Linux Audio

Check our new training course

Loading...
v4.17
 
   1/*
   2 * Copyright (c) 2000-2003 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_format.h"
  21#include "xfs_log_format.h"
  22#include "xfs_shared.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_bit.h"
  25#include "xfs_mount.h"
  26#include "xfs_defer.h"
  27#include "xfs_inode.h"
  28#include "xfs_bmap.h"
  29#include "xfs_bmap_util.h"
  30#include "xfs_alloc.h"
  31#include "xfs_quota.h"
  32#include "xfs_error.h"
  33#include "xfs_trans.h"
  34#include "xfs_buf_item.h"
  35#include "xfs_trans_space.h"
  36#include "xfs_trans_priv.h"
  37#include "xfs_qm.h"
  38#include "xfs_cksum.h"
  39#include "xfs_trace.h"
  40#include "xfs_log.h"
  41#include "xfs_bmap_btree.h"
 
  42
  43/*
  44 * Lock order:
  45 *
  46 * ip->i_lock
  47 *   qi->qi_tree_lock
  48 *     dquot->q_qlock (xfs_dqlock() and friends)
  49 *       dquot->q_flush (xfs_dqflock() and friends)
  50 *       qi->qi_lru_lock
  51 *
  52 * If two dquots need to be locked the order is user before group/project,
  53 * otherwise by the lowest id first, see xfs_dqlock2.
  54 */
  55
  56struct kmem_zone		*xfs_qm_dqtrxzone;
  57static struct kmem_zone		*xfs_qm_dqzone;
  58
  59static struct lock_class_key xfs_dquot_group_class;
  60static struct lock_class_key xfs_dquot_project_class;
  61
  62/*
  63 * This is called to free all the memory associated with a dquot
  64 */
  65void
  66xfs_qm_dqdestroy(
  67	xfs_dquot_t	*dqp)
  68{
  69	ASSERT(list_empty(&dqp->q_lru));
  70
  71	kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
  72	mutex_destroy(&dqp->q_qlock);
  73
  74	XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
  75	kmem_zone_free(xfs_qm_dqzone, dqp);
  76}
  77
  78/*
  79 * If default limits are in force, push them into the dquot now.
  80 * We overwrite the dquot limits only if they are zero and this
  81 * is not the root dquot.
  82 */
  83void
  84xfs_qm_adjust_dqlimits(
  85	struct xfs_mount	*mp,
  86	struct xfs_dquot	*dq)
  87{
 
  88	struct xfs_quotainfo	*q = mp->m_quotainfo;
  89	struct xfs_disk_dquot	*d = &dq->q_core;
  90	struct xfs_def_quota	*defq;
  91	int			prealloc = 0;
  92
  93	ASSERT(d->d_id);
  94	defq = xfs_get_defquota(dq, q);
  95
  96	if (defq->bsoftlimit && !d->d_blk_softlimit) {
  97		d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit);
  98		prealloc = 1;
  99	}
 100	if (defq->bhardlimit && !d->d_blk_hardlimit) {
 101		d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit);
 102		prealloc = 1;
 103	}
 104	if (defq->isoftlimit && !d->d_ino_softlimit)
 105		d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit);
 106	if (defq->ihardlimit && !d->d_ino_hardlimit)
 107		d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit);
 108	if (defq->rtbsoftlimit && !d->d_rtb_softlimit)
 109		d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit);
 110	if (defq->rtbhardlimit && !d->d_rtb_hardlimit)
 111		d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit);
 112
 113	if (prealloc)
 114		xfs_dquot_set_prealloc_limits(dq);
 115}
 116
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 117/*
 118 * Check the limits and timers of a dquot and start or reset timers
 119 * if necessary.
 120 * This gets called even when quota enforcement is OFF, which makes our
 121 * life a little less complicated. (We just don't reject any quota
 122 * reservations in that case, when enforcement is off).
 123 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
 124 * enforcement's off.
 125 * In contrast, warnings are a little different in that they don't
 126 * 'automatically' get started when limits get exceeded.  They do
 127 * get reset to zero, however, when we find the count to be under
 128 * the soft limit (they are only ever set non-zero via userspace).
 129 */
 130void
 131xfs_qm_adjust_dqtimers(
 132	xfs_mount_t		*mp,
 133	xfs_disk_dquot_t	*d)
 134{
 135	ASSERT(d->d_id);
 136
 137#ifdef DEBUG
 138	if (d->d_blk_hardlimit)
 139		ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
 140		       be64_to_cpu(d->d_blk_hardlimit));
 141	if (d->d_ino_hardlimit)
 142		ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
 143		       be64_to_cpu(d->d_ino_hardlimit));
 144	if (d->d_rtb_hardlimit)
 145		ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
 146		       be64_to_cpu(d->d_rtb_hardlimit));
 147#endif
 148
 149	if (!d->d_btimer) {
 150		if ((d->d_blk_softlimit &&
 151		     (be64_to_cpu(d->d_bcount) >
 152		      be64_to_cpu(d->d_blk_softlimit))) ||
 153		    (d->d_blk_hardlimit &&
 154		     (be64_to_cpu(d->d_bcount) >
 155		      be64_to_cpu(d->d_blk_hardlimit)))) {
 156			d->d_btimer = cpu_to_be32(get_seconds() +
 157					mp->m_quotainfo->qi_btimelimit);
 158		} else {
 159			d->d_bwarns = 0;
 160		}
 161	} else {
 162		if ((!d->d_blk_softlimit ||
 163		     (be64_to_cpu(d->d_bcount) <=
 164		      be64_to_cpu(d->d_blk_softlimit))) &&
 165		    (!d->d_blk_hardlimit ||
 166		    (be64_to_cpu(d->d_bcount) <=
 167		     be64_to_cpu(d->d_blk_hardlimit)))) {
 168			d->d_btimer = 0;
 169		}
 170	}
 171
 172	if (!d->d_itimer) {
 173		if ((d->d_ino_softlimit &&
 174		     (be64_to_cpu(d->d_icount) >
 175		      be64_to_cpu(d->d_ino_softlimit))) ||
 176		    (d->d_ino_hardlimit &&
 177		     (be64_to_cpu(d->d_icount) >
 178		      be64_to_cpu(d->d_ino_hardlimit)))) {
 179			d->d_itimer = cpu_to_be32(get_seconds() +
 180					mp->m_quotainfo->qi_itimelimit);
 181		} else {
 182			d->d_iwarns = 0;
 183		}
 184	} else {
 185		if ((!d->d_ino_softlimit ||
 186		     (be64_to_cpu(d->d_icount) <=
 187		      be64_to_cpu(d->d_ino_softlimit)))  &&
 188		    (!d->d_ino_hardlimit ||
 189		     (be64_to_cpu(d->d_icount) <=
 190		      be64_to_cpu(d->d_ino_hardlimit)))) {
 191			d->d_itimer = 0;
 192		}
 193	}
 194
 195	if (!d->d_rtbtimer) {
 196		if ((d->d_rtb_softlimit &&
 197		     (be64_to_cpu(d->d_rtbcount) >
 198		      be64_to_cpu(d->d_rtb_softlimit))) ||
 199		    (d->d_rtb_hardlimit &&
 200		     (be64_to_cpu(d->d_rtbcount) >
 201		      be64_to_cpu(d->d_rtb_hardlimit)))) {
 202			d->d_rtbtimer = cpu_to_be32(get_seconds() +
 203					mp->m_quotainfo->qi_rtbtimelimit);
 204		} else {
 205			d->d_rtbwarns = 0;
 206		}
 207	} else {
 208		if ((!d->d_rtb_softlimit ||
 209		     (be64_to_cpu(d->d_rtbcount) <=
 210		      be64_to_cpu(d->d_rtb_softlimit))) &&
 211		    (!d->d_rtb_hardlimit ||
 212		     (be64_to_cpu(d->d_rtbcount) <=
 213		      be64_to_cpu(d->d_rtb_hardlimit)))) {
 214			d->d_rtbtimer = 0;
 215		}
 216	}
 217}
 218
 219/*
 220 * initialize a buffer full of dquots and log the whole thing
 221 */
 222STATIC void
 223xfs_qm_init_dquot_blk(
 224	xfs_trans_t	*tp,
 225	xfs_mount_t	*mp,
 226	xfs_dqid_t	id,
 227	uint		type,
 228	xfs_buf_t	*bp)
 229{
 230	struct xfs_quotainfo	*q = mp->m_quotainfo;
 231	xfs_dqblk_t	*d;
 232	xfs_dqid_t	curid;
 233	int		i;
 
 
 234
 235	ASSERT(tp);
 236	ASSERT(xfs_buf_islocked(bp));
 237
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 238	d = bp->b_addr;
 239
 240	/*
 241	 * ID of the first dquot in the block - id's are zero based.
 242	 */
 243	curid = id - (id % q->qi_dqperchunk);
 244	memset(d, 0, BBTOB(q->qi_dqchunklen));
 245	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
 246		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 247		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
 248		d->dd_diskdq.d_id = cpu_to_be32(curid);
 249		d->dd_diskdq.d_flags = type;
 250		if (xfs_sb_version_hascrc(&mp->m_sb)) {
 
 
 251			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
 252			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
 253					 XFS_DQUOT_CRC_OFF);
 254		}
 255	}
 256
 257	xfs_trans_dquot_buf(tp, bp,
 258			    (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
 259			    ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
 260			     XFS_BLF_GDQUOT_BUF)));
 261	xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 262}
 263
 264/*
 265 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
 266 * watermarks correspond to the soft and hard limits by default. If a soft limit
 267 * is not specified, we use 95% of the hard limit.
 268 */
 269void
 270xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
 271{
 272	uint64_t space;
 273
 274	dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
 275	dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
 276	if (!dqp->q_prealloc_lo_wmark) {
 277		dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
 278		do_div(dqp->q_prealloc_lo_wmark, 100);
 279		dqp->q_prealloc_lo_wmark *= 95;
 280	}
 281
 282	space = dqp->q_prealloc_hi_wmark;
 283
 284	do_div(space, 100);
 285	dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
 286	dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
 287	dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
 288}
 289
 290/*
 291 * Allocate a block and fill it with dquots.
 292 * This is called when the bmapi finds a hole.
 
