrace = 0;
2312		dqopt->info[cnt].dqi_bgrace = 0;
2313		dqopt->ops[cnt] = NULL;
2314	}
2315
2316	/* Skip syncing and setting flags if quota files are hidden */
2317	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2318		goto put_inodes;
2319
2320	/* Sync the superblock so that buffers with quota data are written to
2321	 * disk (and so userspace sees correct data afterwards). */
2322	if (sb->s_op->sync_fs)
2323		sb->s_op->sync_fs(sb, 1);
2324	sync_blockdev(sb->s_bdev);
2325	/* Now the quota files are just ordinary files and we can set the
2326	 * inode flags back. Moreover we discard the pagecache so that
2327	 * userspace sees the writes we did bypassing the pagecache. We
2328	 * must also discard the blockdev buffers so that we see the
2329	 * changes done by userspace on the next quotaon() */
2330	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2331		if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2332			inode_lock(dqopt->files[cnt]);
2333			truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2334			inode_unlock(dqopt->files[cnt]);
2335		}
2336	if (sb->s_bdev)
2337		invalidate_bdev(sb->s_bdev);
2338put_inodes:
2339	/* We are done when suspending quotas */
2340	if (flags & DQUOT_SUSPENDED)
2341		return 0;
2342
2343	for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2344		if (!sb_has_quota_loaded(sb, cnt))
2345			vfs_cleanup_quota_inode(sb, cnt);
2346	return 0;
2347}
2348EXPORT_SYMBOL(dquot_disable);
2349
2350int dquot_quota_off(struct super_block *sb, int type)
2351{
2352	return dquot_disable(sb, type,
2353			     DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2354}
2355EXPORT_SYMBOL(dquot_quota_off);
2356
2357/*
2358 *	Turn quotas on on a device
2359 */
2360
2361static int vfs_setup_quota_inode(struct inode *inode, int type)
2362{
2363	struct super_block *sb = inode->i_sb;
2364	struct quota_info *dqopt = sb_dqopt(sb);
2365
2366	if (is_bad_inode(inode))
2367		return -EUCLEAN;
2368	if (!S_ISREG(inode->i_mode))
2369		return -EACCES;
2370	if (IS_RDONLY(inode))
2371		return -EROFS;
2372	if (sb_has_quota_loaded(sb, type))
2373		return -EBUSY;
2374
2375	/*
2376	 * Quota files should never be encrypted.  They should be thought of as
2377	 * filesystem metadata, not user data.  New-style internal quota files
2378	 * cannot be encrypted by users anyway, but old-style external quota
2379	 * files could potentially be incorrectly created in an encrypted
2380	 * directory, hence this explicit check.  Some reasons why encrypted
2381	 * quota files don't work include: (1) some filesystems that support
2382	 * encryption don't handle it in their quota_read and quota_write, and
2383	 * (2) cleaning up encrypted quota files at unmount would need special
2384	 * consideration, as quota files are cleaned up later than user files.
2385	 */
2386	if (IS_ENCRYPTED(inode))
2387		return -EINVAL;
2388
2389	dqopt->files[type] = igrab(inode);
2390	if (!dqopt->files[type])
2391		return -EIO;
2392	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2393		/* We don't want quota and atime on quota files (deadlocks
2394		 * possible) Also nobody should write to the file - we use
2395		 * special IO operations which ignore the immutable bit. */
2396		inode_lock(inode);
2397		inode->i_flags |= S_NOQUOTA;
2398		inode_unlock(inode);
2399		/*
2400		 * When S_NOQUOTA is set, remove dquot references as no more
2401		 * references can be added
2402		 */
2403		__dquot_drop(inode);
2404	}
2405	return 0;
2406}
2407
2408int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2409	unsigned int flags)
2410{
2411	struct quota_format_type *fmt = find_quota_format(format_id);
2412	struct quota_info *dqopt = sb_dqopt(sb);
2413	int error;
2414
2415	lockdep_assert_held_write(&sb->s_umount);
2416
2417	/* Just unsuspend quotas? */
2418	if (WARN_ON_ONCE(flags & DQUOT_SUSPENDED))
2419		return -EINVAL;
2420
2421	if (!fmt)
2422		return -ESRCH;
2423	if (!sb->dq_op || !sb->s_qcop ||
2424	    (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2425		error = -EINVAL;
2426		goto out_fmt;
2427	}
