brintos

brintos / linux-shallow public Read only

0
0
Text · 81.3 KiB · 19dcb56 Raw
3048 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.4 * All Rights Reserved.5 */6#include <linux/iversion.h>7 8#include "xfs.h"9#include "xfs_fs.h"10#include "xfs_shared.h"11#include "xfs_format.h"12#include "xfs_log_format.h"13#include "xfs_trans_resv.h"14#include "xfs_mount.h"15#include "xfs_defer.h"16#include "xfs_inode.h"17#include "xfs_dir2.h"18#include "xfs_attr.h"19#include "xfs_bit.h"20#include "xfs_trans_space.h"21#include "xfs_trans.h"22#include "xfs_buf_item.h"23#include "xfs_inode_item.h"24#include "xfs_iunlink_item.h"25#include "xfs_ialloc.h"26#include "xfs_bmap.h"27#include "xfs_bmap_util.h"28#include "xfs_errortag.h"29#include "xfs_error.h"30#include "xfs_quota.h"31#include "xfs_filestream.h"32#include "xfs_trace.h"33#include "xfs_icache.h"34#include "xfs_symlink.h"35#include "xfs_trans_priv.h"36#include "xfs_log.h"37#include "xfs_bmap_btree.h"38#include "xfs_reflink.h"39#include "xfs_ag.h"40#include "xfs_log_priv.h"41#include "xfs_health.h"42#include "xfs_pnfs.h"43#include "xfs_parent.h"44#include "xfs_xattr.h"45#include "xfs_inode_util.h"46 47struct kmem_cache *xfs_inode_cache;48 49/*50 * These two are wrapper routines around the xfs_ilock() routine used to51 * centralize some grungy code.  They are used in places that wish to lock the52 * inode solely for reading the extents.  The reason these places can't just53 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to54 * bringing in of the extents from disk for a file in b-tree format.  If the55 * inode is in b-tree format, then we need to lock the inode exclusively until56 * the extents are read in.  Locking it exclusively all the time would limit57 * our parallelism unnecessarily, though.  What we do instead is check to see58 * if the extents have been read in yet, and only lock the inode exclusively59 * if they have not.60 *61 * The functions return a value which should be given to the corresponding62 * xfs_iunlock() call.63 */64uint65xfs_ilock_data_map_shared(66	struct xfs_inode	*ip)67{68	uint			lock_mode = XFS_ILOCK_SHARED;69 70	if (xfs_need_iread_extents(&ip->i_df))71		lock_mode = XFS_ILOCK_EXCL;72	xfs_ilock(ip, lock_mode);73	return lock_mode;74}75 76uint77xfs_ilock_attr_map_shared(78	struct xfs_inode	*ip)79{80	uint			lock_mode = XFS_ILOCK_SHARED;81 82	if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))83		lock_mode = XFS_ILOCK_EXCL;84	xfs_ilock(ip, lock_mode);85	return lock_mode;86}87 88/*89 * You can't set both SHARED and EXCL for the same lock,90 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,91 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values92 * to set in lock_flags.93 */94static inline void95xfs_lock_flags_assert(96	uint		lock_flags)97{98	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=99		(XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));100	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=101		(XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));102	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=103		(XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));104	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);105	ASSERT(lock_flags != 0);106}107 108/*109 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2110 * multi-reader locks: invalidate_lock and the i_lock.  This routine allows111 * various combinations of the locks to be obtained.112 *113 * The 3 locks should always be ordered so that the IO lock is obtained first,114 * the mmap lock second and the ilock last in order to prevent deadlock.115 *116 * Basic locking order:117 *118 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock119 *120 * mmap_lock locking order:121 *122 * i_rwsem -> page lock -> mmap_lock123 * mmap_lock -> invalidate_lock -> page_lock124 *125 * The difference in mmap_lock locking order mean that we cannot hold the126 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths127 * can fault in pages during copy in/out (for buffered IO) or require the128 * mmap_lock in get_user_pages() to map the user pages into the kernel address129 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page130 * fault because page faults already hold the mmap_lock.131 *132 * Hence to serialise fully against both syscall and mmap based IO, we need to133 * take both the i_rwsem and the invalidate_lock. These locks should *only* be134 * both taken in places where we need to invalidate the page cache in a race135 * free manner (e.g. truncate, hole punch and other extent manipulation136 * functions).137 */138void139xfs_ilock(140	xfs_inode_t		*ip,141	uint			lock_flags)142{143	trace_xfs_ilock(ip, lock_flags, _RET_IP_);144 145	xfs_lock_flags_assert(lock_flags);146 147	if (lock_flags & XFS_IOLOCK_EXCL) {148		down_write_nested(&VFS_I(ip)->i_rwsem,149				  XFS_IOLOCK_DEP(lock_flags));150	} else if (lock_flags & XFS_IOLOCK_SHARED) {151		down_read_nested(&VFS_I(ip)->i_rwsem,152				 XFS_IOLOCK_DEP(lock_flags));153	}154 155	if (lock_flags & XFS_MMAPLOCK_EXCL) {156		down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,157				  XFS_MMAPLOCK_DEP(lock_flags));158	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {159		down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,160				 XFS_MMAPLOCK_DEP(lock_flags));161	}162 163	if (lock_flags & XFS_ILOCK_EXCL)164		down_write_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));165	else if (lock_flags & XFS_ILOCK_SHARED)166		down_read_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));167}168 169/*170 * This is just like xfs_ilock(), except that the caller171 * is guaranteed not to sleep.  It returns 1 if it gets172 * the requested locks and 0 otherwise.  If the IO lock is173 * obtained but the inode lock cannot be, then the IO lock174 * is dropped before returning.175 *176 * ip -- the inode being locked177 * lock_flags -- this parameter indicates the inode's locks to be178 *       to be locked.  See the comment for xfs_ilock() for a list179 *	 of valid values.180 */181int182xfs_ilock_nowait(183	xfs_inode_t		*ip,184	uint			lock_flags)185{186	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);187 188	xfs_lock_flags_assert(lock_flags);189 190	if (lock_flags & XFS_IOLOCK_EXCL) {191		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))192			goto out;193	} else if (lock_flags & XFS_IOLOCK_SHARED) {194		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))195			goto out;196	}197 198	if (lock_flags & XFS_MMAPLOCK_EXCL) {199		if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))200			goto out_undo_iolock;201	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {202		if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))203			goto out_undo_iolock;204	}205 206	if (lock_flags & XFS_ILOCK_EXCL) {207		if (!down_write_trylock(&ip->i_lock))208			goto out_undo_mmaplock;209	} else if (lock_flags & XFS_ILOCK_SHARED) {210		if (!down_read_trylock(&ip->i_lock))211			goto out_undo_mmaplock;212	}213	return 1;214 215out_undo_mmaplock:216	if (lock_flags & XFS_MMAPLOCK_EXCL)217		up_write(&VFS_I(ip)->i_mapping->invalidate_lock);218	else if (lock_flags & XFS_MMAPLOCK_SHARED)219		up_read(&VFS_I(ip)->i_mapping->invalidate_lock);220out_undo_iolock:221	if (lock_flags & XFS_IOLOCK_EXCL)222		up_write(&VFS_I(ip)->i_rwsem);223	else if (lock_flags & XFS_IOLOCK_SHARED)224		up_read(&VFS_I(ip)->i_rwsem);225out:226	return 0;227}228 229/*230 * xfs_iunlock() is used to drop the inode locks acquired with231 * xfs_ilock() and xfs_ilock_nowait().  The caller must pass232 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so233 * that we know which locks to drop.234 *235 * ip -- the inode being unlocked236 * lock_flags -- this parameter indicates the inode's locks to be237 *       to be unlocked.  See the comment for xfs_ilock() for a list238 *	 of valid values for this parameter.239 *240 */241void242xfs_iunlock(243	xfs_inode_t		*ip,244	uint			lock_flags)245{246	xfs_lock_flags_assert(lock_flags);247 248	if (lock_flags & XFS_IOLOCK_EXCL)249		up_write(&VFS_I(ip)->i_rwsem);250	else if (lock_flags & XFS_IOLOCK_SHARED)251		up_read(&VFS_I(ip)->i_rwsem);252 253	if (lock_flags & XFS_MMAPLOCK_EXCL)254		up_write(&VFS_I(ip)->i_mapping->invalidate_lock);255	else if (lock_flags & XFS_MMAPLOCK_SHARED)256		up_read(&VFS_I(ip)->i_mapping->invalidate_lock);257 258	if (lock_flags & XFS_ILOCK_EXCL)259		up_write(&ip->i_lock);260	else if (lock_flags & XFS_ILOCK_SHARED)261		up_read(&ip->i_lock);262 263	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);264}265 266/*267 * give up write locks.  the i/o lock cannot be held nested268 * if it is being demoted.269 */270void271xfs_ilock_demote(272	xfs_inode_t		*ip,273	uint			lock_flags)274{275	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));276	ASSERT((lock_flags &277		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);278 279	if (lock_flags & XFS_ILOCK_EXCL)280		downgrade_write(&ip->i_lock);281	if (lock_flags & XFS_MMAPLOCK_EXCL)282		downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);283	if (lock_flags & XFS_IOLOCK_EXCL)284		downgrade_write(&VFS_I(ip)->i_rwsem);285 286	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);287}288 289void290xfs_assert_ilocked(291	struct xfs_inode	*ip,292	uint			lock_flags)293{294	/*295	 * Sometimes we assert the ILOCK is held exclusively, but we're in296	 * a workqueue, so lockdep doesn't know we're the owner.297	 */298	if (lock_flags & XFS_ILOCK_SHARED)299		rwsem_assert_held(&ip->i_lock);300	else if (lock_flags & XFS_ILOCK_EXCL)301		rwsem_assert_held_write_nolockdep(&ip->i_lock);302 303	if (lock_flags & XFS_MMAPLOCK_SHARED)304		rwsem_assert_held(&VFS_I(ip)->i_mapping->invalidate_lock);305	else if (lock_flags & XFS_MMAPLOCK_EXCL)306		rwsem_assert_held_write(&VFS_I(ip)->i_mapping->invalidate_lock);307 308	if (lock_flags & XFS_IOLOCK_SHARED)309		rwsem_assert_held(&VFS_I(ip)->i_rwsem);310	else if (lock_flags & XFS_IOLOCK_EXCL)311		rwsem_assert_held_write(&VFS_I(ip)->i_rwsem);312}313 314/*315 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when316 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined317 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build318 * errors and warnings.319 */320#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)321static bool322xfs_lockdep_subclass_ok(323	int subclass)324{325	return subclass < MAX_LOCKDEP_SUBCLASSES;326}327#else328#define xfs_lockdep_subclass_ok(subclass)	(true)329#endif330 331/*332 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different333 * value. This can be called for any type of inode lock combination, including334 * parent locking. Care must be taken to ensure we don't overrun the subclass335 * storage fields in the class mask we build.336 */337static inline uint338xfs_lock_inumorder(339	uint	lock_mode,340	uint	subclass)341{342	uint	class = 0;343 344	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |345			      XFS_ILOCK_RTSUM)));346	ASSERT(xfs_lockdep_subclass_ok(subclass));347 348	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {349		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);350		class += subclass << XFS_IOLOCK_SHIFT;351	}352 353	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {354		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);355		class += subclass << XFS_MMAPLOCK_SHIFT;356	}357 358	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {359		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);360		class += subclass << XFS_ILOCK_SHIFT;361	}362 363	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;364}365 366/*367 * The following routine will lock n inodes in exclusive mode.  