xfs_qm.c 59.1 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2 3
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
L
Linus Torvalds 已提交
4
 *
5 6
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
L
Linus Torvalds 已提交
7 8
 * published by the Free Software Foundation.
 *
9 10 11 12
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
L
Linus Torvalds 已提交
13
 *
14 15 16
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
L
Linus Torvalds 已提交
17 18 19
 */
#include "xfs.h"
#include "xfs_fs.h"
20
#include "xfs_bit.h"
L
Linus Torvalds 已提交
21
#include "xfs_log.h"
22
#include "xfs_inum.h"
L
Linus Torvalds 已提交
23 24 25 26 27 28 29 30 31 32
#include "xfs_trans.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_alloc.h"
#include "xfs_quota.h"
#include "xfs_mount.h"
#include "xfs_bmap_btree.h"
#include "xfs_ialloc_btree.h"
#include "xfs_dinode.h"
#include "xfs_inode.h"
33 34
#include "xfs_ialloc.h"
#include "xfs_itable.h"
L
Linus Torvalds 已提交
35 36
#include "xfs_rtalloc.h"
#include "xfs_error.h"
37
#include "xfs_bmap.h"
L
Linus Torvalds 已提交
38 39 40 41 42
#include "xfs_attr.h"
#include "xfs_buf_item.h"
#include "xfs_trans_space.h"
#include "xfs_utils.h"
#include "xfs_qm.h"
C
Christoph Hellwig 已提交
43
#include "xfs_trace.h"
L
Linus Torvalds 已提交
44 45 46 47 48 49 50

/*
 * The global quota manager. There is only one of these for the entire
 * system, _not_ one per file system. XQM keeps track of the overall
 * quota functionality, including maintaining the freelist and hash
 * tables of dquots.
 */
C
Christoph Hellwig 已提交
51
struct mutex	xfs_Gqm_lock;
L
Linus Torvalds 已提交
52
struct xfs_qm	*xfs_Gqm;
53
uint		ndquot;
L
Linus Torvalds 已提交
54 55 56 57 58 59 60 61

kmem_zone_t	*qm_dqzone;
kmem_zone_t	*qm_dqtrxzone;

STATIC void	xfs_qm_list_init(xfs_dqlist_t *, char *, int);
STATIC void	xfs_qm_list_destroy(xfs_dqlist_t *);

STATIC int	xfs_qm_init_quotainos(xfs_mount_t *);
62
STATIC int	xfs_qm_init_quotainfo(xfs_mount_t *);
63
STATIC int	xfs_qm_shake(struct shrinker *, int, gfp_t);
L
Linus Torvalds 已提交
64

65 66 67 68 69
static struct shrinker xfs_qm_shaker = {
	.shrink = xfs_qm_shake,
	.seeks = DEFAULT_SEEKS,
};

L
Linus Torvalds 已提交
70
#ifdef DEBUG
C
Christoph Hellwig 已提交
71
extern struct mutex	qcheck_lock;
L
Linus Torvalds 已提交
72 73 74
#endif

#ifdef QUOTADEBUG
75 76 77 78 79 80 81 82
static void
xfs_qm_dquot_list_print(
	struct xfs_mount *mp)
{
	xfs_dquot_t	*dqp;
	int		i = 0;

	list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist_lock, qi_mplist) {
83
		xfs_debug(mp, "   %d. \"%d (%s)\"   "
84 85 86 87 88 89 90
				  "bcnt = %lld, icnt = %lld, refs = %d",
			i++, be32_to_cpu(dqp->q_core.d_id),
			DQFLAGTO_TYPESTR(dqp),
			(long long)be64_to_cpu(dqp->q_core.d_bcount),
			(long long)be64_to_cpu(dqp->q_core.d_icount),
			dqp->q_nrefs);
	}
L
Linus Torvalds 已提交
91 92
}
#else
93
static void xfs_qm_dquot_list_print(struct xfs_mount *mp) { }
L
Linus Torvalds 已提交
94 95 96 97 98 99 100 101 102
#endif

/*
 * Initialize the XQM structure.
 * Note that there is not one quota manager per file system.
 */
STATIC struct xfs_qm *
xfs_Gqm_init(void)
{
103 104
	xfs_dqhash_t	*udqhash, *gdqhash;
	xfs_qm_t	*xqm;
105 106
	size_t		hsize;
	uint		i;
L
Linus Torvalds 已提交
107 108 109 110

	/*
	 * Initialize the dquot hash tables.
	 */
111
	udqhash = kmem_zalloc_greedy(&hsize,
112
				     XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
113 114 115 116 117
				     XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
	if (!udqhash)
		goto out;

	gdqhash = kmem_zalloc_large(hsize);
J
Julia Lawall 已提交
118
	if (!gdqhash)
119 120
		goto out_free_udqhash;

121
	hsize /= sizeof(xfs_dqhash_t);
122
	ndquot = hsize << 8;
L
Linus Torvalds 已提交
123

124 125 126 127
	xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
	xqm->qm_dqhashmask = hsize - 1;
	xqm->qm_usr_dqhtable = udqhash;
	xqm->qm_grp_dqhtable = gdqhash;
L
Linus Torvalds 已提交
128 129 130 131 132 133 134 135 136 137 138
	ASSERT(xqm->qm_usr_dqhtable != NULL);
	ASSERT(xqm->qm_grp_dqhtable != NULL);

	for (i = 0; i < hsize; i++) {
		xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
		xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
	}

	/*
	 * Freelist of all dquots of all file systems
	 */
139 140 141
	INIT_LIST_HEAD(&xqm->qm_dqfrlist);
	xqm->qm_dqfrlist_cnt = 0;
	mutex_init(&xqm->qm_dqfrlist_lock);
L
Linus Torvalds 已提交
142 143 144 145 146 147 148 149 150 151 152

	/*
	 * dquot zone. we register our own low-memory callback.
	 */
	if (!qm_dqzone) {
		xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
						"xfs_dquots");
		qm_dqzone = xqm->qm_dqzone;
	} else
		xqm->qm_dqzone = qm_dqzone;

153
	register_shrinker(&xfs_qm_shaker);
L
Linus Torvalds 已提交
154 155 156 157 158 159 160 161 162 163 164 165 166 167 168

	/*
	 * The t_dqinfo portion of transactions.
	 */
	if (!qm_dqtrxzone) {
		xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
						   "xfs_dqtrx");
		qm_dqtrxzone = xqm->qm_dqtrxzone;
	} else
		xqm->qm_dqtrxzone = qm_dqtrxzone;

	atomic_set(&xqm->qm_totaldquots, 0);
	xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
	xqm->qm_nrefs = 0;
#ifdef DEBUG
169
	mutex_init(&qcheck_lock);
L
Linus Torvalds 已提交
170 171
#endif
	return xqm;
172 173 174 175 176

 out_free_udqhash:
	kmem_free_large(udqhash);
 out:
	return NULL;
L
Linus Torvalds 已提交
177 178 179 180 181
}

/*
 * Destroy the global quota manager when its reference count goes to zero.
 */
182
STATIC void
L
Linus Torvalds 已提交
183 184 185
xfs_qm_destroy(
	struct xfs_qm	*xqm)
{
186
	struct xfs_dquot *dqp, *n;
L
Linus Torvalds 已提交
187 188 189 190
	int		hsize, i;

	ASSERT(xqm != NULL);
	ASSERT(xqm->qm_nrefs == 0);
191
	unregister_shrinker(&xfs_qm_shaker);
L
Linus Torvalds 已提交
192 193 194 195 196
	hsize = xqm->qm_dqhashmask + 1;
	for (i = 0; i < hsize; i++) {
		xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
		xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
	}
197 198
	kmem_free_large(xqm->qm_usr_dqhtable);
	kmem_free_large(xqm->qm_grp_dqhtable);
L
Linus Torvalds 已提交
199 200 201
	xqm->qm_usr_dqhtable = NULL;
	xqm->qm_grp_dqhtable = NULL;
	xqm->qm_dqhashmask = 0;
202 203 204 205 206 207

	/* frlist cleanup */
	mutex_lock(&xqm->qm_dqfrlist_lock);
	list_for_each_entry_safe(dqp, n, &xqm->qm_dqfrlist, q_freelist) {
		xfs_dqlock(dqp);
#ifdef QUOTADEBUG
208
		xfs_debug(dqp->q_mount, "FREELIST destroy 0x%p", dqp);
209 210 211 212 213 214 215 216
#endif
		list_del_init(&dqp->q_freelist);
		xfs_Gqm->qm_dqfrlist_cnt--;
		xfs_dqunlock(dqp);
		xfs_qm_dqdestroy(dqp);
	}
	mutex_unlock(&xqm->qm_dqfrlist_lock);
	mutex_destroy(&xqm->qm_dqfrlist_lock);
L
Linus Torvalds 已提交
217 218 219
#ifdef DEBUG
	mutex_destroy(&qcheck_lock);
#endif
220
	kmem_free(xqm);
L
Linus Torvalds 已提交
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238
}

/*
 * Called at mount time to let XQM know that another file system is
 * starting quotas. This isn't crucial information as the individual mount
 * structures are pretty independent, but it helps the XQM keep a
 * global view of what's going on.
 */
/* ARGSUSED */
STATIC int
xfs_qm_hold_quotafs_ref(
	struct xfs_mount *mp)
{
	/*
	 * Need to lock the xfs_Gqm structure for things like this. For example,
	 * the structure could disappear between the entry to this routine and
	 * a HOLD operation if not locked.
	 */
239
	mutex_lock(&xfs_Gqm_lock);
L
Linus Torvalds 已提交
240

241
	if (!xfs_Gqm) {
L
Linus Torvalds 已提交
242
		xfs_Gqm = xfs_Gqm_init();
243 244
		if (!xfs_Gqm) {
			mutex_unlock(&xfs_Gqm_lock);
245
			return ENOMEM;
246
		}
247 248
	}

L
Linus Torvalds 已提交
249 250 251 252 253
	/*
	 * We can keep a list of all filesystems with quotas mounted for
	 * debugging and statistical purposes, but ...
	 * Just take a reference and get out.
	 */
254 255
	xfs_Gqm->qm_nrefs++;
	mutex_unlock(&xfs_Gqm_lock);
L
Linus Torvalds 已提交
256 257 258 259 260 261 262 263 264 265 266 267 268 269

	return 0;
}


/*
 * Release the reference that a filesystem took at mount time,
 * so that we know when we need to destroy the entire quota manager.
 */
/* ARGSUSED */
STATIC void
xfs_qm_rele_quotafs_ref(
	struct xfs_mount *mp)
{
270
	xfs_dquot_t	*dqp, *n;
L
Linus Torvalds 已提交
271 272 273 274 275 276 277

	ASSERT(xfs_Gqm);
	ASSERT(xfs_Gqm->qm_nrefs > 0);

	/*
	 * Go thru the freelist and destroy all inactive dquots.
	 */
278
	mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
L
Linus Torvalds 已提交
279

280
	list_for_each_entry_safe(dqp, n, &xfs_Gqm->qm_dqfrlist, q_freelist) {
L
Linus Torvalds 已提交
281 282 283 284
		xfs_dqlock(dqp);
		if (dqp->dq_flags & XFS_DQ_INACTIVE) {
			ASSERT(dqp->q_mount == NULL);
			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
285
			ASSERT(list_empty(&dqp->q_hashlist));
286
			ASSERT(list_empty(&dqp->q_mplist));
287 288
			list_del_init(&dqp->q_freelist);
			xfs_Gqm->qm_dqfrlist_cnt--;
L
Linus Torvalds 已提交
289 290 291 292 293 294
			xfs_dqunlock(dqp);
			xfs_qm_dqdestroy(dqp);
		} else {
			xfs_dqunlock(dqp);
		}
	}
295
	mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
L
Linus Torvalds 已提交
296 297 298 299 300

