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_shared.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_bit.h"
  25#include "xfs_sb.h"
  26#include "xfs_ag.h"
  27#include "xfs_mount.h"
  28#include "xfs_inode.h"
  29#include "xfs_ialloc.h"
  30#include "xfs_itable.h"
  31#include "xfs_quota.h"
  32#include "xfs_error.h"
  33#include "xfs_bmap.h"
  34#include "xfs_bmap_btree.h"
  35#include "xfs_trans.h"
  36#include "xfs_trans_space.h"
  37#include "xfs_qm.h"
  38#include "xfs_trace.h"
  39#include "xfs_icache.h"
  40#include "xfs_cksum.h"
  41#include "xfs_dinode.h"
 
 
 
 
 
 
  42
  43/*
  44 * The global quota manager. There is only one of these for the entire
  45 * system, _not_ one per file system. XQM keeps track of the overall
  46 * quota functionality, including maintaining the freelist and hash
  47 * tables of dquots.
  48 */
  49STATIC int	xfs_qm_init_quotainos(xfs_mount_t *);
  50STATIC int	xfs_qm_init_quotainfo(xfs_mount_t *);
  51
  52
  53STATIC void	xfs_qm_dqfree_one(struct xfs_dquot *dqp);
  54/*
  55 * We use the batch lookup interface to iterate over the dquots as it
  56 * currently is the only interface into the radix tree code that allows
  57 * fuzzy lookups instead of exact matches.  Holding the lock over multiple
  58 * operations is fine as all callers are used either during mount/umount
  59 * or quotaoff.
  60 */
  61#define XFS_DQ_LOOKUP_BATCH	32
  62
  63STATIC int
  64xfs_qm_dquot_walk(
  65	struct xfs_mount	*mp,
  66	int			type,
  67	int			(*execute)(struct xfs_dquot *dqp, void *data),
  68	void			*data)
  69{
  70	struct xfs_quotainfo	*qi = mp->m_quotainfo;
  71	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
  72	uint32_t		next_index;
  73	int			last_error = 0;
  74	int			skipped;
  75	int			nr_found;
  76
  77restart:
  78	skipped = 0;
  79	next_index = 0;
  80	nr_found = 0;
  81
  82	while (1) {
  83		struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
  84		int		error = 0;
  85		int		i;
  86
  87		mutex_lock(&qi->qi_tree_lock);
  88		nr_found = radix_tree_gang_lookup(tree, (void **)batch,
  89					next_index, XFS_DQ_LOOKUP_BATCH);
  90		if (!nr_found) {
  91			mutex_unlock(&qi->qi_tree_lock);
  92			break;
  93		}
  94
  95		for (i = 0; i < nr_found; i++) {
  96			struct xfs_dquot *dqp = batch[i];
  97
  98			next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
  99
 100			error = execute(batch[i], data);
 101			if (error == EAGAIN) {
 102				skipped++;
 103				continue;
 104			}
 105			if (error && last_error != EFSCORRUPTED)
 106				last_error = error;
 107		}
 108
 109		mutex_unlock(&qi->qi_tree_lock);
 110
 111		/* bail out if the filesystem is corrupted.  */
 112		if (last_error == EFSCORRUPTED) {
 113			skipped = 0;
 114			break;
 115		}
 
 
 
 116	}
 117
 118	if (skipped) {
 119		delay(1);
 120		goto restart;
 121	}
 122
 123	return last_error;
 124}
 125
 126
 127/*
 128 * Purge a dquot from all tracking data structures and free it.
 129 */
 130STATIC int
 131xfs_qm_dqpurge(
 132	struct xfs_dquot	*dqp,
 133	void			*data)
 134{
 135	struct xfs_mount	*mp = dqp->q_mount;
 136	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 137
 138	xfs_dqlock(dqp);
 139	if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
 140		xfs_dqunlock(dqp);
 141		return EAGAIN;
 142	}
 143
 144	dqp->dq_flags |= XFS_DQ_FREEING;
 145
 146	xfs_dqflock(dqp);
 147
 148	/*
 149	 * If we are turning this type of quotas off, we don't care
 150	 * about the dirty metadata sitting in this dquot. OTOH, if
 151	 * we're unmounting, we do care, so we flush it and wait.
 152	 */
 153	if (XFS_DQ_IS_DIRTY(dqp)) {
 154		struct xfs_buf	*bp = NULL;
 155		int		error;
 156
 157		/*
 158		 * We don't care about getting disk errors here. We need
 159		 * to purge this dquot anyway, so we go ahead regardless.
 160		 */
 161		error = xfs_qm_dqflush(dqp, &bp);
 162		if (error) {
 163			xfs_warn(mp, "%s: dquot %p flush failed",
 164				__func__, dqp);
 165		} else {
 
 
 
 
 
 
 
 
 
 
 166			error = xfs_bwrite(bp);
 167			xfs_buf_relse(bp);
 168		}
 169		xfs_dqflock(dqp);
 170	}
 
 171
 
 172	ASSERT(atomic_read(&dqp->q_pincount) == 0);
 173	ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
 174	       !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
 175
 176	xfs_dqfunlock(dqp);
 177	xfs_dqunlock(dqp);
 178
 179	radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
 180			  be32_to_cpu(dqp->q_core.d_id));
 181	qi->qi_dquots--;
 182
 183	/*
 184	 * We move dquots to the freelist as soon as their reference count
 185	 * hits zero, so it really should be on the freelist here.
 186	 */
 187	ASSERT(!list_empty(&dqp->q_lru));
 188	list_lru_del(&qi->qi_lru, &dqp->q_lru);
 189	XFS_STATS_DEC(xs_qm_dquot_unused);
 190
 191	xfs_qm_dqdestroy(dqp);
 192	return 0;
 193}
 194
 195/*
 196 * Release the group or project dquot pointers the user dquots maybe carrying
 197 * around as a hint, and proceed to purge the user dquot cache if requested.
 198*/
 199STATIC int
 200xfs_qm_dqpurge_hints(
 201	struct xfs_dquot	*dqp,
 202	void			*data)
 203{
 204	struct xfs_dquot	*gdqp = NULL;
 205	struct xfs_dquot	*pdqp = NULL;
 206	uint			flags = *((uint *)data);
 207
 208	xfs_dqlock(dqp);
 209	if (dqp->dq_flags & XFS_DQ_FREEING) {
 210		xfs_dqunlock(dqp);
 211		return EAGAIN;
 212	}
 213
 214	/* If this quota has a hint attached, prepare for releasing it now */
 215	gdqp = dqp->q_gdquot;
 216	if (gdqp)
 217		dqp->q_gdquot = NULL;
 218
 219	pdqp = dqp->q_pdquot;
 220	if (pdqp)
 221		dqp->q_pdquot = NULL;
 222
 
