brintos

brintos / linux-shallow public Read only

0
0
Text · 47.5 KiB · 3e45b9b Raw
1902 lines · c
1// SPDX-License-Identifier: GPL-2.0-or-later2/*3 * Copyright (C) 2018-2023 Oracle.  All Rights Reserved.4 * Author: Darrick J. Wong <djwong@kernel.org>5 */6#include "xfs.h"7#include "xfs_fs.h"8#include "xfs_shared.h"9#include "xfs_format.h"10#include "xfs_trans_resv.h"11#include "xfs_mount.h"12#include "xfs_defer.h"13#include "xfs_btree.h"14#include "xfs_bit.h"15#include "xfs_log_format.h"16#include "xfs_trans.h"17#include "xfs_sb.h"18#include "xfs_inode.h"19#include "xfs_icache.h"20#include "xfs_inode_buf.h"21#include "xfs_inode_fork.h"22#include "xfs_ialloc.h"23#include "xfs_da_format.h"24#include "xfs_reflink.h"25#include "xfs_alloc.h"26#include "xfs_rmap.h"27#include "xfs_rmap_btree.h"28#include "xfs_bmap.h"29#include "xfs_bmap_btree.h"30#include "xfs_bmap_util.h"31#include "xfs_dir2.h"32#include "xfs_dir2_priv.h"33#include "xfs_quota_defs.h"34#include "xfs_quota.h"35#include "xfs_ag.h"36#include "xfs_rtbitmap.h"37#include "xfs_attr_leaf.h"38#include "xfs_log_priv.h"39#include "xfs_health.h"40#include "xfs_symlink_remote.h"41#include "scrub/xfs_scrub.h"42#include "scrub/scrub.h"43#include "scrub/common.h"44#include "scrub/btree.h"45#include "scrub/trace.h"46#include "scrub/repair.h"47#include "scrub/iscan.h"48#include "scrub/readdir.h"49#include "scrub/tempfile.h"50 51/*52 * Inode Record Repair53 * ===================54 *55 * Roughly speaking, inode problems can be classified based on whether or not56 * they trip the dinode verifiers.  If those trip, then we won't be able to57 * xfs_iget ourselves the inode.58 *59 * Therefore, the xrep_dinode_* functions fix anything that will cause the60 * inode buffer verifier or the dinode verifier.  The xrep_inode_* functions61 * fix things on live incore inodes.  The inode repair functions make decisions62 * with security and usability implications when reviving a file:63 *64 * - Files with zero di_mode or a garbage di_mode are converted to regular file65 *   that only root can read.  This file may not actually contain user data,66 *   if the file was not previously a regular file.  Setuid and setgid bits67 *   are cleared.68 *69 * - Zero-size directories can be truncated to look empty.  It is necessary to70 *   run the bmapbtd and directory repair functions to fully rebuild the71 *   directory.72 *73 * - Zero-size symbolic link targets can be truncated to '?'.  It is necessary74 *   to run the bmapbtd and symlink repair functions to salvage the symlink.75 *76 * - Invalid extent size hints will be removed.77 *78 * - Quotacheck will be scheduled if we repaired an inode that was so badly79 *   damaged that the ondisk inode had to be rebuilt.80 *81 * - Invalid user, group, or project IDs (aka -1U) will be reset to zero.82 *   Setuid and setgid bits are cleared.83 *84 * - Data and attr forks are reset to extents format with zero extents if the85 *   fork data is inconsistent.  It is necessary to run the bmapbtd or bmapbta86 *   repair functions to recover the space mapping.87 *88 * - ACLs will not be recovered if the attr fork is zapped or the extended89 *   attribute structure itself requires salvaging.90 *91 * - If the attr fork is zapped, the user and group ids are reset to root and92 *   the setuid and setgid bits are removed.93 */94 95/*96 * All the information we need to repair the ondisk inode if we can't iget the97 * incore inode.  We don't allocate this buffer unless we're going to perform98 * a repair to the ondisk inode cluster buffer.99 */100struct xrep_inode {101	/* Inode mapping that we saved from the initial lookup attempt. */102	struct xfs_imap		imap;103 104	struct xfs_scrub	*sc;105 106	/* Blocks in use on the data device by data extents or bmbt blocks. */107	xfs_rfsblock_t		data_blocks;108 109	/* Blocks in use on the rt device. */110	xfs_rfsblock_t		rt_blocks;111 112	/* Blocks in use by the attr fork. */113	xfs_rfsblock_t		attr_blocks;114 115	/* Number of data device extents for the data fork. */116	xfs_extnum_t		data_extents;117 118	/*119	 * Number of realtime device extents for the data fork.  If120	 * data_extents and rt_extents indicate that the data fork has extents121	 * on both devices, we'll just back away slowly.122	 */123	xfs_extnum_t		rt_extents;124 125	/* Number of (data device) extents for the attr fork. */126	xfs_aextnum_t		attr_extents;127 128	/* Sick state to set after zapping parts of the inode. */129	unsigned int		ino_sick_mask;130 131	/* Must we remove all access from this file? */132	bool			zap_acls;133 134	/* Inode scanner to see if we can find the ftype from dirents */135	struct xchk_iscan	ftype_iscan;136	uint8_t			alleged_ftype;137};138 139/*140 * Setup function for inode repair.  @imap contains the ondisk inode mapping141 * information so that we can correct the ondisk inode cluster buffer if142 * necessary to make iget work.143 */144int145xrep_setup_inode(146	struct xfs_scrub	*sc,147	const struct xfs_imap	*imap)148{149	struct xrep_inode	*ri;150 151	sc->buf = kzalloc(sizeof(struct xrep_inode), XCHK_GFP_FLAGS);152	if (!sc->buf)153		return -ENOMEM;154 155	ri = sc->buf;156	memcpy(&ri->imap, imap, sizeof(struct xfs_imap));157	ri->sc = sc;158	return 0;159}160 161/*162 * Make sure this ondisk inode can pass the inode buffer verifier.  This is163 * not the same as the dinode verifier.164 */165STATIC void166xrep_dinode_buf_core(167	struct xfs_scrub	*sc,168	struct xfs_buf		*bp,169	unsigned int		ioffset)170{171	struct xfs_dinode	*dip = xfs_buf_offset(bp, ioffset);172	struct xfs_trans	*tp = sc->tp;173	struct xfs_mount	*mp = sc->mp;174	xfs_agino_t		agino;175	bool			crc_ok = false;176	bool			magic_ok = false;177	bool			unlinked_ok = false;178 179	agino = be32_to_cpu(dip->di_next_unlinked);180 181	if (xfs_verify_agino_or_null(bp->b_pag, agino))182		unlinked_ok = true;183 184	if (dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&185	    xfs_dinode_good_version(mp, dip->di_version))186		magic_ok = true;187 188	if (xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,189			XFS_DINODE_CRC_OFF))190		crc_ok = true;191 192	if (magic_ok && unlinked_ok && crc_ok)193		return;194 195	if (!magic_ok) {196		dip->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);197		dip->di_version = 3;198	}199	if (!unlinked_ok)200		dip->di_next_unlinked = cpu_to_be32(NULLAGINO);201	xfs_dinode_calc_crc(mp, dip);202	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);203	xfs_trans_log_buf(tp, bp, ioffset,204				  ioffset + sizeof(struct xfs_dinode) - 1);205}206 207/* Make sure this inode cluster buffer can pass the inode buffer verifier. */208STATIC void209xrep_dinode_buf(210	struct xfs_scrub	*sc,211	struct xfs_buf		*bp)212{213	struct xfs_mount	