 293 */
 294STATIC int
 295xfs_qm_dqalloc(
 296	xfs_trans_t	**tpp,
 297	xfs_mount_t	*mp,
 298	xfs_dquot_t	*dqp,
 299	xfs_inode_t	*quotip,
 300	xfs_fileoff_t	offset_fsb,
 301	xfs_buf_t	**O_bpp)
 302{
 303	xfs_fsblock_t	firstblock;
 304	struct xfs_defer_ops dfops;
 305	xfs_bmbt_irec_t map;
 306	int		nmaps, error;
 307	xfs_buf_t	*bp;
 308	xfs_trans_t	*tp = *tpp;
 309
 310	ASSERT(tp != NULL);
 311
 312	trace_xfs_dqalloc(dqp);
 313
 314	/*
 315	 * Initialize the bmap freelist prior to calling bmapi code.
 316	 */
 317	xfs_defer_init(&dfops, &firstblock);
 
 318	xfs_ilock(quotip, XFS_ILOCK_EXCL);
 319	/*
 320	 * Return if this type of quotas is turned off while we didn't
 321	 * have an inode lock
 322	 */
 323	if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
 324		xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 325		return -ESRCH;
 
 
 326	}
 327
 328	xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
 329	nmaps = 1;
 330	error = xfs_bmapi_write(tp, quotip, offset_fsb,
 331				XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
 332				&firstblock, XFS_QM_DQALLOC_SPACE_RES(mp),
 333				&map, &nmaps, &dfops);
 
 
 
 
 
 
 334	if (error)
 335		goto error0;
 
 336	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
 337	ASSERT(nmaps == 1);
 338	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 339	       (map.br_startblock != HOLESTARTBLOCK));
 340
 341	/*
 342	 * Keep track of the blkno to save a lookup later
 343	 */
 344	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 345
 346	/* now we can just get the buffer (there's nothing to read yet) */
 347	bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
 348			       dqp->q_blkno,
 349			       mp->m_quotainfo->qi_dqchunklen,
 350			       0);
 351	if (!bp) {
 352		error = -ENOMEM;
 353		goto error1;
 354	}
 355	bp->b_ops = &xfs_dquot_buf_ops;
 356
 357	/*
 358	 * Make a chunk of dquots out of this buffer and log
 359	 * the entire thing.
 360	 */
 361	xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
 362			      dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
 363
 364	/*
 365	 * xfs_defer_finish() may commit the current transaction and
 366	 * start a second transaction if the freelist is not empty.
 
 
 
 
 
 
 367	 *
 368	 * Since we still want to modify this buffer, we need to
 369	 * ensure that the buffer is not released on commit of
 370	 * the first transaction and ensure the buffer is added to the
 371	 * second transaction.
 372	 *
 373	 * If there is only one transaction then don't stop the buffer
 374	 * from being released when it commits later on.
 
 
 
 
 
 
 375	 */
 376
 377	xfs_trans_bhold(tp, bp);
 378
 379	error = xfs_defer_finish(tpp, &dfops);
 380	if (error)
 381		goto error1;
 382
 383	/* Transaction was committed? */
 384	if (*tpp != tp) {
 385		tp = *tpp;
 386		xfs_trans_bjoin(tp, bp);
 387	} else {
 388		xfs_trans_bhold_release(tp, bp);
 389	}
 390
 391	*O_bpp = bp;
 392	return 0;
 393
 394error1:
 395	xfs_defer_cancel(&dfops);
 396error0:
 397	return error;
 398}
 399
 400/*
 401 * Maps a dquot to the buffer containing its on-disk version.
 402 * This returns a ptr to the buffer containing the on-disk dquot
 403 * in the bpp param, and a ptr to the on-disk dquot within that buffer
 404 */
 405STATIC int
 406xfs_qm_dqtobp(
 407	xfs_trans_t		**tpp,
 408	xfs_dquot_t		*dqp,
 409	xfs_disk_dquot_t	**O_ddpp,
 410	xfs_buf_t		**O_bpp,
 411	uint			flags)
 412{
 413	struct xfs_bmbt_irec	map;
 414	int			nmaps = 1, error;
 415	struct xfs_buf		*bp;
 416	struct xfs_inode	*quotip;
 417	struct xfs_mount	*mp = dqp->q_mount;
 418	xfs_dqid_t		id = be32_to_cpu(dqp->q_core.d_id);
 419	struct xfs_trans	*tp = (tpp ? *tpp : NULL);
 420	uint			lock_mode;
 421
 422	quotip = xfs_quota_inode(dqp->q_mount, dqp->dq_flags);
 423	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
 424
 425	lock_mode = xfs_ilock_data_map_shared(quotip);
 426	if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
 427		/*
 428		 * Return if this type of quotas is turned off while we
 429		 * didn't have the quota inode lock.
 430		 */
 431		xfs_iunlock(quotip, lock_mode);
 432		return -ESRCH;
 433	}
 434
 435	/*
 436	 * Find the block map; no allocations yet
 437	 */
 438	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
 439			       XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
 440
 441	xfs_iunlock(quotip, lock_mode);
 442	if (error)
 443		return error;
 444
 445	ASSERT(nmaps == 1);
 446	ASSERT(map.br_blockcount == 1);
 447
 448	/*
 449	 * Offset of dquot in the (fixed sized) dquot chunk.
 450	 */
 451	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
 452		sizeof(xfs_dqblk_t);
 453
 454	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
 455	if (map.br_startblock == HOLESTARTBLOCK) {
 456		/*
 457		 * We don't allocate unless we're asked to
 458		 */
 459		if (!(flags & XFS_QMOPT_DQALLOC))
 460			return -ENOENT;
 461
 462		ASSERT(tp);
 463		error = xfs_qm_dqalloc(tpp, mp, dqp, quotip,
 464					dqp->q_fileoffset, &bp);
 465		if (error)
 466			return error;
 467		tp = *tpp;
 468	} else {
 469		trace_xfs_dqtobp_read(dqp);
 470
 471		/*
 472		 * store the blkno etc so that we don't have to do the
 473		 * mapping all the time
 474		 */
 475		dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 476
 477		error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
 478					   dqp->q_blkno,
 479					   mp->m_quotainfo->qi_dqchunklen,
 480					   0, &bp, &xfs_dquot_buf_ops);
 481		if (error) {
 482			ASSERT(bp == NULL);
 483			return error;
 484		}
 485	}
 486
 487	ASSERT(xfs_buf_islocked(bp));
 488	*O_bpp = bp;
 489	*O_ddpp = bp->b_addr + dqp->q_bufoffset;
 490
 491	return 0;
 492}
 493
 494
 495/*
 496 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
 497 * and release the buffer immediately.
 498 *
 499 * If XFS_QMOPT_DQALLOC is set, allocate a dquot on disk if it needed.
 500 */
 501int
 502xfs_qm_dqread(
 503	struct xfs_mount	*mp,
 504	xfs_dqid_t		id,
 505	uint			type,
 506	uint			flags,
 507	struct xfs_dquot	**O_dqpp)
 508{
 509	struct xfs_dquot	*dqp;
 510	struct xfs_disk_dquot	*ddqp;
 511	struct xfs_buf		*bp;
 512	struct xfs_trans	*tp = NULL;
 513	int			error;
 514
 515	dqp = kmem_zone_zalloc(xfs_qm_dqzone, KM_SLEEP);
 516
 517	dqp->dq_flags = type;
 518	dqp->q_core.d_id = cpu_to_be32(id);
 519	dqp->q_mount = mp;
 520	INIT_LIST_HEAD(&dqp->q_lru);
 521	mutex_init(&dqp->q_qlock);
 522	init_waitqueue_head(&dqp->q_pinwait);
 
 
 
 
 
 
 523
 524	/*
 525	 * Because we want to use a counting completion, complete
 526	 * the flush completion once to allow a single access to
 527	 * the flush completion without blocking.
 528	 */
 529	init_completion(&dqp->q_flush);
 530	complete(&dqp->q_flush);
 531
 532	/*
 533	 * Make sure group quotas have a different lock class than user
 534	 * quotas.
 535	 */
 536	switch (type) {
 537	case XFS_DQ_USER:
 538		/* uses the default lock class */
 539		break;
 540	case XFS_DQ_GROUP:
 541		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
 542		break;
 543	case XFS_DQ_PROJ:
 544		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
 545		break;
 546	default:
 547		ASSERT(0);
 548		break;
 549	}
 550
 
 
 551	XFS_STATS_INC(mp, xs_qm_dquot);
 
 
 552
 553	trace_xfs_dqread(dqp);
 
 
 
 
 
 
 
 554
 555	if (flags & XFS_QMOPT_DQALLOC) {
 556		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
 557				XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
 558		if (error)
 559			goto error0;
 560	}
 561
 562	/*
 563	 * get a pointer to the on-disk dquot and the buffer containing it
 564	 * dqp already knows its own type (GROUP/USER).
 
 565	 */
 566	error = xfs_qm_dqtobp(&tp, dqp, &ddqp, &bp, flags);
 567	if (error) {
 568		/*
 569		 * This can happen if quotas got turned off (ESRCH),
 570		 * or if the dquot didn't exist on disk and we ask to
 571		 * allocate (ENOENT).
 572		 */
 573		trace_xfs_dqread_fail(dqp);
 574		goto error1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 575	}
 576
 577	/* copy everything from disk dquot to the incore dquot */
 578	memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
 579	xfs_qm_dquot_logitem_init(dqp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 580
 581	/*
 582	 * Reservation counters are defined as reservation plus current usage
 583	 * to avoid having to add every time.
 584	 */
 585	dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
 586	dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
 587	dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
 588
 589	/* initialize the dquot speculative prealloc thresholds */
 590	xfs_dquot_set_prealloc_limits(dqp);
 