 223	xfs_dqunlock(dqp);
 224
 225	if (gdqp)
 226		xfs_qm_dqrele(gdqp);
 227	if (pdqp)
 228		xfs_qm_dqrele(pdqp);
 229
 230	if (flags & XFS_QMOPT_UQUOTA)
 231		return xfs_qm_dqpurge(dqp, NULL);
 232
 233	return 0;
 234}
 235
 236/*
 237 * Purge the dquot cache.
 238 */
 239void
 240xfs_qm_dqpurge_all(
 241	struct xfs_mount	*mp,
 242	uint			flags)
 243{
 244	/*
 245	 * We have to release group/project dquot hint(s) from the user dquot
 246	 * at first if they are there, otherwise we would run into an infinite
 247	 * loop while walking through radix tree to purge other type of dquots
 248	 * since their refcount is not zero if the user dquot refers to them
 249	 * as hint.
 250	 *
 251	 * Call the special xfs_qm_dqpurge_hints() will end up go through the
 252	 * general xfs_qm_dqpurge() against user dquot cache if requested.
 253	 */
 254	xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge_hints, &flags);
 255
 256	if (flags & XFS_QMOPT_GQUOTA)
 257		xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
 258	if (flags & XFS_QMOPT_PQUOTA)
 259		xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
 260}
 261
 262/*
 263 * Just destroy the quotainfo structure.
 264 */
 265void
 266xfs_qm_unmount(
 267	struct xfs_mount	*mp)
 268{
 269	if (mp->m_quotainfo) {
 270		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
 271		xfs_qm_destroy_quotainfo(mp);
 272	}
 273}
 274
 275
 276/*
 277 * This is called from xfs_mountfs to start quotas and initialize all
 278 * necessary data structures like quotainfo.  This is also responsible for
 279 * running a quotacheck as necessary.  We are guaranteed that the superblock
 280 * is consistently read in at this point.
 281 *
 282 * If we fail here, the mount will continue with quota turned off. We don't
 283 * need to inidicate success or failure at all.
 284 */
 285void
 286xfs_qm_mount_quotas(
 287	xfs_mount_t	*mp)
 288{
 289	int		error = 0;
 290	uint		sbf;
 291
 292	/*
 293	 * If quotas on realtime volumes is not supported, we disable
 294	 * quotas immediately.
 295	 */
 296	if (mp->m_sb.sb_rextents) {
 297		xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
 298		mp->m_qflags = 0;
 299		goto write_changes;
 300	}
 301
 302	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 303
 304	/*
 305	 * Allocate the quotainfo structure inside the mount struct, and
 306	 * create quotainode(s), and change/rev superblock if necessary.
 307	 */
 308	error = xfs_qm_init_quotainfo(mp);
 309	if (error) {
 310		/*
 311		 * We must turn off quotas.
 312		 */
 313		ASSERT(mp->m_quotainfo == NULL);
 314		mp->m_qflags = 0;
 315		goto write_changes;
 316	}
 317	/*
 318	 * If any of the quotas are not consistent, do a quotacheck.
 319	 */
 320	if (XFS_QM_NEED_QUOTACHECK(mp)) {
 321		error = xfs_qm_quotacheck(mp);
 322		if (error) {
 323			/* Quotacheck failed and disabled quotas. */
 324			return;
 325		}
 326	}
 327	/* 
 328	 * If one type of quotas is off, then it will lose its
 329	 * quotachecked status, since we won't be doing accounting for
 330	 * that type anymore.
 331	 */
 332	if (!XFS_IS_UQUOTA_ON(mp))
 333		mp->m_qflags &= ~XFS_UQUOTA_CHKD;
 334	if (!XFS_IS_GQUOTA_ON(mp))
 335		mp->m_qflags &= ~XFS_GQUOTA_CHKD;
 336	if (!XFS_IS_PQUOTA_ON(mp))
 337		mp->m_qflags &= ~XFS_PQUOTA_CHKD;
 338
 339 write_changes:
 340	/*
 341	 * We actually don't have to acquire the m_sb_lock at all.
 342	 * This can only be called from mount, and that's single threaded. XXX
 343	 */
 344	spin_lock(&mp->m_sb_lock);
 345	sbf = mp->m_sb.sb_qflags;
 346	mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
 347	spin_unlock(&mp->m_sb_lock);
 348
 349	if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
 350		if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
 351			/*
 352			 * We could only have been turning quotas off.
 353			 * We aren't in very good shape actually because
 354			 * the incore structures are convinced that quotas are
 355			 * off, but the on disk superblock doesn't know that !
 356			 */
 357			ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
 358			xfs_alert(mp, "%s: Superblock update failed!",
 359				__func__);
 360		}
 361	}
 362
 363	if (error) {
 364		xfs_warn(mp, "Failed to initialize disk quotas.");
 365		return;
 366	}
 367}
 368
 369/*
 370 * Called from the vfsops layer.
 371 */
 372void
 373xfs_qm_unmount_quotas(
 374	xfs_mount_t	*mp)
 375{
 376	/*
 377	 * Release the dquots that root inode, et al might be holding,
 378	 * before we flush quotas and blow away the quotainfo structure.
 379	 */
 380	ASSERT(mp->m_rootip);
 381	xfs_qm_dqdetach(mp->m_rootip);
 382	if (mp->m_rbmip)
 383		xfs_qm_dqdetach(mp->m_rbmip);
 384	if (mp->m_rsumip)
 385		xfs_qm_dqdetach(mp->m_rsumip);
 
 
 
 386
 387	/*
 388	 * Release the quota inodes.
 389	 */
 390	if (mp->m_quotainfo) {
 391		if (mp->m_quotainfo->qi_uquotaip) {
 392			IRELE(mp->m_quotainfo->qi_uquotaip);
 393			mp->m_quotainfo->qi_uquotaip = NULL;
 394		}
 395		if (mp->m_quotainfo->qi_gquotaip) {
 396			IRELE(mp->m_quotainfo->qi_gquotaip);
 397			mp->m_quotainfo->qi_gquotaip = NULL;
 398		}
 399		if (mp->m_quotainfo->qi_pquotaip) {
 400			IRELE(mp->m_quotainfo->qi_pquotaip);
 401			mp->m_quotainfo->qi_pquotaip = NULL;
 402		}
 403	}
 404}
 405
 406STATIC int
 407xfs_qm_dqattach_one(
 408	xfs_inode_t	*ip,
 409	xfs_dqid_t	id,
 410	uint		type,
 411	uint		doalloc,
 412	xfs_dquot_t	*udqhint, /* hint */
 413	xfs_dquot_t	**IO_idqpp)
 414{
 415	xfs_dquot_t	*dqp;
 416	int		error;
 417
 418	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 419	error = 0;
 420
 421	/*
 422	 * See if we already have it in the inode itself. IO_idqpp is
 423	 * &i_udquot or &i_gdquot. This made the code look weird, but
 424	 * made the logic a lot simpler.
 425	 */
 426	dqp = *IO_idqpp;
 427	if (dqp) {
 428		trace_xfs_dqattach_found(dqp);
 429		return 0;
 430	}
 431
 432	/*
 433	 * udqhint is the i_udquot field in inode, and is non-NULL only
 434	 * when the type arg is group/project. Its purpose is to save a
 435	 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
 436	 * the user dquot.
 437	 */
 438	if (udqhint) {
 439		ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 440		xfs_dqlock(udqhint);
 441
 442		/*
 443		 * No need to take dqlock to look at the id.
 444		 *
 445		 * The ID can't change until it gets reclaimed, and it won't
 446		 * be reclaimed as long as we have a ref from inode and we
 447		 * hold the ilock.
 448		 */
 449		if (type == XFS_DQ_GROUP)
 450			dqp = udqhint->q_gdquot;
 451		else
 452			dqp = udqhint->q_pdquot;
 453		if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
 454			ASSERT(*IO_idqpp == NULL);
 455
 456			*IO_idqpp = xfs_qm_dqhold(dqp);
 457			xfs_dqunlock(udqhint);
 458			return 0;
 459		}
 460
 461		/*
 462		 * We can't hold a dquot lock when we call the dqget code.
 463		 * We'll deadlock in no time, because of (not conforming to)
 464		 * lock ordering - the inodelock comes before any dquot lock,
 465		 * and we may drop and reacquire the ilock in xfs_qm_dqget().
 466		 */
 467		xfs_dqunlock(udqhint);
 468	}
 469
 470	/*
 471	 * Find the dquot from somewhere. This bumps the
 472	 * reference count of dquot and returns it locked.
 473	 * This can return ENOENT if dquot didn't exist on
 474	 * disk and we didn't ask it to allocate;
 475	 * ESRCH if quotas got turned off suddenly.
 476	 */
 477	error = xfs_qm_dqget(ip->i_mount, ip, id, type,
 478			     doalloc | XFS_QMOPT_DOWARN, &dqp);
 479	if (error)
 480		return error;
 481
 482	trace_xfs_dqattach_get(dqp);
 483
 484	/*
 485	 * dqget may have dropped and re-acquired the ilock, but it guarantees
 486	 * that the dquot returned is the one that should go in the inode.
 487	 */
 488	*IO_idqpp = dqp;
 489	xfs_dqunlock(dqp);
 490	return 0;
 491}
 492
 493
 494/*
 495 * Given a udquot and group/project type, attach the group/project
 496 * dquot pointer to the udquot as a hint for future lookups.
 497 */
 498STATIC void
 499xfs_qm_dqattach_hint(
 500	struct xfs_inode	*ip,
 501	int			type)
 502{
 503	struct xfs_dquot **dqhintp;
 504	struct xfs_dquot *dqp;
 505	struct xfs_dquot *udq = ip->i_udquot;
 506
 507	ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 508
 509	xfs_dqlock(udq);
 510
 511	if (type == XFS_DQ_GROUP) {
 512		dqp = ip->i_gdquot;
 513		dqhintp = &udq->q_gdquot;
 514	} else {
 515		dqp = ip->i_pdquot;
 516		dqhintp = &udq->q_pdquot;
 517	}
 518
 519	if (*dqhintp) {
 520		struct xfs_dquot *tmp;
 521
 522		if (*dqhintp == dqp)
 523			goto done;
 524
 525		tmp = *dqhintp;
 526		*dqhintp = NULL;
 527		xfs_qm_dqrele(tmp);
 528	}
 529
 530	*dqhintp = xfs_qm_dqhold(dqp);
 531done:
 532	xfs_dqunlock(udq);
 533}
 534
 535static bool
 536xfs_qm_need_dqattach(
 537	struct xfs_inode	*ip)
 538{
 539	struct xfs_mount	*mp = ip->i_mount;
 540
 541	if (!XFS_IS_QUOTA_RUNNING(mp))
 542		return false;
 543	if (!XFS_IS_QUOTA_ON(mp))
 544		return false;
 545	if (!XFS_NOT_DQATTACHED(mp, ip))
 546		return false;
 547	if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
 548		return false;
 
