xfs_qm.c 63.4 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
#include "xfs_trans.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_dir2.h"
#include "xfs_alloc.h"
#include "xfs_dmapi.h"
#include "xfs_quota.h"
#include "xfs_mount.h"
#include "xfs_bmap_btree.h"
32
#include "xfs_alloc_btree.h"
L
Linus Torvalds 已提交
33 34
#include "xfs_ialloc_btree.h"
#include "xfs_dir2_sf.h"
35
#include "xfs_attr_sf.h"
L
Linus Torvalds 已提交
36 37
#include "xfs_dinode.h"
#include "xfs_inode.h"
38 39 40
#include "xfs_btree.h"
#include "xfs_ialloc.h"
#include "xfs_itable.h"
L
Linus Torvalds 已提交
41 42
#include "xfs_rtalloc.h"
#include "xfs_error.h"
43
#include "xfs_bmap.h"
L
Linus Torvalds 已提交
44 45 46 47 48 49
#include "xfs_rw.h"
#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
50
#include "xfs_trace.h"
L
Linus Torvalds 已提交
51 52 53 54 55 56 57

/*
 * 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
58
struct mutex	xfs_Gqm_lock;
L
Linus Torvalds 已提交
59
struct xfs_qm	*xfs_Gqm;
60
uint		ndquot;
L
Linus Torvalds 已提交
61 62 63 64

kmem_zone_t	*qm_dqzone;
kmem_zone_t	*qm_dqtrxzone;

65
static cred_t	xfs_zerocr;
66

L
Linus Torvalds 已提交
67 68 69
STATIC void	xfs_qm_list_init(xfs_dqlist_t *, char *, int);
STATIC void	xfs_qm_list_destroy(xfs_dqlist_t *);

70 71 72
STATIC void	xfs_qm_freelist_init(xfs_frlist_t *);
STATIC void	xfs_qm_freelist_destroy(xfs_frlist_t *);

L
Linus Torvalds 已提交
73
STATIC int	xfs_qm_init_quotainos(xfs_mount_t *);
74
STATIC int	xfs_qm_init_quotainfo(xfs_mount_t *);
A
Al Viro 已提交
75
STATIC int	xfs_qm_shake(int, gfp_t);
L
Linus Torvalds 已提交
76

77 78 79 80 81
static struct shrinker xfs_qm_shaker = {
	.shrink = xfs_qm_shake,
	.seeks = DEFAULT_SEEKS,
};

L
Linus Torvalds 已提交
82
#ifdef DEBUG
C
Christoph Hellwig 已提交
83
extern struct mutex	qcheck_lock;
L
Linus Torvalds 已提交
84 85 86
#endif

#ifdef QUOTADEBUG
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102
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) {
		cmn_err(CE_DEBUG, "   %d. \"%d (%s)\"   "
				  "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
103 104
}
#else
105
static void xfs_qm_dquot_list_print(struct xfs_mount *mp) { }
L
Linus Torvalds 已提交
106 107 108 109 110 111 112 113 114
#endif

/*
 * Initialize the XQM structure.
 * Note that there is not one quota manager per file system.
 */
STATIC struct xfs_qm *
xfs_Gqm_init(void)
{
115 116
	xfs_dqhash_t	*udqhash, *gdqhash;
	xfs_qm_t	*xqm;
117 118
	size_t		hsize;
	uint		i;
L
Linus Torvalds 已提交
119 120 121 122

	/*
	 * Initialize the dquot hash tables.
	 */
123
	udqhash = kmem_zalloc_greedy(&hsize,
124
				     XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
125 126 127 128 129
				     XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
	if (!udqhash)
		goto out;

	gdqhash = kmem_zalloc_large(hsize);
J
Julia Lawall 已提交
130
	if (!gdqhash)
131 132
		goto out_free_udqhash;

133
	hsize /= sizeof(xfs_dqhash_t);
134
	ndquot = hsize << 8;
L
Linus Torvalds 已提交
135

136 137 138 139
	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
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
	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
	 */
	xfs_qm_freelist_init(&(xqm->qm_dqfreelist));

	/*
	 * 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;

163
	register_shrinker(&xfs_qm_shaker);
L
Linus Torvalds 已提交
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178

	/*
	 * 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
179
	mutex_init(&qcheck_lock);
L
Linus Torvalds 已提交
180 181
#endif
	return xqm;
182 183 184 185 186

 out_free_udqhash:
	kmem_free_large(udqhash);
 out:
	return NULL;
L
Linus Torvalds 已提交
187 188 189 190 191
}

/*
 * Destroy the global quota manager when its reference count goes to zero.
 */
192
STATIC void
L
Linus Torvalds 已提交
193 194 195 196 197 198 199
xfs_qm_destroy(
	struct xfs_qm	*xqm)
{
	int		hsize, i;

	ASSERT(xqm != NULL);
	ASSERT(xqm->qm_nrefs == 0);
200
	unregister_shrinker(&xfs_qm_shaker);
L
Linus Torvalds 已提交
201 202 203 204 205
	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]));
	}
206 207
	kmem_free_large(xqm->qm_usr_dqhtable);
	kmem_free_large(xqm->qm_grp_dqhtable);
L
Linus Torvalds 已提交
208 209 210 211 212 213 214
	xqm->qm_usr_dqhtable = NULL;
	xqm->qm_grp_dqhtable = NULL;
	xqm->qm_dqhashmask = 0;
	xfs_qm_freelist_destroy(&(xqm->qm_dqfreelist));
#ifdef DEBUG
	mutex_destroy(&qcheck_lock);
#endif
215
	kmem_free(xqm);
L
Linus Torvalds 已提交
216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
}

/*
 * 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.
	 */
234
	mutex_lock(&xfs_Gqm_lock);
L
Linus Torvalds 已提交
235

236
	if (!xfs_Gqm) {
L
Linus Torvalds 已提交
237
		xfs_Gqm = xfs_Gqm_init();
238 239 240 241
		if (!xfs_Gqm)
			return ENOMEM;
	}

L
Linus Torvalds 已提交
242 243 244 245 246
	/*
	 * We can keep a list of all filesystems with quotas mounted for
	 * debugging and statistical purposes, but ...
	 * Just take a reference and get out.
	 */
247 248
	xfs_Gqm->qm_nrefs++;
	mutex_unlock(&xfs_Gqm_lock);
L
Linus Torvalds 已提交
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280

	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)
{
	xfs_dquot_t	*dqp, *nextdqp;

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

	/*
	 * Go thru the freelist and destroy all inactive dquots.
	 */
	xfs_qm_freelist_lock(xfs_Gqm);

	for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
	     dqp != (xfs_dquot_t *)&(xfs_Gqm->qm_dqfreelist); ) {
		xfs_dqlock(dqp);
		nextdqp = dqp->dq_flnext;
		if (dqp->dq_flags & XFS_DQ_INACTIVE) {
			ASSERT(dqp->q_mount == NULL);
			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
			ASSERT(dqp->HL_PREVP == NULL);
281
			ASSERT(list_empty(&dqp->q_mplist));
L
Linus Torvalds 已提交
282 283 284 285 286 287 288 289 290 291 292 293 294 295
			XQM_FREELIST_REMOVE(dqp);
			xfs_dqunlock(dqp);
			xfs_qm_dqdestroy(dqp);
		} else {
			xfs_dqunlock(dqp);
		}
		dqp = nextdqp;
	}
	xfs_qm_freelist_unlock(xfs_Gqm);