 
 591
 592	/* Mark the buf so that this will stay incore a little longer */
 593	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 594
 595	/*
 596	 * We got the buffer with a xfs_trans_read_buf() (in dqtobp())
 597	 * So we need to release with xfs_trans_brelse().
 598	 * The strategy here is identical to that of inodes; we lock
 599	 * the dquot in xfs_qm_dqget() before making it accessible to
 600	 * others. This is because dquots, like inodes, need a good level of
 601	 * concurrency, and we don't want to take locks on the entire buffers
 602	 * for dquot accesses.
 603	 * Note also that the dquot buffer may even be dirty at this point, if
 604	 * this particular dquot was repaired. We still aren't afraid to
 605	 * brelse it because we have the changes incore.
 606	 */
 607	ASSERT(xfs_buf_islocked(bp));
 608	xfs_trans_brelse(tp, bp);
 609
 610	if (tp) {
 611		error = xfs_trans_commit(tp);
 612		if (error)
 613			goto error0;
 614	}
 615
 616	*O_dqpp = dqp;
 617	return error;
 618
 619error1:
 620	if (tp)
 621		xfs_trans_cancel(tp);
 622error0:
 623	xfs_qm_dqdestroy(dqp);
 624	*O_dqpp = NULL;
 625	return error;
 626}
 627
 628/*
 629 * Advance to the next id in the current chunk, or if at the
 630 * end of the chunk, skip ahead to first id in next allocated chunk
 631 * using the SEEK_DATA interface.
 632 */
 633static int
 634xfs_dq_get_next_id(
 635	struct xfs_mount	*mp,
 636	uint			type,
 637	xfs_dqid_t		*id)
 638{
 639	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
 640	xfs_dqid_t		next_id = *id + 1; /* simple advance */
 641	uint			lock_flags;
 642	struct xfs_bmbt_irec	got;
 643	struct xfs_iext_cursor	cur;
 644	xfs_fsblock_t		start;
 645	int			error = 0;
 646
 647	/* If we'd wrap past the max ID, stop */
 648	if (next_id < *id)
 649		return -ENOENT;
 650
 651	/* If new ID is within the current chunk, advancing it sufficed */
 652	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
 653		*id = next_id;
 654		return 0;
 655	}
 656
 657	/* Nope, next_id is now past the current chunk, so find the next one */
 658	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
 659
 660	lock_flags = xfs_ilock_data_map_shared(quotip);
 661	if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) {
 662		error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
 663		if (error)
 664			return error;
 665	}
 666
 667	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
 668		/* contiguous chunk, bump startoff for the id calculation */
 669		if (got.br_startoff < start)
 670			got.br_startoff = start;
 671		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
 672	} else {
 673		error = -ENOENT;
 674	}
 675
 676	xfs_iunlock(quotip, lock_flags);
 677
 678	return error;
 679}
 680
 681/*
 682 * Given the file system, inode OR id, and type (UDQUOT/GDQUOT), return a
 683 * a locked dquot, doing an allocation (if requested) as needed.
 684 * When both an inode and an id are given, the inode's id takes precedence.
 685 * That is, if the id changes while we don't hold the ilock inside this
 686 * function, the new dquot is returned, not necessarily the one requested
 687 * in the id argument.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 688 */
 689int
 690xfs_qm_dqget(
 691	xfs_mount_t	*mp,
 692	xfs_inode_t	*ip,	  /* locked inode (optional) */
 693	xfs_dqid_t	id,	  /* uid/projid/gid depending on type */
 694	uint		type,	  /* XFS_DQ_USER/XFS_DQ_PROJ/XFS_DQ_GROUP */
 695	uint		flags,	  /* DQALLOC, DQSUSER, DQREPAIR, DOWARN */
 696	xfs_dquot_t	**O_dqpp) /* OUT : locked incore dquot */
 697{
 698	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 699	struct radix_tree_root *tree = xfs_dquot_tree(qi, type);
 700	struct xfs_dquot	*dqp;
 701	int			error;
 702
 703	ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 704	if ((! XFS_IS_UQUOTA_ON(mp) && type == XFS_DQ_USER) ||
 705	    (! XFS_IS_PQUOTA_ON(mp) && type == XFS_DQ_PROJ) ||
 706	    (! XFS_IS_GQUOTA_ON(mp) && type == XFS_DQ_GROUP)) {
 707		return -ESRCH;
 
 
 
 
 708	}
 709
 710	ASSERT(type == XFS_DQ_USER ||
 711	       type == XFS_DQ_PROJ ||
 712	       type == XFS_DQ_GROUP);
 713	if (ip) {
 714		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 715		ASSERT(xfs_inode_dquot(ip, type) == NULL);
 
 
 
 
 
 
 
 716	}
 717
 718restart:
 719	mutex_lock(&qi->qi_tree_lock);
 720	dqp = radix_tree_lookup(tree, id);
 721	if (dqp) {
 722		xfs_dqlock(dqp);
 723		if (dqp->dq_flags & XFS_DQ_FREEING) {
 724			xfs_dqunlock(dqp);
 725			mutex_unlock(&qi->qi_tree_lock);
 726			trace_xfs_dqget_freeing(dqp);
 727			delay(1);
 728			goto restart;
 729		}
 730
 731		/* uninit / unused quota found in radix tree, keep looking  */
 732		if (flags & XFS_QMOPT_DQNEXT) {
 733			if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
 734				xfs_dqunlock(dqp);
 735				mutex_unlock(&qi->qi_tree_lock);
 736				error = xfs_dq_get_next_id(mp, type, &id);
 737				if (error)
 738					return error;
 739				goto restart;
 740			}
 741		}
 
 
 
 742
 743		dqp->q_nrefs++;
 744		mutex_unlock(&qi->qi_tree_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 745
 746		trace_xfs_dqget_hit(dqp);
 747		XFS_STATS_INC(mp, xs_qm_dqcachehits);
 
 
 
 
 
 
 748		*O_dqpp = dqp;
 749		return 0;
 750	}
 751	mutex_unlock(&qi->qi_tree_lock);
 752	XFS_STATS_INC(mp, xs_qm_dqcachemisses);
 753
 754	/*
 755	 * Dquot cache miss. We don't want to keep the inode lock across
 756	 * a (potential) disk read. Also we don't want to deal with the lock
 757	 * ordering between quotainode and this inode. OTOH, dropping the inode
 758	 * lock here means dealing with a chown that can happen before
 759	 * we re-acquire the lock.
 760	 */
 761	if (ip)
 762		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 763
 764	error = xfs_qm_dqread(mp, id, type, flags, &dqp);
 765
 766	if (ip)
 767		xfs_ilock(ip, XFS_ILOCK_EXCL);
 768
 769	/* If we are asked to find next active id, keep looking */
 770	if (error == -ENOENT && (flags & XFS_QMOPT_DQNEXT)) {
 771		error = xfs_dq_get_next_id(mp, type, &id);
 772		if (!error)
 773			goto restart;
 774	}
 775
 776	if (error)
 777		return error;
 778
 779	if (ip) {
 780		/*
 781		 * A dquot could be attached to this inode by now, since
 782		 * we had dropped the ilock.
 783		 */
 784		if (xfs_this_quota_on(mp, type)) {
 785			struct xfs_dquot	*dqp1;
 786
 787			dqp1 = xfs_inode_dquot(ip, type);
 788			if (dqp1) {
 789				xfs_qm_dqdestroy(dqp);
 790				dqp = dqp1;
 791				xfs_dqlock(dqp);
 792				goto dqret;
 793			}
 794		} else {
 795			/* inode stays locked on return */
 796			xfs_qm_dqdestroy(dqp);
 797			return -ESRCH;
 
 
 798		}
 
 
 
 
 799	}
 800
 801	mutex_lock(&qi->qi_tree_lock);
 802	error = radix_tree_insert(tree, id, dqp);
 803	if (unlikely(error)) {
 804		WARN_ON(error != -EEXIST);
 805
 806		/*
 807		 * Duplicate found. Just throw away the new dquot and start
 808		 * over.
 809		 */
 810		mutex_unlock(&qi->qi_tree_lock);
 811		trace_xfs_dqget_dup(dqp);
 812		xfs_qm_dqdestroy(dqp);
 813		XFS_STATS_INC(mp, xs_qm_dquot_dups);
 814		goto restart;
 815	}
 816
 817	/*
 818	 * We return a locked dquot to the caller, with a reference taken
 819	 */
 820	xfs_dqlock(dqp);
 821	dqp->q_nrefs = 1;
 
 822
 823	qi->qi_dquots++;
 824	mutex_unlock(&qi->qi_tree_lock);
 
 
 
 
 
 
 
 
 
 
 
 825
 826	/* If we are asked to find next active id, keep looking */
 827	if (flags & XFS_QMOPT_DQNEXT) {
 828		if (XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
 829			xfs_qm_dqput(dqp);
 830			error = xfs_dq_get_next_id(mp, type, &id);
 831			if (error)
 832				return error;
 833			goto restart;
 
 
 
 834		}
 
 
 835	}
 836
 837 dqret:
 838	ASSERT((ip == NULL) || xfs_isilocked(ip, XFS_ILOCK_EXCL));
 839	trace_xfs_dqget_miss(dqp);
 840	*O_dqpp = dqp;
 841	return 0;
 842}
 843
 844/*
 845 * Release a reference to the dquot (decrement ref-count) and unlock it.
 846 *
 847 * If there is a group quota attached to this dquot, carefully release that
 848 * too without tripping over deadlocks'n'stuff.
 849 */
 850void
 851xfs_qm_dqput(
 852	struct xfs_dquot	*dqp)
 853{
 854	ASSERT(dqp->q_nrefs > 0);
 855	ASSERT(XFS_DQ_IS_LOCKED(dqp));
 856
 857	trace_xfs_dqput(dqp);
 858
 859	if (--dqp->q_nrefs == 0) {
 860		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
 861		trace_xfs_dqput_free(dqp);
 862
 863		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
 864			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
 865	}
 866	xfs_dqunlock(dqp);
 867}
 868
 869/*
 870 * Release a dquot. Flush it if dirty, then dqput() it.
 871 * dquot must not be locked.
 872 */
 873void
 874xfs_qm_dqrele(
 875	xfs_dquot_t	*dqp)
 876{
 877	if (!dqp)
 878		return;
 879
 880	trace_xfs_dqrele(dqp);
 881
 882	xfs_dqlock(dqp);
 883	/*
 884	 * We don't care to flush it if the dquot is dirty here.
 885	 * That will create stutters that we want to avoid.
 886	 * Instead we do a delayed write when we try to reclaim
 887	 * a dirty dquot. Also xfs_sync will take part of the burden...
 888	 */
 889	xfs_qm_dqput(dqp);
 890}
 891
 892/*
 893 * This is the dquot flushing I/O completion routine.  It is called
 894 * from interrupt level when the buffer containing the dquot is
 895 * flushed to disk.  It is responsible for removing the dquot logitem
 896 * from the AIL if it has not been re-logged, and unlocking the dquot's
 897 * flush lock. This behavior is very similar to that of inodes..
 898 */
 899STATIC void
 900xfs_qm_dqflush_done(
 901	struct xfs_buf		*bp,
 902	struct xfs_log_item	*lip)
 903{
 904	xfs_dq_logitem_t	*qip = (struct xfs_dq_logitem *)lip;
 905	xfs_dquot_t		*dqp = qip->qli_dquot;
 906	struct xfs_ail		*ailp = lip->li_ailp;
 