 
 549	return true;
 550}
 551
 552/*
 553 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 554 * into account.
 555 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 556 * Inode may get unlocked and relocked in here, and the caller must deal with
 557 * the consequences.
 558 */
 559int
 560xfs_qm_dqattach_locked(
 561	xfs_inode_t	*ip,
 562	uint		flags)
 563{
 564	xfs_mount_t	*mp = ip->i_mount;
 565	uint		nquotas = 0;
 566	int		error = 0;
 567
 568	if (!xfs_qm_need_dqattach(ip))
 569		return 0;
 570
 571	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 
 572
 573	if (XFS_IS_UQUOTA_ON(mp)) {
 574		error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
 575						flags & XFS_QMOPT_DQALLOC,
 576						NULL, &ip->i_udquot);
 577		if (error)
 578			goto done;
 579		nquotas++;
 580	}
 581
 582	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 583	if (XFS_IS_GQUOTA_ON(mp)) {
 584		error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
 585						flags & XFS_QMOPT_DQALLOC,
 586						ip->i_udquot, &ip->i_gdquot);
 587		/*
 588		 * Don't worry about the udquot that we may have
 589		 * attached above. It'll get detached, if not already.
 590		 */
 591		if (error)
 592			goto done;
 593		nquotas++;
 594	}
 595
 596	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 597	if (XFS_IS_PQUOTA_ON(mp)) {
 598		error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
 599						flags & XFS_QMOPT_DQALLOC,
 600						ip->i_udquot, &ip->i_pdquot);
 601		/*
 602		 * Don't worry about the udquot that we may have
 603		 * attached above. It'll get detached, if not already.
 604		 */
 605		if (error)
 606			goto done;
 607		nquotas++;
 608	}
 609
 
 610	/*
 611	 * Attach this group/project quota to the user quota as a hint.
 612	 * This WON'T, in general, result in a thrash.
 613	 */
 614	if (nquotas > 1 && ip->i_udquot) {
 615		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 616		ASSERT(ip->i_gdquot || !XFS_IS_GQUOTA_ON(mp));
 617		ASSERT(ip->i_pdquot || !XFS_IS_PQUOTA_ON(mp));
 618
 619		/*
 620		 * We do not have i_udquot locked at this point, but this check
 621		 * is OK since we don't depend on the i_gdquot to be accurate
 622		 * 100% all the time. It is just a hint, and this will
 623		 * succeed in general.
 624		 */
 625		if (ip->i_udquot->q_gdquot != ip->i_gdquot)
 626			xfs_qm_dqattach_hint(ip, XFS_DQ_GROUP);
 627
 628		if (ip->i_udquot->q_pdquot != ip->i_pdquot)
 629			xfs_qm_dqattach_hint(ip, XFS_DQ_PROJ);
 630	}
 631
 632 done:
 633#ifdef DEBUG
 634	if (!error) {
 635		if (XFS_IS_UQUOTA_ON(mp))
 636			ASSERT(ip->i_udquot);
 637		if (XFS_IS_GQUOTA_ON(mp))
 638			ASSERT(ip->i_gdquot);
 639		if (XFS_IS_PQUOTA_ON(mp))
 640			ASSERT(ip->i_pdquot);
 641	}
 642	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 643#endif
 644	return error;
 645}
 646
 647int
 648xfs_qm_dqattach(
 649	struct xfs_inode	*ip,
 650	uint			flags)
 651{
 652	int			error;
 653
 654	if (!xfs_qm_need_dqattach(ip))
 655		return 0;
 656
 657	xfs_ilock(ip, XFS_ILOCK_EXCL);
 658	error = xfs_qm_dqattach_locked(ip, flags);
 659	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 660
 661	return error;
 662}
 663
 664/*
 665 * Release dquots (and their references) if any.
 666 * The inode should be locked EXCL except when this's called by
 667 * xfs_ireclaim.
 668 */
 669void
 670xfs_qm_dqdetach(
 671	xfs_inode_t	*ip)
 672{
 673	if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
 674		return;
 675
 676	trace_xfs_dquot_dqdetach(ip);
 677
 678	ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
 679	if (ip->i_udquot) {
 680		xfs_qm_dqrele(ip->i_udquot);
 681		ip->i_udquot = NULL;
 682	}
 683	if (ip->i_gdquot) {
 684		xfs_qm_dqrele(ip->i_gdquot);
 685		ip->i_gdquot = NULL;
 686	}
 687	if (ip->i_pdquot) {
 688		xfs_qm_dqrele(ip->i_pdquot);
 689		ip->i_pdquot = NULL;
 690	}
 691}
 692
 693struct xfs_qm_isolate {
 694	struct list_head	buffers;
 695	struct list_head	dispose;
 696};
 697
 698static enum lru_status
 699xfs_qm_dquot_isolate(
 700	struct list_head	*item,
 701	spinlock_t		*lru_lock,
 702	void			*arg)
 
 703{
 704	struct xfs_dquot	*dqp = container_of(item,
 705						struct xfs_dquot, q_lru);
 706	struct xfs_qm_isolate	*isol = arg;
 707
 708	if (!xfs_dqlock_nowait(dqp))
 709		goto out_miss_busy;
 710
 711	/*
 
 
 
 
 
 
 
 
 712	 * This dquot has acquired a reference in the meantime remove it from
 713	 * the freelist and try again.
 714	 */
 715	if (dqp->q_nrefs) {
 716		xfs_dqunlock(dqp);
 717		XFS_STATS_INC(xs_qm_dqwants);
 718
 719		trace_xfs_dqreclaim_want(dqp);
 720		list_del_init(&dqp->q_lru);
 721		XFS_STATS_DEC(xs_qm_dquot_unused);
 722		return LRU_REMOVED;
 723	}
 724
 725	/*
 726	 * If the dquot is dirty, flush it. If it's already being flushed, just
 727	 * skip it so there is time for the IO to complete before we try to
 728	 * reclaim it again on the next LRU pass.
 729	 */
 730	if (!xfs_dqflock_nowait(dqp)) {
 731		xfs_dqunlock(dqp);
 732		goto out_miss_busy;
 733	}
 734
 735	if (XFS_DQ_IS_DIRTY(dqp)) {
 736		struct xfs_buf	*bp = NULL;
 737		int		error;
 738
 739		trace_xfs_dqreclaim_dirty(dqp);
 740
 741		/* we have to drop the LRU lock to flush the dquot */
 742		spin_unlock(lru_lock);
 743
 744		error = xfs_qm_dqflush(dqp, &bp);
 745		if (error) {
 746			xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
 747				 __func__, dqp);
 748			goto out_unlock_dirty;
 749		}
 750
 
 
 
 
 
 
 
 
 751		xfs_buf_delwri_queue(bp, &isol->buffers);
 752		xfs_buf_relse(bp);
 753		goto out_unlock_dirty;
 754	}
 
 
 755	xfs_dqfunlock(dqp);
 756
 757	/*
 758	 * Prevent lookups now that we are past the point of no return.
 759	 */
 760	dqp->dq_flags |= XFS_DQ_FREEING;
 761	xfs_dqunlock(dqp);
 762
 763	ASSERT(dqp->q_nrefs == 0);
 764	list_move_tail(&dqp->q_lru, &isol->dispose);
 765	XFS_STATS_DEC(xs_qm_dquot_unused);
 766	trace_xfs_dqreclaim_done(dqp);
 767	XFS_STATS_INC(xs_qm_dqreclaims);
 768	return LRU_REMOVED;
 769
 