	/*
	 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
	 * be restarted.
	 */
296 297
	mutex_lock(&xfs_Gqm_lock);
	if (--xfs_Gqm->qm_nrefs == 0) {
L
Linus Torvalds 已提交
298 299 300
		xfs_qm_destroy(xfs_Gqm);
		xfs_Gqm = NULL;
	}
301
	mutex_unlock(&xfs_Gqm_lock);
L
Linus Torvalds 已提交
302 303 304 305 306 307
}

/*
 * Just destroy the quotainfo structure.
 */
void
C
Christoph Hellwig 已提交
308 309
xfs_qm_unmount(
	struct xfs_mount	*mp)
L
Linus Torvalds 已提交
310
{
C
Christoph Hellwig 已提交
311 312
	if (mp->m_quotainfo) {
		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
L
Linus Torvalds 已提交
313
		xfs_qm_destroy_quotainfo(mp);
C
Christoph Hellwig 已提交
314
	}
L
Linus Torvalds 已提交
315 316 317 318 319 320 321 322
}


/*
 * 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.
323 324 325
 *
 * 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
326
 */
327
void
L
Linus Torvalds 已提交
328
xfs_qm_mount_quotas(
C
Christoph Hellwig 已提交
329
	xfs_mount_t	*mp)
L
Linus Torvalds 已提交
330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346
{
	int		error = 0;
	uint		sbf;

	/*
	 * If quotas on realtime volumes is not supported, we disable
	 * quotas immediately.
	 */
	if (mp->m_sb.sb_rextents) {
		cmn_err(CE_NOTE,
			"Cannot turn on quotas for realtime filesystem %s",
			mp->m_fsname);
		mp->m_qflags = 0;
		goto write_changes;
	}

	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
347

L
Linus Torvalds 已提交
348 349 350 351
	/*
	 * Allocate the quotainfo structure inside the mount struct, and
	 * create quotainode(s), and change/rev superblock if necessary.
	 */
352 353
	error = xfs_qm_init_quotainfo(mp);
	if (error) {
L
Linus Torvalds 已提交
354 355 356 357 358 359 360 361 362 363
		/*
		 * 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
364
	if (XFS_QM_NEED_QUOTACHECK(mp)) {
365 366 367 368
		error = xfs_qm_quotacheck(mp);
		if (error) {
			/* Quotacheck failed and disabled quotas. */
			return;
L
Linus Torvalds 已提交
369 370
		}
	}
371 372 373 374 375
	/* 
	 * 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.
	 */
376
	if (!XFS_IS_UQUOTA_ON(mp))
377
		mp->m_qflags &= ~XFS_UQUOTA_CHKD;
378
	if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
379
		mp->m_qflags &= ~XFS_OQUOTA_CHKD;
380

L
Linus Torvalds 已提交
381 382
 write_changes:
	/*
E
Eric Sandeen 已提交
383
	 * We actually don't have to acquire the m_sb_lock at all.
L
Linus Torvalds 已提交
384 385
	 * This can only be called from mount, and that's single threaded. XXX
	 */
E
Eric Sandeen 已提交
386
	spin_lock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
387 388
	sbf = mp->m_sb.sb_qflags;
	mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
E
Eric Sandeen 已提交
389
	spin_unlock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407

	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)));
			xfs_fs_cmn_err(CE_ALERT, mp,
				"XFS mount_quotas: Superblock update failed!");
		}
	}

	if (error) {
		xfs_fs_cmn_err(CE_WARN, mp,
			"Failed to initialize disk quotas.");
C
Christoph Hellwig 已提交
408
		return;
L
Linus Torvalds 已提交
409
	}
C
Christoph Hellwig 已提交
410 411 412 413 414

#ifdef QUOTADEBUG
	if (XFS_IS_QUOTA_ON(mp))
		xfs_qm_internalqcheck(mp);
#endif
L
Linus Torvalds 已提交
415 416 417 418 419
}

/*
 * Called from the vfsops layer.
 */
420
void
L
Linus Torvalds 已提交
421 422 423 424 425 426 427 428 429 430 431 432 433 434 435
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);

	/*
436
	 * Release the quota inodes.
L
Linus Torvalds 已提交
437 438
	 */
	if (mp->m_quotainfo) {
439 440 441
		if (mp->m_quotainfo->qi_uquotaip) {
			IRELE(mp->m_quotainfo->qi_uquotaip);
			mp->m_quotainfo->qi_uquotaip = NULL;
L
Linus Torvalds 已提交
442
		}
443 444 445
		if (mp->m_quotainfo->qi_gquotaip) {
			IRELE(mp->m_quotainfo->qi_gquotaip);
			mp->m_quotainfo->qi_gquotaip = NULL;
L
Linus Torvalds 已提交
446 447 448 449 450 451 452 453
		}
	}
}

/*
 * Flush all dquots of the given file system to disk. The dquots are
 * _not_ purged from memory here, just their data written to disk.
 */
454
STATIC int
L
Linus Torvalds 已提交
455 456
xfs_qm_dqflush_all(
	xfs_mount_t	*mp,
457
	int		sync_mode)
L
Linus Torvalds 已提交
458 459 460 461 462 463 464
{
	int		recl;
	xfs_dquot_t	*dqp;
	int		niters;
	int		error;

	if (mp->m_quotainfo == NULL)
465
		return 0;
L
Linus Torvalds 已提交
466 467
	niters = 0;
again:
468 469
	mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
	list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
470 471 472 473 474
		xfs_dqlock(dqp);
		if (! XFS_DQ_IS_DIRTY(dqp)) {
			xfs_dqunlock(dqp);
			continue;
		}
C
Christoph Hellwig 已提交
475

L
Linus Torvalds 已提交
476
		/* XXX a sentinel would be better */
477
		recl = mp->m_quotainfo->qi_dqreclaims;
478
		if (!xfs_dqflock_nowait(dqp)) {
L
Linus Torvalds 已提交
479 480 481 482 483 484 485 486 487 488 489 490 491
			/*
			 * 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.
		 */
492
		mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
493
		error = xfs_qm_dqflush(dqp, sync_mode);
L
Linus Torvalds 已提交
494 495
		xfs_dqunlock(dqp);
		if (error)
496
			return error;
L
Linus Torvalds 已提交
497

498 499 500
		mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
		if (recl != mp->m_quotainfo->qi_dqreclaims) {
			mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
501 502 503 504 505
			/* XXX restart limit */
			goto again;
		}
	}

506
	mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
507
	/* return ! busy */
508
	return 0;
L
Linus Torvalds 已提交
509 510 511 512 513 514 515 516 517 518 519 520 521
}
/*
 * 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(
	xfs_mount_t	*mp)
{
	xfs_dquot_t	*dqp, *gdqp;
	int		nrecl;

 again:
522 523
	ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
	list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
524 525 526 527 528 529 530 531 532 533 534 535
		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.
			 */
536 537
			nrecl = mp->m_quotainfo->qi_dqreclaims;
			mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
538 539
			xfs_qm_dqput(gdqp);

540 541
			mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
			if (nrecl != mp->m_quotainfo->qi_dqreclaims)
L
Linus Torvalds 已提交
542 543 544 545 546 547 548 549 550 551 552 553 554 555
				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(
	xfs_mount_t	*mp,
556
	uint		flags) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
L
Linus Torvalds 已提交
557
{
558
	xfs_dquot_t	*dqp, *n;
L
Linus Torvalds 已提交
559 560 561 562 563
	uint		dqtype;
	int		nrecl;
	int		nmisses;

	if (mp->m_quotainfo == NULL)
564
		return 0;
L
Linus Torvalds 已提交
565 566

	dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
567
	dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
L
Linus Torvalds 已提交
568 569
	dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;