 907
 908	/*
 909	 * We only want to pull the item from the AIL if its
 910	 * location in the log has not changed since we started the flush.
 911	 * Thus, we only bother if the dquot's lsn has
 912	 * not changed. First we check the lsn outside the lock
 913	 * since it's cheaper, and then we recheck while
 914	 * holding the lock before removing the dquot from the AIL.
 915	 */
 916	if ((lip->li_flags & XFS_LI_IN_AIL) &&
 917	    ((lip->li_lsn == qip->qli_flush_lsn) ||
 918	     (lip->li_flags & XFS_LI_FAILED))) {
 919
 920		/* xfs_trans_ail_delete() drops the AIL lock. */
 921		spin_lock(&ailp->ail_lock);
 
 922		if (lip->li_lsn == qip->qli_flush_lsn) {
 923			xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
 
 
 924		} else {
 925			/*
 926			 * Clear the failed state since we are about to drop the
 927			 * flush lock
 928			 */
 929			if (lip->li_flags & XFS_LI_FAILED)
 930				xfs_clear_li_failed(lip);
 931			spin_unlock(&ailp->ail_lock);
 932		}
 933	}
 934
 935	/*
 936	 * Release the dq's flush lock since we're done with it.
 937	 */
 938	xfs_dqfunlock(dqp);
 939}
 940
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 941/*
 942 * Write a modified dquot to disk.
 943 * The dquot must be locked and the flush lock too taken by caller.
 944 * The flush lock will not be unlocked until the dquot reaches the disk,
 945 * but the dquot is free to be unlocked and modified by the caller
 946 * in the interim. Dquot is still locked on return. This behavior is
 947 * identical to that of inodes.
 948 */
 949int
 950xfs_qm_dqflush(
 951	struct xfs_dquot	*dqp,
 952	struct xfs_buf		**bpp)
 953{
 954	struct xfs_mount	*mp = dqp->q_mount;
 
 955	struct xfs_buf		*bp;
 956	struct xfs_disk_dquot	*ddqp;
 957	xfs_failaddr_t		fa;
 958	int			error;
 959
 960	ASSERT(XFS_DQ_IS_LOCKED(dqp));
 961	ASSERT(!completion_done(&dqp->q_flush));
 962
 963	trace_xfs_dqflush(dqp);
 964
 965	*bpp = NULL;
 966
 967	xfs_qm_dqunpin_wait(dqp);
 968
 969	/*
 970	 * This may have been unpinned because the filesystem is shutting
 971	 * down forcibly. If that's the case we must not write this dquot
 972	 * to disk, because the log record didn't make it to disk.
 973	 *
 974	 * We also have to remove the log item from the AIL in this case,
 975	 * as we wait for an emptry AIL as part of the unmount process.
 976	 */
 977	if (XFS_FORCED_SHUTDOWN(mp)) {
 978		struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
 979		dqp->dq_flags &= ~XFS_DQ_DIRTY;
 980
 981		xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE);
 982
 983		error = -EIO;
 984		goto out_unlock;
 985	}
 986
 987	/*
 988	 * Get the buffer containing the on-disk dquot
 989	 */
 990	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
 991				   mp->m_quotainfo->qi_dqchunklen, 0, &bp,
 992				   &xfs_dquot_buf_ops);
 993	if (error)
 994		goto out_unlock;
 
 
 995
 996	/*
 997	 * Calculate the location of the dquot inside the buffer.
 998	 */
 999	ddqp = bp->b_addr + dqp->q_bufoffset;
1000
1001	/*
1002	 * A simple sanity check in case we got a corrupted dquot..
1003	 */
1004	fa = xfs_dquot_verify(mp, &dqp->q_core, be32_to_cpu(ddqp->d_id), 0, 0);
1005	if (fa) {
1006		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1007				be32_to_cpu(ddqp->d_id), fa);
1008		xfs_buf_relse(bp);
1009		xfs_dqfunlock(dqp);
1010		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1011		return -EIO;
1012	}
1013
1014	/* This is the only portion of data that needs to persist */
1015	memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
 
1016
1017	/*
1018	 * Clear the dirty field and remember the flush lsn for later use.
1019	 */
1020	dqp->dq_flags &= ~XFS_DQ_DIRTY;
1021
1022	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1023					&dqp->q_logitem.qli_item.li_lsn);
1024
1025	/*
1026	 * copy the lsn into the on-disk dquot now while we have the in memory
1027	 * dquot here. This can't be done later in the write verifier as we
1028	 * can't get access to the log item at that point in time.
1029	 *
1030	 * We also calculate the CRC here so that the on-disk dquot in the
1031	 * buffer always has a valid CRC. This ensures there is no possibility
1032	 * of a dquot without an up-to-date CRC getting to disk.
1033	 */
1034	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1035		struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
1036
1037		dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1038		xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1039				 XFS_DQUOT_CRC_OFF);
1040	}
1041
1042	/*
1043	 * Attach an iodone routine so that we can remove this dquot from the
1044	 * AIL and release the flush lock once the dquot is synced to disk.
1045	 */
1046	xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1047				  &dqp->q_logitem.qli_item);
1048
1049	/*
1050	 * If the buffer is pinned then push on the log so we won't
1051	 * get stuck waiting in the write for too long.
1052	 */
1053	if (xfs_buf_ispinned(bp)) {
1054		trace_xfs_dqflush_force(dqp);
1055		xfs_log_force(mp, 0);
1056	}
1057
1058	trace_xfs_dqflush_done(dqp);
1059	*bpp = bp;
1060	return 0;
1061
 
 
 
 
1062out_unlock:
1063	xfs_dqfunlock(dqp);
1064	return -EIO;
1065}
1066
1067/*
1068 * Lock two xfs_dquot structures.
1069 *
1070 * To avoid deadlocks we always lock the quota structure with
1071 * the lowerd id first.
1072 */
1073void
1074xfs_dqlock2(
1075	xfs_dquot_t	*d1,
1076	xfs_dquot_t	*d2)
1077{
1078	if (d1 && d2) {
1079		ASSERT(d1 != d2);
1080		if (be32_to_cpu(d1->q_core.d_id) >
1081		    be32_to_cpu(d2->q_core.d_id)) {
1082			mutex_lock(&d2->q_qlock);
1083			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1084		} else {
1085			mutex_lock(&d1->q_qlock);
1086			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1087		}
1088	} else if (d1) {
1089		mutex_lock(&d1->q_qlock);
1090	} else if (d2) {
1091		mutex_lock(&d2->q_qlock);
1092	}
1093}
1094
1095int __init
1096xfs_qm_init(void)
1097{
1098	xfs_qm_dqzone =
1099		kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1100	if (!xfs_qm_dqzone)
 
1101		goto out;
1102
1103	xfs_qm_dqtrxzone =
1104		kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1105	if (!xfs_qm_dqtrxzone)
1106		goto out_free_dqzone;
 
1107
1108	return 0;
1109
1110out_free_dqzone:
1111	kmem_zone_destroy(xfs_qm_dqzone);
1112out:
1113	return -ENOMEM;
1114}
1115
1116void
1117xfs_qm_exit(void)
1118{
1119	kmem_zone_destroy(xfs_qm_dqtrxzone);
1120	kmem_zone_destroy(xfs_qm_dqzone);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1121}
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2003 Silicon Graphics, Inc.
   4 * All Rights Reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_format.h"
   9#include "xfs_log_format.h"
  10#include "xfs_shared.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_bit.h"
  13#include "xfs_mount.h"
  14#include "xfs_defer.h"
  15#include "xfs_inode.h"
  16#include "xfs_bmap.h"
 
 
  17#include "xfs_quota.h"
 
  18#include "xfs_trans.h"
  19#include "xfs_buf_item.h"
  20#include "xfs_trans_space.h"
  21#include "xfs_trans_priv.h"
  22#include "xfs_qm.h"
 
  23#include "xfs_trace.h"
  24#include "xfs_log.h"
  25#include "xfs_bmap_btree.h"
  26#include "xfs_error.h"
  27
  28/*
  29 * Lock order:
  30 *
  31 * ip->i_lock
  32 *   qi->qi_tree_lock
  33 *     dquot->q_qlock (xfs_dqlock() and friends)
  34 *       dquot->q_flush (xfs_dqflock() and friends)
  35 *       qi->qi_lru_lock
  36 *
  37 * If two dquots need to be locked the order is user before group/project,
  38 * otherwise by the lowest id first, see xfs_dqlock2.
  39 */
  40
  41struct kmem_cache		*xfs_dqtrx_cache;
  42static struct kmem_cache	*xfs_dquot_cache;
  43
  44static struct lock_class_key xfs_dquot_group_class;
  45static struct lock_class_key xfs_dquot_project_class;
  46
  47/*
  48 * This is called to free all the memory associated with a dquot
  49 */
  50void
  51xfs_qm_dqdestroy(
  52	struct xfs_dquot	*dqp)
  53{
  54	ASSERT(list_empty(&dqp->q_lru));
  55
  56	kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
  57	mutex_destroy(&dqp->q_qlock);
  58
  59	XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
  60	kmem_cache_free(xfs_dquot_cache, dqp);
  61}
  62
  63/*
  64 * If default limits are in force, push them into the dquot now.
  65 * We overwrite the dquot limits only if they are zero and this
  66 * is not the root dquot.
  67 */
  68void
  69xfs_qm_adjust_dqlimits(
 
  70	struct xfs_dquot	*dq)
  71{
  72	struct xfs_mount	*mp = dq->q_mount;
  73	struct xfs_quotainfo	*q = mp->m_quotainfo;
 
