Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4 * All Rights Reserved.
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_shared.h"
   9#include "xfs_format.h"
  10#include "xfs_log_format.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_mount.h"
  13#include "xfs_inode.h"
  14#include "xfs_trans.h"
  15#include "xfs_trans_priv.h"
  16#include "xfs_inode_item.h"
  17#include "xfs_quota.h"
  18#include "xfs_trace.h"
  19#include "xfs_icache.h"
  20#include "xfs_bmap_util.h"
  21#include "xfs_dquot_item.h"
  22#include "xfs_dquot.h"
  23#include "xfs_reflink.h"
  24#include "xfs_ialloc.h"
  25#include "xfs_ag.h"
  26
  27#include <linux/iversion.h>
  28
  29/* Radix tree tags for incore inode tree. */
  30
  31/* inode is to be reclaimed */
  32#define XFS_ICI_RECLAIM_TAG	0
  33/* Inode has speculative preallocations (posteof or cow) to clean. */
  34#define XFS_ICI_BLOCKGC_TAG	1
  35
  36/*
  37 * The goal for walking incore inodes.  These can correspond with incore inode
  38 * radix tree tags when convenient.  Avoid existing XFS_IWALK namespace.
  39 */
  40enum xfs_icwalk_goal {
  41	/* Goals that are not related to tags; these must be < 0. */
  42	XFS_ICWALK_DQRELE	= -1,
  43
  44	/* Goals directly associated with tagged inodes. */
  45	XFS_ICWALK_BLOCKGC	= XFS_ICI_BLOCKGC_TAG,
  46	XFS_ICWALK_RECLAIM	= XFS_ICI_RECLAIM_TAG,
  47};
  48
  49#define XFS_ICWALK_NULL_TAG	(-1U)
  50
  51/* Compute the inode radix tree tag for this goal. */
  52static inline unsigned int
  53xfs_icwalk_tag(enum xfs_icwalk_goal goal)
  54{
  55	return goal < 0 ? XFS_ICWALK_NULL_TAG : goal;
  56}
  57
  58static int xfs_icwalk(struct xfs_mount *mp,
  59		enum xfs_icwalk_goal goal, struct xfs_icwalk *icw);
  60static int xfs_icwalk_ag(struct xfs_perag *pag,
  61		enum xfs_icwalk_goal goal, struct xfs_icwalk *icw);
  62
  63/*
  64 * Private inode cache walk flags for struct xfs_icwalk.  Must not
  65 * coincide with XFS_ICWALK_FLAGS_VALID.
  66 */
  67#define XFS_ICWALK_FLAG_DROP_UDQUOT	(1U << 31)
  68#define XFS_ICWALK_FLAG_DROP_GDQUOT	(1U << 30)
  69#define XFS_ICWALK_FLAG_DROP_PDQUOT	(1U << 29)
  70
  71/* Stop scanning after icw_scan_limit inodes. */
  72#define XFS_ICWALK_FLAG_SCAN_LIMIT	(1U << 28)
  73
  74#define XFS_ICWALK_FLAG_RECLAIM_SICK	(1U << 27)
  75#define XFS_ICWALK_FLAG_UNION		(1U << 26) /* union filter algorithm */
  76
  77#define XFS_ICWALK_PRIVATE_FLAGS	(XFS_ICWALK_FLAG_DROP_UDQUOT | \
  78					 XFS_ICWALK_FLAG_DROP_GDQUOT | \
  79					 XFS_ICWALK_FLAG_DROP_PDQUOT | \
  80					 XFS_ICWALK_FLAG_SCAN_LIMIT | \
  81					 XFS_ICWALK_FLAG_RECLAIM_SICK | \
  82					 XFS_ICWALK_FLAG_UNION)
  83
  84/*
  85 * Allocate and initialise an xfs_inode.
  86 */
  87struct xfs_inode *
  88xfs_inode_alloc(
  89	struct xfs_mount	*mp,
  90	xfs_ino_t		ino)
  91{
  92	struct xfs_inode	*ip;
  93
  94	/*
  95	 * XXX: If this didn't occur in transactions, we could drop GFP_NOFAIL
  96	 * and return NULL here on ENOMEM.
  97	 */
  98	ip = kmem_cache_alloc(xfs_inode_zone, GFP_KERNEL | __GFP_NOFAIL);
  99
 100	if (inode_init_always(mp->m_super, VFS_I(ip))) {
 101		kmem_cache_free(xfs_inode_zone, ip);
 102		return NULL;
 103	}
 104
 105	/* VFS doesn't initialise i_mode! */
 106	VFS_I(ip)->i_mode = 0;
 107
 108	XFS_STATS_INC(mp, vn_active);
 109	ASSERT(atomic_read(&ip->i_pincount) == 0);
 110	ASSERT(ip->i_ino == 0);
 111
 112	/* initialise the xfs inode */
 113	ip->i_ino = ino;
 114	ip->i_mount = mp;
 115	memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
 116	ip->i_afp = NULL;
 117	ip->i_cowfp = NULL;
 118	memset(&ip->i_df, 0, sizeof(ip->i_df));
 119	ip->i_flags = 0;
 120	ip->i_delayed_blks = 0;
 121	ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
 122	ip->i_nblocks = 0;
 123	ip->i_forkoff = 0;
 124	ip->i_sick = 0;
 125	ip->i_checked = 0;
 126	INIT_WORK(&ip->i_ioend_work, xfs_end_io);
 127	INIT_LIST_HEAD(&ip->i_ioend_list);
 128	spin_lock_init(&ip->i_ioend_lock);
 129
 130	return ip;
 131}
 132
 133STATIC void
 134xfs_inode_free_callback(
 135	struct rcu_head		*head)
 136{
 137	struct inode		*inode = container_of(head, struct inode, i_rcu);
 138	struct xfs_inode	*ip = XFS_I(inode);
 139
 140	switch (VFS_I(ip)->i_mode & S_IFMT) {
 141	case S_IFREG:
 142	case S_IFDIR:
 143	case S_IFLNK:
 144		xfs_idestroy_fork(&ip->i_df);
 145		break;
 146	}
 147
 148	if (ip->i_afp) {
 149		xfs_idestroy_fork(ip->i_afp);
 150		kmem_cache_free(xfs_ifork_zone, ip->i_afp);
 151	}
 152	if (ip->i_cowfp) {
 153		xfs_idestroy_fork(ip->i_cowfp);
 154		kmem_cache_free(xfs_ifork_zone, ip->i_cowfp);
 155	}
 156	if (ip->i_itemp) {
 157		ASSERT(!test_bit(XFS_LI_IN_AIL,
 158				 &ip->i_itemp->ili_item.li_flags));
 159		xfs_inode_item_destroy(ip);
 160		ip->i_itemp = NULL;
 161	}
 162
 163	kmem_cache_free(xfs_inode_zone, ip);
 164}
 165
 166static void
 167__xfs_inode_free(
 168	struct xfs_inode	*ip)
 169{
 170	/* asserts to verify all state is correct here */
 171	ASSERT(atomic_read(&ip->i_pincount) == 0);
 172	ASSERT(!ip->i_itemp || list_empty(&ip->i_itemp->ili_item.li_bio_list));
 173	XFS_STATS_DEC(ip->i_mount, vn_active);
 174
 175	call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
 176}
 177
 178void
 179xfs_inode_free(
 180	struct xfs_inode	*ip)
 181{
 182	ASSERT(!xfs_iflags_test(ip, XFS_IFLUSHING));
 183
 184	/*
 185	 * Because we use RCU freeing we need to ensure the inode always
 186	 * appears to be reclaimed with an invalid inode number when in the
 187	 * free state. The ip->i_flags_lock provides the barrier against lookup
 188	 * races.
 189	 */
 190	spin_lock(&ip->i_flags_lock);
 191	ip->i_flags = XFS_IRECLAIM;
 192	ip->i_ino = 0;
 193	spin_unlock(&ip->i_flags_lock);
 194
 195	__xfs_inode_free(ip);
 196}
 197
 198/*
 199 * Queue background inode reclaim work if there are reclaimable inodes and there
 200 * isn't reclaim work already scheduled or in progress.
 201 */
 202static void
 203xfs_reclaim_work_queue(
 204	struct xfs_mount        *mp)
 205{
 206
 207	rcu_read_lock();
 208	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
 209		queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
 210			msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
 211	}
 212	rcu_read_unlock();
 213}
 214
 215/*
 216 * Background scanning to trim preallocated space. This is queued based on the
 217 * 'speculative_prealloc_lifetime' tunable (5m by default).