570
	mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
571 572 573 574 575 576 577 578 579 580 581

	/*
	 * 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;
582
	ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
L
Linus Torvalds 已提交
583 584 585 586
	/*
	 * 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.
	 */
587
	list_for_each_entry_safe(dqp, n, &mp->m_quotainfo->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
588 589 590 591 592
		/*
		 * 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.
		 */
593
		if ((dqp->dq_flags & dqtype) == 0)
L
Linus Torvalds 已提交
594 595
			continue;

596
		if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
597 598
			nrecl = mp->m_quotainfo->qi_dqreclaims;
			mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
599
			mutex_lock(&dqp->q_hash->qh_lock);
600
			mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
601 602 603 604 605 606 607

			/*
			 * 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.
			 */
608
			if (nrecl != mp->m_quotainfo->qi_dqreclaims) {
609
				mutex_unlock(&dqp->q_hash->qh_lock);
L
Linus Torvalds 已提交
610 611 612 613 614 615 616 617
				goto again;
			}
		}

		/*
		 * Take the dquot off the mplist and hashlist. It may remain on
		 * freelist in INACTIVE state.
		 */
618
		nmisses += xfs_qm_dqpurge(dqp);
L
Linus Torvalds 已提交
619
	}
620
	mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654
	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
655
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
656
	error = 0;
657

L
Linus Torvalds 已提交
658 659 660 661 662
	/*
	 * 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.
	 */
663 664
	dqp = *IO_idqpp;
	if (dqp) {
C
Christoph Hellwig 已提交
665
		trace_xfs_dqattach_found(dqp);
666
		return 0;
L
Linus Torvalds 已提交
667 668 669 670
	}

	/*
	 * udqhint is the i_udquot field in inode, and is non-NULL only
671
	 * when the type arg is group/project. Its purpose is to save a
L
Linus Torvalds 已提交
672 673 674
	 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
	 * the user dquot.
	 */
675 676
	if (udqhint) {
		ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
L
Linus Torvalds 已提交
677 678
		xfs_dqlock(udqhint);

679 680 681 682 683 684 685 686 687 688 689 690 691 692
		/*
		 * 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
693 694
			xfs_dqunlock(dqp);
			xfs_dqunlock(udqhint);
695
			return 0;
L
Linus Torvalds 已提交
696
		}
697 698 699 700 701 702 703

		/*
		 * 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
704
		xfs_dqunlock(udqhint);
705 706
	}

L
Linus Torvalds 已提交
707 708 709 710 711 712 713
	/*
	 * 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.
	 */
714 715 716
	error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
	if (error)
		return error;
L
Linus Torvalds 已提交
717

C
Christoph Hellwig 已提交
718
	trace_xfs_dqattach_get(dqp);
719

L
Linus Torvalds 已提交
720 721 722 723 724
	/*
	 * 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;
725 726
	xfs_dqunlock(dqp);
	return 0;
L
Linus Torvalds 已提交
727 728 729 730 731 732 733
}


/*
 * 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
734
 * already and getting rid of that gets us into lock ordering constraints.
L
Linus Torvalds 已提交
735 736 737 738 739 740
 * 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,
741
	xfs_dquot_t	*gdq)
L
Linus Torvalds 已提交
742 743 744
{
	xfs_dquot_t	*tmp;

745
	xfs_dqlock(udq);
L
Linus Torvalds 已提交
746 747 748

	if ((tmp = udq->q_gdquot)) {
		if (tmp == gdq) {
749
			xfs_dqunlock(udq);
L
Linus Torvalds 已提交
750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770
			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));
771
		xfs_dqlock(gdq);
L
Linus Torvalds 已提交
772 773 774 775 776 777 778 779 780 781 782 783
	}

	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;
	}
784 785 786

	xfs_dqunlock(gdq);
	xfs_dqunlock(udq);
L
Linus Torvalds 已提交
787 788 789 790
}


/*
791 792
 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 * into account.
L
Linus Torvalds 已提交
793 794 795 796 797
 * 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
798
xfs_qm_dqattach_locked(
L
Linus Torvalds 已提交
799 800 801 802 803 804 805
	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
806 807 808 809 810
	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)
811
		return 0;
L
Linus Torvalds 已提交
812

C
Christoph Hellwig 已提交
813
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
814 815 816 817 818 819 820 821 822

	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
823 824

	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
825 826 827 828 829 830
	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) :
			xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
L
Linus Torvalds 已提交
831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846
						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
847
		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
848 849 850 851 852 853 854 855 856 857 858 859 860 861
		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.
		 */
862
		xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
L
Linus Torvalds 已提交
863 864
	}

C
Christoph Hellwig 已提交
865
 done:
L
Linus Torvalds 已提交
866 867 868 869
#ifdef QUOTADEBUG
	if (! error) {
		if (XFS_IS_UQUOTA_ON(mp))
			ASSERT(ip->i_udquot);
870
		if (XFS_IS_OQUOTA_ON(mp))
L
Linus Torvalds 已提交
871 872
			ASSERT(ip->i_gdquot);
	}
C
Christoph Hellwig 已提交
873
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
874
#endif
C
Christoph Hellwig 已提交
875 876
	return error;
}
L
Linus Torvalds 已提交
877

C
Christoph Hellwig 已提交
878 879 880 881 882 883 884 885 886 887
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
888

889
	return error;
L
Linus Torvalds 已提交
890 891 892 893 894 895 896 897 898 899 900 901 902 903
}

/*
 * 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
904 905
	trace_xfs_dquot_dqdetach(ip);

L
Linus Torvalds 已提交
906 907 908 909 910 911 912 913 914 915 916 917 918 919 920
	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(
	xfs_mount_t	*mp,
921
	int		flags)
L
Linus Torvalds 已提交
922 923 924 925 926
{
	int		recl, restarts;
	xfs_dquot_t	*dqp;
	int		error;

C
Christoph Hellwig 已提交
927
	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
928 929
		return 0;

L
Linus Torvalds 已提交
930 931 932
	restarts = 0;

  again:
933
	mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
934 935 936 937 938 939
	/*
	 * 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.
	 */
940 941
	if (!XFS_IS_QUOTA_ON(mp)) {
		mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
942
		return 0;
L
Linus Torvalds 已提交
943
	}
944 945
	ASSERT(mutex_is_locked(&mp->m_quotainfo->qi_dqlist_lock));
	list_for_each_entry(dqp, &mp->m_quotainfo->qi_dqlist, q_mplist) {
L
Linus Torvalds 已提交
946 947 948 949 950
		/*
		 * 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
951 952
		if (flags & SYNC_TRYLOCK) {
			if (!XFS_DQ_IS_DIRTY(dqp))
L
Linus Torvalds 已提交
953
				continue;
C
Christoph Hellwig 已提交
954
			if (!xfs_qm_dqlock_nowait(dqp))
L
Linus Torvalds 已提交
955 956 957 958 959 960 961 962 963 964 965 966 967 968
				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 */
969
		recl = mp->m_quotainfo->qi_dqreclaims;
970
		if (!xfs_dqflock_nowait(dqp)) {
C
Christoph Hellwig 已提交
971
			if (flags & SYNC_TRYLOCK) {
L
Linus Torvalds 已提交
972 973 974 975 976
				xfs_dqunlock(dqp);
				continue;
			}
			/*
			 * If we can't grab the flush lock then if the caller
977
			 * really wanted us to give this our best shot, so
L
Linus Torvalds 已提交
978 979
			 * see if we can give a push to the buffer before we wait
			 * on the flush lock. At this point, we know that
980
			 * even though the dquot is being flushed,
L
Linus Torvalds 已提交
981 982 983 984 985 986 987 988
			 * 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
		 */
989
		mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
990
		error = xfs_qm_dqflush(dqp, flags);
L
Linus Torvalds 已提交
991 992
		xfs_dqunlock(dqp);
		if (error && XFS_FORCED_SHUTDOWN(mp))
993
			return 0;	/* Need to prevent umount failure */
L
Linus Torvalds 已提交
994
		else if (error)
995
			return error;
L
Linus Torvalds 已提交
996

