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_zone		*xfs_qm_dqtrxzone;
  42static struct kmem_zone		*xfs_qm_dqzone;
  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_qm_dqzone, 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		if (res->timer == 0)
 140			res->warnings = 0;
 141		else
 142			res->timer = 0;
 143	}
 144}
 145
 146/*
 147 * Check the limits and timers of a dquot and start or reset timers
 148 * if necessary.
 149 * This gets called even when quota enforcement is OFF, which makes our
 150 * life a little less complicated. (We just don't reject any quota
 151 * reservations in that case, when enforcement is off).
 152 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
 153 * enforcement's off.
 154 * In contrast, warnings are a little different in that they don't
 155 * 'automatically' get started when limits get exceeded.  They do
 156 * get reset to zero, however, when we find the count to be under
 157 * the soft limit (they are only ever set non-zero via userspace).
 158 */
 159void
 160xfs_qm_adjust_dqtimers(
 161	struct xfs_dquot	*dq)
 162{
 163	struct xfs_mount	*mp = dq->q_mount;
 164	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 165	struct xfs_def_quota	*defq;
 166
 167	ASSERT(dq->q_id);
 168	defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
 169
 170	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
 171	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
 172	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
 173}
 174
 175/*
 176 * initialize a buffer full of dquots and log the whole thing
 177 */
 178STATIC void
 179xfs_qm_init_dquot_blk(
 180	struct xfs_trans	*tp,
 181	struct xfs_mount	*mp,
 182	xfs_dqid_t		id,
 183	xfs_dqtype_t		type,
 184	struct xfs_buf		*bp)
 185{
 186	struct xfs_quotainfo	*q = mp->m_quotainfo;
 187	struct xfs_dqblk	*d;
 188	xfs_dqid_t		curid;
 189	unsigned int		qflag;
 190	unsigned int		blftype;
 191	int			i;
 192
 193	ASSERT(tp);
 194	ASSERT(xfs_buf_islocked(bp));
 195
 196	switch (type) {
 197	case XFS_DQTYPE_USER:
 198		qflag = XFS_UQUOTA_CHKD;
 199		blftype = XFS_BLF_UDQUOT_BUF;
 200		break;
 201	case XFS_DQTYPE_PROJ:
 202		qflag = XFS_PQUOTA_CHKD;
 203		blftype = XFS_BLF_PDQUOT_BUF;
 204		break;
 205	case XFS_DQTYPE_GROUP:
 206		qflag = XFS_GQUOTA_CHKD;
 207		blftype = XFS_BLF_GDQUOT_BUF;
 208		break;
 209	default:
 210		ASSERT(0);
 211		return;
 212	}
 213
 214	d = bp->b_addr;
 215
 216	/*
 217	 * ID of the first dquot in the block - id's are zero based.
 218	 */
 219	curid = id - (id % q->qi_dqperchunk);
 220	memset(d, 0, BBTOB(q->qi_dqchunklen));
 221	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
 222		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 223		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
 224		d->dd_diskdq.d_id = cpu_to_be32(curid);
 225		d->dd_diskdq.d_type = type;
 226		if (curid > 0 && xfs_sb_version_hasbigtime(&mp->m_sb))
 227			d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
 228		if (xfs_sb_version_hascrc(&mp->m_sb)) {
 229			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
 230			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
 231					 XFS_DQUOT_CRC_OFF);
 232		}
 233	}
 234
 235	xfs_trans_dquot_buf(tp, bp, blftype);
 236
 237	/*
 238	 * quotacheck uses delayed writes to update all the dquots on disk in an
 239	 * efficient manner instead of logging the individual dquot changes as
 240	 * they are made. However if we log the buffer allocated here and crash
 241	 * after quotacheck while the logged initialisation is still in the
 242	 * active region of the log, log recovery can replay the dquot buffer
 243	 * initialisation over the top of the checked dquots and corrupt quota
 244	 * accounting.
 245	 *
 246	 * To avoid this problem, quotacheck cannot log the initialised buffer.
 247	 * We must still dirty the buffer and write it back before the
 248	 * allocation transaction clears the log. Therefore, mark the buffer as
 249	 * ordered instead of logging it directly. This is safe for quotacheck
 250	 * because it detects and repairs allocated but initialized dquot blocks
 251	 * in the quota inodes.
 252	 */
 253	if (!(mp->m_qflags & qflag))
 254		xfs_trans_ordered_buf(tp, bp);
 255	else
 256		xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
 257}
 258
 259/*
 260 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
 261 * watermarks correspond to the soft and hard limits by default. If a soft limit
 262 * is not specified, we use 95% of the hard limit.
 263 */
 264void
 265xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
 266{
 267	uint64_t space;
 268
 269	dqp->q_prealloc_hi_wmark = dqp->q_blk.hardlimit;
 270	dqp->q_prealloc_lo_wmark = dqp->q_blk.softlimit;
 271	if (!dqp->q_prealloc_lo_wmark) {
 272		dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
 273		do_div(dqp->q_prealloc_lo_wmark, 100);
 274		dqp->q_prealloc_lo_wmark *= 95;
 275	}
 276
 277	space = dqp->q_prealloc_hi_wmark;
 278
 279	do_div(space, 100);
 280	dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
 281	dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
 282	dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
 283}
 284
 285/*
 286 * Ensure that the given in-core dquot has a buffer on disk backing it, and
 287 * return the buffer locked and held. This is called when the bmapi finds a
 288 * hole.