*mp = sc->mp;214	int			i;215	int			ni;216 217	ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;218	for (i = 0; i < ni; i++)219		xrep_dinode_buf_core(sc, bp, i << mp->m_sb.sb_inodelog);220}221 222/* Reinitialize things that never change in an inode. */223STATIC void224xrep_dinode_header(225	struct xfs_scrub	*sc,226	struct xfs_dinode	*dip)227{228	trace_xrep_dinode_header(sc, dip);229 230	dip->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);231	if (!xfs_dinode_good_version(sc->mp, dip->di_version))232		dip->di_version = 3;233	dip->di_ino = cpu_to_be64(sc->sm->sm_ino);234	uuid_copy(&dip->di_uuid, &sc->mp->m_sb.sb_meta_uuid);235	dip->di_gen = cpu_to_be32(sc->sm->sm_gen);236}237 238/*239 * If this directory entry points to the scrub target inode, then the directory240 * we're scanning is the parent of the scrub target inode.241 */242STATIC int243xrep_dinode_findmode_dirent(244	struct xfs_scrub		*sc,245	struct xfs_inode		*dp,246	xfs_dir2_dataptr_t		dapos,247	const struct xfs_name		*name,248	xfs_ino_t			ino,249	void				*priv)250{251	struct xrep_inode		*ri = priv;252	int				error = 0;253 254	if (xchk_should_terminate(ri->sc, &error))255		return error;256 257	if (ino != sc->sm->sm_ino)258		return 0;259 260	/* Ignore garbage directory entry names. */261	if (name->len == 0 || !xfs_dir2_namecheck(name->name, name->len))262		return -EFSCORRUPTED;263 264	/* Don't pick up dot or dotdot entries; we only want child dirents. */265	if (xfs_dir2_samename(name, &xfs_name_dotdot) ||266	    xfs_dir2_samename(name, &xfs_name_dot))267		return 0;268 269	/*270	 * Uhoh, more than one parent for this inode and they don't agree on271	 * the file type?272	 */273	if (ri->alleged_ftype != XFS_DIR3_FT_UNKNOWN &&274	    ri->alleged_ftype != name->type) {275		trace_xrep_dinode_findmode_dirent_inval(ri->sc, dp, name->type,276				ri->alleged_ftype);277		return -EFSCORRUPTED;278	}279 280	/* We found a potential parent; remember the ftype. */281	trace_xrep_dinode_findmode_dirent(ri->sc, dp, name->type);282	ri->alleged_ftype = name->type;283	return 0;284}285 286/* Try to lock a directory, or wait a jiffy. */287static inline int288xrep_dinode_ilock_nowait(289	struct xfs_inode	*dp,290	unsigned int		lock_mode)291{292	if (xfs_ilock_nowait(dp, lock_mode))293		return true;294 295	schedule_timeout_killable(1);296	return false;297}298 299/*300 * Try to lock a directory to look for ftype hints.  Since we already hold the301 * AGI buffer, we cannot block waiting for the ILOCK because rename can take302 * the ILOCK and then try to lock AGIs.303 */304STATIC int305xrep_dinode_trylock_directory(306	struct xrep_inode	*ri,307	struct xfs_inode	*dp,308	unsigned int		*lock_modep)309{310	unsigned long		deadline = jiffies + msecs_to_jiffies(30000);311	unsigned int		lock_mode;312	int			error = 0;313 314	do {315		if (xchk_should_terminate(ri->sc, &error))316			return error;317 318		if (xfs_need_iread_extents(&dp->i_df))319			lock_mode = XFS_ILOCK_EXCL;320		else321			lock_mode = XFS_ILOCK_SHARED;322 323		if (xrep_dinode_ilock_nowait(dp, lock_mode)) {324			*lock_modep = lock_mode;325			return 0;326		}327	} while (!time_is_before_jiffies(deadline));328	return -EBUSY;329}330 331/*332 * If this is a directory, walk the dirents looking for any that point to the333 * scrub target inode.334 */335STATIC int336xrep_dinode_findmode_walk_directory(337	struct xrep_inode	*ri,338	struct xfs_inode	*dp)339{340	struct xfs_scrub	*sc = ri->sc;341	unsigned int		lock_mode;342	int			error = 0;343 344	/* Ignore temporary repair directories. */345	if (xrep_is_tempfile(dp))346		return 0;347 348	/*349	 * Scan the directory to see if there it contains an entry pointing to350	 * the directory that we are repairing.351	 */352	error = xrep_dinode_trylock_directory(ri, dp, &lock_mode);353	if (error)354		return error;355 356	/*357	 * If this directory is known to be sick, we cannot scan it reliably358	 * and must abort.359	 */360	if (xfs_inode_has_sickness(dp, XFS_SICK_INO_CORE |361				       XFS_SICK_INO_BMBTD |362				       XFS_SICK_INO_DIR)) {363		error = -EFSCORRUPTED;364		goto out_unlock;365	}366 367	/*368	 * We cannot complete our parent pointer scan if a directory looks as369	 * though it has been zapped by the inode record repair code.370	 */371	if (xchk_dir_looks_zapped(dp)) {372		error = -EBUSY;373		goto out_unlock;374	}375 376	error = xchk_dir_walk(sc, dp, xrep_dinode_findmode_dirent, ri);377	if (error)378		goto out_unlock;379 380out_unlock:381	xfs_iunlock(dp, lock_mode);382	return error;383}384 385/*386 * Try to find the mode of the inode being repaired by looking for directories387 * that point down to this file.388 */389STATIC int390xrep_dinode_find_mode(391	struct xrep_inode	*ri,392	uint16_t		*mode)393{394	struct xfs_scrub	*sc = ri->sc;395	struct xfs_inode	*dp;396	int			error;397 398	/* No ftype means we have no other metadata to consult. */399	if (!xfs_has_ftype(sc->mp)) {400		*mode = S_IFREG;401		return 0;402	}403 404	/*405	 * Scan all directories for parents that might point down to this406	 * inode.  Skip the inode being repaired during the scan since it407	 * cannot be its own parent.  Note that we still hold the AGI locked408	 * so there's a real possibility that _iscan_iter can return EBUSY.409	 */410	xchk_iscan_start(sc, 5000, 100, &ri->ftype_iscan);411	xchk_iscan_set_agi_trylock(&ri->ftype_iscan);412	ri->ftype_iscan.skip_ino = sc->sm->sm_ino;413	ri->alleged_ftype = XFS_DIR3_FT_UNKNOWN;414	while ((error = xchk_iscan_iter(&ri->ftype_iscan, &dp)) == 1) {415		if (S_ISDIR(VFS_I(dp)->i_mode))416			error = xrep_dinode_findmode_walk_directory(ri, dp);417		xchk_iscan_mark_visited(&ri->ftype_iscan, dp);418		xchk_irele(sc, dp);419		if (error < 0)420			break;421		if (xchk_should_terminate(sc, &error))422			break;423	}424	xchk_iscan_iter_finish(&ri->ftype_iscan);425	xchk_iscan_teardown(&ri->ftype_iscan);426 427	if (error == -EBUSY) {428		if (ri->alleged_ftype != XFS_DIR3_FT_UNKNOWN) {429			/*430			 * If we got an EBUSY after finding at least one431			 * dirent, that means the scan found an inode on the432			 * inactivation list and could not open it.  Accept the433			 * alleged ftype and install a new mode below.434			 */435			error = 0;436		} else if (!(sc->flags & XCHK_TRY_HARDER)) {437			/*438			 * Otherwise, retry the operation one time to see if439			 * the reason for the delay is an inode from the same440			 * cluster buffer waiting on the inactivation list.441			 */442			error = -EDEADLOCK;443		}444	}445	if (error)446		return error;447 448	/*449	 * Convert the discovered ftype into the file mode.  