997 998
		mutex_lock(&mp->m_quotainfo->qi_dqlist_lock);
		if (recl != mp->m_quotainfo->qi_dqreclaims) {
L
Linus Torvalds 已提交
999 1000 1001
			if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
				break;

1002
			mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
L
Linus Torvalds 已提交
1003 1004 1005 1006
			goto again;
		}
	}

1007
	mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
1008
	return 0;
L
Linus Torvalds 已提交
1009 1010
}

1011 1012 1013 1014 1015 1016 1017
/*
 * 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
1018 1019 1020 1021 1022

/*
 * This initializes all the quota information that's kept in the
 * mount structure
 */
1023
STATIC int
L
Linus Torvalds 已提交
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036
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))) {
1037
		return error;
L
Linus Torvalds 已提交
1038 1039 1040 1041 1042 1043 1044 1045 1046
	}

	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))) {
1047
		kmem_free(qinf);
L
Linus Torvalds 已提交
1048
		mp->m_quotainfo = NULL;
1049
		return error;
L
Linus Torvalds 已提交
1050 1051
	}

1052 1053 1054
	INIT_LIST_HEAD(&qinf->qi_dqlist);
	mutex_init(&qinf->qi_dqlist_lock);
	lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
1055

L
Linus Torvalds 已提交
1056 1057 1058
	qinf->qi_dqreclaims = 0;

	/* mutex used to serialize quotaoffs */
1059
	mutex_init(&qinf->qi_quotaofflock);
L
Linus Torvalds 已提交
1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077

	/* 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,
1078 1079 1080
			     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
1081 1082 1083 1084 1085 1086 1087 1088 1089 1090
			     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.
		 */
1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
		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
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121
 
		/*
		 * 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;
1122
		qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
L
Linus Torvalds 已提交
1123 1124
	}

1125
	return 0;
L
Linus Torvalds 已提交
1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150
}


/*
 * 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);

1151 1152
	ASSERT(list_empty(&qi->qi_dqlist));
	mutex_destroy(&qi->qi_dqlist_lock);
L
Linus Torvalds 已提交
1153 1154

	if (qi->qi_uquotaip) {
1155
		IRELE(qi->qi_uquotaip);
L
Linus Torvalds 已提交
1156 1157 1158
		qi->qi_uquotaip = NULL; /* paranoia */
	}
	if (qi->qi_gquotaip) {
1159
		IRELE(qi->qi_gquotaip);
L
Linus Torvalds 已提交
1160 1161 1162
		qi->qi_gquotaip = NULL;
	}
	mutex_destroy(&qi->qi_quotaofflock);
1163
	kmem_free(qi);
L
Linus Torvalds 已提交
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177
	mp->m_quotainfo = NULL;
}



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

/* ARGSUSED */
STATIC void
xfs_qm_list_init(
	xfs_dqlist_t	*list,
	char		*str,
	int		n)
{
1178
	mutex_init(&list->qh_lock);
L
Linus Torvalds 已提交
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206
	list->qh_next = NULL;
	list->qh_version = 0;
	list->qh_nelems = 0;
}

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


/*
 * Stripped down version of dqattach. This doesn't attach, or even look at the
 * dquots attached to the inode. The rationale is that there won't be any
 * attached at the time this is called from quotacheck.
 */
STATIC int
xfs_qm_dqget_noattach(
	xfs_inode_t	*ip,
	xfs_dquot_t	**O_udqpp,
	xfs_dquot_t	**O_gdqpp)
{
	int		error;
	xfs_mount_t	*mp;
	xfs_dquot_t	*udqp, *gdqp;

C
Christoph Hellwig 已提交
1207
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224
	mp = ip->i_mount;
	udqp = NULL;
	gdqp = NULL;

	if (XFS_IS_UQUOTA_ON(mp)) {
		ASSERT(ip->i_udquot == NULL);
		/*
		 * We want the dquot allocated if it doesn't exist.
		 */
		if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
					 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
					 &udqp))) {
			/*
			 * Shouldn't be able to turn off quotas here.
			 */
			ASSERT(error != ESRCH);
			ASSERT(error != ENOENT);
1225
			return error;
L
Linus Torvalds 已提交
1226 1227 1228 1229
		}
		ASSERT(udqp);
	}

1230
	if (XFS_IS_OQUOTA_ON(mp)) {
L
Linus Torvalds 已提交
1231 1232 1233
		ASSERT(ip->i_gdquot == NULL);
		if (udqp)
			xfs_dqunlock(udqp);
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243
		error = XFS_IS_GQUOTA_ON(mp) ?
				xfs_qm_dqget(mp, ip,
					     ip->i_d.di_gid, XFS_DQ_GROUP,
					     XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
					     &gdqp) :
				xfs_qm_dqget(mp, ip,
					     ip->i_d.di_projid, XFS_DQ_PROJ,
					     XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
					     &gdqp);
		if (error) {
L
Linus Torvalds 已提交
1244 1245 1246 1247
			if (udqp)
				xfs_qm_dqrele(udqp);
			ASSERT(error != ESRCH);
			ASSERT(error != ENOENT);
1248
			return error;
L
Linus Torvalds 已提交
1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
		}
		ASSERT(gdqp);

		/* Reacquire the locks in the right order */
		if (udqp) {
			if (! xfs_qm_dqlock_nowait(udqp)) {
				xfs_dqunlock(gdqp);
				xfs_dqlock(udqp);
				xfs_dqlock(gdqp);
			}
		}
	}

	*O_udqpp = udqp;
	*O_gdqpp = gdqp;

#ifdef QUOTADEBUG
	if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
	if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
#endif
1269
	return 0;
L
Linus Torvalds 已提交
1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286
}

/*
 * 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;

1287
	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
L
Linus Torvalds 已提交
1288 1289 1290 1291 1292 1293
	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);
1294
		return error;
L
Linus Torvalds 已提交
1295 1296
	}

D
David Chinner 已提交
1297
	if ((error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0,
1298
				   &xfs_zerocr, 0, 1, ip, &committed))) {
L
Linus Torvalds 已提交
1299 1300
		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
				 XFS_TRANS_ABORT);
1301
		return error;
L
Linus Torvalds 已提交
1302 1303 1304 1305 1306 1307
	}

	/*
	 * Keep an extra reference to this quota inode. This inode is
	 * locked exclusively and joined to the transaction already.
	 */
C
Christoph Hellwig 已提交
1308
	ASSERT(xfs_isilocked(*ip, XFS_ILOCK_EXCL));
1309
	IHOLD(*ip);
L
Linus Torvalds 已提交
1310 1311 1312 1313 1314 1315