We assume the368 * caller calls us with the inodes in i_ino order.369 *370 * We need to detect deadlock where an inode that we lock is in the AIL and we371 * start waiting for another inode that is locked by a thread in a long running372 * transaction (such as truncate). This can result in deadlock since the long373 * running trans might need to wait for the inode we just locked in order to374 * push the tail and free space in the log.375 *376 * xfs_lock_inodes() can only be used to lock one type of lock at a time -377 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we378 * lock more than one at a time, lockdep will report false positives saying we379 * have violated locking orders.380 */381void382xfs_lock_inodes(383	struct xfs_inode	**ips,384	int			inodes,385	uint			lock_mode)386{387	int			attempts = 0;388	uint			i;389	int			j;390	bool			try_lock;391	struct xfs_log_item	*lp;392 393	/*394	 * Currently supports between 2 and 5 inodes with exclusive locking.  We395	 * support an arbitrary depth of locking here, but absolute limits on396	 * inodes depend on the type of locking and the limits placed by397	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by398	 * the asserts.399	 */400	ASSERT(ips && inodes >= 2 && inodes <= 5);401	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |402			    XFS_ILOCK_EXCL));403	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |404			      XFS_ILOCK_SHARED)));405	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||406		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);407	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||408		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);409 410	if (lock_mode & XFS_IOLOCK_EXCL) {411		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));412	} else if (lock_mode & XFS_MMAPLOCK_EXCL)413		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));414 415again:416	try_lock = false;417	i = 0;418	for (; i < inodes; i++) {419		ASSERT(ips[i]);420 421		if (i && (ips[i] == ips[i - 1]))	/* Already locked */422			continue;423 424		/*425		 * If try_lock is not set yet, make sure all locked inodes are426		 * not in the AIL.  If any are, set try_lock to be used later.427		 */428		if (!try_lock) {429			for (j = (i - 1); j >= 0 && !try_lock; j--) {430				lp = &ips[j]->i_itemp->ili_item;431				if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))432					try_lock = true;433			}434		}435 436		/*437		 * If any of the previous locks we have locked is in the AIL,438		 * we must TRY to get the second and subsequent locks. If439		 * we can't get any, we must release all we have440		 * and try again.441		 */442		if (!try_lock) {443			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));444			continue;445		}446 447		/* try_lock means we have an inode locked that is in the AIL. */448		ASSERT(i != 0);449		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))450			continue;451 452		/*453		 * Unlock all previous guys and try again.  xfs_iunlock will try454		 * to push the tail if the inode is in the AIL.455		 */456		attempts++;457		for (j = i - 1; j >= 0; j--) {458			/*459			 * Check to see if we've already unlocked this one.  Not460			 * the first one going back, and the inode ptr is the461			 * same.462			 */463			if (j != (i - 1) && ips[j] == ips[j + 1])464				continue;465 466			xfs_iunlock(ips[j], lock_mode);467		}468 469		if ((attempts % 5) == 0) {470			delay(1); /* Don't just spin the CPU */471		}472		goto again;473	}474}475 476/*477 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and478 * mmaplock must be double-locked separately since we use i_rwsem and479 * invalidate_lock for that. We now support taking one lock EXCL and the480 * other SHARED.481 */482void483xfs_lock_two_inodes(484	struct xfs_inode	*ip0,485	uint			ip0_mode,486	struct xfs_inode	*ip1,487	uint			ip1_mode)488{489	int			attempts = 0;490	struct xfs_log_item	*lp;491 492	ASSERT(hweight32(ip0_mode) == 1);493	ASSERT(hweight32(ip1_mode) == 1);494	ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));495	ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));496	ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));497	ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));498	ASSERT(ip0->i_ino != ip1->i_ino);499 500	if (ip0->i_ino > ip1->i_ino) {501		swap(ip0, ip1);502		swap(ip0_mode, ip1_mode);503	}504 505 again:506	xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));507 508	/*509	 * If the first lock we have locked is in the AIL, we must TRY to get510	 * the second lock. If we can't get it, we must release the first one511	 * and try again.512	 */513	lp = &ip0->i_itemp->ili_item;514	if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {515		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {516			xfs_iunlock(ip0, ip0_mode);517			if ((++attempts % 5) == 0)518				delay(1); /* Don't just spin the CPU */519			goto again;520		}521	} else {522		xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));523	}524}525 526/*527 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match528 * is allowed, otherwise it has to be an exact match. If a CI match is found,529 * ci_name->name will point to a the actual name (caller must free) or530 * will be set to NULL if an exact match is found.531 */532int533xfs_lookup(534	struct xfs_inode	*dp,535	const struct xfs_name	*name,536	struct xfs_inode	**ipp,537	struct xfs_name		*ci_name)538{539	xfs_ino_t		inum;540	int			error;541 542	trace_xfs_lookup(dp, name);543 544	if (xfs_is_shutdown(dp->i_mount))545		return -EIO;546	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))547		return -EIO;548 549	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);550	if (error)551		goto out_unlock;552 553	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);554	if (error)555		goto out_free_name;556 557	return 0;558 559out_free_name:560	if (ci_name)561		kfree(ci_name->name);562out_unlock:563	*ipp = NULL;564	return error;565}566 567/*568 * Initialise a newly allocated inode and return the in-core inode to the569 * caller locked exclusively.570 *571 * Caller is responsible for unlocking the inode manually upon return572 */573int574xfs_icreate(575	struct xfs_trans	*tp,576	xfs_ino_t		ino,577	const struct xfs_icreate_args *args,578	struct xfs_inode	**ipp)579{580	struct xfs_mount	*mp = tp->t_mountp;581	struct xfs_inode	*ip = NULL;582	int			error;583 584	/*585	 * Get the in-core inode with the lock held exclusively to prevent586	 * others from looking at until we're done.587	 */588	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);589	if (error)590		return error;591 592	ASSERT(ip != NULL);593	xfs_trans_ijoin(tp, ip, 0);594	xfs_inode_init(tp, args, ip);595 596	/* now that we have an i_mode we can setup the inode structure */597	xfs_setup_inode(ip);598 599	*ipp = ip;600	return 0;601}602 603/* Return dquots for the ids that will be assigned to a new file. */604int605xfs_icreate_dqalloc(606	const struct xfs_icreate_args	*args,607	struct xfs_dquot		**udqpp,608	struct xfs_dquot		**gdqpp,609	struct xfs_dquot		**pdqpp)610{611	struct inode			*dir = VFS_I(args->pip);612	kuid_t				uid = GLOBAL_ROOT_UID;613	kgid_t				gid = GLOBAL_ROOT_GID;614	prid_t				prid = 0;615	unsigned int			flags = XFS_QMOPT_QUOTALL;616 617	if (args->idmap) {618		/*619		 * The uid/gid computation code must match what the VFS uses to620		 * assign i_[ug]id.  INHERIT adjusts the gid computation for621		 * setgid/grpid systems.622		 */623		uid = mapped_fsuid(args->idmap, i_user_ns(dir));624		gid = mapped_fsgid(args->idmap, i_user_ns(dir));625		prid = xfs_get_initial_prid(args->pip);626		flags |= XFS_QMOPT_INHERIT;627	}628 629	*udqpp = *gdqpp = *pdqpp = NULL;630 631	return xfs_qm_vop_dqalloc(args->pip, uid, gid, prid, flags, udqpp,632			gdqpp, pdqpp);633}634 635int636xfs_create(637	const struct xfs_icreate_args *args,638	struct xfs_name		*name,639	struct xfs_inode	**ipp)640{641	struct xfs_inode	*dp = args->pip;642	struct xfs_dir_update	du = {643		.dp		= dp,644		.name		= name,645	};646	struct xfs_mount	*mp = dp->i_mount;647	struct xfs_trans	*tp = NULL;648	struct xfs_dquot	*udqp;649	struct xfs_dquot	*gdqp;650	struct xfs_dquot	*pdqp;651	struct xfs_trans_res	*tres;652	xfs_ino_t		ino;653	bool			unlock_dp_on_error = false;654	bool			is_dir = S_ISDIR(args->mode);655	uint			resblks;656	int			error;657 658	trace_xfs_create(dp, name);659 660	if (xfs_is_shutdown(mp))661		return -EIO;662	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))663		return -EIO;664 665	/* Make sure that we have allocated dquot(s) on disk. */666	error = xfs_icreate_dqalloc(args, &udqp, &gdqp, &pdqp);667	if (error)668		return error;669 670	if (is_dir) {671		resblks = xfs_mkdir_space_res(mp, name->len);672		tres = &M_RES(mp)->tr_mkdir;673	} else {674		resblks = xfs_create_space_res(mp, name->len);675		tres = &M_RES(mp)->tr_create;676	}677 678	error = xfs_parent_start(mp, &du.ppargs);679	if (error)680		goto out_release_dquots;681 682	/*683	 * Initially assume that the file does not exist and684	 * reserve the resources for that case.  If that is not685	 * the case we'll drop the one we have and get a more686	 * appropriate transaction later.687	 */688	error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,689			&tp);690	if (error == -ENOSPC) {691		/* flush outstanding delalloc blocks and retry */692		xfs_flush_inodes(mp);693		error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,694				resblks, &tp);695	}696	if (error)697		goto out_parent;698 699	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);700	unlock_dp_on_error = true;701 702	/*703	 * A newly created regular or special file just has one directory704	 * entry pointing to them, but a directory also the "." entry705	 * pointing to itself.706	 */707	error = xfs_dialloc(&tp, args, &ino);708	if (!error)709		error = xfs_icreate(tp, ino, args, &du.ip);710	if (error)711		goto out_trans_cancel;712 713	/*714	 * Now we join the directory inode to the transaction.  We do not do it715	 * earlier because xfs_dialloc might commit the previous transaction716	 * (and release all the locks).  An error from here on will result in717	 * the transaction cancel unlocking dp so don't do it explicitly in the718	 * error path.719	 */720	xfs_trans_ijoin(tp, dp, 0);721 722	error = xfs_dir_create_child(tp, resblks, &du);723	if (error)724		goto out_trans_cancel;725 726	/*727	 * If this is a synchronous mount, make sure that the728	 * create transaction goes to disk before returning to729	 * the user.730	 */731	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))732		xfs_trans_set_sync(tp);733 734	/*735	 * Attach the dquot(s) to the inodes and modify them incore.736	 * These ids of the inode couldn't have changed since the new737	 * inode has been locked ever since it was created.738	 */739	xfs_qm_vop_create_dqattach(tp, du.ip, udqp, gdqp, pdqp);740 741	error = xfs_trans_commit(tp);742	if (error)743		goto out_release_inode;744 745	xfs_qm_dqrele(udqp);746	xfs_qm_dqrele(gdqp);747	xfs_qm_dqrele(pdqp);748 749	*ipp = du.ip;750	xfs_iunlock(du.ip, XFS_ILOCK_EXCL);751	xfs_iunlock(dp, XFS_ILOCK_EXCL);752	xfs_parent_finish(mp, du.ppargs);753	return 0;754 755 out_trans_cancel:756	xfs_trans_cancel(tp);757 out_release_inode:758	/*759	 * Wait until after the current transaction is aborted to finish the760	 * setup of the inode and release the inode.  