	/*
	 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
	 * be restarted.
	 */
301 302
	mutex_lock(&xfs_Gqm_lock);
	if (--xfs_Gqm->qm_nrefs == 0) {
L
Linus Torvalds 已提交
303 304 305
		xfs_qm_destroy(xfs_Gqm);
		xfs_Gqm = NULL;
	}
306
	mutex_unlock(&xfs_Gqm_lock);
L
Linus Torvalds 已提交
307 308 309 310 311 312
}

/*
 * Just destroy the quotainfo structure.
 */
void
C
Christoph Hellwig 已提交
313 314
xfs_qm_unmount(
	struct xfs_mount	*mp)
L
Linus Torvalds 已提交
315
{
C
Christoph Hellwig 已提交
316
	if (mp->m_quotainfo) {
317
		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
L
Linus Torvalds 已提交
318
		xfs_qm_destroy_quotainfo(mp);
C
Christoph Hellwig 已提交
319
	}
L
Linus Torvalds 已提交
320 321 322 323 324 325 326 327
}


/*
 * This is called from xfs_mountfs to start quotas and initialize all
 * necessary data structures like quotainfo.  This is also responsible for
 * running a quotacheck as necessary.  We are guaranteed that the superblock
 * is consistently read in at this point.
328 329 330
 *
 * If we fail here, the mount will continue with quota turned off. We don't
 * need to inidicate success or failure at all.
L
Linus Torvalds 已提交
331
 */
332
void
L
Linus Torvalds 已提交
333
xfs_qm_mount_quotas(
C
Christoph Hellwig 已提交
334
	xfs_mount_t	*mp)
L
Linus Torvalds 已提交
335 336 337 338 339 340 341 342 343
{
	int		error = 0;
	uint		sbf;

	/*
	 * If quotas on realtime volumes is not supported, we disable
	 * quotas immediately.
	 */
	if (mp->m_sb.sb_rextents) {
344
		xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
L
Linus Torvalds 已提交
345 346 347 348 349
		mp->m_qflags = 0;
		goto write_changes;
	}

	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
350

L
Linus Torvalds 已提交
351 352 353 354
	/*
	 * Allocate the quotainfo structure inside the mount struct, and
	 * create quotainode(s), and change/rev superblock if necessary.
	 */
355 356
	error = xfs_qm_init_quotainfo(mp);
	if (error) {
L
Linus Torvalds 已提交
357 358 359 360 361 362 363 364 365 366
		/*
		 * We must turn off quotas.
		 */
		ASSERT(mp->m_quotainfo == NULL);
		mp->m_qflags = 0;
		goto write_changes;
	}
	/*
	 * If any of the quotas are not consistent, do a quotacheck.
	 */
C
Christoph Hellwig 已提交
367
	if (XFS_QM_NEED_QUOTACHECK(mp)) {
368 369 370 371
		error = xfs_qm_quotacheck(mp);
		if (error) {
			/* Quotacheck failed and disabled quotas. */
			return;
L
Linus Torvalds 已提交
372 373
		}
	}
374 375 376 377 378
	/* 
	 * If one type of quotas is off, then it will lose its
	 * quotachecked status, since we won't be doing accounting for
	 * that type anymore.
	 */
379
	if (!XFS_IS_UQUOTA_ON(mp))
380
		mp->m_qflags &= ~XFS_UQUOTA_CHKD;
381
	if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
382
		mp->m_qflags &= ~XFS_OQUOTA_CHKD;
383

L
Linus Torvalds 已提交
384 385
 write_changes:
	/*
E
Eric Sandeen 已提交
386
	 * We actually don't have to acquire the m_sb_lock at all.
L
Linus Torvalds 已提交
387 388
	 * This can only be called from mount, and that's single threaded. XXX
	 */
E
Eric Sandeen 已提交
389
	spin_lock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
390 391
	sbf = mp->m_sb.sb_qflags;
	mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
E
Eric Sandeen 已提交
392
	spin_unlock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
393 394 395 396 397 398 399 400 401 402

	if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
		if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
			/*
			 * We could only have been turning quotas off.
			 * We aren't in very good shape actually because
			 * the incore structures are convinced that quotas are
			 * off, but the on disk superblock doesn't know that !
			 */
			ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
403 404
			xfs_alert(mp, "%s: Superblock update failed!",
				__func__);
L
Linus Torvalds 已提交
405 406 407 408
		}
	}

	if (error) {
409
		xfs_warn(mp, "Failed to initialize disk quotas.");
C
Christoph Hellwig 已提交
410
		return;
L
Linus Torvalds 已提交
411
	}
C
Christoph Hellwig 已提交
412 413 414 415 416

#ifdef QUOTADEBUG
	if (XFS_IS_QUOTA_ON(mp))
		xfs_qm_internalqcheck(mp);
#endif
L
Linus Torvalds 已提交
417 418 419 420 421
}

/*
 * Called from the vfsops layer.
 */
422
void
L
Linus Torvalds 已提交
423 424 425 426 427 428 429 430 431 432 433 434 435 436 437
xfs_qm_unmount_quotas(
	xfs_mount_t	*mp)
{
	/*
	 * Release the dquots that root inode, et al might be holding,
	 * before we flush quotas and blow away the quotainfo structure.
	 */
	ASSERT(mp->m_rootip);
	xfs_qm_dqdetach(mp->m_rootip);
	if (mp->m_rbmip)
		xfs_qm_dqdetach(mp->m_rbmip);
	if (mp->m_rsumip)
		xfs_qm_dqdetach(mp->m_rsumip);

	/*
438
	 * Release the quota inodes.
L
Linus Torvalds 已提交
439 440
	 */
	if (mp->m_quotainfo) {
441 442 443
		if (mp->m_quotainfo->qi_uquotaip) {
			IRELE(mp->m_quotainfo->qi_uquotaip);
			mp->m_quotainfo->qi_uquotaip = NULL;
L
Linus Torvalds 已提交
444
		}
445 446 447
		if (mp->m_quotainfo->qi_gquotaip) {
			IRELE(mp->m_quotainfo->qi_gquotaip);
			mp->m_quotainfo->qi_gquotaip = NULL;
L
Linus Torvalds 已提交
448 449 450 451 452 453 454 455
		}
	}
}

/*
 * Flush all dquots of the given file system to disk. The dquots are
 * _not_ purged from memory here, just their data written to disk.
 */
456
STATIC int
L
Linus Torvalds 已提交
457
xfs_qm_dqflush_all(
458 459
	struct xfs_mount	*mp,
	int			sync_mode)
L
Linus Torvalds 已提交
460
{
461 462 463 464
	struct xfs_quotainfo	*q = mp->m_quotainfo;
	int			recl;
	struct xfs_dquot	*dqp;
	int			error;
L
Linus Torvalds 已提交
465

466
	if (!q)
467
		return 0;
L
Linus Torvalds 已提交
468
again:
469 470
	mutex_lock(&q->qi_dqlist_lock);
	list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
471 472 473 474 475
		xfs_dqlock(dqp);
		if (! XFS_DQ_IS_DIRTY(dqp)) {
			xfs_dqunlock(dqp);
			continue;
		}
C
Christoph Hellwig 已提交
476

L
Linus Torvalds 已提交
477
		/* XXX a sentinel would be better */
478
		recl = q->qi_dqreclaims;
479
		if (!xfs_dqflock_nowait(dqp)) {
L
Linus Torvalds 已提交
480 481 482 483 484 485 486 487 488 489 490 491 492
			/*
			 * If we can't grab the flush lock then check
			 * to see if the dquot has been flushed delayed
			 * write.  If so, grab its buffer and send it
			 * out immediately.  We'll be able to acquire
			 * the flush lock when the I/O completes.
			 */
			xfs_qm_dqflock_pushbuf_wait(dqp);
		}
		/*
		 * Let go of the mplist lock. We don't want to hold it
		 * across a disk write.
		 */
493
		mutex_unlock(&q->qi_dqlist_lock);
494
		error = xfs_qm_dqflush(dqp, sync_mode);
L
Linus Torvalds 已提交
495 496
		xfs_dqunlock(dqp);
		if (error)
497
			return error;
L
Linus Torvalds 已提交
498

499 500 501
		mutex_lock(&q->qi_dqlist_lock);
		if (recl != q->qi_dqreclaims) {
			mutex_unlock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
502 503 504 505 506
			/* XXX restart limit */
			goto again;
		}
	}

507
	mutex_unlock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
508
	/* return ! busy */
509
	return 0;
L
Linus Torvalds 已提交
510 511 512 513 514 515 516
}
/*
 * Release the group dquot pointers the user dquots may be
 * carrying around as a hint. mplist is locked on entry and exit.
 */
STATIC void
xfs_qm_detach_gdquots(
517
	struct xfs_mount	*mp)
L
Linus Torvalds 已提交
518
{
519 520 521
	struct xfs_quotainfo	*q = mp->m_quotainfo;
	struct xfs_dquot	*dqp, *gdqp;
	int			nrecl;
L
Linus Torvalds 已提交
522 523

 again:
524 525
	ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
	list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
526 527 528 529 530 531 532 533 534 535 536 537
		xfs_dqlock(dqp);
		if ((gdqp = dqp->q_gdquot)) {
			xfs_dqlock(gdqp);
			dqp->q_gdquot = NULL;
		}
		xfs_dqunlock(dqp);

		if (gdqp) {
			/*
			 * Can't hold the mplist lock across a dqput.
			 * XXXmust convert to marker based iterations here.
			 */
538 539
			nrecl = q->qi_dqreclaims;
			mutex_unlock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
540 541
			xfs_qm_dqput(gdqp);

542 543
			mutex_lock(&q->qi_dqlist_lock);
			if (nrecl != q->qi_dqreclaims)
L
Linus Torvalds 已提交
544 545 546 547 548 549 550 551 552 553 554 555 556
				goto again;
		}
	}
}

/*
 * Go through all the incore dquots of this file system and take them
 * off the mplist and hashlist, if the dquot type matches the dqtype
 * parameter. This is used when turning off quota accounting for
 * users and/or groups, as well as when the filesystem is unmounting.
 */
STATIC int
xfs_qm_dqpurge_int(
557 558
	struct xfs_mount	*mp,
	uint			flags)
L
Linus Torvalds 已提交
559
{
560 561 562 563 564
	struct xfs_quotainfo	*q = mp->m_quotainfo;
	struct xfs_dquot	*dqp, *n;
	uint			dqtype;
	int			nrecl;
	int			nmisses;
L
Linus Torvalds 已提交
565

566
	if (!q)
567
		return 0;
L
Linus Torvalds 已提交
568 569

	dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
570
	dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
L
Linus Torvalds 已提交
571 572
	dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;

573
	mutex_lock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
574 575 576 577 578 579 580 581 582 583 584

	/*
	 * In the first pass through all incore dquots of this filesystem,
	 * we release the group dquot pointers the user dquots may be
	 * carrying around as a hint. We need to do this irrespective of
	 * what's being turned off.
	 */
	xfs_qm_detach_gdquots(mp);

      again:
	nmisses = 0;
585
	ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
L
Linus Torvalds 已提交
586 587 588 589
	/*
	 * Try to get rid of all of the unwanted dquots. The idea is to
	 * get them off mplist and hashlist, but leave them on freelist.
	 */
590
	list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
591 592 593 594 595
		/*
		 * It's OK to look at the type without taking dqlock here.
		 * We're holding the mplist lock here, and that's needed for
		 * a dqreclaim.
		 */
596
		if ((dqp->dq_flags & dqtype) == 0)
L
Linus Torvalds 已提交
597 598
			continue;