2428	/* Filesystems outside of init_user_ns not yet supported */
2429	if (sb->s_user_ns != &init_user_ns) {
2430		error = -EINVAL;
2431		goto out_fmt;
2432	}
2433	/* Usage always has to be set... */
2434	if (!(flags & DQUOT_USAGE_ENABLED)) {
2435		error = -EINVAL;
2436		goto out_fmt;
2437	}
2438	if (sb_has_quota_loaded(sb, type)) {
2439		error = -EBUSY;
2440		goto out_fmt;
2441	}
2442
2443	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2444		/* As we bypass the pagecache we must now flush all the
2445		 * dirty data and invalidate caches so that kernel sees
2446		 * changes from userspace. It is not enough to just flush
2447		 * the quota file since if blocksize < pagesize, invalidation
2448		 * of the cache could fail because of other unrelated dirty
2449		 * data */
2450		sync_filesystem(sb);
2451		invalidate_bdev(sb->s_bdev);
2452	}
2453
2454	error = -EINVAL;
2455	if (!fmt->qf_ops->check_quota_file(sb, type))
2456		goto out_fmt;
2457
2458	dqopt->ops[type] = fmt->qf_ops;
2459	dqopt->info[type].dqi_format = fmt;
2460	dqopt->info[type].dqi_fmt_id = format_id;
2461	INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2462	error = dqopt->ops[type]->read_file_info(sb, type);
2463	if (error < 0)
2464		goto out_fmt;
2465	if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2466		spin_lock(&dq_data_lock);
2467		dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2468		spin_unlock(&dq_data_lock);
2469	}
2470	spin_lock(&dq_state_lock);
2471	dqopt->flags |= dquot_state_flag(flags, type);
2472	spin_unlock(&dq_state_lock);
2473
2474	error = add_dquot_ref(sb, type);
2475	if (error)
2476		dquot_disable(sb, type,
2477			      DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2478
2479	return error;
2480out_fmt:
2481	put_quota_format(fmt);
2482
2483	return error;
2484}
2485EXPORT_SYMBOL(dquot_load_quota_sb);
2486
2487/*
2488 * More powerful function for turning on quotas on given quota inode allowing
2489 * setting of individual quota flags
2490 */
2491int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2492	unsigned int flags)
2493{
2494	int err;
2495
2496	err = vfs_setup_quota_inode(inode, type);
2497	if (err < 0)
2498		return err;
2499	err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2500	if (err < 0)
2501		vfs_cleanup_quota_inode(inode->i_sb, type);
2502	return err;
2503}
2504EXPORT_SYMBOL(dquot_load_quota_inode);
2505
2506/* Reenable quotas on remount RW */
2507int dquot_resume(struct super_block *sb, int type)
2508{
2509	struct quota_info *dqopt = sb_dqopt(sb);
2510	int ret = 0, cnt;
2511	unsigned int flags;
2512
2513	/* s_umount should be held in exclusive mode */
2514	if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2515		up_read(&sb->s_umount);
2516
2517	for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2518		if (type != -1 && cnt != type)
2519			continue;
2520		if (!sb_has_quota_suspended(sb, cnt))
2521			continue;
2522
2523		spin_lock(&dq_state_lock);
2524		flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2525							DQUOT_LIMITS_ENABLED,
2526							cnt);
2527		dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2528		spin_unlock(&dq_state_lock);
2529
2530		flags = dquot_generic_flag(flags, cnt);
2531		ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2532					  flags);
2533		if (ret < 0)
2534			vfs_cleanup_quota_inode(sb, cnt);
2535	}
2536
2537	return ret;
2538}
2539EXPORT_SYMBOL(dquot_resume);
2540
2541int dquot_quota_on(struct super_block *sb, int type, int format_id,
2542		   const struct path *path)
2543{
2544	int error = security_quota_on(path->dentry);
2545	if (error)
2546		return error;
2547	/* Quota file not on the same filesystem? */
2548	if (path->dentry->d_sb != sb)
2549		error = -EXDEV;
2550	else
2551		error = dquot_load_quota_inode(d_inode(path->dentry), type,
2552					     format_id, DQUOT_USAGE_ENABLED |
2553					     DQUOT_LIMITS_ENABLED);
2554	return error;
2555}
2556EXPORT_SYMBOL(dquot_quota_on);
2557
2558/*
2559 * This function is used when filesystem needs to initialize quotas
2560 * during mount time.
