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#include "xfs_health.h"
  28
  29/*
  30 * Lock order:
  31 *
  32 * ip->i_lock
  33 *   qi->qi_tree_lock
  34 *     dquot->q_qlock (xfs_dqlock() and friends)
  35 *       dquot->q_flush (xfs_dqflock() and friends)
  36 *       qi->qi_lru_lock
  37 *
  38 * If two dquots need to be locked the order is user before group/project,
  39 * otherwise by the lowest id first, see xfs_dqlock2.
  40 */
  41
  42struct kmem_cache		*xfs_dqtrx_cache;
  43static struct kmem_cache	*xfs_dquot_cache;
  44
  45static struct lock_class_key xfs_dquot_group_class;
  46static struct lock_class_key xfs_dquot_project_class;
  47
  48/* Record observations of quota corruption with the health tracking system. */
  49static void
  50xfs_dquot_mark_sick(
  51	struct xfs_dquot	*dqp)
  52{
  53	struct xfs_mount	*mp = dqp->q_mount;
  54
  55	switch (dqp->q_type) {
  56	case XFS_DQTYPE_USER:
  57		xfs_fs_mark_sick(mp, XFS_SICK_FS_UQUOTA);
  58		break;
  59	case XFS_DQTYPE_GROUP:
  60		xfs_fs_mark_sick(mp, XFS_SICK_FS_GQUOTA);
  61		break;
  62	case XFS_DQTYPE_PROJ:
  63		xfs_fs_mark_sick(mp, XFS_SICK_FS_PQUOTA);
  64		break;
  65	default:
  66		ASSERT(0);
  67		break;
  68	}
  69}
  70
  71/*
  72 * Detach the dquot buffer if it's still attached, because we can get called
  73 * through dqpurge after a log shutdown.  Caller must hold the dqflock or have
  74 * otherwise isolated the dquot.
  75 */
  76void
  77xfs_dquot_detach_buf(
  78	struct xfs_dquot	*dqp)
  79{
  80	struct xfs_dq_logitem	*qlip = &dqp->q_logitem;
  81	struct xfs_buf		*bp = NULL;
  82
  83	spin_lock(&qlip->qli_lock);
  84	if (qlip->qli_item.li_buf) {
  85		bp = qlip->qli_item.li_buf;
  86		qlip->qli_item.li_buf = NULL;
  87	}
  88	spin_unlock(&qlip->qli_lock);
  89	if (bp) {
  90		xfs_buf_lock(bp);
  91		list_del_init(&qlip->qli_item.li_bio_list);
  92		xfs_buf_relse(bp);
  93	}
  94}
  95
  96/*
  97 * This is called to free all the memory associated with a dquot
  98 */
  99void
 100xfs_qm_dqdestroy(
 101	struct xfs_dquot	*dqp)
 102{
 103	ASSERT(list_empty(&dqp->q_lru));
 104	ASSERT(dqp->q_logitem.qli_item.li_buf == NULL);
 105
 106	kvfree(dqp->q_logitem.qli_item.li_lv_shadow);
 107	mutex_destroy(&dqp->q_qlock);
 108
 109	XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
 110	kmem_cache_free(xfs_dquot_cache, dqp);
 111}
 112
 113/*
 114 * If default limits are in force, push them into the dquot now.
 115 * We overwrite the dquot limits only if they are zero and this
 116 * is not the root dquot.
 117 */
 118void
 119xfs_qm_adjust_dqlimits(
 120	struct xfs_dquot	*dq)
 121{
 122	struct xfs_mount	*mp = dq->q_mount;
 123	struct xfs_quotainfo	*q = mp->m_quotainfo;
 124	struct xfs_def_quota	*defq;
 125	int			prealloc = 0;
 126
 127	ASSERT(dq->q_id);
 128	defq = xfs_get_defquota(q, xfs_dquot_type(dq));
 129
 130	if (!dq->q_blk.softlimit) {
 131		dq->q_blk.softlimit = defq->blk.soft;
 132		prealloc = 1;
 133	}
 134	if (!dq->q_blk.hardlimit) {
 135		dq->q_blk.hardlimit = defq->blk.hard;
 136		prealloc = 1;
 137	}
 138	if (!dq->q_ino.softlimit)
 139		dq->q_ino.softlimit = defq->ino.soft;
 140	if (!dq->q_ino.hardlimit)
 141		dq->q_ino.hardlimit = defq->ino.hard;
 142	if (!dq->q_rtb.softlimit)
 143		dq->q_rtb.softlimit = defq->rtb.soft;
 144	if (!dq->q_rtb.hardlimit)
 145		dq->q_rtb.hardlimit = defq->rtb.hard;
 146
 147	if (prealloc)
 148		xfs_dquot_set_prealloc_limits(dq);
 149}
 150
 151/* Set the expiration time of a quota's grace period. */
 152time64_t
 153xfs_dquot_set_timeout(
 154	struct xfs_mount	*mp,
 155	time64_t		timeout)
 156{
 157	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 158
 159	return clamp_t(time64_t, timeout, qi->qi_expiry_min,
 160					  qi->qi_expiry_max);
 161}
 162
 163/* Set the length of the default grace period. */
 164time64_t
 165xfs_dquot_set_grace_period(
 166	time64_t		grace)
 167{
 168	return clamp_t(time64_t, grace, XFS_DQ_GRACE_MIN, XFS_DQ_GRACE_MAX);
 169}
 170
 171/*
 172 * Determine if this quota counter is over either limit and set the quota
 173 * timers as appropriate.
 174 */
 175static inline void
 176xfs_qm_adjust_res_timer(
 177	struct xfs_mount	*mp,
 178	struct xfs_dquot_res	*res,
 179	struct xfs_quota_limits	*qlim)
 180{
 181	ASSERT(res->hardlimit == 0 || res->softlimit <= res->hardlimit);
 182
 183	if ((res->softlimit && res->count > res->softlimit) ||
 184	    (res->hardlimit && res->count > res->hardlimit)) {
 185		if (res->timer == 0)
 186			res->timer = xfs_dquot_set_timeout(mp,
 187					ktime_get_real_seconds() + qlim->time);
 188	} else {
 189		res->timer = 0;
 190	}
 191}
 192
 193/*
 194 * Check the limits and timers of a dquot and start or reset timers
 195 * if necessary.
 196 * This gets called even when quota enforcement is OFF, which makes our
 197 * life a little less complicated. (We just don't reject any quota
 198 * reservations in that case, when enforcement is off).
 199 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
 200 * enforcement's off.
 201 * In contrast, warnings are a little different in that they don't
 202 * 'automatically' get started when limits get exceeded.  They do
 203 * get reset to zero, however, when we find the count to be under
 204 * the soft limit (they are only ever set non-zero via userspace).