This prevents recursive761	 * transactions and deadlocks from xfs_inactive.762	 */763	if (du.ip) {764		xfs_iunlock(du.ip, XFS_ILOCK_EXCL);765		xfs_finish_inode_setup(du.ip);766		xfs_irele(du.ip);767	}768 out_parent:769	xfs_parent_finish(mp, du.ppargs);770 out_release_dquots:771	xfs_qm_dqrele(udqp);772	xfs_qm_dqrele(gdqp);773	xfs_qm_dqrele(pdqp);774 775	if (unlock_dp_on_error)776		xfs_iunlock(dp, XFS_ILOCK_EXCL);777	return error;778}779 780int781xfs_create_tmpfile(782	const struct xfs_icreate_args *args,783	struct xfs_inode	**ipp)784{785	struct xfs_inode	*dp = args->pip;786	struct xfs_mount	*mp = dp->i_mount;787	struct xfs_inode	*ip = NULL;788	struct xfs_trans	*tp = NULL;789	struct xfs_dquot	*udqp;790	struct xfs_dquot	*gdqp;791	struct xfs_dquot	*pdqp;792	struct xfs_trans_res	*tres;793	xfs_ino_t		ino;794	uint			resblks;795	int			error;796 797	ASSERT(args->flags & XFS_ICREATE_TMPFILE);798 799	if (xfs_is_shutdown(mp))800		return -EIO;801 802	/* Make sure that we have allocated dquot(s) on disk. */803	error = xfs_icreate_dqalloc(args, &udqp, &gdqp, &pdqp);804	if (error)805		return error;806 807	resblks = XFS_IALLOC_SPACE_RES(mp);808	tres = &M_RES(mp)->tr_create_tmpfile;809 810	error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,811			&tp);812	if (error)813		goto out_release_dquots;814 815	error = xfs_dialloc(&tp, args, &ino);816	if (!error)817		error = xfs_icreate(tp, ino, args, &ip);818	if (error)819		goto out_trans_cancel;820 821	if (xfs_has_wsync(mp))822		xfs_trans_set_sync(tp);823 824	/*825	 * Attach the dquot(s) to the inodes and modify them incore.826	 * These ids of the inode couldn't have changed since the new827	 * inode has been locked ever since it was created.828	 */829	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);830 831	error = xfs_iunlink(tp, ip);832	if (error)833		goto out_trans_cancel;834 835	error = xfs_trans_commit(tp);836	if (error)837		goto out_release_inode;838 839	xfs_qm_dqrele(udqp);840	xfs_qm_dqrele(gdqp);841	xfs_qm_dqrele(pdqp);842 843	*ipp = ip;844	xfs_iunlock(ip, XFS_ILOCK_EXCL);845	return 0;846 847 out_trans_cancel:848	xfs_trans_cancel(tp);849 out_release_inode:850	/*851	 * Wait until after the current transaction is aborted to finish the852	 * setup of the inode and release the inode.  This prevents recursive853	 * transactions and deadlocks from xfs_inactive.854	 */855	if (ip) {856		xfs_iunlock(ip, XFS_ILOCK_EXCL);857		xfs_finish_inode_setup(ip);858		xfs_irele(ip);859	}860 out_release_dquots:861	xfs_qm_dqrele(udqp);862	xfs_qm_dqrele(gdqp);863	xfs_qm_dqrele(pdqp);864 865	return error;866}867 868int869xfs_link(870	struct xfs_inode	*tdp,871	struct xfs_inode	*sip,872	struct xfs_name		*target_name)873{874	struct xfs_dir_update	du = {875		.dp		= tdp,876		.name		= target_name,877		.ip		= sip,878	};879	struct xfs_mount	*mp = tdp->i_mount;880	struct xfs_trans	*tp;881	int			error, nospace_error = 0;882	int			resblks;883 884	trace_xfs_link(tdp, target_name);885 886	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));887 888	if (xfs_is_shutdown(mp))889		return -EIO;890	if (xfs_ifork_zapped(tdp, XFS_DATA_FORK))891		return -EIO;892 893	error = xfs_qm_dqattach(sip);894	if (error)895		goto std_return;896 897	error = xfs_qm_dqattach(tdp);898	if (error)899		goto std_return;900 901	error = xfs_parent_start(mp, &du.ppargs);902	if (error)903		goto std_return;904 905	resblks = xfs_link_space_res(mp, target_name->len);906	error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,907			&tp, &nospace_error);908	if (error)909		goto out_parent;910 911	/*912	 * We don't allow reservationless or quotaless hardlinking when parent913	 * pointers are enabled because we can't back out if the xattrs must914	 * grow.915	 */916	if (du.ppargs && nospace_error) {917		error = nospace_error;918		goto error_return;919	}920 921	/*922	 * If we are using project inheritance, we only allow hard link923	 * creation in our tree when the project IDs are the same; else924	 * the tree quota mechanism could be circumvented.925	 */926	if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&927		     tdp->i_projid != sip->i_projid)) {928		/*929		 * Project quota setup skips special files which can930		 * leave inodes in a PROJINHERIT directory without a931		 * project ID set. We need to allow links to be made932		 * to these "project-less" inodes because userspace933		 * expects them to succeed after project ID setup,934		 * but everything else should be rejected.935		 */936		if (!special_file(VFS_I(sip)->i_mode) ||937		    sip->i_projid != 0) {938			error = -EXDEV;939			goto error_return;940		}941	}942 943	error = xfs_dir_add_child(tp, resblks, &du);944	if (error)945		goto error_return;946 947	/*948	 * If this is a synchronous mount, make sure that the949	 * link transaction goes to disk before returning to950	 * the user.951	 */952	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))953		xfs_trans_set_sync(tp);954 955	error = xfs_trans_commit(tp);956	xfs_iunlock(tdp, XFS_ILOCK_EXCL);957	xfs_iunlock(sip, XFS_ILOCK_EXCL);958	xfs_parent_finish(mp, du.ppargs);959	return error;960 961 error_return:962	xfs_trans_cancel(tp);963	xfs_iunlock(tdp, XFS_ILOCK_EXCL);964	xfs_iunlock(sip, XFS_ILOCK_EXCL);965 out_parent:966	xfs_parent_finish(mp, du.ppargs);967 std_return:968	if (error == -ENOSPC && nospace_error)969		error = nospace_error;970	return error;971}972 973/* Clear the reflink flag and the cowblocks tag if possible. */974static void975xfs_itruncate_clear_reflink_flags(976	struct xfs_inode	*ip)977{978	struct xfs_ifork	*dfork;979	struct xfs_ifork	*cfork;980 981	if (!xfs_is_reflink_inode(ip))982		return;983	dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);984	cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);985	if (dfork->if_bytes == 0 && cfork->if_bytes == 0)986		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;987	if (cfork->if_bytes == 0)988		xfs_inode_clear_cowblocks_tag(ip);989}990 991/*992 * Free up the underlying blocks past new_size.  The new size must be smaller993 * than the current size.  This routine can be used both for the attribute and994 * data fork, and does not modify the inode size, which is left to the caller.995 *996 * The transaction passed to this routine must have made a permanent log997 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the998 * given transaction and start new ones, so make sure everything involved in999 * the transaction is tidy before calling here.  Some transaction will be1000 * returned to the caller to be committed.  The incoming transaction must1001 * already include the inode, and both inode locks must be held exclusively.1002 * The inode must also be "held" within the transaction.  On return the inode1003 * will be "held" within the returned transaction.  This routine does NOT1004 * require any disk space to be reserved for it within the transaction.1005 *1006 * If we get an error, we must return with the inode locked and linked into the1007 * current transaction. This keeps things simple for the higher level code,1008 * because it always knows that the inode is locked and held in the transaction1009 * that returns to it whether errors occur or not.  We don't mark the inode1010 * dirty on error so that transactions can be easily aborted if possible.1011 */1012int1013xfs_itruncate_extents_flags(1014	struct xfs_trans	**tpp,1015	struct xfs_inode	*ip,1016	int			whichfork,1017	xfs_fsize_t		new_size,1018	int			flags)1019{1020	struct xfs_mount	*mp = ip->i_mount;1021	struct xfs_trans	*tp = *tpp;1022	xfs_fileoff_t		first_unmap_block;1023	int			error = 0;1024 1025	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);1026	if (atomic_read(&VFS_I(ip)->i_count))1027		xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);1028	ASSERT(new_size <= XFS_ISIZE(ip));1029	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);1030	ASSERT(ip->i_itemp != NULL);1031	ASSERT(ip->i_itemp->ili_lock_flags == 0);1032	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));1033 1034	trace_xfs_itruncate_extents_start(ip, new_size);1035 1036	flags |= xfs_bmapi_aflag(whichfork);1037 1038	/*1039	 * Since it is possible for space to become allocated beyond1040	 * the end of the file (in a crash where the space is allocated1041	 * but the inode size is not yet updated), simply remove any1042	 * blocks which show up between the new EOF and the maximum1043	 * possible file size.1044	 *1045	 * We have to free all the blocks to the bmbt maximum offset, even if1046	 * the page cache can't scale that far.1047	 */1048	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);1049	if (!xfs_verify_fileoff(mp, first_unmap_block)) {1050		WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);1051		return 0;1052	}1053 1054	error = xfs_bunmapi_range(&tp, ip, flags, first_unmap_block,1055			XFS_MAX_FILEOFF);1056	if (error)1057		goto out;1058 1059	if (whichfork == XFS_DATA_FORK) {1060		/* Remove all pending CoW reservations. */1061		error = xfs_reflink_cancel_cow_blocks(ip, &tp,1062				first_unmap_block, XFS_MAX_FILEOFF, true);1063		if (error)1064			goto out;1065 1066		xfs_itruncate_clear_reflink_flags(ip);1067	}1068 1069	/*1070	 * Always re-log the inode so that our permanent transaction can keep1071	 * on rolling it forward in the log.1072	 */1073	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);1074 1075	trace_xfs_itruncate_extents_end(ip, new_size);1076 1077out:1078	*tpp = tp;1079	return error;1080}1081 1082/*1083 * Mark all the buffers attached to this directory stale.  In theory we should1084 * never be freeing a directory with any blocks at all, but this covers the1085 * case where we've recovered a directory swap with a "temporary" directory1086 * created by online repair and now need to dump it.1087 */1088STATIC void1089xfs_inactive_dir(1090	struct xfs_inode	*dp)1091{1092	struct xfs_iext_cursor	icur;1093	struct xfs_bmbt_irec	got;1094	struct xfs_mount	*mp = dp->i_mount;1095	struct xfs_da_geometry	*geo = mp->m_dir_geo;1096	struct xfs_ifork	*ifp = xfs_ifork_ptr(dp, XFS_DATA_FORK);1097	xfs_fileoff_t		off;1098 1099	/*1100	 * Invalidate each directory block.  All directory blocks are of1101	 * fsbcount length and alignment, so we only need to walk those same1102	 * offsets.  We hold the only reference to this inode, so we must wait1103	 * for the buffer locks.1104	 */1105	for_each_xfs_iext(ifp, &icur, &got) {1106		for (off = round_up(got.br_startoff, geo->fsbcount);1107		     off < got.br_startoff + got.br_blockcount;1108		     off += geo->fsbcount) {1109			struct xfs_buf	*bp = NULL;1110			xfs_fsblock_t	fsbno;1111			int		error;1112 1113			fsbno = (off - got.br_startoff) + got.br_startblock;1114			error = xfs_buf_incore(mp->m_ddev_targp,1115					XFS_FSB_TO_DADDR(mp, fsbno),1116					XFS_FSB_TO_BB(mp, geo->fsbcount),1117					XBF_LIVESCAN, &bp);1118			if (error)1119				continue;1120 1121			xfs_buf_stale(bp);1122			xfs_buf_relse(bp);1123		}1124	}1125}1126 1127/*1128 * xfs_inactive_truncate1129 *1130 * Called to perform a truncate when an inode becomes unlinked.1131 */1132STATIC int1133xfs_inactive_truncate(1134	struct xfs_inode *ip)1135{1136	struct xfs_mount	*mp = ip->i_mount;1137	struct xfs_trans	*tp;1138	int			error;1139 1140	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);1141	if (error) {1142		ASSERT(xfs_is_shutdown(mp));1143		return error;1144	}1145	xfs_ilock(ip, XFS_ILOCK_EXCL);1146	xfs_trans_ijoin(tp, ip, 0);1147 1148	/*1149	 * Log the inode size first to prevent stale data exposure in the event1150	 * of a system crash before the truncate completes. See the related1151	 * comment in xfs_vn_setattr_size() for details.1152	 */1153	ip->i_disk_size = 0;1154	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);1155 1156	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);1157	if (error)1158		goto error_trans_cancel;1159 1160	ASSERT(ip->i_df.if_nextents == 0);1161 1162	error = xfs_trans_commit(tp);1163	if (error)1164		goto error_unlock;1165 1166	xfs_iunlock(ip, XFS_ILOCK_EXCL);1167	return 0;1168 1169error_trans_cancel:1170	xfs_trans_cancel(tp);1171error_unlock:1172	xfs_iunlock(ip, XFS_ILOCK_EXCL);1173	return error;1174}1175 1176/*1177 * xfs_inactive_ifree()1178 *1179 * Perform the inode free when an inode is unlinked.1180 */1181STATIC int1182xfs_inactive_ifree(1183	struct xfs_inode *ip)1184{1185	struct xfs_mount	*mp = ip->i_mount;1186	struct xfs_trans	*tp;1187	int			error;1188 1189	/*1190	 * We try to use a per-AG reservation for any block needed by the finobt1191	 * tree, but as the finobt feature predates the per-AG reservation1192	 * support a degraded file system might not have enough space for the1193	 * reservation at mount time.  