 
 770out_miss_busy:
 771	trace_xfs_dqreclaim_busy(dqp);
 772	XFS_STATS_INC(xs_qm_dqreclaim_misses);
 773	return LRU_SKIP;
 774
 775out_unlock_dirty:
 776	trace_xfs_dqreclaim_busy(dqp);
 777	XFS_STATS_INC(xs_qm_dqreclaim_misses);
 778	xfs_dqunlock(dqp);
 779	spin_lock(lru_lock);
 780	return LRU_RETRY;
 781}
 782
 783static unsigned long
 784xfs_qm_shrink_scan(
 785	struct shrinker		*shrink,
 786	struct shrink_control	*sc)
 787{
 788	struct xfs_quotainfo	*qi = container_of(shrink,
 789					struct xfs_quotainfo, qi_shrinker);
 790	struct xfs_qm_isolate	isol;
 791	unsigned long		freed;
 792	int			error;
 793	unsigned long		nr_to_scan = sc->nr_to_scan;
 794
 795	if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
 796		return 0;
 797
 798	INIT_LIST_HEAD(&isol.buffers);
 799	INIT_LIST_HEAD(&isol.dispose);
 800
 801	freed = list_lru_walk_node(&qi->qi_lru, sc->nid, xfs_qm_dquot_isolate, &isol,
 802					&nr_to_scan);
 803
 804	error = xfs_buf_delwri_submit(&isol.buffers);
 805	if (error)
 806		xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
 807
 808	while (!list_empty(&isol.dispose)) {
 809		struct xfs_dquot	*dqp;
 810
 811		dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
 812		list_del_init(&dqp->q_lru);
 813		xfs_qm_dqfree_one(dqp);
 814	}
 815
 816	return freed;
 817}
 818
 819static unsigned long
 820xfs_qm_shrink_count(
 821	struct shrinker		*shrink,
 822	struct shrink_control	*sc)
 823{
 824	struct xfs_quotainfo	*qi = container_of(shrink,
 825					struct xfs_quotainfo, qi_shrinker);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 826
 827	return list_lru_count_node(&qi->qi_lru, sc->nid);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 828}
 829
 830/*
 831 * This initializes all the quota information that's kept in the
 832 * mount structure
 833 */
 834STATIC int
 835xfs_qm_init_quotainfo(
 836	xfs_mount_t	*mp)
 837{
 838	xfs_quotainfo_t *qinf;
 839	int		error;
 840	xfs_dquot_t	*dqp;
 841
 842	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 843
 844	qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
 
 845
 846	error = -list_lru_init(&qinf->qi_lru);
 847	if (error)
 848		goto out_free_qinf;
 849
 850	/*
 851	 * See if quotainodes are setup, and if not, allocate them,
 852	 * and change the superblock accordingly.
 853	 */
 854	error = xfs_qm_init_quotainos(mp);
 
 
 
 855	if (error)
 856		goto out_free_lru;
 857
 858	INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
 859	INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
 860	INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
 861	mutex_init(&qinf->qi_tree_lock);
 862
 863	/* mutex used to serialize quotaoffs */
 864	mutex_init(&qinf->qi_quotaofflock);
 865
 866	/* Precalc some constants */
 867	qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 868	qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(mp,
 869							qinf->qi_dqchunklen);
 
 
 
 
 
 
 
 
 
 
 870
 871	mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
 872
 873	/*
 874	 * We try to get the limits from the superuser's limits fields.
 875	 * This is quite hacky, but it is standard quota practice.
 876	 *
 877	 * We look at the USR dquot with id == 0 first, but if user quotas
 878	 * are not enabled we goto the GRP dquot with id == 0.
 879	 * We don't really care to keep separate default limits for user
 880	 * and group quotas, at least not at this point.
 881	 *
 882	 * Since we may not have done a quotacheck by this point, just read
 883	 * the dquot without attaching it to any hashtables or lists.
 884	 */
 885	error = xfs_qm_dqread(mp, 0,
 886			XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
 887			 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
 888			  XFS_DQ_PROJ),
 889			XFS_QMOPT_DOWARN, &dqp);
 890	if (!error) {
 891		xfs_disk_dquot_t	*ddqp = &dqp->q_core;
 
 
 892
 893		/*
 894		 * The warnings and timers set the grace period given to
 895		 * a user or group before he or she can not perform any
 896		 * more writing. If it is zero, a default is used.
 897		 */
 898		qinf->qi_btimelimit = ddqp->d_btimer ?
 899			be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
 900		qinf->qi_itimelimit = ddqp->d_itimer ?
 901			be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
 902		qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
 903			be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
 904		qinf->qi_bwarnlimit = ddqp->d_bwarns ?
 905			be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
 906		qinf->qi_iwarnlimit = ddqp->d_iwarns ?
 907			be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
 908		qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
 909			be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
 910		qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 911		qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 912		qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 913		qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 914		qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 915		qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 916
 917		xfs_qm_dqdestroy(dqp);
 918	} else {
 919		qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
 920		qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
 921		qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
 922		qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
 923		qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
 924		qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
 925	}
 926
 927	qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
 928	qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
 929	qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
 930	qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
 931	register_shrinker(&qinf->qi_shrinker);
 932	return 0;
 933
 
 
 
 
 934out_free_lru:
 935	list_lru_destroy(&qinf->qi_lru);
 936out_free_qinf:
 937	kmem_free(qinf);
 938	mp->m_quotainfo = NULL;
 939	return error;
 940}
 941
 942
 943/*
 944 * Gets called when unmounting a filesystem or when all quotas get
 945 * turned off.
 946 * This purges the quota inodes, destroys locks and frees itself.
 947 */
 948void
 949xfs_qm_destroy_quotainfo(
 950	xfs_mount_t	*mp)
 951{
 952	xfs_quotainfo_t *qi;
 953
 954	qi = mp->m_quotainfo;
 955	ASSERT(qi != NULL);
 956
 957	unregister_shrinker(&qi->qi_shrinker);
 958	list_lru_destroy(&qi->qi_lru);
 959
 960	if (qi->qi_uquotaip) {
 961		IRELE(qi->qi_uquotaip);
 962		qi->qi_uquotaip = NULL; /* paranoia */
 963	}
 964	if (qi->qi_gquotaip) {
 965		IRELE(qi->qi_gquotaip);
 966		qi->qi_gquotaip = NULL;
 967	}
 968	if (qi->qi_pquotaip) {
 969		IRELE(qi->qi_pquotaip);
 970		qi->qi_pquotaip = NULL;
 971	}
 972	mutex_destroy(&qi->qi_quotaofflock);
 973	kmem_free(qi);
 974	mp->m_quotainfo = NULL;
 975}
 976
 
 
 
 
 
 
 
 
 
 
 
 977/*
 978 * Create an inode and return with a reference already taken, but unlocked
 979 * This is how we create quota inodes
 980 */
 981STATIC int
 982xfs_qm_qino_alloc(
 983	xfs_mount_t	*mp,
 984	xfs_inode_t	**ip,
 985	__int64_t	sbfields,
 986	uint		flags)
 987{
 988	xfs_trans_t	*tp;
 989	int		error;
 990	int		committed;
 991
 992	*ip = NULL;
 993	/*
 994	 * With superblock that doesn't have separate pquotino, we
 995	 * share an inode between gquota and pquota. If the on-disk
 996	 * superblock has GQUOTA and the filesystem is now mounted
 997	 * with PQUOTA, just use sb_gquotino for sb_pquotino and
 998	 * vice-versa.
 999	 */
1000	if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
1001			(flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
1002		xfs_ino_t ino = NULLFSINO;
1003
1004		if ((flags & XFS_QMOPT_PQUOTA) &&
1005			     (mp->m_sb.sb_gquotino != NULLFSINO)) {
1006			ino = mp->m_sb.sb_gquotino;
1007			ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
 
 
 
 
1008		} else if ((flags & XFS_QMOPT_GQUOTA) &&
1009			     (mp->m_sb.sb_pquotino != NULLFSINO)) {
1010			ino = mp->m_sb.sb_pquotino;
1011			ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
 
 
 
 
1012		}
1013		if (ino != NULLFSINO) {
1014			error = xfs_iget(mp, NULL, ino, 0, 0, ip);
1015			if (error)
1016				return error;
 
1017			mp->m_sb.sb_gquotino = NULLFSINO;
1018			mp->m_sb.sb_pquotino = NULLFSINO;
 
1019		}
1020	}
1021
1022	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1023	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
1024				  XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
1025	if (error) {
1026		xfs_trans_cancel(tp, 0);
1027		return error;
1028	}
1029
1030	if (!*ip) {
1031		error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
1032								&committed);
 
 
 
 
 
 
 
1033		if (error) {
1034			xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1035					 XFS_TRANS_ABORT);
1036			return error;
1037		}
 
 
1038	}
1039
1040	/*
1041	 * Make the changes in the superblock, and log those too.
1042	 * sbfields arg may contain fields other than *QUOTINO;
1043	 * VERSIONNUM for example.
1044	 */
1045	spin_lock(&mp->m_sb_lock);
1046	if (flags & XFS_QMOPT_SBVERSION) {
1047		ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1048		ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1049			XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
1050				(XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1051				 XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1052				 XFS_SB_QFLAGS));
1053
1054		xfs_sb_version_addquota(&mp->m_sb);
1055		mp->m_sb.sb_uquotino = NULLFSINO;
1056		mp->m_sb.sb_gquotino = NULLFSINO;
1057		mp->m_sb.sb_pquotino = NULLFSINO;
1058
1059		/* qflags will get updated fully _after_ quotacheck */
1060		mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
1061	}
1062	if (flags & XFS_QMOPT_UQUOTA)
1063		mp->m_sb.sb_uquotino = (*ip)->i_ino;
1064	else if (flags & XFS_QMOPT_GQUOTA)
1065		mp->m_sb.sb_gquotino = (*ip)->i_ino;
1066	else
1067		mp->m_sb.sb_pquotino = (*ip)->i_ino;
1068	spin_unlock(&mp->m_sb_lock);
1069	xfs_mod_sb(tp, sbfields);
1070
1071	if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
 