 205 */
 206void
 207xfs_qm_adjust_dqtimers(
 208	struct xfs_dquot	*dq)
 209{
 210	struct xfs_mount	*mp = dq->q_mount;
 211	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 212	struct xfs_def_quota	*defq;
 213
 214	ASSERT(dq->q_id);
 215	defq = xfs_get_defquota(qi, xfs_dquot_type(dq));
 216
 217	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_blk, &defq->blk);
 218	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_ino, &defq->ino);
 219	xfs_qm_adjust_res_timer(dq->q_mount, &dq->q_rtb, &defq->rtb);
 220}
 221
 222/*
 223 * initialize a buffer full of dquots and log the whole thing
 224 */
 225void
 226xfs_qm_init_dquot_blk(
 227	struct xfs_trans	*tp,
 228	xfs_dqid_t		id,
 229	xfs_dqtype_t		type,
 230	struct xfs_buf		*bp)
 231{
 232	struct xfs_mount	*mp = tp->t_mountp;
 233	struct xfs_quotainfo	*q = mp->m_quotainfo;
 234	struct xfs_dqblk	*d;
 235	xfs_dqid_t		curid;
 236	unsigned int		qflag;
 237	unsigned int		blftype;
 238	int			i;
 239
 240	ASSERT(tp);
 241	ASSERT(xfs_buf_islocked(bp));
 242
 243	switch (type) {
 244	case XFS_DQTYPE_USER:
 245		qflag = XFS_UQUOTA_CHKD;
 246		blftype = XFS_BLF_UDQUOT_BUF;
 247		break;
 248	case XFS_DQTYPE_PROJ:
 249		qflag = XFS_PQUOTA_CHKD;
 250		blftype = XFS_BLF_PDQUOT_BUF;
 251		break;
 252	case XFS_DQTYPE_GROUP:
 253		qflag = XFS_GQUOTA_CHKD;
 254		blftype = XFS_BLF_GDQUOT_BUF;
 255		break;
 256	default:
 257		ASSERT(0);
 258		return;
 259	}
 260
 261	d = bp->b_addr;
 262
 263	/*
 264	 * ID of the first dquot in the block - id's are zero based.
 265	 */
 266	curid = id - (id % q->qi_dqperchunk);
 267	memset(d, 0, BBTOB(q->qi_dqchunklen));
 268	for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
 269		d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 270		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
 271		d->dd_diskdq.d_id = cpu_to_be32(curid);
 272		d->dd_diskdq.d_type = type;
 273		if (curid > 0 && xfs_has_bigtime(mp))
 274			d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
 275		if (xfs_has_crc(mp)) {
 276			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
 277			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
 278					 XFS_DQUOT_CRC_OFF);
 279		}
 280	}
 281
 282	xfs_trans_dquot_buf(tp, bp, blftype);
 283
 284	/*
 285	 * quotacheck uses delayed writes to update all the dquots on disk in an
 286	 * efficient manner instead of logging the individual dquot changes as
 287	 * they are made. However if we log the buffer allocated here and crash
 288	 * after quotacheck while the logged initialisation is still in the
 289	 * active region of the log, log recovery can replay the dquot buffer
 290	 * initialisation over the top of the checked dquots and corrupt quota
 291	 * accounting.
 292	 *
 293	 * To avoid this problem, quotacheck cannot log the initialised buffer.
 294	 * We must still dirty the buffer and write it back before the
 295	 * allocation transaction clears the log. Therefore, mark the buffer as
 296	 * ordered instead of logging it directly. This is safe for quotacheck
 297	 * because it detects and repairs allocated but initialized dquot blocks
 298	 * in the quota inodes.
 299	 */
 300	if (!(mp->m_qflags & qflag))
 301		xfs_trans_ordered_buf(tp, bp);
 302	else
 303		xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
 304}
 305
 306static void
 307xfs_dquot_set_prealloc(
 308	struct xfs_dquot_pre		*pre,
 309	const struct xfs_dquot_res	*res)
 310{
 311	xfs_qcnt_t			space;
 312
 313	pre->q_prealloc_hi_wmark = res->hardlimit;
 314	pre->q_prealloc_lo_wmark = res->softlimit;
 315
 316	space = div_u64(pre->q_prealloc_hi_wmark, 100);
 317	if (!pre->q_prealloc_lo_wmark)
 318		pre->q_prealloc_lo_wmark = space * 95;
 319
 320	pre->q_low_space[XFS_QLOWSP_1_PCNT] = space;
 321	pre->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
 322	pre->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
 323}
 324
 325/*
 326 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
 327 * watermarks correspond to the soft and hard limits by default. If a soft limit
 328 * is not specified, we use 95% of the hard limit.
 329 */
 330void
 331xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
 332{
 333	xfs_dquot_set_prealloc(&dqp->q_blk_prealloc, &dqp->q_blk);
 334	xfs_dquot_set_prealloc(&dqp->q_rtb_prealloc, &dqp->q_rtb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 335}
 336
 337/*
 338 * Ensure that the given in-core dquot has a buffer on disk backing it, and
 339 * return the buffer locked and held. This is called when the bmapi finds a
 340 * hole.
 341 */
 342STATIC int
 343xfs_dquot_disk_alloc(
 344	struct xfs_dquot	*dqp,
 345	struct xfs_buf		**bpp)
 346{
 347	struct xfs_bmbt_irec	map;
 348	struct xfs_trans	*tp;
 349	struct xfs_mount	*mp = dqp->q_mount;
 350	struct xfs_buf		*bp;
 351	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
 352	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
 353	int			nmaps = 1;
 354	int			error;
 355
 356	trace_xfs_dqalloc(dqp);
 357
 358	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
 359			XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
 360	if (error)
 361		return error;
 362
 363	xfs_ilock(quotip, XFS_ILOCK_EXCL);
 364	xfs_trans_ijoin(tp, quotip, 0);
 365
 366	if (!xfs_this_quota_on(dqp->q_mount, qtype)) {
 367		/*
 368		 * Return if this type of quotas is turned off while we didn't
 369		 * have an inode lock
 370		 */
 371		error = -ESRCH;
 372		goto err_cancel;
 373	}
 374
 375	error = xfs_iext_count_extend(tp, quotip, XFS_DATA_FORK,
 376			XFS_IEXT_ADD_NOSPLIT_CNT);
 
 
 
 377	if (error)
 378		goto err_cancel;
 379
 380	/* Create the block mapping. */
 381	error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
 382			XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA, 0, &map,
 383			&nmaps);
 384	if (error)
 385		goto err_cancel;
 386
 387	ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
 
 388	ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 389	       (map.br_startblock != HOLESTARTBLOCK));
 390
 391	/*
 392	 * Keep track of the blkno to save a lookup later
 393	 */
 394	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 395
 396	/* now we can just get the buffer (there's nothing to read yet) */
 397	error = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
 398			mp->m_quotainfo->qi_dqchunklen, 0, &bp);
 399	if (error)
 400		goto err_cancel;
 401	bp->b_ops = &xfs_dquot_buf_ops;
 402
 403	/*
 404	 * Make a chunk of dquots out of this buffer and log
 405	 * the entire thing.
 406	 */
 407	xfs_qm_init_dquot_blk(tp, dqp->q_id, qtype, bp);
 408	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 409
 410	/*
 411	 * Hold the buffer and join it to the dfops so that we'll still own
 412	 * the buffer when we return to the caller.  The buffer disposal on
 413	 * error must be paid attention to very carefully, as it has been
 414	 * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
 415	 * code when allocating a new dquot record" in 2005, and the later
 416	 * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
 417	 * the buffer locked across the _defer_finish call.  We can now do
 418	 * this correctly with xfs_defer_bjoin.