599
		if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
600 601
			nrecl = q->qi_dqreclaims;
			mutex_unlock(&q->qi_dqlist_lock);
602
			mutex_lock(&dqp->q_hash->qh_lock);
603
			mutex_lock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
604 605 606 607 608 609 610

			/*
			 * XXXTheoretically, we can get into a very long
			 * ping pong game here.
			 * No one can be adding dquots to the mplist at
			 * this point, but somebody might be taking things off.
			 */
611
			if (nrecl != q->qi_dqreclaims) {
612
				mutex_unlock(&dqp->q_hash->qh_lock);
L
Linus Torvalds 已提交
613 614 615 616 617 618 619 620
				goto again;
			}
		}

		/*
		 * Take the dquot off the mplist and hashlist. It may remain on
		 * freelist in INACTIVE state.
		 */
621
		nmisses += xfs_qm_dqpurge(dqp);
L
Linus Torvalds 已提交
622
	}
623
	mutex_unlock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657
	return nmisses;
}

int
xfs_qm_dqpurge_all(
	xfs_mount_t	*mp,
	uint		flags)
{
	int		ndquots;

	/*
	 * Purge the dquot cache.
	 * None of the dquots should really be busy at this point.
	 */
	if (mp->m_quotainfo) {
		while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
			delay(ndquots * 10);
		}
	}
	return 0;
}

STATIC int
xfs_qm_dqattach_one(
	xfs_inode_t	*ip,
	xfs_dqid_t	id,
	uint		type,
	uint		doalloc,
	xfs_dquot_t	*udqhint, /* hint */
	xfs_dquot_t	**IO_idqpp)
{
	xfs_dquot_t	*dqp;
	int		error;

C
Christoph Hellwig 已提交
658
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
659
	error = 0;
660

L
Linus Torvalds 已提交
661 662 663 664 665
	/*
	 * See if we already have it in the inode itself. IO_idqpp is
	 * &i_udquot or &i_gdquot. This made the code look weird, but
	 * made the logic a lot simpler.
	 */
666 667
	dqp = *IO_idqpp;
	if (dqp) {
C
Christoph Hellwig 已提交
668
		trace_xfs_dqattach_found(dqp);
669
		return 0;
L
Linus Torvalds 已提交
670 671 672 673
	}

	/*
	 * udqhint is the i_udquot field in inode, and is non-NULL only
674
	 * when the type arg is group/project. Its purpose is to save a
L
Linus Torvalds 已提交
675 676 677
	 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
	 * the user dquot.
	 */
678 679
	if (udqhint) {
		ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
L
Linus Torvalds 已提交
680 681
		xfs_dqlock(udqhint);

682 683 684 685 686 687 688 689 690 691 692 693 694 695
		/*
		 * No need to take dqlock to look at the id.
		 *
		 * The ID can't change until it gets reclaimed, and it won't
		 * be reclaimed as long as we have a ref from inode and we
		 * hold the ilock.
		 */
		dqp = udqhint->q_gdquot;
		if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
			xfs_dqlock(dqp);
			XFS_DQHOLD(dqp);
			ASSERT(*IO_idqpp == NULL);
			*IO_idqpp = dqp;

L
Linus Torvalds 已提交
696 697
			xfs_dqunlock(dqp);
			xfs_dqunlock(udqhint);
698
			return 0;
L
Linus Torvalds 已提交
699
		}
700 701 702 703 704 705 706

		/*
		 * We can't hold a dquot lock when we call the dqget code.
		 * We'll deadlock in no time, because of (not conforming to)
		 * lock ordering - the inodelock comes before any dquot lock,
		 * and we may drop and reacquire the ilock in xfs_qm_dqget().
		 */
L
Linus Torvalds 已提交
707
		xfs_dqunlock(udqhint);
708 709
	}

L
Linus Torvalds 已提交
710 711 712 713 714 715 716
	/*
	 * Find the dquot from somewhere. This bumps the
	 * reference count of dquot and returns it locked.
	 * This can return ENOENT if dquot didn't exist on
	 * disk and we didn't ask it to allocate;
	 * ESRCH if quotas got turned off suddenly.
	 */
717 718 719
	error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
	if (error)
		return error;
L
Linus Torvalds 已提交
720

C
Christoph Hellwig 已提交
721
	trace_xfs_dqattach_get(dqp);
722

L
Linus Torvalds 已提交
723 724 725 726 727
	/*
	 * dqget may have dropped and re-acquired the ilock, but it guarantees
	 * that the dquot returned is the one that should go in the inode.
	 */
	*IO_idqpp = dqp;
728 729
	xfs_dqunlock(dqp);
	return 0;
L
Linus Torvalds 已提交
730 731 732 733 734 735 736
}


/*
 * Given a udquot and gdquot, attach a ptr to the group dquot in the
 * udquot as a hint for future lookups. The idea sounds simple, but the
 * execution isn't, because the udquot might have a group dquot attached
737
 * already and getting rid of that gets us into lock ordering constraints.
L
Linus Torvalds 已提交
738 739 740 741 742 743
 * The process is complicated more by the fact that the dquots may or may not
 * be locked on entry.
 */
STATIC void
xfs_qm_dqattach_grouphint(
	xfs_dquot_t	*udq,
744
	xfs_dquot_t	*gdq)
L
Linus Torvalds 已提交
745 746 747
{
	xfs_dquot_t	*tmp;

748
	xfs_dqlock(udq);
L
Linus Torvalds 已提交
749 750 751

	if ((tmp = udq->q_gdquot)) {
		if (tmp == gdq) {
752
			xfs_dqunlock(udq);
L
Linus Torvalds 已提交
753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773
			return;
		}

		udq->q_gdquot = NULL;
		/*
		 * We can't keep any dqlocks when calling dqrele,
		 * because the freelist lock comes before dqlocks.
		 */
		xfs_dqunlock(udq);
		/*
		 * we took a hard reference once upon a time in dqget,
		 * so give it back when the udquot no longer points at it
		 * dqput() does the unlocking of the dquot.
		 */
		xfs_qm_dqrele(tmp);

		xfs_dqlock(udq);
		xfs_dqlock(gdq);

	} else {
		ASSERT(XFS_DQ_IS_LOCKED(udq));
774
		xfs_dqlock(gdq);
L
Linus Torvalds 已提交
775 776 777 778 779 780 781 782 783 784 785 786
	}

	ASSERT(XFS_DQ_IS_LOCKED(udq));
	ASSERT(XFS_DQ_IS_LOCKED(gdq));
	/*
	 * Somebody could have attached a gdquot here,
	 * when we dropped the uqlock. If so, just do nothing.
	 */
	if (udq->q_gdquot == NULL) {
		XFS_DQHOLD(gdq);
		udq->q_gdquot = gdq;
	}
787 788 789

	xfs_dqunlock(gdq);
	xfs_dqunlock(udq);
L
Linus Torvalds 已提交
790 791 792 793
}


/*
794 795
 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 * into account.
L
Linus Torvalds 已提交
796 797 798 799 800
 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 * Inode may get unlocked and relocked in here, and the caller must deal with
 * the consequences.
 */
int
C
Christoph Hellwig 已提交
801
xfs_qm_dqattach_locked(
L
Linus Torvalds 已提交
802 803 804 805 806 807 808
	xfs_inode_t	*ip,
	uint		flags)
{
	xfs_mount_t	*mp = ip->i_mount;
	uint		nquotas = 0;
	int		error = 0;

C
Christoph Hellwig 已提交
809 810 811 812 813
	if (!XFS_IS_QUOTA_RUNNING(mp) ||
	    !XFS_IS_QUOTA_ON(mp) ||
	    !XFS_NOT_DQATTACHED(mp, ip) ||
	    ip->i_ino == mp->m_sb.sb_uquotino ||
	    ip->i_ino == mp->m_sb.sb_gquotino)
814
		return 0;
L
Linus Torvalds 已提交
815

C
Christoph Hellwig 已提交
816
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
817 818 819 820 821 822 823 824 825

	if (XFS_IS_UQUOTA_ON(mp)) {
		error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
						flags & XFS_QMOPT_DQALLOC,
						NULL, &ip->i_udquot);
		if (error)
			goto done;
		nquotas++;
	}
C
Christoph Hellwig 已提交
826 827

	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
828 829 830 831 832
	if (XFS_IS_OQUOTA_ON(mp)) {
		error = XFS_IS_GQUOTA_ON(mp) ?
			xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
						flags & XFS_QMOPT_DQALLOC,
						ip->i_udquot, &ip->i_gdquot) :
833
			xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
L
Linus Torvalds 已提交
834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849
						flags & XFS_QMOPT_DQALLOC,
						ip->i_udquot, &ip->i_gdquot);
		/*
		 * Don't worry about the udquot that we may have
		 * attached above. It'll get detached, if not already.
		 */
		if (error)
			goto done;
		nquotas++;
	}

	/*
	 * Attach this group quota to the user quota as a hint.
	 * This WON'T, in general, result in a thrash.
	 */
	if (nquotas == 2) {
C
Christoph Hellwig 已提交
850
		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
851 852 853 854 855 856 857 858 859 860 861 862 863 864
		ASSERT(ip->i_udquot);
		ASSERT(ip->i_gdquot);

		/*
		 * We may or may not have the i_udquot locked at this point,
		 * but this check is OK since we don't depend on the i_gdquot to
		 * be accurate 100% all the time. It is just a hint, and this
		 * will succeed in general.
		 */
		if (ip->i_udquot->q_gdquot == ip->i_gdquot)
			goto done;
		/*
		 * Attach i_gdquot to the gdquot hint inside the i_udquot.
		 */
865
		xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
L
Linus Torvalds 已提交
866 867
	}

C
Christoph Hellwig 已提交
868
 done:
L
Linus Torvalds 已提交
869 870 871 872
#ifdef QUOTADEBUG
	if (! error) {
		if (XFS_IS_UQUOTA_ON(mp))
			ASSERT(ip->i_udquot);
873
		if (XFS_IS_OQUOTA_ON(mp))
L
Linus Torvalds 已提交
874 875
			ASSERT(ip->i_gdquot);
	}
C
Christoph Hellwig 已提交
876
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
877
#endif
C
Christoph Hellwig 已提交
878 879
	return error;
}
L
Linus Torvalds 已提交
880

C
Christoph Hellwig 已提交
881 882 883 884 885 886 887 888 889 890
int
xfs_qm_dqattach(
	struct xfs_inode	*ip,
	uint			flags)
{
	int			error;

	xfs_ilock(ip, XFS_ILOCK_EXCL);
	error = xfs_qm_dqattach_locked(ip, flags);
	xfs_iunlock(ip, XFS_ILOCK_EXCL);
L
Linus Torvalds 已提交
891

892
	return error;
L
Linus Torvalds 已提交
893 894 895 896 897 898 899 900 901 902 903 904 905 906
}

/*
 * Release dquots (and their references) if any.
 * The inode should be locked EXCL except when this's called by
 * xfs_ireclaim.
 */
void
xfs_qm_dqdetach(
	xfs_inode_t	*ip)
{
	if (!(ip->i_udquot || ip->i_gdquot))
		return;

C
Christoph Hellwig 已提交
907 908
	trace_xfs_dquot_dqdetach(ip);

