1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
 
 
 
 
  20#include "xfs_bit.h"
  21#include "xfs_log.h"
  22#include "xfs_inum.h"
  23#include "xfs_trans.h"
  24#include "xfs_sb.h"
  25#include "xfs_ag.h"
  26#include "xfs_alloc.h"
  27#include "xfs_quota.h"
  28#include "xfs_mount.h"
  29#include "xfs_bmap_btree.h"
  30#include "xfs_ialloc_btree.h"
  31#include "xfs_dinode.h"
  32#include "xfs_inode.h"
  33#include "xfs_ialloc.h"
  34#include "xfs_itable.h"
  35#include "xfs_rtalloc.h"
  36#include "xfs_error.h"
  37#include "xfs_bmap.h"
  38#include "xfs_attr.h"
  39#include "xfs_buf_item.h"
 
  40#include "xfs_trans_space.h"
  41#include "xfs_utils.h"
  42#include "xfs_qm.h"
  43#include "xfs_trace.h"
 
 
  44
  45/*
  46 * The global quota manager. There is only one of these for the entire
  47 * system, _not_ one per file system. XQM keeps track of the overall
  48 * quota functionality, including maintaining the freelist and hash
  49 * tables of dquots.
  50 */
  51struct mutex	xfs_Gqm_lock;
  52struct xfs_qm	*xfs_Gqm;
  53uint		ndquot;
  54
  55kmem_zone_t	*qm_dqzone;
  56kmem_zone_t	*qm_dqtrxzone;
  57
  58STATIC void	xfs_qm_list_init(xfs_dqlist_t *, char *, int);
  59STATIC void	xfs_qm_list_destroy(xfs_dqlist_t *);
  60
  61STATIC int	xfs_qm_init_quotainos(xfs_mount_t *);
  62STATIC int	xfs_qm_init_quotainfo(xfs_mount_t *);
  63STATIC int	xfs_qm_shake(struct shrinker *, struct shrink_control *);
  64
  65static struct shrinker xfs_qm_shaker = {
  66	.shrink = xfs_qm_shake,
  67	.seeks = DEFAULT_SEEKS,
  68};
  69
 
 
  70/*
  71 * Initialize the XQM structure.
  72 * Note that there is not one quota manager per file system.
 
 
 
  73 */
  74STATIC struct xfs_qm *
  75xfs_Gqm_init(void)
  76{
  77	xfs_dqhash_t	*udqhash, *gdqhash;
  78	xfs_qm_t	*xqm;
  79	size_t		hsize;
  80	uint		i;
  81
  82	/*
  83	 * Initialize the dquot hash tables.
  84	 */
  85	udqhash = kmem_zalloc_greedy(&hsize,
  86				     XFS_QM_HASHSIZE_LOW * sizeof(xfs_dqhash_t),
  87				     XFS_QM_HASHSIZE_HIGH * sizeof(xfs_dqhash_t));
  88	if (!udqhash)
  89		goto out;
  90
  91	gdqhash = kmem_zalloc_large(hsize);
  92	if (!gdqhash)
  93		goto out_free_udqhash;
  94
  95	hsize /= sizeof(xfs_dqhash_t);
  96	ndquot = hsize << 8;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  97
  98	xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
  99	xqm->qm_dqhashmask = hsize - 1;
 100	xqm->qm_usr_dqhtable = udqhash;
 101	xqm->qm_grp_dqhtable = gdqhash;
 102	ASSERT(xqm->qm_usr_dqhtable != NULL);
 103	ASSERT(xqm->qm_grp_dqhtable != NULL);
 104
 105	for (i = 0; i < hsize; i++) {
 106		xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
 107		xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
 108	}
 109
 110	/*
 111	 * Freelist of all dquots of all file systems
 112	 */
 113	INIT_LIST_HEAD(&xqm->qm_dqfrlist);
 114	xqm->qm_dqfrlist_cnt = 0;
 115	mutex_init(&xqm->qm_dqfrlist_lock);
 
 
 116
 117	/*
 118	 * dquot zone. we register our own low-memory callback.
 119	 */
 120	if (!qm_dqzone) {
 121		xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
 122						"xfs_dquots");
 123		qm_dqzone = xqm->qm_dqzone;
 124	} else
 125		xqm->qm_dqzone = qm_dqzone;
 126
 127	register_shrinker(&xfs_qm_shaker);
 
 
 
 
 
 
 
 
 128
 129	/*
 130	 * The t_dqinfo portion of transactions.
 131	 */
 132	if (!qm_dqtrxzone) {
 133		xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
 134						   "xfs_dqtrx");
 135		qm_dqtrxzone = xqm->qm_dqtrxzone;
 136	} else
 137		xqm->qm_dqtrxzone = qm_dqtrxzone;
 138
 139	atomic_set(&xqm->qm_totaldquots, 0);
 140	xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
 141	xqm->qm_nrefs = 0;
 142	return xqm;
 143
 144 out_free_udqhash:
 145	kmem_free_large(udqhash);
 146 out:
 147	return NULL;
 148}
 149
 
 150/*
 151 * Destroy the global quota manager when its reference count goes to zero.
 152 */
 153STATIC void
 154xfs_qm_destroy(
 155	struct xfs_qm	*xqm)
 
 156{
 157	struct xfs_dquot *dqp, *n;
 158	int		hsize, i;
 159
 160	ASSERT(xqm != NULL);
 161	ASSERT(xqm->qm_nrefs == 0);
 162	unregister_shrinker(&xfs_qm_shaker);
 163	hsize = xqm->qm_dqhashmask + 1;
 164	for (i = 0; i < hsize; i++) {
 165		xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
 166		xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
 167	}
 168	kmem_free_large(xqm->qm_usr_dqhtable);
 169	kmem_free_large(xqm->qm_grp_dqhtable);
 170	xqm->qm_usr_dqhtable = NULL;
 171	xqm->qm_grp_dqhtable = NULL;
 172	xqm->qm_dqhashmask = 0;
 173
 174	/* frlist cleanup */
 175	mutex_lock(&xqm->qm_dqfrlist_lock);
 176	list_for_each_entry_safe(dqp, n, &xqm->qm_dqfrlist, q_freelist) {
 177		xfs_dqlock(dqp);
 178		list_del_init(&dqp->q_freelist);
 179		xfs_Gqm->qm_dqfrlist_cnt--;
 180		xfs_dqunlock(dqp);
 181		xfs_qm_dqdestroy(dqp);
 182	}
 183	mutex_unlock(&xqm->qm_dqfrlist_lock);
 184	mutex_destroy(&xqm->qm_dqfrlist_lock);
 185	kmem_free(xqm);
 186}
 187
 188/*
 189 * Called at mount time to let XQM know that another file system is
 190 * starting quotas. This isn't crucial information as the individual mount
 191 * structures are pretty independent, but it helps the XQM keep a
 192 * global view of what's going on.
 193 */
 194/* ARGSUSED */
 195STATIC int
 196xfs_qm_hold_quotafs_ref(
 197	struct xfs_mount *mp)
 198{
 199	/*
 200	 * Need to lock the xfs_Gqm structure for things like this. For example,
 201	 * the structure could disappear between the entry to this routine and
 202	 * a HOLD operation if not locked.
 203	 */
 204	mutex_lock(&xfs_Gqm_lock);
 
 
 205
 206	if (!xfs_Gqm) {
 207		xfs_Gqm = xfs_Gqm_init();
 208		if (!xfs_Gqm) {
 209			mutex_unlock(&xfs_Gqm_lock);
 210			return ENOMEM;
 
 
 
 
 
 
 211		}
 
 212	}
 213
 
 
 
 
 
 
 
 
 
 
 
 214	/*
 215	 * We can keep a list of all filesystems with quotas mounted for
 216	 * debugging and statistical purposes, but ...
 217	 * Just take a reference and get out.
 218	 */
 219	xfs_Gqm->qm_nrefs++;
 220	mutex_unlock(&xfs_Gqm_lock);
 