 419	 *
 420	 * Above, we allocated a disk block for the dquot information and used
 421	 * get_buf to initialize the dquot. If the _defer_finish fails, the old
 422	 * transaction is gone but the new buffer is not joined or held to any
 423	 * transaction, so we must _buf_relse it.
 424	 *
 425	 * If everything succeeds, the caller of this function is returned a
 426	 * buffer that is locked and held to the transaction.  The caller
 427	 * is responsible for unlocking any buffer passed back, either
 428	 * manually or by committing the transaction.  On error, the buffer is
 429	 * released and not passed back.
 430	 *
 431	 * Keep the quota inode ILOCKed until after the transaction commit to
 432	 * maintain the atomicity of bmap/rmap updates.
 433	 */
 434	xfs_trans_bhold(tp, bp);
 435	error = xfs_trans_commit(tp);
 436	xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 437	if (error) {
 438		xfs_buf_relse(bp);
 439		return error;
 440	}
 441
 442	*bpp = bp;
 443	return 0;
 444
 445err_cancel:
 446	xfs_trans_cancel(tp);
 447	xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 448	return error;
 449}
 450
 451/*
 452 * Read in the in-core dquot's on-disk metadata and return the buffer.
 453 * Returns ENOENT to signal a hole.
 454 */
 455STATIC int
 456xfs_dquot_disk_read(
 457	struct xfs_mount	*mp,
 458	struct xfs_dquot	*dqp,
 459	struct xfs_buf		**bpp)
 460{
 461	struct xfs_bmbt_irec	map;
 462	struct xfs_buf		*bp;
 463	xfs_dqtype_t		qtype = xfs_dquot_type(dqp);
 464	struct xfs_inode	*quotip = xfs_quota_inode(mp, qtype);
 465	uint			lock_mode;
 466	int			nmaps = 1;
 467	int			error;
 468
 469	lock_mode = xfs_ilock_data_map_shared(quotip);
 470	if (!xfs_this_quota_on(mp, qtype)) {
 471		/*
 472		 * Return if this type of quotas is turned off while we
 473		 * didn't have the quota inode lock.
 474		 */
 475		xfs_iunlock(quotip, lock_mode);
 476		return -ESRCH;
 477	}
 478
 479	/*
 480	 * Find the block map; no allocations yet
 481	 */
 482	error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
 483			XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
 484	xfs_iunlock(quotip, lock_mode);
 485	if (error)
 486		return error;
 487
 488	ASSERT(nmaps == 1);
 489	ASSERT(map.br_blockcount >= 1);
 490	ASSERT(map.br_startblock != DELAYSTARTBLOCK);
 491	if (map.br_startblock == HOLESTARTBLOCK)
 492		return -ENOENT;
 493
 494	trace_xfs_dqtobp_read(dqp);
 495
 496	/*
 497	 * store the blkno etc so that we don't have to do the
 498	 * mapping all the time
 499	 */
 500	dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 501
 502	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
 503			mp->m_quotainfo->qi_dqchunklen, 0, &bp,
 504			&xfs_dquot_buf_ops);
 505	if (xfs_metadata_is_sick(error))
 506		xfs_dquot_mark_sick(dqp);
 507	if (error) {
 508		ASSERT(bp == NULL);
 509		return error;
 510	}
 511
 512	ASSERT(xfs_buf_islocked(bp));
 513	xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 514	*bpp = bp;
 515
 516	return 0;
 517}
 518
 519/* Allocate and initialize everything we need for an incore dquot. */
 520STATIC struct xfs_dquot *
 521xfs_dquot_alloc(
 522	struct xfs_mount	*mp,
 523	xfs_dqid_t		id,
 524	xfs_dqtype_t		type)
 525{
 526	struct xfs_dquot	*dqp;
 527
 528	dqp = kmem_cache_zalloc(xfs_dquot_cache, GFP_KERNEL | __GFP_NOFAIL);
 529
 530	dqp->q_type = type;
 531	dqp->q_id = id;
 532	dqp->q_mount = mp;
 533	INIT_LIST_HEAD(&dqp->q_lru);
 534	mutex_init(&dqp->q_qlock);
 535	init_waitqueue_head(&dqp->q_pinwait);
 536	dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
 537	/*
 538	 * Offset of dquot in the (fixed sized) dquot chunk.
 539	 */
 540	dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
 541			sizeof(struct xfs_dqblk);
 542
 543	/*
 544	 * Because we want to use a counting completion, complete
 545	 * the flush completion once to allow a single access to
 546	 * the flush completion without blocking.
 547	 */
 548	init_completion(&dqp->q_flush);
 549	complete(&dqp->q_flush);
 550
 551	/*
 552	 * Make sure group quotas have a different lock class than user
 553	 * quotas.
 554	 */
 555	switch (type) {
 556	case XFS_DQTYPE_USER:
 557		/* uses the default lock class */
 558		break;
 559	case XFS_DQTYPE_GROUP:
 560		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
 561		break;
 562	case XFS_DQTYPE_PROJ:
 563		lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
 564		break;
 565	default:
 566		ASSERT(0);
 567		break;
 568	}
 569
 570	xfs_qm_dquot_logitem_init(dqp);
 571
 572	XFS_STATS_INC(mp, xs_qm_dquot);
 573	return dqp;
 574}
 575
 576/* Check the ondisk dquot's id and type match what the incore dquot expects. */
 577static bool
 578xfs_dquot_check_type(
 579	struct xfs_dquot	*dqp,
 580	struct xfs_disk_dquot	*ddqp)
 581{
 582	uint8_t			ddqp_type;
 583	uint8_t			dqp_type;
 584
 585	ddqp_type = ddqp->d_type & XFS_DQTYPE_REC_MASK;
 586	dqp_type = xfs_dquot_type(dqp);
 587
 588	if (be32_to_cpu(ddqp->d_id) != dqp->q_id)
 589		return false;
 590
 591	/*
 592	 * V5 filesystems always expect an exact type match.  V4 filesystems
 593	 * expect an exact match for user dquots and for non-root group and
 594	 * project dquots.
 595	 */
 596	if (xfs_has_crc(dqp->q_mount) ||
 597	    dqp_type == XFS_DQTYPE_USER || dqp->q_id != 0)
 598		return ddqp_type == dqp_type;
 599
 600	/*
 601	 * V4 filesystems support either group or project quotas, but not both
 602	 * at the same time.  The non-user quota file can be switched between
 603	 * group and project quota uses depending on the mount options, which
 604	 * means that we can encounter the other type when we try to load quota
 605	 * defaults.  Quotacheck will soon reset the entire quota file
 606	 * (including the root dquot) anyway, but don't log scary corruption
 607	 * reports to dmesg.