L
Linus Torvalds 已提交
909 910 911 912 913 914 915 916 917 918 919 920 921 922
	ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
	ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
	if (ip->i_udquot) {
		xfs_qm_dqrele(ip->i_udquot);
		ip->i_udquot = NULL;
	}
	if (ip->i_gdquot) {
		xfs_qm_dqrele(ip->i_gdquot);
		ip->i_gdquot = NULL;
	}
}

int
xfs_qm_sync(
923 924
	struct xfs_mount	*mp,
	int			flags)
L
Linus Torvalds 已提交
925
{
926 927 928 929
	struct xfs_quotainfo	*q = mp->m_quotainfo;
	int			recl, restarts;
	struct xfs_dquot	*dqp;
	int			error;
L
Linus Torvalds 已提交
930

C
Christoph Hellwig 已提交
931
	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
932 933
		return 0;

L
Linus Torvalds 已提交
934 935 936
	restarts = 0;

  again:
937
	mutex_lock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
938 939 940 941 942 943
	/*
	 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
	 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
	 * when we have the mplist lock, we know that dquots will be consistent
	 * as long as we have it locked.
	 */
944
	if (!XFS_IS_QUOTA_ON(mp)) {
945
		mutex_unlock(&q->qi_dqlist_lock);
946
		return 0;
L
Linus Torvalds 已提交
947
	}
948 949
	ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
	list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
950 951 952 953 954
		/*
		 * If this is vfs_sync calling, then skip the dquots that
		 * don't 'seem' to be dirty. ie. don't acquire dqlock.
		 * This is very similar to what xfs_sync does with inodes.
		 */
C
Christoph Hellwig 已提交
955 956
		if (flags & SYNC_TRYLOCK) {
			if (!XFS_DQ_IS_DIRTY(dqp))
L
Linus Torvalds 已提交
957
				continue;
C
Christoph Hellwig 已提交
958
			if (!xfs_qm_dqlock_nowait(dqp))
L
Linus Torvalds 已提交
959 960 961 962 963 964 965 966 967 968 969 970 971 972
				continue;
		} else {
			xfs_dqlock(dqp);
		}

		/*
		 * Now, find out for sure if this dquot is dirty or not.
		 */
		if (! XFS_DQ_IS_DIRTY(dqp)) {
			xfs_dqunlock(dqp);
			continue;
		}

		/* XXX a sentinel would be better */
973
		recl = q->qi_dqreclaims;
974
		if (!xfs_dqflock_nowait(dqp)) {
C
Christoph Hellwig 已提交
975
			if (flags & SYNC_TRYLOCK) {
L
Linus Torvalds 已提交
976 977 978 979 980
				xfs_dqunlock(dqp);
				continue;
			}
			/*
			 * If we can't grab the flush lock then if the caller
981
			 * really wanted us to give this our best shot, so
L
Linus Torvalds 已提交
982 983
			 * see if we can give a push to the buffer before we wait
			 * on the flush lock. At this point, we know that
984
			 * even though the dquot is being flushed,
L
Linus Torvalds 已提交
985 986 987 988 989 990 991 992
			 * it has (new) dirty data.
			 */
			xfs_qm_dqflock_pushbuf_wait(dqp);
		}
		/*
		 * Let go of the mplist lock. We don't want to hold it
		 * across a disk write
		 */
993
		mutex_unlock(&q->qi_dqlist_lock);
994
		error = xfs_qm_dqflush(dqp, flags);
L
Linus Torvalds 已提交
995 996
		xfs_dqunlock(dqp);
		if (error && XFS_FORCED_SHUTDOWN(mp))
997
			return 0;	/* Need to prevent umount failure */
L
Linus Torvalds 已提交
998
		else if (error)
999
			return error;
L
Linus Torvalds 已提交
1000

1001 1002
		mutex_lock(&q->qi_dqlist_lock);
		if (recl != q->qi_dqreclaims) {
L
Linus Torvalds 已提交
1003 1004 1005
			if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
				break;

1006
			mutex_unlock(&q->qi_dqlist_lock);
L
Linus Torvalds 已提交
1007 1008 1009 1010
			goto again;
		}
	}

1011
	mutex_unlock(&q->qi_dqlist_lock);
1012
	return 0;
L
Linus Torvalds 已提交
1013 1014
}

1015 1016 1017 1018 1019 1020 1021
/*
 * The hash chains and the mplist use the same xfs_dqhash structure as
 * their list head, but we can take the mplist qh_lock and one of the
 * hash qh_locks at the same time without any problem as they aren't
 * related.
 */
static struct lock_class_key xfs_quota_mplist_class;
L
Linus Torvalds 已提交
1022 1023 1024 1025 1026

/*
 * This initializes all the quota information that's kept in the
 * mount structure
 */
1027
STATIC int
L
Linus Torvalds 已提交
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040
xfs_qm_init_quotainfo(
	xfs_mount_t	*mp)
{
	xfs_quotainfo_t *qinf;
	int		error;
	xfs_dquot_t	*dqp;

	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	/*
	 * Tell XQM that we exist as soon as possible.
	 */
	if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1041
		return error;
L
Linus Torvalds 已提交
1042 1043 1044 1045 1046 1047 1048 1049 1050
	}

	qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);

	/*
	 * See if quotainodes are setup, and if not, allocate them,
	 * and change the superblock accordingly.
	 */
	if ((error = xfs_qm_init_quotainos(mp))) {
1051
		kmem_free(qinf);
L
Linus Torvalds 已提交
1052
		mp->m_quotainfo = NULL;
1053
		return error;
L
Linus Torvalds 已提交
1054 1055
	}

1056 1057 1058
	INIT_LIST_HEAD(&qinf->qi_dqlist);
	mutex_init(&qinf->qi_dqlist_lock);
	lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
1059

L
Linus Torvalds 已提交
1060 1061 1062
	qinf->qi_dqreclaims = 0;

	/* mutex used to serialize quotaoffs */
1063
	mutex_init(&qinf->qi_quotaofflock);
L
Linus Torvalds 已提交
1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081

	/* Precalc some constants */
	qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
	ASSERT(qinf->qi_dqchunklen);
	qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
	do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));

	mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);

	/*
	 * We try to get the limits from the superuser's limits fields.
	 * This is quite hacky, but it is standard quota practice.
	 * We look at the USR dquot with id == 0 first, but if user quotas
	 * are not enabled we goto the GRP dquot with id == 0.
	 * We don't really care to keep separate default limits for user
	 * and group quotas, at least not at this point.
	 */
	error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1082 1083 1084
			     XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER : 
			     (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
				XFS_DQ_PROJ),
L
Linus Torvalds 已提交
1085 1086 1087 1088 1089 1090 1091 1092 1093 1094
			     XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
			     &dqp);
	if (! error) {
		xfs_disk_dquot_t	*ddqp = &dqp->q_core;

		/*
		 * The warnings and timers set the grace period given to
		 * a user or group before he or she can not perform any
		 * more writing. If it is zero, a default is used.
		 */
1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112
		qinf->qi_btimelimit = ddqp->d_btimer ?
			be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
		qinf->qi_itimelimit = ddqp->d_itimer ?
			be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
		qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
			be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
		qinf->qi_bwarnlimit = ddqp->d_bwarns ?
			be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
		qinf->qi_iwarnlimit = ddqp->d_iwarns ?
			be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
		qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
			be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
		qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
		qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
		qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
		qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
		qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
		qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
L
Linus Torvalds 已提交
1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125
 
		/*
		 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
		 * we don't want this dquot cached. We haven't done a
		 * quotacheck yet, and quotacheck doesn't like incore dquots.
		 */
		xfs_qm_dqdestroy(dqp);
	} else {
		qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
		qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
		qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
		qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
		qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1126
		qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
L
Linus Torvalds 已提交
1127 1128
	}

1129
	return 0;
L
Linus Torvalds 已提交
1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
}


/*
 * Gets called when unmounting a filesystem or when all quotas get
 * turned off.
 * This purges the quota inodes, destroys locks and frees itself.
 */
void
xfs_qm_destroy_quotainfo(
	xfs_mount_t	*mp)
{
	xfs_quotainfo_t *qi;

	qi = mp->m_quotainfo;
	ASSERT(qi != NULL);
	ASSERT(xfs_Gqm != NULL);

	/*
	 * Release the reference that XQM kept, so that we know
	 * when the XQM structure should be freed. We cannot assume
	 * that xfs_Gqm is non-null after this point.
	 */
	xfs_qm_rele_quotafs_ref(mp);

1155 1156
	ASSERT(list_empty(&qi->qi_dqlist));
	mutex_destroy(&qi->qi_dqlist_lock);
L
Linus Torvalds 已提交
1157 1158

	if (qi->qi_uquotaip) {
1159
		IRELE(qi->qi_uquotaip);
L
Linus Torvalds 已提交
1160 1161 1162
		qi->qi_uquotaip = NULL; /* paranoia */
	}
	if (qi->qi_gquotaip) {
1163
		IRELE(qi->qi_gquotaip);
L
Linus Torvalds 已提交
1164 1165 1166
		qi->qi_gquotaip = NULL;
	}
	mutex_destroy(&qi->qi_quotaofflock);
1167
	kmem_free(qi);
L
Linus Torvalds 已提交
1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181
	mp->m_quotainfo = NULL;
}



/* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */

/* ARGSUSED */
STATIC void
xfs_qm_list_init(
	xfs_dqlist_t	*list,
	char		*str,
	int		n)
{
1182
	mutex_init(&list->qh_lock);
1183
	INIT_LIST_HEAD(&list->qh_list);
L
Linus Torvalds 已提交
1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209
	list->qh_version = 0;
	list->qh_nelems = 0;
}

STATIC void
xfs_qm_list_destroy(
	xfs_dqlist_t	*list)
{
	mutex_destroy(&(list->qh_lock));
}

/*
 * Create an inode and return with a reference already taken, but unlocked
 * This is how we create quota inodes
 */
STATIC int
xfs_qm_qino_alloc(
	xfs_mount_t	*mp,
	xfs_inode_t	**ip,
	__int64_t	sbfields,
	uint		flags)
{
	xfs_trans_t	*tp;
	int		error;
	int		committed;

1210
	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
L
Linus Torvalds 已提交
1211 1212 1213 1214 1215 1216
	if ((error = xfs_trans_reserve(tp,
				      XFS_QM_QINOCREATE_SPACE_RES(mp),
				      XFS_CREATE_LOG_RES(mp), 0,
				      XFS_TRANS_PERM_LOG_RES,
				      XFS_CREATE_LOG_COUNT))) {
		xfs_trans_cancel(tp, 0);
1217
		return error;
L
Linus Torvalds 已提交
1218 1219
	}

C
Christoph Hellwig 已提交
1220 1221
	error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, &committed);
	if (error) {
L
Linus Torvalds 已提交
1222 1223
		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
				 XFS_TRANS_ABORT);
1224
		return error;
L
Linus Torvalds 已提交
1225 1226 1227 1228 1229 1230 1231
	}

	/*
	 * Make the changes in the superblock, and log those too.
	 * sbfields arg may contain fields other than *QUOTINO;
	 * VERSIONNUM for example.
	 */
E
Eric Sandeen 已提交
1232
	spin_lock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
1233
	if (flags & XFS_QMOPT_SBVERSION) {
1234
		ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
L
Linus Torvalds 已提交
1235 1236 1237 1238 1239
		ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
				   XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
		       (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
			XFS_SB_GQUOTINO | XFS_SB_QFLAGS));