 221
 
 222	return 0;
 223}
 224
 225
 226/*
 227 * Release the reference that a filesystem took at mount time,
 228 * so that we know when we need to destroy the entire quota manager.
 229 */
 230/* ARGSUSED */
 231STATIC void
 232xfs_qm_rele_quotafs_ref(
 233	struct xfs_mount *mp)
 234{
 235	xfs_dquot_t	*dqp, *n;
 236
 237	ASSERT(xfs_Gqm);
 238	ASSERT(xfs_Gqm->qm_nrefs > 0);
 239
 240	/*
 241	 * Go thru the freelist and destroy all inactive dquots.
 242	 */
 243	mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
 244
 245	list_for_each_entry_safe(dqp, n, &xfs_Gqm->qm_dqfrlist, q_freelist) {
 246		xfs_dqlock(dqp);
 247		if (dqp->dq_flags & XFS_DQ_INACTIVE) {
 248			ASSERT(dqp->q_mount == NULL);
 249			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
 250			ASSERT(list_empty(&dqp->q_hashlist));
 251			ASSERT(list_empty(&dqp->q_mplist));
 252			list_del_init(&dqp->q_freelist);
 253			xfs_Gqm->qm_dqfrlist_cnt--;
 254			xfs_dqunlock(dqp);
 255			xfs_qm_dqdestroy(dqp);
 256		} else {
 257			xfs_dqunlock(dqp);
 258		}
 259	}
 260	mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
 261
 262	/*
 263	 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
 264	 * be restarted.
 265	 */
 266	mutex_lock(&xfs_Gqm_lock);
 267	if (--xfs_Gqm->qm_nrefs == 0) {
 268		xfs_qm_destroy(xfs_Gqm);
 269		xfs_Gqm = NULL;
 270	}
 271	mutex_unlock(&xfs_Gqm_lock);
 272}
 273
 274/*
 275 * Just destroy the quotainfo structure.
 276 */
 277void
 278xfs_qm_unmount(
 279	struct xfs_mount	*mp)
 280{
 281	if (mp->m_quotainfo) {
 282		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
 283		xfs_qm_destroy_quotainfo(mp);
 284	}
 285}
 286
 287
 288/*
 289 * This is called from xfs_mountfs to start quotas and initialize all
 290 * necessary data structures like quotainfo.  This is also responsible for
 291 * running a quotacheck as necessary.  We are guaranteed that the superblock
 292 * is consistently read in at this point.
 293 *
 294 * If we fail here, the mount will continue with quota turned off. We don't
 295 * need to inidicate success or failure at all.
 296 */
 297void
 298xfs_qm_mount_quotas(
 299	xfs_mount_t	*mp)
 300{
 301	int		error = 0;
 302	uint		sbf;
 303
 304	/*
 305	 * If quotas on realtime volumes is not supported, we disable
 306	 * quotas immediately.
 307	 */
 308	if (mp->m_sb.sb_rextents) {
 309		xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
 310		mp->m_qflags = 0;
 311		goto write_changes;
 312	}
 313
 314	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 315
 316	/*
 317	 * Allocate the quotainfo structure inside the mount struct, and
 318	 * create quotainode(s), and change/rev superblock if necessary.
 319	 */
 320	error = xfs_qm_init_quotainfo(mp);
 321	if (error) {
 322		/*
 323		 * We must turn off quotas.
 324		 */
 325		ASSERT(mp->m_quotainfo == NULL);
 326		mp->m_qflags = 0;
 327		goto write_changes;
 328	}
 329	/*
 330	 * If any of the quotas are not consistent, do a quotacheck.
 331	 */
 332	if (XFS_QM_NEED_QUOTACHECK(mp)) {
 333		error = xfs_qm_quotacheck(mp);
 334		if (error) {
 335			/* Quotacheck failed and disabled quotas. */
 336			return;
 337		}
 338	}
 339	/* 
 340	 * If one type of quotas is off, then it will lose its
 341	 * quotachecked status, since we won't be doing accounting for
 342	 * that type anymore.
 343	 */
 344	if (!XFS_IS_UQUOTA_ON(mp))
 345		mp->m_qflags &= ~XFS_UQUOTA_CHKD;
 346	if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
 347		mp->m_qflags &= ~XFS_OQUOTA_CHKD;
 348
 349 write_changes:
 350	/*
 351	 * We actually don't have to acquire the m_sb_lock at all.
 352	 * This can only be called from mount, and that's single threaded. XXX
 353	 */
 354	spin_lock(&mp->m_sb_lock);
 355	sbf = mp->m_sb.sb_qflags;
 356	mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
 357	spin_unlock(&mp->m_sb_lock);
 358
 359	if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
 360		if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
 361			/*
 362			 * We could only have been turning quotas off.
 363			 * We aren't in very good shape actually because
 364			 * the incore structures are convinced that quotas are
 365			 * off, but the on disk superblock doesn't know that !
 366			 */
 367			ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
 368			xfs_alert(mp, "%s: Superblock update failed!",
 369				__func__);
 370		}
 371	}
 372
 373	if (error) {
 374		xfs_warn(mp, "Failed to initialize disk quotas.");
 375		return;
 376	}
 377}
 378
 379/*
 380 * Called from the vfsops layer.
 381 */
 382void
 383xfs_qm_unmount_quotas(
 384	xfs_mount_t	*mp)
 385{
 386	/*
 387	 * Release the dquots that root inode, et al might be holding,
 388	 * before we flush quotas and blow away the quotainfo structure.
 389	 */
 390	ASSERT(mp->m_rootip);
 391	xfs_qm_dqdetach(mp->m_rootip);
 392	if (mp->m_rbmip)
 393		xfs_qm_dqdetach(mp->m_rbmip);
 394	if (mp->m_rsumip)
 395		xfs_qm_dqdetach(mp->m_rsumip);
 396
 397	/*
 398	 * Release the quota inodes.
 399	 */
 400	if (mp->m_quotainfo) {
 401		if (mp->m_quotainfo->qi_uquotaip) {
 402			IRELE(mp->m_quotainfo->qi_uquotaip);
 403			mp->m_quotainfo->qi_uquotaip = NULL;
 404		}
 405		if (mp->m_quotainfo->qi_gquotaip) {
 406			IRELE(mp->m_quotainfo->qi_gquotaip);
 407			mp->m_quotainfo->qi_gquotaip = NULL;
 408		}
 409	}
 410}
 411
 412/*
 413 * Flush all dquots of the given file system to disk. The dquots are
 414 * _not_ purged from memory here, just their data written to disk.
 415 */
 416STATIC int
 417xfs_qm_dqflush_all(
 418	struct xfs_mount	*mp,
 419	int			sync_mode)
 420{
 421	struct xfs_quotainfo	*q = mp->m_quotainfo;
 422	int			recl;
 423	struct xfs_dquot	*dqp;
 424	int			error;
 425
 426	if (!q)
 427		return 0;
 428again:
 429	mutex_lock(&q->qi_dqlist_lock);
 430	list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
 431		xfs_dqlock(dqp);
 432		if (! XFS_DQ_IS_DIRTY(dqp)) {
 433			xfs_dqunlock(dqp);
 434			continue;
 435		}
 436
 437		/* XXX a sentinel would be better */
 438		recl = q->qi_dqreclaims;
 439		if (!xfs_dqflock_nowait(dqp)) {
 440			/*
 441			 * If we can't grab the flush lock then check
 442			 * to see if the dquot has been flushed delayed
 443			 * write.  If so, grab its buffer and send it
 444			 * out immediately.  We'll be able to acquire
 445			 * the flush lock when the I/O completes.
 446			 */
 447			xfs_qm_dqflock_pushbuf_wait(dqp);
 448		}
 449		/*
 450		 * Let go of the mplist lock. We don't want to hold it
 451		 * across a disk write.
 452		 */
 453		mutex_unlock(&q->qi_dqlist_lock);
 454		error = xfs_qm_dqflush(dqp, sync_mode);
 455		xfs_dqunlock(dqp);
 456		if (error)
 457			return error;
 458
 459		mutex_lock(&q->qi_dqlist_lock);
 460		if (recl != q->qi_dqreclaims) {
 461			mutex_unlock(&q->qi_dqlist_lock);
 462			/* XXX restart limit */
 463			goto again;
 464		}
 465	}
 466
 467	mutex_unlock(&q->qi_dqlist_lock);
 468	/* return ! busy */
 469	return 0;
 470}
 471/*
 472 * Release the group dquot pointers the user dquots may be
 473 * carrying around as a hint. mplist is locked on entry and exit.
 474 */
 475STATIC void
 476xfs_qm_detach_gdquots(
 477	struct xfs_mount	*mp)
 478{
 479	struct xfs_quotainfo	*q = mp->m_quotainfo;
 480	struct xfs_dquot	*dqp, *gdqp;
 481	int			nrecl;
 482
 483 again:
 484	ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
 485	list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
 486		xfs_dqlock(dqp);
 487		if ((gdqp = dqp->q_gdquot)) {
 488			xfs_dqlock(gdqp);
 489			dqp->q_gdquot = NULL;
 490		}
 491		xfs_dqunlock(dqp);
 492
 493		if (gdqp) {
 494			/*
 495			 * Can't hold the mplist lock across a dqput.
 496			 * XXXmust convert to marker based iterations here.
 497			 */
 498			nrecl = q->qi_dqreclaims;
 499			mutex_unlock(&q->qi_dqlist_lock);
 500			xfs_qm_dqput(gdqp);
 501
 502			mutex_lock(&q->qi_dqlist_lock);
 503			if (nrecl != q->qi_dqreclaims)
 504				goto again;
 505		}
 506	}
 507}
 508
 509/*
 510 * Go through all the incore dquots of this file system and take them
 511 * off the mplist and hashlist, if the dquot type matches the dqtype
 512 * parameter. This is used when turning off quota accounting for
 513 * users and/or groups, as well as when the filesystem is unmounting.
 514 */
 515STATIC int
 516xfs_qm_dqpurge_int(
 517	struct xfs_mount	*mp,
 518	uint			flags)
 519{
 520	struct xfs_quotainfo	*q = mp->m_quotainfo;
 521	struct xfs_dquot	*dqp, *n;
 522	uint			dqtype;
 523	int			nrecl;
 524	int			nmisses;
 525
 526	if (!q)
 527		return 0;
 528
 529	dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
 530	dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
 531	dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
 532
 533	mutex_lock(&q->qi_dqlist_lock);
 534
 535	/*
 536	 * In the first pass through all incore dquots of this filesystem,
 537	 * we release the group dquot pointers the user dquots may be
 538	 * carrying around as a hint. We need to do this irrespective of
 539	 * what's being turned off.
 540	 */
 541	xfs_qm_detach_gdquots(mp);
 542
 543      again:
 544	nmisses = 0;
 545	ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
 546	/*
 547	 * Try to get rid of all of the unwanted dquots. The idea is to
 548	 * get them off mplist and hashlist, but leave them on freelist.
 549	 */
 550	list_for_each_entry_safe(dqp, n, &q->qi_dqlist, q_mplist) {
 551		/*
 552		 * It's OK to look at the type without taking dqlock here.
 553		 * We're holding the mplist lock here, and that's needed for
 554		 * a dqreclaim.
 555		 */
 556		if ((dqp->dq_flags & dqtype) == 0)
 557			continue;
 558
 559		if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
 560			nrecl = q->qi_dqreclaims;
 561			mutex_unlock(&q->qi_dqlist_lock);
 562			mutex_lock(&dqp->q_hash->qh_lock);
 563			mutex_lock(&q->qi_dqlist_lock);
 564
 565			/*
 566			 * XXXTheoretically, we can get into a very long
 567			 * ping pong game here.
 568			 * No one can be adding dquots to the mplist at
 569			 * this point, but somebody might be taking things off.
 570			 */
 571			if (nrecl != q->qi_dqreclaims) {
 572				mutex_unlock(&dqp->q_hash->qh_lock);
 573				goto again;
 574			}
 575		}
 576
 577		/*
 578		 * Take the dquot off the mplist and hashlist. It may remain on
 579		 * freelist in INACTIVE state.
 580		 */
 581		nmisses += xfs_qm_dqpurge(dqp);
 582	}
 583	mutex_unlock(&q->qi_dqlist_lock);
 584	return nmisses;
 585}
 586
 587int
 588xfs_qm_dqpurge_all(
 589	xfs_mount_t	*mp,
 590	uint		flags)
 591{
 592	int		ndquots;
 593
 594	/*
 595	 * Purge the dquot cache.
 596	 * None of the dquots should really be busy at this point.
 597	 */
 598	if (mp->m_quotainfo) {
 599		while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
 600			delay(ndquots * 10);
 601		}
 602	}
 603	return 0;
 604}
 605
 606STATIC int
 607xfs_qm_dqattach_one(
 608	xfs_inode_t	*ip,
 609	xfs_dqid_t	id,
 610	uint		type,
 611	uint		doalloc,
 612	xfs_dquot_t	*udqhint, /* hint */
 613	xfs_dquot_t	**IO_idqpp)
 614{
 615	xfs_dquot_t	*dqp;
 616	int		error;
 617
 618	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 619	error = 0;
 620
 621	/*
 622	 * See if we already have it in the inode itself. IO_idqpp is
 623	 * &i_udquot or &i_gdquot. This made the code look weird, but
 624	 * made the logic a lot simpler.
 625	 */
 626	dqp = *IO_idqpp;
 627	if (dqp) {
 628		trace_xfs_dqattach_found(dqp);
 629		return 0;
 630	}
 631
 632	/*
 633	 * udqhint is the i_udquot field in inode, and is non-NULL only
 634	 * when the type arg is group/project. Its purpose is to save a
 635	 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
 636	 * the user dquot.
 637	 */
 638	if (udqhint) {
 639		ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 640		xfs_dqlock(udqhint);
 641
 642		/*
 643		 * No need to take dqlock to look at the id.
 644		 *
 645		 * The ID can't change until it gets reclaimed, and it won't
 646		 * be reclaimed as long as we have a ref from inode and we
 647		 * hold the ilock.
 648		 */
 649		dqp = udqhint->q_gdquot;
 650		if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
 651			xfs_dqlock(dqp);
 652			XFS_DQHOLD(dqp);
 653			ASSERT(*IO_idqpp == NULL);
 654			*IO_idqpp = dqp;
 655
 656			xfs_dqunlock(dqp);
 657			xfs_dqunlock(udqhint);
 658			return 0;
 659		}
 660
 661		/*
 662		 * We can't hold a dquot lock when we call the dqget code.
 663		 * We'll deadlock in no time, because of (not conforming to)
 664		 * lock ordering - the inodelock comes before any dquot lock,
 665		 * and we may drop and reacquire the ilock in xfs_qm_dqget().
 666		 */
 667		xfs_dqunlock(udqhint);
 668	}
 669
 670	/*
 671	 * Find the dquot from somewhere. This bumps the
 672	 * reference count of dquot and returns it locked.
 673	 * This can return ENOENT if dquot didn't exist on
 674	 * disk and we didn't ask it to allocate;
 675	 * ESRCH if quotas got turned off suddenly.
 676	 */
 677	error = xfs_qm_dqget(ip->i_mount, ip, id, type, XFS_QMOPT_DOWARN, &dqp);
 678	if (error)
 679		return error;
 680
 681	trace_xfs_dqattach_get(dqp);
 682
 683	/*
 684	 * dqget may have dropped and re-acquired the ilock, but it guarantees
 685	 * that the dquot returned is the one that should go in the inode.
 686	 */
 687	*IO_idqpp = dqp;
 688	xfs_dqunlock(dqp);
 689	return 0;
 690}
 691
 692
 693/*
 694 * Given a udquot and gdquot, attach a ptr to the group dquot in the
 695 * udquot as a hint for future lookups. The idea sounds simple, but the
 696 * execution isn't, because the udquot might have a group dquot attached
 697 * already and getting rid of that gets us into lock ordering constraints.
 698 * The process is complicated more by the fact that the dquots may or may not
 699 * be locked on entry.
 700 */
 701STATIC void
 702xfs_qm_dqattach_grouphint(
 703	xfs_dquot_t	*udq,
 704	xfs_dquot_t	*gdq)
 705{
 706	xfs_dquot_t	*tmp;
 707
 708	xfs_dqlock(udq);
 709
 710	if ((tmp = udq->q_gdquot)) {
 711		if (tmp == gdq) {
 712			xfs_dqunlock(udq);
 713			return;
 714		}
 715
 716		udq->q_gdquot = NULL;
 717		/*
 718		 * We can't keep any dqlocks when calling dqrele,
 719		 * because the freelist lock comes before dqlocks.
 720		 */
 721		xfs_dqunlock(udq);
 722		/*
 723		 * we took a hard reference once upon a time in dqget,
 724		 * so give it back when the udquot no longer points at it
 725		 * dqput() does the unlocking of the dquot.
 726		 */
 727		xfs_qm_dqrele(tmp);
 728
 729		xfs_dqlock(udq);
 730		xfs_dqlock(gdq);
 731
 732	} else {
 733		ASSERT(XFS_DQ_IS_LOCKED(udq));
 734		xfs_dqlock(gdq);
 735	}
 736
 737	ASSERT(XFS_DQ_IS_LOCKED(udq));
 738	ASSERT(XFS_DQ_IS_LOCKED(gdq));
 739	/*
 740	 * Somebody could have attached a gdquot here,
 741	 * when we dropped the uqlock. If so, just do nothing.
 742	 */
 743	if (udq->q_gdquot == NULL) {
 744		XFS_DQHOLD(gdq);
 745		udq->q_gdquot = gdq;
 746	}
 747
 748	xfs_dqunlock(gdq);
 749	xfs_dqunlock(udq);
 