	/*
	 * 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
1316
	spin_lock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
1317 1318 1319 1320
	if (flags & XFS_QMOPT_SBVERSION) {
#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
		unsigned oldv = mp->m_sb.sb_versionnum;
#endif
1321
		ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
L
Linus Torvalds 已提交
1322 1323 1324 1325 1326
		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));

1327
		xfs_sb_version_addquota(&mp->m_sb);
L
Linus Torvalds 已提交
1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342
		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 defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
		cmn_err(CE_NOTE,
			"Old superblock version %x, converting to %x.",
			oldv, mp->m_sb.sb_versionnum);
#endif
	}
	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
1343
	spin_unlock(&mp->m_sb_lock);
L
Linus Torvalds 已提交
1344 1345
	xfs_mod_sb(tp, sbfields);

1346
	if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
L
Linus Torvalds 已提交
1347
		xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
1348
		return error;
L
Linus Torvalds 已提交
1349
	}
1350
	return 0;
L
Linus Torvalds 已提交
1351 1352 1353
}


1354
STATIC void
L
Linus Torvalds 已提交
1355 1356 1357 1358 1359 1360 1361 1362 1363
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
1364 1365
	trace_xfs_reset_dqcounts(bp, _RET_IP_);

L
Linus Torvalds 已提交
1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379
	/*
	 * 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));
	ASSERT(XFS_QM_DQPERBLK(mp) == j);
#endif
	ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
	for (j = 0; j < XFS_QM_DQPERBLK(mp); j++) {
		/*
		 * Do a sanity check, and if needed, repair the dqblk. Don't
		 * output any warnings because it's perfectly possible to
1380
		 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
L
Linus Torvalds 已提交
1381 1382 1383
		 */
		(void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
				      "xfs_quotacheck");
1384 1385 1386 1387 1388 1389 1390 1391 1392
		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
1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
		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;
	int		notcommitted;
	int		incr;
1409
	int		type;
L
Linus Torvalds 已提交
1410 1411 1412 1413 1414

	ASSERT(blkcnt > 0);
	notcommitted = 0;
	incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
		XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
1415 1416
	type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
		(flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
L
Linus Torvalds 已提交
1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434
	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),
			      (int)XFS_QI_DQCHUNKLEN(mp), 0, &bp);
		if (error)
			break;

1435
		xfs_qm_reset_dqcounts(mp, bp, firstid, type);
L
Linus Torvalds 已提交
1436 1437 1438 1439 1440 1441 1442
		xfs_bdwrite(mp, bp);
		/*
		 * goto the next block.
		 */
		bno++;
		firstid += XFS_QM_DQPERBLK(mp);
	}
1443
	return error;
L
Linus Torvalds 已提交
1444 1445 1446
}

/*
1447
 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
L
Linus Torvalds 已提交
1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466
 * 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;
	/*
1467
	 * This looks racy, but we can't keep an inode lock across a
L
Linus Torvalds 已提交
1468 1469 1470 1471
	 * 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)
1472
		return 0;
L
Linus Torvalds 已提交
1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489

	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,
1490
				  0, map, &nmaps, NULL, NULL);
L
Linus Torvalds 已提交
1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538
		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 *
				XFS_QM_DQPERBLK(mp);
			/*
			 * 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--) {
					xfs_baread(mp->m_ddev_targp,
					       XFS_FSB_TO_DADDR(mp, rablkno),
					       (int)XFS_QI_DQCHUNKLEN(mp));
					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);

1539
	kmem_free(map);
L
Linus Torvalds 已提交
1540

1541
	return error;
L
Linus Torvalds 已提交
1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
}

/*
 * Called by dqusage_adjust in doing a quotacheck.
 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
 * this updates its incore copy 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.
 */
STATIC void
xfs_qm_quotacheck_dqadjust(
	xfs_dquot_t		*dqp,
	xfs_qcnt_t		nblks,
	xfs_qcnt_t		rtblks)
{
	ASSERT(XFS_DQ_IS_LOCKED(dqp));
C
Christoph Hellwig 已提交
1558 1559 1560

	trace_xfs_dqadjust(dqp);

L
Linus Torvalds 已提交
1561 1562 1563 1564
	/*
	 * Adjust the inode count and the block count to reflect this inode's
	 * resource usage.
	 */
1565
	be64_add_cpu(&dqp->q_core.d_icount, 1);
L
Linus Torvalds 已提交
1566 1567
	dqp->q_res_icount++;
	if (nblks) {
1568
		be64_add_cpu(&dqp->q_core.d_bcount, nblks);
L
Linus Torvalds 已提交
1569 1570 1571
		dqp->q_res_bcount += nblks;
	}
	if (rtblks) {
1572
		be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
L
Linus Torvalds 已提交
1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592
		dqp->q_res_rtbcount += rtblks;
	}

	/*
	 * Set default limits, adjust timers (since we changed usages)
	 */
	if (! XFS_IS_SUSER_DQUOT(dqp)) {
		xfs_qm_adjust_dqlimits(dqp->q_mount, &dqp->q_core);
		xfs_qm_adjust_dqtimers(dqp->q_mount, &dqp->q_core);
	}

	dqp->dq_flags |= XFS_DQ_DIRTY;
}

STATIC int
xfs_qm_get_rtblks(
	xfs_inode_t	*ip,
	xfs_qcnt_t	*O_rtblks)
{
	xfs_filblks_t	rtblks;			/* total rt blks */
1593
	xfs_extnum_t	idx;			/* extent record index */
L
Linus Torvalds 已提交
1594 1595 1596 1597 1598 1599 1600 1601
	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)))
1602
			return error;
L
Linus Torvalds 已提交
1603 1604
	}
	rtblks = 0;
1605
	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1606 1607
	for (idx = 0; idx < nextents; idx++)
		rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
L
Linus Torvalds 已提交
1608
	*O_rtblks = (xfs_qcnt_t)rtblks;
1609
	return 0;
L
Linus Torvalds 已提交
1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652
}

/*
 * 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 */
	void		*private_data,	/* not used */
	xfs_daddr_t	bno,		/* starting block of inode cluster */
	int		*ubused,	/* not used */
	void		*dip,		/* on-disk inode pointer (not used) */
	int		*res)		/* result code value */
{
	xfs_inode_t	*ip;
	xfs_dquot_t	*udqp, *gdqp;
	xfs_qcnt_t	nblks, rtblks;
	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.
	 */
	if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip, bno))) {
		*res = BULKSTAT_RV_NOTHING;
1653
		return error;
L
Linus Torvalds 已提交
1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664
	}

	/*
	 * Obtain the locked dquots. In case of an error (eg. allocation
	 * fails for ENOSPC), we return the negative of the error number
	 * to bulkstat, so that it can get propagated to quotacheck() and
	 * making us disable quotas for the file system.
	 */
	if ((error = xfs_qm_dqget_noattach(ip, &udqp, &gdqp))) {
		xfs_iput(ip, XFS_ILOCK_EXCL);
		*res = BULKSTAT_RV_GIVEUP;
1665
		return error;
L
Linus Torvalds 已提交
1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681
	}

	rtblks = 0;
	if (! XFS_IS_REALTIME_INODE(ip)) {
		nblks = (xfs_qcnt_t)ip->i_d.di_nblocks;
	} else {
		/*
		 * Walk thru the extent list and count the realtime blocks.
		 */
		if ((error = xfs_qm_get_rtblks(ip, &rtblks))) {
			xfs_iput(ip, XFS_ILOCK_EXCL);
			if (udqp)
				xfs_qm_dqput(udqp);
			if (gdqp)
				xfs_qm_dqput(gdqp);
			*res = BULKSTAT_RV_GIVEUP;
1682
			return error;
L
Linus Torvalds 已提交
1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703
		}
		nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
	}
	ASSERT(ip->i_delayed_blks == 0);