 
1072		xfs_alert(mp, "%s failed (error %d)!", __func__, error);
1073		return error;
1074	}
1075	return 0;
 
 
 
 
1076}
1077
1078
1079STATIC void
1080xfs_qm_reset_dqcounts(
1081	xfs_mount_t	*mp,
1082	xfs_buf_t	*bp,
1083	xfs_dqid_t	id,
1084	uint		type)
1085{
1086	struct xfs_dqblk	*dqb;
1087	int			j;
1088
1089	trace_xfs_reset_dqcounts(bp, _RET_IP_);
1090
1091	/*
1092	 * Reset all counters and timers. They'll be
1093	 * started afresh by xfs_qm_quotacheck.
1094	 */
1095#ifdef DEBUG
1096	j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1097	do_div(j, sizeof(xfs_dqblk_t));
1098	ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1099#endif
1100	dqb = bp->b_addr;
1101	for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1102		struct xfs_disk_dquot	*ddq;
1103
1104		ddq = (struct xfs_disk_dquot *)&dqb[j];
1105
1106		/*
1107		 * Do a sanity check, and if needed, repair the dqblk. Don't
1108		 * output any warnings because it's perfectly possible to
1109		 * find uninitialised dquot blks. See comment in xfs_dqcheck.
 
 
 
 
 
 
 
 
 
1110		 */
1111		xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1112			    "xfs_quotacheck");
1113		ddq->d_bcount = 0;
1114		ddq->d_icount = 0;
1115		ddq->d_rtbcount = 0;
1116		ddq->d_btimer = 0;
1117		ddq->d_itimer = 0;
1118		ddq->d_rtbtimer = 0;
1119		ddq->d_bwarns = 0;
1120		ddq->d_iwarns = 0;
1121		ddq->d_rtbwarns = 0;
1122
1123		if (xfs_sb_version_hascrc(&mp->m_sb)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1124			xfs_update_cksum((char *)&dqb[j],
1125					 sizeof(struct xfs_dqblk),
1126					 XFS_DQUOT_CRC_OFF);
1127		}
1128	}
1129}
1130
1131STATIC int
1132xfs_qm_dqiter_bufs(
1133	struct xfs_mount	*mp,
1134	xfs_dqid_t		firstid,
1135	xfs_fsblock_t		bno,
1136	xfs_filblks_t		blkcnt,
1137	uint			flags,
1138	struct list_head	*buffer_list)
1139{
1140	struct xfs_buf		*bp;
1141	int			error;
1142	int			type;
1143
1144	ASSERT(blkcnt > 0);
1145	type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1146		(flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1147	error = 0;
1148
1149	/*
1150	 * Blkcnt arg can be a very big number, and might even be
1151	 * larger than the log itself. So, we have to break it up into
1152	 * manageable-sized transactions.
1153	 * Note that we don't start a permanent transaction here; we might
1154	 * not be able to get a log reservation for the whole thing up front,
1155	 * and we don't really care to either, because we just discard
1156	 * everything if we were to crash in the middle of this loop.
1157	 */
1158	while (blkcnt--) {
1159		error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1160			      XFS_FSB_TO_DADDR(mp, bno),
1161			      mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1162			      &xfs_dquot_buf_ops);
1163
1164		/*
1165		 * CRC and validation errors will return a EFSCORRUPTED here. If
1166		 * this occurs, re-read without CRC validation so that we can
1167		 * repair the damage via xfs_qm_reset_dqcounts(). This process
1168		 * will leave a trace in the log indicating corruption has
1169		 * been detected.
1170		 */
1171		if (error == EFSCORRUPTED) {
1172			error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1173				      XFS_FSB_TO_DADDR(mp, bno),
1174				      mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1175				      NULL);
1176		}
1177
1178		if (error)
1179			break;
1180
 
 
 
 
 
 
1181		xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1182		xfs_buf_delwri_queue(bp, buffer_list);
1183		xfs_buf_relse(bp);
1184
1185		/* goto the next block. */
1186		bno++;
1187		firstid += mp->m_quotainfo->qi_dqperchunk;
1188	}
1189
1190	return error;
1191}
1192
1193/*
1194 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1195 * caller supplied function for every chunk of dquots that we find.
1196 */
1197STATIC int
1198xfs_qm_dqiterate(
1199	struct xfs_mount	*mp,
1200	struct xfs_inode	*qip,
1201	uint			flags,
1202	struct list_head	*buffer_list)
1203{
1204	struct xfs_bmbt_irec	*map;
1205	int			i, nmaps;	/* number of map entries */
1206	int			error;		/* return value */
1207	xfs_fileoff_t		lblkno;
1208	xfs_filblks_t		maxlblkcnt;
1209	xfs_dqid_t		firstid;
1210	xfs_fsblock_t		rablkno;
1211	xfs_filblks_t		rablkcnt;
1212
1213	error = 0;
1214	/*
1215	 * This looks racy, but we can't keep an inode lock across a
1216	 * trans_reserve. But, this gets called during quotacheck, and that
1217	 * happens only at mount time which is single threaded.
1218	 */
1219	if (qip->i_d.di_nblocks == 0)
1220		return 0;
1221
1222	map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
 
1223
1224	lblkno = 0;
1225	maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1226	do {
1227		uint		lock_mode;
1228
1229		nmaps = XFS_DQITER_MAP_SIZE;
1230		/*
1231		 * We aren't changing the inode itself. Just changing
1232		 * some of its data. No new blocks are added here, and
1233		 * the inode is never added to the transaction.
1234		 */
1235		lock_mode = xfs_ilock_data_map_shared(qip);
1236		error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1237				       map, &nmaps, 0);
1238		xfs_iunlock(qip, lock_mode);
1239		if (error)
1240			break;
1241
1242		ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1243		for (i = 0; i < nmaps; i++) {
1244			ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1245			ASSERT(map[i].br_blockcount);
1246
1247
1248			lblkno += map[i].br_blockcount;
1249
1250			if (map[i].br_startblock == HOLESTARTBLOCK)
1251				continue;
1252
1253			firstid = (xfs_dqid_t) map[i].br_startoff *
1254				mp->m_quotainfo->qi_dqperchunk;
1255			/*
1256			 * Do a read-ahead on the next extent.
1257			 */
1258			if ((i+1 < nmaps) &&
1259			    (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1260				rablkcnt =  map[i+1].br_blockcount;
1261				rablkno = map[i+1].br_startblock;
1262				while (rablkcnt--) {
1263					xfs_buf_readahead(mp->m_ddev_targp,
1264					       XFS_FSB_TO_DADDR(mp, rablkno),
1265					       mp->m_quotainfo->qi_dqchunklen,
1266					       NULL);
1267					rablkno++;
1268				}
1269			}
1270			/*
1271			 * Iterate thru all the blks in the extent and
1272			 * reset the counters of all the dquots inside them.
1273			 */
1274			error = xfs_qm_dqiter_bufs(mp, firstid,
1275						   map[i].br_startblock,
1276						   map[i].br_blockcount,
1277						   flags, buffer_list);
1278			if (error)
1279				goto out;
1280		}
1281	} while (nmaps > 0);
1282
1283out:
1284	kmem_free(map);
1285	return error;
1286}
1287
1288/*
1289 * Called by dqusage_adjust in doing a quotacheck.
1290 *
1291 * Given the inode, and a dquot id this updates both the incore dqout as well
1292 * as the buffer copy. This is so that once the quotacheck is done, we can
1293 * just log all the buffers, as opposed to logging numerous updates to
1294 * individual dquots.
1295 */
1296STATIC int
1297xfs_qm_quotacheck_dqadjust(
1298	struct xfs_inode	*ip,
1299	xfs_dqid_t		id,
1300	uint			type,
1301	xfs_qcnt_t		nblks,
1302	xfs_qcnt_t		rtblks)
1303{
1304	struct xfs_mount	*mp = ip->i_mount;
1305	struct xfs_dquot	*dqp;
 
1306	int			error;
1307
1308	error = xfs_qm_dqget(mp, ip, id, type,
1309			     XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1310	if (error) {
1311		/*
1312		 * Shouldn't be able to turn off quotas here.
1313		 */
1314		ASSERT(error != ESRCH);
1315		ASSERT(error != ENOENT);
1316		return error;
1317	}
1318
 
 
 