 289 */
 290STATIC int
 291xfs_dquot_disk_alloc(
 292	struct xfs_trans	**tpp,
 293	struct xfs_dquot	*dqp,
 294	struct xfs_buf		**bpp)
 295{
 296	struct xfs_bmbt_irec	map;
 297	struct xfs_trans	*tp = *tpp;
 298	struct xfs_mount	*mp = tp->t_mountp;
 299	struct xfs_buf		*bp;
 300	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
 301	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
 302	int			nmaps = 1;
 303	int			error;
 304
 305	trace_xfs_dqalloc(dqp);
 306
 307	xfs_ilock(quotip, XFS_ILOCK_EXCL);
 308	if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
 309		/*
 310		 * Return if this type of quotas is turned off while we didn't
 311		 * have an inode lock
 312		 */
 313		xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 314		return -ESRCH;
 315	}
 316
 317	xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
 318
 319	error = xfs_iext_count_may_overflow(quotip, XFS_DATA_FORK,
 320			XFS_IEXT_ADD_NOSPLIT_CNT);
 321	if (error)
 322		return error;
 323
 324	/* Create the block mapping. */
 
 325	error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
 326			XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
 327			&nmaps);
 328	if (error)
 329		return error;
 330	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
 331	ASSERT(nmaps == 1);
 332	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 333	       (map.br_startblock != HOLESTARTBLOCK));
 334
 335	/*
 336	 * Keep track of the blkno to save a lookup later
 337	 */
 338	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 339
 340	/* now we can just get the buffer (there's nothing to read yet) */
 341	error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
 342			mp->m_quotainfo->qi_dqchunklen, 0, &bp);
 343	if (error)
 344		return error;
 345	bp->b_ops = &xfs_dquot_buf_ops;
 346
 347	/*
 348	 * Make a chunk of dquots out of this buffer and log
 349	 * the entire thing.
 350	 */
 351	xfs_qm_init_dquot_blk(tp, mp, dqp->q_id, qtype, bp);
 352	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 353
 354	/*
 355	 * Hold the buffer and join it to the dfops so that we'll still own
 356	 * the buffer when we return to the caller.  The buffer disposal on
 357	 * error must be paid attention to very carefully, as it has been
 358	 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
 359	 * code when allocating a new dquot record" in 2005, and the later
 360	 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
 361	 * the buffer locked across the _defer_finish call.  We can now do
 362	 * this correctly with xfs_defer_bjoin.
 363	 *
 364	 * Above, we allocated a disk block for the dquot information and used
 365	 * get_buf to initialize the dquot. If the _defer_finish fails, the old
 366	 * transaction is gone but the new buffer is not joined or held to any
 367	 * transaction, so we must _buf_relse it.
 368	 *
 369	 * If everything succeeds, the caller of this function is returned a
 370	 * buffer that is locked and held to the transaction.  The caller
 371	 * is responsible for unlocking any buffer passed back, either
 372	 * manually or by committing the transaction.  On error, the buffer is
 373	 * released and not passed back.
 374	 */
 375	xfs_trans_bhold(tp, bp);
 376	error = xfs_defer_finish(tpp);
 377	if (error) {
 378		xfs_trans_bhold_release(*tpp, bp);
 379		xfs_trans_brelse(*tpp, bp);
 380		return error;
 381	}
 382	*bpp = bp;
 383	return 0;
 384}
 385
 386/*
 387 * Read in the in-core dquot's on-disk metadata and return the buffer.
 388 * Returns ENOENT to signal a hole.
 389 */
 390STATIC int
 391xfs_dquot_disk_read(
 392	struct xfs_mount	*mp,
 393	struct xfs_dquot	*dqp,
 394	struct xfs_buf		**bpp)
 395{
 396	struct xfs_bmbt_irec	map;
 397	struct xfs_buf		*bp;
 398	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
 399	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
 400	uint			lock_mode;
 401	int			nmaps = 1;
 402	int			error;
 403
 404	lock_mode = xfs_ilock_data_map_shared(quotip);
 405	if (!xfs_this_quota_on(mp, qtype)) {
 406		/*
 407		 * Return if this type of quotas is turned off while we
 408		 * didn't have the quota inode lock.