 608	 */
 609	return ddqp_type == XFS_DQTYPE_GROUP || ddqp_type == XFS_DQTYPE_PROJ;
 610}
 611
 612/* Copy the in-core quota fields in from the on-disk buffer. */
 613STATIC int
 614xfs_dquot_from_disk(
 615	struct xfs_dquot	*dqp,
 616	struct xfs_buf		*bp)
 617{
 618	struct xfs_dqblk	*dqb = xfs_buf_offset(bp, dqp->q_bufoffset);
 619	struct xfs_disk_dquot	*ddqp = &dqb->dd_diskdq;
 620
 621	/*
 622	 * Ensure that we got the type and ID we were looking for.
 623	 * Everything else was checked by the dquot buffer verifier.
 624	 */
 625	if (!xfs_dquot_check_type(dqp, ddqp)) {
 626		xfs_alert_tag(bp->b_mount, XFS_PTAG_VERIFIER_ERROR,
 627			  "Metadata corruption detected at %pS, quota %u",
 628			  __this_address, dqp->q_id);
 629		xfs_alert(bp->b_mount, "Unmount and run xfs_repair");
 630		xfs_dquot_mark_sick(dqp);
 631		return -EFSCORRUPTED;
 632	}
 633
 634	/* copy everything from disk dquot to the incore dquot */
 635	dqp->q_type = ddqp->d_type;
 636	dqp->q_blk.hardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 637	dqp->q_blk.softlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 638	dqp->q_ino.hardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 639	dqp->q_ino.softlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 640	dqp->q_rtb.hardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 641	dqp->q_rtb.softlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 642
 643	dqp->q_blk.count = be64_to_cpu(ddqp->d_bcount);
 644	dqp->q_ino.count = be64_to_cpu(ddqp->d_icount);
 645	dqp->q_rtb.count = be64_to_cpu(ddqp->d_rtbcount);
 646
 647	dqp->q_blk.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_btimer);
 648	dqp->q_ino.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_itimer);
 649	dqp->q_rtb.timer = xfs_dquot_from_disk_ts(ddqp, ddqp->d_rtbtimer);
 650
 651	/*
 652	 * Reservation counters are defined as reservation plus current usage
 653	 * to avoid having to add every time.
 654	 */
 655	dqp->q_blk.reserved = dqp->q_blk.count;
 656	dqp->q_ino.reserved = dqp->q_ino.count;
 657	dqp->q_rtb.reserved = dqp->q_rtb.count;
 658
 659	/* initialize the dquot speculative prealloc thresholds */
 660	xfs_dquot_set_prealloc_limits(dqp);
 661	return 0;
 662}
 663
 664/* Copy the in-core quota fields into the on-disk buffer. */
 665void
 666xfs_dquot_to_disk(
 667	struct xfs_disk_dquot	*ddqp,
 668	struct xfs_dquot	*dqp)
 669{
 670	ddqp->d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 671	ddqp->d_version = XFS_DQUOT_VERSION;
 672	ddqp->d_type = dqp->q_type;
 673	ddqp->d_id = cpu_to_be32(dqp->q_id);
 674	ddqp->d_pad0 = 0;
 675	ddqp->d_pad = 0;
 676
 677	ddqp->d_blk_hardlimit = cpu_to_be64(dqp->q_blk.hardlimit);
 678	ddqp->d_blk_softlimit = cpu_to_be64(dqp->q_blk.softlimit);
 679	ddqp->d_ino_hardlimit = cpu_to_be64(dqp->q_ino.hardlimit);
 680	ddqp->d_ino_softlimit = cpu_to_be64(dqp->q_ino.softlimit);
 681	ddqp->d_rtb_hardlimit = cpu_to_be64(dqp->q_rtb.hardlimit);
 682	ddqp->d_rtb_softlimit = cpu_to_be64(dqp->q_rtb.softlimit);
 683
 684	ddqp->d_bcount = cpu_to_be64(dqp->q_blk.count);
 685	ddqp->d_icount = cpu_to_be64(dqp->q_ino.count);
 686	ddqp->d_rtbcount = cpu_to_be64(dqp->q_rtb.count);
 687
 688	ddqp->d_bwarns = 0;
 689	ddqp->d_iwarns = 0;
 690	ddqp->d_rtbwarns = 0;
 691
 692	ddqp->d_btimer = xfs_dquot_to_disk_ts(dqp, dqp->q_blk.timer);
 693	ddqp->d_itimer = xfs_dquot_to_disk_ts(dqp, dqp->q_ino.timer);
 694	ddqp->d_rtbtimer = xfs_dquot_to_disk_ts(dqp, dqp->q_rtb.timer);
 695}
 696
 697/*
 698 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
 699 * and release the buffer immediately.  If @can_alloc is true, fill any
 700 * holes in the on-disk metadata.
 701 */
 702static int
 703xfs_qm_dqread(
 704	struct xfs_mount	*mp,
 705	xfs_dqid_t		id,
 706	xfs_dqtype_t		type,
 707	bool			can_alloc,
 708	struct xfs_dquot	**dqpp)
 709{
 710	struct xfs_dquot	*dqp;
 711	struct xfs_buf		*bp;
 712	int			error;
 713
 714	dqp = xfs_dquot_alloc(mp, id, type);
 715	trace_xfs_dqread(dqp);
 716
 717	/* Try to read the buffer, allocating if necessary. */
 718	error = xfs_dquot_disk_read(mp, dqp, &bp);
 719	if (error == -ENOENT && can_alloc)
 720		error = xfs_dquot_disk_alloc(dqp, &bp);
 721	if (error)
 722		goto err;
 723
 724	/*
 725	 * At this point we should have a clean locked buffer.  Copy the data
 726	 * to the incore dquot and release the buffer since the incore dquot
 727	 * has its own locking protocol so we needn't tie up the buffer any
 728	 * further.
 729	 */
 730	ASSERT(xfs_buf_islocked(bp));
 731	error = xfs_dquot_from_disk(dqp, bp);
 732	xfs_buf_relse(bp);
 733	if (error)
 734		goto err;
 735
 736	*dqpp = dqp;
 737	return error;
 738
 739err:
 740	trace_xfs_dqread_fail(dqp);
 741	xfs_qm_dqdestroy(dqp);
 742	*dqpp = NULL;
 743	return error;
 744}
 745
 746/*
 747 * Advance to the next id in the current chunk, or if at the
 748 * end of the chunk, skip ahead to first id in next allocated chunk
 749 * using the SEEK_DATA interface.
 750 */
 751static int
 752xfs_dq_get_next_id(
 753	struct xfs_mount	*mp,
 754	xfs_dqtype_t		type,
 755	xfs_dqid_t		*id)
 756{
 757	struct xfs_inode	*quotip = xfs_quota_inode(mp, type);
 758	xfs_dqid_t		next_id = *id + 1; /* simple advance */
 759	uint			lock_flags;
 760	struct xfs_bmbt_irec	got;
 761	struct xfs_iext_cursor	cur;
 762	xfs_fsblock_t		start;
 763	int			error = 0;
 764
 765	/* If we'd wrap past the max ID, stop */
 766	if (next_id < *id)
 767		return -ENOENT;
 768
 769	/* If new ID is within the current chunk, advancing it sufficed */
 770	if (next_id % mp->m_quotainfo->qi_dqperchunk) {
 771		*id = next_id;
 772		return 0;
 773	}
 774
 775	/* Nope, next_id is now past the current chunk, so find the next one */
 776	start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
 777
 778	lock_flags = xfs_ilock_data_map_shared(quotip);
 779	error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
 780	if (error)
 781		return error;
 782
 783	if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
 784		/* contiguous chunk, bump startoff for the id calculation */
 785		if (got.br_startoff < start)
 786			got.br_startoff = start;
 787		*id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
 788	} else {
 789		error = -ENOENT;
 790	}
 791
 792	xfs_iunlock(quotip, lock_flags);
 793
 794	return error;
 795}
 796
 797/*
 798 * Look up the dquot in the in-core cache.  If found, the dquot is returned
 799 * locked and ready to go.