2561 */
2562int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2563		int format_id, int type)
2564{
2565	struct dentry *dentry;
2566	int error;
2567
2568	dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2569	if (IS_ERR(dentry))
2570		return PTR_ERR(dentry);
2571
2572	error = security_quota_on(dentry);
2573	if (!error)
2574		error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2575				DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2576
2577	dput(dentry);
2578	return error;
2579}
2580EXPORT_SYMBOL(dquot_quota_on_mount);
2581
2582static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2583{
2584	int ret;
2585	int type;
2586	struct quota_info *dqopt = sb_dqopt(sb);
2587
2588	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2589		return -ENOSYS;
2590	/* Accounting cannot be turned on while fs is mounted */
2591	flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2592	if (!flags)
2593		return -EINVAL;
2594	for (type = 0; type < MAXQUOTAS; type++) {
2595		if (!(flags & qtype_enforce_flag(type)))
2596			continue;
2597		/* Can't enforce without accounting */
2598		if (!sb_has_quota_usage_enabled(sb, type)) {
2599			ret = -EINVAL;
2600			goto out_err;
2601		}
2602		if (sb_has_quota_limits_enabled(sb, type)) {
2603			ret = -EBUSY;
2604			goto out_err;
2605		}
2606		spin_lock(&dq_state_lock);
2607		dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2608		spin_unlock(&dq_state_lock);
2609	}
2610	return 0;
2611out_err:
2612	/* Backout enforcement enablement we already did */
2613	for (type--; type >= 0; type--)  {
2614		if (flags & qtype_enforce_flag(type))
2615			dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2616	}
2617	/* Error code translation for better compatibility with XFS */
2618	if (ret == -EBUSY)
2619		ret = -EEXIST;
2620	return ret;
2621}
2622
2623static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2624{
2625	int ret;
2626	int type;
2627	struct quota_info *dqopt = sb_dqopt(sb);
2628
2629	if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2630		return -ENOSYS;
2631	/*
2632	 * We don't support turning off accounting via quotactl. In principle
2633	 * quota infrastructure can do this but filesystems don't expect
2634	 * userspace to be able to do it.
2635	 */
2636	if (flags &
2637		  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2638		return -EOPNOTSUPP;
2639
2640	/* Filter out limits not enabled */
2641	for (type = 0; type < MAXQUOTAS; type++)
2642		if (!sb_has_quota_limits_enabled(sb, type))
2643			flags &= ~qtype_enforce_flag(type);
2644	/* Nothing left? */
2645	if (!flags)
2646		return -EEXIST;
2647	for (type = 0; type < MAXQUOTAS; type++) {
2648		if (flags & qtype_enforce_flag(type)) {
2649			ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2650			if (ret < 0)
2651				goto out_err;
2652		}
2653	}
2654	return 0;
2655out_err:
2656	/* Backout enforcement disabling we already did */
2657	for (type--; type >= 0; type--)  {
2658		if (flags & qtype_enforce_flag(type)) {
2659			spin_lock(&dq_state_lock);
2660			dqopt->flags |=
2661				dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2662			spin_unlock(&dq_state_lock);
2663		}
2664	}
2665	return ret;
2666}
2667
2668/* Generic routine for getting common part of quota structure */
2669static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2670{
2671	struct mem_dqblk *dm = &dquot->dq_dqb;
2672
2673	memset(di, 0, sizeof(*di));
2674	spin_lock(&dquot->dq_dqb_lock);
2675	di->d_spc_hardlimit = dm->dqb_bhardlimit;
2676	di->d_spc_softlimit = dm->dqb_bsoftlimit;
2677	di->d_ino_hardlimit = dm->dqb_ihardlimit;
2678	di->d_ino_softlimit = dm->dqb_isoftlimit;
2679	di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2680	di->d_ino_count = dm->dqb_curinodes;
2681	di->d_spc_timer = dm->dqb_btime;
2682	di->d_ino_timer = dm->dqb_itime;
2683	spin_unlock(&dquot->dq_dqb_lock);
2684}
2685