	/*
	 * We can't release the inode while holding its dquot locks.
	 * The inode can go into inactive and might try to acquire the dquotlocks.
	 * So, just unlock here and do a vn_rele at the end.
	 */
	xfs_iunlock(ip, XFS_ILOCK_EXCL);

	/*
	 * 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.
	 *
1704
	 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
L
Linus Torvalds 已提交
1705 1706 1707 1708 1709 1710 1711
	 * and quotaoffs don't race. (Quotachecks happen at mount time only).
	 */
	if (XFS_IS_UQUOTA_ON(mp)) {
		ASSERT(udqp);
		xfs_qm_quotacheck_dqadjust(udqp, nblks, rtblks);
		xfs_qm_dqput(udqp);
	}
1712
	if (XFS_IS_OQUOTA_ON(mp)) {
L
Linus Torvalds 已提交
1713 1714 1715 1716 1717 1718 1719 1720
		ASSERT(gdqp);
		xfs_qm_quotacheck_dqadjust(gdqp, nblks, rtblks);
		xfs_qm_dqput(gdqp);
	}
	/*
	 * Now release the inode. This will send it to 'inactive', and
	 * possibly even free blocks.
	 */
1721
	IRELE(ip);
L
Linus Torvalds 已提交
1722 1723 1724 1725 1726

	/*
	 * Goto next inode.
	 */
	*res = BULKSTAT_RV_DIDONE;
1727
	return 0;
L
Linus Torvalds 已提交
1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755
}

/*
 * 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;

	ASSERT(XFS_QI_UQIP(mp) || XFS_QI_GQIP(mp));
	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	/*
	 * There should be no cached dquots. The (simplistic) quotacheck
	 * algorithm doesn't like that.
	 */
1756
	ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
L
Linus Torvalds 已提交
1757 1758 1759 1760

	cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);

	/*
1761
	 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
L
Linus Torvalds 已提交
1762 1763 1764 1765 1766 1767 1768 1769 1770 1771
	 * their counters to zero. We need a clean slate.
	 * We don't log our changes till later.
	 */
	if ((uip = XFS_QI_UQIP(mp))) {
		if ((error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA)))
			goto error_return;
		flags |= XFS_UQUOTA_CHKD;
	}

	if ((gip = XFS_QI_GQIP(mp))) {
1772 1773
		if ((error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
					XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA)))
L
Linus Torvalds 已提交
1774
			goto error_return;
1775
		flags |= XFS_OQUOTA_CHKD;
L
Linus Torvalds 已提交
1776 1777 1778 1779 1780 1781 1782 1783 1784
	}

	do {
		/*
		 * Iterate thru all the inodes in the file system,
		 * adjusting the corresponding dquot counters in core.
		 */
		if ((error = xfs_bulkstat(mp, &lastino, &count,
				     xfs_qm_dqusage_adjust, NULL,
1785
				     structsz, NULL, BULKSTAT_FG_IGET, &done)))
L
Linus Torvalds 已提交
1786 1787 1788 1789
			break;

	} while (! done);

1790 1791 1792 1793 1794 1795
	/*
	 * 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)
1796
		error = xfs_qm_dqflush_all(mp, 0);
1797

L
Linus Torvalds 已提交
1798 1799 1800 1801 1802 1803 1804 1805
	/*
	 * 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) {
1806
		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_QUOTAOFF);
L
Linus Torvalds 已提交
1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823
		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.
	 */
1824
	mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
L
Linus Torvalds 已提交
1825 1826
	mp->m_qflags |= flags;

1827
	xfs_qm_dquot_list_print(mp);
L
Linus Torvalds 已提交
1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839

 error_return:
	if (error) {
		cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
			"Disabling quotas.",
			mp->m_fsname, error);
		/*
		 * We must turn off quotas.
		 */
		ASSERT(mp->m_quotainfo != NULL);
		ASSERT(xfs_Gqm != NULL);
		xfs_qm_destroy_quotainfo(mp);
1840 1841 1842 1843
		if (xfs_mount_reset_sbqflags(mp)) {
			cmn_err(CE_WARN, "XFS quotacheck %s: "
				"Failed to reset quota flags.", mp->m_fsname);
		}
L
Linus Torvalds 已提交
1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870
	} else {
		cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
	}
	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
	 */
1871
	if (xfs_sb_version_hasquota(&mp->m_sb)) {
L
Linus Torvalds 已提交
1872 1873 1874 1875 1876 1877 1878
		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,
					     0, 0, &uip, 0)))
				return XFS_ERROR(error);
		}
1879
		if (XFS_IS_OQUOTA_ON(mp) &&
L
Linus Torvalds 已提交
1880 1881 1882 1883 1884
		    mp->m_sb.sb_gquotino != NULLFSINO) {
			ASSERT(mp->m_sb.sb_gquotino > 0);
			if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
					     0, 0, &gip, 0))) {
				if (uip)
1885
					IRELE(uip);
L
Linus Torvalds 已提交
1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908
				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;
	}
1909 1910 1911 1912 1913 1914
	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
1915
			if (uip)
1916
				IRELE(uip);
L
Linus Torvalds 已提交
1917 1918 1919 1920 1921 1922 1923 1924

			return XFS_ERROR(error);
		}
	}

	XFS_QI_UQIP(mp) = uip;
	XFS_QI_GQIP(mp) = gip;

1925
	return 0;
L
Linus Torvalds 已提交
1926 1927 1928
}


1929

L
Linus Torvalds 已提交
1930
/*
1931 1932
 * Just pop the least recently used dquot off the freelist and
 * recycle it. The returned dquot is locked.
L
Linus Torvalds 已提交
1933
 */
1934 1935
STATIC xfs_dquot_t *
xfs_qm_dqreclaim_one(void)
L
Linus Torvalds 已提交
1936
{
1937 1938
	xfs_dquot_t	*dqpout;
	xfs_dquot_t	*dqp;
L
Linus Torvalds 已提交
1939 1940 1941
	int		restarts;

	restarts = 0;
1942
	dqpout = NULL;
L
Linus Torvalds 已提交
1943

1944 1945
	/* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
startagain:
L
Linus Torvalds 已提交
1946 1947
	xfs_qm_freelist_lock(xfs_Gqm);

1948
	FOREACH_DQUOT_IN_FREELIST(dqp, &(xfs_Gqm->qm_dqfreelist)) {
1949
		struct xfs_mount *mp = dqp->q_mount;
L
Linus Torvalds 已提交
1950 1951 1952 1953
		xfs_dqlock(dqp);

		/*
		 * We are racing with dqlookup here. Naturally we don't
1954 1955 1956
		 * 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
1957 1958
		 */
		if (dqp->dq_flags & XFS_DQ_WANT) {
1959 1960 1961 1962
			ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));

			trace_xfs_dqreclaim_want(dqp);