1240
		xfs_sb_version_addquota(&mp->m_sb);
L
Linus Torvalds 已提交
1241 1242 1243 1244 1245 1246 1247 1248 1249 1250
		mp->m_sb.sb_uquotino = NULLFSINO;
		mp->m_sb.sb_gquotino = NULLFSINO;

		/* qflags will get updated _after_ quotacheck */
		mp->m_sb.sb_qflags = 0;
	}
	if (flags & XFS_QMOPT_UQUOTA)
		mp->m_sb.sb_uquotino = (*ip)->i_ino;
	else
		mp->m_sb.sb_gquotino = (*ip)->i_ino;
E
Eric Sandeen 已提交
1251
	spin_unlock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
1252 1253
	xfs_mod_sb(tp, sbfields);

1254
	if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1255
		xfs_alert(mp, "%s failed (error %d)!", __func__, error);
1256
		return error;
L
Linus Torvalds 已提交
1257
	}
1258
	return 0;
L
Linus Torvalds 已提交
1259 1260 1261
}


1262
STATIC void
L
Linus Torvalds 已提交
1263 1264 1265 1266 1267 1268 1269 1270 1271
xfs_qm_reset_dqcounts(
	xfs_mount_t	*mp,
	xfs_buf_t	*bp,
	xfs_dqid_t	id,
	uint		type)
{
	xfs_disk_dquot_t	*ddq;
	int			j;

C
Christoph Hellwig 已提交
1272 1273
	trace_xfs_reset_dqcounts(bp, _RET_IP_);

L
Linus Torvalds 已提交
1274 1275 1276 1277 1278 1279 1280
	/*
	 * Reset all counters and timers. They'll be
	 * started afresh by xfs_qm_quotacheck.
	 */
#ifdef DEBUG
	j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
	do_div(j, sizeof(xfs_dqblk_t));
1281
	ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
L
Linus Torvalds 已提交
1282 1283
#endif
	ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1284
	for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
L
Linus Torvalds 已提交
1285 1286 1287
		/*
		 * Do a sanity check, and if needed, repair the dqblk. Don't
		 * output any warnings because it's perfectly possible to
1288
		 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
L
Linus Torvalds 已提交
1289
		 */
1290
		(void) xfs_qm_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
L
Linus Torvalds 已提交
1291
				      "xfs_quotacheck");
1292 1293 1294 1295 1296 1297 1298 1299 1300
		ddq->d_bcount = 0;
		ddq->d_icount = 0;
		ddq->d_rtbcount = 0;
		ddq->d_btimer = 0;
		ddq->d_itimer = 0;
		ddq->d_rtbtimer = 0;
		ddq->d_bwarns = 0;
		ddq->d_iwarns = 0;
		ddq->d_rtbwarns = 0;
L
Linus Torvalds 已提交
1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314
		ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
	}
}

STATIC int
xfs_qm_dqiter_bufs(
	xfs_mount_t	*mp,
	xfs_dqid_t	firstid,
	xfs_fsblock_t	bno,
	xfs_filblks_t	blkcnt,
	uint		flags)
{
	xfs_buf_t	*bp;
	int		error;
1315
	int		type;
L
Linus Torvalds 已提交
1316 1317

	ASSERT(blkcnt > 0);
1318 1319
	type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
		(flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
L
Linus Torvalds 已提交
1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333
	error = 0;

	/*
	 * Blkcnt arg can be a very big number, and might even be
	 * larger than the log itself. So, we have to break it up into
	 * manageable-sized transactions.
	 * Note that we don't start a permanent transaction here; we might
	 * not be able to get a log reservation for the whole thing up front,
	 * and we don't really care to either, because we just discard
	 * everything if we were to crash in the middle of this loop.
	 */
	while (blkcnt--) {
		error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
			      XFS_FSB_TO_DADDR(mp, bno),
1334
			      mp->m_quotainfo->qi_dqchunklen, 0, &bp);
L
Linus Torvalds 已提交
1335 1336 1337
		if (error)
			break;

1338
		xfs_qm_reset_dqcounts(mp, bp, firstid, type);
L
Linus Torvalds 已提交
1339 1340 1341 1342 1343
		xfs_bdwrite(mp, bp);
		/*
		 * goto the next block.
		 */
		bno++;
1344
		firstid += mp->m_quotainfo->qi_dqperchunk;
L
Linus Torvalds 已提交
1345
	}
1346
	return error;
L
Linus Torvalds 已提交
1347 1348 1349
}

/*
1350
 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
L
Linus Torvalds 已提交
1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369
 * caller supplied function for every chunk of dquots that we find.
 */
STATIC int
xfs_qm_dqiterate(
	xfs_mount_t	*mp,
	xfs_inode_t	*qip,
	uint		flags)
{
	xfs_bmbt_irec_t		*map;
	int			i, nmaps;	/* number of map entries */
	int			error;		/* return value */
	xfs_fileoff_t		lblkno;
	xfs_filblks_t		maxlblkcnt;
	xfs_dqid_t		firstid;
	xfs_fsblock_t		rablkno;
	xfs_filblks_t		rablkcnt;

	error = 0;
	/*
1370
	 * This looks racy, but we can't keep an inode lock across a
L
Linus Torvalds 已提交
1371 1372 1373 1374
	 * trans_reserve. But, this gets called during quotacheck, and that
	 * happens only at mount time which is single threaded.
	 */
	if (qip->i_d.di_nblocks == 0)
1375
		return 0;
L
Linus Torvalds 已提交
1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392

	map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);

	lblkno = 0;
	maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
	do {
		nmaps = XFS_DQITER_MAP_SIZE;
		/*
		 * We aren't changing the inode itself. Just changing
		 * some of its data. No new blocks are added here, and
		 * the inode is never added to the transaction.
		 */
		xfs_ilock(qip, XFS_ILOCK_SHARED);
		error = xfs_bmapi(NULL, qip, lblkno,
				  maxlblkcnt - lblkno,
				  XFS_BMAPI_METADATA,
				  NULL,
1393
				  0, map, &nmaps, NULL);
L
Linus Torvalds 已提交
1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409
		xfs_iunlock(qip, XFS_ILOCK_SHARED);
		if (error)
			break;

		ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
		for (i = 0; i < nmaps; i++) {
			ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
			ASSERT(map[i].br_blockcount);


			lblkno += map[i].br_blockcount;

			if (map[i].br_startblock == HOLESTARTBLOCK)
				continue;

			firstid = (xfs_dqid_t) map[i].br_startoff *
1410
				mp->m_quotainfo->qi_dqperchunk;
L
Linus Torvalds 已提交
1411 1412 1413 1414 1415 1416 1417 1418
			/*
			 * Do a read-ahead on the next extent.
			 */
			if ((i+1 < nmaps) &&
			    (map[i+1].br_startblock != HOLESTARTBLOCK)) {
				rablkcnt =  map[i+1].br_blockcount;
				rablkno = map[i+1].br_startblock;
				while (rablkcnt--) {
C
Christoph Hellwig 已提交
1419
					xfs_buf_readahead(mp->m_ddev_targp,
L
Linus Torvalds 已提交
1420
					       XFS_FSB_TO_DADDR(mp, rablkno),
1421
					       mp->m_quotainfo->qi_dqchunklen);
L
Linus Torvalds 已提交
1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441
					rablkno++;
				}
			}
			/*
			 * Iterate thru all the blks in the extent and
			 * reset the counters of all the dquots inside them.
			 */
			if ((error = xfs_qm_dqiter_bufs(mp,
						       firstid,
						       map[i].br_startblock,
						       map[i].br_blockcount,
						       flags))) {
				break;
			}
		}

		if (error)
			break;
	} while (nmaps > 0);

1442
	kmem_free(map);
L
Linus Torvalds 已提交
1443

1444
	return error;
L
Linus Torvalds 已提交
1445 1446 1447 1448
}

/*
 * Called by dqusage_adjust in doing a quotacheck.
1449 1450 1451 1452 1453
 *
 * Given the inode, and a dquot id this updates both the incore dqout as well
 * as the buffer copy. This is so that once the quotacheck is done, we can
 * just log all the buffers, as opposed to logging numerous updates to
 * individual dquots.
L
Linus Torvalds 已提交
1454
 */
1455
STATIC int
L
Linus Torvalds 已提交
1456
xfs_qm_quotacheck_dqadjust(
1457 1458 1459
	struct xfs_inode	*ip,
	xfs_dqid_t		id,
	uint			type,
L
Linus Torvalds 已提交
1460 1461 1462
	xfs_qcnt_t		nblks,
	xfs_qcnt_t		rtblks)
{
1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476
	struct xfs_mount	*mp = ip->i_mount;
	struct xfs_dquot	*dqp;
	int			error;

	error = xfs_qm_dqget(mp, ip, id, type,
			     XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
	if (error) {
		/*
		 * Shouldn't be able to turn off quotas here.
		 */
		ASSERT(error != ESRCH);
		ASSERT(error != ENOENT);
		return error;
	}
C
Christoph Hellwig 已提交
1477 1478 1479

	trace_xfs_dqadjust(dqp);

L
Linus Torvalds 已提交
1480 1481 1482 1483
	/*
	 * Adjust the inode count and the block count to reflect this inode's
	 * resource usage.
	 */
1484
	be64_add_cpu(&dqp->q_core.d_icount, 1);
L
Linus Torvalds 已提交
1485 1486
	dqp->q_res_icount++;
	if (nblks) {
1487
		be64_add_cpu(&dqp->q_core.d_bcount, nblks);
L
Linus Torvalds 已提交
1488 1489 1490
		dqp->q_res_bcount += nblks;
	}
	if (rtblks) {
1491
		be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
L
Linus Torvalds 已提交
1492 1493 1494 1495 1496
		dqp->q_res_rtbcount += rtblks;
	}

	/*
	 * Set default limits, adjust timers (since we changed usages)
1497 1498
	 *
	 * There are no timers for the default values set in the root dquot.
L
Linus Torvalds 已提交
1499
	 */
1500
	if (dqp->q_core.d_id) {
1501 1502
		xfs_qm_adjust_dqlimits(mp, &dqp->q_core);
		xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
L
Linus Torvalds 已提交
1503 1504 1505
	}

	dqp->dq_flags |= XFS_DQ_DIRTY;
1506 1507
	xfs_qm_dqput(dqp);
	return 0;
L
Linus Torvalds 已提交
1508 1509 1510 1511 1512 1513 1514 1515
}

STATIC int
xfs_qm_get_rtblks(
	xfs_inode_t	*ip,
	xfs_qcnt_t	*O_rtblks)
{
	xfs_filblks_t	rtblks;			/* total rt blks */
1516
	xfs_extnum_t	idx;			/* extent record index */
L
Linus Torvalds 已提交
1517 1518 1519 1520 1521 1522 1523 1524
	xfs_ifork_t	*ifp;			/* inode fork pointer */
	xfs_extnum_t	nextents;		/* number of extent entries */
	int		error;

	ASSERT(XFS_IS_REALTIME_INODE(ip));
	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
		if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1525
			return error;
L
Linus Torvalds 已提交
1526 1527
	}
	rtblks = 0;
1528
	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1529 1530
	for (idx = 0; idx < nextents; idx++)
		rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
L
Linus Torvalds 已提交
1531
	*O_rtblks = (xfs_qcnt_t)rtblks;
1532
	return 0;
L
Linus Torvalds 已提交
1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549
}