 800 */
 801static struct xfs_dquot *
 802xfs_qm_dqget_cache_lookup(
 803	struct xfs_mount	*mp,
 804	struct xfs_quotainfo	*qi,
 805	struct radix_tree_root	*tree,
 806	xfs_dqid_t		id)
 807{
 808	struct xfs_dquot	*dqp;
 809
 810restart:
 811	mutex_lock(&qi->qi_tree_lock);
 812	dqp = radix_tree_lookup(tree, id);
 813	if (!dqp) {
 814		mutex_unlock(&qi->qi_tree_lock);
 815		XFS_STATS_INC(mp, xs_qm_dqcachemisses);
 816		return NULL;
 817	}
 818
 819	xfs_dqlock(dqp);
 820	if (dqp->q_flags & XFS_DQFLAG_FREEING) {
 821		xfs_dqunlock(dqp);
 822		mutex_unlock(&qi->qi_tree_lock);
 823		trace_xfs_dqget_freeing(dqp);
 824		delay(1);
 825		goto restart;
 826	}
 827
 828	dqp->q_nrefs++;
 829	mutex_unlock(&qi->qi_tree_lock);
 830
 831	trace_xfs_dqget_hit(dqp);
 832	XFS_STATS_INC(mp, xs_qm_dqcachehits);
 833	return dqp;
 834}
 835
 836/*
 837 * Try to insert a new dquot into the in-core cache.  If an error occurs the
 838 * caller should throw away the dquot and start over.  Otherwise, the dquot
 839 * is returned locked (and held by the cache) as if there had been a cache
 840 * hit.
 841 *
 842 * The insert needs to be done under memalloc_nofs context because the radix
 843 * tree can do memory allocation during insert. The qi->qi_tree_lock is taken in
 844 * memory reclaim when freeing unused dquots, so we cannot have the radix tree
 845 * node allocation recursing into filesystem reclaim whilst we hold the
 846 * qi_tree_lock.
 847 */
 848static int
 849xfs_qm_dqget_cache_insert(
 850	struct xfs_mount	*mp,
 851	struct xfs_quotainfo	*qi,
 852	struct radix_tree_root	*tree,
 853	xfs_dqid_t		id,
 854	struct xfs_dquot	*dqp)
 855{
 856	unsigned int		nofs_flags;
 857	int			error;
 858
 859	nofs_flags = memalloc_nofs_save();
 860	mutex_lock(&qi->qi_tree_lock);
 861	error = radix_tree_insert(tree, id, dqp);
 862	if (unlikely(error)) {
 863		/* Duplicate found!  Caller must try again. */
 864		trace_xfs_dqget_dup(dqp);
 865		goto out_unlock;
 866	}
 867
 868	/* Return a locked dquot to the caller, with a reference taken. */
 869	xfs_dqlock(dqp);
 870	dqp->q_nrefs = 1;
 871	qi->qi_dquots++;
 872
 873out_unlock:
 874	mutex_unlock(&qi->qi_tree_lock);
 875	memalloc_nofs_restore(nofs_flags);
 876	return error;
 877}
 878
 879/* Check our input parameters. */
 880static int
 881xfs_qm_dqget_checks(
 882	struct xfs_mount	*mp,
 883	xfs_dqtype_t		type)
 884{
 885	switch (type) {
 886	case XFS_DQTYPE_USER:
 887		if (!XFS_IS_UQUOTA_ON(mp))
 888			return -ESRCH;
 889		return 0;
 890	case XFS_DQTYPE_GROUP:
 891		if (!XFS_IS_GQUOTA_ON(mp))
 892			return -ESRCH;
 893		return 0;
 894	case XFS_DQTYPE_PROJ:
 895		if (!XFS_IS_PQUOTA_ON(mp))
 896			return -ESRCH;
 897		return 0;
 898	default:
 899		WARN_ON_ONCE(0);
 900		return -EINVAL;
 901	}
 902}
 903
 904/*
 905 * Given the file system, id, and type (UDQUOT/GDQUOT/PDQUOT), return a
 906 * locked dquot, doing an allocation (if requested) as needed.
 907 */
 908int
 909xfs_qm_dqget(
 910	struct xfs_mount	*mp,
 911	xfs_dqid_t		id,
 912	xfs_dqtype_t		type,
 913	bool			can_alloc,
 914	struct xfs_dquot	**O_dqpp)
 915{
 916	struct xfs_quotainfo	*qi = mp->m_quotainfo;
 917	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
 918	struct xfs_dquot	*dqp;
 919	int			error;
 920
 921	error = xfs_qm_dqget_checks(mp, type);
 922	if (error)
 923		return error;
 924
 925restart:
 926	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
 927	if (dqp) {
 928		*O_dqpp = dqp;
 929		return 0;
 930	}
 931
 932	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
 933	if (error)
 934		return error;
 935
 936	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
 937	if (error) {
 938		/*
 939		 * Duplicate found. Just throw away the new dquot and start
 940		 * over.
 941		 */
 942		xfs_qm_dqdestroy(dqp);
 943		XFS_STATS_INC(mp, xs_qm_dquot_dups);
 944		goto restart;
 945	}
 946
 947	trace_xfs_dqget_miss(dqp);
 948	*O_dqpp = dqp;
 949	return 0;
 950}
 951
 952/*
 953 * Given a dquot id and type, read and initialize a dquot from the on-disk
 954 * metadata.  This function is only for use during quota initialization so
 955 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
 956 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
 957 */
 958int
 959xfs_qm_dqget_uncached(
 960	struct xfs_mount	*mp,
 961	xfs_dqid_t		id,
 962	xfs_dqtype_t		type,
 963	struct xfs_dquot	**dqpp)
 964{
 965	int			error;
 966
 967	error = xfs_qm_dqget_checks(mp, type);
 968	if (error)
 969		return error;
 970
 971	return xfs_qm_dqread(mp, id, type, 0, dqpp);
 972}
 973
 974/* Return the quota id for a given inode and type. */
 975xfs_dqid_t
 976xfs_qm_id_for_quotatype(
 977	struct xfs_inode	*ip,
 978	xfs_dqtype_t		type)
 979{
 980	switch (type) {
 981	case XFS_DQTYPE_USER:
 982		return i_uid_read(VFS_I(ip));
 983	case XFS_DQTYPE_GROUP:
 984		return i_gid_read(VFS_I(ip));
 985	case XFS_DQTYPE_PROJ:
 986		return ip->i_projid;
 987	}
 988	ASSERT(0);
 989	return 0;
 990}
 991
 992/*
 993 * Return the dquot for a given inode and type.  If @can_alloc is true, then
 994 * allocate blocks if needed.  The inode's ILOCK must be held and it must not
 995 * have already had an inode attached.