In that case try to dip into the reserved1194	 * pool and pray.1195	 *1196	 * Send a warning if the reservation does happen to fail, as the inode1197	 * now remains allocated and sits on the unlinked list until the fs is1198	 * repaired.1199	 */1200	if (unlikely(mp->m_finobt_nores)) {1201		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,1202				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,1203				&tp);1204	} else {1205		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);1206	}1207	if (error) {1208		if (error == -ENOSPC) {1209			xfs_warn_ratelimited(mp,1210			"Failed to remove inode(s) from unlinked list. "1211			"Please free space, unmount and run xfs_repair.");1212		} else {1213			ASSERT(xfs_is_shutdown(mp));1214		}1215		return error;1216	}1217 1218	/*1219	 * We do not hold the inode locked across the entire rolling transaction1220	 * here. We only need to hold it for the first transaction that1221	 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the1222	 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode1223	 * here breaks the relationship between cluster buffer invalidation and1224	 * stale inode invalidation on cluster buffer item journal commit1225	 * completion, and can result in leaving dirty stale inodes hanging1226	 * around in memory.1227	 *1228	 * We have no need for serialising this inode operation against other1229	 * operations - we freed the inode and hence reallocation is required1230	 * and that will serialise on reallocating the space the deferops need1231	 * to free. Hence we can unlock the inode on the first commit of1232	 * the transaction rather than roll it right through the deferops. This1233	 * avoids relogging the XFS_ISTALE inode.1234	 *1235	 * We check that xfs_ifree() hasn't grown an internal transaction roll1236	 * by asserting that the inode is still locked when it returns.1237	 */1238	xfs_ilock(ip, XFS_ILOCK_EXCL);1239	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);1240 1241	error = xfs_ifree(tp, ip);1242	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);1243	if (error) {1244		/*1245		 * If we fail to free the inode, shut down.  The cancel1246		 * might do that, we need to make sure.  Otherwise the1247		 * inode might be lost for a long time or forever.1248		 */1249		if (!xfs_is_shutdown(mp)) {1250			xfs_notice(mp, "%s: xfs_ifree returned error %d",1251				__func__, error);1252			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);1253		}1254		xfs_trans_cancel(tp);1255		return error;1256	}1257 1258	/*1259	 * Credit the quota account(s). The inode is gone.1260	 */1261	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);1262 1263	return xfs_trans_commit(tp);1264}1265 1266/*1267 * Returns true if we need to update the on-disk metadata before we can free1268 * the memory used by this inode.  Updates include freeing post-eof1269 * preallocations; freeing COW staging extents; and marking the inode free in1270 * the inobt if it is on the unlinked list.1271 */1272bool1273xfs_inode_needs_inactive(1274	struct xfs_inode	*ip)1275{1276	struct xfs_mount	*mp = ip->i_mount;1277	struct xfs_ifork	*cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);1278 1279	/*1280	 * If the inode is already free, then there can be nothing1281	 * to clean up here.1282	 */1283	if (VFS_I(ip)->i_mode == 0)1284		return false;1285 1286	/*1287	 * If this is a read-only mount, don't do this (would generate I/O)1288	 * unless we're in log recovery and cleaning the iunlinked list.1289	 */1290	if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))1291		return false;1292 1293	/* If the log isn't running, push inodes straight to reclaim. */1294	if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))1295		return false;1296 1297	/* Metadata inodes require explicit resource cleanup. */1298	if (xfs_is_metadata_inode(ip))1299		return false;1300 1301	/* Want to clean out the cow blocks if there are any. */1302	if (cow_ifp && cow_ifp->if_bytes > 0)1303		return true;1304 1305	/* Unlinked files must be freed. */1306	if (VFS_I(ip)->i_nlink == 0)1307		return true;1308 1309	/*1310	 * This file isn't being freed, so check if there are post-eof blocks1311	 * to free.1312	 *1313	 * Note: don't bother with iolock here since lockdep complains about1314	 * acquiring it in reclaim context. We have the only reference to the1315	 * inode at this point anyways.1316	 */1317	return xfs_can_free_eofblocks(ip);1318}1319 1320/*1321 * Save health status somewhere, if we're dumping an inode with uncorrected1322 * errors and online repair isn't running.1323 */1324static inline void1325xfs_inactive_health(1326	struct xfs_inode	*ip)1327{1328	struct xfs_mount	*mp = ip->i_mount;1329	struct xfs_perag	*pag;1330	unsigned int		sick;1331	unsigned int		checked;1332 1333	xfs_inode_measure_sickness(ip, &sick, &checked);1334	if (!sick)1335		return;1336 1337	trace_xfs_inode_unfixed_corruption(ip, sick);1338 1339	if (sick & XFS_SICK_INO_FORGET)1340		return;1341 1342	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));1343	if (!pag) {1344		/* There had better still be a perag structure! */1345		ASSERT(0);1346		return;1347	}1348 1349	xfs_ag_mark_sick(pag, XFS_SICK_AG_INODES);1350	xfs_perag_put(pag);1351}1352 1353/*1354 * xfs_inactive1355 *1356 * This is called when the vnode reference count for the vnode1357 * goes to zero.  If the file has been unlinked, then it must1358 * now be truncated.  Also, we clear all of the read-ahead state1359 * kept for the inode here since the file is now closed.1360 */1361int1362xfs_inactive(1363	xfs_inode_t	*ip)1364{1365	struct xfs_mount	*mp;1366	int			error = 0;1367	int			truncate = 0;1368 1369	/*1370	 * If the inode is already free, then there can be nothing1371	 * to clean up here.1372	 */1373	if (VFS_I(ip)->i_mode == 0) {1374		ASSERT(ip->i_df.if_broot_bytes == 0);1375		goto out;1376	}1377 1378	mp = ip->i_mount;1379	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));1380 1381	xfs_inactive_health(ip);1382 1383	/*1384	 * If this is a read-only mount, don't do this (would generate I/O)1385	 * unless we're in log recovery and cleaning the iunlinked list.1386	 */1387	if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))1388		goto out;1389 1390	/* Metadata inodes require explicit resource cleanup. */1391	if (xfs_is_metadata_inode(ip))1392		goto out;1393 1394	/* Try to clean out the cow blocks if there are any. */1395	if (xfs_inode_has_cow_data(ip))1396		xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);1397 1398	if (VFS_I(ip)->i_nlink != 0) {1399		/*1400		 * Note: don't bother with iolock here since lockdep complains1401		 * about acquiring it in reclaim context. We have the only1402		 * reference to the inode at this point anyways.1403		 */1404		if (xfs_can_free_eofblocks(ip))1405			error = xfs_free_eofblocks(ip);1406 1407		goto out;1408	}1409 1410	if (S_ISREG(VFS_I(ip)->i_mode) &&1411	    (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||1412	     xfs_inode_has_filedata(ip)))1413		truncate = 1;1414 1415	if (xfs_iflags_test(ip, XFS_IQUOTAUNCHECKED)) {1416		/*1417		 * If this inode is being inactivated during a quotacheck and1418		 * has not yet been scanned by quotacheck, we /must/ remove1419		 * the dquots from the inode before inactivation changes the1420		 * block and inode counts.  Most probably this is a result of1421		 * reloading the incore iunlinked list to purge unrecovered1422		 * unlinked inodes.1423		 */1424		xfs_qm_dqdetach(ip);1425	} else {1426		error = xfs_qm_dqattach(ip);1427		if (error)1428			goto out;1429	}1430 1431	if (S_ISDIR(VFS_I(ip)->i_mode) && ip->i_df.if_nextents > 0) {1432		xfs_inactive_dir(ip);1433		truncate = 1;1434	}1435 1436	if (S_ISLNK(VFS_I(ip)->i_mode))1437		error = xfs_inactive_symlink(ip);1438	else if (truncate)1439		error = xfs_inactive_truncate(ip);1440	if (error)1441		goto out;1442 1443	/*1444	 * If there are attributes associated with the file then blow them away1445	 * now.  The code calls a routine that recursively deconstructs the1446	 * attribute fork. If also blows away the in-core attribute fork.1447	 */1448	if (xfs_inode_has_attr_fork(ip)) {1449		error = xfs_attr_inactive(ip);1450		if (error)1451			goto out;1452	}1453 1454	ASSERT(ip->i_forkoff == 0);1455 1456	/*1457	 * Free the inode.1458	 */1459	error = xfs_inactive_ifree(ip);1460 1461out:1462	/*1463	 * We're done making metadata updates for this inode, so we can release1464	 * the attached dquots.1465	 */1466	xfs_qm_dqdetach(ip);1467	return error;1468}1469 1470/*1471 * Find an inode on the unlinked list. This does not take references to the1472 * inode as we have existence guarantees by holding the AGI buffer lock and that1473 * only unlinked, referenced inodes can be on the unlinked inode list.  If we1474 * don't find the inode in cache, then let the caller handle the situation.1475 */1476struct xfs_inode *1477xfs_iunlink_lookup(1478	struct xfs_perag	*pag,1479	xfs_agino_t		agino)1480{1481	struct xfs_inode	*ip;1482 1483	rcu_read_lock();1484	ip = radix_tree_lookup(&pag->pag_ici_root, agino);1485	if (!ip) {1486		/* Caller can handle inode not being in memory. */1487		rcu_read_unlock();1488		return NULL;1489	}1490 1491	/*1492	 * Inode in RCU freeing limbo should not happen.  Warn about this and1493	 * let the caller handle the failure.1494	 */1495	if (WARN_ON_ONCE(!ip->i_ino)) {1496		rcu_read_unlock();1497		return NULL;1498	}1499	ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));1500	rcu_read_unlock();1501	return ip;1502}1503 1504/*1505 * Load the inode @next_agino into the cache and set its prev_unlinked pointer1506 * to @prev_agino.  Caller must hold the AGI to synchronize with other changes1507 * to the unlinked list.1508 */1509int1510xfs_iunlink_reload_next(1511	struct xfs_trans	*tp,1512	struct xfs_buf		*agibp,1513	xfs_agino_t		prev_agino,1514	xfs_agino_t		next_agino)1515{1516	struct xfs_perag	*pag = agibp->b_pag;1517	struct xfs_mount	*mp = pag->pag_mount;1518	struct xfs_inode	*next_ip = NULL;1519	xfs_ino_t		ino;1520	int			error;1521 1522	ASSERT(next_agino != NULLAGINO);1523 1524#ifdef DEBUG1525	rcu_read_lock();1526	next_ip = radix_tree_lookup(&pag->pag_ici_root, next_agino);1527	ASSERT(next_ip == NULL);1528	rcu_read_unlock();1529#endif1530 1531	xfs_info_ratelimited(mp,1532 "Found unrecovered unlinked inode 0x%x in AG 0x%x.  Initiating recovery.",1533			next_agino, pag->pag_agno);1534 1535	/*1536	 * Use an untrusted lookup just to be cautious in case the AGI has been1537	 * corrupted and now points at a free inode.  