  74	struct xfs_def_quota	*defq;
  75	int			prealloc = 0;
  76
  77	ASSERT(dq->q_id);
  78	defq = xfs_get_defquota(q, xfs_dquot_type(dq));
  79
  80	if (!dq->q_blk.softlimit) {
  81		dq->q_blk.softlimit = defq->blk.soft;
  82		prealloc = 1;
  83	}
  84	if (!dq->q_blk.hardlimit) {
  85		dq->q_blk.hardlimit = defq->blk.hard;
  86		prealloc = 1;
  87	}
  88	if (!dq->q_ino.softlimit)
  89		dq->q_ino.softlimit = defq->ino.soft;
  90	if (!dq->q_ino.hardlimit)
  91		dq->q_ino.hardlimit = defq->ino.hard;
  92	if (!dq->q_rtb.softlimit)
  93		dq->q_rtb.softlimit = defq->rtb.soft;
  94	if (!dq->q_rtb.hardlimit)
  95		dq->q_rtb.hardlimit = defq->rtb.hard;
  96
  97	if (prealloc)
  98		xfs_dquot_set_prealloc_limits(dq);
  99}
 100
 101/* Set the expiration time of a quota's grace period. */
 102time64_t
 103xfs_dquot_set_timeout(
 104	struct xfs_mount	*mp,
 105	time64_t		timeout)
 106{
 107	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 108
 109	return clamp_t(time64_t, timeout, qi->qi_expiry_min,
 110					  qi->qi_expiry_max);
 111}
 112
 113/* Set the length of the default grace period. */
 114time64_t
 115xfs_dquot_set_grace_period(
 116	time64_t		grace)
 117{
 118	return clamp_t(time64_t, grace, XFS_DQ_GRACE_MIN, XFS_DQ_GRACE_MAX);
 119}
 120
 121/*
 122 * Determine if this quota counter is over either limit and set the quota
 123 * timers as appropriate.
 124 */
 125static inline void
 126xfs_qm_adjust_res_timer(
 127	struct xfs_mount	*mp,
 128	struct xfs_dquot_res	*res,
 129	struct xfs_quota_limits	*qlim)
 130{
 131	ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
 132
 133	if ((res->softlimit && res->count > res->softlimit) ||
 134	    (res->hardlimit && res->count > res->hardlimit)) {
 135		if (res->timer == 0)
 136			res->timer = xfs_dquot_set_timeout(mp,
 137					ktime_get_real_seconds() + qlim->time);
 138	} else {
 139		res->timer = 0;
 140	}
 141}
 142
 143/*
 144 * Check the limits and timers of a dquot and start or reset timers
 145 * if necessary.
 146 * This gets called even when quota enforcement is OFF, which makes our
 147 * life a little less complicated. (We just don't reject any quota
 148 * reservations in that case, when enforcement is off).
 149 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
 150 * enforcement's off.
 151 * In contrast, warnings are a little different in that they don't
 152 * 'automatically' get started when limits get exceeded.  They do
 153 * get reset to zero, however, when we find the count to be under
 154 * the soft limit (they are only ever set non-zero via userspace).
 155 */
 156void
 157xfs_qm_adjust_dqtimers(
 158	struct xfs_dquot	*dq)
 
 159{
 160	struct xfs_mount	*mp = dq->q_mount;
 161	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 162	struct xfs_def_quota	*defq;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 163
 164	ASSERT(dq->q_id);
 165	defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 166
 167	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
 168	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
 169	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 170}
 171
 172/*
 173 * initialize a buffer full of dquots and log the whole thing
 174 */
 175STATIC void
 176xfs_qm_init_dquot_blk(
 177	struct xfs_trans	*tp,
 178	struct xfs_mount	*mp,
 179	xfs_dqid_t		id,
 180	xfs_dqtype_t		type,
 181	struct xfs_buf		*bp)
 182{
 183	struct xfs_quotainfo	*q = mp->m_quotainfo;
 184	struct xfs_dqblk	*d;
 185	xfs_dqid_t		curid;
 186	unsigned int		qflag;
 187	unsigned int		blftype;
 188	int			i;
 189
 190	ASSERT(tp);
 191	ASSERT(xfs_buf_islocked(bp));
 192
 193	switch (type) {
 194	case XFS_DQTYPE_USER:
 195		qflag = XFS_UQUOTA_CHKD;
 196		blftype = XFS_BLF_UDQUOT_BUF;
 197		break;
 198	case XFS_DQTYPE_PROJ:
 199		qflag = XFS_PQUOTA_CHKD;
 200		blftype = XFS_BLF_PDQUOT_BUF;
 201		break;
 202	case XFS_DQTYPE_GROUP:
 203		qflag = XFS_GQUOTA_CHKD;
 204		blftype = XFS_BLF_GDQUOT_BUF;
 205		break;
 206	default:
 207		ASSERT(0);
 208		return;
 209	}
 210
 211	d = bp->b_addr;
 212
 213	/*
 214	 * ID of the first dquot in the block - id's are zero based.
 215	 */
 216	curid = id - (id % q->qi_dqperchunk);
 217	memset(d, 0, BBTOB(q->qi_dqchunklen));
 218	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
 219		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 220		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
 221		d->dd_diskdq.d_id = cpu_to_be32(curid);
 222		d->dd_diskdq.d_type = type;
 223		if (curid > 0 && xfs_has_bigtime(mp))
 224			d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
 225		if (xfs_has_crc(mp)) {
 226			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
 227			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
 228					 XFS_DQUOT_CRC_OFF);
 229		}
 230	}
 231
 232	xfs_trans_dquot_buf(tp, bp, blftype);
 233
 234	/*
 235	 * quotacheck uses delayed writes to update all the dquots on disk in an
 236	 * efficient manner instead of logging the individual dquot changes as
 237	 * they are made. However if we log the buffer allocated here and crash
 238	 * after quotacheck while the logged initialisation is still in the
 239	 * active region of the log, log recovery can replay the dquot buffer
 240	 * initialisation over the top of the checked dquots and corrupt quota
 241	 * accounting.
 242	 *
 243	 * To avoid this problem, quotacheck cannot log the initialised buffer.
 244	 * We must still dirty the buffer and write it back before the
 245	 * allocation transaction clears the log. Therefore, mark the buffer as
 246	 * ordered instead of logging it directly. This is safe for quotacheck
 247	 * because it detects and repairs allocated but initialized dquot blocks
 248	 * in the quota inodes.
 249	 */
 250	if (!(mp->m_qflags & qflag))
 251		xfs_trans_ordered_buf(tp, bp);
 252	else
 253		xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
 254}
 255
 256/*
 257 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
 258 * watermarks correspond to the soft and hard limits by default. If a soft limit
 259 * is not specified, we use 95% of the hard limit.
 260 */
 261void
 262xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
 263{
 264	uint64_t space;
 265
 266	dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
 267	dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
 268	if (!dqp->q_prealloc_lo_wmark) {
 269		dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
 270		do_div(dqp->q_prealloc_lo_wmark, 100);
 271		dqp->q_prealloc_lo_wmark *= 95;
 272	}
 273
 274	space = dqp->q_prealloc_hi_wmark;
 275
 276	do_div(space, 100);
 277	dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
 278	dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
 279	dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
 280}
 281
 282/*
 283 * Ensure that the given in-core dquot has a buffer on disk backing it, and
 284 * return the buffer locked and held. This is called when the bmapi finds a
 285 * hole.
 286 */
 287STATIC int
 288xfs_dquot_disk_alloc(
 289	struct xfs_dquot	*dqp,
 290	struct xfs_buf		**bpp)
 291{
 292	struct xfs_bmbt_irec	map;
 293	struct xfs_trans	*tp;
 294	struct xfs_mount	*mp = dqp->q_mount;
 295	struct xfs_buf		*bp;
 296	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
 297	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
 298	int			nmaps = 1;
 299	int			error;
 
 
 
 
 300
 301	trace_xfs_dqalloc(dqp);
 302
 303	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
 304			XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
 305	if (error)
 306		return error;
 307
 308	xfs_ilock(quotip, XFS_ILOCK_EXCL);
 309	xfs_trans_ijoin(tp, quotip, 0);
 310
 311	if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
 312		/*
 313		 * Return if this type of quotas is turned off while we didn't
 314		 * have an inode lock
 315		 */
 316		error = -ESRCH;
 317		goto err_cancel;
 318	}
 319
 320	error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
 321			XFS_IEXT_ADD_NOSPLIT_CNT);
 322	if (error == -EFBIG)
 323		error = xfs_iext_count_upgrade(tp, quotip,
 324				XFS_IEXT_ADD_NOSPLIT_CNT);
 325	if (error)
 326		goto err_cancel;
 327
 328	/* Create the block mapping. */
 329	error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
 330			XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
 331			&nmaps);
 332	if (error)
 333		goto err_cancel;
 334
 335	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
 336	ASSERT(nmaps == 1);
 337	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 338	       (map.br_startblock != HOLESTARTBLOCK));
 339
 340	/*
 341	 * Keep track of the blkno to save a lookup later
 342	 */
 343	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 344
 345	/* now we can just get the buffer (there's nothing to read yet) */
 346	error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
 347			mp->m_quotainfo->qi_dqchunklen, 0, &bp);
 348	if (error)
 349		goto err_cancel;
 
 
 
 
 350	bp->b_ops = &xfs_dquot_buf_ops;
 351
 352	/*
 353	 * Make a chunk of dquots out of this buffer and log
 354	 * the entire thing.
 355	 */
 356	xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
 357	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 358
 359	/*
 360	 * Hold the buffer and join it to the dfops so that we'll still own
 361	 * the buffer when we return to the caller.  The buffer disposal on
 362	 * error must be paid attention to very carefully, as it has been
 363	 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
 364	 * code when allocating a new dquot record" in 2005, and the later
 365	 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
 366	 * the buffer locked across the _defer_finish call.  We can now do
 367	 * this correctly with xfs_defer_bjoin.
 368	 *
 369	 * Above, we allocated a disk block for the dquot information and used
 370	 * get_buf to initialize the dquot. If the _defer_finish fails, the old
 371	 * transaction is gone but the new buffer is not joined or held to any
 372	 * transaction, so we must _buf_relse it.
 373	 *
 374	 * If everything succeeds, the caller of this function is returned a
 375	 * buffer that is locked and held to the transaction.  The caller
 376	 * is responsible for unlocking any buffer passed back, either
 377	 * manually or by committing the transaction.  On error, the buffer is
 378	 * released and not passed back.
 379	 *
 380	 * Keep the quota inode ILOCKed until after the transaction commit to
 381	 * maintain the atomicity of bmap/rmap updates.
 382	 */
 
 383	xfs_trans_bhold(tp, bp);
 384	error = xfs_trans_commit(tp);
 385	xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 386	if (error) {
 387		xfs_buf_relse(bp);
 388		return error;
 
 
 
 
 
 
 389	}
 390
 391	*bpp = bp;
 392	return 0;
 393
 394err_cancel:
 395	xfs_trans_cancel(tp);
 396	xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 397	return error;
 398}
 399
 400/*
 401 * Read in the in-core dquot's on-disk metadata and return the buffer.
 402 * Returns ENOENT to signal a hole.
 