 218 */
 219static inline void
 220xfs_blockgc_queue(
 221	struct xfs_perag	*pag)
 222{
 223	rcu_read_lock();
 224	if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG))
 225		queue_delayed_work(pag->pag_mount->m_gc_workqueue,
 226				   &pag->pag_blockgc_work,
 227				   msecs_to_jiffies(xfs_blockgc_secs * 1000));
 228	rcu_read_unlock();
 229}
 230
 231/* Set a tag on both the AG incore inode tree and the AG radix tree. */
 232static void
 233xfs_perag_set_inode_tag(
 234	struct xfs_perag	*pag,
 235	xfs_agino_t		agino,
 236	unsigned int		tag)
 237{
 238	struct xfs_mount	*mp = pag->pag_mount;
 239	bool			was_tagged;
 240
 241	lockdep_assert_held(&pag->pag_ici_lock);
 242
 243	was_tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
 244	radix_tree_tag_set(&pag->pag_ici_root, agino, tag);
 245
 246	if (tag == XFS_ICI_RECLAIM_TAG)
 247		pag->pag_ici_reclaimable++;
 248
 249	if (was_tagged)
 250		return;
 251
 252	/* propagate the tag up into the perag radix tree */
 253	spin_lock(&mp->m_perag_lock);
 254	radix_tree_tag_set(&mp->m_perag_tree, pag->pag_agno, tag);
 255	spin_unlock(&mp->m_perag_lock);
 256
 257	/* start background work */
 258	switch (tag) {
 259	case XFS_ICI_RECLAIM_TAG:
 260		xfs_reclaim_work_queue(mp);
 261		break;
 262	case XFS_ICI_BLOCKGC_TAG:
 263		xfs_blockgc_queue(pag);
 264		break;
 265	}
 266
 267	trace_xfs_perag_set_inode_tag(mp, pag->pag_agno, tag, _RET_IP_);
 268}
 269
 270/* Clear a tag on both the AG incore inode tree and the AG radix tree. */
 271static void
 272xfs_perag_clear_inode_tag(
 273	struct xfs_perag	*pag,
 274	xfs_agino_t		agino,
 275	unsigned int		tag)
 276{
 277	struct xfs_mount	*mp = pag->pag_mount;
 278
 279	lockdep_assert_held(&pag->pag_ici_lock);
 280
 281	/*
 282	 * Reclaim can signal (with a null agino) that it cleared its own tag
 283	 * by removing the inode from the radix tree.
 284	 */
 285	if (agino != NULLAGINO)
 286		radix_tree_tag_clear(&pag->pag_ici_root, agino, tag);
 287	else
 288		ASSERT(tag == XFS_ICI_RECLAIM_TAG);
 289
 290	if (tag == XFS_ICI_RECLAIM_TAG)
 291		pag->pag_ici_reclaimable--;
 292
 293	if (radix_tree_tagged(&pag->pag_ici_root, tag))
 294		return;
 295
 296	/* clear the tag from the perag radix tree */
 297	spin_lock(&mp->m_perag_lock);
 298	radix_tree_tag_clear(&mp->m_perag_tree, pag->pag_agno, tag);
 299	spin_unlock(&mp->m_perag_lock);
 300
 301	trace_xfs_perag_clear_inode_tag(mp, pag->pag_agno, tag, _RET_IP_);
 302}
 303
 304/*
 305 * We set the inode flag atomically with the radix tree tag.
 306 * Once we get tag lookups on the radix tree, this inode flag
 307 * can go away.
 308 */
 309void
 310xfs_inode_mark_reclaimable(
 311	struct xfs_inode	*ip)
 312{
 313	struct xfs_mount	*mp = ip->i_mount;
 314	struct xfs_perag	*pag;
 315
 316	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 317	spin_lock(&pag->pag_ici_lock);
 318	spin_lock(&ip->i_flags_lock);
 319
 320	xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
 321			XFS_ICI_RECLAIM_TAG);
 322	__xfs_iflags_set(ip, XFS_IRECLAIMABLE);
 323
 324	spin_unlock(&ip->i_flags_lock);
 325	spin_unlock(&pag->pag_ici_lock);
 326	xfs_perag_put(pag);
 327}
 328
 329static inline void
 330xfs_inew_wait(
 331	struct xfs_inode	*ip)
 332{
 333	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_INEW_BIT);
 334	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_INEW_BIT);
 335
 336	do {
 337		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
 338		if (!xfs_iflags_test(ip, XFS_INEW))
 339			break;
 340		schedule();
 341	} while (true);
 342	finish_wait(wq, &wait.wq_entry);
 343}
 344
 345/*
 346 * When we recycle a reclaimable inode, we need to re-initialise the VFS inode
 347 * part of the structure. This is made more complex by the fact we store
 348 * information about the on-disk values in the VFS inode and so we can't just
 349 * overwrite the values unconditionally. Hence we save the parameters we
 350 * need to retain across reinitialisation, and rewrite them into the VFS inode
 351 * after reinitialisation even if it fails.
 352 */
 353static int
 354xfs_reinit_inode(
 355	struct xfs_mount	*mp,
 356	struct inode		*inode)
 357{
 358	int			error;
 359	uint32_t		nlink = inode->i_nlink;
 360	uint32_t		generation = inode->i_generation;
 361	uint64_t		version = inode_peek_iversion(inode);
 362	umode_t			mode = inode->i_mode;
 363	dev_t			dev = inode->i_rdev;
 364	kuid_t			uid = inode->i_uid;
 365	kgid_t			gid = inode->i_gid;
 366
 367	error = inode_init_always(mp->m_super, inode);
 368
 369	set_nlink(inode, nlink);
 370	inode->i_generation = generation;
 371	inode_set_iversion_queried(inode, version);
 372	inode->i_mode = mode;
 373	inode->i_rdev = dev;
 374	inode->i_uid = uid;
 375	inode->i_gid = gid;
 376	return error;
 377}
 378
 379/*
 380 * Carefully nudge an inode whose VFS state has been torn down back into a
 381 * usable state.  Drops the i_flags_lock and the rcu read lock.
 382 */
 383static int
 384xfs_iget_recycle(
 385	struct xfs_perag	*pag,
 386	struct xfs_inode	*ip) __releases(&ip->i_flags_lock)
 387{
 388	struct xfs_mount	*mp = ip->i_mount;
 389	struct inode		*inode = VFS_I(ip);
 390	int			error;
 391
 392	trace_xfs_iget_recycle(ip);
 393
 394	/*
 395	 * We need to make it look like the inode is being reclaimed to prevent
 396	 * the actual reclaim workers from stomping over us while we recycle
 397	 * the inode.  We can't clear the radix tree tag yet as it requires
 398	 * pag_ici_lock to be held exclusive.
 399	 */
 400	ip->i_flags |= XFS_IRECLAIM;
 401
 402	spin_unlock(&ip->i_flags_lock);
 403	rcu_read_unlock();
 404
 405	ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 406	error = xfs_reinit_inode(mp, inode);
 407	if (error) {
 408		bool	wake;
 409
 410		/*
 411		 * Re-initializing the inode failed, and we are in deep
 412		 * trouble.  Try to re-add it to the reclaim list.
 413		 */
 414		rcu_read_lock();
 415		spin_lock(&ip->i_flags_lock);
 416		wake = !!__xfs_iflags_test(ip, XFS_INEW);
 417		ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
 418		if (wake)
 419			wake_up_bit(&ip->i_flags, __XFS_INEW_BIT);
 420		ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
 421		spin_unlock(&ip->i_flags_lock);
 422		rcu_read_unlock();
 423
 424		trace_xfs_iget_recycle_fail(ip);
 425		return error;
 426	}
 427
 428	spin_lock(&pag->pag_ici_lock);
 429	spin_lock(&ip->i_flags_lock);
 430
 431	/*
 432	 * Clear the per-lifetime state in the inode as we are now effectively
 433	 * a new inode and need to return to the initial state before reuse
 434	 * occurs.
 435	 */
 436	ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
 437	ip->i_flags |= XFS_INEW;
 438	xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
 439			XFS_ICI_RECLAIM_TAG);
 440	inode->i_state = I_NEW;
 441	spin_unlock(&ip->i_flags_lock);
 442	spin_unlock(&pag->pag_ici_lock);
 443
 444	return 0;
 445}
 446
 447/*
 448 * If we are allocating a new inode, then check what was returned is
 449 * actually a free, empty inode. If we are not allocating an inode,
 450 * then check we didn't find a free inode.