That shouldn't happen,1538	 * but we'd rather shut down now since we're already running in a weird1539	 * situation.1540	 */1541	ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, next_agino);1542	error = xfs_iget(mp, tp, ino, XFS_IGET_UNTRUSTED, 0, &next_ip);1543	if (error) {1544		xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);1545		return error;1546	}1547 1548	/* If this is not an unlinked inode, something is very wrong. */1549	if (VFS_I(next_ip)->i_nlink != 0) {1550		xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);1551		error = -EFSCORRUPTED;1552		goto rele;1553	}1554 1555	next_ip->i_prev_unlinked = prev_agino;1556	trace_xfs_iunlink_reload_next(next_ip);1557rele:1558	ASSERT(!(VFS_I(next_ip)->i_state & I_DONTCACHE));1559	if (xfs_is_quotacheck_running(mp) && next_ip)1560		xfs_iflags_set(next_ip, XFS_IQUOTAUNCHECKED);1561	xfs_irele(next_ip);1562	return error;1563}1564 1565/*1566 * Look up the inode number specified and if it is not already marked XFS_ISTALE1567 * mark it stale. We should only find clean inodes in this lookup that aren't1568 * already stale.1569 */1570static void1571xfs_ifree_mark_inode_stale(1572	struct xfs_perag	*pag,1573	struct xfs_inode	*free_ip,1574	xfs_ino_t		inum)1575{1576	struct xfs_mount	*mp = pag->pag_mount;1577	struct xfs_inode_log_item *iip;1578	struct xfs_inode	*ip;1579 1580retry:1581	rcu_read_lock();1582	ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));1583 1584	/* Inode not in memory, nothing to do */1585	if (!ip) {1586		rcu_read_unlock();1587		return;1588	}1589 1590	/*1591	 * because this is an RCU protected lookup, we could find a recently1592	 * freed or even reallocated inode during the lookup. We need to check1593	 * under the i_flags_lock for a valid inode here. Skip it if it is not1594	 * valid, the wrong inode or stale.1595	 */1596	spin_lock(&ip->i_flags_lock);1597	if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))1598		goto out_iflags_unlock;1599 1600	/*1601	 * Don't try to lock/unlock the current inode, but we _cannot_ skip the1602	 * other inodes that we did not find in the list attached to the buffer1603	 * and are not already marked stale. If we can't lock it, back off and1604	 * retry.1605	 */1606	if (ip != free_ip) {1607		if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {1608			spin_unlock(&ip->i_flags_lock);1609			rcu_read_unlock();1610			delay(1);1611			goto retry;1612		}1613	}1614	ip->i_flags |= XFS_ISTALE;1615 1616	/*1617	 * If the inode is flushing, it is already attached to the buffer.  All1618	 * we needed to do here is mark the inode stale so buffer IO completion1619	 * will remove it from the AIL.1620	 */1621	iip = ip->i_itemp;1622	if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {1623		ASSERT(!list_empty(&iip->ili_item.li_bio_list));1624		ASSERT(iip->ili_last_fields);1625		goto out_iunlock;1626	}1627 1628	/*1629	 * Inodes not attached to the buffer can be released immediately.1630	 * Everything else has to go through xfs_iflush_abort() on journal1631	 * commit as the flock synchronises removal of the inode from the1632	 * cluster buffer against inode reclaim.1633	 */1634	if (!iip || list_empty(&iip->ili_item.li_bio_list))1635		goto out_iunlock;1636 1637	__xfs_iflags_set(ip, XFS_IFLUSHING);1638	spin_unlock(&ip->i_flags_lock);1639	rcu_read_unlock();1640 1641	/* we have a dirty inode in memory that has not yet been flushed. */1642	spin_lock(&iip->ili_lock);1643	iip->ili_last_fields = iip->ili_fields;1644	iip->ili_fields = 0;1645	iip->ili_fsync_fields = 0;1646	spin_unlock(&iip->ili_lock);1647	ASSERT(iip->ili_last_fields);1648 1649	if (ip != free_ip)1650		xfs_iunlock(ip, XFS_ILOCK_EXCL);1651	return;1652 1653out_iunlock:1654	if (ip != free_ip)1655		xfs_iunlock(ip, XFS_ILOCK_EXCL);1656out_iflags_unlock:1657	spin_unlock(&ip->i_flags_lock);1658	rcu_read_unlock();1659}1660 1661/*1662 * A big issue when freeing the inode cluster is that we _cannot_ skip any1663 * inodes that are in memory - they all must be marked stale and attached to1664 * the cluster buffer.1665 */1666static int1667xfs_ifree_cluster(1668	struct xfs_trans	*tp,1669	struct xfs_perag	*pag,1670	struct xfs_inode	*free_ip,1671	struct xfs_icluster	*xic)1672{1673	struct xfs_mount	*mp = free_ip->i_mount;1674	struct xfs_ino_geometry	*igeo = M_IGEO(mp);1675	struct xfs_buf		*bp;1676	xfs_daddr_t		blkno;1677	xfs_ino_t		inum = xic->first_ino;1678	int			nbufs;1679	int			i, j;1680	int			ioffset;1681	int			error;1682 1683	nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;1684 1685	for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {1686		/*1687		 * The allocation bitmap tells us which inodes of the chunk were1688		 * physically allocated. Skip the cluster if an inode falls into1689		 * a sparse region.1690		 */1691		ioffset = inum - xic->first_ino;1692		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {1693			ASSERT(ioffset % igeo->inodes_per_cluster == 0);1694			continue;1695		}1696 1697		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),1698					 XFS_INO_TO_AGBNO(mp, inum));1699 1700		/*1701		 * We obtain and lock the backing buffer first in the process1702		 * here to ensure dirty inodes attached to the buffer remain in1703		 * the flushing state while we mark them stale.1704		 *1705		 * If we scan the in-memory inodes first, then buffer IO can1706		 * complete before we get a lock on it, and hence we may fail1707		 * to mark all the active inodes on the buffer stale.1708		 */1709		error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,1710				mp->m_bsize * igeo->blocks_per_cluster,1711				XBF_UNMAPPED, &bp);1712		if (error)1713			return error;1714 1715		/*1716		 * This buffer may not have been correctly initialised as we1717		 * didn't read it from disk. That's not important because we are1718		 * only using to mark the buffer as stale in the log, and to1719		 * attach stale cached inodes on it.1720		 *1721		 * For the inode that triggered the cluster freeing, this1722		 * attachment may occur in xfs_inode_item_precommit() after we1723		 * have marked this buffer stale.  If this buffer was not in1724		 * memory before xfs_ifree_cluster() started, it will not be1725		 * marked XBF_DONE and this will cause problems later in1726		 * xfs_inode_item_precommit() when we trip over a (stale, !done)1727		 * buffer to attached to the transaction.1728		 *1729		 * Hence we have to mark the buffer as XFS_DONE here. This is1730		 * safe because we are also marking the buffer as XBF_STALE and1731		 * XFS_BLI_STALE. That means it will never be dispatched for1732		 * IO and it won't be unlocked until the cluster freeing has1733		 * been committed to the journal and the buffer unpinned. If it1734		 * is written, we want to know about it, and we want it to1735		 * fail. We can acheive this by adding a write verifier to the1736		 * buffer.1737		 */1738		bp->b_flags |= XBF_DONE;1739		bp->b_ops = &xfs_inode_buf_ops;1740 1741		/*1742		 * Now we need to set all the cached clean inodes as XFS_ISTALE,1743		 * too. This requires lookups, and will skip inodes that we've1744		 * already marked XFS_ISTALE.1745		 */1746		for (i = 0; i < igeo->inodes_per_cluster; i++)1747			xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);1748 1749		xfs_trans_stale_inode_buf(tp, bp);1750		xfs_trans_binval(tp, bp);1751	}1752	return 0;1753}1754 1755/*1756 * This is called to return an inode to the inode free list.  The inode should1757 * already be truncated to 0 length and have no pages associated with it.  This1758 * routine also assumes that the inode is already a part of the transaction.1759 *1760 * The on-disk copy of the inode will have been added to the list of unlinked1761 * inodes in the AGI. We need to remove the inode from that list atomically with1762 * respect to freeing it here.1763 */1764int1765xfs_ifree(1766	struct xfs_trans	*tp,1767	struct xfs_inode	*ip)1768{1769	struct xfs_mount	*mp = ip->i_mount;1770	struct xfs_perag	*pag;1771	struct xfs_icluster	xic = { 0 };1772	struct xfs_inode_log_item *iip = ip->i_itemp;1773	int			error;1774 1775	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);1776	ASSERT(VFS_I(ip)->i_nlink == 0);1777	ASSERT(ip->i_df.if_nextents == 0);1778	ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));1779	ASSERT(ip->i_nblocks == 0);1780 1781	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));1782 1783	error = xfs_inode_uninit(tp, pag, ip, &xic);1784	if (error)1785		goto out;1786 1787	if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))1788		xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);1789 1790	/* Don't attempt to replay owner changes for a deleted inode */1791	spin_lock(&iip->ili_lock);1792	iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);1793	spin_unlock(&iip->ili_lock);1794 1795	if (xic.deleted)1796		error = xfs_ifree_cluster(tp, pag, ip, &xic);1797out:1798	xfs_perag_put(pag);1799	return error;1800}1801 1802/*1803 * This is called to unpin an inode.  The caller must have the inode locked1804 * in at least shared mode so that the buffer cannot be subsequently pinned1805 * once someone is waiting for it to be unpinned.1806 */1807static void1808xfs_iunpin(1809	struct xfs_inode	*ip)1810{1811	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);1812 1813	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);1814 1815	/* Give the log a push to start the unpinning I/O */1816	xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);1817 1818}1819 1820static void1821__xfs_iunpin_wait(1822	struct xfs_inode	*ip)1823{1824	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);1825	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);1826 1827	xfs_iunpin(ip);1828 1829	do {1830		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);1831		if (xfs_ipincount(ip))1832			io_schedule();1833	} while (xfs_ipincount(ip));1834	finish_wait(wq, &wait.wq_entry);1835}1836 1837void1838xfs_iunpin_wait(1839	struct xfs_inode	*ip)1840{1841	if (xfs_ipincount(ip))1842		__xfs_iunpin_wait(ip);1843}1844 1845/*1846 * Removing an inode from the namespace involves removing the directory entry1847 * and dropping the link count on the inode. Removing the directory entry can1848 * result in locking an AGF (directory blocks were freed) and removing a link1849 * count can result in placing the inode on an unlinked list which results in1850 * locking an AGI.1851 *1852 * The big problem here is that we have an ordering constraint on AGF and AGI1853 * locking - inode allocation locks the AGI, then can allocate a new extent for1854 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode1855 * removes the inode from the unlinked list, requiring that we lock the AGI1856 * first, and then freeing the inode can result in an inode chunk being freed1857 * and hence freeing disk space requiring that we lock an AGF.1858 *1859 * Hence the ordering that is imposed by other parts of the code is AGI before1860 * AGF. This means we cannot remove the directory entry before we drop the inode1861 * reference count and put it on the unlinked list as this results in a lock1862 * order of AGF then AGI, and this can deadlock against inode allocation and1863 * freeing. Therefore we must drop the link counts before we remove the1864 * directory entry.1865 *1866 * This is still safe from a transactional point of view - it is not until we1867 * get to xfs_defer_finish() that we have the possibility of multiple1868 * transactions in this operation. Hence as long as we remove the directory1869 * entry and drop the link count in the first transaction of the remove1870 * operation, there are no transactional constraints on the ordering here.1871 */1872int1873xfs_remove(1874	struct xfs_inode	*dp,1875	struct xfs_name		*name,1876	struct xfs_inode	*ip)1877{1878	struct xfs_dir_update	du = {1879		.dp		= dp,1880		.name		= name,1881		.ip		= ip,1882	};1883	struct xfs_mount	*mp = dp->i_mount;1884	struct xfs_trans	*tp = NULL;1885	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);1886	int			dontcare;1887	int                     error = 0;1888	uint			resblks;1889 1890	trace_xfs_remove(dp, name);1891 1892	if (xfs_is_shutdown(mp))1893		return -EIO;1894	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))1895		return -EIO;1896 1897	error = xfs_qm_dqattach(dp);1898	if (error)1899		goto std_return;1900 1901	error = xfs_qm_dqattach(ip);1902	if (error)1903		goto std_return;1904 1905	error = xfs_parent_start(mp, &du.ppargs);1906	if (error)1907		goto std_return;1908 1909	/*1910	 * We try to get the real space reservation first, allowing for1911	 * directory btree deletion(s) implying possible bmap insert(s).  