 996 */
 997int
 998xfs_qm_dqget_inode(
 999	struct xfs_inode	*ip,
1000	xfs_dqtype_t		type,
1001	bool			can_alloc,
1002	struct xfs_dquot	**O_dqpp)
1003{
1004	struct xfs_mount	*mp = ip->i_mount;
1005	struct xfs_quotainfo	*qi = mp->m_quotainfo;
1006	struct radix_tree_root	*tree = xfs_dquot_tree(qi, type);
1007	struct xfs_dquot	*dqp;
1008	xfs_dqid_t		id;
1009	int			error;
1010
1011	error = xfs_qm_dqget_checks(mp, type);
1012	if (error)
1013		return error;
1014
1015	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1016	ASSERT(xfs_inode_dquot(ip, type) == NULL);
1017	ASSERT(!xfs_is_metadir_inode(ip));
1018
1019	id = xfs_qm_id_for_quotatype(ip, type);
1020
1021restart:
1022	dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
1023	if (dqp) {
1024		*O_dqpp = dqp;
1025		return 0;
1026	}
1027
1028	/*
1029	 * Dquot cache miss. We don't want to keep the inode lock across
1030	 * a (potential) disk read. Also we don't want to deal with the lock
1031	 * ordering between quotainode and this inode. OTOH, dropping the inode
1032	 * lock here means dealing with a chown that can happen before
1033	 * we re-acquire the lock.
1034	 */
1035	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1036	error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
1037	xfs_ilock(ip, XFS_ILOCK_EXCL);
1038	if (error)
1039		return error;
1040
1041	/*
1042	 * A dquot could be attached to this inode by now, since we had
1043	 * dropped the ilock.
1044	 */
1045	if (xfs_this_quota_on(mp, type)) {
1046		struct xfs_dquot	*dqp1;
1047
1048		dqp1 = xfs_inode_dquot(ip, type);
1049		if (dqp1) {
1050			xfs_qm_dqdestroy(dqp);
1051			dqp = dqp1;
1052			xfs_dqlock(dqp);
1053			goto dqret;
1054		}
1055	} else {
1056		/* inode stays locked on return */
1057		xfs_qm_dqdestroy(dqp);
1058		return -ESRCH;
1059	}
1060
1061	error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
1062	if (error) {
1063		/*
1064		 * Duplicate found. Just throw away the new dquot and start
1065		 * over.
1066		 */
1067		xfs_qm_dqdestroy(dqp);
1068		XFS_STATS_INC(mp, xs_qm_dquot_dups);
1069		goto restart;
1070	}
1071
1072dqret:
1073	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1074	trace_xfs_dqget_miss(dqp);
1075	*O_dqpp = dqp;
1076	return 0;
1077}
1078
1079/*
1080 * Starting at @id and progressing upwards, look for an initialized incore
1081 * dquot, lock it, and return it.
1082 */
1083int
1084xfs_qm_dqget_next(
1085	struct xfs_mount	*mp,
1086	xfs_dqid_t		id,
1087	xfs_dqtype_t		type,
1088	struct xfs_dquot	**dqpp)
1089{
1090	struct xfs_dquot	*dqp;
1091	int			error = 0;
1092
1093	*dqpp = NULL;
1094	for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
1095		error = xfs_qm_dqget(mp, id, type, false, &dqp);
1096		if (error == -ENOENT)
1097			continue;
1098		else if (error != 0)
1099			break;
1100
1101		if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
1102			*dqpp = dqp;
1103			return 0;
1104		}
1105
1106		xfs_qm_dqput(dqp);
1107	}
1108
1109	return error;
1110}
1111
1112/*
1113 * Release a reference to the dquot (decrement ref-count) and unlock it.
1114 *
1115 * If there is a group quota attached to this dquot, carefully release that
1116 * too without tripping over deadlocks'n'stuff.
1117 */
1118void
1119xfs_qm_dqput(
1120	struct xfs_dquot	*dqp)
1121{
1122	ASSERT(dqp->q_nrefs > 0);
1123	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1124
1125	trace_xfs_dqput(dqp);
1126
1127	if (--dqp->q_nrefs == 0) {
1128		struct xfs_quotainfo	*qi = dqp->q_mount->m_quotainfo;
1129		trace_xfs_dqput_free(dqp);
1130
1131		if (list_lru_add_obj(&qi->qi_lru, &dqp->q_lru))
1132			XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
1133	}
1134	xfs_dqunlock(dqp);
1135}
1136
1137/*
1138 * Release a dquot. Flush it if dirty, then dqput() it.
1139 * dquot must not be locked.
1140 */
1141void
1142xfs_qm_dqrele(
1143	struct xfs_dquot	*dqp)
1144{
1145	if (!dqp)
1146		return;
1147
1148	trace_xfs_dqrele(dqp);
1149
1150	xfs_dqlock(dqp);
1151	/*
1152	 * We don't care to flush it if the dquot is dirty here.
1153	 * That will create stutters that we want to avoid.
1154	 * Instead we do a delayed write when we try to reclaim
1155	 * a dirty dquot. Also xfs_sync will take part of the burden...
1156	 */
1157	xfs_qm_dqput(dqp);
1158}
1159
1160/*
1161 * This is the dquot flushing I/O completion routine.  It is called
1162 * from interrupt level when the buffer containing the dquot is
1163 * flushed to disk.  It is responsible for removing the dquot logitem
1164 * from the AIL if it has not been re-logged, and unlocking the dquot's
1165 * flush lock. This behavior is very similar to that of inodes..
1166 */
1167static void
1168xfs_qm_dqflush_done(
1169	struct xfs_log_item	*lip)
1170{
1171	struct xfs_dq_logitem	*qlip =
1172			container_of(lip, struct xfs_dq_logitem, qli_item);
1173	struct xfs_dquot	*dqp = qlip->qli_dquot;
1174	struct xfs_ail		*ailp = lip->li_ailp;
1175	struct xfs_buf		*bp = NULL;
1176	xfs_lsn_t		tail_lsn;
1177
1178	/*
1179	 * We only want to pull the item from the AIL if its
1180	 * location in the log has not changed since we started the flush.
1181	 * Thus, we only bother if the dquot's lsn has
1182	 * not changed. First we check the lsn outside the lock
1183	 * since it's cheaper, and then we recheck while
1184	 * holding the lock before removing the dquot from the AIL.