 451 *
 452 * Returns:
 453 *	0		if the inode free state matches the lookup context
 454 *	-ENOENT		if the inode is free and we are not allocating
 455 *	-EFSCORRUPTED	if there is any state mismatch at all
 456 */
 457static int
 458xfs_iget_check_free_state(
 459	struct xfs_inode	*ip,
 460	int			flags)
 461{
 462	if (flags & XFS_IGET_CREATE) {
 463		/* should be a free inode */
 464		if (VFS_I(ip)->i_mode != 0) {
 465			xfs_warn(ip->i_mount,
 466"Corruption detected! Free inode 0x%llx not marked free! (mode 0x%x)",
 467				ip->i_ino, VFS_I(ip)->i_mode);
 468			return -EFSCORRUPTED;
 469		}
 470
 471		if (ip->i_nblocks != 0) {
 472			xfs_warn(ip->i_mount,
 473"Corruption detected! Free inode 0x%llx has blocks allocated!",
 474				ip->i_ino);
 475			return -EFSCORRUPTED;
 476		}
 477		return 0;
 478	}
 479
 480	/* should be an allocated inode */
 481	if (VFS_I(ip)->i_mode == 0)
 482		return -ENOENT;
 483
 484	return 0;
 485}
 486
 487/*
 488 * Check the validity of the inode we just found it the cache
 489 */
 490static int
 491xfs_iget_cache_hit(
 492	struct xfs_perag	*pag,
 493	struct xfs_inode	*ip,
 494	xfs_ino_t		ino,
 495	int			flags,
 496	int			lock_flags) __releases(RCU)
 497{
 498	struct inode		*inode = VFS_I(ip);
 499	struct xfs_mount	*mp = ip->i_mount;
 500	int			error;
 501
 502	/*
 503	 * check for re-use of an inode within an RCU grace period due to the
 504	 * radix tree nodes not being updated yet. We monitor for this by
 505	 * setting the inode number to zero before freeing the inode structure.
 506	 * If the inode has been reallocated and set up, then the inode number
 507	 * will not match, so check for that, too.
 508	 */
 509	spin_lock(&ip->i_flags_lock);
 510	if (ip->i_ino != ino)
 511		goto out_skip;
 512
 513	/*
 514	 * If we are racing with another cache hit that is currently
 515	 * instantiating this inode or currently recycling it out of
 516	 * reclaimable state, wait for the initialisation to complete
 517	 * before continuing.
 518	 *
 519	 * XXX(hch): eventually we should do something equivalent to
 520	 *	     wait_on_inode to wait for these flags to be cleared
 521	 *	     instead of polling for it.
 522	 */
 523	if (ip->i_flags & (XFS_INEW | XFS_IRECLAIM))
 524		goto out_skip;
 525
 526	/*
 527	 * Check the inode free state is valid. This also detects lookup
 528	 * racing with unlinks.
 529	 */
 530	error = xfs_iget_check_free_state(ip, flags);
 531	if (error)
 532		goto out_error;
 533
 534	/* Skip inodes that have no vfs state. */
 535	if ((flags & XFS_IGET_INCORE) &&
 536	    (ip->i_flags & XFS_IRECLAIMABLE))
 537		goto out_skip;
 538
 539	/* The inode fits the selection criteria; process it. */
 540	if (ip->i_flags & XFS_IRECLAIMABLE) {
 541		/* Drops i_flags_lock and RCU read lock. */
 542		error = xfs_iget_recycle(pag, ip);
 543		if (error)
 544			return error;
 545	} else {
 546		/* If the VFS inode is being torn down, pause and try again. */
 547		if (!igrab(inode))
 548			goto out_skip;
 549
 550		/* We've got a live one. */
 551		spin_unlock(&ip->i_flags_lock);
 552		rcu_read_unlock();
 553		trace_xfs_iget_hit(ip);
 554	}
 555
 556	if (lock_flags != 0)
 557		xfs_ilock(ip, lock_flags);
 558
 559	if (!(flags & XFS_IGET_INCORE))
 560		xfs_iflags_clear(ip, XFS_ISTALE);
 561	XFS_STATS_INC(mp, xs_ig_found);
 562
 563	return 0;
 564
 565out_skip:
 566	trace_xfs_iget_skip(ip);
 567	XFS_STATS_INC(mp, xs_ig_frecycle);
 568	error = -EAGAIN;
 569out_error:
 570	spin_unlock(&ip->i_flags_lock);
 571	rcu_read_unlock();
 572	return error;
 573}
 574
 575static int
 576xfs_iget_cache_miss(
 577	struct xfs_mount	*mp,
 578	struct xfs_perag	*pag,
 579	xfs_trans_t		*tp,
 580	xfs_ino_t		ino,
 581	struct xfs_inode	**ipp,
 582	int			flags,
 583	int			lock_flags)
 584{
 585	struct xfs_inode	*ip;
 586	int			error;
 587	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ino);
 588	int			iflags;
 589
 590	ip = xfs_inode_alloc(mp, ino);
 591	if (!ip)
 592		return -ENOMEM;
 593
 594	error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, flags);
 595	if (error)
 596		goto out_destroy;
 597
 598	/*
 599	 * For version 5 superblocks, if we are initialising a new inode and we
 600	 * are not utilising the XFS_MOUNT_IKEEP inode cluster mode, we can
 601	 * simply build the new inode core with a random generation number.
 602	 *
 603	 * For version 4 (and older) superblocks, log recovery is dependent on
 604	 * the i_flushiter field being initialised from the current on-disk
 605	 * value and hence we must also read the inode off disk even when
 606	 * initializing new inodes.
 607	 */
 608	if (xfs_sb_version_has_v3inode(&mp->m_sb) &&
 609	    (flags & XFS_IGET_CREATE) && !(mp->m_flags & XFS_MOUNT_IKEEP)) {
 610		VFS_I(ip)->i_generation = prandom_u32();
 611	} else {
 612		struct xfs_buf		*bp;
 613
 614		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp);
 615		if (error)
 616			goto out_destroy;
 617
 618		error = xfs_inode_from_disk(ip,
 619				xfs_buf_offset(bp, ip->i_imap.im_boffset));
 620		if (!error)
 621			xfs_buf_set_ref(bp, XFS_INO_REF);
 622		xfs_trans_brelse(tp, bp);
 623
 624		if (error)
 625			goto out_destroy;
 626	}
 627
 628	trace_xfs_iget_miss(ip);
 629
 630	/*
 631	 * Check the inode free state is valid. This also detects lookup
 632	 * racing with unlinks.
 633	 */
 634	error = xfs_iget_check_free_state(ip, flags);
 635	if (error)
 636		goto out_destroy;
 637
 638	/*
 639	 * Preload the radix tree so we can insert safely under the
 640	 * write spinlock. Note that we cannot sleep inside the preload
 641	 * region. Since we can be called from transaction context, don't
 642	 * recurse into the file system.
 643	 */
 644	if (radix_tree_preload(GFP_NOFS)) {
 645		error = -EAGAIN;
 646		goto out_destroy;
 647	}
 648
 649	/*
 650	 * Because the inode hasn't been added to the radix-tree yet it can't
 651	 * be found by another thread, so we can do the non-sleeping lock here.
 652	 */
 653	if (lock_flags) {
 654		if (!xfs_ilock_nowait(ip, lock_flags))
 655			BUG();
 656	}
 657
 658	/*
 659	 * These values must be set before inserting the inode into the radix
 660	 * tree as the moment it is inserted a concurrent lookup (allowed by the
 661	 * RCU locking mechanism) can find it and that lookup must see that this
 662	 * is an inode currently under construction (i.e. that XFS_INEW is set).
 663	 * The ip->i_flags_lock that protects the XFS_INEW flag forms the
 664	 * memory barrier that ensures this detection works correctly at lookup
 665	 * time.