 409		 */
 410		xfs_iunlock(quotip, lock_mode);
 411		return -ESRCH;
 412	}
 413
 414	/*
 415	 * Find the block map; no allocations yet
 416	 */
 417	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
 418			XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
 419	xfs_iunlock(quotip, lock_mode);
 420	if (error)
 421		return error;
 422
 423	ASSERT(nmaps == 1);
 424	ASSERT(map.br_blockcount >= 1);
 425	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
 426	if (map.br_startblock == HOLESTARTBLOCK)
 427		return -ENOENT;
 428
 429	trace_xfs_dqtobp_read(dqp);
 430
 431	/*
 432	 * store the blkno etc so that we don't have to do the
 433	 * mapping all the time
 434	 */
 435	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 436
 437	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
 438			mp->m_quotainfo->qi_dqchunklen, 0, &bp,
 439			&xfs_dquot_buf_ops);
 440	if (error) {
 441		ASSERT(bp == NULL);
 442		return error;
 443	}
 444
 445	ASSERT(xfs_buf_islocked(bp));
 446	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 447	*bpp = bp;
 448
 449	return 0;
 450}
 451
 452/* Allocate and initialize everything we need for an incore dquot. */
 453STATIC struct xfs_dquot *
 454xfs_dquot_alloc(
 455	struct xfs_mount	*mp,
 456	xfs_dqid_t		id,
 457	xfs_dqtype_t		type)
 458{
 459	struct xfs_dquot	*dqp;
 460
 461	dqp = kmem_cache_zalloc(xfs_qm_dqzone, GFP_KERNEL | __GFP_NOFAIL);
 462
 463	dqp->q_type = type;
 464	dqp->q_id = id;
 465	dqp->q_mount = mp;
 466	INIT_LIST_HEAD(&dqp->q_lru);
 467	mutex_init(&dqp->q_qlock);
 468	init_waitqueue_head(&dqp->q_pinwait);
 469	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
 470	/*
 471	 * Offset of dquot in the (fixed sized) dquot chunk.
 472	 */
 473	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
 474			sizeof(xfs_dqblk_t);
 475
 476	/*
 477	 * Because we want to use a counting completion, complete
 478	 * the flush completion once to allow a single access to
 479	 * the flush completion without blocking.
 480	 */
 481	init_completion(&dqp->q_flush);
 482	complete(&dqp->q_flush);
 483
 484	/*
 485	 * Make sure group quotas have a different lock class than user
 486	 * quotas.
 487	 */
 488	switch (type) {
 489	case XFS_DQTYPE_USER:
 490		/* uses the default lock class */
 491		break;
 492	case XFS_DQTYPE_GROUP:
 493		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
 494		break;
 495	case XFS_DQTYPE_PROJ:
 496		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
 497		break;
 498	default:
 499		ASSERT(0);
 500		break;
 501	}
 502
 503	xfs_qm_dquot_logitem_init(dqp);
 504
 505	XFS_STATS_INC(mp, xs_qm_dquot);
 506	return dqp;
 507}
 508
 509/* Check the ondisk dquot's id and type match what the incore dquot expects. */
 510static bool
 511xfs_dquot_check_type(
 512	struct xfs_dquot	*dqp,
 513	struct xfs_disk_dquot	*ddqp)
 514{
 515	uint8_t			ddqp_type;
 516	uint8_t			dqp_type;
 517
 518	ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
 519	dqp_type = xfs_dquot_type(dqp);
 520
 521	if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
 522		return false;
 523
 524	/*
 525	 * V5 filesystems always expect an exact type match.  V4 filesystems
 526	 * expect an exact match for user dquots and for non-root group and
 527	 * project dquots.
 528	 */
 529	if (xfs_sb_version_hascrc(&dqp->q_mount->m_sb) ||
 530	    dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
 531		return ddqp_type == dqp_type;
 532
 533	/*
 534	 * V4 filesystems support either group or project quotas, but not both
 535	 * at the same time.  The non-user quota file can be switched between
 536	 * group and project quota uses depending on the mount options, which
 537	 * means that we can encounter the other type when we try to load quota
 538	 * defaults.  Quotacheck will soon reset the the entire quota file
 539	 * (including the root dquot) anyway, but don't log scary corruption
 540	 * reports to dmesg.
 541	 */
 542	return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
 543}
 544
 545/* Copy the in-core quota fields in from the on-disk buffer. */
 546STATIC int
 547xfs_dquot_from_disk(
 548	struct xfs_dquot	*dqp,
 549	struct xfs_buf		*bp)
 550{
 551	struct xfs_disk_dquot	*ddqp = bp->b_addr + dqp->q_bufoffset;
 552
 553	/*
 554	 * Ensure that we got the type and ID we were looking for.
 555	 * Everything else was checked by the dquot buffer verifier.
 556	 */
 557	if (!xfs_dquot_check_type(dqp, ddqp)) {
 
 558		xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
 559			  "Metadata corruption detected at %pS, quota %u",
 560			  __this_address, dqp->q_id);
 561		xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
 562		return -EFSCORRUPTED;
 563	}
 564
 565	/* copy everything from disk dquot to the incore dquot */
 566	dqp->q_type = ddqp->d_type;
 567	dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 568	dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 569	dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 570	dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 571	dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 572	dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 573
 574	dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
 575	dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
 576	dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
 577
 578	dqp->q_blk.warnings = be16_to_cpu(ddqp->d_bwarns);
 579	dqp->q_ino.warnings = be16_to_cpu(ddqp->d_iwarns);
 580	dqp->q_rtb.warnings = be16_to_cpu(ddqp->d_rtbwarns);
 581
 582	dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
 583	dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
 584	dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
 585
 586	/*
 587	 * Reservation counters are defined as reservation plus current usage
 588	 * to avoid having to add every time.