If all else fails,450	 * return S_IFREG.451	 */452	switch (ri->alleged_ftype) {453	case XFS_DIR3_FT_DIR:454		*mode = S_IFDIR;455		break;456	case XFS_DIR3_FT_WHT:457	case XFS_DIR3_FT_CHRDEV:458		*mode = S_IFCHR;459		break;460	case XFS_DIR3_FT_BLKDEV:461		*mode = S_IFBLK;462		break;463	case XFS_DIR3_FT_FIFO:464		*mode = S_IFIFO;465		break;466	case XFS_DIR3_FT_SOCK:467		*mode = S_IFSOCK;468		break;469	case XFS_DIR3_FT_SYMLINK:470		*mode = S_IFLNK;471		break;472	default:473		*mode = S_IFREG;474		break;475	}476	return 0;477}478 479/* Turn di_mode into /something/ recognizable.  Returns true if we succeed. */480STATIC int481xrep_dinode_mode(482	struct xrep_inode	*ri,483	struct xfs_dinode	*dip)484{485	struct xfs_scrub	*sc = ri->sc;486	uint16_t		mode = be16_to_cpu(dip->di_mode);487	int			error;488 489	trace_xrep_dinode_mode(sc, dip);490 491	if (mode == 0 || xfs_mode_to_ftype(mode) != XFS_DIR3_FT_UNKNOWN)492		return 0;493 494	/* Try to fix the mode.  If we cannot, then leave everything alone. */495	error = xrep_dinode_find_mode(ri, &mode);496	switch (error) {497	case -EINTR:498	case -EBUSY:499	case -EDEADLOCK:500		/* temporary failure or fatal signal */501		return error;502	case 0:503		/* found mode */504		break;505	default:506		/* some other error, assume S_IFREG */507		mode = S_IFREG;508		break;509	}510 511	/* bad mode, so we set it to a file that only root can read */512	dip->di_mode = cpu_to_be16(mode);513	dip->di_uid = 0;514	dip->di_gid = 0;515	ri->zap_acls = true;516	return 0;517}518 519/* Fix unused link count fields having nonzero values. */520STATIC void521xrep_dinode_nlinks(522	struct xfs_dinode	*dip)523{524	if (dip->di_version > 1)525		dip->di_onlink = 0;526	else527		dip->di_nlink = 0;528}529 530/* Fix any conflicting flags that the verifiers complain about. */531STATIC void532xrep_dinode_flags(533	struct xfs_scrub	*sc,534	struct xfs_dinode	*dip,535	bool			isrt)536{537	struct xfs_mount	*mp = sc->mp;538	uint64_t		flags2 = be64_to_cpu(dip->di_flags2);539	uint16_t		flags = be16_to_cpu(dip->di_flags);540	uint16_t		mode = be16_to_cpu(dip->di_mode);541 542	trace_xrep_dinode_flags(sc, dip);543 544	if (isrt)545		flags |= XFS_DIFLAG_REALTIME;546	else547		flags &= ~XFS_DIFLAG_REALTIME;548 549	/*550	 * For regular files on a reflink filesystem, set the REFLINK flag to551	 * protect shared extents.  A later stage will actually check those552	 * extents and clear the flag if possible.553	 */554	if (xfs_has_reflink(mp) && S_ISREG(mode))555		flags2 |= XFS_DIFLAG2_REFLINK;556	else557		flags2 &= ~(XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE);558	if (flags & XFS_DIFLAG_REALTIME)559		flags2 &= ~XFS_DIFLAG2_REFLINK;560	if (!xfs_has_bigtime(mp))561		flags2 &= ~XFS_DIFLAG2_BIGTIME;562	if (!xfs_has_large_extent_counts(mp))563		flags2 &= ~XFS_DIFLAG2_NREXT64;564	if (flags2 & XFS_DIFLAG2_NREXT64)565		dip->di_nrext64_pad = 0;566	else if (dip->di_version >= 3)567		dip->di_v3_pad = 0;568	dip->di_flags = cpu_to_be16(flags);569	dip->di_flags2 = cpu_to_be64(flags2);570}571 572/*573 * Blow out symlink; now it points nowhere.  We don't have to worry about574 * incore state because this inode is failing the verifiers.575 */576STATIC void577xrep_dinode_zap_symlink(578	struct xrep_inode	*ri,579	struct xfs_dinode	*dip)580{581	struct xfs_scrub	*sc = ri->sc;582	char			*p;583 584	trace_xrep_dinode_zap_symlink(sc, dip);585 586	dip->di_format = XFS_DINODE_FMT_LOCAL;587	dip->di_size = cpu_to_be64(1);588	p = XFS_DFORK_PTR(dip, XFS_DATA_FORK);589	*p = '?';590	ri->ino_sick_mask |= XFS_SICK_INO_SYMLINK_ZAPPED;591}592 593/*594 * Blow out dir, make the parent point to the root.  In the future repair will595 * reconstruct this directory for us.  Note that there's no in-core directory596 * inode because the sf verifier tripped, so we don't have to worry about the597 * dentry cache.598 */599STATIC void600xrep_dinode_zap_dir(601	struct xrep_inode	*ri,602	struct xfs_dinode	*dip)603{604	struct xfs_scrub	*sc = ri->sc;605	struct xfs_mount	*mp = sc->mp;606	struct xfs_dir2_sf_hdr	*sfp;607	int			i8count;608 609	trace_xrep_dinode_zap_dir(sc, dip);610 611	dip->di_format = XFS_DINODE_FMT_LOCAL;612	i8count = mp->m_sb.sb_rootino > XFS_DIR2_MAX_SHORT_INUM;613	sfp = XFS_DFORK_PTR(dip, XFS_DATA_FORK);614	sfp->count = 0;615	sfp->i8count = i8count;616	xfs_dir2_sf_put_parent_ino(sfp, mp->m_sb.sb_rootino);617	dip->di_size = cpu_to_be64(xfs_dir2_sf_hdr_size(i8count));618	ri->ino_sick_mask |= XFS_SICK_INO_DIR_ZAPPED;619}620 621/* Make sure we don't have a garbage file size. */622STATIC void623xrep_dinode_size(624	struct xrep_inode	*ri,625	struct xfs_dinode	*dip)626{627	struct xfs_scrub	*sc = ri->sc;628	uint64_t		size = be64_to_cpu(dip->di_size);629	uint16_t		mode = be16_to_cpu(dip->di_mode);630 631	trace_xrep_dinode_size(sc, dip);632 633	switch (mode & S_IFMT) {634	case S_IFIFO:635	case S_IFCHR:636	case S_IFBLK:637	case S_IFSOCK:638		/* di_size can't be nonzero for special files */639		dip->di_size = 0;640		break;641	case S_IFREG:642		/* Regular files can't be larger than 2^63-1 bytes. */643		dip->di_size = cpu_to_be64(size & ~(1ULL << 63));644		break;645	case S_IFLNK:646		/*647		 * Truncate ridiculously oversized symlinks.  If the size is648		 * zero, reset it to point to the current directory.  Both of649		 * these conditions trigger dinode verifier errors, so there650		 * is no in-core state to reset.651		 */652		if (size > XFS_SYMLINK_MAXLEN)653			dip->di_size = cpu_to_be64(XFS_SYMLINK_MAXLEN);654		else if (size == 0)655			xrep_dinode_zap_symlink(ri, dip);656		break;657	case S_IFDIR:658		/*659		 * Directories can't have a size larger than 32G.  If the size660		 * is zero, reset it to an empty directory.  Both of these661		 * conditions trigger dinode verifier errors, so there is no662		 * in-core state to reset.663		 */664		if (size > XFS_DIR2_SPACE_SIZE)665			dip->di_size = cpu_to_be64(XFS_DIR2_SPACE_SIZE);666		else if (size == 0)667			xrep_dinode_zap_dir(ri, dip);668		break;669	}670}671 672/* Fix extent size hints. */673STATIC void674xrep_dinode_extsize_hints(675	struct xfs_scrub	*sc,676	struct xfs_dinode	*dip)677{678	struct xfs_mount	*mp = sc->mp;679	uint64_t		flags2 = be64_to_cpu(dip->di_flags2);680	uint16_t		flags = be16_to_cpu(dip->di_flags);681	uint16_t		mode = be16_to_cpu(dip->di_mode);682 683	xfs_failaddr_t		fa;684 685	trace_xrep_dinode_extsize_hints(sc, dip);686 687	fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),688			mode, flags);689	if (fa) {690		dip->di_extsize = 0;691		dip->di_flags &= ~cpu_to_be16(XFS_DIFLAG_EXTSIZE |692					      XFS_DIFLAG_EXTSZINHERIT);693	}694 695	if (dip->di_version < 3)696		return;697 698	fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),699			mode, flags, flags2);700	if (fa) {701		dip->di_cowextsize = 0;702		dip->di_flags2 &= ~cpu_to_be64(XFS_DIFLAG2_COWEXTSIZE);703	}704}705 706/* Count extents and blocks for an inode given an rmap. */707STATIC int708xrep_dinode_walk_rmap(709	struct xfs_btree_cur		*cur,710	const struct xfs_rmap_irec	*rec,711	void				*priv)712{713	struct xrep_inode		*ri = priv;714	int				error = 0;715 716	if (xchk_should_terminate(ri->sc, &error))717		return error;718 719	/* We only care about this inode. */720	if (rec->rm_owner != ri->sc->sm->sm_ino)721		return 0;722 723	if (rec->rm_flags & XFS_RMAP_ATTR_FORK) {724		ri->attr_blocks += rec->rm_blockcount;725		if (!