2686int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2687		    struct qc_dqblk *di)
2688{
2689	struct dquot *dquot;
2690
2691	dquot = dqget(sb, qid);
2692	if (IS_ERR(dquot))
2693		return PTR_ERR(dquot);
2694	do_get_dqblk(dquot, di);
2695	dqput(dquot);
2696
2697	return 0;
2698}
2699EXPORT_SYMBOL(dquot_get_dqblk);
2700
2701int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2702			 struct qc_dqblk *di)
2703{
2704	struct dquot *dquot;
2705	int err;
2706
2707	if (!sb->dq_op->get_next_id)
2708		return -ENOSYS;
2709	err = sb->dq_op->get_next_id(sb, qid);
2710	if (err < 0)
2711		return err;
2712	dquot = dqget(sb, *qid);
2713	if (IS_ERR(dquot))
2714		return PTR_ERR(dquot);
2715	do_get_dqblk(dquot, di);
2716	dqput(dquot);
2717
2718	return 0;
2719}
2720EXPORT_SYMBOL(dquot_get_next_dqblk);
2721
2722#define VFS_QC_MASK \
2723	(QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2724	 QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2725	 QC_SPC_TIMER | QC_INO_TIMER)
2726
2727/* Generic routine for setting common part of quota structure */
2728static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2729{
2730	struct mem_dqblk *dm = &dquot->dq_dqb;
2731	int check_blim = 0, check_ilim = 0;
2732	struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2733	int ret;
2734
2735	if (di->d_fieldmask & ~VFS_QC_MASK)
2736		return -EINVAL;
2737
2738	if (((di->d_fieldmask & QC_SPC_SOFT) &&
2739	     di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2740	    ((di->d_fieldmask & QC_SPC_HARD) &&
2741	     di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2742	    ((di->d_fieldmask & QC_INO_SOFT) &&
2743	     (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2744	    ((di->d_fieldmask & QC_INO_HARD) &&
2745	     (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2746		return -ERANGE;
2747
2748	spin_lock(&dquot->dq_dqb_lock);
2749	if (di->d_fieldmask & QC_SPACE) {
2750		dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2751		check_blim = 1;
2752		set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2753	}
2754
2755	if (di->d_fieldmask & QC_SPC_SOFT)
2756		dm->dqb_bsoftlimit = di->d_spc_softlimit;
2757	if (di->d_fieldmask & QC_SPC_HARD)
2758		dm->dqb_bhardlimit = di->d_spc_hardlimit;
2759	if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2760		check_blim = 1;
2761		set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2762	}
2763
2764	if (di->d_fieldmask & QC_INO_COUNT) {
2765		dm->dqb_curinodes = di->d_ino_count;
2766		check_ilim = 1;
2767		set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2768	}
2769
2770	if (di->d_fieldmask & QC_INO_SOFT)
2771		dm->dqb_isoftlimit = di->d_ino_softlimit;
2772	if (di->d_fieldmask & QC_INO_HARD)
2773		dm->dqb_ihardlimit = di->d_ino_hardlimit;
2774	if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2775		check_ilim = 1;
2776		set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2777	}
2778
2779	if (di->d_fieldmask & QC_SPC_TIMER) {
2780		dm->dqb_btime = di->d_spc_timer;
2781		check_blim = 1;
2782		set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2783	}
2784
2785	if (di->d_fieldmask & QC_INO_TIMER) {
2786		dm->dqb_itime = di->d_ino_timer;
2787		check_ilim = 1;
2788		set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2789	}
2790
2791	if (check_blim) {
2792		if (!dm->dqb_bsoftlimit ||
2793		    dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2794			dm->dqb_btime = 0;
2795			clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2796		} else if (!(di->d_fieldmask & QC_SPC_TIMER))
2797			/* Set grace only if user hasn't provided his own... */
2798			dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2799	}
2800	if (check_ilim) {
2801		if (!dm->dqb_isoftlimit ||
2802		    dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2803			dm->dqb_itime = 0;
2804			clear_bit(DQ_INODES_B, &dquot->dq_flags);