 666	 */
 667	iflags = XFS_INEW;
 668	if (flags & XFS_IGET_DONTCACHE)
 669		d_mark_dontcache(VFS_I(ip));
 670	ip->i_udquot = NULL;
 671	ip->i_gdquot = NULL;
 672	ip->i_pdquot = NULL;
 673	xfs_iflags_set(ip, iflags);
 674
 675	/* insert the new inode */
 676	spin_lock(&pag->pag_ici_lock);
 677	error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
 678	if (unlikely(error)) {
 679		WARN_ON(error != -EEXIST);
 680		XFS_STATS_INC(mp, xs_ig_dup);
 681		error = -EAGAIN;
 682		goto out_preload_end;
 683	}
 684	spin_unlock(&pag->pag_ici_lock);
 685	radix_tree_preload_end();
 686
 687	*ipp = ip;
 688	return 0;
 689
 690out_preload_end:
 691	spin_unlock(&pag->pag_ici_lock);
 692	radix_tree_preload_end();
 693	if (lock_flags)
 694		xfs_iunlock(ip, lock_flags);
 695out_destroy:
 696	__destroy_inode(VFS_I(ip));
 697	xfs_inode_free(ip);
 698	return error;
 699}
 700
 701/*
 702 * Look up an inode by number in the given file system.  The inode is looked up
 703 * in the cache held in each AG.  If the inode is found in the cache, initialise
 704 * the vfs inode if necessary.
 705 *
 706 * If it is not in core, read it in from the file system's device, add it to the
 707 * cache and initialise the vfs inode.
 708 *
 709 * The inode is locked according to the value of the lock_flags parameter.
 710 * Inode lookup is only done during metadata operations and not as part of the
 711 * data IO path. Hence we only allow locking of the XFS_ILOCK during lookup.
 712 */
 713int
 714xfs_iget(
 715	struct xfs_mount	*mp,
 716	struct xfs_trans	*tp,
 717	xfs_ino_t		ino,
 718	uint			flags,
 719	uint			lock_flags,
 720	struct xfs_inode	**ipp)
 721{
 722	struct xfs_inode	*ip;
 723	struct xfs_perag	*pag;
 724	xfs_agino_t		agino;
 725	int			error;
 726
 727	ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
 728
 729	/* reject inode numbers outside existing AGs */
 730	if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
 731		return -EINVAL;
 732
 733	XFS_STATS_INC(mp, xs_ig_attempts);
 734
 735	/* get the perag structure and ensure that it's inode capable */
 736	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
 737	agino = XFS_INO_TO_AGINO(mp, ino);
 738
 739again:
 740	error = 0;
 741	rcu_read_lock();
 742	ip = radix_tree_lookup(&pag->pag_ici_root, agino);
 743
 744	if (ip) {
 745		error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
 746		if (error)
 747			goto out_error_or_again;
 748	} else {
 749		rcu_read_unlock();
 750		if (flags & XFS_IGET_INCORE) {
 751			error = -ENODATA;
 752			goto out_error_or_again;
 753		}
 754		XFS_STATS_INC(mp, xs_ig_missed);
 755
 756		error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
 757							flags, lock_flags);
 758		if (error)
 759			goto out_error_or_again;
 760	}
 761	xfs_perag_put(pag);
 762
 763	*ipp = ip;
 764
 765	/*
 766	 * If we have a real type for an on-disk inode, we can setup the inode
 767	 * now.	 If it's a new inode being created, xfs_ialloc will handle it.
 768	 */
 769	if (xfs_iflags_test(ip, XFS_INEW) && VFS_I(ip)->i_mode != 0)
 770		xfs_setup_existing_inode(ip);
 771	return 0;
 772
 773out_error_or_again:
 774	if (!(flags & XFS_IGET_INCORE) && error == -EAGAIN) {
 775		delay(1);
 776		goto again;
 777	}
 778	xfs_perag_put(pag);
 779	return error;
 780}
 781
 782/*
 783 * "Is this a cached inode that's also allocated?"
 784 *
 785 * Look up an inode by number in the given file system.  If the inode is
 786 * in cache and isn't in purgatory, return 1 if the inode is allocated
 787 * and 0 if it is not.  For all other cases (not in cache, being torn
 788 * down, etc.), return a negative error code.
 789 *
 790 * The caller has to prevent inode allocation and freeing activity,
 791 * presumably by locking the AGI buffer.   This is to ensure that an
 792 * inode cannot transition from allocated to freed until the caller is
 793 * ready to allow that.  If the inode is in an intermediate state (new,
 794 * reclaimable, or being reclaimed), -EAGAIN will be returned; if the
 795 * inode is not in the cache, -ENOENT will be returned.  The caller must
 796 * deal with these scenarios appropriately.
 797 *
 798 * This is a specialized use case for the online scrubber; if you're
 799 * reading this, you probably want xfs_iget.
 800 */
 801int
 802xfs_icache_inode_is_allocated(
 803	struct xfs_mount	*mp,
 804	struct xfs_trans	*tp,
 805	xfs_ino_t		ino,
 806	bool			*inuse)
 807{
 808	struct xfs_inode	*ip;
 809	int			error;
 810
 811	error = xfs_iget(mp, tp, ino, XFS_IGET_INCORE, 0, &ip);
 812	if (error)
 813		return error;
 814
 815	*inuse = !!(VFS_I(ip)->i_mode);
 816	xfs_irele(ip);
 817	return 0;
 818}
 819
 820#ifdef CONFIG_XFS_QUOTA
 821/* Decide if we want to grab this inode to drop its dquots. */
 822static bool
 823xfs_dqrele_igrab(
 824	struct xfs_inode	*ip)
 825{
 826	bool			ret = false;
 827
 828	ASSERT(rcu_read_lock_held());
 829
 830	/* Check for stale RCU freed inode */
 831	spin_lock(&ip->i_flags_lock);
 832	if (!ip->i_ino)
 833		goto out_unlock;
 834
 835	/*
 836	 * Skip inodes that are anywhere in the reclaim machinery because we
 837	 * drop dquots before tagging an inode for reclamation.
 838	 */
 839	if (ip->i_flags & (XFS_IRECLAIM | XFS_IRECLAIMABLE))
 840		goto out_unlock;
 841
 842	/*
 843	 * The inode looks alive; try to grab a VFS reference so that it won't
 844	 * get destroyed.  If we got the reference, return true to say that
 845	 * we grabbed the inode.
 846	 *
 847	 * If we can't get the reference, then we know the inode had its VFS
 848	 * state torn down and hasn't yet entered the reclaim machinery.  Since
 849	 * we also know that dquots are detached from an inode before it enters
 850	 * reclaim, we can skip the inode.
 851	 */
 852	ret = igrab(VFS_I(ip)) != NULL;
 853
 854out_unlock:
 855	spin_unlock(&ip->i_flags_lock);
 856	return ret;
 857}
 858
 859/* Drop this inode's dquots. */
 860static void
 861xfs_dqrele_inode(
 862	struct xfs_inode	*ip,
 863	struct xfs_icwalk	*icw)
 864{
 865	if (xfs_iflags_test(ip, XFS_INEW))
 866		xfs_inew_wait(ip);
 867
 868	xfs_ilock(ip, XFS_ILOCK_EXCL);
 869	if (icw->icw_flags & XFS_ICWALK_FLAG_DROP_UDQUOT) {
 870		xfs_qm_dqrele(ip->i_udquot);
 871		ip->i_udquot = NULL;
 872	}
 873	if (icw->icw_flags & XFS_ICWALK_FLAG_DROP_GDQUOT) {
 874		xfs_qm_dqrele(ip->i_gdquot);
 875		ip->i_gdquot = NULL;
 876	}
 877	if (icw->icw_flags & XFS_ICWALK_FLAG_DROP_PDQUOT) {
 878		xfs_qm_dqrele(ip->i_pdquot);
 879		ip->i_pdquot = NULL;
 880	}
 881	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 882	xfs_irele(ip);
 883}
 884
 885/*
 886 * Detach all dquots from incore inodes if we can.  The caller must already
 887 * have dropped the relevant XFS_[UGP]QUOTA_ACTIVE flags so that dquots will
 888 * not get reattached.
 889 */
 890int
 891xfs_dqrele_all_inodes(
 892	struct xfs_mount	*mp,
 893	unsigned int		qflags)
 894{
 895	struct xfs_icwalk	icw = { .icw_flags = 0 };
 896
 897	if (qflags & XFS_UQUOTA_ACCT)
 898		icw.icw_flags |= XFS_ICWALK_FLAG_DROP_UDQUOT;
 899	if (qflags & XFS_GQUOTA_ACCT)
 900		icw.icw_flags |= XFS_ICWALK_FLAG_DROP_GDQUOT;
 901	if (qflags & XFS_PQUOTA_ACCT)
 902		icw.icw_flags |= XFS_ICWALK_FLAG_DROP_PDQUOT;
 903
 904	return xfs_icwalk(mp, XFS_ICWALK_DQRELE, &icw);
 905}
 906#else
 907# define xfs_dqrele_igrab(ip)		(false)
 908# define xfs_dqrele_inode(ip, priv)	((void)0)
 909#endif /* CONFIG_XFS_QUOTA */
 910
 911/*
 912 * Grab the inode for reclaim exclusively.