 
 
 
 
 
 
 750}
 751
 752
 753/*
 754 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 755 * into account.
 756 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 757 * Inode may get unlocked and relocked in here, and the caller must deal with
 758 * the consequences.
 759 */
 760int
 761xfs_qm_dqattach_locked(
 762	xfs_inode_t	*ip,
 763	uint		flags)
 764{
 765	xfs_mount_t	*mp = ip->i_mount;
 766	uint		nquotas = 0;
 767	int		error = 0;
 768
 769	if (!XFS_IS_QUOTA_RUNNING(mp) ||
 770	    !XFS_IS_QUOTA_ON(mp) ||
 771	    !XFS_NOT_DQATTACHED(mp, ip) ||
 772	    ip->i_ino == mp->m_sb.sb_uquotino ||
 773	    ip->i_ino == mp->m_sb.sb_gquotino)
 774		return 0;
 775
 776	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 777
 778	if (XFS_IS_UQUOTA_ON(mp)) {
 779		error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
 780						flags & XFS_QMOPT_DQALLOC,
 781						NULL, &ip->i_udquot);
 782		if (error)
 783			goto done;
 784		nquotas++;
 785	}
 786
 787	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 788	if (XFS_IS_OQUOTA_ON(mp)) {
 789		error = XFS_IS_GQUOTA_ON(mp) ?
 790			xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
 791						flags & XFS_QMOPT_DQALLOC,
 792						ip->i_udquot, &ip->i_gdquot) :
 793			xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
 794						flags & XFS_QMOPT_DQALLOC,
 795						ip->i_udquot, &ip->i_gdquot);
 796		/*
 797		 * Don't worry about the udquot that we may have
 798		 * attached above. It'll get detached, if not already.
 799		 */
 800		if (error)
 801			goto done;
 802		nquotas++;
 803	}
 804
 805	/*
 806	 * Attach this group quota to the user quota as a hint.
 807	 * This WON'T, in general, result in a thrash.
 808	 */
 809	if (nquotas == 2) {
 810		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 811		ASSERT(ip->i_udquot);
 812		ASSERT(ip->i_gdquot);
 
 813
 814		/*
 815		 * We may or may not have the i_udquot locked at this point,
 816		 * but this check is OK since we don't depend on the i_gdquot to
 817		 * be accurate 100% all the time. It is just a hint, and this
 818		 * will succeed in general.
 819		 */
 820		if (ip->i_udquot->q_gdquot == ip->i_gdquot)
 821			goto done;
 822		/*
 823		 * Attach i_gdquot to the gdquot hint inside the i_udquot.
 824		 */
 825		xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
 826	}
 827
 828 done:
 829#ifdef DEBUG
 830	if (!error) {
 831		if (XFS_IS_UQUOTA_ON(mp))
 832			ASSERT(ip->i_udquot);
 833		if (XFS_IS_OQUOTA_ON(mp))
 834			ASSERT(ip->i_gdquot);
 835	}
 836	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 837#endif
 838	return error;
 839}
 840
 841int
 842xfs_qm_dqattach(
 843	struct xfs_inode	*ip,
 844	uint			flags)
 845{
 846	int			error;
 847
 
 
 
 848	xfs_ilock(ip, XFS_ILOCK_EXCL);
 849	error = xfs_qm_dqattach_locked(ip, flags);
 850	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 851
 852	return error;
 853}
 854
 855/*
 856 * Release dquots (and their references) if any.
 857 * The inode should be locked EXCL except when this's called by
 858 * xfs_ireclaim.
 859 */
 860void
 861xfs_qm_dqdetach(
 862	xfs_inode_t	*ip)
 863{
 864	if (!(ip->i_udquot || ip->i_gdquot))
 865		return;
 866
 867	trace_xfs_dquot_dqdetach(ip);
 868
 869	ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
 870	ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
 871	if (ip->i_udquot) {
 872		xfs_qm_dqrele(ip->i_udquot);
 873		ip->i_udquot = NULL;
 874	}
 875	if (ip->i_gdquot) {
 876		xfs_qm_dqrele(ip->i_gdquot);
 877		ip->i_gdquot = NULL;
 878	}
 
 
 