L
Linus Torvalds 已提交
1963 1964 1965
			xfs_dqunlock(dqp);
			xfs_qm_freelist_unlock(xfs_Gqm);
			if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1966
				return NULL;
L
Linus Torvalds 已提交
1967
			XQM_STATS_INC(xqmstats.xs_qm_dqwants);
1968
			goto startagain;
L
Linus Torvalds 已提交
1969 1970 1971 1972 1973 1974 1975 1976
		}

		/*
		 * 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) {
1977
			ASSERT(mp == NULL);
L
Linus Torvalds 已提交
1978 1979
			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
			ASSERT(dqp->HL_PREVP == NULL);
1980
			ASSERT(list_empty(&dqp->q_mplist));
1981 1982 1983
			XQM_FREELIST_REMOVE(dqp);
			xfs_dqunlock(dqp);
			dqpout = dqp;
L
Linus Torvalds 已提交
1984
			XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
1985
			break;
L
Linus Torvalds 已提交
1986 1987
		}

1988
		ASSERT(dqp->q_hash);
1989
		ASSERT(!list_empty(&dqp->q_mplist));
1990

L
Linus Torvalds 已提交
1991 1992 1993 1994
		/*
		 * 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.
		 */
1995
		if (!xfs_dqflock_nowait(dqp)) {
L
Linus Torvalds 已提交
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
			xfs_dqunlock(dqp);
			continue;
		}

		/*
		 * 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)) {
2007
			int	error;
C
Christoph Hellwig 已提交
2008

2009
			trace_xfs_dqreclaim_dirty(dqp);
C
Christoph Hellwig 已提交
2010

L
Linus Torvalds 已提交
2011 2012
			/*
			 * We flush it delayed write, so don't bother
2013
			 * releasing the freelist lock.
L
Linus Torvalds 已提交
2014
			 */
2015
			error = xfs_qm_dqflush(dqp, 0);
2016
			if (error) {
2017
				xfs_fs_cmn_err(CE_WARN, mp,
2018
			"xfs_qm_dqreclaim: dquot %p flush failed", dqp);
2019
			}
L
Linus Torvalds 已提交
2020 2021 2022
			xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
			continue;
		}
2023

L
Linus Torvalds 已提交
2024 2025 2026 2027 2028 2029 2030
		/*
		 * 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.
		 */
2031
		if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
2032 2033
			restarts++;
			goto dqfunlock;
L
Linus Torvalds 已提交
2034
		}
2035

L
Linus Torvalds 已提交
2036 2037 2038 2039 2040
		/*
		 * 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.
		 */
2041
		if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) {
2042 2043
			restarts++;
			mutex_unlock(&dqp->q_hash->qh_lock);
L
Linus Torvalds 已提交
2044 2045 2046
			xfs_dqfunlock(dqp);
			xfs_dqunlock(dqp);
			xfs_qm_freelist_unlock(xfs_Gqm);
2047 2048 2049
			if (restarts++ >= XFS_QM_RECLAIM_MAX_RESTARTS)
				return NULL;
			goto startagain;
L
Linus Torvalds 已提交
2050
		}
C
Christoph Hellwig 已提交
2051

L
Linus Torvalds 已提交
2052
		ASSERT(dqp->q_nrefs == 0);
2053 2054 2055
		list_del_init(&dqp->q_mplist);
		mp->m_quotainfo->qi_dquots--;
		mp->m_quotainfo->qi_dqreclaims++;
2056
		XQM_HASHLIST_REMOVE(dqp->q_hash, dqp);
L
Linus Torvalds 已提交
2057
		XQM_FREELIST_REMOVE(dqp);
2058 2059 2060 2061 2062
		dqpout = dqp;
		mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
		mutex_unlock(&dqp->q_hash->qh_lock);
dqfunlock:
		xfs_dqfunlock(dqp);
L
Linus Torvalds 已提交
2063
		xfs_dqunlock(dqp);
2064 2065 2066 2067
		if (dqpout)
			break;
		if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
			return NULL;
L
Linus Torvalds 已提交
2068 2069
	}
	xfs_qm_freelist_unlock(xfs_Gqm);
2070
	return dqpout;
L
Linus Torvalds 已提交
2071 2072
}

2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096
/*
 * 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
2097 2098 2099 2100 2101 2102

/*
 * The kmem_shake interface is invoked when memory is running low.
 */
/* ARGSUSED */
STATIC int
A
Al Viro 已提交
2103
xfs_qm_shake(int nr_to_scan, gfp_t gfp_mask)
L
Linus Torvalds 已提交
2104 2105 2106 2107
{
	int	ndqused, nfree, n;

	if (!kmem_shake_allow(gfp_mask))
2108
		return 0;
L
Linus Torvalds 已提交
2109
	if (!xfs_Gqm)
2110
		return 0;
L
Linus Torvalds 已提交
2111 2112 2113 2114 2115 2116 2117 2118

	nfree = xfs_Gqm->qm_dqfreelist.qh_nelems; /* free dquots */
	/* incore dquots in all f/s's */
	ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;

	ASSERT(ndqused >= 0);

	if (nfree <= ndqused && nfree < ndquot)
2119
		return 0;
L
Linus Torvalds 已提交
2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159

	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;
2160
			return B_FALSE;
L
Linus Torvalds 已提交
2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172
		}
		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);

2173
	return B_TRUE;
L
Linus Torvalds 已提交
2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197
}


/*
 * 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
	cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
#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);
2198
		return error;
L
Linus Torvalds 已提交
2199 2200 2201
	}

	xfs_mod_sb(tp, flags);
2202
	error = xfs_trans_commit(tp, 0);
L
Linus Torvalds 已提交
2203

2204
	return error;
L
Linus Torvalds 已提交
2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222
}


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


/*
 * Given an inode, a uid and gid (from cred_t) make sure that we have
 * 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
2223 2224 2225 2226 2227 2228 2229
	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
2230
{
C
Christoph Hellwig 已提交
2231 2232 2233 2234
	struct xfs_mount	*mp = ip->i_mount;
	struct xfs_dquot	*uq, *gq;
	int			error;
	uint			lockflags;
L
Linus Torvalds 已提交
2235

C
Christoph Hellwig 已提交
2236
	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
L
Linus Torvalds 已提交
2237 2238 2239 2240 2241
		return 0;

	lockflags = XFS_ILOCK_EXCL;
	xfs_ilock(ip, lockflags);

2242
	if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
L
Linus Torvalds 已提交
2243 2244 2245 2246 2247 2248 2249
		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
2250 2251
		error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
		if (error) {
L
Linus Torvalds 已提交
2252
			xfs_iunlock(ip, lockflags);
2253
			return error;
L
Linus Torvalds 已提交
2254 2255 2256 2257
		}
	}

	uq = gq = NULL;
2258
	if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
L
Linus Torvalds 已提交
2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275
		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);
2276
				return error;
L
Linus Torvalds 已提交
2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295
			}
			/*
			 * 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);
		}
	}
2296
	if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
L
Linus Torvalds 已提交
2297 2298 2299 2300 2301 2302 2303 2304 2305 2306
		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);
2307
				return error;
L
Linus Torvalds 已提交
2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318
			}
			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);
		}
2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341
	} else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
		if (ip->i_d.di_projid != prid) {
			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
2342 2343
	}
	if (uq)
C
Christoph Hellwig 已提交
2344
		trace_xfs_dquot_dqalloc(ip);
L
Linus Torvalds 已提交
2345 2346 2347 2348 2349 2350 2351 2352 2353 2354

	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);
2355
	return 0;
L
Linus Torvalds 已提交
2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369
}