 913 *
 914 * We have found this inode via a lookup under RCU, so the inode may have
 915 * already been freed, or it may be in the process of being recycled by
 916 * xfs_iget(). In both cases, the inode will have XFS_IRECLAIM set. If the inode
 917 * has been fully recycled by the time we get the i_flags_lock, XFS_IRECLAIMABLE
 918 * will not be set. Hence we need to check for both these flag conditions to
 919 * avoid inodes that are no longer reclaim candidates.
 920 *
 921 * Note: checking for other state flags here, under the i_flags_lock or not, is
 922 * racy and should be avoided. Those races should be resolved only after we have
 923 * ensured that we are able to reclaim this inode and the world can see that we
 924 * are going to reclaim it.
 925 *
 926 * Return true if we grabbed it, false otherwise.
 927 */
 928static bool
 929xfs_reclaim_igrab(
 930	struct xfs_inode	*ip,
 931	struct xfs_icwalk	*icw)
 932{
 933	ASSERT(rcu_read_lock_held());
 934
 935	spin_lock(&ip->i_flags_lock);
 936	if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
 937	    __xfs_iflags_test(ip, XFS_IRECLAIM)) {
 938		/* not a reclaim candidate. */
 939		spin_unlock(&ip->i_flags_lock);
 940		return false;
 941	}
 942
 943	/* Don't reclaim a sick inode unless the caller asked for it. */
 944	if (ip->i_sick &&
 945	    (!icw || !(icw->icw_flags & XFS_ICWALK_FLAG_RECLAIM_SICK))) {
 946		spin_unlock(&ip->i_flags_lock);
 947		return false;
 948	}
 949
 950	__xfs_iflags_set(ip, XFS_IRECLAIM);
 951	spin_unlock(&ip->i_flags_lock);
 952	return true;
 953}
 954
 955/*
 956 * Inode reclaim is non-blocking, so the default action if progress cannot be
 957 * made is to "requeue" the inode for reclaim by unlocking it and clearing the
 958 * XFS_IRECLAIM flag.  If we are in a shutdown state, we don't care about
 959 * blocking anymore and hence we can wait for the inode to be able to reclaim
 960 * it.
 961 *
 962 * We do no IO here - if callers require inodes to be cleaned they must push the
 963 * AIL first to trigger writeback of dirty inodes.  This enables writeback to be
 964 * done in the background in a non-blocking manner, and enables memory reclaim
 965 * to make progress without blocking.
 966 */
 967static void
 968xfs_reclaim_inode(
 969	struct xfs_inode	*ip,
 970	struct xfs_perag	*pag)
 971{
 972	xfs_ino_t		ino = ip->i_ino; /* for radix_tree_delete */
 973
 974	if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
 975		goto out;
 976	if (xfs_iflags_test_and_set(ip, XFS_IFLUSHING))
 977		goto out_iunlock;
 978
 979	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 980		xfs_iunpin_wait(ip);
 981		xfs_iflush_abort(ip);
 982		goto reclaim;
 983	}
 984	if (xfs_ipincount(ip))
 985		goto out_clear_flush;
 986	if (!xfs_inode_clean(ip))
 987		goto out_clear_flush;
 988
 989	xfs_iflags_clear(ip, XFS_IFLUSHING);
 990reclaim:
 991
 992	/*
 993	 * Because we use RCU freeing we need to ensure the inode always appears
 994	 * to be reclaimed with an invalid inode number when in the free state.
 995	 * We do this as early as possible under the ILOCK so that
 996	 * xfs_iflush_cluster() and xfs_ifree_cluster() can be guaranteed to
 997	 * detect races with us here. By doing this, we guarantee that once
 998	 * xfs_iflush_cluster() or xfs_ifree_cluster() has locked XFS_ILOCK that
 999	 * it will see either a valid inode that will serialise correctly, or it
1000	 * will see an invalid inode that it can skip.
1001	 */
1002	spin_lock(&ip->i_flags_lock);
1003	ip->i_flags = XFS_IRECLAIM;
1004	ip->i_ino = 0;
1005	ip->i_sick = 0;
1006	ip->i_checked = 0;
1007	spin_unlock(&ip->i_flags_lock);
1008
1009	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1010
1011	XFS_STATS_INC(ip->i_mount, xs_ig_reclaims);
1012	/*
1013	 * Remove the inode from the per-AG radix tree.
1014	 *
1015	 * Because radix_tree_delete won't complain even if the item was never
1016	 * added to the tree assert that it's been there before to catch
1017	 * problems with the inode life time early on.
1018	 */
1019	spin_lock(&pag->pag_ici_lock);
1020	if (!radix_tree_delete(&pag->pag_ici_root,
1021				XFS_INO_TO_AGINO(ip->i_mount, ino)))
1022		ASSERT(0);
1023	xfs_perag_clear_inode_tag(pag, NULLAGINO, XFS_ICI_RECLAIM_TAG);
1024	spin_unlock(&pag->pag_ici_lock);
1025
1026	/*
1027	 * Here we do an (almost) spurious inode lock in order to coordinate
1028	 * with inode cache radix tree lookups.  This is because the lookup
1029	 * can reference the inodes in the cache without taking references.
1030	 *
1031	 * We make that OK here by ensuring that we wait until the inode is
1032	 * unlocked after the lookup before we go ahead and free it.
1033	 */
1034	xfs_ilock(ip, XFS_ILOCK_EXCL);
1035	ASSERT(!ip->i_udquot && !ip->i_gdquot && !ip->i_pdquot);
1036	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1037	ASSERT(xfs_inode_clean(ip));
1038
1039	__xfs_inode_free(ip);
1040	return;
1041
1042out_clear_flush:
1043	xfs_iflags_clear(ip, XFS_IFLUSHING);
1044out_iunlock:
1045	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1046out:
1047	xfs_iflags_clear(ip, XFS_IRECLAIM);
1048}
1049
1050/* Reclaim sick inodes if we're unmounting or the fs went down. */
1051static inline bool
1052xfs_want_reclaim_sick(
1053	struct xfs_mount	*mp)
1054{
1055	return (mp->m_flags & XFS_MOUNT_UNMOUNTING) ||
1056	       (mp->m_flags & XFS_MOUNT_NORECOVERY) ||
1057	       XFS_FORCED_SHUTDOWN(mp);
1058}
1059
1060void
1061xfs_reclaim_inodes(
1062	struct xfs_mount	*mp)
1063{
1064	struct xfs_icwalk	icw = {
1065		.icw_flags	= 0,
1066	};
1067
1068	if (xfs_want_reclaim_sick(mp))
1069		icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1070
1071	while (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
1072		xfs_ail_push_all_sync(mp->m_ail);
1073		xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1074	}
1075}
1076
1077/*
1078 * The shrinker infrastructure determines how many inodes we should scan for
1079 * reclaim. We want as many clean inodes ready to reclaim as possible, so we
1080 * push the AIL here. We also want to proactively free up memory if we can to
1081 * minimise the amount of work memory reclaim has to do so we kick the
1082 * background reclaim if it isn't already scheduled.
1083 */
1084long
1085xfs_reclaim_inodes_nr(
1086	struct xfs_mount	*mp,
1087	unsigned long		nr_to_scan)
1088{
1089	struct xfs_icwalk	icw = {
1090		.icw_flags	= XFS_ICWALK_FLAG_SCAN_LIMIT,
1091		.icw_scan_limit	= min_t(unsigned long, LONG_MAX, nr_to_scan),
1092	};
1093
1094	if (xfs_want_reclaim_sick(mp))
1095		icw.icw_flags |= XFS_ICWALK_FLAG_RECLAIM_SICK;
1096
1097	/* kick background reclaimer and push the AIL */
1098	xfs_reclaim_work_queue(mp);
1099	xfs_ail_push_all(mp->m_ail);
1100
1101	xfs_icwalk(mp, XFS_ICWALK_RECLAIM, &icw);
1102	return 0;
1103}
1104
1105/*
1106 * Return the number of reclaimable inodes in the filesystem for
1107 * the shrinker to determine how much to reclaim.