If we1912	 * can't get the space reservation then we use 0 instead, and avoid the1913	 * bmap btree insert(s) in the directory code by, if the bmap insert1914	 * tries to happen, instead trimming the LAST block from the directory.1915	 *1916	 * Ignore EDQUOT and ENOSPC being returned via nospace_error because1917	 * the directory code can handle a reservationless update and we don't1918	 * want to prevent a user from trying to free space by deleting things.1919	 */1920	resblks = xfs_remove_space_res(mp, name->len);1921	error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,1922			&tp, &dontcare);1923	if (error) {1924		ASSERT(error != -ENOSPC);1925		goto out_parent;1926	}1927 1928	error = xfs_dir_remove_child(tp, resblks, &du);1929	if (error)1930		goto out_trans_cancel;1931 1932	/*1933	 * If this is a synchronous mount, make sure that the1934	 * remove transaction goes to disk before returning to1935	 * the user.1936	 */1937	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))1938		xfs_trans_set_sync(tp);1939 1940	error = xfs_trans_commit(tp);1941	if (error)1942		goto out_unlock;1943 1944	if (is_dir && xfs_inode_is_filestream(ip))1945		xfs_filestream_deassociate(ip);1946 1947	xfs_iunlock(ip, XFS_ILOCK_EXCL);1948	xfs_iunlock(dp, XFS_ILOCK_EXCL);1949	xfs_parent_finish(mp, du.ppargs);1950	return 0;1951 1952 out_trans_cancel:1953	xfs_trans_cancel(tp);1954 out_unlock:1955	xfs_iunlock(ip, XFS_ILOCK_EXCL);1956	xfs_iunlock(dp, XFS_ILOCK_EXCL);1957 out_parent:1958	xfs_parent_finish(mp, du.ppargs);1959 std_return:1960	return error;1961}1962 1963static inline void1964xfs_iunlock_rename(1965	struct xfs_inode	**i_tab,1966	int			num_inodes)1967{1968	int			i;1969 1970	for (i = num_inodes - 1; i >= 0; i--) {1971		/* Skip duplicate inodes if src and target dps are the same */1972		if (!i_tab[i] || (i > 0 && i_tab[i] == i_tab[i - 1]))1973			continue;1974		xfs_iunlock(i_tab[i], XFS_ILOCK_EXCL);1975	}1976}1977 1978/*1979 * Enter all inodes for a rename transaction into a sorted array.1980 */1981#define __XFS_SORT_INODES	51982STATIC void1983xfs_sort_for_rename(1984	struct xfs_inode	*dp1,	/* in: old (source) directory inode */1985	struct xfs_inode	*dp2,	/* in: new (target) directory inode */1986	struct xfs_inode	*ip1,	/* in: inode of old entry */1987	struct xfs_inode	*ip2,	/* in: inode of new entry */1988	struct xfs_inode	*wip,	/* in: whiteout inode */1989	struct xfs_inode	**i_tab,/* out: sorted array of inodes */1990	int			*num_inodes)  /* in/out: inodes in array */1991{1992	int			i;1993 1994	ASSERT(*num_inodes == __XFS_SORT_INODES);1995	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));1996 1997	/*1998	 * i_tab contains a list of pointers to inodes.  We initialize1999	 * the table here & we'll sort it.  We will then use it to2000	 * order the acquisition of the inode locks.2001	 *2002	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.2003	 */2004	i = 0;2005	i_tab[i++] = dp1;2006	i_tab[i++] = dp2;2007	i_tab[i++] = ip1;2008	if (ip2)2009		i_tab[i++] = ip2;2010	if (wip)2011		i_tab[i++] = wip;2012	*num_inodes = i;2013 2014	xfs_sort_inodes(i_tab, *num_inodes);2015}2016 2017void2018xfs_sort_inodes(2019	struct xfs_inode	**i_tab,2020	unsigned int		num_inodes)2021{2022	int			i, j;2023 2024	ASSERT(num_inodes <= __XFS_SORT_INODES);2025 2026	/*2027	 * Sort the elements via bubble sort.  (Remember, there are at2028	 * most 5 elements to sort, so this is adequate.)2029	 */2030	for (i = 0; i < num_inodes; i++) {2031		for (j = 1; j < num_inodes; j++) {2032			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino)2033				swap(i_tab[j], i_tab[j - 1]);2034		}2035	}2036}2037 2038/*2039 * xfs_rename_alloc_whiteout()2040 *2041 * Return a referenced, unlinked, unlocked inode that can be used as a2042 * whiteout in a rename transaction. We use a tmpfile inode here so that if we2043 * crash between allocating the inode and linking it into the rename transaction2044 * recovery will free the inode and we won't leak it.2045 */2046static int2047xfs_rename_alloc_whiteout(2048	struct mnt_idmap	*idmap,2049	struct xfs_name		*src_name,2050	struct xfs_inode	*dp,2051	struct xfs_inode	**wip)2052{2053	struct xfs_icreate_args	args = {2054		.idmap		= idmap,2055		.pip		= dp,2056		.mode		= S_IFCHR | WHITEOUT_MODE,2057		.flags		= XFS_ICREATE_TMPFILE,2058	};2059	struct xfs_inode	*tmpfile;2060	struct qstr		name;2061	int			error;2062 2063	error = xfs_create_tmpfile(&args, &tmpfile);2064	if (error)2065		return error;2066 2067	name.name = src_name->name;2068	name.len = src_name->len;2069	error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);2070	if (error) {2071		xfs_finish_inode_setup(tmpfile);2072		xfs_irele(tmpfile);2073		return error;2074	}2075 2076	/*2077	 * Prepare the tmpfile inode as if it were created through the VFS.2078	 * Complete the inode setup and flag it as linkable.  nlink is already2079	 * zero, so we can skip the drop_nlink.2080	 */2081	xfs_setup_iops(tmpfile);2082	xfs_finish_inode_setup(tmpfile);2083	VFS_I(tmpfile)->i_state |= I_LINKABLE;2084 2085	*wip = tmpfile;2086	return 0;2087}2088 2089/*2090 * xfs_rename2091 */2092int2093xfs_rename(2094	struct mnt_idmap	*idmap,2095	struct xfs_inode	*src_dp,2096	struct xfs_name		*src_name,2097	struct xfs_inode	*src_ip,2098	struct xfs_inode	*target_dp,2099	struct xfs_name		*target_name,2100	struct xfs_inode	*target_ip,2101	unsigned int		flags)2102{2103	struct xfs_dir_update	du_src = {2104		.dp		= src_dp,2105		.name		= src_name,2106		.ip		= src_ip,2107	};2108	struct xfs_dir_update	du_tgt = {2109		.dp		= target_dp,2110		.name		= target_name,2111		.ip		= target_ip,2112	};2113	struct xfs_dir_update	du_wip = { };2114	struct xfs_mount	*mp = src_dp->i_mount;2115	struct xfs_trans	*tp;2116	struct xfs_inode	*inodes[__XFS_SORT_INODES];2117	int			i;2118	int			num_inodes = __XFS_SORT_INODES;2119	bool			new_parent = (src_dp != target_dp);2120	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);2121	int			spaceres;2122	bool			retried = false;2123	int			error, nospace_error = 0;2124 2125	trace_xfs_rename(src_dp, target_dp, src_name, target_name);2126 2127	if ((flags & RENAME_EXCHANGE) && !target_ip)2128		return -EINVAL;2129 2130	/*2131	 * If we are doing a whiteout operation, allocate the whiteout inode2132	 * we will be placing at the target and ensure the type is set2133	 * appropriately.2134	 */2135	if (flags & RENAME_WHITEOUT) {2136		error = xfs_rename_alloc_whiteout(idmap, src_name, target_dp,2137				&du_wip.ip);2138		if (error)2139			return error;2140 2141		/* setup target dirent info as whiteout */2142		src_name->type = XFS_DIR3_FT_CHRDEV;2143	}2144 2145	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, du_wip.ip,2146			inodes, &num_inodes);2147 2148	error = xfs_parent_start(mp, &du_src.ppargs);2149	if (error)2150		goto out_release_wip;2151 2152	if (du_wip.ip) {2153		error = xfs_parent_start(mp, &du_wip.ppargs);2154		if (error)2155			goto out_src_ppargs;2156	}2157 2158	if (target_ip) {2159		error = xfs_parent_start(mp, &du_tgt.ppargs);2160		if (error)2161			goto out_wip_ppargs;2162	}2163 2164retry:2165	nospace_error = 0;2166	spaceres = xfs_rename_space_res(mp, src_name->len, target_ip != NULL,2167			target_name->len, du_wip.ip != NULL);2168	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);2169	if (error == -ENOSPC) {2170		nospace_error = error;2171		spaceres = 0;2172		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,2173				&tp);2174	}2175	if (error)2176		goto out_tgt_ppargs;2177 2178	/*2179	 * We don't allow reservationless renaming when parent pointers are2180	 * enabled because we can't back out if the xattrs must grow.2181	 */2182	if (du_src.ppargs && nospace_error) {2183		error = nospace_error;2184		xfs_trans_cancel(tp);2185		goto out_tgt_ppargs;2186	}2187 2188	/*2189	 * Attach the dquots to the inodes2190	 */2191	error = xfs_qm_vop_rename_dqattach(inodes);2192	if (error) {2193		xfs_trans_cancel(tp);2194		goto out_tgt_ppargs;2195	}2196 2197	/*2198	 * Lock all the participating inodes. Depending upon whether2199	 * the target_name exists in the target directory, and2200	 * whether the target directory is the same as the source2201	 * directory, we can lock from 2 to 5 inodes.2202	 */2203	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);2204 2205	/*2206	 * Join all the inodes to the transaction.2207	 */2208	xfs_trans_ijoin(tp, src_dp, 0);2209	if (new_parent)2210		xfs_trans_ijoin(tp, target_dp, 0);2211	xfs_trans_ijoin(tp, src_ip, 0);2212	if (target_ip)2213		xfs_trans_ijoin(tp, target_ip, 0);2214	if (du_wip.ip)2215		xfs_trans_ijoin(tp, du_wip.ip, 0);2216 2217	/*2218	 * If we are using project inheritance, we only allow renames2219	 * into our tree when the project IDs are the same; else the2220	 * tree quota mechanism would be circumvented.2221	 */2222	if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&2223		     target_dp->i_projid != src_ip->i_projid)) {2224		error = -EXDEV;2225		goto out_trans_cancel;2226	}2227 2228	/* RENAME_EXCHANGE is unique from here on. */2229	if (flags & RENAME_EXCHANGE) {2230		error = xfs_dir_exchange_children(tp, &du_src, &du_tgt,2231				spaceres);2232		if (error)2233			goto out_trans_cancel;2234		goto out_commit;2235	}2236 2237	/*2238	 * Try to reserve quota to handle an expansion of the target directory.2239	 * We'll allow the rename to continue in reservationless mode if we hit2240	 * a space usage constraint.  If we trigger reservationless mode, save2241	 * the errno if there isn't any free space in the target directory.2242	 */2243	if (spaceres != 0) {2244		error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,2245				0, false);2246		if (error == -EDQUOT || error == -ENOSPC) {2247			if (!retried) {2248				xfs_trans_cancel(tp);2249				xfs_iunlock_rename(inodes, num_inodes);2250				xfs_blockgc_free_quota(target_dp, 0);2251				retried = true;2252				goto retry;2253			}2254 2255			nospace_error = error;2256			spaceres = 0;2257			error = 0;2258		}2259		if (error)2260			goto out_trans_cancel;2261	}2262 2263	/*2264	 * We don't allow quotaless renaming when parent pointers are enabled2265	 * because we can't back out if the xattrs must grow.2266	 */2267	if (du_src.ppargs && nospace_error) {2268		error = nospace_error;2269		goto out_trans_cancel;2270	}2271 2272	/*2273	 * Lock the AGI buffers we need to handle bumping the nlink of the2274	 * whiteout inode off the unlinked list and to handle dropping the2275	 * nlink of the target inode.  