(rec->rm_flags & XFS_RMAP_BMBT_BLOCK))726			ri->attr_extents++;727 728		return 0;729	}730 731	ri->data_blocks += rec->rm_blockcount;732	if (!(rec->rm_flags & XFS_RMAP_BMBT_BLOCK))733		ri->data_extents++;734 735	return 0;736}737 738/* Count extents and blocks for an inode from all AG rmap data. */739STATIC int740xrep_dinode_count_ag_rmaps(741	struct xrep_inode	*ri,742	struct xfs_perag	*pag)743{744	struct xfs_btree_cur	*cur;745	struct xfs_buf		*agf;746	int			error;747 748	error = xfs_alloc_read_agf(pag, ri->sc->tp, 0, &agf);749	if (error)750		return error;751 752	cur = xfs_rmapbt_init_cursor(ri->sc->mp, ri->sc->tp, agf, pag);753	error = xfs_rmap_query_all(cur, xrep_dinode_walk_rmap, ri);754	xfs_btree_del_cursor(cur, error);755	xfs_trans_brelse(ri->sc->tp, agf);756	return error;757}758 759/* Count extents and blocks for a given inode from all rmap data. */760STATIC int761xrep_dinode_count_rmaps(762	struct xrep_inode	*ri)763{764	struct xfs_perag	*pag;765	xfs_agnumber_t		agno;766	int			error;767 768	if (!xfs_has_rmapbt(ri->sc->mp) || xfs_has_realtime(ri->sc->mp))769		return -EOPNOTSUPP;770 771	for_each_perag(ri->sc->mp, agno, pag) {772		error = xrep_dinode_count_ag_rmaps(ri, pag);773		if (error) {774			xfs_perag_rele(pag);775			return error;776		}777	}778 779	/* Can't have extents on both the rt and the data device. */780	if (ri->data_extents && ri->rt_extents)781		return -EFSCORRUPTED;782 783	trace_xrep_dinode_count_rmaps(ri->sc,784			ri->data_blocks, ri->rt_blocks, ri->attr_blocks,785			ri->data_extents, ri->rt_extents, ri->attr_extents);786	return 0;787}788 789/* Return true if this extents-format ifork looks like garbage. */790STATIC bool791xrep_dinode_bad_extents_fork(792	struct xfs_scrub	*sc,793	struct xfs_dinode	*dip,794	unsigned int		dfork_size,795	int			whichfork)796{797	struct xfs_bmbt_irec	new;798	struct xfs_bmbt_rec	*dp;799	xfs_extnum_t		nex;800	bool			isrt;801	unsigned int		i;802 803	nex = xfs_dfork_nextents(dip, whichfork);804	if (nex > dfork_size / sizeof(struct xfs_bmbt_rec))805		return true;806 807	dp = XFS_DFORK_PTR(dip, whichfork);808 809	isrt = dip->di_flags & cpu_to_be16(XFS_DIFLAG_REALTIME);810	for (i = 0; i < nex; i++, dp++) {811		xfs_failaddr_t	fa;812 813		xfs_bmbt_disk_get_all(dp, &new);814		fa = xfs_bmap_validate_extent_raw(sc->mp, isrt, whichfork,815				&new);816		if (fa)817			return true;818	}819 820	return false;821}822 823/* Return true if this btree-format ifork looks like garbage. */824STATIC bool825xrep_dinode_bad_bmbt_fork(826	struct xfs_scrub	*sc,827	struct xfs_dinode	*dip,828	unsigned int		dfork_size,829	int			whichfork)830{831	struct xfs_bmdr_block	*dfp;832	xfs_extnum_t		nex;833	unsigned int		i;834	unsigned int		dmxr;835	unsigned int		nrecs;836	unsigned int		level;837 838	nex = xfs_dfork_nextents(dip, whichfork);839	if (nex <= dfork_size / sizeof(struct xfs_bmbt_rec))840		return true;841 842	if (dfork_size < sizeof(struct xfs_bmdr_block))843		return true;844 845	dfp = XFS_DFORK_PTR(dip, whichfork);846	nrecs = be16_to_cpu(dfp->bb_numrecs);847	level = be16_to_cpu(dfp->bb_level);848 849	if (nrecs == 0 || xfs_bmdr_space_calc(nrecs) > dfork_size)850		return true;851	if (level == 0 || level >= XFS_BM_MAXLEVELS(sc->mp, whichfork))852		return true;853 854	dmxr = xfs_bmdr_maxrecs(dfork_size, 0);855	for (i = 1; i <= nrecs; i++) {856		struct xfs_bmbt_key	*fkp;857		xfs_bmbt_ptr_t		*fpp;858		xfs_fileoff_t		fileoff;859		xfs_fsblock_t		fsbno;860 861		fkp = xfs_bmdr_key_addr(dfp, i);862		fileoff = be64_to_cpu(fkp->br_startoff);863		if (!xfs_verify_fileoff(sc->mp, fileoff))864			return true;865 866		fpp = xfs_bmdr_ptr_addr(dfp, i, dmxr);867		fsbno = be64_to_cpu(*fpp);868		if (!xfs_verify_fsbno(sc->mp, fsbno))869			return true;870	}871 872	return false;873}874 875/*876 * Check the data fork for things that will fail the ifork verifiers or the877 * ifork formatters.878 */879STATIC bool880xrep_dinode_check_dfork(881	struct xfs_scrub	*sc,882	struct xfs_dinode	*dip,883	uint16_t		mode)884{885	void			*dfork_ptr;886	int64_t			data_size;887	unsigned int		fmt;888	unsigned int		dfork_size;889 890	/*891	 * Verifier functions take signed int64_t, so check for bogus negative892	 * values first.893	 */894	data_size = be64_to_cpu(dip->di_size);895	if (data_size < 0)896		return true;897 898	fmt = XFS_DFORK_FORMAT(dip, XFS_DATA_FORK);899	switch (mode & S_IFMT) {900	case S_IFIFO:901	case S_IFCHR:902	case S_IFBLK:903	case S_IFSOCK:904		if (fmt != XFS_DINODE_FMT_DEV)905			return true;906		break;907	case S_IFREG:908		if (fmt == XFS_DINODE_FMT_LOCAL)909			return true;910		fallthrough;911	case S_IFLNK:912	case S_IFDIR:913		switch (fmt) {914		case XFS_DINODE_FMT_LOCAL:915		case XFS_DINODE_FMT_EXTENTS:916		case XFS_DINODE_FMT_BTREE:917			break;918		default:919			return true;920		}921		break;922	default:923		return true;924	}925 926	dfork_size = XFS_DFORK_SIZE(dip, sc->mp, XFS_DATA_FORK);927	dfork_ptr = XFS_DFORK_PTR(dip, XFS_DATA_FORK);928 929	switch (fmt) {930	case XFS_DINODE_FMT_DEV:931		break;932	case XFS_DINODE_FMT_LOCAL:933		/* dir/symlink structure cannot be larger than the fork */934		if (data_size > dfork_size)935			return true;936		/* directory structure must pass verification. */937		if (S_ISDIR(mode) &&938		    xfs_dir2_sf_verify(sc->mp, dfork_ptr, data_size) != NULL)939			return true;940		/* symlink structure must pass verification. */941		if (S_ISLNK(mode) &&942		    xfs_symlink_shortform_verify(dfork_ptr, data_size) != NULL)943			return true;944		break;945	case XFS_DINODE_FMT_EXTENTS:946		if (xrep_dinode_bad_extents_fork(sc, dip, dfork_size,947				XFS_DATA_FORK))948			return true;949		break;950	case XFS_DINODE_FMT_BTREE:951		if (xrep_dinode_bad_bmbt_fork(sc, dip, dfork_size,952				XFS_DATA_FORK))953			return true;954		break;955	default:956		return true;957	}958 959	return false;960}961 962static void963xrep_dinode_set_data_nextents(964	struct xfs_dinode	*dip,965	xfs_extnum_t		nextents)966{967	if (xfs_dinode_has_large_extent_counts(dip))968		dip->di_big_nextents = cpu_to_be64(nextents);969	else970		dip->di_nextents = cpu_to_be32(nextents);971}972 973static void974xrep_dinode_set_attr_nextents(975	struct xfs_dinode	*dip,976	xfs_extnum_t		nextents)977{978	if (xfs_dinode_has_large_extent_counts(dip))979		dip->di_big_anextents = cpu_to_be32(nextents);980	