1108 */
1109long
1110xfs_reclaim_inodes_count(
1111	struct xfs_mount	*mp)
1112{
1113	struct xfs_perag	*pag;
1114	xfs_agnumber_t		ag = 0;
1115	long			reclaimable = 0;
1116
1117	while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1118		ag = pag->pag_agno + 1;
1119		reclaimable += pag->pag_ici_reclaimable;
1120		xfs_perag_put(pag);
1121	}
1122	return reclaimable;
1123}
1124
1125STATIC bool
1126xfs_icwalk_match_id(
1127	struct xfs_inode	*ip,
1128	struct xfs_icwalk	*icw)
1129{
1130	if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1131	    !uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1132		return false;
1133
1134	if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) &&
1135	    !gid_eq(VFS_I(ip)->i_gid, icw->icw_gid))
1136		return false;
1137
1138	if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) &&
1139	    ip->i_projid != icw->icw_prid)
1140		return false;
1141
1142	return true;
1143}
1144
1145/*
1146 * A union-based inode filtering algorithm. Process the inode if any of the
1147 * criteria match. This is for global/internal scans only.
1148 */
1149STATIC bool
1150xfs_icwalk_match_id_union(
1151	struct xfs_inode	*ip,
1152	struct xfs_icwalk	*icw)
1153{
1154	if ((icw->icw_flags & XFS_ICWALK_FLAG_UID) &&
1155	    uid_eq(VFS_I(ip)->i_uid, icw->icw_uid))
1156		return true;
1157
1158	if ((icw->icw_flags & XFS_ICWALK_FLAG_GID) &&
1159	    gid_eq(VFS_I(ip)->i_gid, icw->icw_gid))
1160		return true;
1161
1162	if ((icw->icw_flags & XFS_ICWALK_FLAG_PRID) &&
1163	    ip->i_projid == icw->icw_prid)
1164		return true;
1165
1166	return false;
1167}
1168
1169/*
1170 * Is this inode @ip eligible for eof/cow block reclamation, given some
1171 * filtering parameters @icw?  The inode is eligible if @icw is null or
1172 * if the predicate functions match.
1173 */
1174static bool
1175xfs_icwalk_match(
1176	struct xfs_inode	*ip,
1177	struct xfs_icwalk	*icw)
1178{
1179	bool			match;
1180
1181	if (!icw)
1182		return true;
1183
1184	if (icw->icw_flags & XFS_ICWALK_FLAG_UNION)
1185		match = xfs_icwalk_match_id_union(ip, icw);
1186	else
1187		match = xfs_icwalk_match_id(ip, icw);
1188	if (!match)
1189		return false;
1190
1191	/* skip the inode if the file size is too small */
1192	if ((icw->icw_flags & XFS_ICWALK_FLAG_MINFILESIZE) &&
1193	    XFS_ISIZE(ip) < icw->icw_min_file_size)
1194		return false;
1195
1196	return true;
1197}
1198
1199/*
1200 * This is a fast pass over the inode cache to try to get reclaim moving on as
1201 * many inodes as possible in a short period of time. It kicks itself every few
1202 * seconds, as well as being kicked by the inode cache shrinker when memory
1203 * goes low.
1204 */
1205void
1206xfs_reclaim_worker(
1207	struct work_struct *work)
1208{
1209	struct xfs_mount *mp = container_of(to_delayed_work(work),
1210					struct xfs_mount, m_reclaim_work);
1211
1212	xfs_icwalk(mp, XFS_ICWALK_RECLAIM, NULL);
1213	xfs_reclaim_work_queue(mp);
1214}
1215
1216STATIC int
1217xfs_inode_free_eofblocks(
1218	struct xfs_inode	*ip,
1219	struct xfs_icwalk	*icw,
1220	unsigned int		*lockflags)
1221{
1222	bool			wait;
1223
1224	wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1225
1226	if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS))
1227		return 0;
1228
1229	/*
1230	 * If the mapping is dirty the operation can block and wait for some
1231	 * time. Unless we are waiting, skip it.
1232	 */
1233	if (!wait && mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1234		return 0;
1235
1236	if (!xfs_icwalk_match(ip, icw))
1237		return 0;
1238
1239	/*
1240	 * If the caller is waiting, return -EAGAIN to keep the background
1241	 * scanner moving and revisit the inode in a subsequent pass.
1242	 */
1243	if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1244		if (wait)
1245			return -EAGAIN;
1246		return 0;
1247	}
1248	*lockflags |= XFS_IOLOCK_EXCL;
1249
1250	if (xfs_can_free_eofblocks(ip, false))
1251		return xfs_free_eofblocks(ip);
1252
1253	/* inode could be preallocated or append-only */
1254	trace_xfs_inode_free_eofblocks_invalid(ip);
1255	xfs_inode_clear_eofblocks_tag(ip);
1256	return 0;
1257}
1258
1259static void
1260xfs_blockgc_set_iflag(
1261	struct xfs_inode	*ip,
1262	unsigned long		iflag)
1263{
1264	struct xfs_mount	*mp = ip->i_mount;
1265	struct xfs_perag	*pag;
1266
1267	ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
1268
1269	/*
1270	 * Don't bother locking the AG and looking up in the radix trees
1271	 * if we already know that we have the tag set.
1272	 */
1273	if (ip->i_flags & iflag)
1274		return;
1275	spin_lock(&ip->i_flags_lock);
1276	ip->i_flags |= iflag;
1277	spin_unlock(&ip->i_flags_lock);
1278
1279	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1280	spin_lock(&pag->pag_ici_lock);
1281
1282	xfs_perag_set_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1283			XFS_ICI_BLOCKGC_TAG);
1284
1285	spin_unlock(&pag->pag_ici_lock);
1286	xfs_perag_put(pag);
1287}
1288
1289void
1290xfs_inode_set_eofblocks_tag(
1291	xfs_inode_t	*ip)
1292{
1293	trace_xfs_inode_set_eofblocks_tag(ip);
1294	return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS);
1295}
1296
1297static void
1298xfs_blockgc_clear_iflag(
1299	struct xfs_inode	*ip,
1300	unsigned long		iflag)
1301{
1302	struct xfs_mount	*mp = ip->i_mount;
1303	struct xfs_perag	*pag;
1304	bool			clear_tag;
1305
1306	ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
1307
1308	spin_lock(&ip->i_flags_lock);
1309	ip->i_flags &= ~iflag;
1310	clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0;
1311	spin_unlock(&ip->i_flags_lock);
1312
1313	if (!clear_tag)
1314		return;
1315
1316	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1317	spin_lock(&pag->pag_ici_lock);
1318
1319	xfs_perag_clear_inode_tag(pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
1320			XFS_ICI_BLOCKGC_TAG);
1321
1322	spin_unlock(&pag->pag_ici_lock);
1323	xfs_perag_put(pag);
1324}
1325
1326void
1327xfs_inode_clear_eofblocks_tag(
1328	xfs_inode_t	*ip)
1329{
1330	trace_xfs_inode_clear_eofblocks_tag(ip);
1331	return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS);
1332}
1333
1334/*
1335 * Set ourselves up to free CoW blocks from this file.  If it's already clean
1336 * then we can bail out quickly, but otherwise we must back off if the file
1337 * is undergoing some kind of write.
1338 */
1339static bool
1340xfs_prep_free_cowblocks(
1341	struct xfs_inode	*ip)
1342{
1343	/*
1344	 * Just clear the tag if we have an empty cow fork or none at all. It's
1345	 * possible the inode was fully unshared since it was originally tagged.
1346	 */
1347	if (!xfs_inode_has_cow_data(ip)) {
1348		trace_xfs_inode_free_cowblocks_invalid(ip);
1349		xfs_inode_clear_cowblocks_tag(ip);
1350		return false;
1351	}
1352
1353	/*
1354	 * If the mapping is dirty or under writeback we cannot touch the
1355	 * CoW fork.  Leave it alone if we're in the midst of a directio.
1356	 */
1357	if ((VFS_I(ip)->i_state & I_DIRTY_PAGES) ||
1358	    mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY) ||
1359	    mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_WRITEBACK) ||
1360	    atomic_read(&VFS_I(ip)->i_dio_count))
1361		return false;
1362
1363	return true;
1364}
1365
1366/*
1367 * Automatic CoW Reservation Freeing
1368 *
1369 * These functions automatically garbage collect leftover CoW reservations
1370 * that were made on behalf of a cowextsize hint when we start to run out
1371 * of quota or when the reservations sit around for too long.  If the file
1372 * has dirty pages or is undergoing writeback, its CoW reservations will
1373 * be retained.