 
 879}
 880
 881int
 882xfs_qm_sync(
 883	struct xfs_mount	*mp,
 884	int			flags)
 885{
 886	struct xfs_quotainfo	*q = mp->m_quotainfo;
 887	int			recl, restarts;
 888	struct xfs_dquot	*dqp;
 889	int			error;
 890
 891	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
 892		return 0;
 
 
 
 
 
 
 
 
 
 893
 894	restarts = 0;
 
 895
 896  again:
 897	mutex_lock(&q->qi_dqlist_lock);
 898	/*
 899	 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
 900	 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
 901	 * when we have the mplist lock, we know that dquots will be consistent
 902	 * as long as we have it locked.
 903	 */
 904	if (!XFS_IS_QUOTA_ON(mp)) {
 905		mutex_unlock(&q->qi_dqlist_lock);
 906		return 0;
 907	}
 908	ASSERT(mutex_is_locked(&q->qi_dqlist_lock));
 909	list_for_each_entry(dqp, &q->qi_dqlist, q_mplist) {
 910		/*
 911		 * If this is vfs_sync calling, then skip the dquots that
 912		 * don't 'seem' to be dirty. ie. don't acquire dqlock.
 913		 * This is very similar to what xfs_sync does with inodes.
 914		 */
 915		if (flags & SYNC_TRYLOCK) {
 916			if (!XFS_DQ_IS_DIRTY(dqp))
 917				continue;
 918			if (!xfs_qm_dqlock_nowait(dqp))
 919				continue;
 920		} else {
 921			xfs_dqlock(dqp);
 922		}
 923
 924		/*
 925		 * Now, find out for sure if this dquot is dirty or not.
 926		 */
 927		if (! XFS_DQ_IS_DIRTY(dqp)) {
 928			xfs_dqunlock(dqp);
 929			continue;
 930		}
 931
 932		/* XXX a sentinel would be better */
 933		recl = q->qi_dqreclaims;
 934		if (!xfs_dqflock_nowait(dqp)) {
 935			if (flags & SYNC_TRYLOCK) {
 936				xfs_dqunlock(dqp);
 937				continue;
 938			}
 939			/*
 940			 * If we can't grab the flush lock then if the caller
 941			 * really wanted us to give this our best shot, so
 942			 * see if we can give a push to the buffer before we wait
 943			 * on the flush lock. At this point, we know that
 944			 * even though the dquot is being flushed,
 945			 * it has (new) dirty data.
 946			 */
 947			xfs_qm_dqflock_pushbuf_wait(dqp);
 948		}
 949		/*
 950		 * Let go of the mplist lock. We don't want to hold it
 951		 * across a disk write
 952		 */
 953		mutex_unlock(&q->qi_dqlist_lock);
 954		error = xfs_qm_dqflush(dqp, flags);
 955		xfs_dqunlock(dqp);
 956		if (error && XFS_FORCED_SHUTDOWN(mp))
 957			return 0;	/* Need to prevent umount failure */
 958		else if (error)
 959			return error;
 
 
 960
 961		mutex_lock(&q->qi_dqlist_lock);
 962		if (recl != q->qi_dqreclaims) {
 963			if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
 964				break;
 965
 966			mutex_unlock(&q->qi_dqlist_lock);
 967			goto again;
 
 
 
 
 
 
 968		}
 
 
 
 
 969	}
 
 970
 971	mutex_unlock(&q->qi_dqlist_lock);
 972	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 973}
 974
 975/*
 976 * The hash chains and the mplist use the same xfs_dqhash structure as
 977 * their list head, but we can take the mplist qh_lock and one of the
 978 * hash qh_locks at the same time without any problem as they aren't
 979 * related.
 980 */
 981static struct lock_class_key xfs_quota_mplist_class;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 982
 983/*
 984 * This initializes all the quota information that's kept in the
 985 * mount structure
 986 */
 987STATIC int
 988xfs_qm_init_quotainfo(
 989	xfs_mount_t	*mp)
 990{
 991	xfs_quotainfo_t *qinf;
 992	int		error;
 993	xfs_dquot_t	*dqp;
 994
 995	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 996
 997	/*
 998	 * Tell XQM that we exist as soon as possible.
 999	 */
1000	if ((error = xfs_qm_hold_quotafs_ref(mp))) {
1001		return error;
1002	}
1003
1004	qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1005
 
 
 
 
1006	/*
1007	 * See if quotainodes are setup, and if not, allocate them,
1008	 * and change the superblock accordingly.
1009	 */
1010	if ((error = xfs_qm_init_quotainos(mp))) {
1011		kmem_free(qinf);
1012		mp->m_quotainfo = NULL;
1013		return error;
1014	}
1015
1016	INIT_LIST_HEAD(&qinf->qi_dqlist);
1017	mutex_init(&qinf->qi_dqlist_lock);
1018	lockdep_set_class(&qinf->qi_dqlist_lock, &xfs_quota_mplist_class);
1019
1020	qinf->qi_dqreclaims = 0;
 
 
 
1021
1022	/* mutex used to serialize quotaoffs */
1023	mutex_init(&qinf->qi_quotaofflock);
1024
1025	/* Precalc some constants */
1026	qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1027	ASSERT(qinf->qi_dqchunklen);
1028	qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1029	do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1030
1031	mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1032
1033	/*
1034	 * We try to get the limits from the superuser's limits fields.
1035	 * This is quite hacky, but it is standard quota practice.
1036	 * We look at the USR dquot with id == 0 first, but if user quotas
1037	 * are not enabled we goto the GRP dquot with id == 0.
1038	 * We don't really care to keep separate default limits for user
1039	 * and group quotas, at least not at this point.
1040	 */
1041	error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
1042			     XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER : 
1043			     (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1044				XFS_DQ_PROJ),
1045			     XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1046			     &dqp);
1047	if (! error) {
 
 
 
1048		xfs_disk_dquot_t	*ddqp = &dqp->q_core;
1049
1050		/*
1051		 * The warnings and timers set the grace period given to
1052		 * a user or group before he or she can not perform any
1053		 * more writing. If it is zero, a default is used.
1054		 */
1055		qinf->qi_btimelimit = ddqp->d_btimer ?
1056			be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
1057		qinf->qi_itimelimit = ddqp->d_itimer ?
1058			be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
1059		qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
1060			be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
1061		qinf->qi_bwarnlimit = ddqp->d_bwarns ?
1062			be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
1063		qinf->qi_iwarnlimit = ddqp->d_iwarns ?
1064			be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
1065		qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
1066			be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
1067		qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
1068		qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
1069		qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
1070		qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
1071		qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
1072		qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
1073 
1074		/*
1075		 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1076		 * we don't want this dquot cached. We haven't done a
1077		 * quotacheck yet, and quotacheck doesn't like incore dquots.
1078		 */
1079		xfs_qm_dqdestroy(dqp);
1080	} else {
1081		qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1082		qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1083		qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1084		qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1085		qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
1086		qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
1087	}
1088
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1089	return 0;
1090}
1091
 
 
 
 
 
 
 
 
 
 
 
1092
1093/*
1094 * Gets called when unmounting a filesystem or when all quotas get
1095 * turned off.
1096 * This purges the quota inodes, destroys locks and frees itself.
1097 */
1098void
1099xfs_qm_destroy_quotainfo(
1100	xfs_mount_t	*mp)
1101{
1102	xfs_quotainfo_t *qi;
1103
1104	qi = mp->m_quotainfo;
1105	ASSERT(qi != NULL);
1106	ASSERT(xfs_Gqm != NULL);
1107
1108	/*
1109	 * Release the reference that XQM kept, so that we know
1110	 * when the XQM structure should be freed. We cannot assume
1111	 * that xfs_Gqm is non-null after this point.
1112	 */
1113	xfs_qm_rele_quotafs_ref(mp);
1114
1115	ASSERT(list_empty(&qi->qi_dqlist));
1116	mutex_destroy(&qi->qi_dqlist_lock);
1117
1118	if (qi->qi_uquotaip) {
1119		IRELE(qi->qi_uquotaip);
1120		qi->qi_uquotaip = NULL; /* paranoia */
1121	}
1122	if (qi->qi_gquotaip) {
1123		IRELE(qi->qi_gquotaip);
1124		qi->qi_gquotaip = NULL;
1125	}
1126	mutex_destroy(&qi->qi_quotaofflock);
1127	kmem_free(qi);
1128	mp->m_quotainfo = NULL;
1129}
1130
1131
1132
1133/* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1134
1135/* ARGSUSED */
1136STATIC void
1137xfs_qm_list_init(
1138	xfs_dqlist_t	*list,
1139	char		*str,
1140	int		n)
1141{
1142	mutex_init(&list->qh_lock);
1143	INIT_LIST_HEAD(&list->qh_list);
1144	list->qh_version = 0;
1145	list->qh_nelems = 0;
1146}
1147
1148STATIC void
1149xfs_qm_list_destroy(
1150	xfs_dqlist_t	*list)
1151{
1152	mutex_destroy(&(list->qh_lock));
1153}
1154
1155/*
1156 * Create an inode and return with a reference already taken, but unlocked
1157 * This is how we create quota inodes
1158 */
1159STATIC int
1160xfs_qm_qino_alloc(
1161	xfs_mount_t	*mp,
1162	xfs_inode_t	**ip,
1163	__int64_t	sbfields,
1164	uint		flags)
1165{
1166	xfs_trans_t	*tp;
1167	int		error;
1168	int		committed;
1169
1170	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1171	if ((error = xfs_trans_reserve(tp,
1172				      XFS_QM_QINOCREATE_SPACE_RES(mp),
1173				      XFS_CREATE_LOG_RES(mp), 0,
1174				      XFS_TRANS_PERM_LOG_RES,
1175				      XFS_CREATE_LOG_COUNT))) {
1176		xfs_trans_cancel(tp, 0);
1177		return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1178	}
1179
1180	error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, &committed);
1181	if (error) {
1182		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1183				 XFS_TRANS_ABORT);
1184		return error;
 
 
 
 
 
 
 