 589	 */
 590	dqp->q_blk.reserved = dqp->q_blk.count;
 591	dqp->q_ino.reserved = dqp->q_ino.count;
 592	dqp->q_rtb.reserved = dqp->q_rtb.count;
 593
 594	/* initialize the dquot speculative prealloc thresholds */
 595	xfs_dquot_set_prealloc_limits(dqp);
 596	return 0;
 597}
 598
 599/* Copy the in-core quota fields into the on-disk buffer. */
 600void
 601xfs_dquot_to_disk(
 602	struct xfs_disk_dquot	*ddqp,
 603	struct xfs_dquot	*dqp)
 604{
 605	ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 606	ddqp->d_version = XFS_DQUOT_VERSION;
 607	ddqp->d_type = dqp->q_type;
 608	ddqp->d_id = cpu_to_be32(dqp->q_id);
 609	ddqp->d_pad0 = 0;
 610	ddqp->d_pad = 0;
 611
 612	ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
 613	ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
 614	ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
 615	ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
 616	ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
 617	ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
 618
 619	ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
 620	ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
 621	ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
 622
 623	ddqp->d_bwarns = cpu_to_be16(dqp->q_blk.warnings);
 624	ddqp->d_iwarns = cpu_to_be16(dqp->q_ino.warnings);
 625	ddqp->d_rtbwarns = cpu_to_be16(dqp->q_rtb.warnings);
 626
 627	ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
 628	ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
 629	ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
 630}
 631
 632/* Allocate and initialize the dquot buffer for this in-core dquot. */
 633static int
 634xfs_qm_dqread_alloc(
 635	struct xfs_mount	*mp,
 636	struct xfs_dquot	*dqp,
 637	struct xfs_buf		**bpp)
 638{
 639	struct xfs_trans	*tp;
 640	int			error;
 641
 642	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
 643			XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
 644	if (error)
 645		goto err;
 646
 647	error = xfs_dquot_disk_alloc(&tp, dqp, bpp);
 648	if (error)
 649		goto err_cancel;
 650
 651	error = xfs_trans_commit(tp);
 652	if (error) {
 653		/*
 654		 * Buffer was held to the transaction, so we have to unlock it
 655		 * manually here because we're not passing it back.
 656		 */
 657		xfs_buf_relse(*bpp);
 658		*bpp = NULL;
 659		goto err;
 660	}
 661	return 0;
 662
 663err_cancel:
 664	xfs_trans_cancel(tp);
 665err:
 666	return error;
 667}
 668
 669/*
 670 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
 671 * and release the buffer immediately.  If @can_alloc is true, fill any
 672 * holes in the on-disk metadata.
 673 */
 674static int
 675xfs_qm_dqread(
 676	struct xfs_mount	*mp,
 677	xfs_dqid_t		id,
 678	xfs_dqtype_t		type,
 679	bool			can_alloc,
 680	struct xfs_dquot	**dqpp)
 681{
 682	struct xfs_dquot	*dqp;
 683	struct xfs_buf		*bp;
 684	int			error;
 685
 686	dqp = xfs_dquot_alloc(mp, id, type);
 687	trace_xfs_dqread(dqp);
 688
 689	/* Try to read the buffer, allocating if necessary. */
 690	error = xfs_dquot_disk_read(mp, dqp, &bp);
 691	if (error == -ENOENT && can_alloc)
 692		error = xfs_qm_dqread_alloc(mp, dqp, &bp);
 693	if (error)
 694		goto err;
 695
 696	/*
 697	 * At this point we should have a clean locked buffer.  Copy the data
 698	 * to the incore dquot and release the buffer since the incore dquot
 699	 * has its own locking protocol so we needn't tie up the buffer any
 700	 * further.
 701	 */
 702	ASSERT(xfs_buf_islocked(bp));
 703	error = xfs_dquot_from_disk(dqp, bp);
 704	xfs_buf_relse(bp);
 705	if (error)
 706		goto err;
 707
 708	*dqpp = dqp;
 709	return error;
 710
 711err:
 712	trace_xfs_dqread_fail(dqp);
 713	xfs_qm_dqdestroy(dqp);
 714	*dqpp = NULL;
 715	return error;
 716}
 717
 718/*
 719 * Advance to the next id in the current chunk, or if at the
 720 * end of the chunk, skip ahead to first id in next allocated chunk
 721 * using the SEEK_DATA interface.
 722 */
 723static int
 724xfs_dq_get_next_id(
 725	struct xfs_mount	*mp,
 726	xfs_dqtype_t		type,
 727	xfs_dqid_t		*id)
 728{
 729	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
 730	xfs_dqid_t		next_id = *id + 1; /* simple advance */
 731	uint			lock_flags;
 732	struct xfs_bmbt_irec	got;
 733	struct xfs_iext_cursor	cur;
 734	xfs_fsblock_t		start;
 735	int			error = 0;
 736
 737	/* If we'd wrap past the max ID, stop */
 738	if (next_id < *id)
 739		return -ENOENT;
 740
 741	/* If new ID is within the current chunk, advancing it sufficed */
 742	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
 743		*id = next_id;
 744		return 0;
 745	}
 746
 747	/* Nope, next_id is now past the current chunk, so find the next one */
 748	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
 749
 750	lock_flags = xfs_ilock_data_map_shared(quotip);
 751	error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
 752	if (error)
 753		return error;
 
 
 754
 755	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
 756		/* contiguous chunk, bump startoff for the id calculation */
 757		if (got.br_startoff < start)
 758			got.br_startoff = start;
 759		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
 760	} else {
 761		error = -ENOENT;
 762	}
 763
 764	xfs_iunlock(quotip, lock_flags);
 765
 766	return error;
 767}
 768
 769/*
 770 * Look up the dquot in the in-core cache.  If found, the dquot is returned
 771 * locked and ready to go.