1374 *
1375 * The actual garbage collection piggybacks off the same code that runs
1376 * the speculative EOF preallocation garbage collector.
1377 */
1378STATIC int
1379xfs_inode_free_cowblocks(
1380	struct xfs_inode	*ip,
1381	struct xfs_icwalk	*icw,
1382	unsigned int		*lockflags)
1383{
1384	bool			wait;
1385	int			ret = 0;
1386
1387	wait = icw && (icw->icw_flags & XFS_ICWALK_FLAG_SYNC);
1388
1389	if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS))
1390		return 0;
1391
1392	if (!xfs_prep_free_cowblocks(ip))
1393		return 0;
1394
1395	if (!xfs_icwalk_match(ip, icw))
1396		return 0;
1397
1398	/*
1399	 * If the caller is waiting, return -EAGAIN to keep the background
1400	 * scanner moving and revisit the inode in a subsequent pass.
1401	 */
1402	if (!(*lockflags & XFS_IOLOCK_EXCL) &&
1403	    !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1404		if (wait)
1405			return -EAGAIN;
1406		return 0;
1407	}
1408	*lockflags |= XFS_IOLOCK_EXCL;
1409
1410	if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) {
1411		if (wait)
1412			return -EAGAIN;
1413		return 0;
1414	}
1415	*lockflags |= XFS_MMAPLOCK_EXCL;
1416
1417	/*
1418	 * Check again, nobody else should be able to dirty blocks or change
1419	 * the reflink iflag now that we have the first two locks held.
1420	 */
1421	if (xfs_prep_free_cowblocks(ip))
1422		ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
1423	return ret;
1424}
1425
1426void
1427xfs_inode_set_cowblocks_tag(
1428	xfs_inode_t	*ip)
1429{
1430	trace_xfs_inode_set_cowblocks_tag(ip);
1431	return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS);
1432}
1433
1434void
1435xfs_inode_clear_cowblocks_tag(
1436	xfs_inode_t	*ip)
1437{
1438	trace_xfs_inode_clear_cowblocks_tag(ip);
1439	return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS);
1440}
1441
1442/* Disable post-EOF and CoW block auto-reclamation. */
1443void
1444xfs_blockgc_stop(
1445	struct xfs_mount	*mp)
1446{
1447	struct xfs_perag	*pag;
1448	xfs_agnumber_t		agno;
1449
1450	for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1451		cancel_delayed_work_sync(&pag->pag_blockgc_work);
1452}
1453
1454/* Enable post-EOF and CoW block auto-reclamation. */
1455void
1456xfs_blockgc_start(
1457	struct xfs_mount	*mp)
1458{
1459	struct xfs_perag	*pag;
1460	xfs_agnumber_t		agno;
1461
1462	for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
1463		xfs_blockgc_queue(pag);
1464}
1465
1466/* Don't try to run block gc on an inode that's in any of these states. */
1467#define XFS_BLOCKGC_NOGRAB_IFLAGS	(XFS_INEW | \
1468					 XFS_IRECLAIMABLE | \
1469					 XFS_IRECLAIM)
1470/*
1471 * Decide if the given @ip is eligible for garbage collection of speculative
1472 * preallocations, and grab it if so.  Returns true if it's ready to go or
1473 * false if we should just ignore it.
1474 */
1475static bool
1476xfs_blockgc_igrab(
1477	struct xfs_inode	*ip)
1478{
1479	struct inode		*inode = VFS_I(ip);
1480
1481	ASSERT(rcu_read_lock_held());
1482
1483	/* Check for stale RCU freed inode */
1484	spin_lock(&ip->i_flags_lock);
1485	if (!ip->i_ino)
1486		goto out_unlock_noent;
1487
1488	if (ip->i_flags & XFS_BLOCKGC_NOGRAB_IFLAGS)
1489		goto out_unlock_noent;
1490	spin_unlock(&ip->i_flags_lock);
1491
1492	/* nothing to sync during shutdown */
1493	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1494		return false;
1495
1496	/* If we can't grab the inode, it must on it's way to reclaim. */
1497	if (!igrab(inode))
1498		return false;
1499
1500	/* inode is valid */
1501	return true;
1502
1503out_unlock_noent:
1504	spin_unlock(&ip->i_flags_lock);
1505	return false;
1506}
1507
1508/* Scan one incore inode for block preallocations that we can remove. */
1509static int
1510xfs_blockgc_scan_inode(
1511	struct xfs_inode	*ip,
1512	struct xfs_icwalk	*icw)
1513{
1514	unsigned int		lockflags = 0;
1515	int			error;
1516
1517	error = xfs_inode_free_eofblocks(ip, icw, &lockflags);
1518	if (error)
1519		goto unlock;
1520
1521	error = xfs_inode_free_cowblocks(ip, icw, &lockflags);
1522unlock:
1523	if (lockflags)
1524		xfs_iunlock(ip, lockflags);
1525	xfs_irele(ip);
1526	return error;
1527}
1528
1529/* Background worker that trims preallocated space. */
1530void
1531xfs_blockgc_worker(
1532	struct work_struct	*work)
1533{
1534	struct xfs_perag	*pag = container_of(to_delayed_work(work),
1535					struct xfs_perag, pag_blockgc_work);
1536	struct xfs_mount	*mp = pag->pag_mount;
1537	int			error;
1538
1539	if (!sb_start_write_trylock(mp->m_super))
1540		return;
1541	error = xfs_icwalk_ag(pag, XFS_ICWALK_BLOCKGC, NULL);
1542	if (error)
1543		xfs_info(mp, "AG %u preallocation gc worker failed, err=%d",
1544				pag->pag_agno, error);
1545	sb_end_write(mp->m_super);
1546	xfs_blockgc_queue(pag);
1547}
1548
1549/*
1550 * Try to free space in the filesystem by purging eofblocks and cowblocks.
1551 */
1552int
1553xfs_blockgc_free_space(
1554	struct xfs_mount	*mp,
1555	struct xfs_icwalk	*icw)
1556{
1557	trace_xfs_blockgc_free_space(mp, icw, _RET_IP_);
1558
1559	return xfs_icwalk(mp, XFS_ICWALK_BLOCKGC, icw);
1560}
1561
1562/*
1563 * Run cow/eofblocks scans on the supplied dquots.  We don't know exactly which
1564 * quota caused an allocation failure, so we make a best effort by including
1565 * each quota under low free space conditions (less than 1% free space) in the
1566 * scan.
1567 *
1568 * Callers must not hold any inode's ILOCK.  If requesting a synchronous scan
1569 * (XFS_ICWALK_FLAG_SYNC), the caller also must not hold any inode's IOLOCK or
1570 * MMAPLOCK.
1571 */
1572int
1573xfs_blockgc_free_dquots(
1574	struct xfs_mount	*mp,
1575	struct xfs_dquot	*udqp,
1576	struct xfs_dquot	*gdqp,
1577	struct xfs_dquot	*pdqp,
1578	unsigned int		iwalk_flags)
1579{
1580	struct xfs_icwalk	icw = {0};
1581	bool			do_work = false;
1582
1583	if (!udqp && !gdqp && !pdqp)
1584		return 0;
1585
1586	/*
1587	 * Run a scan to free blocks using the union filter to cover all
1588	 * applicable quotas in a single scan.