else981		dip->di_anextents = cpu_to_be16(nextents);982}983 984/* Reset the data fork to something sane. */985STATIC void986xrep_dinode_zap_dfork(987	struct xrep_inode	*ri,988	struct xfs_dinode	*dip,989	uint16_t		mode)990{991	struct xfs_scrub	*sc = ri->sc;992 993	trace_xrep_dinode_zap_dfork(sc, dip);994 995	ri->ino_sick_mask |= XFS_SICK_INO_BMBTD_ZAPPED;996 997	xrep_dinode_set_data_nextents(dip, 0);998	ri->data_blocks = 0;999	ri->rt_blocks = 0;1000 1001	/* Special files always get reset to DEV */1002	switch (mode & S_IFMT) {1003	case S_IFIFO:1004	case S_IFCHR:1005	case S_IFBLK:1006	case S_IFSOCK:1007		dip->di_format = XFS_DINODE_FMT_DEV;1008		dip->di_size = 0;1009		return;1010	}1011 1012	/*1013	 * If we have data extents, reset to an empty map and hope the user1014	 * will run the bmapbtd checker next.1015	 */1016	if (ri->data_extents || ri->rt_extents || S_ISREG(mode)) {1017		dip->di_format = XFS_DINODE_FMT_EXTENTS;1018		return;1019	}1020 1021	/* Otherwise, reset the local format to the minimum. */1022	switch (mode & S_IFMT) {1023	case S_IFLNK:1024		xrep_dinode_zap_symlink(ri, dip);1025		break;1026	case S_IFDIR:1027		xrep_dinode_zap_dir(ri, dip);1028		break;1029	}1030}1031 1032/*1033 * Check the attr fork for things that will fail the ifork verifiers or the1034 * ifork formatters.1035 */1036STATIC bool1037xrep_dinode_check_afork(1038	struct xfs_scrub		*sc,1039	struct xfs_dinode		*dip)1040{1041	struct xfs_attr_sf_hdr		*afork_ptr;1042	size_t				attr_size;1043	unsigned int			afork_size;1044 1045	if (XFS_DFORK_BOFF(dip) == 0)1046		return dip->di_aformat != XFS_DINODE_FMT_EXTENTS ||1047		       xfs_dfork_attr_extents(dip) != 0;1048 1049	afork_size = XFS_DFORK_SIZE(dip, sc->mp, XFS_ATTR_FORK);1050	afork_ptr = XFS_DFORK_PTR(dip, XFS_ATTR_FORK);1051 1052	switch (XFS_DFORK_FORMAT(dip, XFS_ATTR_FORK)) {1053	case XFS_DINODE_FMT_LOCAL:1054		/* Fork has to be large enough to extract the xattr size. */1055		if (afork_size < sizeof(struct xfs_attr_sf_hdr))1056			return true;1057 1058		/* xattr structure cannot be larger than the fork */1059		attr_size = be16_to_cpu(afork_ptr->totsize);1060		if (attr_size > afork_size)1061			return true;1062 1063		/* xattr structure must pass verification. */1064		return xfs_attr_shortform_verify(afork_ptr, attr_size) != NULL;1065	case XFS_DINODE_FMT_EXTENTS:1066		if (xrep_dinode_bad_extents_fork(sc, dip, afork_size,1067					XFS_ATTR_FORK))1068			return true;1069		break;1070	case XFS_DINODE_FMT_BTREE:1071		if (xrep_dinode_bad_bmbt_fork(sc, dip, afork_size,1072					XFS_ATTR_FORK))1073			return true;1074		break;1075	default:1076		return true;1077	}1078 1079	return false;1080}1081 1082/*1083 * Reset the attr fork to empty.  Since the attr fork could have contained1084 * ACLs, make the file readable only by root.1085 */1086STATIC void1087xrep_dinode_zap_afork(1088	struct xrep_inode	*ri,1089	struct xfs_dinode	*dip,1090	uint16_t		mode)1091{1092	struct xfs_scrub	*sc = ri->sc;1093 1094	trace_xrep_dinode_zap_afork(sc, dip);1095 1096	ri->ino_sick_mask |= XFS_SICK_INO_BMBTA_ZAPPED;1097 1098	dip->di_aformat = XFS_DINODE_FMT_EXTENTS;1099	xrep_dinode_set_attr_nextents(dip, 0);1100	ri->attr_blocks = 0;1101 1102	/*1103	 * If the data fork is in btree format, removing the attr fork entirely1104	 * might cause verifier failures if the next level down in the bmbt1105	 * could now fit in the data fork area.1106	 */1107	if (dip->di_format != XFS_DINODE_FMT_BTREE)1108		dip->di_forkoff = 0;1109	dip->di_mode = cpu_to_be16(mode & ~0777);1110	dip->di_uid = 0;1111	dip->di_gid = 0;1112}1113 1114/* Make sure the fork offset is a sensible value. */1115STATIC void1116xrep_dinode_ensure_forkoff(1117	struct xrep_inode	*ri,1118	struct xfs_dinode	*dip,1119	uint16_t		mode)1120{1121	struct xfs_bmdr_block	*bmdr;1122	struct xfs_scrub	*sc = ri->sc;1123	xfs_extnum_t		attr_extents, data_extents;1124	size_t			bmdr_minsz = xfs_bmdr_space_calc(1);1125	unsigned int		lit_sz = XFS_LITINO(sc->mp);1126	unsigned int		afork_min, dfork_min;1127 1128	trace_xrep_dinode_ensure_forkoff(sc, dip);1129 1130	/*1131	 * Before calling this function, xrep_dinode_core ensured that both1132	 * forks actually fit inside their respective literal areas.  If this1133	 * was not the case, the fork was reset to FMT_EXTENTS with zero1134	 * records.  If the rmapbt scan found attr or data fork blocks, this1135	 * will be noted in the dinode_stats, and we must leave enough room1136	 * for the bmap repair code to reconstruct the mapping structure.1137	 *1138	 * First, compute the minimum space required for the attr fork.1139	 */1140	switch (dip->di_aformat) {1141	case XFS_DINODE_FMT_LOCAL:1142		/*1143		 * If we still have a shortform xattr structure at all, that1144		 * means the attr fork area was exactly large enough to fit1145		 * the sf structure.1146		 */1147		afork_min = XFS_DFORK_SIZE(dip, sc->mp, XFS_ATTR_FORK);1148		break;1149	case XFS_DINODE_FMT_EXTENTS:1150		attr_extents = xfs_dfork_attr_extents(dip);1151		if (attr_extents) {1152			/*1153			 * We must maintain sufficient space to hold the entire1154			 * extent map array in the data fork.  Note that we1155			 * previously zapped the fork if it had no chance of1156			 * fitting in the inode.1157			 */1158			afork_min = sizeof(struct xfs_bmbt_rec) * attr_extents;1159		} else if (ri->attr_extents > 0) {1160			/*1161			 * The attr fork thinks it has zero extents, but we1162			 * found some xattr extents.  We need to leave enough1163			 * empty space here so that the incore attr fork will1164			 * get created (and hence trigger the attr fork bmap1165			 * repairer).1166			 */1167			afork_min = bmdr_minsz;1168		} else {1169			/* No extents on disk or found in rmapbt. */1170			afork_min = 0;1171		}1172		break;1173	case XFS_DINODE_FMT_BTREE:1174		/* Must have space for btree header and key/pointers. */1175		bmdr = XFS_DFORK_PTR(dip, XFS_ATTR_FORK);1176		afork_min = xfs_bmap_broot_space(sc->mp, bmdr);1177		break;1178	default:1179		/* We should never see any other formats. */1180		afork_min = 0;1181		break;1182	}1183 1184	/* Compute the minimum space required for the data fork. */1185	switch (dip->di_format) {1186	case XFS_DINODE_FMT_DEV:1187		dfork_min = sizeof(__be32);1188		break;1189	case XFS_DINODE_FMT_UUID:1190		dfork_min = sizeof(uuid_t);1191		break;1192	case XFS_DINODE_FMT_LOCAL:1193		/*1194		 * If we still have a shortform data fork at all, that means1195		 * the data fork area was large enough to fit whatever was in1196		 * there.1197		 */1198		dfork_min = be64_to_cpu(dip->di_size);1199		break;1200	case XFS_DINODE_FMT_EXTENTS:1201		data_extents = xfs_dfork_data_extents(dip);1202		if (data_extents) {1203			/*1204			 * We must maintain sufficient space to hold the entire1205			 * extent map array in the data fork.  