1185	}
1186
1187	/*
1188	 * Make the changes in the superblock, and log those too.
1189	 * sbfields arg may contain fields other than *QUOTINO;
1190	 * VERSIONNUM for example.
1191	 */
1192	spin_lock(&mp->m_sb_lock);
1193	if (flags & XFS_QMOPT_SBVERSION) {
1194		ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1195		ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1196				   XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1197		       (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1198			XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1199
1200		xfs_sb_version_addquota(&mp->m_sb);
1201		mp->m_sb.sb_uquotino = NULLFSINO;
1202		mp->m_sb.sb_gquotino = NULLFSINO;
 
1203
1204		/* qflags will get updated _after_ quotacheck */
1205		mp->m_sb.sb_qflags = 0;
1206	}
1207	if (flags & XFS_QMOPT_UQUOTA)
1208		mp->m_sb.sb_uquotino = (*ip)->i_ino;
1209	else
1210		mp->m_sb.sb_gquotino = (*ip)->i_ino;
 
 
1211	spin_unlock(&mp->m_sb_lock);
1212	xfs_mod_sb(tp, sbfields);
1213
1214	if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
 
 
1215		xfs_alert(mp, "%s failed (error %d)!", __func__, error);
1216		return error;
1217	}
1218	return 0;
 
 
1219}
1220
1221
1222STATIC void
1223xfs_qm_reset_dqcounts(
1224	xfs_mount_t	*mp,
1225	xfs_buf_t	*bp,
1226	xfs_dqid_t	id,
1227	uint		type)
1228{
1229	xfs_disk_dquot_t	*ddq;
1230	int			j;
 
1231
1232	trace_xfs_reset_dqcounts(bp, _RET_IP_);
1233
1234	/*
1235	 * Reset all counters and timers. They'll be
1236	 * started afresh by xfs_qm_quotacheck.
1237	 */
1238#ifdef DEBUG
1239	j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1240	do_div(j, sizeof(xfs_dqblk_t));
1241	ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1242#endif
1243	ddq = bp->b_addr;
1244	for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
 
 
 
 
1245		/*
1246		 * Do a sanity check, and if needed, repair the dqblk. Don't
1247		 * output any warnings because it's perfectly possible to
1248		 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
 
 
 
 
 
 
 
 
 
1249		 */
1250		(void) xfs_qm_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1251				      "xfs_quotacheck");
1252		ddq->d_bcount = 0;
1253		ddq->d_icount = 0;
1254		ddq->d_rtbcount = 0;
1255		ddq->d_btimer = 0;
1256		ddq->d_itimer = 0;
1257		ddq->d_rtbtimer = 0;
1258		ddq->d_bwarns = 0;
1259		ddq->d_iwarns = 0;
1260		ddq->d_rtbwarns = 0;
1261		ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
 
 
 
 
 
1262	}
1263}
1264
1265STATIC int
1266xfs_qm_dqiter_bufs(
1267	xfs_mount_t	*mp,
1268	xfs_dqid_t	firstid,
1269	xfs_fsblock_t	bno,
1270	xfs_filblks_t	blkcnt,
1271	uint		flags)
 
1272{
1273	xfs_buf_t	*bp;
1274	int		error;
1275	int		type;
1276
1277	ASSERT(blkcnt > 0);
1278	type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1279		(flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1280	error = 0;
1281
1282	/*
1283	 * Blkcnt arg can be a very big number, and might even be
1284	 * larger than the log itself. So, we have to break it up into
1285	 * manageable-sized transactions.
1286	 * Note that we don't start a permanent transaction here; we might
1287	 * not be able to get a log reservation for the whole thing up front,
1288	 * and we don't really care to either, because we just discard
1289	 * everything if we were to crash in the middle of this loop.
1290	 */
1291	while (blkcnt--) {
1292		error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1293			      XFS_FSB_TO_DADDR(mp, bno),
1294			      mp->m_quotainfo->qi_dqchunklen, 0, &bp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1295		if (error)
1296			break;
1297
1298		xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1299		xfs_bdwrite(mp, bp);
1300		/*
1301		 * goto the next block.
 
 
1302		 */
 
 
 
 
 
 
1303		bno++;
1304		firstid += mp->m_quotainfo->qi_dqperchunk;
1305	}
 
1306	return error;
1307}
1308
1309/*
1310 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1311 * caller supplied function for every chunk of dquots that we find.
1312 */
1313STATIC int
1314xfs_qm_dqiterate(
1315	xfs_mount_t	*mp,
1316	xfs_inode_t	*qip,
1317	uint		flags)
 
1318{
1319	xfs_bmbt_irec_t		*map;
1320	int			i, nmaps;	/* number of map entries */
1321	int			error;		/* return value */
1322	xfs_fileoff_t		lblkno;
1323	xfs_filblks_t		maxlblkcnt;
1324	xfs_dqid_t		firstid;
1325	xfs_fsblock_t		rablkno;
1326	xfs_filblks_t		rablkcnt;
1327
1328	error = 0;
1329	/*
1330	 * This looks racy, but we can't keep an inode lock across a
1331	 * trans_reserve. But, this gets called during quotacheck, and that
1332	 * happens only at mount time which is single threaded.
1333	 */
1334	if (qip->i_d.di_nblocks == 0)
1335		return 0;
1336
1337	map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1338
1339	lblkno = 0;
1340	maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1341	do {
 
 
1342		nmaps = XFS_DQITER_MAP_SIZE;
1343		/*
1344		 * We aren't changing the inode itself. Just changing
1345		 * some of its data. No new blocks are added here, and
1346		 * the inode is never added to the transaction.
1347		 */
1348		xfs_ilock(qip, XFS_ILOCK_SHARED);
1349		error = xfs_bmapi(NULL, qip, lblkno,
1350				  maxlblkcnt - lblkno,
1351				  XFS_BMAPI_METADATA,
1352				  NULL,
1353				  0, map, &nmaps, NULL);
1354		xfs_iunlock(qip, XFS_ILOCK_SHARED);
1355		if (error)
1356			break;
1357
1358		ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1359		for (i = 0; i < nmaps; i++) {
1360			ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1361			ASSERT(map[i].br_blockcount);
1362
1363
1364			lblkno += map[i].br_blockcount;
1365
1366			if (map[i].br_startblock == HOLESTARTBLOCK)
1367				continue;
1368
1369			firstid = (xfs_dqid_t) map[i].br_startoff *
1370				mp->m_quotainfo->qi_dqperchunk;
1371			/*
1372			 * Do a read-ahead on the next extent.
1373			 */
1374			if ((i+1 < nmaps) &&
1375			    (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1376				rablkcnt =  map[i+1].br_blockcount;
1377				rablkno = map[i+1].br_startblock;
1378				while (rablkcnt--) {
1379					xfs_buf_readahead(mp->m_ddev_targp,
1380					       XFS_FSB_TO_DADDR(mp, rablkno),
1381					       mp->m_quotainfo->qi_dqchunklen);
 
1382					rablkno++;
1383				}
1384			}
1385			/*
1386			 * Iterate thru all the blks in the extent and
1387			 * reset the counters of all the dquots inside them.
1388			 */
1389			if ((error = xfs_qm_dqiter_bufs(mp,
1390						       firstid,
1391						       map[i].br_startblock,
1392						       map[i].br_blockcount,
1393						       flags))) {
1394				break;
1395			}
1396		}
1397
1398		if (error)
1399			break;
1400	} while (nmaps > 0);
1401
 
1402	kmem_free(map);
1403
1404	return error;
1405}
1406
1407/*
1408 * Called by dqusage_adjust in doing a quotacheck.
1409 *
1410 * Given the inode, and a dquot id this updates both the incore dqout as well
1411 * as the buffer copy. This is so that once the quotacheck is done, we can
1412 * just log all the buffers, as opposed to logging numerous updates to
1413 * individual dquots.
1414 */
1415STATIC int
1416xfs_qm_quotacheck_dqadjust(
1417	struct xfs_inode	*ip,
1418	xfs_dqid_t		id,
1419	uint			type,
1420	xfs_qcnt_t		nblks,
1421	xfs_qcnt_t		rtblks)
1422{
1423	struct xfs_mount	*mp = ip->i_mount;
1424	struct xfs_dquot	*dqp;
1425	int			error;
1426
1427	error = xfs_qm_dqget(mp, ip, id, type,
1428			     XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1429	if (error) {
1430		/*
1431		 * Shouldn't be able to turn off quotas here.
1432		 */
1433		ASSERT(error != ESRCH);
1434		ASSERT(error != ENOENT);
1435		return error;
1436	}
1437
1438	trace_xfs_dqadjust(dqp);
1439
1440	/*
1441	 * Adjust the inode count and the block count to reflect this inode's
1442	 * resource usage.
1443	 */
1444	be64_add_cpu(&dqp->q_core.d_icount, 1);
1445	dqp->q_res_icount++;
1446	if (nblks) {
1447		be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1448		dqp->q_res_bcount += nblks;
1449	}
1450	if (rtblks) {
1451		be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1452		dqp->q_res_rtbcount += rtblks;
1453	}
1454
1455	/*
1456	 * Set default limits, adjust timers (since we changed usages)
1457	 *
1458	 * There are no timers for the default values set in the root dquot.
1459	 */
1460	if (dqp->q_core.d_id) {
1461		xfs_qm_adjust_dqlimits(mp, &dqp->q_core);
1462		xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1463	}
1464
1465	dqp->dq_flags |= XFS_DQ_DIRTY;
1466	xfs_qm_dqput(dqp);
1467	return 0;
1468}
1469
1470STATIC int
1471xfs_qm_get_rtblks(
1472	xfs_inode_t	*ip,
1473	xfs_qcnt_t	*O_rtblks)
1474{
1475	xfs_filblks_t	rtblks;			/* total rt blks */
1476	xfs_extnum_t	idx;			/* extent record index */
1477	xfs_ifork_t	*ifp;			/* inode fork pointer */
1478	xfs_extnum_t	nextents;		/* number of extent entries */
1479	int		error;
1480
1481	ASSERT(XFS_IS_REALTIME_INODE(ip));
1482	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1483	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1484		if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1485			return error;
1486	}
1487	rtblks = 0;
1488	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1489	for (idx = 0; idx < nextents; idx++)
1490		rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1491	*O_rtblks = (xfs_qcnt_t)rtblks;
1492	return 0;
1493}
1494
1495/*
1496 * callback routine supplied to bulkstat(). Given an inumber, find its
1497 * dquots and update them to account for resources taken by that inode.
1498 */
1499/* ARGSUSED */
1500STATIC int
1501xfs_qm_dqusage_adjust(
1502	xfs_mount_t	*mp,		/* mount point for filesystem */
1503	xfs_ino_t	ino,		/* inode number to get data for */
1504	void		__user *buffer,	/* not used */
1505	int		ubsize,		/* not used */
1506	int		*ubused,	/* not used */
1507	int		*res)		/* result code value */
1508{
1509	xfs_inode_t	*ip;
1510	xfs_qcnt_t	nblks, rtblks = 0;
 