2805		} else if (!(di->d_fieldmask & QC_INO_TIMER))
2806			/* Set grace only if user hasn't provided his own... */
2807			dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2808	}
2809	if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2810	    dm->dqb_isoftlimit)
2811		clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2812	else
2813		set_bit(DQ_FAKE_B, &dquot->dq_flags);
2814	spin_unlock(&dquot->dq_dqb_lock);
2815	ret = mark_dquot_dirty(dquot);
2816	if (ret < 0)
2817		return ret;
2818	return 0;
2819}
2820
2821int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2822		  struct qc_dqblk *di)
2823{
2824	struct dquot *dquot;
2825	int rc;
2826
2827	dquot = dqget(sb, qid);
2828	if (IS_ERR(dquot)) {
2829		rc = PTR_ERR(dquot);
2830		goto out;
2831	}
2832	rc = do_set_dqblk(dquot, di);
2833	dqput(dquot);
2834out:
2835	return rc;
2836}
2837EXPORT_SYMBOL(dquot_set_dqblk);
2838
2839/* Generic routine for getting common part of quota file information */
2840int dquot_get_state(struct super_block *sb, struct qc_state *state)
2841{
2842	struct mem_dqinfo *mi;
2843	struct qc_type_state *tstate;
2844	struct quota_info *dqopt = sb_dqopt(sb);
2845	int type;
2846
2847	memset(state, 0, sizeof(*state));
2848	for (type = 0; type < MAXQUOTAS; type++) {
2849		if (!sb_has_quota_active(sb, type))
2850			continue;
2851		tstate = state->s_state + type;
2852		mi = sb_dqopt(sb)->info + type;
2853		tstate->flags = QCI_ACCT_ENABLED;
2854		spin_lock(&dq_data_lock);
2855		if (mi->dqi_flags & DQF_SYS_FILE)
2856			tstate->flags |= QCI_SYSFILE;
2857		if (mi->dqi_flags & DQF_ROOT_SQUASH)
2858			tstate->flags |= QCI_ROOT_SQUASH;
2859		if (sb_has_quota_limits_enabled(sb, type))
2860			tstate->flags |= QCI_LIMITS_ENFORCED;
2861		tstate->spc_timelimit = mi->dqi_bgrace;
2862		tstate->ino_timelimit = mi->dqi_igrace;
2863		if (dqopt->files[type]) {
2864			tstate->ino = dqopt->files[type]->i_ino;
2865			tstate->blocks = dqopt->files[type]->i_blocks;
2866		}
2867		tstate->nextents = 1;	/* We don't know... */
2868		spin_unlock(&dq_data_lock);
2869	}
2870	return 0;
2871}
2872EXPORT_SYMBOL(dquot_get_state);
2873
2874/* Generic routine for setting common part of quota file information */
2875int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2876{
2877	struct mem_dqinfo *mi;
2878
2879	if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2880	    (ii->i_fieldmask & QC_RT_SPC_TIMER))
2881		return -EINVAL;
2882	if (!sb_has_quota_active(sb, type))
2883		return -ESRCH;
2884	mi = sb_dqopt(sb)->info + type;
2885	if (ii->i_fieldmask & QC_FLAGS) {
2886		if ((ii->i_flags & QCI_ROOT_SQUASH &&
2887		     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2888			return -EINVAL;
2889	}
2890	spin_lock(&dq_data_lock);
2891	if (ii->i_fieldmask & QC_SPC_TIMER)
2892		mi->dqi_bgrace = ii->i_spc_timelimit;
2893	if (ii->i_fieldmask & QC_INO_TIMER)
2894		mi->dqi_igrace = ii->i_ino_timelimit;
2895	if (ii->i_fieldmask & QC_FLAGS) {
2896		if (ii->i_flags & QCI_ROOT_SQUASH)
2897			mi->dqi_flags |= DQF_ROOT_SQUASH;
2898		else
2899			mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2900	}
2901	spin_unlock(&dq_data_lock);
2902	mark_info_dirty(sb, type);
2903	/* Force write to disk */
2904	return sb->dq_op->write_info(sb, type);
2905}
2906EXPORT_SYMBOL(dquot_set_dqinfo);
2907
2908const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2909	.quota_enable	= dquot_quota_enable,
2910	.quota_disable	= dquot_quota_disable,
2911	.quota_sync	= dquot_quota_sync,
2912	.get_state	= dquot_get_state,
2913	.set_info	= dquot_set_dqinfo,
2914	.get_dqblk	= dquot_get_dqblk,
2915	.get_nextdqblk	= dquot_get_next_dqblk,
2916	.set_dqblk	= dquot_set_dqblk
2917};
2918EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2919
2920static int do_proc_dqstats(struct ctl_table *table, int write,
2921		     void *buffer, size_t *lenp, loff_t *ppos)
2922{