 772 */
 773static struct xfs_dquot *
 774xfs_qm_dqget_cache_lookup(
 775	struct xfs_mount	*mp,
 776	struct xfs_quotainfo	*qi,
 777	struct radix_tree_root	*tree,
 778	xfs_dqid_t		id)
 779{
 780	struct xfs_dquot	*dqp;
 781
 782restart:
 783	mutex_lock(&qi->qi_tree_lock);
 784	dqp = radix_tree_lookup(tree, id);
 785	if (!dqp) {
 786		mutex_unlock(&qi->qi_tree_lock);
 787		XFS_STATS_INC(mp, xs_qm_dqcachemisses);
 788		return NULL;
 789	}
 790
 791	xfs_dqlock(dqp);
 792	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
 793		xfs_dqunlock(dqp);
 794		mutex_unlock(&qi->qi_tree_lock);
 795		trace_xfs_dqget_freeing(dqp);
 796		delay(1);
 797		goto restart;
 798	}
 799
 800	dqp->q_nrefs++;
 801	mutex_unlock(&qi->qi_tree_lock);
 802
 803	trace_xfs_dqget_hit(dqp);
 804	XFS_STATS_INC(mp, xs_qm_dqcachehits);
 805	return dqp;
 806}
 807
 808/*
 809 * Try to insert a new dquot into the in-core cache.  If an error occurs the
 810 * caller should throw away the dquot and start over.  Otherwise, the dquot
 811 * is returned locked (and held by the cache) as if there had been a cache
 812 * hit.
 813 */
 814static int
 815xfs_qm_dqget_cache_insert(
 816	struct xfs_mount	*mp,
 817	struct xfs_quotainfo	*qi,
 818	struct radix_tree_root	*tree,
 819	xfs_dqid_t		id,
 820	struct xfs_dquot	*dqp)
 821{
 822	int			error;
 823
 824	mutex_lock(&qi->qi_tree_lock);
 825	error = radix_tree_insert(tree, id, dqp);
 826	if (unlikely(error)) {
 827		/* Duplicate found!  Caller must try again. */
 828		WARN_ON(error != -EEXIST);
 829		mutex_unlock(&qi->qi_tree_lock);
 830		trace_xfs_dqget_dup(dqp);
 831		return error;
 832	}
 833
 834	/* Return a locked dquot to the caller, with a reference taken. */
 835	xfs_dqlock(dqp);
 836	dqp->q_nrefs = 1;
 837
 838	qi->qi_dquots++;
 839	mutex_unlock(&qi->qi_tree_lock);
 840
 841	return 0;
 842}
 843
 844/* Check our input parameters. */
 845static int
 846xfs_qm_dqget_checks(
 847	struct xfs_mount	*mp,
 848	xfs_dqtype_t		type)
 849{
 850	if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp)))
 851		return -ESRCH;
 852
 853	switch (type) {
 854	case XFS_DQTYPE_USER:
 855		if (!XFS_IS_UQUOTA_ON(mp))
 856			return -ESRCH;
 857		return 0;
 858	case XFS_DQTYPE_GROUP:
 859		if (!XFS_IS_GQUOTA_ON(mp))
 860			return -ESRCH;
 861		return 0;
 862	case XFS_DQTYPE_PROJ:
 863		if (!XFS_IS_PQUOTA_ON(mp))
 864			return -ESRCH;
 865		return 0;
 866	default:
 867		WARN_ON_ONCE(0);
 868		return -EINVAL;
 869	}
 870}
 871
 872/*
 873 * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
 874 * locked dquot, doing an allocation (if requested) as needed.
 875 */
 876int
 877xfs_qm_dqget(
 878	struct xfs_mount	*mp,
 879	xfs_dqid_t		id,
 880	xfs_dqtype_t		type,
 881	bool			can_alloc,
 882	struct xfs_dquot	**O_dqpp)
 883{
 884	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 885	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 886	struct xfs_dquot	*dqp;
 887	int			error;
 888
 889	error = xfs_qm_dqget_checks(mp, type);
 890	if (error)
 891		return error;
 892
 893restart:
 894	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
 895	if (dqp) {
 896		*O_dqpp = dqp;
 897		return 0;
 898	}
 899
 900	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
 901	if (error)
 902		return error;
 903
 904	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
 905	if (error) {
 906		/*
 907		 * Duplicate found. Just throw away the new dquot and start
 908		 * over.