/*
 * 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;
2370 2371 2372
	uint		bfield = XFS_IS_REALTIME_INODE(ip) ?
				 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;

C
Christoph Hellwig 已提交
2373

C
Christoph Hellwig 已提交
2374
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
L
Linus Torvalds 已提交
2375 2376 2377 2378 2379 2380 2381
	ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));

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

2382 2383
	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
2384 2385

	/* the sparkling new dquot */
2386 2387
	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
2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398

	/*
	 * 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;

2399
	return prevdq;
L
Linus Torvalds 已提交
2400 2401 2402
}

/*
2403
 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
L
Linus Torvalds 已提交
2404 2405 2406 2407 2408 2409 2410 2411 2412
 */
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
2413
	xfs_mount_t	*mp = ip->i_mount;
2414
	uint		delblks, blkflags, prjflags = 0;
L
Linus Torvalds 已提交
2415
	xfs_dquot_t	*unresudq, *unresgdq, *delblksudq, *delblksgdq;
C
Christoph Hellwig 已提交
2416 2417
	int		error;

L
Linus Torvalds 已提交
2418

C
Christoph Hellwig 已提交
2419
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
L
Linus Torvalds 已提交
2420 2421 2422 2423
	ASSERT(XFS_IS_QUOTA_RUNNING(mp));

	delblks = ip->i_delayed_blks;
	delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2424 2425
	blkflags = XFS_IS_REALTIME_INODE(ip) ?
			XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
L
Linus Torvalds 已提交
2426 2427

	if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2428
	    ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
L
Linus Torvalds 已提交
2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439
		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;
		}
	}
2440
	if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2441 2442 2443 2444 2445 2446 2447
		if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
		     ip->i_d.di_projid != be32_to_cpu(gdqp->q_core.d_id))
			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))) {
2448 2449 2450 2451 2452
			delblksgdq = gdqp;
			if (delblks) {
				ASSERT(ip->i_gdquot);
				unresgdq = ip->i_gdquot;
			}
L
Linus Torvalds 已提交
2453 2454 2455 2456 2457
		}
	}

	if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
				delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2458
				flags | blkflags | prjflags)))
L
Linus Torvalds 已提交
2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474
		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,
2475
				flags | blkflags | prjflags)))
L
Linus Torvalds 已提交
2476 2477 2478
			return (error);
		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
				unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2479
				blkflags);
L
Linus Torvalds 已提交
2480 2481 2482 2483 2484 2485 2486
	}

	return (0);
}

int
xfs_qm_vop_rename_dqattach(
C
Christoph Hellwig 已提交
2487
	struct xfs_inode	**i_tab)
L
Linus Torvalds 已提交
2488
{
C
Christoph Hellwig 已提交
2489 2490
	struct xfs_mount	*mp = i_tab[0]->i_mount;
	int			i;
L
Linus Torvalds 已提交
2491

C
Christoph Hellwig 已提交
2492
	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2493
		return 0;
L
Linus Torvalds 已提交
2494

C
Christoph Hellwig 已提交
2495 2496 2497 2498
	for (i = 0; (i < 4 && i_tab[i]); i++) {
		struct xfs_inode	*ip = i_tab[i];
		int			error;

L
Linus Torvalds 已提交
2499 2500 2501
		/*
		 * Watch out for duplicate entries in the table.
		 */
C
Christoph Hellwig 已提交
2502 2503
		if (i == 0 || ip != i_tab[i-1]) {
			if (XFS_NOT_DQATTACHED(mp, ip)) {
L
Linus Torvalds 已提交
2504 2505
				error = xfs_qm_dqattach(ip, 0);
				if (error)
2506
					return error;
L
Linus Torvalds 已提交
2507 2508 2509
			}
		}
	}
2510
	return 0;
L
Linus Torvalds 已提交
2511 2512 2513
}

void
C
Christoph Hellwig 已提交
2514 2515 2516 2517 2518
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
2519
{
C
Christoph Hellwig 已提交
2520 2521 2522
	struct xfs_mount	*mp = tp->t_mountp;

	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
L
Linus Torvalds 已提交
2523 2524
		return;

C
Christoph Hellwig 已提交
2525
	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
C
Christoph Hellwig 已提交
2526
	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
L
Linus Torvalds 已提交
2527 2528 2529 2530 2531 2532 2533

	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
2534
		ASSERT(XFS_IS_UQUOTA_ON(mp));
2535
		ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
L
Linus Torvalds 已提交
2536 2537 2538 2539 2540 2541 2542 2543
		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
2544 2545
		ASSERT(XFS_IS_OQUOTA_ON(mp));
		ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2546 2547
			ip->i_d.di_gid : ip->i_d.di_projid) ==
				be32_to_cpu(gdqp->q_core.d_id));
L
Linus Torvalds 已提交
2548 2549 2550 2551 2552
		xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
	}
}

/* ------------- list stuff -----------------*/
2553
STATIC void
L
Linus Torvalds 已提交
2554 2555 2556
xfs_qm_freelist_init(xfs_frlist_t *ql)
{
	ql->qh_next = ql->qh_prev = (xfs_dquot_t *) ql;
2557
	mutex_init(&ql->qh_lock);
L
Linus Torvalds 已提交
2558 2559 2560 2561
	ql->qh_version = 0;
	ql->qh_nelems = 0;
}

2562
STATIC void
L
Linus Torvalds 已提交
2563 2564 2565 2566
xfs_qm_freelist_destroy(xfs_frlist_t *ql)
{
	xfs_dquot_t	*dqp, *nextdqp;

2567
	mutex_lock(&ql->qh_lock);
L
Linus Torvalds 已提交
2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579
	for (dqp = ql->qh_next;
	     dqp != (xfs_dquot_t *)ql; ) {
		xfs_dqlock(dqp);
		nextdqp = dqp->dq_flnext;
#ifdef QUOTADEBUG
		cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
#endif
		XQM_FREELIST_REMOVE(dqp);
		xfs_dqunlock(dqp);
		xfs_qm_dqdestroy(dqp);
		dqp = nextdqp;
	}
2580
	mutex_unlock(&ql->qh_lock);
L
Linus Torvalds 已提交
2581 2582 2583 2584 2585
	mutex_destroy(&ql->qh_lock);

	ASSERT(ql->qh_nelems == 0);
}

2586
STATIC void
L
Linus Torvalds 已提交
2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614
xfs_qm_freelist_insert(xfs_frlist_t *ql, xfs_dquot_t *dq)
{
	dq->dq_flnext = ql->qh_next;
	dq->dq_flprev = (xfs_dquot_t *)ql;
	ql->qh_next = dq;
	dq->dq_flnext->dq_flprev = dq;
	xfs_Gqm->qm_dqfreelist.qh_nelems++;
	xfs_Gqm->qm_dqfreelist.qh_version++;
}

void
xfs_qm_freelist_unlink(xfs_dquot_t *dq)
{
	xfs_dquot_t *next = dq->dq_flnext;
	xfs_dquot_t *prev = dq->dq_flprev;

	next->dq_flprev = prev;
	prev->dq_flnext = next;
	dq->dq_flnext = dq->dq_flprev = dq;
	xfs_Gqm->qm_dqfreelist.qh_nelems--;
	xfs_Gqm->qm_dqfreelist.qh_version++;
}

void
xfs_qm_freelist_append(xfs_frlist_t *ql, xfs_dquot_t *dq)
{
	xfs_qm_freelist_insert((xfs_frlist_t *)ql->qh_prev, dq);
}