2923	unsigned int type = (unsigned long *)table->data - dqstats.stat;
2924	s64 value = percpu_counter_sum(&dqstats.counter[type]);
2925
2926	/* Filter negative values for non-monotonic counters */
2927	if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2928			  type == DQST_FREE_DQUOTS))
2929		value = 0;
2930
2931	/* Update global table */
2932	dqstats.stat[type] = value;
2933	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2934}
2935
2936static struct ctl_table fs_dqstats_table[] = {
2937	{
2938		.procname	= "lookups",
2939		.data		= &dqstats.stat[DQST_LOOKUPS],
2940		.maxlen		= sizeof(unsigned long),
2941		.mode		= 0444,
2942		.proc_handler	= do_proc_dqstats,
2943	},
2944	{
2945		.procname	= "drops",
2946		.data		= &dqstats.stat[DQST_DROPS],
2947		.maxlen		= sizeof(unsigned long),
2948		.mode		= 0444,
2949		.proc_handler	= do_proc_dqstats,
2950	},
2951	{
2952		.procname	= "reads",
2953		.data		= &dqstats.stat[DQST_READS],
2954		.maxlen		= sizeof(unsigned long),
2955		.mode		= 0444,
2956		.proc_handler	= do_proc_dqstats,
2957	},
2958	{
2959		.procname	= "writes",
2960		.data		= &dqstats.stat[DQST_WRITES],
2961		.maxlen		= sizeof(unsigned long),
2962		.mode		= 0444,
2963		.proc_handler	= do_proc_dqstats,
2964	},
2965	{
2966		.procname	= "cache_hits",
2967		.data		= &dqstats.stat[DQST_CACHE_HITS],
2968		.maxlen		= sizeof(unsigned long),
2969		.mode		= 0444,
2970		.proc_handler	= do_proc_dqstats,
2971	},
2972	{
2973		.procname	= "allocated_dquots",
2974		.data		= &dqstats.stat[DQST_ALLOC_DQUOTS],
2975		.maxlen		= sizeof(unsigned long),
2976		.mode		= 0444,
2977		.proc_handler	= do_proc_dqstats,
2978	},
2979	{
2980		.procname	= "free_dquots",
2981		.data		= &dqstats.stat[DQST_FREE_DQUOTS],
2982		.maxlen		= sizeof(unsigned long),
2983		.mode		= 0444,
2984		.proc_handler	= do_proc_dqstats,
2985	},
2986	{
2987		.procname	= "syncs",
2988		.data		= &dqstats.stat[DQST_SYNCS],
2989		.maxlen		= sizeof(unsigned long),
2990		.mode		= 0444,
2991		.proc_handler	= do_proc_dqstats,
2992	},
2993#ifdef CONFIG_PRINT_QUOTA_WARNING
2994	{
2995		.procname	= "warnings",
2996		.data		= &flag_print_warnings,
2997		.maxlen		= sizeof(int),
2998		.mode		= 0644,
2999		.proc_handler	= proc_dointvec,
3000	},
3001#endif
3002};
3003
3004static int __init dquot_init(void)
3005{
3006	int i, ret;
3007	unsigned long nr_hash, order;
3008	struct shrinker *dqcache_shrinker;
3009
3010	printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
3011
3012	register_sysctl_init("fs/quota", fs_dqstats_table);
3013
3014	dquot_cachep = kmem_cache_create("dquot",
3015			sizeof(struct dquot), sizeof(unsigned long) * 4,
3016			(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
3017				SLAB_PANIC),
3018			NULL);
3019
3020	order = 0;
3021	dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
3022	if (!dquot_hash)
3023		panic("Cannot create dquot hash table");
3024
3025	ret = percpu_counter_init_many(dqstats.counter, 0, GFP_KERNEL,
3026				       _DQST_DQSTAT_LAST);
3027	if (ret)
3028		panic("Cannot create dquot stat counters");
3029
3030	/* Find power-of-two hlist_heads which can fit into allocation */
3031	nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
3032	dq_hash_bits = ilog2(nr_hash);
3033
3034	nr_hash = 1UL << dq_hash_bits;
3035	dq_hash_mask = nr_hash - 1;
3036	for (i = 0; i < nr_hash; i++)
3037		INIT_HLIST_HEAD(dquot_hash + i);
3038
3039	pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
3040		" %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
3041
3042	dqcache_shrinker = shrinker_alloc(0, "dquota-cache");
3043	if (!dqcache_shrinker)
3044		panic("Cannot allocate dquot shrinker");
3045
3046	dqcache_shrinker->count_objects = dqcache_shrink_count;
3047	dqcache_shrinker->scan_objects = dqcache_shrink_scan;
3048
3049	shrinker_register(dqcache_shrinker);
3050
3051	return 0;
3052}
3053fs_initcall(dquot_init);
3054