/*
 * callback routine supplied to bulkstat(). Given an inumber, find its
 * dquots and update them to account for resources taken by that inode.
 */
/* ARGSUSED */
STATIC int
xfs_qm_dqusage_adjust(
	xfs_mount_t	*mp,		/* mount point for filesystem */
	xfs_ino_t	ino,		/* inode number to get data for */
	void		__user *buffer,	/* not used */
	int		ubsize,		/* not used */
	int		*ubused,	/* not used */
	int		*res)		/* result code value */
{
	xfs_inode_t	*ip;
1550
	xfs_qcnt_t	nblks, rtblks = 0;
L
Linus Torvalds 已提交
1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569
	int		error;

	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	/*
	 * rootino must have its resources accounted for, not so with the quota
	 * inodes.
	 */
	if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
		*res = BULKSTAT_RV_NOTHING;
		return XFS_ERROR(EINVAL);
	}

	/*
	 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
	 * interface expects the inode to be exclusively locked because that's
	 * the case in all other instances. It's OK that we do this because
	 * quotacheck is done only at mount time.
	 */
1570 1571
	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
	if (error) {
L
Linus Torvalds 已提交
1572
		*res = BULKSTAT_RV_NOTHING;
1573
		return error;
L
Linus Torvalds 已提交
1574 1575
	}

1576
	ASSERT(ip->i_delayed_blks == 0);
L
Linus Torvalds 已提交
1577

1578
	if (XFS_IS_REALTIME_INODE(ip)) {
L
Linus Torvalds 已提交
1579 1580 1581
		/*
		 * Walk thru the extent list and count the realtime blocks.
		 */
1582 1583 1584
		error = xfs_qm_get_rtblks(ip, &rtblks);
		if (error)
			goto error0;
L
Linus Torvalds 已提交
1585 1586
	}

1587
	nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
L
Linus Torvalds 已提交
1588 1589 1590 1591 1592 1593 1594 1595 1596 1597

	/*
	 * Add the (disk blocks and inode) resources occupied by this
	 * inode to its dquots. We do this adjustment in the incore dquot,
	 * and also copy the changes to its buffer.
	 * We don't care about putting these changes in a transaction
	 * envelope because if we crash in the middle of a 'quotacheck'
	 * we have to start from the beginning anyway.
	 * Once we're done, we'll log all the dquot bufs.
	 *
1598
	 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
L
Linus Torvalds 已提交
1599 1600 1601
	 * and quotaoffs don't race. (Quotachecks happen at mount time only).
	 */
	if (XFS_IS_UQUOTA_ON(mp)) {
1602 1603 1604 1605
		error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
						   XFS_DQ_USER, nblks, rtblks);
		if (error)
			goto error0;
L
Linus Torvalds 已提交
1606
	}
1607 1608 1609 1610 1611 1612

	if (XFS_IS_GQUOTA_ON(mp)) {
		error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
						   XFS_DQ_GROUP, nblks, rtblks);
		if (error)
			goto error0;
L
Linus Torvalds 已提交
1613 1614
	}

1615
	if (XFS_IS_PQUOTA_ON(mp)) {
1616
		error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1617 1618 1619 1620 1621 1622 1623
						   XFS_DQ_PROJ, nblks, rtblks);
		if (error)
			goto error0;
	}

	xfs_iunlock(ip, XFS_ILOCK_EXCL);
	IRELE(ip);
L
Linus Torvalds 已提交
1624
	*res = BULKSTAT_RV_DIDONE;
1625
	return 0;
1626 1627 1628 1629 1630 1631

error0:
	xfs_iunlock(ip, XFS_ILOCK_EXCL);
	IRELE(ip);
	*res = BULKSTAT_RV_GIVEUP;
	return error;
L
Linus Torvalds 已提交
1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652
}

/*
 * Walk thru all the filesystem inodes and construct a consistent view
 * of the disk quota world. If the quotacheck fails, disable quotas.
 */
int
xfs_qm_quotacheck(
	xfs_mount_t	*mp)
{
	int		done, count, error;
	xfs_ino_t	lastino;
	size_t		structsz;
	xfs_inode_t	*uip, *gip;
	uint		flags;

	count = INT_MAX;
	structsz = 1;
	lastino = 0;
	flags = 0;

1653
	ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
L
Linus Torvalds 已提交
1654 1655 1656 1657 1658 1659
	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	/*
	 * There should be no cached dquots. The (simplistic) quotacheck
	 * algorithm doesn't like that.
	 */
1660
	ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
L
Linus Torvalds 已提交
1661

1662
	xfs_notice(mp, "Quotacheck needed: Please wait.");
L
Linus Torvalds 已提交
1663 1664

	/*
1665
	 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
L
Linus Torvalds 已提交
1666 1667 1668
	 * their counters to zero. We need a clean slate.
	 * We don't log our changes till later.
	 */
1669 1670 1671 1672
	uip = mp->m_quotainfo->qi_uquotaip;
	if (uip) {
		error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
		if (error)
L
Linus Torvalds 已提交
1673 1674 1675 1676
			goto error_return;
		flags |= XFS_UQUOTA_CHKD;
	}

1677 1678 1679 1680 1681
	gip = mp->m_quotainfo->qi_gquotaip;
	if (gip) {
		error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
					XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
		if (error)
L
Linus Torvalds 已提交
1682
			goto error_return;
1683
		flags |= XFS_OQUOTA_CHKD;
L
Linus Torvalds 已提交
1684 1685 1686 1687 1688 1689 1690
	}

	do {
		/*
		 * Iterate thru all the inodes in the file system,
		 * adjusting the corresponding dquot counters in core.
		 */
1691 1692 1693 1694
		error = xfs_bulkstat(mp, &lastino, &count,
				     xfs_qm_dqusage_adjust,
				     structsz, NULL, &done);
		if (error)
L
Linus Torvalds 已提交
1695 1696
			break;

1697
	} while (!done);
L
Linus Torvalds 已提交
1698

1699 1700 1701 1702 1703 1704
	/*
	 * We've made all the changes that we need to make incore.
	 * Flush them down to disk buffers if everything was updated
	 * successfully.
	 */
	if (!error)
1705
		error = xfs_qm_dqflush_all(mp, 0);
1706

L
Linus Torvalds 已提交
1707 1708 1709 1710 1711 1712 1713 1714
	/*
	 * We can get this error if we couldn't do a dquot allocation inside
	 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
	 * dirty dquots that might be cached, we just want to get rid of them
	 * and turn quotaoff. The dquots won't be attached to any of the inodes
	 * at this point (because we intentionally didn't in dqget_noattach).
	 */
	if (error) {
1715
		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
L
Linus Torvalds 已提交
1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732
		goto error_return;
	}

	/*
	 * We didn't log anything, because if we crashed, we'll have to
	 * start the quotacheck from scratch anyway. However, we must make
	 * sure that our dquot changes are secure before we put the
	 * quotacheck'd stamp on the superblock. So, here we do a synchronous
	 * flush.
	 */
	XFS_bflush(mp->m_ddev_targp);

	/*
	 * If one type of quotas is off, then it will lose its
	 * quotachecked status, since we won't be doing accounting for
	 * that type anymore.
	 */
1733
	mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
L
Linus Torvalds 已提交
1734 1735
	mp->m_qflags |= flags;

1736
	xfs_qm_dquot_list_print(mp);
L
Linus Torvalds 已提交
1737 1738 1739

 error_return:
	if (error) {
1740 1741 1742
		xfs_warn(mp,
	"Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
			error);
L
Linus Torvalds 已提交
1743 1744 1745 1746 1747 1748
		/*
		 * We must turn off quotas.
		 */
		ASSERT(mp->m_quotainfo != NULL);
		ASSERT(xfs_Gqm != NULL);
		xfs_qm_destroy_quotainfo(mp);
1749
		if (xfs_mount_reset_sbqflags(mp)) {
1750 1751
			xfs_warn(mp,
				"Quotacheck: Failed to reset quota flags.");
1752
		}
1753 1754
	} else
		xfs_notice(mp, "Quotacheck: Done.");
L
Linus Torvalds 已提交
1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778
	return (error);
}

/*
 * This is called after the superblock has been read in and we're ready to
 * iget the quota inodes.
 */
STATIC int
xfs_qm_init_quotainos(
	xfs_mount_t	*mp)
{
	xfs_inode_t	*uip, *gip;
	int		error;
	__int64_t	sbflags;
	uint		flags;

	ASSERT(mp->m_quotainfo);
	uip = gip = NULL;
	sbflags = 0;
	flags = 0;

	/*
	 * Get the uquota and gquota inodes
	 */
1779
	if (xfs_sb_version_hasquota(&mp->m_sb)) {
L
Linus Torvalds 已提交
1780 1781 1782 1783
		if (XFS_IS_UQUOTA_ON(mp) &&
		    mp->m_sb.sb_uquotino != NULLFSINO) {
			ASSERT(mp->m_sb.sb_uquotino > 0);
			if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1784
					     0, 0, &uip)))
L
Linus Torvalds 已提交
1785 1786
				return XFS_ERROR(error);
		}
1787
		if (XFS_IS_OQUOTA_ON(mp) &&
L
Linus Torvalds 已提交
1788 1789 1790
		    mp->m_sb.sb_gquotino != NULLFSINO) {
			ASSERT(mp->m_sb.sb_gquotino > 0);
			if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1791
					     0, 0, &gip))) {
L
Linus Torvalds 已提交
1792
				if (uip)
1793
					IRELE(uip);
L
Linus Torvalds 已提交
1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816
				return XFS_ERROR(error);
			}
		}
	} else {
		flags |= XFS_QMOPT_SBVERSION;
		sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
			    XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
	}

	/*
	 * Create the two inodes, if they don't exist already. The changes
	 * made above will get added to a transaction and logged in one of
	 * the qino_alloc calls below.  If the device is readonly,
	 * temporarily switch to read-write to do this.
	 */
	if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
		if ((error = xfs_qm_qino_alloc(mp, &uip,
					      sbflags | XFS_SB_UQUOTINO,
					      flags | XFS_QMOPT_UQUOTA)))
			return XFS_ERROR(error);

		flags &= ~XFS_QMOPT_SBVERSION;
	}
1817 1818 1819 1820 1821 1822
	if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
		flags |= (XFS_IS_GQUOTA_ON(mp) ?
				XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
		error = xfs_qm_qino_alloc(mp, &gip,
					  sbflags | XFS_SB_GQUOTINO, flags);
		if (error) {
L
Linus Torvalds 已提交
1823
			if (uip)
1824
				IRELE(uip);
L
Linus Torvalds 已提交
1825 1826 1827 1828 1829

			return XFS_ERROR(error);
		}
	}

1830 1831
	mp->m_quotainfo->qi_uquotaip = uip;
	mp->m_quotainfo->qi_gquotaip = gip;
L
Linus Torvalds 已提交
1832

1833
	return 0;
L
Linus Torvalds 已提交
1834 1835 1836
}


1837

L
Linus Torvalds 已提交
1838
/*
1839 1840
 * Just pop the least recently used dquot off the freelist and
 * recycle it. The returned dquot is locked.
L
Linus Torvalds 已提交
1841
 */
1842 1843
STATIC xfs_dquot_t *
xfs_qm_dqreclaim_one(void)
L
Linus Torvalds 已提交
1844
{
1845 1846
	xfs_dquot_t	*dqpout;
	xfs_dquot_t	*dqp;
L
Linus Torvalds 已提交
1847
	int		restarts;
D
Dave Chinner 已提交
1848
	int		startagain;
L
Linus Torvalds 已提交
1849 1850

	restarts = 0;
1851
	dqpout = NULL;
L
Linus Torvalds 已提交
1852

1853
	/* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
D
Dave Chinner 已提交
1854 1855
again:
	startagain = 0;
1856
	mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
L
Linus Torvalds 已提交
1857