1185	 */
1186	if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1187	    (lip->li_lsn == qlip->qli_flush_lsn ||
1188	     test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1189
1190		spin_lock(&ailp->ail_lock);
1191		xfs_clear_li_failed(lip);
1192		if (lip->li_lsn == qlip->qli_flush_lsn) {
1193			/* xfs_ail_update_finish() drops the AIL lock */
1194			tail_lsn = xfs_ail_delete_one(ailp, lip);
1195			xfs_ail_update_finish(ailp, tail_lsn);
1196		} else {
1197			spin_unlock(&ailp->ail_lock);
1198		}
1199	}
1200
1201	/*
1202	 * If this dquot hasn't been dirtied since initiating the last dqflush,
1203	 * release the buffer reference.  We already unlinked this dquot item
1204	 * from the buffer.
1205	 */
1206	spin_lock(&qlip->qli_lock);
1207	if (!qlip->qli_dirty) {
1208		bp = lip->li_buf;
1209		lip->li_buf = NULL;
1210	}
1211	spin_unlock(&qlip->qli_lock);
1212	if (bp)
1213		xfs_buf_rele(bp);
1214
1215	/*
1216	 * Release the dq's flush lock since we're done with it.
1217	 */
1218	xfs_dqfunlock(dqp);
1219}
1220
1221void
1222xfs_buf_dquot_iodone(
1223	struct xfs_buf		*bp)
1224{
1225	struct xfs_log_item	*lip, *n;
1226
1227	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
1228		list_del_init(&lip->li_bio_list);
1229		xfs_qm_dqflush_done(lip);
1230	}
1231}
1232
1233void
1234xfs_buf_dquot_io_fail(
1235	struct xfs_buf		*bp)
1236{
1237	struct xfs_log_item	*lip;
1238
1239	spin_lock(&bp->b_mount->m_ail->ail_lock);
1240	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1241		set_bit(XFS_LI_FAILED, &lip->li_flags);
1242	spin_unlock(&bp->b_mount->m_ail->ail_lock);
1243}
1244
1245/* Check incore dquot for errors before we flush. */
1246static xfs_failaddr_t
1247xfs_qm_dqflush_check(
1248	struct xfs_dquot	*dqp)
1249{
1250	xfs_dqtype_t		type = xfs_dquot_type(dqp);
1251
1252	if (type != XFS_DQTYPE_USER &&
1253	    type != XFS_DQTYPE_GROUP &&
1254	    type != XFS_DQTYPE_PROJ)
1255		return __this_address;
1256
1257	if (dqp->q_id == 0)
1258		return NULL;
1259
1260	if (dqp->q_blk.softlimit && dqp->q_blk.count > dqp->q_blk.softlimit &&
1261	    !dqp->q_blk.timer)
1262		return __this_address;
1263
1264	if (dqp->q_ino.softlimit && dqp->q_ino.count > dqp->q_ino.softlimit &&
1265	    !dqp->q_ino.timer)
1266		return __this_address;
1267
1268	if (dqp->q_rtb.softlimit && dqp->q_rtb.count > dqp->q_rtb.softlimit &&
1269	    !dqp->q_rtb.timer)
1270		return __this_address;
1271
1272	/* bigtime flag should never be set on root dquots */
1273	if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
1274		if (!xfs_has_bigtime(dqp->q_mount))
1275			return __this_address;
1276		if (dqp->q_id == 0)
1277			return __this_address;
1278	}
1279
1280	return NULL;
1281}
1282
1283/*
1284 * Get the buffer containing the on-disk dquot.
1285 *
1286 * Requires dquot flush lock, will clear the dirty flag, delete the quota log
1287 * item from the AIL, and shut down the system if something goes wrong.
1288 */
1289static int
1290xfs_dquot_read_buf(
1291	struct xfs_trans	*tp,
1292	struct xfs_dquot	*dqp,
1293	struct xfs_buf		**bpp)
1294{
1295	struct xfs_mount	*mp = dqp->q_mount;
1296	struct xfs_buf		*bp = NULL;
1297	int			error;
1298
1299	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, dqp->q_blkno,
1300				   mp->m_quotainfo->qi_dqchunklen, 0,
1301				   &bp, &xfs_dquot_buf_ops);
1302	if (xfs_metadata_is_sick(error))
1303		xfs_dquot_mark_sick(dqp);
1304	if (error)
1305		goto out_abort;
1306
1307	*bpp = bp;
1308	return 0;
1309
1310out_abort:
1311	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1312	xfs_trans_ail_delete(&dqp->q_logitem.qli_item, 0);
1313	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1314	return error;
1315}
1316
1317/*
1318 * Attach a dquot buffer to this dquot to avoid allocating a buffer during a
1319 * dqflush, since dqflush can be called from reclaim context.  Caller must hold
1320 * the dqlock.
1321 */
1322int
1323xfs_dquot_attach_buf(
1324	struct xfs_trans	*tp,
1325	struct xfs_dquot	*dqp)
1326{
1327	struct xfs_dq_logitem	*qlip = &dqp->q_logitem;
1328	struct xfs_log_item	*lip = &qlip->qli_item;
1329	int			error;
1330
1331	spin_lock(&qlip->qli_lock);
1332	if (!lip->li_buf) {
1333		struct xfs_buf	*bp = NULL;
1334
1335		spin_unlock(&qlip->qli_lock);
1336		error = xfs_dquot_read_buf(tp, dqp, &bp);
1337		if (error)
1338			return error;
1339
1340		/*
1341		 * Hold the dquot buffer so that we retain our ref to it after
1342		 * detaching it from the transaction, then give that ref to the
1343		 * dquot log item so that the AIL does not have to read the
1344		 * dquot buffer to push this item.
1345		 */
1346		xfs_buf_hold(bp);
1347		xfs_trans_brelse(tp, bp);
1348
1349		spin_lock(&qlip->qli_lock);
1350		lip->li_buf = bp;
1351	}
1352	qlip->qli_dirty = true;
1353	spin_unlock(&qlip->qli_lock);
1354
1355	return 0;
1356}
1357
1358/*
1359 * Get a new reference the dquot buffer attached to this dquot for a dqflush
1360 * operation.
1361 *
1362 * Returns 0 and a NULL bp if none was attached to the dquot; 0 and a locked
1363 * bp; or -EAGAIN if the buffer could not be locked.
1364 */
1365int
1366xfs_dquot_use_attached_buf(
1367	struct xfs_dquot	*dqp,
1368	struct xfs_buf		**bpp)
1369{
1370	struct xfs_buf		*bp = dqp->q_logitem.qli_item.li_buf;
1371
1372	/*
1373	 * A NULL buffer can happen if the dquot dirty flag was set but the
1374	 * filesystem shut down before transaction commit happened.  In that
1375	 * case we're not going to flush anyway.
1376	 */
1377	if (!bp) {
1378		ASSERT(xfs_is_shutdown(dqp->q_mount));
1379
1380		*bpp = NULL;
1381		return 0;
1382	}
1383
1384	if (!xfs_buf_trylock(bp))
1385		return -EAGAIN;
1386
1387	xfs_buf_hold(bp);
1388	*bpp = bp;
1389	return 0;
1390}
1391
1392/*
1393 * Write a modified dquot to disk.
1394 * The dquot must be locked and the flush lock too taken by caller.
1395 * The flush lock will not be unlocked until the dquot reaches the disk,
1396 * but the dquot is free to be unlocked and modified by the caller
1397 * in the interim. Dquot is still locked on return. This behavior is
1398 * identical to that of inodes.