1511	int		error;
1512
1513	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1514
1515	/*
1516	 * rootino must have its resources accounted for, not so with the quota
1517	 * inodes.
1518	 */
1519	if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1520		*res = BULKSTAT_RV_NOTHING;
1521		return XFS_ERROR(EINVAL);
1522	}
1523
1524	/*
1525	 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1526	 * interface expects the inode to be exclusively locked because that's
1527	 * the case in all other instances. It's OK that we do this because
1528	 * quotacheck is done only at mount time.
1529	 */
1530	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
 
1531	if (error) {
1532		*res = BULKSTAT_RV_NOTHING;
1533		return error;
1534	}
1535
1536	ASSERT(ip->i_delayed_blks == 0);
1537
1538	if (XFS_IS_REALTIME_INODE(ip)) {
1539		/*
1540		 * Walk thru the extent list and count the realtime blocks.
1541		 */
1542		error = xfs_qm_get_rtblks(ip, &rtblks);
1543		if (error)
1544			goto error0;
 
 
 
1545	}
1546
1547	nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1548
1549	/*
1550	 * Add the (disk blocks and inode) resources occupied by this
1551	 * inode to its dquots. We do this adjustment in the incore dquot,
1552	 * and also copy the changes to its buffer.
1553	 * We don't care about putting these changes in a transaction
1554	 * envelope because if we crash in the middle of a 'quotacheck'
1555	 * we have to start from the beginning anyway.
1556	 * Once we're done, we'll log all the dquot bufs.
1557	 *
1558	 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1559	 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1560	 */
1561	if (XFS_IS_UQUOTA_ON(mp)) {
1562		error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1563						   XFS_DQ_USER, nblks, rtblks);
1564		if (error)
1565			goto error0;
1566	}
1567
1568	if (XFS_IS_GQUOTA_ON(mp)) {
1569		error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1570						   XFS_DQ_GROUP, nblks, rtblks);
1571		if (error)
1572			goto error0;
1573	}
1574
1575	if (XFS_IS_PQUOTA_ON(mp)) {
1576		error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1577						   XFS_DQ_PROJ, nblks, rtblks);
1578		if (error)
1579			goto error0;
1580	}
1581
1582	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1583	IRELE(ip);
1584	*res = BULKSTAT_RV_DIDONE;
1585	return 0;
1586
1587error0:
1588	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1589	IRELE(ip);
1590	*res = BULKSTAT_RV_GIVEUP;
1591	return error;
1592}
1593
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1594/*
1595 * Walk thru all the filesystem inodes and construct a consistent view
1596 * of the disk quota world. If the quotacheck fails, disable quotas.
1597 */
1598int
1599xfs_qm_quotacheck(
1600	xfs_mount_t	*mp)
1601{
1602	int		done, count, error;
1603	xfs_ino_t	lastino;
1604	size_t		structsz;
1605	xfs_inode_t	*uip, *gip;
1606	uint		flags;
 
 
 
1607
1608	count = INT_MAX;
1609	structsz = 1;
1610	lastino = 0;
1611	flags = 0;
1612
1613	ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1614	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1615
1616	/*
1617	 * There should be no cached dquots. The (simplistic) quotacheck
1618	 * algorithm doesn't like that.
1619	 */
1620	ASSERT(list_empty(&mp->m_quotainfo->qi_dqlist));
1621
1622	xfs_notice(mp, "Quotacheck needed: Please wait.");
1623
1624	/*
1625	 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1626	 * their counters to zero. We need a clean slate.
1627	 * We don't log our changes till later.
1628	 */
1629	uip = mp->m_quotainfo->qi_uquotaip;
1630	if (uip) {
1631		error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA);
 
1632		if (error)
1633			goto error_return;
1634		flags |= XFS_UQUOTA_CHKD;
1635	}
1636
1637	gip = mp->m_quotainfo->qi_gquotaip;
1638	if (gip) {
1639		error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1640					XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1641		if (error)
1642			goto error_return;
1643		flags |= XFS_OQUOTA_CHKD;
 
 
 
 
 
 
 
 
1644	}
1645
1646	do {
1647		/*
1648		 * Iterate thru all the inodes in the file system,
1649		 * adjusting the corresponding dquot counters in core.
1650		 */
1651		error = xfs_bulkstat(mp, &lastino, &count,
1652				     xfs_qm_dqusage_adjust,
1653				     structsz, NULL, &done);
1654		if (error)
1655			break;
1656
1657	} while (!done);
1658
1659	/*
1660	 * We've made all the changes that we need to make incore.
1661	 * Flush them down to disk buffers if everything was updated
1662	 * successfully.
1663	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1664	if (!error)
1665		error = xfs_qm_dqflush_all(mp, 0);
1666
1667	/*
1668	 * We can get this error if we couldn't do a dquot allocation inside
1669	 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1670	 * dirty dquots that might be cached, we just want to get rid of them
1671	 * and turn quotaoff. The dquots won't be attached to any of the inodes
1672	 * at this point (because we intentionally didn't in dqget_noattach).
1673	 */
1674	if (error) {
1675		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1676		goto error_return;
1677	}
1678
1679	/*
1680	 * We didn't log anything, because if we crashed, we'll have to
1681	 * start the quotacheck from scratch anyway. However, we must make
1682	 * sure that our dquot changes are secure before we put the
1683	 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1684	 * flush.
1685	 */
1686	XFS_bflush(mp->m_ddev_targp);
1687
1688	/*
1689	 * If one type of quotas is off, then it will lose its
1690	 * quotachecked status, since we won't be doing accounting for
1691	 * that type anymore.
1692	 */
1693	mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
1694	mp->m_qflags |= flags;
1695
1696 error_return:
 
 
1697	if (error) {
1698		xfs_warn(mp,
1699	"Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1700			error);
1701		/*
1702		 * We must turn off quotas.
1703		 */
1704		ASSERT(mp->m_quotainfo != NULL);
1705		ASSERT(xfs_Gqm != NULL);
1706		xfs_qm_destroy_quotainfo(mp);
1707		if (xfs_mount_reset_sbqflags(mp)) {
1708			xfs_warn(mp,
1709				"Quotacheck: Failed to reset quota flags.");
1710		}
1711	} else
1712		xfs_notice(mp, "Quotacheck: Done.");
1713	return (error);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1714}
1715
1716/*
1717 * This is called after the superblock has been read in and we're ready to
1718 * iget the quota inodes.
1719 */
1720STATIC int
1721xfs_qm_init_quotainos(
1722	xfs_mount_t	*mp)
1723{
1724	xfs_inode_t	*uip, *gip;
1725	int		error;
1726	__int64_t	sbflags;
1727	uint		flags;
 
1728
1729	ASSERT(mp->m_quotainfo);
1730	uip = gip = NULL;
1731	sbflags = 0;
1732	flags = 0;
1733
1734	/*
1735	 * Get the uquota and gquota inodes
1736	 */
1737	if (xfs_sb_version_hasquota(&mp->m_sb)) {
1738		if (XFS_IS_UQUOTA_ON(mp) &&
1739		    mp->m_sb.sb_uquotino != NULLFSINO) {
1740			ASSERT(mp->m_sb.sb_uquotino > 0);
1741			if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1742					     0, 0, &uip)))
1743				return XFS_ERROR(error);
 
1744		}
1745		if (XFS_IS_OQUOTA_ON(mp) &&
1746		    mp->m_sb.sb_gquotino != NULLFSINO) {
1747			ASSERT(mp->m_sb.sb_gquotino > 0);
1748			if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1749					     0, 0, &gip))) {
1750				if (uip)
1751					IRELE(uip);
1752				return XFS_ERROR(error);
1753			}
 
 
 
 
 
 
1754		}
1755	} else {
1756		flags |= XFS_QMOPT_SBVERSION;
1757		sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1758			    XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1759	}
1760
1761	/*
1762	 * Create the two inodes, if they don't exist already. The changes
1763	 * made above will get added to a transaction and logged in one of
1764	 * the qino_alloc calls below.  If the device is readonly,
1765	 * temporarily switch to read-write to do this.
1766	 */
1767	if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1768		if ((error = xfs_qm_qino_alloc(mp, &uip,
1769					      sbflags | XFS_SB_UQUOTINO,
1770					      flags | XFS_QMOPT_UQUOTA)))
1771			return XFS_ERROR(error);
1772
1773		flags &= ~XFS_QMOPT_SBVERSION;
1774	}
1775	if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1776		flags |= (XFS_IS_GQUOTA_ON(mp) ?
1777				XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1778		error = xfs_qm_qino_alloc(mp, &gip,
1779					  sbflags | XFS_SB_GQUOTINO, flags);
1780		if (error) {
1781			if (uip)
1782				IRELE(uip);
1783
1784			return XFS_ERROR(error);
1785		}
 
 
 
 
 
1786	}
1787
1788	mp->m_quotainfo->qi_uquotaip = uip;
1789	mp->m_quotainfo->qi_gquotaip = gip;
 