1858
	list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
1859
		struct xfs_mount *mp = dqp->q_mount;
L
Linus Torvalds 已提交
1860 1861 1862 1863
		xfs_dqlock(dqp);

		/*
		 * We are racing with dqlookup here. Naturally we don't
1864 1865 1866
		 * want to reclaim a dquot that lookup wants. We release the
		 * freelist lock and start over, so that lookup will grab
		 * both the dquot and the freelistlock.
L
Linus Torvalds 已提交
1867 1868
		 */
		if (dqp->dq_flags & XFS_DQ_WANT) {
1869 1870 1871
			ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));

			trace_xfs_dqreclaim_want(dqp);
L
Linus Torvalds 已提交
1872
			XQM_STATS_INC(xqmstats.xs_qm_dqwants);
D
Dave Chinner 已提交
1873 1874 1875
			restarts++;
			startagain = 1;
			goto dqunlock;
L
Linus Torvalds 已提交
1876 1877 1878 1879 1880 1881 1882 1883
		}

		/*
		 * If the dquot is inactive, we are assured that it is
		 * not on the mplist or the hashlist, and that makes our
		 * life easier.
		 */
		if (dqp->dq_flags & XFS_DQ_INACTIVE) {
1884
			ASSERT(mp == NULL);
L
Linus Torvalds 已提交
1885
			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
1886
			ASSERT(list_empty(&dqp->q_hashlist));
1887
			ASSERT(list_empty(&dqp->q_mplist));
1888 1889
			list_del_init(&dqp->q_freelist);
			xfs_Gqm->qm_dqfrlist_cnt--;
1890
			dqpout = dqp;
L
Linus Torvalds 已提交
1891
			XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
D
Dave Chinner 已提交
1892
			goto dqunlock;
L
Linus Torvalds 已提交
1893 1894
		}

1895
		ASSERT(dqp->q_hash);
1896
		ASSERT(!list_empty(&dqp->q_mplist));
1897

L
Linus Torvalds 已提交
1898
		/*
D
Dave Chinner 已提交
1899 1900
		 * Try to grab the flush lock. If this dquot is in the process
		 * of getting flushed to disk, we don't want to reclaim it.
L
Linus Torvalds 已提交
1901
		 */
D
Dave Chinner 已提交
1902 1903
		if (!xfs_dqflock_nowait(dqp))
			goto dqunlock;
L
Linus Torvalds 已提交
1904 1905 1906 1907 1908 1909 1910 1911

		/*
		 * We have the flush lock so we know that this is not in the
		 * process of being flushed. So, if this is dirty, flush it
		 * DELWRI so that we don't get a freelist infested with
		 * dirty dquots.
		 */
		if (XFS_DQ_IS_DIRTY(dqp)) {
1912
			int	error;
C
Christoph Hellwig 已提交
1913

1914
			trace_xfs_dqreclaim_dirty(dqp);
C
Christoph Hellwig 已提交
1915

L
Linus Torvalds 已提交
1916 1917
			/*
			 * We flush it delayed write, so don't bother
1918
			 * releasing the freelist lock.
L
Linus Torvalds 已提交
1919
			 */
1920
			error = xfs_qm_dqflush(dqp, 0);
1921
			if (error) {
1922 1923
				xfs_warn(mp, "%s: dquot %p flush failed",
					__func__, dqp);
1924
			}
D
Dave Chinner 已提交
1925
			goto dqunlock;
L
Linus Torvalds 已提交
1926
		}
1927

L
Linus Torvalds 已提交
1928 1929 1930 1931 1932 1933 1934
		/*
		 * We're trying to get the hashlock out of order. This races
		 * with dqlookup; so, we giveup and goto the next dquot if
		 * we couldn't get the hashlock. This way, we won't starve
		 * a dqlookup process that holds the hashlock that is
		 * waiting for the freelist lock.
		 */
1935
		if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
1936 1937
			restarts++;
			goto dqfunlock;
L
Linus Torvalds 已提交
1938
		}
1939

L
Linus Torvalds 已提交
1940 1941 1942 1943 1944
		/*
		 * This races with dquot allocation code as well as dqflush_all
		 * and reclaim code. So, if we failed to grab the mplist lock,
		 * giveup everything and start over.
		 */
1945
		if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) {
1946
			restarts++;
D
Dave Chinner 已提交
1947 1948
			startagain = 1;
			goto qhunlock;
L
Linus Torvalds 已提交
1949
		}
C
Christoph Hellwig 已提交
1950

L
Linus Torvalds 已提交
1951
		ASSERT(dqp->q_nrefs == 0);
1952 1953 1954
		list_del_init(&dqp->q_mplist);
		mp->m_quotainfo->qi_dquots--;
		mp->m_quotainfo->qi_dqreclaims++;
1955 1956
		list_del_init(&dqp->q_hashlist);
		dqp->q_hash->qh_version++;
1957 1958
		list_del_init(&dqp->q_freelist);
		xfs_Gqm->qm_dqfrlist_cnt--;
1959 1960
		dqpout = dqp;
		mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
D
Dave Chinner 已提交
1961
qhunlock:
1962 1963 1964
		mutex_unlock(&dqp->q_hash->qh_lock);
dqfunlock:
		xfs_dqfunlock(dqp);
D
Dave Chinner 已提交
1965
dqunlock:
L
Linus Torvalds 已提交
1966
		xfs_dqunlock(dqp);
1967 1968 1969
		if (dqpout)
			break;
		if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
D
Dave Chinner 已提交
1970 1971 1972 1973 1974
			break;
		if (startagain) {
			mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
			goto again;
		}
L
Linus Torvalds 已提交
1975
	}
1976
	mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1977
	return dqpout;
L
Linus Torvalds 已提交
1978 1979
}

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
/*
 * Traverse the freelist of dquots and attempt to reclaim a maximum of
 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
 * favor the lookup function ...
 */
STATIC int
xfs_qm_shake_freelist(
	int	howmany)
{
	int		nreclaimed = 0;
	xfs_dquot_t	*dqp;

	if (howmany <= 0)
		return 0;

	while (nreclaimed < howmany) {
		dqp = xfs_qm_dqreclaim_one();
		if (!dqp)
			return nreclaimed;
		xfs_qm_dqdestroy(dqp);
		nreclaimed++;
	}
	return nreclaimed;
}
L
Linus Torvalds 已提交
2004 2005 2006 2007 2008 2009

/*
 * The kmem_shake interface is invoked when memory is running low.
 */
/* ARGSUSED */
STATIC int
2010 2011 2012 2013
xfs_qm_shake(
	struct shrinker	*shrink,
	int		nr_to_scan,
	gfp_t		gfp_mask)
L
Linus Torvalds 已提交
2014 2015 2016 2017
{
	int	ndqused, nfree, n;

	if (!kmem_shake_allow(gfp_mask))
2018
		return 0;
L
Linus Torvalds 已提交
2019
	if (!xfs_Gqm)
2020
		return 0;
L
Linus Torvalds 已提交
2021

2022
	nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
L
Linus Torvalds 已提交
2023 2024 2025 2026 2027 2028
	/* incore dquots in all f/s's */
	ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;

	ASSERT(ndqused >= 0);

	if (nfree <= ndqused && nfree < ndquot)
2029
		return 0;
L
Linus Torvalds 已提交
2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069

	ndqused *= xfs_Gqm->qm_dqfree_ratio;	/* target # of free dquots */
	n = nfree - ndqused - ndquot;		/* # over target */

	return xfs_qm_shake_freelist(MAX(nfree, n));
}


/*------------------------------------------------------------------*/

/*
 * Return a new incore dquot. Depending on the number of
 * dquots in the system, we either allocate a new one on the kernel heap,
 * or reclaim a free one.
 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
 * to reclaim an existing one from the freelist.
 */
boolean_t
xfs_qm_dqalloc_incore(
	xfs_dquot_t **O_dqpp)
{
	xfs_dquot_t	*dqp;

	/*
	 * Check against high water mark to see if we want to pop
	 * a nincompoop dquot off the freelist.
	 */
	if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
		/*
		 * Try to recycle a dquot from the freelist.
		 */
		if ((dqp = xfs_qm_dqreclaim_one())) {
			XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
			/*
			 * Just zero the core here. The rest will get
			 * reinitialized by caller. XXX we shouldn't even
			 * do this zero ...
			 */
			memset(&dqp->q_core, 0, sizeof(dqp->q_core));
			*O_dqpp = dqp;
2070
			return B_FALSE;
L
Linus Torvalds 已提交
2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082
		}
		XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
	}

	/*
	 * Allocate a brand new dquot on the kernel heap and return it
	 * to the caller to initialize.
	 */
	ASSERT(xfs_Gqm->qm_dqzone != NULL);
	*O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
	atomic_inc(&xfs_Gqm->qm_totaldquots);

2083
	return B_TRUE;
L
Linus Torvalds 已提交
2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099
}


/*
 * Start a transaction and write the incore superblock changes to
 * disk. flags parameter indicates which fields have changed.
 */
int
xfs_qm_write_sb_changes(
	xfs_mount_t	*mp,
	__int64_t	flags)
{
	xfs_trans_t	*tp;
	int		error;

#ifdef QUOTADEBUG
2100
	xfs_notice(mp, "Writing superblock quota changes");
L
Linus Torvalds 已提交
2101 2102 2103 2104 2105 2106 2107
#endif
	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
	if ((error = xfs_trans_reserve(tp, 0,
				      mp->m_sb.sb_sectsize + 128, 0,
				      0,
				      XFS_DEFAULT_LOG_COUNT))) {
		xfs_trans_cancel(tp, 0);
2108
		return error;
L
Linus Torvalds 已提交
2109 2110 2111
	}

	xfs_mod_sb(tp, flags);
2112
	error = xfs_trans_commit(tp, 0);
L
Linus Torvalds 已提交
2113

2114
	return error;
L
Linus Torvalds 已提交
2115 2116 2117 2118 2119 2120 2121
}


/* --------------- utility functions for vnodeops ---------------- */


/*
C
Christoph Hellwig 已提交
2122
 * Given an inode, a uid, gid and prid make sure that we have
L
Linus Torvalds 已提交
2123 2124 2125 2126 2127 2128 2129 2130 2131 2132
 * allocated relevant dquot(s) on disk, and that we won't exceed inode
 * quotas by creating this file.
 * This also attaches dquot(s) to the given inode after locking it,
 * and returns the dquots corresponding to the uid and/or gid.
 *
 * in	: inode (unlocked)
 * out	: udquot, gdquot with references taken and unlocked
 */
int
xfs_qm_vop_dqalloc(
C
Christoph Hellwig 已提交
2133 2134 2135 2136 2137 2138 2139
	struct xfs_inode	*ip,
	uid_t			uid,
	gid_t			gid,
	prid_t			prid,
	uint			flags,
	struct xfs_dquot	**O_udqpp,
	struct xfs_dquot	**O_gdqpp)
L
Linus Torvalds 已提交
2140
{
C
Christoph Hellwig 已提交
2141 2142 2143 2144
	struct xfs_mount	*mp = ip->i_mount;
	struct xfs_dquot	*uq, *gq;
	int			error;
	uint			lockflags;
L
Linus Torvalds 已提交
2145

C
Christoph Hellwig 已提交
2146
	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
L
Linus Torvalds 已提交
2147 2148 2149 2150 2151
		return 0;

	lockflags = XFS_ILOCK_EXCL;
	xfs_ilock(ip, lockflags);

2152
	if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
L
Linus Torvalds 已提交
2153 2154 2155 2156 2157 2158 2159
		gid = ip->i_d.di_gid;