 909		 */
 910		xfs_qm_dqdestroy(dqp);
 911		XFS_STATS_INC(mp, xs_qm_dquot_dups);
 912		goto restart;
 913	}
 914
 915	trace_xfs_dqget_miss(dqp);
 916	*O_dqpp = dqp;
 917	return 0;
 918}
 919
 920/*
 921 * Given a dquot id and type, read and initialize a dquot from the on-disk
 922 * metadata.  This function is only for use during quota initialization so
 923 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
 924 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
 925 */
 926int
 927xfs_qm_dqget_uncached(
 928	struct xfs_mount	*mp,
 929	xfs_dqid_t		id,
 930	xfs_dqtype_t		type,
 931	struct xfs_dquot	**dqpp)
 932{
 933	int			error;
 934
 935	error = xfs_qm_dqget_checks(mp, type);
 936	if (error)
 937		return error;
 938
 939	return xfs_qm_dqread(mp, id, type, 0, dqpp);
 940}
 941
 942/* Return the quota id for a given inode and type. */
 943xfs_dqid_t
 944xfs_qm_id_for_quotatype(
 945	struct xfs_inode	*ip,
 946	xfs_dqtype_t		type)
 947{
 948	switch (type) {
 949	case XFS_DQTYPE_USER:
 950		return i_uid_read(VFS_I(ip));
 951	case XFS_DQTYPE_GROUP:
 952		return i_gid_read(VFS_I(ip));
 953	case XFS_DQTYPE_PROJ:
 954		return ip->i_projid;
 955	}
 956	ASSERT(0);
 957	return 0;
 958}
 959
 960/*
 961 * Return the dquot for a given inode and type.  If @can_alloc is true, then
 962 * allocate blocks if needed.  The inode's ILOCK must be held and it must not
 963 * have already had an inode attached.
 964 */
 965int
 966xfs_qm_dqget_inode(
 967	struct xfs_inode	*ip,
 968	xfs_dqtype_t		type,
 969	bool			can_alloc,
 970	struct xfs_dquot	**O_dqpp)
 971{
 972	struct xfs_mount	*mp = ip->i_mount;
 973	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 974	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 975	struct xfs_dquot	*dqp;
 976	xfs_dqid_t		id;
 977	int			error;
 978
 979	error = xfs_qm_dqget_checks(mp, type);
 980	if (error)
 981		return error;
 982
 983	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 984	ASSERT(xfs_inode_dquot(ip, type) == NULL);
 985
 986	id = xfs_qm_id_for_quotatype(ip, type);
 987
 988restart:
 989	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
 990	if (dqp) {
 991		*O_dqpp = dqp;
 992		return 0;
 993	}
 994
 995	/*
 996	 * Dquot cache miss. We don't want to keep the inode lock across
 997	 * a (potential) disk read. Also we don't want to deal with the lock
 998	 * ordering between quotainode and this inode. OTOH, dropping the inode
 999	 * lock here means dealing with a chown that can happen before
1000	 * we re-acquire the lock.
1001	 */
1002	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1003	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
1004	xfs_ilock(ip, XFS_ILOCK_EXCL);
1005	if (error)
1006		return error;
1007
1008	/*
1009	 * A dquot could be attached to this inode by now, since we had
1010	 * dropped the ilock.
1011	 */
1012	if (xfs_this_quota_on(mp, type)) {
1013		struct xfs_dquot	*dqp1;
1014
1015		dqp1 = xfs_inode_dquot(ip, type);
1016		if (dqp1) {
1017			xfs_qm_dqdestroy(dqp);
1018			dqp = dqp1;
1019			xfs_dqlock(dqp);
1020			goto dqret;
1021		}
1022	} else {
1023		/* inode stays locked on return */
1024		xfs_qm_dqdestroy(dqp);
1025		return -ESRCH;
1026	}
1027
1028	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1029	if (error) {
1030		/*
1031		 * Duplicate found. Just throw away the new dquot and start
1032		 * over.
1033		 */
1034		xfs_qm_dqdestroy(dqp);
1035		XFS_STATS_INC(mp, xs_qm_dquot_dups);
1036		goto restart;
1037	}
1038
1039dqret:
1040	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1041	trace_xfs_dqget_miss(dqp);
1042	*O_dqpp = dqp;
1043	return 0;
1044}
1045
1046/*
1047 * Starting at @id and progressing upwards, look for an initialized incore
1048 * dquot, lock it, and return it.
1049 */
1050int
1051xfs_qm_dqget_next(
1052	struct xfs_mount	*mp,
1053	xfs_dqid_t		id,
1054	xfs_dqtype_t		type,
1055	struct xfs_dquot	**dqpp)
1056{
1057	struct xfs_dquot	*dqp;
1058	int			error = 0;
1059
1060	*dqpp = NULL;
1061	for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1062		error = xfs_qm_dqget(mp, id, type, false, &dqp);
1063		if (error == -ENOENT)
1064			continue;
1065		else if (error != 0)
1066			break;
1067
1068		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1069			*dqpp = dqp;
1070			return 0;
1071		}
1072
1073		xfs_qm_dqput(dqp);
1074	}
1075
1076	return error;
1077}
1078
1079/*
1080 * Release a reference to the dquot (decrement ref-count) and unlock it.
1081 *
1082 * If there is a group quota attached to this dquot, carefully release that
1083 * too without tripping over deadlocks'n'stuff.
1084 */
1085void
1086xfs_qm_dqput(
1087	struct xfs_dquot	*dqp)
1088{
1089	ASSERT(dqp->q_nrefs > 0);
1090	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1091
1092	trace_xfs_dqput(dqp);
1093
1094	if (--dqp->q_nrefs == 0) {
1095		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
1096		trace_xfs_dqput_free(dqp);
1097
1098		if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
1099			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1100	}
1101	xfs_dqunlock(dqp);
1102}
1103
1104/*
1105 * Release a dquot. Flush it if dirty, then dqput() it.
1106 * dquot must not be locked.
1107 */
1108void
1109xfs_qm_dqrele(
1110	struct xfs_dquot	*dqp)
1111{
1112	if (!dqp)
1113		return;
1114
1115	trace_xfs_dqrele(dqp);
1116
1117	xfs_dqlock(dqp);
1118	/*
1119	 * We don't care to flush it if the dquot is dirty here.