1589	 */
1590	icw.icw_flags = XFS_ICWALK_FLAG_UNION | iwalk_flags;
1591
1592	if (XFS_IS_UQUOTA_ENFORCED(mp) && udqp && xfs_dquot_lowsp(udqp)) {
1593		icw.icw_uid = make_kuid(mp->m_super->s_user_ns, udqp->q_id);
1594		icw.icw_flags |= XFS_ICWALK_FLAG_UID;
1595		do_work = true;
1596	}
1597
1598	if (XFS_IS_UQUOTA_ENFORCED(mp) && gdqp && xfs_dquot_lowsp(gdqp)) {
1599		icw.icw_gid = make_kgid(mp->m_super->s_user_ns, gdqp->q_id);
1600		icw.icw_flags |= XFS_ICWALK_FLAG_GID;
1601		do_work = true;
1602	}
1603
1604	if (XFS_IS_PQUOTA_ENFORCED(mp) && pdqp && xfs_dquot_lowsp(pdqp)) {
1605		icw.icw_prid = pdqp->q_id;
1606		icw.icw_flags |= XFS_ICWALK_FLAG_PRID;
1607		do_work = true;
1608	}
1609
1610	if (!do_work)
1611		return 0;
1612
1613	return xfs_blockgc_free_space(mp, &icw);
1614}
1615
1616/* Run cow/eofblocks scans on the quotas attached to the inode. */
1617int
1618xfs_blockgc_free_quota(
1619	struct xfs_inode	*ip,
1620	unsigned int		iwalk_flags)
1621{
1622	return xfs_blockgc_free_dquots(ip->i_mount,
1623			xfs_inode_dquot(ip, XFS_DQTYPE_USER),
1624			xfs_inode_dquot(ip, XFS_DQTYPE_GROUP),
1625			xfs_inode_dquot(ip, XFS_DQTYPE_PROJ), iwalk_flags);
1626}
1627
1628/* XFS Inode Cache Walking Code */
1629
1630/*
1631 * The inode lookup is done in batches to keep the amount of lock traffic and
1632 * radix tree lookups to a minimum. The batch size is a trade off between
1633 * lookup reduction and stack usage. This is in the reclaim path, so we can't
1634 * be too greedy.
1635 */
1636#define XFS_LOOKUP_BATCH	32
1637
1638
1639/*
1640 * Decide if we want to grab this inode in anticipation of doing work towards
1641 * the goal.
1642 */
1643static inline bool
1644xfs_icwalk_igrab(
1645	enum xfs_icwalk_goal	goal,
1646	struct xfs_inode	*ip,
1647	struct xfs_icwalk	*icw)
1648{
1649	switch (goal) {
1650	case XFS_ICWALK_DQRELE:
1651		return xfs_dqrele_igrab(ip);
1652	case XFS_ICWALK_BLOCKGC:
1653		return xfs_blockgc_igrab(ip);
1654	case XFS_ICWALK_RECLAIM:
1655		return xfs_reclaim_igrab(ip, icw);
1656	default:
1657		return false;
1658	}
1659}
1660
1661/*
1662 * Process an inode.  Each processing function must handle any state changes
1663 * made by the icwalk igrab function.  Return -EAGAIN to skip an inode.
1664 */
1665static inline int
1666xfs_icwalk_process_inode(
1667	enum xfs_icwalk_goal	goal,
1668	struct xfs_inode	*ip,
1669	struct xfs_perag	*pag,
1670	struct xfs_icwalk	*icw)
1671{
1672	int			error = 0;
1673
1674	switch (goal) {
1675	case XFS_ICWALK_DQRELE:
1676		xfs_dqrele_inode(ip, icw);
1677		break;
1678	case XFS_ICWALK_BLOCKGC:
1679		error = xfs_blockgc_scan_inode(ip, icw);
1680		break;
1681	case XFS_ICWALK_RECLAIM:
1682		xfs_reclaim_inode(ip, pag);
1683		break;
1684	}
1685	return error;
1686}
1687
1688/*
1689 * For a given per-AG structure @pag and a goal, grab qualifying inodes and
1690 * process them in some manner.
1691 */
1692static int
1693xfs_icwalk_ag(
1694	struct xfs_perag	*pag,
1695	enum xfs_icwalk_goal	goal,
1696	struct xfs_icwalk	*icw)
1697{
1698	struct xfs_mount	*mp = pag->pag_mount;
1699	uint32_t		first_index;
1700	int			last_error = 0;
1701	int			skipped;
1702	bool			done;
1703	int			nr_found;
1704
1705restart:
1706	done = false;
1707	skipped = 0;
1708	if (goal == XFS_ICWALK_RECLAIM)
1709		first_index = READ_ONCE(pag->pag_ici_reclaim_cursor);
1710	else
1711		first_index = 0;
1712	nr_found = 0;
1713	do {
1714		struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1715		unsigned int	tag = xfs_icwalk_tag(goal);
1716		int		error = 0;
1717		int		i;
1718
1719		rcu_read_lock();
1720
1721		if (tag == XFS_ICWALK_NULL_TAG)
1722			nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
1723					(void **)batch, first_index,
1724					XFS_LOOKUP_BATCH);
1725		else
1726			nr_found = radix_tree_gang_lookup_tag(
1727					&pag->pag_ici_root,
1728					(void **) batch, first_index,
1729					XFS_LOOKUP_BATCH, tag);
1730
1731		if (!nr_found) {
1732			done = true;
1733			rcu_read_unlock();
1734			break;
1735		}
1736
1737		/*
1738		 * Grab the inodes before we drop the lock. if we found
1739		 * nothing, nr == 0 and the loop will be skipped.
1740		 */
1741		for (i = 0; i < nr_found; i++) {
1742			struct xfs_inode *ip = batch[i];
1743
1744			if (done || !xfs_icwalk_igrab(goal, ip, icw))
1745				batch[i] = NULL;
1746
1747			/*
1748			 * Update the index for the next lookup. Catch
1749			 * overflows into the next AG range which can occur if
1750			 * we have inodes in the last block of the AG and we
1751			 * are currently pointing to the last inode.
1752			 *
1753			 * Because we may see inodes that are from the wrong AG
1754			 * due to RCU freeing and reallocation, only update the
1755			 * index if it lies in this AG. It was a race that lead
1756			 * us to see this inode, so another lookup from the
1757			 * same index will not find it again.
1758			 */
1759			if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
1760				continue;
1761			first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1762			if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1763				done = true;
1764		}
1765
1766		/* unlock now we've grabbed the inodes. */
1767		rcu_read_unlock();
1768
1769		for (i = 0; i < nr_found; i++) {
1770			if (!batch[i])
1771				continue;
1772			error = xfs_icwalk_process_inode(goal, batch[i], pag,
1773					icw);
1774			if (error == -EAGAIN) {
1775				skipped++;
1776				continue;
1777			}
1778			if (error && last_error != -EFSCORRUPTED)
1779				last_error = error;
1780		}
1781
1782		/* bail out if the filesystem is corrupted.  */
1783		if (error == -EFSCORRUPTED)
1784			break;
1785
1786		cond_resched();
1787
1788		if (icw && (icw->icw_flags & XFS_ICWALK_FLAG_SCAN_LIMIT)) {
1789			icw->icw_scan_limit -= XFS_LOOKUP_BATCH;
1790			if (icw->icw_scan_limit <= 0)
1791				break;
1792		}
1793	} while (nr_found && !done);
1794
1795	if (goal == XFS_ICWALK_RECLAIM) {
1796		if (done)
1797			first_index = 0;
1798		WRITE_ONCE(pag->pag_ici_reclaim_cursor, first_index);
1799	}
1800
1801	if (skipped) {
1802		delay(1);
1803		goto restart;
1804	}
1805	return last_error;
1806}
1807
1808/* Fetch the next (possibly tagged) per-AG structure. */
1809static inline struct xfs_perag *
1810xfs_icwalk_get_perag(
1811	struct xfs_mount	*mp,
1812	xfs_agnumber_t		agno,
1813	enum xfs_icwalk_goal	goal)
1814{
1815	unsigned int		tag = xfs_icwalk_tag(goal);
1816
1817	if (tag == XFS_ICWALK_NULL_TAG)
1818		return xfs_perag_get(mp, agno);
1819	return xfs_perag_get_tag(mp, agno, tag);
1820}
1821
1822/* Walk all incore inodes to achieve a given goal. */
1823static int
1824xfs_icwalk(
1825	struct xfs_mount	*mp,
1826	enum xfs_icwalk_goal	goal,
1827	struct xfs_icwalk	*icw)
1828{
1829	struct xfs_perag	*pag;
1830	int			error = 0;
1831	int			last_error = 0;
1832	xfs_agnumber_t		agno = 0;
1833
1834	while ((pag = xfs_icwalk_get_perag(mp, agno, goal))) {
1835		agno = pag->pag_agno + 1;
1836		error = xfs_icwalk_ag(pag, goal, icw);
1837		xfs_perag_put(pag);
1838		if (error) {
1839			last_error = error;
1840			if (error == -EFSCORRUPTED)
1841				break;
1842		}
1843	}
1844	return last_error;
1845	BUILD_BUG_ON(XFS_ICWALK_PRIVATE_FLAGS & XFS_ICWALK_FLAGS_VALID);
1846}