1399 */
1400int
1401xfs_qm_dqflush(
1402	struct xfs_dquot	*dqp,
1403	struct xfs_buf		*bp)
1404{
1405	struct xfs_mount	*mp = dqp->q_mount;
1406	struct xfs_dq_logitem	*qlip = &dqp->q_logitem;
1407	struct xfs_log_item	*lip = &qlip->qli_item;
1408	struct xfs_dqblk	*dqblk;
1409	xfs_failaddr_t		fa;
1410	int			error;
1411
1412	ASSERT(XFS_DQ_IS_LOCKED(dqp));
1413	ASSERT(!completion_done(&dqp->q_flush));
1414
1415	trace_xfs_dqflush(dqp);
1416
 
 
1417	xfs_qm_dqunpin_wait(dqp);
1418
 
 
 
 
 
 
 
 
 
 
 
 
 
1419	fa = xfs_qm_dqflush_check(dqp);
1420	if (fa) {
1421		xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1422				dqp->q_id, fa);
 
1423		xfs_dquot_mark_sick(dqp);
1424		error = -EFSCORRUPTED;
1425		goto out_abort;
1426	}
1427
1428	/* Flush the incore dquot to the ondisk buffer. */
1429	dqblk = xfs_buf_offset(bp, dqp->q_bufoffset);
1430	xfs_dquot_to_disk(&dqblk->dd_diskdq, dqp);
1431
1432	/*
1433	 * Clear the dirty field and remember the flush lsn for later use.
1434	 */
1435	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1436
1437	/*
1438	 * We hold the dquot lock, so nobody can dirty it while we're
1439	 * scheduling the write out.  Clear the dirty-since-flush flag.
1440	 */
1441	spin_lock(&qlip->qli_lock);
1442	qlip->qli_dirty = false;
1443	spin_unlock(&qlip->qli_lock);
1444
1445	xfs_trans_ail_copy_lsn(mp->m_ail, &qlip->qli_flush_lsn, &lip->li_lsn);
1446
1447	/*
1448	 * copy the lsn into the on-disk dquot now while we have the in memory
1449	 * dquot here. This can't be done later in the write verifier as we
1450	 * can't get access to the log item at that point in time.
1451	 *
1452	 * We also calculate the CRC here so that the on-disk dquot in the
1453	 * buffer always has a valid CRC. This ensures there is no possibility
1454	 * of a dquot without an up-to-date CRC getting to disk.
1455	 */
1456	if (xfs_has_crc(mp)) {
1457		dqblk->dd_lsn = cpu_to_be64(lip->li_lsn);
1458		xfs_update_cksum((char *)dqblk, sizeof(struct xfs_dqblk),
1459				 XFS_DQUOT_CRC_OFF);
1460	}
1461
1462	/*
1463	 * Attach the dquot to the buffer so that we can remove this dquot from
1464	 * the AIL and release the flush lock once the dquot is synced to disk.
1465	 */
1466	bp->b_flags |= _XBF_DQUOTS;
1467	list_add_tail(&lip->li_bio_list, &bp->b_li_list);
1468
1469	/*
1470	 * If the buffer is pinned then push on the log so we won't
1471	 * get stuck waiting in the write for too long.
1472	 */
1473	if (xfs_buf_ispinned(bp)) {
1474		trace_xfs_dqflush_force(dqp);
1475		xfs_log_force(mp, 0);
1476	}
1477
1478	trace_xfs_dqflush_done(dqp);
 
1479	return 0;
1480
1481out_abort:
1482	dqp->q_flags &= ~XFS_DQFLAG_DIRTY;
1483	xfs_trans_ail_delete(lip, 0);
1484	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 
1485	xfs_dqfunlock(dqp);
1486	return error;
1487}
1488
1489/*
1490 * Lock two xfs_dquot structures.
1491 *
1492 * To avoid deadlocks we always lock the quota structure with
1493 * the lowerd id first.
1494 */
1495void
1496xfs_dqlock2(
1497	struct xfs_dquot	*d1,
1498	struct xfs_dquot	*d2)
1499{
1500	if (d1 && d2) {
1501		ASSERT(d1 != d2);
1502		if (d1->q_id > d2->q_id) {
1503			mutex_lock(&d2->q_qlock);
1504			mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1505		} else {
1506			mutex_lock(&d1->q_qlock);
1507			mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1508		}
1509	} else if (d1) {
1510		mutex_lock(&d1->q_qlock);
1511	} else if (d2) {
1512		mutex_lock(&d2->q_qlock);
1513	}
1514}
1515
1516static int
1517xfs_dqtrx_cmp(
1518	const void		*a,
1519	const void		*b)
1520{
1521	const struct xfs_dqtrx	*qa = a;
1522	const struct xfs_dqtrx	*qb = b;
1523
1524	if (qa->qt_dquot->q_id > qb->qt_dquot->q_id)
1525		return 1;
1526	if (qa->qt_dquot->q_id < qb->qt_dquot->q_id)
1527		return -1;
1528	return 0;
1529}
1530
1531void
1532xfs_dqlockn(
1533	struct xfs_dqtrx	*q)
1534{
1535	unsigned int		i;
1536
1537	BUILD_BUG_ON(XFS_QM_TRANS_MAXDQS > MAX_LOCKDEP_SUBCLASSES);
1538
1539	/* Sort in order of dquot id, do not allow duplicates */
1540	for (i = 0; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++) {
1541		unsigned int	j;
1542
1543		for (j = 0; j < i; j++)
1544			ASSERT(q[i].qt_dquot != q[j].qt_dquot);
1545	}
1546	if (i == 0)
1547		return;
1548
1549	sort(q, i, sizeof(struct xfs_dqtrx), xfs_dqtrx_cmp, NULL);
1550
1551	mutex_lock(&q[0].qt_dquot->q_qlock);
1552	for (i = 1; i < XFS_QM_TRANS_MAXDQS && q[i].qt_dquot != NULL; i++)
1553		mutex_lock_nested(&q[i].qt_dquot->q_qlock,
1554				XFS_QLOCK_NESTED + i - 1);
1555}
1556
1557int __init
1558xfs_qm_init(void)
1559{
1560	xfs_dquot_cache = kmem_cache_create("xfs_dquot",
1561					  sizeof(struct xfs_dquot),
1562					  0, 0, NULL);
1563	if (!xfs_dquot_cache)
1564		goto out;
1565
1566	xfs_dqtrx_cache = kmem_cache_create("xfs_dqtrx",
1567					     sizeof(struct xfs_dquot_acct),
1568					     0, 0, NULL);
1569	if (!xfs_dqtrx_cache)
1570		goto out_free_dquot_cache;
1571
1572	return 0;
1573
1574out_free_dquot_cache:
1575	kmem_cache_destroy(xfs_dquot_cache);
1576out:
1577	return -ENOMEM;
1578}
1579
1580void
1581xfs_qm_exit(void)
1582{
1583	kmem_cache_destroy(xfs_dqtrx_cache);
1584	kmem_cache_destroy(xfs_dquot_cache);
1585}