Note that we1206			 * previously zapped the fork if it had no chance of1207			 * fitting in the inode.1208			 */1209			dfork_min = sizeof(struct xfs_bmbt_rec) * data_extents;1210		} else if (ri->data_extents > 0 || ri->rt_extents > 0) {1211			/*1212			 * The data fork thinks it has zero extents, but we1213			 * found some data extents.  We need to leave enough1214			 * empty space here so that the data fork bmap repair1215			 * will recover the mappings.1216			 */1217			dfork_min = bmdr_minsz;1218		} else {1219			/* No extents on disk or found in rmapbt. */1220			dfork_min = 0;1221		}1222		break;1223	case XFS_DINODE_FMT_BTREE:1224		/* Must have space for btree header and key/pointers. */1225		bmdr = XFS_DFORK_PTR(dip, XFS_DATA_FORK);1226		dfork_min = xfs_bmap_broot_space(sc->mp, bmdr);1227		break;1228	default:1229		dfork_min = 0;1230		break;1231	}1232 1233	/*1234	 * Round all values up to the nearest 8 bytes, because that is the1235	 * precision of di_forkoff.1236	 */1237	afork_min = roundup(afork_min, 8);1238	dfork_min = roundup(dfork_min, 8);1239	bmdr_minsz = roundup(bmdr_minsz, 8);1240 1241	ASSERT(dfork_min <= lit_sz);1242	ASSERT(afork_min <= lit_sz);1243 1244	/*1245	 * If the data fork was zapped and we don't have enough space for the1246	 * recovery fork, move the attr fork up.1247	 */1248	if (dip->di_format == XFS_DINODE_FMT_EXTENTS &&1249	    xfs_dfork_data_extents(dip) == 0 &&1250	    (ri->data_extents > 0 || ri->rt_extents > 0) &&1251	    bmdr_minsz > XFS_DFORK_DSIZE(dip, sc->mp)) {1252		if (bmdr_minsz + afork_min > lit_sz) {1253			/*1254			 * The attr for and the stub fork we need to recover1255			 * the data fork won't both fit.  Zap the attr fork.1256			 */1257			xrep_dinode_zap_afork(ri, dip, mode);1258			afork_min = bmdr_minsz;1259		} else {1260			void	*before, *after;1261 1262			/* Otherwise, just slide the attr fork up. */1263			before = XFS_DFORK_APTR(dip);1264			dip->di_forkoff = bmdr_minsz >> 3;1265			after = XFS_DFORK_APTR(dip);1266			memmove(after, before, XFS_DFORK_ASIZE(dip, sc->mp));1267		}1268	}1269 1270	/*1271	 * If the attr fork was zapped and we don't have enough space for the1272	 * recovery fork, move the attr fork down.1273	 */1274	if (dip->di_aformat == XFS_DINODE_FMT_EXTENTS &&1275	    xfs_dfork_attr_extents(dip) == 0 &&1276	    ri->attr_extents > 0 &&1277	    bmdr_minsz > XFS_DFORK_ASIZE(dip, sc->mp)) {1278		if (dip->di_format == XFS_DINODE_FMT_BTREE) {1279			/*1280			 * If the data fork is in btree format then we can't1281			 * adjust forkoff because that runs the risk of1282			 * violating the extents/btree format transition rules.1283			 */1284		} else if (bmdr_minsz + dfork_min > lit_sz) {1285			/*1286			 * If we can't move the attr fork, too bad, we lose the1287			 * attr fork and leak its blocks.1288			 */1289			xrep_dinode_zap_afork(ri, dip, mode);1290		} else {1291			/*1292			 * Otherwise, just slide the attr fork down.  The attr1293			 * fork is empty, so we don't have any old contents to1294			 * move here.1295			 */1296			dip->di_forkoff = (lit_sz - bmdr_minsz) >> 3;1297		}1298	}1299}1300 1301/*1302 * Zap the data/attr forks if we spot anything that isn't going to pass the1303 * ifork verifiers or the ifork formatters, because we need to get the inode1304 * into good enough shape that the higher level repair functions can run.1305 */1306STATIC void1307xrep_dinode_zap_forks(1308	struct xrep_inode	*ri,1309	struct xfs_dinode	*dip)1310{1311	struct xfs_scrub	*sc = ri->sc;1312	xfs_extnum_t		data_extents;1313	xfs_extnum_t		attr_extents;1314	xfs_filblks_t		nblocks;1315	uint16_t		mode;1316	bool			zap_datafork = false;1317	bool			zap_attrfork = ri->zap_acls;1318 1319	trace_xrep_dinode_zap_forks(sc, dip);1320 1321	mode = be16_to_cpu(dip->di_mode);1322 1323	data_extents = xfs_dfork_data_extents(dip);1324	attr_extents = xfs_dfork_attr_extents(dip);1325	nblocks = be64_to_cpu(dip->di_nblocks);1326 1327	/* Inode counters don't make sense? */1328	if (data_extents > nblocks)1329		zap_datafork = true;1330	if (attr_extents > nblocks)1331		zap_attrfork = true;1332	if (data_extents + attr_extents > nblocks)1333		zap_datafork = zap_attrfork = true;1334 1335	if (!zap_datafork)1336		zap_datafork = xrep_dinode_check_dfork(sc, dip, mode);1337	if (!zap_attrfork)1338		zap_attrfork = xrep_dinode_check_afork(sc, dip);1339 1340	/* Zap whatever's bad. */1341	if (zap_attrfork)1342		xrep_dinode_zap_afork(ri, dip, mode);1343	if (zap_datafork)1344		xrep_dinode_zap_dfork(ri, dip, mode);1345	xrep_dinode_ensure_forkoff(ri, dip, mode);1346 1347	/*1348	 * Zero di_nblocks if we don't have any extents at all to satisfy the1349	 * buffer verifier.1350	 */1351	data_extents = xfs_dfork_data_extents(dip);1352	attr_extents = xfs_dfork_attr_extents(dip);1353	if (data_extents + attr_extents == 0)1354		dip->di_nblocks = 0;1355}1356 1357/* Inode didn't pass dinode verifiers, so fix the raw buffer and retry iget. */1358STATIC int1359xrep_dinode_core(1360	struct xrep_inode	*ri)1361{1362	struct xfs_scrub	*sc = ri->sc;1363	struct xfs_buf		*bp;1364	struct xfs_dinode	*dip;1365	xfs_ino_t		ino = sc->sm->sm_ino;1366	int			error;1367	int			iget_error;1368 1369	/* Figure out what this inode had mapped in both forks. */1370	error = xrep_dinode_count_rmaps(ri);1371	if (error)1372		return error;1373 1374	/* Read the inode cluster buffer. */1375	error = xfs_trans_read_buf(sc->mp, sc->tp, sc->mp->m_ddev_targp,1376			ri->imap.im_blkno, ri->imap.im_len, XBF_UNMAPPED, &bp,1377			NULL);1378	if (error)1379		return error;1380 1381	/* Make sure we can pass the inode buffer verifier. */1382	xrep_dinode_buf(sc, bp);1383	bp->b_ops = &xfs_inode_buf_ops;1384 1385	/* Fix everything the verifier will complain about. */1386	dip = xfs_buf_offset(bp, ri->imap.im_boffset);1387	xrep_dinode_header(sc, dip);1388	iget_error = xrep_dinode_mode(ri, dip);1389	if (iget_error)1390		goto write;1391	xrep_dinode_nlinks(dip);1392	xrep_dinode_flags(sc, dip, ri->rt_extents > 0);1393	xrep_dinode_size(ri, dip);1394	xrep_dinode_extsize_hints(sc, dip);1395	xrep_dinode_zap_forks(ri, dip);1396 1397write:1398	/* Write out the inode. */1399	trace_xrep_dinode_fixed(sc, dip);1400	xfs_dinode_calc_crc(sc->mp, dip);1401	xfs_trans_buf_set_type(sc->tp, bp, XFS_BLFT_DINO_BUF);1402	xfs_trans_log_buf(sc->tp, bp, ri->imap.im_boffset,1403			ri->imap.im_boffset + sc->mp->m_sb.sb_inodesize - 1);1404 1405	/*1406	 * In theory, we've fixed the ondisk inode record enough that we should1407	 * be able to load the inode into the cache.  Try to iget that inode1408	 * now while we hold the AGI and the inode cluster buffer and take the1409	 * IOLOCK so that we can continue with repairs without anyone else1410	 * accessing the inode.  