1120	 * That will create stutters that we want to avoid.
1121	 * Instead we do a delayed write when we try to reclaim
1122	 * a dirty dquot. Also xfs_sync will take part of the burden...
1123	 */
1124	xfs_qm_dqput(dqp);
1125}
1126
1127/*
1128 * This is the dquot flushing I/O completion routine.  It is called
1129 * from interrupt level when the buffer containing the dquot is
1130 * flushed to disk.  It is responsible for removing the dquot logitem
1131 * from the AIL if it has not been re-logged, and unlocking the dquot's
1132 * flush lock. This behavior is very similar to that of inodes..
1133 */
1134static void
1135xfs_qm_dqflush_done(
1136	struct xfs_log_item	*lip)
1137{
1138	struct xfs_dq_logitem	*qip = (struct xfs_dq_logitem *)lip;
1139	struct xfs_dquot	*dqp = qip->qli_dquot;
1140	struct xfs_ail		*ailp = lip->li_ailp;
1141	xfs_lsn_t		tail_lsn;
1142
1143	/*
1144	 * We only want to pull the item from the AIL if its
1145	 * location in the log has not changed since we started the flush.
1146	 * Thus, we only bother if the dquot's lsn has
1147	 * not changed. First we check the lsn outside the lock
1148	 * since it's cheaper, and then we recheck while
1149	 * holding the lock before removing the dquot from the AIL.
1150	 */
1151	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1152	    ((lip->li_lsn == qip->qli_flush_lsn) ||
1153	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1154
1155		spin_lock(&ailp->ail_lock);
1156		xfs_clear_li_failed(lip);
1157		if (lip->li_lsn == qip->qli_flush_lsn) {
1158			/* xfs_ail_update_finish() drops the AIL lock */
1159			tail_lsn = xfs_ail_delete_one(ailp, lip);
1160			xfs_ail_update_finish(ailp, tail_lsn);
1161		} else {
1162			spin_unlock(&ailp->ail_lock);
1163		}
1164	}
1165
1166	/*
1167	 * Release the dq's flush lock since we're done with it.
1168	 */
1169	xfs_dqfunlock(dqp);
1170}
1171
1172void
1173xfs_buf_dquot_iodone(
1174	struct xfs_buf		*bp)
1175{
1176	struct xfs_log_item	*lip, *n;
1177
1178	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1179		list_del_init(&lip->li_bio_list);
1180		xfs_qm_dqflush_done(lip);
1181	}
1182}
1183
1184void
1185xfs_buf_dquot_io_fail(
1186	struct xfs_buf		*bp)
1187{
1188	struct xfs_log_item	*lip;
1189
1190	spin_lock(&bp->b_mount->m_ail->ail_lock);
1191	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1192		xfs_set_li_failed(lip, bp);
1193	spin_unlock(&bp->b_mount->m_ail->ail_lock);
1194}
1195
1196/* Check incore dquot for errors before we flush. */
1197static xfs_failaddr_t
1198xfs_qm_dqflush_check(
1199	struct xfs_dquot	*dqp)
1200{
1201	xfs_dqtype_t		type = xfs_dquot_type(dqp);
1202
1203	if (type != XFS_DQTYPE_USER &&
1204	    type != XFS_DQTYPE_GROUP &&
1205	    type != XFS_DQTYPE_PROJ)
1206		return __this_address;
1207
1208	if (dqp->q_id == 0)
1209		return NULL;
1210
1211	if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1212	    !dqp->q_blk.timer)
1213		return __this_address;
1214
1215	if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1216	    !dqp->q_ino.timer)
1217		return __this_address;
1218
1219	if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1220	    !dqp->q_rtb.timer)
1221		return __this_address;
1222
1223	/* bigtime flag should never be set on root dquots */
1224	if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1225		if (!xfs_sb_version_hasbigtime(&dqp->q_mount->m_sb))
1226			return __this_address;
1227		if (dqp->q_id == 0)
1228			return __this_address;
1229	}
1230
1231	return NULL;
1232}
1233
1234/*
1235 * Write a modified dquot to disk.
1236 * The dquot must be locked and the flush lock too taken by caller.
1237 * The flush lock will not be unlocked until the dquot reaches the disk,
1238 * but the dquot is free to be unlocked and modified by the caller
1239 * in the interim. Dquot is still locked on return. This behavior is
1240 * identical to that of inodes.
1241 */
1242int
1243xfs_qm_dqflush(
1244	struct xfs_dquot	*dqp,
1245	struct xfs_buf		**bpp)
1246{
1247	struct xfs_mount	*mp = dqp->q_mount;
1248	struct xfs_log_item	*lip = &dqp->q_logitem.qli_item;
1249	struct xfs_buf		*bp;
1250	struct xfs_dqblk	*dqblk;
1251	xfs_failaddr_t		fa;
1252	int			error;
1253
1254	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1255	ASSERT(!completion_done(&dqp->q_flush));
1256
1257	trace_xfs_dqflush(dqp);
1258
1259	*bpp = NULL;
1260
1261	xfs_qm_dqunpin_wait(dqp);
1262
1263	/*
1264	 * Get the buffer containing the on-disk dquot
1265	 */
1266	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1267				   mp->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK,
1268				   &bp, &xfs_dquot_buf_ops);
1269	if (error == -EAGAIN)
1270		goto out_unlock;
1271	if (error)
1272		goto out_abort;
1273
1274	fa = xfs_qm_dqflush_check(dqp);
1275	if (fa) {
1276		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1277				dqp->q_id, fa);
1278		xfs_buf_relse(bp);
1279		error = -EFSCORRUPTED;
1280		goto out_abort;
1281	}
1282
1283	/* Flush the incore dquot to the ondisk buffer. */
1284	dqblk = bp->b_addr + dqp->q_bufoffset;
1285	xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1286
1287	/*
1288	 * Clear the dirty field and remember the flush lsn for later use.