 
1319	trace_xfs_dqadjust(dqp);
1320
1321	/*
1322	 * Adjust the inode count and the block count to reflect this inode's
1323	 * resource usage.
1324	 */
1325	be64_add_cpu(&dqp->q_core.d_icount, 1);
1326	dqp->q_res_icount++;
1327	if (nblks) {
1328		be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1329		dqp->q_res_bcount += nblks;
1330	}
1331	if (rtblks) {
1332		be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1333		dqp->q_res_rtbcount += rtblks;
1334	}
1335
1336	/*
1337	 * Set default limits, adjust timers (since we changed usages)
1338	 *
1339	 * There are no timers for the default values set in the root dquot.
1340	 */
1341	if (dqp->q_core.d_id) {
1342		xfs_qm_adjust_dqlimits(mp, dqp);
1343		xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1344	}
1345
1346	dqp->dq_flags |= XFS_DQ_DIRTY;
1347	xfs_qm_dqput(dqp);
1348	return 0;
1349}
1350
1351STATIC int
1352xfs_qm_get_rtblks(
1353	xfs_inode_t	*ip,
1354	xfs_qcnt_t	*O_rtblks)
1355{
1356	xfs_filblks_t	rtblks;			/* total rt blks */
1357	xfs_extnum_t	idx;			/* extent record index */
1358	xfs_ifork_t	*ifp;			/* inode fork pointer */
1359	xfs_extnum_t	nextents;		/* number of extent entries */
1360	int		error;
1361
1362	ASSERT(XFS_IS_REALTIME_INODE(ip));
1363	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1364	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1365		if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1366			return error;
1367	}
1368	rtblks = 0;
1369	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1370	for (idx = 0; idx < nextents; idx++)
1371		rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1372	*O_rtblks = (xfs_qcnt_t)rtblks;
1373	return 0;
1374}
1375
1376/*
1377 * callback routine supplied to bulkstat(). Given an inumber, find its
1378 * dquots and update them to account for resources taken by that inode.
1379 */
1380/* ARGSUSED */
1381STATIC int
1382xfs_qm_dqusage_adjust(
1383	xfs_mount_t	*mp,		/* mount point for filesystem */
1384	xfs_ino_t	ino,		/* inode number to get data for */
1385	void		__user *buffer,	/* not used */
1386	int		ubsize,		/* not used */
1387	int		*ubused,	/* not used */
1388	int		*res)		/* result code value */
1389{
1390	xfs_inode_t	*ip;
1391	xfs_qcnt_t	nblks, rtblks = 0;
1392	int		error;
1393
1394	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1395
1396	/*
1397	 * rootino must have its resources accounted for, not so with the quota
1398	 * inodes.
1399	 */
1400	if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1401		*res = BULKSTAT_RV_NOTHING;
1402		return XFS_ERROR(EINVAL);
1403	}
1404
1405	/*
1406	 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1407	 * interface expects the inode to be exclusively locked because that's
1408	 * the case in all other instances. It's OK that we do this because
1409	 * quotacheck is done only at mount time.
1410	 */
1411	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1412	if (error) {
1413		*res = BULKSTAT_RV_NOTHING;
 
1414		return error;
 
 
 
 
 
 
 
 
 
 
 
 
1415	}
1416
 
 
 
 
1417	ASSERT(ip->i_delayed_blks == 0);
1418
 
1419	if (XFS_IS_REALTIME_INODE(ip)) {
1420		/*
1421		 * Walk thru the extent list and count the realtime blocks.
1422		 */
1423		error = xfs_qm_get_rtblks(ip, &rtblks);
1424		if (error)
1425			goto error0;
 
1426	}
1427
1428	nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
 
1429
1430	/*
1431	 * Add the (disk blocks and inode) resources occupied by this
1432	 * inode to its dquots. We do this adjustment in the incore dquot,
1433	 * and also copy the changes to its buffer.
1434	 * We don't care about putting these changes in a transaction
1435	 * envelope because if we crash in the middle of a 'quotacheck'
1436	 * we have to start from the beginning anyway.
1437	 * Once we're done, we'll log all the dquot bufs.
1438	 *
1439	 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1440	 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1441	 */
1442	if (XFS_IS_UQUOTA_ON(mp)) {
1443		error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1444						   XFS_DQ_USER, nblks, rtblks);
1445		if (error)
1446			goto error0;
1447	}
1448
1449	if (XFS_IS_GQUOTA_ON(mp)) {
1450		error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1451						   XFS_DQ_GROUP, nblks, rtblks);
1452		if (error)
1453			goto error0;
1454	}
1455
1456	if (XFS_IS_PQUOTA_ON(mp)) {
1457		error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1458						   XFS_DQ_PROJ, nblks, rtblks);
1459		if (error)
1460			goto error0;
1461	}
1462
1463	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1464	IRELE(ip);
1465	*res = BULKSTAT_RV_DIDONE;
1466	return 0;
1467
1468error0:
1469	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1470	IRELE(ip);
1471	*res = BULKSTAT_RV_GIVEUP;
1472	return error;
1473}
1474
1475STATIC int
1476xfs_qm_flush_one(
1477	struct xfs_dquot	*dqp,
1478	void			*data)
1479{
 
1480	struct list_head	*buffer_list = data;
1481	struct xfs_buf		*bp = NULL;
1482	int			error = 0;
1483
1484	xfs_dqlock(dqp);
1485	if (dqp->dq_flags & XFS_DQ_FREEING)
1486		goto out_unlock;
1487	if (!XFS_DQ_IS_DIRTY(dqp))
1488		goto out_unlock;
1489
1490	xfs_dqflock(dqp);
1491	error = xfs_qm_dqflush(dqp, &bp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1492	if (error)
1493		goto out_unlock;
 
 
 
 
1494
1495	xfs_buf_delwri_queue(bp, buffer_list);
 
 
1496	xfs_buf_relse(bp);
1497out_unlock:
1498	xfs_dqunlock(dqp);
1499	return error;
1500}
1501
1502/*
1503 * Walk thru all the filesystem inodes and construct a consistent view
1504 * of the disk quota world. If the quotacheck fails, disable quotas.
1505 */
1506int
1507xfs_qm_quotacheck(
1508	xfs_mount_t	*mp)
1509{
1510	int			done, count, error, error2;
1511	xfs_ino_t		lastino;
1512	size_t			structsz;
1513	uint			flags;
1514	LIST_HEAD		(buffer_list);
1515	struct xfs_inode	*uip = mp->m_quotainfo->qi_uquotaip;
1516	struct xfs_inode	*gip = mp->m_quotainfo->qi_gquotaip;
1517	struct xfs_inode	*pip = mp->m_quotainfo->qi_pquotaip;
1518
1519	count = INT_MAX;
1520	structsz = 1;
1521	lastino = 0;
1522	flags = 0;
1523
1524	ASSERT(uip || gip || pip);
1525	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1526
1527	xfs_notice(mp, "Quotacheck needed: Please wait.");
1528
1529	/*
1530	 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1531	 * their counters to zero. We need a clean slate.
1532	 * We don't log our changes till later.
1533	 */
1534	if (uip) {
1535		error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1536					 &buffer_list);
1537		if (error)
1538			goto error_return;
1539		flags |= XFS_UQUOTA_CHKD;
1540	}
1541
1542	if (gip) {
1543		error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1544					 &buffer_list);
1545		if (error)
1546			goto error_return;
1547		flags |= XFS_GQUOTA_CHKD;
1548	}
1549
1550	if (pip) {
1551		error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1552					 &buffer_list);
1553		if (error)
1554			goto error_return;
1555		flags |= XFS_PQUOTA_CHKD;
1556	}
1557
1558	do {
1559		/*
1560		 * Iterate thru all the inodes in the file system,
1561		 * adjusting the corresponding dquot counters in core.
1562		 */
1563		error = xfs_bulkstat(mp, &lastino, &count,
1564				     xfs_qm_dqusage_adjust,
1565				     structsz, NULL, &done);
1566		if (error)
1567			break;
1568
1569	} while (!done);
 
 
 
 
 