	/*
	 * Attach the dquot(s) to this inode, doing a dquot allocation
	 * if necessary. The dquot(s) will not be locked.
	 */
	if (XFS_NOT_DQATTACHED(mp, ip)) {
C
Christoph Hellwig 已提交
2160 2161
		error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
		if (error) {
L
Linus Torvalds 已提交
2162
			xfs_iunlock(ip, lockflags);
2163
			return error;
L
Linus Torvalds 已提交
2164 2165 2166 2167
		}
	}

	uq = gq = NULL;
2168
	if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
L
Linus Torvalds 已提交
2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185
		if (ip->i_d.di_uid != uid) {
			/*
			 * What we need is the dquot that has this uid, and
			 * if we send the inode to dqget, the uid of the inode
			 * takes priority over what's sent in the uid argument.
			 * We must unlock inode here before calling dqget if
			 * we're not sending the inode, because otherwise
			 * we'll deadlock by doing trans_reserve while
			 * holding ilock.
			 */
			xfs_iunlock(ip, lockflags);
			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
						 XFS_DQ_USER,
						 XFS_QMOPT_DQALLOC |
						 XFS_QMOPT_DOWARN,
						 &uq))) {
				ASSERT(error != ENOENT);
2186
				return error;
L
Linus Torvalds 已提交
2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205
			}
			/*
			 * Get the ilock in the right order.
			 */
			xfs_dqunlock(uq);
			lockflags = XFS_ILOCK_SHARED;
			xfs_ilock(ip, lockflags);
		} else {
			/*
			 * Take an extra reference, because we'll return
			 * this to caller
			 */
			ASSERT(ip->i_udquot);
			uq = ip->i_udquot;
			xfs_dqlock(uq);
			XFS_DQHOLD(uq);
			xfs_dqunlock(uq);
		}
	}
2206
	if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
L
Linus Torvalds 已提交
2207 2208 2209 2210 2211 2212 2213 2214 2215 2216
		if (ip->i_d.di_gid != gid) {
			xfs_iunlock(ip, lockflags);
			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
						 XFS_DQ_GROUP,
						 XFS_QMOPT_DQALLOC |
						 XFS_QMOPT_DOWARN,
						 &gq))) {
				if (uq)
					xfs_qm_dqrele(uq);
				ASSERT(error != ENOENT);
2217
				return error;
L
Linus Torvalds 已提交
2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228
			}
			xfs_dqunlock(gq);
			lockflags = XFS_ILOCK_SHARED;
			xfs_ilock(ip, lockflags);
		} else {
			ASSERT(ip->i_gdquot);
			gq = ip->i_gdquot;
			xfs_dqlock(gq);
			XFS_DQHOLD(gq);
			xfs_dqunlock(gq);
		}
2229
	} else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2230
		if (xfs_get_projid(ip) != prid) {
2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251
			xfs_iunlock(ip, lockflags);
			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
						 XFS_DQ_PROJ,
						 XFS_QMOPT_DQALLOC |
						 XFS_QMOPT_DOWARN,
						 &gq))) {
				if (uq)
					xfs_qm_dqrele(uq);
				ASSERT(error != ENOENT);
				return (error);
			}
			xfs_dqunlock(gq);
			lockflags = XFS_ILOCK_SHARED;
			xfs_ilock(ip, lockflags);
		} else {
			ASSERT(ip->i_gdquot);
			gq = ip->i_gdquot;
			xfs_dqlock(gq);
			XFS_DQHOLD(gq);
			xfs_dqunlock(gq);
		}
L
Linus Torvalds 已提交
2252 2253
	}
	if (uq)
C
Christoph Hellwig 已提交
2254
		trace_xfs_dquot_dqalloc(ip);
L
Linus Torvalds 已提交
2255 2256 2257 2258 2259 2260 2261 2262 2263 2264

	xfs_iunlock(ip, lockflags);
	if (O_udqpp)
		*O_udqpp = uq;
	else if (uq)
		xfs_qm_dqrele(uq);
	if (O_gdqpp)
		*O_gdqpp = gq;
	else if (gq)
		xfs_qm_dqrele(gq);
2265
	return 0;
L
Linus Torvalds 已提交
2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279
}

/*
 * Actually transfer ownership, and do dquot modifications.
 * These were already reserved.
 */
xfs_dquot_t *
xfs_qm_vop_chown(
	xfs_trans_t	*tp,
	xfs_inode_t	*ip,
	xfs_dquot_t	**IO_olddq,
	xfs_dquot_t	*newdq)
{
	xfs_dquot_t	*prevdq;
2280 2281 2282
	uint		bfield = XFS_IS_REALTIME_INODE(ip) ?
				 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;

C
Christoph Hellwig 已提交
2283

C
Christoph Hellwig 已提交
2284
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
2285 2286 2287 2288 2289 2290 2291
	ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));

	/* old dquot */
	prevdq = *IO_olddq;
	ASSERT(prevdq);
	ASSERT(prevdq != newdq);

2292 2293
	xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
	xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
L
Linus Torvalds 已提交
2294 2295

	/* the sparkling new dquot */
2296 2297
	xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
	xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
L
Linus Torvalds 已提交
2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308

	/*
	 * Take an extra reference, because the inode
	 * is going to keep this dquot pointer even
	 * after the trans_commit.
	 */
	xfs_dqlock(newdq);
	XFS_DQHOLD(newdq);
	xfs_dqunlock(newdq);
	*IO_olddq = newdq;

2309
	return prevdq;
L
Linus Torvalds 已提交
2310 2311 2312
}

/*
2313
 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
L
Linus Torvalds 已提交
2314 2315 2316 2317 2318 2319 2320 2321 2322
 */
int
xfs_qm_vop_chown_reserve(
	xfs_trans_t	*tp,
	xfs_inode_t	*ip,
	xfs_dquot_t	*udqp,
	xfs_dquot_t	*gdqp,
	uint		flags)
{
C
Christoph Hellwig 已提交
2323
	xfs_mount_t	*mp = ip->i_mount;
2324
	uint		delblks, blkflags, prjflags = 0;
L
Linus Torvalds 已提交
2325
	xfs_dquot_t	*unresudq, *unresgdq, *delblksudq, *delblksgdq;
C
Christoph Hellwig 已提交
2326 2327
	int		error;

L
Linus Torvalds 已提交
2328

C
Christoph Hellwig 已提交
2329
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
L
Linus Torvalds 已提交
2330 2331 2332 2333
	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	delblks = ip->i_delayed_blks;
	delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2334 2335
	blkflags = XFS_IS_REALTIME_INODE(ip) ?
			XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
L
Linus Torvalds 已提交
2336 2337

	if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2338
	    ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
L
Linus Torvalds 已提交
2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349
		delblksudq = udqp;
		/*
		 * If there are delayed allocation blocks, then we have to
		 * unreserve those from the old dquot, and add them to the
		 * new dquot.
		 */
		if (delblks) {
			ASSERT(ip->i_udquot);
			unresudq = ip->i_udquot;
		}
	}
2350
	if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2351
		if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2352
		     xfs_get_projid(ip) != be32_to_cpu(gdqp->q_core.d_id))
2353 2354 2355 2356 2357
			prjflags = XFS_QMOPT_ENOSPC;

		if (prjflags ||
		    (XFS_IS_GQUOTA_ON(ip->i_mount) &&
		     ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
2358 2359 2360 2361 2362
			delblksgdq = gdqp;
			if (delblks) {
				ASSERT(ip->i_gdquot);
				unresgdq = ip->i_gdquot;
			}
L
Linus Torvalds 已提交
2363 2364 2365 2366 2367
		}
	}

	if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
				delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2368
				flags | blkflags | prjflags)))
L
Linus Torvalds 已提交
2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384
		return (error);

	/*
	 * Do the delayed blks reservations/unreservations now. Since, these
	 * are done without the help of a transaction, if a reservation fails
	 * its previous reservations won't be automatically undone by trans
	 * code. So, we have to do it manually here.
	 */
	if (delblks) {
		/*
		 * Do the reservations first. Unreservation can't fail.
		 */
		ASSERT(delblksudq || delblksgdq);
		ASSERT(unresudq || unresgdq);
		if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
				delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2385
				flags | blkflags | prjflags)))
L
Linus Torvalds 已提交
2386 2387 2388
			return (error);
		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
				unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2389
				blkflags);
L
Linus Torvalds 已提交
2390 2391 2392 2393 2394 2395 2396
	}

	return (0);
}

int
xfs_qm_vop_rename_dqattach(
C
Christoph Hellwig 已提交
2397
	struct xfs_inode	**i_tab)
L
Linus Torvalds 已提交
2398
{
C
Christoph Hellwig 已提交
2399 2400
	struct xfs_mount	*mp = i_tab[0]->i_mount;
	int			i;
L
Linus Torvalds 已提交
2401

C
Christoph Hellwig 已提交
2402
	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2403
		return 0;
L
Linus Torvalds 已提交
2404

C
Christoph Hellwig 已提交
2405 2406 2407 2408
	for (i = 0; (i < 4 && i_tab[i]); i++) {
		struct xfs_inode	*ip = i_tab[i];
		int			error;

L
Linus Torvalds 已提交
2409 2410 2411
		/*
		 * Watch out for duplicate entries in the table.
		 */
C
Christoph Hellwig 已提交
2412 2413
		if (i == 0 || ip != i_tab[i-1]) {
			if (XFS_NOT_DQATTACHED(mp, ip)) {
L
Linus Torvalds 已提交
2414 2415
				error = xfs_qm_dqattach(ip, 0);
				if (error)
2416
					return error;
L
Linus Torvalds 已提交
2417 2418 2419
			}
		}
	}
2420
	return 0;
L
Linus Torvalds 已提交
2421 2422 2423
}

void
C
Christoph Hellwig 已提交
2424 2425 2426 2427 2428
xfs_qm_vop_create_dqattach(
	struct xfs_trans	*tp,
	struct xfs_inode	*ip,
	struct xfs_dquot	*udqp,
	struct xfs_dquot	*gdqp)
L
Linus Torvalds 已提交
2429
{
C
Christoph Hellwig 已提交
2430 2431 2432
	struct xfs_mount	*mp = tp->t_mountp;

	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
L
Linus Torvalds 已提交
2433 2434
		return;

C
Christoph Hellwig 已提交
2435
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
C
Christoph Hellwig 已提交
2436
	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
L
Linus Torvalds 已提交
2437 2438 2439 2440 2441 2442 2443

	if (udqp) {
		xfs_dqlock(udqp);
		XFS_DQHOLD(udqp);
		xfs_dqunlock(udqp);
		ASSERT(ip->i_udquot == NULL);
		ip->i_udquot = udqp;
C
Christoph Hellwig 已提交
2444
		ASSERT(XFS_IS_UQUOTA_ON(mp));
2445
		ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
L
Linus Torvalds 已提交
2446 2447 2448 2449 2450 2451 2452 2453
		xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
	}
	if (gdqp) {
		xfs_dqlock(gdqp);
		XFS_DQHOLD(gdqp);
		xfs_dqunlock(gdqp);
		ASSERT(ip->i_gdquot == NULL);
		ip->i_gdquot = gdqp;
C
Christoph Hellwig 已提交
2454 2455
		ASSERT(XFS_IS_OQUOTA_ON(mp));
		ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2456
			ip->i_d.di_gid : xfs_get_projid(ip)) ==
2457
				be32_to_cpu(gdqp->q_core.d_id));
L
Linus Torvalds 已提交
2458 2459 2460 2461
		xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
	}
}