Per locking order rules, do this in2276	 * increasing AG order and before directory block allocation tries to2277	 * grab AGFs because we grab AGIs before AGFs.2278	 *2279	 * The (vfs) caller must ensure that if src is a directory then2280	 * target_ip is either null or an empty directory.2281	 */2282	for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {2283		if (inodes[i] == du_wip.ip ||2284		    (inodes[i] == target_ip &&2285		     (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {2286			struct xfs_perag	*pag;2287			struct xfs_buf		*bp;2288 2289			pag = xfs_perag_get(mp,2290					XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));2291			error = xfs_read_agi(pag, tp, 0, &bp);2292			xfs_perag_put(pag);2293			if (error)2294				goto out_trans_cancel;2295		}2296	}2297 2298	error = xfs_dir_rename_children(tp, &du_src, &du_tgt, spaceres,2299			&du_wip);2300	if (error)2301		goto out_trans_cancel;2302 2303	if (du_wip.ip) {2304		/*2305		 * Now we have a real link, clear the "I'm a tmpfile" state2306		 * flag from the inode so it doesn't accidentally get misused in2307		 * future.2308		 */2309		VFS_I(du_wip.ip)->i_state &= ~I_LINKABLE;2310	}2311 2312out_commit:2313	/*2314	 * If this is a synchronous mount, make sure that the rename2315	 * transaction goes to disk before returning to the user.2316	 */2317	if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))2318		xfs_trans_set_sync(tp);2319 2320	error = xfs_trans_commit(tp);2321	nospace_error = 0;2322	goto out_unlock;2323 2324out_trans_cancel:2325	xfs_trans_cancel(tp);2326out_unlock:2327	xfs_iunlock_rename(inodes, num_inodes);2328out_tgt_ppargs:2329	xfs_parent_finish(mp, du_tgt.ppargs);2330out_wip_ppargs:2331	xfs_parent_finish(mp, du_wip.ppargs);2332out_src_ppargs:2333	xfs_parent_finish(mp, du_src.ppargs);2334out_release_wip:2335	if (du_wip.ip)2336		xfs_irele(du_wip.ip);2337	if (error == -ENOSPC && nospace_error)2338		error = nospace_error;2339	return error;2340}2341 2342static int2343xfs_iflush(2344	struct xfs_inode	*ip,2345	struct xfs_buf		*bp)2346{2347	struct xfs_inode_log_item *iip = ip->i_itemp;2348	struct xfs_dinode	*dip;2349	struct xfs_mount	*mp = ip->i_mount;2350	int			error;2351 2352	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);2353	ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));2354	ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||2355	       ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));2356	ASSERT(iip->ili_item.li_buf == bp);2357 2358	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);2359 2360	/*2361	 * We don't flush the inode if any of the following checks fail, but we2362	 * do still update the log item and attach to the backing buffer as if2363	 * the flush happened. This is a formality to facilitate predictable2364	 * error handling as the caller will shutdown and fail the buffer.2365	 */2366	error = -EFSCORRUPTED;2367	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),2368			       mp, XFS_ERRTAG_IFLUSH_1)) {2369		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,2370			"%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,2371			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);2372		goto flush_out;2373	}2374	if (S_ISREG(VFS_I(ip)->i_mode)) {2375		if (XFS_TEST_ERROR(2376		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&2377		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE,2378		    mp, XFS_ERRTAG_IFLUSH_3)) {2379			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,2380				"%s: Bad regular inode %llu, ptr "PTR_FMT,2381				__func__, ip->i_ino, ip);2382			goto flush_out;2383		}2384	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {2385		if (XFS_TEST_ERROR(2386		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&2387		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&2388		    ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,2389		    mp, XFS_ERRTAG_IFLUSH_4)) {2390			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,2391				"%s: Bad directory inode %llu, ptr "PTR_FMT,2392				__func__, ip->i_ino, ip);2393			goto flush_out;2394		}2395	}2396	if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >2397				ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {2398		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,2399			"%s: detected corrupt incore inode %llu, "2400			"total extents = %llu nblocks = %lld, ptr "PTR_FMT,2401			__func__, ip->i_ino,2402			ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),2403			ip->i_nblocks, ip);2404		goto flush_out;2405	}2406	if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,2407				mp, XFS_ERRTAG_IFLUSH_6)) {2408		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,2409			"%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,2410			__func__, ip->i_ino, ip->i_forkoff, ip);2411		goto flush_out;2412	}2413 2414	/*2415	 * Inode item log recovery for v2 inodes are dependent on the flushiter2416	 * count for correct sequencing.  We bump the flush iteration count so2417	 * we can detect flushes which postdate a log record during recovery.2418	 * This is redundant as we now log every change and hence this can't2419	 * happen but we need to still do it to ensure backwards compatibility2420	 * with old kernels that predate logging all inode changes.2421	 */2422	if (!xfs_has_v3inodes(mp))2423		ip->i_flushiter++;2424 2425	/*2426	 * If there are inline format data / attr forks attached to this inode,2427	 * make sure they are not corrupt.2428	 */2429	if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&2430	    xfs_ifork_verify_local_data(ip))2431		goto flush_out;2432	if (xfs_inode_has_attr_fork(ip) &&2433	    ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&2434	    xfs_ifork_verify_local_attr(ip))2435		goto flush_out;2436 2437	/*2438	 * Copy the dirty parts of the inode into the on-disk inode.  We always2439	 * copy out the core of the inode, because if the inode is dirty at all2440	 * the core must be.2441	 */2442	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);2443 2444	/* Wrap, we never let the log put out DI_MAX_FLUSH */2445	if (!xfs_has_v3inodes(mp)) {2446		if (ip->i_flushiter == DI_MAX_FLUSH)2447			ip->i_flushiter = 0;2448	}2449 2450	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);2451	if (xfs_inode_has_attr_fork(ip))2452		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);2453 2454	/*2455	 * We've recorded everything logged in the inode, so we'd like to clear2456	 * the ili_fields bits so we don't log and flush things unnecessarily.2457	 * However, we can't stop logging all this information until the data2458	 * we've copied into the disk buffer is written to disk.  If we did we2459	 * might overwrite the copy of the inode in the log with all the data2460	 * after re-logging only part of it, and in the face of a crash we2461	 * wouldn't have all the data we need to recover.2462	 *2463	 * What we do is move the bits to the ili_last_fields field.  When2464	 * logging the inode, these bits are moved back to the ili_fields field.2465	 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since2466	 * we know that the information those bits represent is permanently on2467	 * disk.  As long as the flush completes before the inode is logged2468	 * again, then both ili_fields and ili_last_fields will be cleared.2469	 */2470	error = 0;2471flush_out:2472	spin_lock(&iip->ili_lock);2473	iip->ili_last_fields = iip->ili_fields;2474	iip->ili_fields = 0;2475	iip->ili_fsync_fields = 0;2476	set_bit(XFS_LI_FLUSHING, &iip->ili_item.li_flags);2477	spin_unlock(&iip->ili_lock);2478 2479	/*2480	 * Store the current LSN of the inode so that we can tell whether the2481	 * item has moved in the AIL from xfs_buf_inode_iodone().2482	 */2483	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,2484				&iip->ili_item.li_lsn);2485 2486	/* generate the checksum. */2487	xfs_dinode_calc_crc(mp, dip);2488	if (error)2489		xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);2490	return error;2491}2492 2493/*2494 * Non-blocking flush of dirty inode metadata into the backing buffer.2495 *2496 * The caller must have a reference to the inode and hold the cluster buffer2497 * locked. The function will walk across all the inodes on the cluster buffer it2498 * can find and lock without blocking, and flush them to the cluster buffer.2499 *2500 * On successful flushing of at least one inode, the caller must write out the2501 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and2502 * the caller needs to release the buffer. On failure, the filesystem will be2503 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED2504 * will be returned.2505 */2506int2507xfs_iflush_cluster(2508	struct xfs_buf		*bp)2509{2510	struct xfs_mount	*mp = bp->b_mount;2511	struct xfs_log_item	*lip, *n;2512	struct xfs_inode	*ip;2513	struct xfs_inode_log_item *iip;2514	int			clcount = 0;2515	int			error = 0;2516 2517	/*2518	 * We must use the safe variant here as on shutdown xfs_iflush_abort()2519	 * will remove itself from the list.2520	 */2521	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {2522		iip = (struct xfs_inode_log_item *)lip;2523		ip = iip->ili_inode;2524 2525		/*2526		 * Quick and dirty check to avoid locks if possible.2527		 */2528		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))2529			continue;2530		if (xfs_ipincount(ip))2531			continue;2532 2533		/*2534		 * The inode is still attached to the buffer, which means it is2535		 * dirty but reclaim might try to grab it. Check carefully for2536		 * that, and grab the ilock while still holding the i_flags_lock2537		 * to guarantee reclaim will not be able to reclaim this inode2538		 * once we drop the i_flags_lock.2539		 */2540		spin_lock(&ip->i_flags_lock);2541		ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));2542		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {2543			spin_unlock(&ip->i_flags_lock);2544			continue;2545		}2546 2547		/*2548		 * ILOCK will pin the inode against reclaim and prevent2549		 * concurrent transactions modifying the inode while we are2550		 * flushing the inode. If we get the lock, set the flushing2551		 * state before we drop the i_flags_lock.2552		 */2553		if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {2554			spin_unlock(&ip->i_flags_lock);2555			continue;2556		}2557		__xfs_iflags_set(ip, XFS_IFLUSHING);2558		spin_unlock(&ip->i_flags_lock);2559 2560		/*2561		 * Abort flushing this inode if we are shut down because the2562		 * inode may not currently be in the AIL. This can occur when2563		 * log I/O failure unpins the inode without inserting into the2564		 * AIL, leaving a dirty/unpinned inode attached to the buffer2565		 * that otherwise looks like it should be flushed.2566		 */2567		if (xlog_is_shutdown(mp->m_log)) {2568			xfs_iunpin_wait(ip);2569			xfs_iflush_abort(ip);2570			xfs_iunlock(ip, XFS_ILOCK_SHARED);2571			error = -EIO;2572			continue;2573		}2574 2575		/* don't block waiting on a log force to unpin dirty inodes */2576		if (xfs_ipincount(ip)) {2577			xfs_iflags_clear(ip, XFS_IFLUSHING);2578			xfs_iunlock(ip, XFS_ILOCK_SHARED);2579			continue;2580		}2581 2582		if (!xfs_inode_clean(ip))2583			error = xfs_iflush(ip, bp);2584		else2585			xfs_iflags_clear(ip, XFS_IFLUSHING);2586		xfs_iunlock(ip, XFS_ILOCK_SHARED);2587		if (error)2588			break;2589		clcount++;2590	}2591 2592	if (error) {2593		/*2594		 * Shutdown first so we kill the log before we release this2595		 * buffer. If it is an INODE_ALLOC buffer and pins the tail2596		 * of the log, failing it before the _log_ is shut down can2597		 * result in the log tail being moved forward in the journal2598		 * on disk because log writes can still be taking place. Hence2599		 * unpinning the tail will allow the ICREATE intent to be2600		 * removed from the log an recovery will fail with uninitialised2601		 * inode cluster buffers.2602		 */2603		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);2604		bp->b_flags |= XBF_ASYNC;2605		xfs_buf_ioend_fail(bp);2606		return error;2607	}2608 2609	if (!clcount)2610		return -EAGAIN;2611 2612	XFS_STATS_INC(mp, xs_icluster_flushcnt);2613	XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);2614	return 0;2615 2616}2617 2618/* Release an inode. */2619void2620xfs_irele(2621	struct xfs_inode	*ip)2622{2623	trace_xfs_irele(ip, _RET_IP_);2624	iput(VFS_I(ip));2625}2626 2627/*2628 * Ensure all commited transactions touching the inode are written to the log.2629 */2630int2631xfs_log_force_inode(2632	struct xfs_inode	*ip)2633{2634	xfs_csn_t		seq = 0;2635 2636	xfs_ilock(ip, XFS_ILOCK_SHARED);2637	if (xfs_ipincount(ip))2638		seq = ip->i_itemp->ili_commit_seq;2639	xfs_iunlock(ip, XFS_ILOCK_SHARED);2640 2641	if (!seq)2642		return 0;2643	return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);2644}2645 2646/*2647 * Grab the exclusive iolock for a data copy from src to dest, making sure to2648 * abide vfs locking order (lowest pointer value goes first) and breaking the2649 * layout leases before proceeding.  