 403 */
 404STATIC int
 405xfs_dquot_disk_read(
 406	struct xfs_mount	*mp,
 407	struct xfs_dquot	*dqp,
 408	struct xfs_buf		**bpp)
 
 
 409{
 410	struct xfs_bmbt_irec	map;
 
 411	struct xfs_buf		*bp;
 412	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
 413	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
 
 
 414	uint			lock_mode;
 415	int			nmaps = 1;
 416	int			error;
 
 417
 418	lock_mode = xfs_ilock_data_map_shared(quotip);
 419	if (!xfs_this_quota_on(mp, qtype)) {
 420		/*
 421		 * Return if this type of quotas is turned off while we
 422		 * didn't have the quota inode lock.
 423		 */
 424		xfs_iunlock(quotip, lock_mode);
 425		return -ESRCH;
 426	}
 427
 428	/*
 429	 * Find the block map; no allocations yet
 430	 */
 431	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
 432			XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
 
 433	xfs_iunlock(quotip, lock_mode);
 434	if (error)
 435		return error;
 436
 437	ASSERT(nmaps == 1);
 438	ASSERT(map.br_blockcount >= 1);
 
 
 
 
 
 
 
 439	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
 440	if (map.br_startblock == HOLESTARTBLOCK)
 441		return -ENOENT;
 
 
 
 
 442
 443	trace_xfs_dqtobp_read(dqp);
 
 
 
 
 
 
 
 444
 445	/*
 446	 * store the blkno etc so that we don't have to do the
 447	 * mapping all the time
 448	 */
 449	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 450
 451	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
 452			mp->m_quotainfo->qi_dqchunklen, 0, &bp,
 453			&xfs_dquot_buf_ops);
 454	if (error) {
 455		ASSERT(bp == NULL);
 456		return error;
 
 
 457	}
 458
 459	ASSERT(xfs_buf_islocked(bp));
 460	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 461	*bpp = bp;
 462
 463	return 0;
 464}
 465
 466/* Allocate and initialize everything we need for an incore dquot. */
 467STATIC struct xfs_dquot *
 468xfs_dquot_alloc(
 
 
 
 
 
 
 469	struct xfs_mount	*mp,
 470	xfs_dqid_t		id,
 471	xfs_dqtype_t		type)
 
 
 472{
 473	struct xfs_dquot	*dqp;
 
 
 
 
 474
 475	dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
 476
 477	dqp->q_type = type;
 478	dqp->q_id = id;
 479	dqp->q_mount = mp;
 480	INIT_LIST_HEAD(&dqp->q_lru);
 481	mutex_init(&dqp->q_qlock);
 482	init_waitqueue_head(&dqp->q_pinwait);
 483	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
 484	/*
 485	 * Offset of dquot in the (fixed sized) dquot chunk.
 486	 */
 487	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
 488			sizeof(struct xfs_dqblk);
 489
 490	/*
 491	 * Because we want to use a counting completion, complete
 492	 * the flush completion once to allow a single access to
 493	 * the flush completion without blocking.
 494	 */
 495	init_completion(&dqp->q_flush);
 496	complete(&dqp->q_flush);
 497
 498	/*
 499	 * Make sure group quotas have a different lock class than user
 500	 * quotas.
 501	 */
 502	switch (type) {
 503	case XFS_DQTYPE_USER:
 504		/* uses the default lock class */
 505		break;
 506	case XFS_DQTYPE_GROUP:
 507		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
 508		break;
 509	case XFS_DQTYPE_PROJ:
 510		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
 511		break;
 512	default:
 513		ASSERT(0);
 514		break;
 515	}
 516
 517	xfs_qm_dquot_logitem_init(dqp);
 518
 519	XFS_STATS_INC(mp, xs_qm_dquot);
 520	return dqp;
 521}
 522
 523/* Check the ondisk dquot's id and type match what the incore dquot expects. */
 524static bool
 525xfs_dquot_check_type(
 526	struct xfs_dquot	*dqp,
 527	struct xfs_disk_dquot	*ddqp)
 528{
 529	uint8_t			ddqp_type;
 530	uint8_t			dqp_type;
 531
 532	ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
 533	dqp_type = xfs_dquot_type(dqp);
 534
 535	if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
 536		return false;
 
 537
 538	/*
 539	 * V5 filesystems always expect an exact type match.  V4 filesystems
 540	 * expect an exact match for user dquots and for non-root group and
 541	 * project dquots.
 542	 */
 543	if (xfs_has_crc(dqp->q_mount) ||
 544	    dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
 545		return ddqp_type == dqp_type;
 546
 547	/*
 548	 * V4 filesystems support either group or project quotas, but not both
 549	 * at the same time.  The non-user quota file can be switched between
 550	 * group and project quota uses depending on the mount options, which
 551	 * means that we can encounter the other type when we try to load quota
 552	 * defaults.  Quotacheck will soon reset the entire quota file
 553	 * (including the root dquot) anyway, but don't log scary corruption
 554	 * reports to dmesg.
 555	 */
 556	return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
 557}
 558
 559/* Copy the in-core quota fields in from the on-disk buffer. */
 560STATIC int
 561xfs_dquot_from_disk(
 562	struct xfs_dquot	*dqp,
 563	struct xfs_buf		*bp)
 564{
 565	struct xfs_disk_dquot	*ddqp = bp->b_addr + dqp->q_bufoffset;
 566
 567	/*
 568	 * Ensure that we got the type and ID we were looking for.
 569	 * Everything else was checked by the dquot buffer verifier.
 570	 */
 571	if (!xfs_dquot_check_type(dqp, ddqp)) {
 572		xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
 573			  "Metadata corruption detected at %pS, quota %u",
 574			  __this_address, dqp->q_id);
 575		xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
 576		return -EFSCORRUPTED;
 577	}
 578
 579	/* copy everything from disk dquot to the incore dquot */
 580	dqp->q_type = ddqp->d_type;
 581	dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 582	dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 583	dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 584	dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 585	dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 586	dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 587
 588	dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
 589	dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
 590	dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
 591
 592	dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
 593	dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
 594	dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
 595
 596	/*
 597	 * Reservation counters are defined as reservation plus current usage
 598	 * to avoid having to add every time.
 599	 */
 600	dqp->q_blk.reserved = dqp->q_blk.count;
 601	dqp->q_ino.reserved = dqp->q_ino.count;
 602	dqp->q_rtb.reserved = dqp->q_rtb.count;
 603
 604	/* initialize the dquot speculative prealloc thresholds */
 605	xfs_dquot_set_prealloc_limits(dqp);
 606	return 0;
 607}
 608
 609/* Copy the in-core quota fields into the on-disk buffer. */
 610void
 611xfs_dquot_to_disk(
 612	struct xfs_disk_dquot	*ddqp,
 613	struct xfs_dquot	*dqp)
 614{
 615	ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 616	ddqp->d_version = XFS_DQUOT_VERSION;
 617	ddqp->d_type = dqp->q_type;
 618	ddqp->d_id = cpu_to_be32(dqp->q_id);
 619	ddqp->d_pad0 = 0;
 620	ddqp->d_pad = 0;
 621
 622	ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
 623	ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
 624	ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
 625	ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
 626	ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
 627	ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
 628
 629	ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
 630	ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
 631	ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
 632
 633	ddqp->d_bwarns = 0;
 634	ddqp->d_iwarns = 0;
 635	ddqp->d_rtbwarns = 0;
 636
 637	ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
 638	ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
 639	ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
 640}
 641
 642/*
 643 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
 644 * and release the buffer immediately.  If @can_alloc is true, fill any
 645 * holes in the on-disk metadata.
 646 */
 647static int
 648xfs_qm_dqread(
 649	struct xfs_mount	*mp,
 650	xfs_dqid_t		id,
 651	xfs_dqtype_t		type,
 652	bool			can_alloc,
 653	struct xfs_dquot	**dqpp)
 654{
 655	struct xfs_dquot	*dqp;
 656	struct xfs_buf		*bp;
 657	int			error;
 658
 659	dqp = xfs_dquot_alloc(mp, id, type);
 660	trace_xfs_dqread(dqp);
 661
 662	/* Try to read the buffer, allocating if necessary. */
 663	error = xfs_dquot_disk_read(mp, dqp, &bp);
 664	if (error == -ENOENT && can_alloc)
 665		error = xfs_dquot_disk_alloc(dqp, &bp);
 666	if (error)
 667		goto err;
 668
 669	/*
 670	 * At this point we should have a clean locked buffer.  Copy the data
 671	 * to the incore dquot and release the buffer since the incore dquot
 672	 * has its own locking protocol so we needn't tie up the buffer any
 673	 * further.
 
 
 
 
 
 
 674	 */
 675	ASSERT(xfs_buf_islocked(bp));
 676	error = xfs_dquot_from_disk(dqp, bp);
 677	xfs_buf_relse(bp);
 678	if (error)
 679		goto err;
 
 
 
 680
 681	*dqpp = dqp;
 682	return error;
 683
 684err:
 685	trace_xfs_dqread_fail(dqp);
 
 
 686	xfs_qm_dqdestroy(dqp);
 687	*dqpp = NULL;
 688	return error;
 689}
 690
 691/*
 692 * Advance to the next id in the current chunk, or if at the
 693 * end of the chunk, skip ahead to first id in next allocated chunk
 694 * using the SEEK_DATA interface.
 695 */
 696static int
 697xfs_dq_get_next_id(
 698	struct xfs_mount	*mp,
 699	xfs_dqtype_t		type,
 700	xfs_dqid_t		*id)
 701{
 702	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
 703	xfs_dqid_t		next_id = *id + 1; /* simple advance */
 704	uint			lock_flags;
 705	struct xfs_bmbt_irec	got;
 706	struct xfs_iext_cursor	cur;
 707	xfs_fsblock_t		start;
 708	int			error = 0;
 709
 710	/* If we'd wrap past the max ID, stop */
 711	if (next_id < *id)
 712		return -ENOENT;
 713
 714	/* If new ID is within the current chunk, advancing it sufficed */
 715	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
 716		*id = next_id;
 717		return 0;
 718	}
 719
 720	/* Nope, next_id is now past the current chunk, so find the next one */
 721	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
 722
 723	lock_flags = xfs_ilock_data_map_shared(quotip);
 724	error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
 725	if (error)
 726		return error;
 