1790
1791	return 0;
1792}
1793
1794
1795
1796/*
1797 * Just pop the least recently used dquot off the freelist and
1798 * recycle it. The returned dquot is locked.
1799 */
1800STATIC xfs_dquot_t *
1801xfs_qm_dqreclaim_one(void)
1802{
1803	xfs_dquot_t	*dqpout;
1804	xfs_dquot_t	*dqp;
1805	int		restarts;
1806	int		startagain;
1807
1808	restarts = 0;
1809	dqpout = NULL;
1810
1811	/* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
1812again:
1813	startagain = 0;
1814	mutex_lock(&xfs_Gqm->qm_dqfrlist_lock);
1815
1816	list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) {
1817		struct xfs_mount *mp = dqp->q_mount;
1818		xfs_dqlock(dqp);
1819
1820		/*
1821		 * We are racing with dqlookup here. Naturally we don't
1822		 * want to reclaim a dquot that lookup wants. We release the
1823		 * freelist lock and start over, so that lookup will grab
1824		 * both the dquot and the freelistlock.
1825		 */
1826		if (dqp->dq_flags & XFS_DQ_WANT) {
1827			ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
1828
1829			trace_xfs_dqreclaim_want(dqp);
1830			XQM_STATS_INC(xqmstats.xs_qm_dqwants);
1831			restarts++;
1832			startagain = 1;
1833			goto dqunlock;
1834		}
1835
1836		/*
1837		 * If the dquot is inactive, we are assured that it is
1838		 * not on the mplist or the hashlist, and that makes our
1839		 * life easier.
1840		 */
1841		if (dqp->dq_flags & XFS_DQ_INACTIVE) {
1842			ASSERT(mp == NULL);
1843			ASSERT(! XFS_DQ_IS_DIRTY(dqp));
1844			ASSERT(list_empty(&dqp->q_hashlist));
1845			ASSERT(list_empty(&dqp->q_mplist));
1846			list_del_init(&dqp->q_freelist);
1847			xfs_Gqm->qm_dqfrlist_cnt--;
1848			dqpout = dqp;
1849			XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
1850			goto dqunlock;
1851		}
1852
1853		ASSERT(dqp->q_hash);
1854		ASSERT(!list_empty(&dqp->q_mplist));
1855
1856		/*
1857		 * Try to grab the flush lock. If this dquot is in the process
1858		 * of getting flushed to disk, we don't want to reclaim it.
1859		 */
1860		if (!xfs_dqflock_nowait(dqp))
1861			goto dqunlock;
1862
1863		/*
1864		 * We have the flush lock so we know that this is not in the
1865		 * process of being flushed. So, if this is dirty, flush it
1866		 * DELWRI so that we don't get a freelist infested with
1867		 * dirty dquots.
1868		 */
1869		if (XFS_DQ_IS_DIRTY(dqp)) {
1870			int	error;
1871
1872			trace_xfs_dqreclaim_dirty(dqp);
1873
1874			/*
1875			 * We flush it delayed write, so don't bother
1876			 * releasing the freelist lock.
1877			 */
1878			error = xfs_qm_dqflush(dqp, 0);
1879			if (error) {
1880				xfs_warn(mp, "%s: dquot %p flush failed",
1881					__func__, dqp);
1882			}
1883			goto dqunlock;
1884		}
1885
1886		/*
1887		 * We're trying to get the hashlock out of order. This races
1888		 * with dqlookup; so, we giveup and goto the next dquot if
1889		 * we couldn't get the hashlock. This way, we won't starve
1890		 * a dqlookup process that holds the hashlock that is
1891		 * waiting for the freelist lock.
1892		 */
1893		if (!mutex_trylock(&dqp->q_hash->qh_lock)) {
1894			restarts++;
1895			goto dqfunlock;
1896		}
1897
1898		/*
1899		 * This races with dquot allocation code as well as dqflush_all
1900		 * and reclaim code. So, if we failed to grab the mplist lock,
1901		 * giveup everything and start over.
1902		 */
1903		if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) {
1904			restarts++;
1905			startagain = 1;
1906			goto qhunlock;
1907		}
1908
1909		ASSERT(dqp->q_nrefs == 0);
1910		list_del_init(&dqp->q_mplist);
1911		mp->m_quotainfo->qi_dquots--;
1912		mp->m_quotainfo->qi_dqreclaims++;
1913		list_del_init(&dqp->q_hashlist);
1914		dqp->q_hash->qh_version++;
1915		list_del_init(&dqp->q_freelist);
1916		xfs_Gqm->qm_dqfrlist_cnt--;
1917		dqpout = dqp;
1918		mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock);
1919qhunlock:
1920		mutex_unlock(&dqp->q_hash->qh_lock);
1921dqfunlock:
1922		xfs_dqfunlock(dqp);
1923dqunlock:
1924		xfs_dqunlock(dqp);
1925		if (dqpout)
1926			break;
1927		if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
1928			break;
1929		if (startagain) {
1930			mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1931			goto again;
1932		}
1933	}
1934	mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock);
1935	return dqpout;
1936}
1937
1938/*
1939 * Traverse the freelist of dquots and attempt to reclaim a maximum of
1940 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
1941 * favor the lookup function ...
1942 */
1943STATIC int
1944xfs_qm_shake_freelist(
1945	int	howmany)
1946{
1947	int		nreclaimed = 0;
1948	xfs_dquot_t	*dqp;
1949
1950	if (howmany <= 0)
1951		return 0;
1952
1953	while (nreclaimed < howmany) {
1954		dqp = xfs_qm_dqreclaim_one();
1955		if (!dqp)
1956			return nreclaimed;
1957		xfs_qm_dqdestroy(dqp);
1958		nreclaimed++;
1959	}
1960	return nreclaimed;
1961}
1962
1963/*
1964 * The kmem_shake interface is invoked when memory is running low.
1965 */
1966/* ARGSUSED */
1967STATIC int
1968xfs_qm_shake(
1969	struct shrinker	*shrink,
1970	struct shrink_control *sc)
1971{
1972	int	ndqused, nfree, n;
1973	gfp_t gfp_mask = sc->gfp_mask;
1974
1975	if (!kmem_shake_allow(gfp_mask))
1976		return 0;
1977	if (!xfs_Gqm)
1978		return 0;
1979
1980	nfree = xfs_Gqm->qm_dqfrlist_cnt; /* free dquots */
1981	/* incore dquots in all f/s's */
1982	ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
1983
1984	ASSERT(ndqused >= 0);
1985
1986	if (nfree <= ndqused && nfree < ndquot)
1987		return 0;
1988
1989	ndqused *= xfs_Gqm->qm_dqfree_ratio;	/* target # of free dquots */
1990	n = nfree - ndqused - ndquot;		/* # over target */
1991
1992	return xfs_qm_shake_freelist(MAX(nfree, n));
1993}
1994
1995
1996/*------------------------------------------------------------------*/
1997
1998/*
1999 * Return a new incore dquot. Depending on the number of
2000 * dquots in the system, we either allocate a new one on the kernel heap,
2001 * or reclaim a free one.
2002 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2003 * to reclaim an existing one from the freelist.
2004 */
2005boolean_t
2006xfs_qm_dqalloc_incore(
2007	xfs_dquot_t **O_dqpp)
2008{
2009	xfs_dquot_t	*dqp;
2010
2011	/*
2012	 * Check against high water mark to see if we want to pop
2013	 * a nincompoop dquot off the freelist.
2014	 */
2015	if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2016		/*
2017		 * Try to recycle a dquot from the freelist.
2018		 */
2019		if ((dqp = xfs_qm_dqreclaim_one())) {
2020			XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2021			/*
2022			 * Just zero the core here. The rest will get
2023			 * reinitialized by caller. XXX we shouldn't even
2024			 * do this zero ...
2025			 */
2026			memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2027			*O_dqpp = dqp;
2028			return B_FALSE;
2029		}
2030		XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2031	}
2032
2033	/*
2034	 * Allocate a brand new dquot on the kernel heap and return it
2035	 * to the caller to initialize.
2036	 */
2037	ASSERT(xfs_Gqm->qm_dqzone != NULL);
2038	*O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2039	atomic_inc(&xfs_Gqm->qm_totaldquots);
2040
2041	return B_TRUE;
2042}
2043
2044
2045/*
2046 * Start a transaction and write the incore superblock changes to
2047 * disk. flags parameter indicates which fields have changed.
2048 */
2049int
2050xfs_qm_write_sb_changes(
2051	xfs_mount_t	*mp,
2052	__int64_t	flags)
2053{
2054	xfs_trans_t	*tp;
2055	int		error;
2056
2057	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2058	if ((error = xfs_trans_reserve(tp, 0,
2059				      mp->m_sb.sb_sectsize + 128, 0,
2060				      0,
2061				      XFS_DEFAULT_LOG_COUNT))) {
2062		xfs_trans_cancel(tp, 0);
2063		return error;
2064	}
2065
2066	xfs_mod_sb(tp, flags);
2067	error = xfs_trans_commit(tp, 0);
2068
2069	return error;
2070}
2071
2072
2073/* --------------- utility functions for vnodeops ---------------- */
2074
2075
2076/*
2077 * Given an inode, a uid, gid and prid make sure that we have
2078 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2079 * quotas by creating this file.
2080 * This also attaches dquot(s) to the given inode after locking it,
2081 * and returns the dquots corresponding to the uid and/or gid.
2082 *
2083 * in	: inode (unlocked)
2084 * out	: udquot, gdquot with references taken and unlocked
2085 */
2086int
2087xfs_qm_vop_dqalloc(
2088	struct xfs_inode	*ip,
2089	uid_t			uid,
2090	gid_t			gid,
2091	prid_t			prid,
2092	uint			flags,
2093	struct xfs_dquot	**O_udqpp,
2094	struct xfs_dquot	**O_gdqpp)
 
2095{
2096	struct xfs_mount	*mp = ip->i_mount;
2097	struct xfs_dquot	*uq, *gq;
 