 
1570
1571	/*
1572	 * We've made all the changes that we need to make incore.  Flush them
1573	 * down to disk buffers if everything was updated successfully.
1574	 */
1575	if (XFS_IS_UQUOTA_ON(mp)) {
1576		error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1577					  &buffer_list);
1578	}
1579	if (XFS_IS_GQUOTA_ON(mp)) {
1580		error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1581					   &buffer_list);
1582		if (!error)
1583			error = error2;
1584	}
1585	if (XFS_IS_PQUOTA_ON(mp)) {
1586		error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1587					   &buffer_list);
1588		if (!error)
1589			error = error2;
1590	}
1591
1592	error2 = xfs_buf_delwri_submit(&buffer_list);
1593	if (!error)
1594		error = error2;
1595
1596	/*
1597	 * We can get this error if we couldn't do a dquot allocation inside
1598	 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1599	 * dirty dquots that might be cached, we just want to get rid of them
1600	 * and turn quotaoff. The dquots won't be attached to any of the inodes
1601	 * at this point (because we intentionally didn't in dqget_noattach).
1602	 */
1603	if (error) {
1604		xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1605		goto error_return;
1606	}
1607
1608	/*
1609	 * If one type of quotas is off, then it will lose its
1610	 * quotachecked status, since we won't be doing accounting for
1611	 * that type anymore.
1612	 */
1613	mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1614	mp->m_qflags |= flags;
1615
1616 error_return:
1617	while (!list_empty(&buffer_list)) {
1618		struct xfs_buf *bp =
1619			list_first_entry(&buffer_list, struct xfs_buf, b_list);
1620		list_del_init(&bp->b_list);
1621		xfs_buf_relse(bp);
1622	}
1623
1624	if (error) {
1625		xfs_warn(mp,
1626	"Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1627			error);
1628		/*
1629		 * We must turn off quotas.
1630		 */
1631		ASSERT(mp->m_quotainfo != NULL);
1632		xfs_qm_destroy_quotainfo(mp);
1633		if (xfs_mount_reset_sbqflags(mp)) {
1634			xfs_warn(mp,
1635				"Quotacheck: Failed to reset quota flags.");
1636		}
1637	} else
 
1638		xfs_notice(mp, "Quotacheck: Done.");
1639	return (error);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1640}
1641
1642/*
1643 * This is called after the superblock has been read in and we're ready to
1644 * iget the quota inodes.
1645 */
1646STATIC int
1647xfs_qm_init_quotainos(
1648	xfs_mount_t	*mp)
1649{
1650	struct xfs_inode	*uip = NULL;
1651	struct xfs_inode	*gip = NULL;
1652	struct xfs_inode	*pip = NULL;
1653	int			error;
1654	__int64_t		sbflags = 0;
1655	uint			flags = 0;
1656
1657	ASSERT(mp->m_quotainfo);
1658
1659	/*
1660	 * Get the uquota and gquota inodes
1661	 */
1662	if (xfs_sb_version_hasquota(&mp->m_sb)) {
1663		if (XFS_IS_UQUOTA_ON(mp) &&
1664		    mp->m_sb.sb_uquotino != NULLFSINO) {
1665			ASSERT(mp->m_sb.sb_uquotino > 0);
1666			error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1667					     0, 0, &uip);
1668			if (error)
1669				return XFS_ERROR(error);
1670		}
1671		if (XFS_IS_GQUOTA_ON(mp) &&
1672		    mp->m_sb.sb_gquotino != NULLFSINO) {
1673			ASSERT(mp->m_sb.sb_gquotino > 0);
1674			error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1675					     0, 0, &gip);
1676			if (error)
1677				goto error_rele;
1678		}
1679		if (XFS_IS_PQUOTA_ON(mp) &&
1680		    mp->m_sb.sb_pquotino != NULLFSINO) {
1681			ASSERT(mp->m_sb.sb_pquotino > 0);
1682			error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1683					     0, 0, &pip);
1684			if (error)
1685				goto error_rele;
1686		}
1687	} else {
1688		flags |= XFS_QMOPT_SBVERSION;
1689		sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1690			    XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1691			    XFS_SB_QFLAGS);
1692	}
1693
1694	/*
1695	 * Create the three inodes, if they don't exist already. The changes
1696	 * made above will get added to a transaction and logged in one of
1697	 * the qino_alloc calls below.  If the device is readonly,
1698	 * temporarily switch to read-write to do this.
1699	 */
1700	if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1701		error = xfs_qm_qino_alloc(mp, &uip,
1702					      sbflags | XFS_SB_UQUOTINO,
1703					      flags | XFS_QMOPT_UQUOTA);
1704		if (error)
1705			goto error_rele;
1706
1707		flags &= ~XFS_QMOPT_SBVERSION;
1708	}
1709	if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1710		error = xfs_qm_qino_alloc(mp, &gip,
1711					  sbflags | XFS_SB_GQUOTINO,
1712					  flags | XFS_QMOPT_GQUOTA);
1713		if (error)
1714			goto error_rele;
1715
1716		flags &= ~XFS_QMOPT_SBVERSION;
1717	}
1718	if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1719		error = xfs_qm_qino_alloc(mp, &pip,
1720					  sbflags | XFS_SB_PQUOTINO,
1721					  flags | XFS_QMOPT_PQUOTA);
1722		if (error)
1723			goto error_rele;
1724	}
1725
1726	mp->m_quotainfo->qi_uquotaip = uip;
1727	mp->m_quotainfo->qi_gquotaip = gip;
1728	mp->m_quotainfo->qi_pquotaip = pip;
1729
1730	return 0;
1731
1732error_rele:
1733	if (uip)
1734		IRELE(uip);
1735	if (gip)
1736		IRELE(gip);
1737	if (pip)
1738		IRELE(pip);
1739	return XFS_ERROR(error);
1740}
1741
1742STATIC void
1743xfs_qm_dqfree_one(
1744	struct xfs_dquot	*dqp)
1745{
1746	struct xfs_mount	*mp = dqp->q_mount;
1747	struct xfs_quotainfo	*qi = mp->m_quotainfo;
1748
1749	mutex_lock(&qi->qi_tree_lock);
1750	radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1751			  be32_to_cpu(dqp->q_core.d_id));
1752
1753	qi->qi_dquots--;
1754	mutex_unlock(&qi->qi_tree_lock);
1755
1756	xfs_qm_dqdestroy(dqp);
1757}
1758
1759/*
1760 * Start a transaction and write the incore superblock changes to
1761 * disk. flags parameter indicates which fields have changed.
1762 */
1763int
1764xfs_qm_write_sb_changes(
1765	xfs_mount_t	*mp,
1766	__int64_t	flags)
1767{
1768	xfs_trans_t	*tp;
1769	int		error;
1770
1771	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1772	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1773	if (error) {
1774		xfs_trans_cancel(tp, 0);
1775		return error;
1776	}
1777
1778	xfs_mod_sb(tp, flags);
1779	error = xfs_trans_commit(tp, 0);
1780
1781	return error;
1782}
1783
1784
1785/* --------------- utility functions for vnodeops ---------------- */
1786
1787
1788/*
1789 * Given an inode, a uid, gid and prid make sure that we have
1790 * allocated relevant dquot(s) on disk, and that we won't exceed inode
1791 * quotas by creating this file.
1792 * This also attaches dquot(s) to the given inode after locking it,
1793 * and returns the dquots corresponding to the uid and/or gid.
1794 *
1795 * in	: inode (unlocked)
1796 * out	: udquot, gdquot with references taken and unlocked
1797 */
1798int
1799xfs_qm_vop_dqalloc(
1800	struct xfs_inode	*ip,
1801	xfs_dqid_t		uid,
1802	xfs_dqid_t		gid,
1803	prid_t			prid,
1804	uint			flags,
1805	struct xfs_dquot	**O_udqpp,
1806	struct xfs_dquot	**O_gdqpp,
1807	struct xfs_dquot	**O_pdqpp)
1808{
1809	struct xfs_mount	*mp = ip->i_mount;
 
 
1810	struct xfs_dquot	*uq = NULL;
1811	struct xfs_dquot	*gq = NULL;
1812	struct xfs_dquot	*pq = NULL;
1813	int			error;
1814	uint			lockflags;
1815
1816	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1817		return 0;
1818
 
 
1819	lockflags = XFS_ILOCK_EXCL;
1820	xfs_ilock(ip, lockflags);
1821
1822	if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1823		gid = ip->i_d.di_gid;
1824
1825	/*
1826	 * Attach the dquot(s) to this inode, doing a dquot allocation
1827	 * if necessary. The dquot(s) will not be locked.
1828	 */
1829	if (XFS_NOT_DQATTACHED(mp, ip)) {
1830		error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1831		if (error) {
1832			xfs_iunlock(ip, lockflags);
1833			return error;
1834		}
1835	}
1836
1837	if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1838		if (ip->i_d.di_uid != uid) {
 
1839			/*
1840			 * What we need is the dquot that has this uid, and
1841			 * if we send the inode to dqget, the uid of the inode
1842			 * takes priority over what's sent in the uid argument.
1843			 * We must unlock inode here before calling dqget if
1844			 * we're not sending the inode, because otherwise
1845			 * we'll deadlock by doing trans_reserve while
1846			 * holding ilock.
1847			 */
1848			xfs_iunlock(ip, lockflags);
1849			error = xfs_qm_dqget(mp, NULL, uid,
1850						 XFS_DQ_USER,
1851						 XFS_QMOPT_DQALLOC |
1852						 XFS_QMOPT_DOWARN,
1853						 &uq);
1854			if (error) {
1855				ASSERT(error != ENOENT);
1856				return error;
1857			}
1858			/*
1859			 * Get the ilock in the right order.
1860			 */
1861			xfs_dqunlock(uq);
1862			lockflags = XFS_ILOCK_SHARED;
1863			xfs_ilock(ip, lockflags);
1864		} else {
1865			/*
1866			 * Take an extra reference, because we'll return
1867			 * this to caller
1868			 */
1869			ASSERT(ip->i_udquot);
1870			uq = xfs_qm_dqhold(ip->i_udquot);
1871		}
1872	}
1873	if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1874		if (ip->i_d.di_gid != gid) {
 