If iget fails, we still need to commit the1411	 * changes.1412	 */1413	if (!iget_error)1414		iget_error = xchk_iget(sc, ino, &sc->ip);1415	if (!iget_error)1416		xchk_ilock(sc, XFS_IOLOCK_EXCL);1417 1418	/*1419	 * Commit the inode cluster buffer updates and drop the AGI buffer that1420	 * we've been holding since scrub setup.  From here on out, repairs1421	 * deal only with the cached inode.1422	 */1423	error = xrep_trans_commit(sc);1424	if (error)1425		return error;1426 1427	if (iget_error)1428		return iget_error;1429 1430	error = xchk_trans_alloc(sc, 0);1431	if (error)1432		return error;1433 1434	error = xrep_ino_dqattach(sc);1435	if (error)1436		return error;1437 1438	xchk_ilock(sc, XFS_ILOCK_EXCL);1439	if (ri->ino_sick_mask)1440		xfs_inode_mark_sick(sc->ip, ri->ino_sick_mask);1441	return 0;1442}1443 1444/* Fix everything xfs_dinode_verify cares about. */1445STATIC int1446xrep_dinode_problems(1447	struct xrep_inode	*ri)1448{1449	struct xfs_scrub	*sc = ri->sc;1450	int			error;1451 1452	error = xrep_dinode_core(ri);1453	if (error)1454		return error;1455 1456	/* We had to fix a totally busted inode, schedule quotacheck. */1457	if (XFS_IS_UQUOTA_ON(sc->mp))1458		xrep_force_quotacheck(sc, XFS_DQTYPE_USER);1459	if (XFS_IS_GQUOTA_ON(sc->mp))1460		xrep_force_quotacheck(sc, XFS_DQTYPE_GROUP);1461	if (XFS_IS_PQUOTA_ON(sc->mp))1462		xrep_force_quotacheck(sc, XFS_DQTYPE_PROJ);1463 1464	return 0;1465}1466 1467/*1468 * Fix problems that the verifiers don't care about.  In general these are1469 * errors that don't cause problems elsewhere in the kernel that we can easily1470 * detect, so we don't check them all that rigorously.1471 */1472 1473/* Make sure block and extent counts are ok. */1474STATIC int1475xrep_inode_blockcounts(1476	struct xfs_scrub	*sc)1477{1478	struct xfs_ifork	*ifp;1479	xfs_filblks_t		count;1480	xfs_filblks_t		acount;1481	xfs_extnum_t		nextents;1482	int			error;1483 1484	trace_xrep_inode_blockcounts(sc);1485 1486	/* Set data fork counters from the data fork mappings. */1487	error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_DATA_FORK,1488			&nextents, &count);1489	if (error)1490		return error;1491	if (xfs_is_reflink_inode(sc->ip)) {1492		/*1493		 * data fork blockcount can exceed physical storage if a user1494		 * reflinks the same block over and over again.1495		 */1496		;1497	} else if (XFS_IS_REALTIME_INODE(sc->ip)) {1498		if (count >= sc->mp->m_sb.sb_rblocks)1499			return -EFSCORRUPTED;1500	} else {1501		if (count >= sc->mp->m_sb.sb_dblocks)1502			return -EFSCORRUPTED;1503	}1504	error = xrep_ino_ensure_extent_count(sc, XFS_DATA_FORK, nextents);1505	if (error)1506		return error;1507	sc->ip->i_df.if_nextents = nextents;1508 1509	/* Set attr fork counters from the attr fork mappings. */1510	ifp = xfs_ifork_ptr(sc->ip, XFS_ATTR_FORK);1511	if (ifp) {1512		error = xfs_bmap_count_blocks(sc->tp, sc->ip, XFS_ATTR_FORK,1513				&nextents, &acount);1514		if (error)1515			return error;1516		if (count >= sc->mp->m_sb.sb_dblocks)1517			return -EFSCORRUPTED;1518		error = xrep_ino_ensure_extent_count(sc, XFS_ATTR_FORK,1519				nextents);1520		if (error)1521			return error;1522		ifp->if_nextents = nextents;1523	} else {1524		acount = 0;1525	}1526 1527	sc->ip->i_nblocks = count + acount;1528	return 0;1529}1530 1531/* Check for invalid uid/gid/prid. */1532STATIC void1533xrep_inode_ids(1534	struct xfs_scrub	*sc)1535{1536	bool			dirty = false;1537 1538	trace_xrep_inode_ids(sc);1539 1540	if (!uid_valid(VFS_I(sc->ip)->i_uid)) {1541		i_uid_write(VFS_I(sc->ip), 0);1542		dirty = true;1543		if (XFS_IS_UQUOTA_ON(sc->mp))1544			xrep_force_quotacheck(sc, XFS_DQTYPE_USER);1545	}1546 1547	if (!gid_valid(VFS_I(sc->ip)->i_gid)) {1548		i_gid_write(VFS_I(sc->ip), 0);1549		dirty = true;1550		if (XFS_IS_GQUOTA_ON(sc->mp))1551			xrep_force_quotacheck(sc, XFS_DQTYPE_GROUP);1552	}1553 1554	if (sc->ip->i_projid == -1U) {1555		sc->ip->i_projid = 0;1556		dirty = true;1557		if (XFS_IS_PQUOTA_ON(sc->mp))1558			xrep_force_quotacheck(sc, XFS_DQTYPE_PROJ);1559	}1560 1561	/* strip setuid/setgid if we touched any of the ids */1562	if (dirty)1563		VFS_I(sc->ip)->i_mode &= ~(S_ISUID | S_ISGID);1564}1565 1566static inline void1567xrep_clamp_timestamp(1568	struct xfs_inode	*ip,1569	struct timespec64	*ts)1570{1571	ts->tv_nsec = clamp_t(long, ts->tv_nsec, 0, NSEC_PER_SEC);1572	*ts = timestamp_truncate(*ts, VFS_I(ip));1573}1574 1575/* Nanosecond counters can't have more than 1 billion. */1576STATIC void1577xrep_inode_timestamps(1578	struct xfs_inode	*ip)1579{1580	struct timespec64	tstamp;1581	struct inode		*inode = VFS_I(ip);1582 1583	tstamp = inode_get_atime(inode);1584	xrep_clamp_timestamp(ip, &tstamp);1585	inode_set_atime_to_ts(inode, tstamp);1586 1587	tstamp = inode_get_mtime(inode);1588	xrep_clamp_timestamp(ip, &tstamp);1589	inode_set_mtime_to_ts(inode, tstamp);1590 1591	tstamp = inode_get_ctime(inode);1592	xrep_clamp_timestamp(ip, &tstamp);1593	inode_set_ctime_to_ts(inode, tstamp);1594 1595	xrep_clamp_timestamp(ip, &ip->i_crtime);1596}1597 1598/* Fix inode flags that don't make sense together. */1599STATIC void1600xrep_inode_flags(1601	struct xfs_scrub	*sc)1602{1603	uint16_t		mode;1604 1605	trace_xrep_inode_flags(sc);1606 1607	mode = VFS_I(sc->ip)->i_mode;1608 1609	/* Clear junk flags */1610	if (sc->ip->i_diflags & ~XFS_DIFLAG_ANY)1611		sc->ip->i_diflags &= ~XFS_DIFLAG_ANY;1612 1613	/* NEWRTBM only applies to realtime bitmaps */1614	if (sc->ip->i_ino == sc->mp->m_sb.sb_rbmino)1615		sc->ip->i_diflags |= XFS_DIFLAG_NEWRTBM;1616	else1617		sc->ip->i_diflags &= ~XFS_DIFLAG_NEWRTBM;1618 1619	/* These only make sense for directories. */1620	if (!S_ISDIR(mode))1621		sc->ip->i_diflags &= ~(XFS_DIFLAG_RTINHERIT |1622					  XFS_DIFLAG_EXTSZINHERIT |1623					  XFS_DIFLAG_PROJINHERIT |1624					  XFS_DIFLAG_NOSYMLINKS);1625 1626	/* These only make sense for files. */1627	if (!S_ISREG(mode))1628		sc->ip->i_diflags &= ~(XFS_DIFLAG_REALTIME |1629					  XFS_DIFLAG_EXTSIZE);1630 1631	/* These only make sense for non-rt files. */1632	if (sc->ip->i_diflags & XFS_DIFLAG_REALTIME)1633		sc->ip->i_diflags &= ~XFS_DIFLAG_FILESTREAM;1634 1635	/* Immutable and append only?  Drop the append. */1636	if ((sc->ip->i_diflags & XFS_DIFLAG_IMMUTABLE) &&1637	    (sc->ip->i_diflags & XFS_DIFLAG_APPEND))1638		sc->ip->i_diflags &= ~XFS_DIFLAG_APPEND;1639 1640	/* Clear junk flags. */1641	if (sc->ip->i_diflags2 & ~XFS_DIFLAG2_ANY)1642		sc->ip->i_diflags2 &= ~XFS_DIFLAG2_ANY;1643 1644	/* No reflink flag unless we support it and it's a file. */1645	if (!xfs_has_reflink(sc->mp) || !S_ISREG(mode))1646		sc->ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;1647 1648	/* DAX only applies to files and dirs. */1649	if (!