 
 727
 728	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
 729		/* contiguous chunk, bump startoff for the id calculation */
 730		if (got.br_startoff < start)
 731			got.br_startoff = start;
 732		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
 733	} else {
 734		error = -ENOENT;
 735	}
 736
 737	xfs_iunlock(quotip, lock_flags);
 738
 739	return error;
 740}
 741
 742/*
 743 * Look up the dquot in the in-core cache.  If found, the dquot is returned
 744 * locked and ready to go.
 745 */
 746static struct xfs_dquot *
 747xfs_qm_dqget_cache_lookup(
 748	struct xfs_mount	*mp,
 749	struct xfs_quotainfo	*qi,
 750	struct radix_tree_root	*tree,
 751	xfs_dqid_t		id)
 752{
 753	struct xfs_dquot	*dqp;
 754
 755restart:
 756	mutex_lock(&qi->qi_tree_lock);
 757	dqp = radix_tree_lookup(tree, id);
 758	if (!dqp) {
 759		mutex_unlock(&qi->qi_tree_lock);
 760		XFS_STATS_INC(mp, xs_qm_dqcachemisses);
 761		return NULL;
 762	}
 763
 764	xfs_dqlock(dqp);
 765	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
 766		xfs_dqunlock(dqp);
 767		mutex_unlock(&qi->qi_tree_lock);
 768		trace_xfs_dqget_freeing(dqp);
 769		delay(1);
 770		goto restart;
 771	}
 772
 773	dqp->q_nrefs++;
 774	mutex_unlock(&qi->qi_tree_lock);
 775
 776	trace_xfs_dqget_hit(dqp);
 777	XFS_STATS_INC(mp, xs_qm_dqcachehits);
 778	return dqp;
 779}
 780
 781/*
 782 * Try to insert a new dquot into the in-core cache.  If an error occurs the
 783 * caller should throw away the dquot and start over.  Otherwise, the dquot
 784 * is returned locked (and held by the cache) as if there had been a cache
 785 * hit.
 786 */
 787static int
 788xfs_qm_dqget_cache_insert(
 789	struct xfs_mount	*mp,
 790	struct xfs_quotainfo	*qi,
 791	struct radix_tree_root	*tree,
 792	xfs_dqid_t		id,
 793	struct xfs_dquot	*dqp)
 794{
 795	int			error;
 796
 797	mutex_lock(&qi->qi_tree_lock);
 798	error = radix_tree_insert(tree, id, dqp);
 799	if (unlikely(error)) {
 800		/* Duplicate found!  Caller must try again. */
 801		WARN_ON(error != -EEXIST);
 802		mutex_unlock(&qi->qi_tree_lock);
 803		trace_xfs_dqget_dup(dqp);
 804		return error;
 805	}
 806
 807	/* Return a locked dquot to the caller, with a reference taken. */
 808	xfs_dqlock(dqp);
 809	dqp->q_nrefs = 1;
 810
 811	qi->qi_dquots++;
 812	mutex_unlock(&qi->qi_tree_lock);
 813
 814	return 0;
 815}
 816
 817/* Check our input parameters. */
 818static int
 819xfs_qm_dqget_checks(
 820	struct xfs_mount	*mp,
 821	xfs_dqtype_t		type)
 822{
 823	switch (type) {
 824	case XFS_DQTYPE_USER:
 825		if (!XFS_IS_UQUOTA_ON(mp))
 826			return -ESRCH;
 827		return 0;
 828	case XFS_DQTYPE_GROUP:
 829		if (!XFS_IS_GQUOTA_ON(mp))
 830			return -ESRCH;
 831		return 0;
 832	case XFS_DQTYPE_PROJ:
 833		if (!XFS_IS_PQUOTA_ON(mp))
 834			return -ESRCH;
 835		return 0;
 836	default:
 837		WARN_ON_ONCE(0);
 838		return -EINVAL;
 839	}
 840}
 841
 842/*
 843 * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
 844 * locked dquot, doing an allocation (if requested) as needed.
 845 */
 846int
 847xfs_qm_dqget(
 848	struct xfs_mount	*mp,
 849	xfs_dqid_t		id,
 850	xfs_dqtype_t		type,
 851	bool			can_alloc,
 852	struct xfs_dquot	**O_dqpp)
 
 853{
 854	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 855	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 856	struct xfs_dquot	*dqp;
 857	int			error;
 858
 859	error = xfs_qm_dqget_checks(mp, type);
 860	if (error)
 861		return error;
 862
 863restart:
 864	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
 865	if (dqp) {
 866		*O_dqpp = dqp;
 867		return 0;
 868	}
 869
 870	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
 871	if (error)
 872		return error;
 873
 874	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
 875	if (error) {
 876		/*
 877		 * Duplicate found. Just throw away the new dquot and start
 878		 * over.
 879		 */
 880		xfs_qm_dqdestroy(dqp);
 881		XFS_STATS_INC(mp, xs_qm_dquot_dups);
 882		goto restart;
 883	}
 884
 885	trace_xfs_dqget_miss(dqp);
 886	*O_dqpp = dqp;
 887	return 0;
 888}
 
 
 
 
 
 
 
 
 889
 890/*
 891 * Given a dquot id and type, read and initialize a dquot from the on-disk
 892 * metadata.  This function is only for use during quota initialization so
 893 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
 894 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
 895 */
 896int
 897xfs_qm_dqget_uncached(
 898	struct xfs_mount	*mp,
 899	xfs_dqid_t		id,
 900	xfs_dqtype_t		type,
 901	struct xfs_dquot	**dqpp)
 902{
 903	int			error;
 904
 905	error = xfs_qm_dqget_checks(mp, type);
 906	if (error)
 907		return error;
 908
 909	return xfs_qm_dqread(mp, id, type, 0, dqpp);
 910}
 911
 912/* Return the quota id for a given inode and type. */
 913xfs_dqid_t
 914xfs_qm_id_for_quotatype(
 915	struct xfs_inode	*ip,
 916	xfs_dqtype_t		type)
 917{
 918	switch (type) {
 919	case XFS_DQTYPE_USER:
 920		return i_uid_read(VFS_I(ip));
 921	case XFS_DQTYPE_GROUP:
 922		return i_gid_read(VFS_I(ip));
 923	case XFS_DQTYPE_PROJ:
 924		return ip->i_projid;
 925	}
 926	ASSERT(0);
 927	return 0;
 928}
 929
 930/*
 931 * Return the dquot for a given inode and type.  If @can_alloc is true, then
 932 * allocate blocks if needed.  The inode's ILOCK must be held and it must not
 933 * have already had an inode attached.
 934 */
 935int
 936xfs_qm_dqget_inode(
 937	struct xfs_inode	*ip,
 938	xfs_dqtype_t		type,
 939	bool			can_alloc,
 940	struct xfs_dquot	**O_dqpp)
 941{
 942	struct xfs_mount	*mp = ip->i_mount;
 943	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 944	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 945	struct xfs_dquot	*dqp;
 946	xfs_dqid_t		id;
 947	int			error;
 948
 949	error = xfs_qm_dqget_checks(mp, type);
 950	if (error)
 951		return error;
 952
 953	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 954	ASSERT(xfs_inode_dquot(ip, type) == NULL);
 955
 956	id = xfs_qm_id_for_quotatype(ip, type);
 957
 958restart:
 959	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
 960	if (dqp) {
 961		*O_dqpp = dqp;
 962		return 0;
 963	}
 
 
 964
 965	/*
 966	 * Dquot cache miss. We don't want to keep the inode lock across
 967	 * a (potential) disk read. Also we don't want to deal with the lock
 968	 * ordering between quotainode and this inode. OTOH, dropping the inode
 969	 * lock here means dealing with a chown that can happen before
 970	 * we re-acquire the lock.
 971	 */
 972	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 973	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
 974	xfs_ilock(ip, XFS_ILOCK_EXCL);
 
 
 
 
 
 
 
 
 
 
 
 
 975	if (error)
 976		return error;
 977
 978	/*
 979	 * A dquot could be attached to this inode by now, since we had
 980	 * dropped the ilock.
 981	 */
 982	if (xfs_this_quota_on(mp, type)) {
 983		struct xfs_dquot	*dqp1;
 
 984
 985		dqp1 = xfs_inode_dquot(ip, type);
 986		if (dqp1) {
 
 
 
 
 
 
 
 987			xfs_qm_dqdestroy(dqp);
 988			dqp = dqp1;
 989			xfs_dqlock(dqp);
 990			goto dqret;
 991		}
 992	} else {
 993		/* inode stays locked on return */
 994		xfs_qm_dqdestroy(dqp);
 995		return -ESRCH;
 996	}
 997
 998	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
 999	if (error) {
 
 
 
1000		/*
1001		 * Duplicate found. Just throw away the new dquot and start
1002		 * over.
1003		 */
 
 
1004		xfs_qm_dqdestroy(dqp);
1005		XFS_STATS_INC(mp, xs_qm_dquot_dups);
1006		goto restart;
1007	}
1008
1009dqret:
1010	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1011	trace_xfs_dqget_miss(dqp);
1012	*O_dqpp = dqp;
1013	return 0;
1014}
1015
1016/*
1017 * Starting at @id and progressing upwards, look for an initialized incore
1018 * dquot, lock it, and return it.
1019 */
1020int
1021xfs_qm_dqget_next(
1022	struct xfs_mount	*mp,
1023	xfs_dqid_t		id,
1024	xfs_dqtype_t		type,
1025	struct xfs_dquot	**dqpp)
1026{
1027	struct xfs_dquot	*dqp;
1028	int			error = 0;
1029
1030	*dqpp = NULL;
1031	for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1032		error = xfs_qm_dqget(mp, id, type, false, &dqp);
1033		if (error == -ENOENT)
1034			continue;
1035		else if (error != 0)
1036			break;
1037
1038		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1039			*dqpp = dqp;
1040			return 0;
1041		}
1042
1043		xfs_qm_dqput(dqp);
1044	}
1045
1046	return error;
 
 
 