1875			xfs_iunlock(ip, lockflags);
1876			error = xfs_qm_dqget(mp, NULL, gid,
1877						 XFS_DQ_GROUP,
1878						 XFS_QMOPT_DQALLOC |
1879						 XFS_QMOPT_DOWARN,
1880						 &gq);
1881			if (error) {
1882				ASSERT(error != ENOENT);
1883				goto error_rele;
1884			}
1885			xfs_dqunlock(gq);
1886			lockflags = XFS_ILOCK_SHARED;
1887			xfs_ilock(ip, lockflags);
1888		} else {
1889			ASSERT(ip->i_gdquot);
1890			gq = xfs_qm_dqhold(ip->i_gdquot);
1891		}
1892	}
1893	if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1894		if (xfs_get_projid(ip) != prid) {
 
1895			xfs_iunlock(ip, lockflags);
1896			error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1897						 XFS_DQ_PROJ,
1898						 XFS_QMOPT_DQALLOC |
1899						 XFS_QMOPT_DOWARN,
1900						 &pq);
1901			if (error) {
1902				ASSERT(error != ENOENT);
1903				goto error_rele;
1904			}
1905			xfs_dqunlock(pq);
1906			lockflags = XFS_ILOCK_SHARED;
1907			xfs_ilock(ip, lockflags);
1908		} else {
1909			ASSERT(ip->i_pdquot);
1910			pq = xfs_qm_dqhold(ip->i_pdquot);
1911		}
1912	}
1913	if (uq)
1914		trace_xfs_dquot_dqalloc(ip);
1915
1916	xfs_iunlock(ip, lockflags);
1917	if (O_udqpp)
1918		*O_udqpp = uq;
1919	else if (uq)
1920		xfs_qm_dqrele(uq);
1921	if (O_gdqpp)
1922		*O_gdqpp = gq;
1923	else if (gq)
1924		xfs_qm_dqrele(gq);
1925	if (O_pdqpp)
1926		*O_pdqpp = pq;
1927	else if (pq)
1928		xfs_qm_dqrele(pq);
1929	return 0;
1930
1931error_rele:
1932	if (gq)
1933		xfs_qm_dqrele(gq);
1934	if (uq)
1935		xfs_qm_dqrele(uq);
1936	return error;
1937}
1938
1939/*
1940 * Actually transfer ownership, and do dquot modifications.
1941 * These were already reserved.
1942 */
1943xfs_dquot_t *
1944xfs_qm_vop_chown(
1945	xfs_trans_t	*tp,
1946	xfs_inode_t	*ip,
1947	xfs_dquot_t	**IO_olddq,
1948	xfs_dquot_t	*newdq)
1949{
1950	xfs_dquot_t	*prevdq;
1951	uint		bfield = XFS_IS_REALTIME_INODE(ip) ?
1952				 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1953
1954
1955	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1956	ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1957
1958	/* old dquot */
1959	prevdq = *IO_olddq;
1960	ASSERT(prevdq);
1961	ASSERT(prevdq != newdq);
1962
1963	xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1964	xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
 
 
 
 
 
1965
1966	/* the sparkling new dquot */
1967	xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1968	xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1969
1970	/*
1971	 * Take an extra reference, because the inode is going to keep
1972	 * this dquot pointer even after the trans_commit.
1973	 */
1974	*IO_olddq = xfs_qm_dqhold(newdq);
1975
1976	return prevdq;
1977}
1978
1979/*
1980 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1981 */
1982int
1983xfs_qm_vop_chown_reserve(
1984	struct xfs_trans	*tp,
1985	struct xfs_inode	*ip,
1986	struct xfs_dquot	*udqp,
1987	struct xfs_dquot	*gdqp,
1988	struct xfs_dquot	*pdqp,
1989	uint			flags)
1990{
1991	struct xfs_mount	*mp = ip->i_mount;
1992	uint			delblks, blkflags, prjflags = 0;
1993	struct xfs_dquot	*udq_unres = NULL;
1994	struct xfs_dquot	*gdq_unres = NULL;
1995	struct xfs_dquot	*pdq_unres = NULL;
1996	struct xfs_dquot	*udq_delblks = NULL;
1997	struct xfs_dquot	*gdq_delblks = NULL;
1998	struct xfs_dquot	*pdq_delblks = NULL;
1999	int			error;
2000
2001
2002	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2003	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2004
2005	delblks = ip->i_delayed_blks;
2006	blkflags = XFS_IS_REALTIME_INODE(ip) ?
2007			XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2008
2009	if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2010	    ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
2011		udq_delblks = udqp;
2012		/*
2013		 * If there are delayed allocation blocks, then we have to
2014		 * unreserve those from the old dquot, and add them to the
2015		 * new dquot.
2016		 */
2017		if (delblks) {
2018			ASSERT(ip->i_udquot);
2019			udq_unres = ip->i_udquot;
2020		}
2021	}
2022	if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
2023	    ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
2024		gdq_delblks = gdqp;
2025		if (delblks) {
2026			ASSERT(ip->i_gdquot);
2027			gdq_unres = ip->i_gdquot;
2028		}
2029	}
2030
2031	if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
2032	    xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
2033		prjflags = XFS_QMOPT_ENOSPC;
2034		pdq_delblks = pdqp;
2035		if (delblks) {
2036			ASSERT(ip->i_pdquot);
2037			pdq_unres = ip->i_pdquot;
2038		}
2039	}
2040
2041	error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2042				udq_delblks, gdq_delblks, pdq_delblks,
2043				ip->i_d.di_nblocks, 1,
2044				flags | blkflags | prjflags);
2045	if (error)
2046		return error;
2047
2048	/*
2049	 * Do the delayed blks reservations/unreservations now. Since, these
2050	 * are done without the help of a transaction, if a reservation fails
2051	 * its previous reservations won't be automatically undone by trans
2052	 * code. So, we have to do it manually here.
2053	 */
2054	if (delblks) {
2055		/*
2056		 * Do the reservations first. Unreservation can't fail.
2057		 */
2058		ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
2059		ASSERT(udq_unres || gdq_unres || pdq_unres);
2060		error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2061			    udq_delblks, gdq_delblks, pdq_delblks,
2062			    (xfs_qcnt_t)delblks, 0,
2063			    flags | blkflags | prjflags);
2064		if (error)
2065			return error;
2066		xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2067				udq_unres, gdq_unres, pdq_unres,
2068				-((xfs_qcnt_t)delblks), 0, blkflags);
2069	}
2070
2071	return (0);
2072}
2073
2074int
2075xfs_qm_vop_rename_dqattach(
2076	struct xfs_inode	**i_tab)
2077{
2078	struct xfs_mount	*mp = i_tab[0]->i_mount;
2079	int			i;
2080
2081	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2082		return 0;
2083
2084	for (i = 0; (i < 4 && i_tab[i]); i++) {
2085		struct xfs_inode	*ip = i_tab[i];
2086		int			error;
2087
2088		/*
2089		 * Watch out for duplicate entries in the table.
2090		 */
2091		if (i == 0 || ip != i_tab[i-1]) {
2092			if (XFS_NOT_DQATTACHED(mp, ip)) {
2093				error = xfs_qm_dqattach(ip, 0);
2094				if (error)
2095					return error;
2096			}
2097		}
2098	}
2099	return 0;
2100}
2101
2102void
2103xfs_qm_vop_create_dqattach(
2104	struct xfs_trans	*tp,
2105	struct xfs_inode	*ip,
2106	struct xfs_dquot	*udqp,
2107	struct xfs_dquot	*gdqp,
2108	struct xfs_dquot	*pdqp)
2109{
2110	struct xfs_mount	*mp = tp->t_mountp;
2111
2112	if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2113		return;
2114
2115	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2116	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2117
2118	if (udqp && XFS_IS_UQUOTA_ON(mp)) {
2119		ASSERT(ip->i_udquot == NULL);
2120		ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2121
2122		ip->i_udquot = xfs_qm_dqhold(udqp);
2123		xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2124	}
2125	if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
2126		ASSERT(ip->i_gdquot == NULL);
2127		ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
 
2128		ip->i_gdquot = xfs_qm_dqhold(gdqp);
2129		xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2130	}
2131	if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
2132		ASSERT(ip->i_pdquot == NULL);
2133		ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
2134
2135		ip->i_pdquot = xfs_qm_dqhold(pdqp);
2136		xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
2137	}
 
 
2138}
2139