(S_ISREG(mode) || S_ISDIR(mode)))1650		sc->ip->i_diflags2 &= ~XFS_DIFLAG2_DAX;1651 1652	/* No reflink files on the realtime device. */1653	if (sc->ip->i_diflags & XFS_DIFLAG_REALTIME)1654		sc->ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;1655}1656 1657/*1658 * Fix size problems with block/node format directories.  If we fail to find1659 * the extent list, just bail out and let the bmapbtd repair functions clean1660 * up that mess.1661 */1662STATIC void1663xrep_inode_blockdir_size(1664	struct xfs_scrub	*sc)1665{1666	struct xfs_iext_cursor	icur;1667	struct xfs_bmbt_irec	got;1668	struct xfs_ifork	*ifp;1669	xfs_fileoff_t		off;1670	int			error;1671 1672	trace_xrep_inode_blockdir_size(sc);1673 1674	error = xfs_iread_extents(sc->tp, sc->ip, XFS_DATA_FORK);1675	if (error)1676		return;1677 1678	/* Find the last block before 32G; this is the dir size. */1679	ifp = xfs_ifork_ptr(sc->ip, XFS_DATA_FORK);1680	off = XFS_B_TO_FSB(sc->mp, XFS_DIR2_SPACE_SIZE);1681	if (!xfs_iext_lookup_extent_before(sc->ip, ifp, &off, &icur, &got)) {1682		/* zero-extents directory? */1683		return;1684	}1685 1686	off = got.br_startoff + got.br_blockcount;1687	sc->ip->i_disk_size = min_t(loff_t, XFS_DIR2_SPACE_SIZE,1688			XFS_FSB_TO_B(sc->mp, off));1689}1690 1691/* Fix size problems with short format directories. */1692STATIC void1693xrep_inode_sfdir_size(1694	struct xfs_scrub	*sc)1695{1696	struct xfs_ifork	*ifp;1697 1698	trace_xrep_inode_sfdir_size(sc);1699 1700	ifp = xfs_ifork_ptr(sc->ip, XFS_DATA_FORK);1701	sc->ip->i_disk_size = ifp->if_bytes;1702}1703 1704/*1705 * Fix any irregularities in a directory inode's size now that we can iterate1706 * extent maps and access other regular inode data.1707 */1708STATIC void1709xrep_inode_dir_size(1710	struct xfs_scrub	*sc)1711{1712	trace_xrep_inode_dir_size(sc);1713 1714	switch (sc->ip->i_df.if_format) {1715	case XFS_DINODE_FMT_EXTENTS:1716	case XFS_DINODE_FMT_BTREE:1717		xrep_inode_blockdir_size(sc);1718		break;1719	case XFS_DINODE_FMT_LOCAL:1720		xrep_inode_sfdir_size(sc);1721		break;1722	}1723}1724 1725/* Fix extent size hint problems. */1726STATIC void1727xrep_inode_extsize(1728	struct xfs_scrub	*sc)1729{1730	/* Fix misaligned extent size hints on a directory. */1731	if ((sc->ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&1732	    (sc->ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) &&1733	    xfs_extlen_to_rtxmod(sc->mp, sc->ip->i_extsize) > 0) {1734		sc->ip->i_extsize = 0;1735		sc->ip->i_diflags &= ~XFS_DIFLAG_EXTSZINHERIT;1736	}1737}1738 1739/* Ensure this file has an attr fork if it needs to hold a parent pointer. */1740STATIC int1741xrep_inode_pptr(1742	struct xfs_scrub	*sc)1743{1744	struct xfs_mount	*mp = sc->mp;1745	struct xfs_inode	*ip = sc->ip;1746	struct inode		*inode = VFS_I(ip);1747 1748	if (!xfs_has_parent(mp))1749		return 0;1750 1751	/*1752	 * Unlinked inodes that cannot be added to the directory tree will not1753	 * have a parent pointer.1754	 */1755	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))1756		return 0;1757 1758	/* The root directory doesn't have a parent pointer. */1759	if (ip == mp->m_rootip)1760		return 0;1761 1762	/*1763	 * Metadata inodes are rooted in the superblock and do not have any1764	 * parents.1765	 */1766	if (xfs_is_metadata_inode(ip))1767		return 0;1768 1769	/* Inode already has an attr fork; no further work possible here. */1770	if (xfs_inode_has_attr_fork(ip))1771		return 0;1772 1773	return xfs_bmap_add_attrfork(sc->tp, ip,1774			sizeof(struct xfs_attr_sf_hdr), true);1775}1776 1777/* Fix any irregularities in an inode that the verifiers don't catch. */1778STATIC int1779xrep_inode_problems(1780	struct xfs_scrub	*sc)1781{1782	int			error;1783 1784	error = xrep_inode_blockcounts(sc);1785	if (error)1786		return error;1787	error = xrep_inode_pptr(sc);1788	if (error)1789		return error;1790	xrep_inode_timestamps(sc->ip);1791	xrep_inode_flags(sc);1792	xrep_inode_ids(sc);1793	/*1794	 * We can now do a better job fixing the size of a directory now that1795	 * we can scan the data fork extents than we could in xrep_dinode_size.1796	 */1797	if (S_ISDIR(VFS_I(sc->ip)->i_mode))1798		xrep_inode_dir_size(sc);1799	xrep_inode_extsize(sc);1800 1801	trace_xrep_inode_fixed(sc);1802	xfs_trans_log_inode(sc->tp, sc->ip, XFS_ILOG_CORE);1803	return xrep_roll_trans(sc);1804}1805 1806/*1807 * Make sure this inode's unlinked list pointers are consistent with its1808 * link count.1809 */1810STATIC int1811xrep_inode_unlinked(1812	struct xfs_scrub	*sc)1813{1814	unsigned int		nlink = VFS_I(sc->ip)->i_nlink;1815	int			error;1816 1817	/*1818	 * If this inode is linked from the directory tree and on the unlinked1819	 * list, remove it from the unlinked list.1820	 */1821	if (nlink > 0 && xfs_inode_on_unlinked_list(sc->ip)) {1822		struct xfs_perag	*pag;1823		int			error;1824 1825		pag = xfs_perag_get(sc->mp,1826				XFS_INO_TO_AGNO(sc->mp, sc->ip->i_ino));1827		error = xfs_iunlink_remove(sc->tp, pag, sc->ip);1828		xfs_perag_put(pag);1829		if (error)1830			return error;1831	}1832 1833	/*1834	 * If this inode is not linked from the directory tree yet not on the1835	 * unlinked list, put it on the unlinked list.1836	 */1837	if (nlink == 0 && !xfs_inode_on_unlinked_list(sc->ip)) {1838		error = xfs_iunlink(sc->tp, sc->ip);1839		if (error)1840			return error;1841	}1842 1843	return 0;1844}1845 1846/* Repair an inode's fields. */1847int1848xrep_inode(1849	struct xfs_scrub	*sc)1850{1851	int			error = 0;1852 1853	/*1854	 * No inode?  That means we failed the _iget verifiers.  Repair all1855	 * the things that the inode verifiers care about, then retry _iget.1856	 */1857	if (!sc->ip) {1858		struct xrep_inode	*ri = sc->buf;1859 1860		ASSERT(ri != NULL);1861 1862		error = xrep_dinode_problems(ri);1863		if (error == -EBUSY) {1864			/*1865			 * Directory scan to recover inode mode encountered a1866			 * busy inode, so we did not continue repairing things.1867			 */1868			return 0;1869		}1870		if (error)1871			return error;1872 1873		/* By this point we had better have a working incore inode. */1874		if (!sc->ip)1875			return -EFSCORRUPTED;1876	}1877 1878	xfs_trans_ijoin(sc->tp, sc->ip, 0);1879 1880	/* If we found corruption of any kind, try to fix it. */1881	if ((sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT) ||1882	    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_XCORRUPT)) {1883		error = xrep_inode_problems(sc);1884		if (error)1885			return error;1886	}1887 1888	/* See if we can clear the reflink flag. */1889	if (xfs_is_reflink_inode(sc->ip)) {1890		error = xfs_reflink_clear_inode_flag(sc->ip, &sc->tp);1891		if (error)1892			return error;1893	}1894 1895	/* Reconnect incore unlinked list */1896	error = xrep_inode_unlinked(sc);1897	if (error)1898		return error;1899 1900	return xrep_defer_finish(sc);1901}1902