1289	 */
1290	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1291
1292	xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1293					&dqp->q_logitem.qli_item.li_lsn);
1294
1295	/*
1296	 * copy the lsn into the on-disk dquot now while we have the in memory
1297	 * dquot here. This can't be done later in the write verifier as we
1298	 * can't get access to the log item at that point in time.
1299	 *
1300	 * We also calculate the CRC here so that the on-disk dquot in the
1301	 * buffer always has a valid CRC. This ensures there is no possibility
1302	 * of a dquot without an up-to-date CRC getting to disk.
1303	 */
1304	if (xfs_sb_version_hascrc(&mp->m_sb)) {
1305		dqblk->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1306		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1307				 XFS_DQUOT_CRC_OFF);
1308	}
1309
1310	/*
1311	 * Attach the dquot to the buffer so that we can remove this dquot from
1312	 * the AIL and release the flush lock once the dquot is synced to disk.
1313	 */
1314	bp->b_flags |= _XBF_DQUOTS;
1315	list_add_tail(&dqp->q_logitem.qli_item.li_bio_list, &bp->b_li_list);
1316
1317	/*
1318	 * If the buffer is pinned then push on the log so we won't
1319	 * get stuck waiting in the write for too long.
1320	 */
1321	if (xfs_buf_ispinned(bp)) {
1322		trace_xfs_dqflush_force(dqp);
1323		xfs_log_force(mp, 0);
1324	}
1325
1326	trace_xfs_dqflush_done(dqp);
1327	*bpp = bp;
1328	return 0;
1329
1330out_abort:
1331	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1332	xfs_trans_ail_delete(lip, 0);
1333	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1334out_unlock:
1335	xfs_dqfunlock(dqp);
1336	return error;
1337}
1338
1339/*
1340 * Lock two xfs_dquot structures.
1341 *
1342 * To avoid deadlocks we always lock the quota structure with
1343 * the lowerd id first.
1344 */
1345void
1346xfs_dqlock2(
1347	struct xfs_dquot	*d1,
1348	struct xfs_dquot	*d2)
1349{
1350	if (d1 && d2) {
1351		ASSERT(d1 != d2);
1352		if (d1->q_id > d2->q_id) {
1353			mutex_lock(&d2->q_qlock);
1354			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1355		} else {
1356			mutex_lock(&d1->q_qlock);
1357			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1358		}
1359	} else if (d1) {
1360		mutex_lock(&d1->q_qlock);
1361	} else if (d2) {
1362		mutex_lock(&d2->q_qlock);
1363	}
1364}
1365
1366int __init
1367xfs_qm_init(void)
1368{
1369	xfs_qm_dqzone = kmem_cache_create("xfs_dquot",
1370					  sizeof(struct xfs_dquot),
1371					  0, 0, NULL);
1372	if (!xfs_qm_dqzone)
1373		goto out;
1374
1375	xfs_qm_dqtrxzone = kmem_cache_create("xfs_dqtrx",
1376					     sizeof(struct xfs_dquot_acct),
1377					     0, 0, NULL);
1378	if (!xfs_qm_dqtrxzone)
1379		goto out_free_dqzone;
1380
1381	return 0;
1382
1383out_free_dqzone:
1384	kmem_cache_destroy(xfs_qm_dqzone);
1385out:
1386	return -ENOMEM;
1387}
1388
1389void
1390xfs_qm_exit(void)
1391{
1392	kmem_cache_destroy(xfs_qm_dqtrxzone);
1393	kmem_cache_destroy(xfs_qm_dqzone);
1394}
1395
1396/*
1397 * Iterate every dquot of a particular type.  The caller must ensure that the
1398 * particular quota type is active.  iter_fn can return negative error codes,
1399 * or -ECANCELED to indicate that it wants to stop iterating.
1400 */
1401int
1402xfs_qm_dqiterate(
1403	struct xfs_mount	*mp,
1404	xfs_dqtype_t		type,
1405	xfs_qm_dqiterate_fn	iter_fn,
1406	void			*priv)
1407{
1408	struct xfs_dquot	*dq;
1409	xfs_dqid_t		id = 0;
1410	int			error;
1411
1412	do {
1413		error = xfs_qm_dqget_next(mp, id, type, &dq);
1414		if (error == -ENOENT)
1415			return 0;
1416		if (error)
1417			return error;
1418
1419		error = iter_fn(dq, type, priv);
1420		id = dq->q_id;
1421		xfs_qm_dqput(dq);
1422	} while (error == 0 && id != 0);
1423
1424	return error;
1425}