 
2098	int			error;
2099	uint			lockflags;
2100
2101	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2102		return 0;
2103
2104	lockflags = XFS_ILOCK_EXCL;
2105	xfs_ilock(ip, lockflags);
2106
2107	if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
2108		gid = ip->i_d.di_gid;
2109
2110	/*
2111	 * Attach the dquot(s) to this inode, doing a dquot allocation
2112	 * if necessary. The dquot(s) will not be locked.
2113	 */
2114	if (XFS_NOT_DQATTACHED(mp, ip)) {
2115		error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
2116		if (error) {
2117			xfs_iunlock(ip, lockflags);
2118			return error;
2119		}
2120	}
2121
2122	uq = gq = NULL;
2123	if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
2124		if (ip->i_d.di_uid != uid) {
2125			/*
2126			 * What we need is the dquot that has this uid, and
2127			 * if we send the inode to dqget, the uid of the inode
2128			 * takes priority over what's sent in the uid argument.
2129			 * We must unlock inode here before calling dqget if
2130			 * we're not sending the inode, because otherwise
2131			 * we'll deadlock by doing trans_reserve while
2132			 * holding ilock.
2133			 */
2134			xfs_iunlock(ip, lockflags);
2135			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2136						 XFS_DQ_USER,
2137						 XFS_QMOPT_DQALLOC |
2138						 XFS_QMOPT_DOWARN,
2139						 &uq))) {
2140				ASSERT(error != ENOENT);
2141				return error;
2142			}
2143			/*
2144			 * Get the ilock in the right order.
2145			 */
2146			xfs_dqunlock(uq);
2147			lockflags = XFS_ILOCK_SHARED;
2148			xfs_ilock(ip, lockflags);
2149		} else {
2150			/*
2151			 * Take an extra reference, because we'll return
2152			 * this to caller
2153			 */
2154			ASSERT(ip->i_udquot);
2155			uq = ip->i_udquot;
2156			xfs_dqlock(uq);
2157			XFS_DQHOLD(uq);
2158			xfs_dqunlock(uq);
2159		}
2160	}
2161	if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
2162		if (ip->i_d.di_gid != gid) {
2163			xfs_iunlock(ip, lockflags);
2164			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2165						 XFS_DQ_GROUP,
2166						 XFS_QMOPT_DQALLOC |
2167						 XFS_QMOPT_DOWARN,
2168						 &gq))) {
2169				if (uq)
2170					xfs_qm_dqrele(uq);
2171				ASSERT(error != ENOENT);
2172				return error;
2173			}
2174			xfs_dqunlock(gq);
2175			lockflags = XFS_ILOCK_SHARED;
2176			xfs_ilock(ip, lockflags);
2177		} else {
2178			ASSERT(ip->i_gdquot);
2179			gq = ip->i_gdquot;
2180			xfs_dqlock(gq);
2181			XFS_DQHOLD(gq);
2182			xfs_dqunlock(gq);
2183		}
2184	} else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
 
2185		if (xfs_get_projid(ip) != prid) {
2186			xfs_iunlock(ip, lockflags);
2187			if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2188						 XFS_DQ_PROJ,
2189						 XFS_QMOPT_DQALLOC |
2190						 XFS_QMOPT_DOWARN,
2191						 &gq))) {
2192				if (uq)
2193					xfs_qm_dqrele(uq);
2194				ASSERT(error != ENOENT);
2195				return (error);
2196			}
2197			xfs_dqunlock(gq);
2198			lockflags = XFS_ILOCK_SHARED;
2199			xfs_ilock(ip, lockflags);
2200		} else {
2201			ASSERT(ip->i_gdquot);
2202			gq = ip->i_gdquot;
2203			xfs_dqlock(gq);
2204			XFS_DQHOLD(gq);
2205			xfs_dqunlock(gq);
2206		}
2207	}
2208	if (uq)
2209		trace_xfs_dquot_dqalloc(ip);
2210
2211	xfs_iunlock(ip, lockflags);
2212	if (O_udqpp)
2213		*O_udqpp = uq;
2214	else if (uq)
2215		xfs_qm_dqrele(uq);
2216	if (O_gdqpp)
2217		*O_gdqpp = gq;
2218	else if (gq)
2219		xfs_qm_dqrele(gq);
 
 
 
 
2220	return 0;
 
 
 
 
 
2221}
2222
2223/*
2224 * Actually transfer ownership, and do dquot modifications.
2225 * These were already reserved.
2226 */
2227xfs_dquot_t *
2228xfs_qm_vop_chown(
2229	xfs_trans_t	*tp,
2230	xfs_inode_t	*ip,
2231	xfs_dquot_t	**IO_olddq,
2232	xfs_dquot_t	*newdq)
2233{
2234	xfs_dquot_t	*prevdq;
2235	uint		bfield = XFS_IS_REALTIME_INODE(ip) ?
2236				 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2237
2238
2239	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2240	ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2241
2242	/* old dquot */
2243	prevdq = *IO_olddq;
2244	ASSERT(prevdq);
2245	ASSERT(prevdq != newdq);
2246
2247	xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2248	xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
2249
2250	/* the sparkling new dquot */
2251	xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2252	xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
2253
2254	/*
2255	 * Take an extra reference, because the inode
2256	 * is going to keep this dquot pointer even
2257	 * after the trans_commit.
2258	 */
2259	xfs_dqlock(newdq);
2260	XFS_DQHOLD(newdq);
2261	xfs_dqunlock(newdq);
2262	*IO_olddq = newdq;
2263
2264	return prevdq;
2265}
2266
2267/*
2268 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
2269 */
2270int
2271xfs_qm_vop_chown_reserve(
2272	xfs_trans_t	*tp,
2273	xfs_inode_t	*ip,
2274	xfs_dquot_t	*udqp,
2275	xfs_dquot_t	*gdqp,
2276	uint		flags)
 
2277{
2278	xfs_mount_t	*mp = ip->i_mount;
2279	uint		delblks, blkflags, prjflags = 0;
2280	xfs_dquot_t	*unresudq, *unresgdq, *delblksudq, *delblksgdq;
2281	int		error;
 
 
 
 
 
2282
2283
2284	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2285	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2286
2287	delblks = ip->i_delayed_blks;
2288	delblksudq = delblksgdq = unresudq = unresgdq = NULL;
2289	blkflags = XFS_IS_REALTIME_INODE(ip) ?
2290			XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2291
2292	if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2293	    ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
2294		delblksudq = udqp;
2295		/*
2296		 * If there are delayed allocation blocks, then we have to
2297		 * unreserve those from the old dquot, and add them to the
2298		 * new dquot.
2299		 */
2300		if (delblks) {
2301			ASSERT(ip->i_udquot);
2302			unresudq = ip->i_udquot;
2303		}
2304	}
2305	if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
2306		if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2307		     xfs_get_projid(ip) != be32_to_cpu(gdqp->q_core.d_id))
2308			prjflags = XFS_QMOPT_ENOSPC;
2309
2310		if (prjflags ||
2311		    (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2312		     ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
2313			delblksgdq = gdqp;
2314			if (delblks) {
2315				ASSERT(ip->i_gdquot);
2316				unresgdq = ip->i_gdquot;
2317			}
2318		}
2319	}
2320
2321	if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2322				delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
2323				flags | blkflags | prjflags)))
2324		return (error);
 
 
 
 
 
 
 
 
 
 
 
 
2325
2326	/*
2327	 * Do the delayed blks reservations/unreservations now. Since, these
2328	 * are done without the help of a transaction, if a reservation fails
2329	 * its previous reservations won't be automatically undone by trans
2330	 * code. So, we have to do it manually here.
2331	 */
2332	if (delblks) {
2333		/*
2334		 * Do the reservations first. Unreservation can't fail.
2335		 */
2336		ASSERT(delblksudq || delblksgdq);
2337		ASSERT(unresudq || unresgdq);
2338		if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2339				delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
2340				flags | blkflags | prjflags)))
2341			return (error);
 
 
2342		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2343				unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
2344				blkflags);
2345	}
2346
2347	return (0);
2348}
2349
2350int
2351xfs_qm_vop_rename_dqattach(
2352	struct xfs_inode	**i_tab)
2353{
2354	struct xfs_mount	*mp = i_tab[0]->i_mount;
2355	int			i;
2356
2357	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2358		return 0;
2359
2360	for (i = 0; (i < 4 && i_tab[i]); i++) {
2361		struct xfs_inode	*ip = i_tab[i];
2362		int			error;
2363
2364		/*
2365		 * Watch out for duplicate entries in the table.
2366		 */
2367		if (i == 0 || ip != i_tab[i-1]) {
2368			if (XFS_NOT_DQATTACHED(mp, ip)) {
2369				error = xfs_qm_dqattach(ip, 0);
2370				if (error)
2371					return error;
2372			}
2373		}
2374	}
2375	return 0;
2376}
2377
2378void
2379xfs_qm_vop_create_dqattach(
2380	struct xfs_trans	*tp,
2381	struct xfs_inode	*ip,
2382	struct xfs_dquot	*udqp,
2383	struct xfs_dquot	*gdqp)
 
2384{
2385	struct xfs_mount	*mp = tp->t_mountp;
2386
2387	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2388		return;
2389
2390	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2391	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2392
2393	if (udqp) {
2394		xfs_dqlock(udqp);
2395		XFS_DQHOLD(udqp);
2396		xfs_dqunlock(udqp);
2397		ASSERT(ip->i_udquot == NULL);
2398		ip->i_udquot = udqp;
2399		ASSERT(XFS_IS_UQUOTA_ON(mp));
2400		ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
 
 
2401		xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2402	}
2403	if (gdqp) {
2404		xfs_dqlock(gdqp);
2405		XFS_DQHOLD(gdqp);
2406		xfs_dqunlock(gdqp);
2407		ASSERT(ip->i_gdquot == NULL);
2408		ip->i_gdquot = gdqp;
2409		ASSERT(XFS_IS_OQUOTA_ON(mp));
2410		ASSERT((XFS_IS_GQUOTA_ON(mp) ?
2411			ip->i_d.di_gid : xfs_get_projid(ip)) ==
2412				be32_to_cpu(gdqp->q_core.d_id));
2413		xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
 
 
 
 
 
 
 
2414	}
2415}
2416