 
1047}
1048
1049/*
1050 * Release a reference to the dquot (decrement ref-count) and unlock it.
1051 *
1052 * If there is a group quota attached to this dquot, carefully release that
1053 * too without tripping over deadlocks'n'stuff.
1054 */
1055void
1056xfs_qm_dqput(
1057	struct xfs_dquot	*dqp)
1058{
1059	ASSERT(dqp->q_nrefs > 0);
1060	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1061
1062	trace_xfs_dqput(dqp);
1063
1064	if (--dqp->q_nrefs == 0) {
1065		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
1066		trace_xfs_dqput_free(dqp);
1067
1068		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1069			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1070	}
1071	xfs_dqunlock(dqp);
1072}
1073
1074/*
1075 * Release a dquot. Flush it if dirty, then dqput() it.
1076 * dquot must not be locked.
1077 */
1078void
1079xfs_qm_dqrele(
1080	struct xfs_dquot	*dqp)
1081{
1082	if (!dqp)
1083		return;
1084
1085	trace_xfs_dqrele(dqp);
1086
1087	xfs_dqlock(dqp);
1088	/*
1089	 * We don't care to flush it if the dquot is dirty here.
1090	 * That will create stutters that we want to avoid.
1091	 * Instead we do a delayed write when we try to reclaim
1092	 * a dirty dquot. Also xfs_sync will take part of the burden...
1093	 */
1094	xfs_qm_dqput(dqp);
1095}
1096
1097/*
1098 * This is the dquot flushing I/O completion routine.  It is called
1099 * from interrupt level when the buffer containing the dquot is
1100 * flushed to disk.  It is responsible for removing the dquot logitem
1101 * from the AIL if it has not been re-logged, and unlocking the dquot's
1102 * flush lock. This behavior is very similar to that of inodes..
1103 */
1104static void
1105xfs_qm_dqflush_done(
 
1106	struct xfs_log_item	*lip)
1107{
1108	struct xfs_dq_logitem	*qip = (struct xfs_dq_logitem *)lip;
1109	struct xfs_dquot	*dqp = qip->qli_dquot;
1110	struct xfs_ail		*ailp = lip->li_ailp;
1111	xfs_lsn_t		tail_lsn;
1112
1113	/*
1114	 * We only want to pull the item from the AIL if its
1115	 * location in the log has not changed since we started the flush.
1116	 * Thus, we only bother if the dquot's lsn has
1117	 * not changed. First we check the lsn outside the lock
1118	 * since it's cheaper, and then we recheck while
1119	 * holding the lock before removing the dquot from the AIL.
1120	 */
1121	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1122	    ((lip->li_lsn == qip->qli_flush_lsn) ||
1123	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1124
 
1125		spin_lock(&ailp->ail_lock);
1126		xfs_clear_li_failed(lip);
1127		if (lip->li_lsn == qip->qli_flush_lsn) {
1128			/* xfs_ail_update_finish() drops the AIL lock */
1129			tail_lsn = xfs_ail_delete_one(ailp, lip);
1130			xfs_ail_update_finish(ailp, tail_lsn);
1131		} else {
 
 
 
 
 
 
1132			spin_unlock(&ailp->ail_lock);
1133		}
1134	}
1135
1136	/*
1137	 * Release the dq's flush lock since we're done with it.
1138	 */
1139	xfs_dqfunlock(dqp);
1140}
1141
1142void
1143xfs_buf_dquot_iodone(
1144	struct xfs_buf		*bp)
1145{
1146	struct xfs_log_item	*lip, *n;
1147
1148	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1149		list_del_init(&lip->li_bio_list);
1150		xfs_qm_dqflush_done(lip);
1151	}
1152}
1153
1154void
1155xfs_buf_dquot_io_fail(
1156	struct xfs_buf		*bp)
1157{
1158	struct xfs_log_item	*lip;
1159
1160	spin_lock(&bp->b_mount->m_ail->ail_lock);
1161	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1162		xfs_set_li_failed(lip, bp);
1163	spin_unlock(&bp->b_mount->m_ail->ail_lock);
1164}
1165
1166/* Check incore dquot for errors before we flush. */
1167static xfs_failaddr_t
1168xfs_qm_dqflush_check(
1169	struct xfs_dquot	*dqp)
1170{
1171	xfs_dqtype_t		type = xfs_dquot_type(dqp);
1172
1173	if (type != XFS_DQTYPE_USER &&
1174	    type != XFS_DQTYPE_GROUP &&
1175	    type != XFS_DQTYPE_PROJ)
1176		return __this_address;
1177
1178	if (dqp->q_id == 0)
1179		return NULL;
1180
1181	if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1182	    !dqp->q_blk.timer)
1183		return __this_address;
1184
1185	if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1186	    !dqp->q_ino.timer)
1187		return __this_address;
1188
1189	if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1190	    !dqp->q_rtb.timer)
1191		return __this_address;
1192
1193	/* bigtime flag should never be set on root dquots */
1194	if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1195		if (!xfs_has_bigtime(dqp->q_mount))
1196			return __this_address;
1197		if (dqp->q_id == 0)
1198			return __this_address;
1199	}
1200
1201	return NULL;
1202}
1203
1204/*
1205 * Write a modified dquot to disk.
1206 * The dquot must be locked and the flush lock too taken by caller.
1207 * The flush lock will not be unlocked until the dquot reaches the disk,
1208 * but the dquot is free to be unlocked and modified by the caller
1209 * in the interim. Dquot is still locked on return. This behavior is
1210 * identical to that of inodes.
1211 */
1212int
1213xfs_qm_dqflush(
1214	struct xfs_dquot	*dqp,
1215	struct xfs_buf		**bpp)
1216{
1217	struct xfs_mount	*mp = dqp->q_mount;
1218	struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
1219	struct xfs_buf		*bp;
1220	struct xfs_dqblk	*dqblk;
1221	xfs_failaddr_t		fa;
1222	int			error;
1223
1224	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1225	ASSERT(!completion_done(&dqp->q_flush));
1226
1227	trace_xfs_dqflush(dqp);
1228
1229	*bpp = NULL;
1230
1231	xfs_qm_dqunpin_wait(dqp);
1232
1233	/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1234	 * Get the buffer containing the on-disk dquot
1235	 */
1236	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1237				   mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1238				   &bp, &xfs_dquot_buf_ops);
1239	if (error == -EAGAIN)
1240		goto out_unlock;
1241	if (error)
1242		goto out_abort;
1243
1244	fa = xfs_qm_dqflush_check(dqp);
 
 
 
 
 
 
 
 
1245	if (fa) {
1246		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1247				dqp->q_id, fa);
1248		xfs_buf_relse(bp);
1249		error = -EFSCORRUPTED;
1250		goto out_abort;
 
1251	}
1252
1253	/* Flush the incore dquot to the ondisk buffer. */
1254	dqblk = bp->b_addr + dqp->q_bufoffset;
1255	xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1256
1257	/*
1258	 * Clear the dirty field and remember the flush lsn for later use.
1259	 */
1260	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1261
1262	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1263					&dqp->q_logitem.qli_item.li_lsn);
1264
1265	/*
1266	 * copy the lsn into the on-disk dquot now while we have the in memory
1267	 * dquot here. This can't be done later in the write verifier as we
1268	 * can't get access to the log item at that point in time.
1269	 *
1270	 * We also calculate the CRC here so that the on-disk dquot in the
1271	 * buffer always has a valid CRC. This ensures there is no possibility
1272	 * of a dquot without an up-to-date CRC getting to disk.
1273	 */
1274	if (xfs_has_crc(mp)) {
1275		dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1276		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
 
 
1277				 XFS_DQUOT_CRC_OFF);
1278	}
1279
1280	/*
1281	 * Attach the dquot to the buffer so that we can remove this dquot from
1282	 * the AIL and release the flush lock once the dquot is synced to disk.
1283	 */
1284	bp->b_flags |= _XBF_DQUOTS;
1285	list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1286
1287	/*
1288	 * If the buffer is pinned then push on the log so we won't
1289	 * get stuck waiting in the write for too long.
1290	 */
1291	if (xfs_buf_ispinned(bp)) {
1292		trace_xfs_dqflush_force(dqp);
1293		xfs_log_force(mp, 0);
1294	}
1295
1296	trace_xfs_dqflush_done(dqp);
1297	*bpp = bp;
1298	return 0;
1299
1300out_abort:
1301	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1302	xfs_trans_ail_delete(lip, 0);
1303	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1304out_unlock:
1305	xfs_dqfunlock(dqp);
1306	return error;
1307}
1308
1309/*
1310 * Lock two xfs_dquot structures.
1311 *
1312 * To avoid deadlocks we always lock the quota structure with
1313 * the lowerd id first.
1314 */
1315void
1316xfs_dqlock2(
1317	struct xfs_dquot	*d1,
1318	struct xfs_dquot	*d2)
1319{
1320	if (d1 && d2) {
1321		ASSERT(d1 != d2);
1322		if (d1->q_id > d2->q_id) {
 
1323			mutex_lock(&d2->q_qlock);
1324			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1325		} else {
1326			mutex_lock(&d1->q_qlock);
1327			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1328		}
1329	} else if (d1) {
1330		mutex_lock(&d1->q_qlock);
1331	} else if (d2) {
1332		mutex_lock(&d2->q_qlock);
1333	}
1334}
1335
1336int __init
1337xfs_qm_init(void)
1338{
1339	xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1340					  sizeof(struct xfs_dquot),
1341					  0, 0, NULL);
1342	if (!xfs_dquot_cache)
1343		goto out;
1344
1345	xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1346					     sizeof(struct xfs_dquot_acct),
1347					     0, 0, NULL);
1348	if (!xfs_dqtrx_cache)
1349		goto out_free_dquot_cache;
1350
1351	return 0;
1352
1353out_free_dquot_cache:
1354	kmem_cache_destroy(xfs_dquot_cache);
1355out:
1356	return -ENOMEM;
1357}
1358
1359void
1360xfs_qm_exit(void)
1361{
1362	kmem_cache_destroy(xfs_dqtrx_cache);
1363	kmem_cache_destroy(xfs_dquot_cache);
1364}
1365
1366/*
1367 * Iterate every dquot of a particular type.  The caller must ensure that the
1368 * particular quota type is active.  iter_fn can return negative error codes,
1369 * or -ECANCELED to indicate that it wants to stop iterating.
1370 */
1371int
1372xfs_qm_dqiterate(
1373	struct xfs_mount	*mp,
1374	xfs_dqtype_t		type,
1375	xfs_qm_dqiterate_fn	iter_fn,
1376	void			*priv)
1377{
1378	struct xfs_dquot	*dq;
1379	xfs_dqid_t		id = 0;
1380	int			error;
1381
1382	do {
1383		error = xfs_qm_dqget_next(mp, id, type, &dq);
1384		if (error == -ENOENT)
1385			return 0;
1386		if (error)
1387			return error;
1388
1389		error = iter_fn(dq, type, priv);
1390		id = dq->q_id;
1391		xfs_qm_dqput(dq);
1392	} while (error == 0 && id != 0);
1393
1394	return error;
1395}