The loop is needed because we cannot call2650 * the blocking break_layout() with the iolocks held, and therefore have to2651 * back out both locks.2652 */2653static int2654xfs_iolock_two_inodes_and_break_layout(2655	struct inode		*src,2656	struct inode		*dest)2657{2658	int			error;2659 2660	if (src > dest)2661		swap(src, dest);2662 2663retry:2664	/* Wait to break both inodes' layouts before we start locking. */2665	error = break_layout(src, true);2666	if (error)2667		return error;2668	if (src != dest) {2669		error = break_layout(dest, true);2670		if (error)2671			return error;2672	}2673 2674	/* Lock one inode and make sure nobody got in and leased it. */2675	inode_lock(src);2676	error = break_layout(src, false);2677	if (error) {2678		inode_unlock(src);2679		if (error == -EWOULDBLOCK)2680			goto retry;2681		return error;2682	}2683 2684	if (src == dest)2685		return 0;2686 2687	/* Lock the other inode and make sure nobody got in and leased it. */2688	inode_lock_nested(dest, I_MUTEX_NONDIR2);2689	error = break_layout(dest, false);2690	if (error) {2691		inode_unlock(src);2692		inode_unlock(dest);2693		if (error == -EWOULDBLOCK)2694			goto retry;2695		return error;2696	}2697 2698	return 0;2699}2700 2701static int2702xfs_mmaplock_two_inodes_and_break_dax_layout(2703	struct xfs_inode	*ip1,2704	struct xfs_inode	*ip2)2705{2706	int			error;2707	bool			retry;2708	struct page		*page;2709 2710	if (ip1->i_ino > ip2->i_ino)2711		swap(ip1, ip2);2712 2713again:2714	retry = false;2715	/* Lock the first inode */2716	xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);2717	error = xfs_break_dax_layouts(VFS_I(ip1), &retry);2718	if (error || retry) {2719		xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);2720		if (error == 0 && retry)2721			goto again;2722		return error;2723	}2724 2725	if (ip1 == ip2)2726		return 0;2727 2728	/* Nested lock the second inode */2729	xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));2730	/*2731	 * We cannot use xfs_break_dax_layouts() directly here because it may2732	 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable2733	 * for this nested lock case.2734	 */2735	page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);2736	if (page && page_ref_count(page) != 1) {2737		xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);2738		xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);2739		goto again;2740	}2741 2742	return 0;2743}2744 2745/*2746 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or2747 * mmap activity.2748 */2749int2750xfs_ilock2_io_mmap(2751	struct xfs_inode	*ip1,2752	struct xfs_inode	*ip2)2753{2754	int			ret;2755 2756	ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));2757	if (ret)2758		return ret;2759 2760	if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {2761		ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);2762		if (ret) {2763			inode_unlock(VFS_I(ip2));2764			if (ip1 != ip2)2765				inode_unlock(VFS_I(ip1));2766			return ret;2767		}2768	} else2769		filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,2770					    VFS_I(ip2)->i_mapping);2771 2772	return 0;2773}2774 2775/* Unlock both inodes to allow IO and mmap activity. */2776void2777xfs_iunlock2_io_mmap(2778	struct xfs_inode	*ip1,2779	struct xfs_inode	*ip2)2780{2781	if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {2782		xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);2783		if (ip1 != ip2)2784			xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);2785	} else2786		filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,2787					      VFS_I(ip2)->i_mapping);2788 2789	inode_unlock(VFS_I(ip2));2790	if (ip1 != ip2)2791		inode_unlock(VFS_I(ip1));2792}2793 2794/* Drop the MMAPLOCK and the IOLOCK after a remap completes. */2795void2796xfs_iunlock2_remapping(2797	struct xfs_inode	*ip1,2798	struct xfs_inode	*ip2)2799{2800	xfs_iflags_clear(ip1, XFS_IREMAPPING);2801 2802	if (ip1 != ip2)2803		xfs_iunlock(ip1, XFS_MMAPLOCK_SHARED);2804	xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);2805 2806	if (ip1 != ip2)2807		inode_unlock_shared(VFS_I(ip1));2808	inode_unlock(VFS_I(ip2));2809}2810 2811/*2812 * Reload the incore inode list for this inode.  Caller should ensure that2813 * the link count cannot change, either by taking ILOCK_SHARED or otherwise2814 * preventing other threads from executing.2815 */2816int2817xfs_inode_reload_unlinked_bucket(2818	struct xfs_trans	*tp,2819	struct xfs_inode	*ip)2820{2821	struct xfs_mount	*mp = tp->t_mountp;2822	struct xfs_buf		*agibp;2823	struct xfs_agi		*agi;2824	struct xfs_perag	*pag;2825	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, ip->i_ino);2826	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ip->i_ino);2827	xfs_agino_t		prev_agino, next_agino;2828	unsigned int		bucket;2829	bool			foundit = false;2830	int			error;2831 2832	/* Grab the first inode in the list */2833	pag = xfs_perag_get(mp, agno);2834	error = xfs_ialloc_read_agi(pag, tp, 0, &agibp);2835	xfs_perag_put(pag);2836	if (error)2837		return error;2838 2839	/*2840	 * We've taken ILOCK_SHARED and the AGI buffer lock to stabilize the2841	 * incore unlinked list pointers for this inode.  Check once more to2842	 * see if we raced with anyone else to reload the unlinked list.2843	 */2844	if (!xfs_inode_unlinked_incomplete(ip)) {2845		foundit = true;2846		goto out_agibp;2847	}2848 2849	bucket = agino % XFS_AGI_UNLINKED_BUCKETS;2850	agi = agibp->b_addr;2851 2852	trace_xfs_inode_reload_unlinked_bucket(ip);2853 2854	xfs_info_ratelimited(mp,2855 "Found unrecovered unlinked inode 0x%x in AG 0x%x.  Initiating list recovery.",2856			agino, agno);2857 2858	prev_agino = NULLAGINO;2859	next_agino = be32_to_cpu(agi->agi_unlinked[bucket]);2860	while (next_agino != NULLAGINO) {2861		struct xfs_inode	*next_ip = NULL;2862 2863		/* Found this caller's inode, set its backlink. */2864		if (next_agino == agino) {2865			next_ip = ip;2866			next_ip->i_prev_unlinked = prev_agino;2867			foundit = true;2868			goto next_inode;2869		}2870 2871		/* Try in-memory lookup first. */2872		next_ip = xfs_iunlink_lookup(pag, next_agino);2873		if (next_ip)2874			goto next_inode;2875 2876		/* Inode not in memory, try reloading it. */2877		error = xfs_iunlink_reload_next(tp, agibp, prev_agino,2878				next_agino);2879		if (error)2880			break;2881 2882		/* Grab the reloaded inode. */2883		next_ip = xfs_iunlink_lookup(pag, next_agino);2884		if (!next_ip) {2885			/* No incore inode at all?  We reloaded it... */2886			ASSERT(next_ip != NULL);2887			error = -EFSCORRUPTED;2888			break;2889		}2890 2891next_inode:2892		prev_agino = next_agino;2893		next_agino = next_ip->i_next_unlinked;2894	}2895 2896out_agibp:2897	xfs_trans_brelse(tp, agibp);2898	/* Should have found this inode somewhere in the iunlinked bucket. */2899	if (!error && !foundit)2900		error = -EFSCORRUPTED;2901	return error;2902}2903 2904/* Decide if this inode is missing its unlinked list and reload it. */2905int2906xfs_inode_reload_unlinked(2907	struct xfs_inode	*ip)2908{2909	struct xfs_trans	*tp;2910	int			error;2911 2912	error = xfs_trans_alloc_empty(ip->i_mount, &tp);2913	if (error)2914		return error;2915 2916	xfs_ilock(ip, XFS_ILOCK_SHARED);2917	if (xfs_inode_unlinked_incomplete(ip))2918		error = xfs_inode_reload_unlinked_bucket(tp, ip);2919	xfs_iunlock(ip, XFS_ILOCK_SHARED);2920	xfs_trans_cancel(tp);2921 2922	return error;2923}2924 2925/* Has this inode fork been zapped by repair? */2926bool2927xfs_ifork_zapped(2928	const struct xfs_inode	*ip,2929	int			whichfork)2930{2931	unsigned int		datamask = 0;2932 2933	switch (whichfork) {2934	case XFS_DATA_FORK:2935		switch (ip->i_vnode.i_mode & S_IFMT) {2936		case S_IFDIR:2937			datamask = XFS_SICK_INO_DIR_ZAPPED;2938			break;2939		case S_IFLNK:2940			datamask = XFS_SICK_INO_SYMLINK_ZAPPED;2941			break;2942		}2943		return ip->i_sick & (XFS_SICK_INO_BMBTD_ZAPPED | datamask);2944	case XFS_ATTR_FORK:2945		return ip->i_sick & XFS_SICK_INO_BMBTA_ZAPPED;2946	default:2947		return false;2948	}2949}2950 2951/* Compute the number of data and realtime blocks used by a file. */2952void2953xfs_inode_count_blocks(2954	struct xfs_trans	*tp,2955	struct xfs_inode	*ip,2956	xfs_filblks_t		*dblocks,2957	xfs_filblks_t		*rblocks)2958{2959	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);2960 2961	*rblocks = 0;2962	if (XFS_IS_REALTIME_INODE(ip))2963		xfs_bmap_count_leaves(ifp, rblocks);2964	*dblocks = ip->i_nblocks - *rblocks;2965}2966 2967static void2968xfs_wait_dax_page(2969	struct inode		*inode)2970{2971	struct xfs_inode        *ip = XFS_I(inode);2972 2973	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);2974	schedule();2975	xfs_ilock(ip, XFS_MMAPLOCK_EXCL);2976}2977 2978int2979xfs_break_dax_layouts(2980	struct inode		*inode,2981	bool			*retry)2982{2983	struct page		*page;2984 2985	xfs_assert_ilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL);2986 2987	page = dax_layout_busy_page(inode->i_mapping);2988	if (!page)2989		return 0;2990 2991	*retry = true;2992	return ___wait_var_event(&page->_refcount,2993			atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,2994			0, 0, xfs_wait_dax_page(inode));2995}2996 2997int2998xfs_break_layouts(2999	struct inode		*inode,3000	uint			*iolock,3001	enum layout_break_reason reason)3002{3003	bool			retry;3004	int			error;3005 3006	xfs_assert_ilocked(XFS_I(inode), XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL);3007 3008	do {3009		retry = false;3010		switch (reason) {3011		case BREAK_UNMAP:3012			error = xfs_break_dax_layouts(inode, &retry);3013			if (error || retry)3014				break;3015			fallthrough;3016		case BREAK_WRITE:3017			error = xfs_break_leased_layouts(inode, iolock, &retry);3018			break;3019		default:3020			WARN_ON_ONCE(1);3021			error = -EINVAL;3022		}3023	} while (error == 0 && retry);3024 3025	return error;3026}3027 3028/* Returns the size of fundamental allocation unit for a file, in bytes. */3029unsigned int3030xfs_inode_alloc_unitsize(3031	struct xfs_inode	*ip)3032{3033	unsigned int		blocks = 1;3034 3035	if (XFS_IS_REALTIME_INODE(ip))3036		blocks = ip->i_mount->m_sb.sb_rextsize;3037 3038	return XFS_FSB_TO_B(ip->i_mount, blocks);3039}3040 3041/* Should we always be using copy on write for file writes? */3042bool3043xfs_is_always_cow_inode(3044	struct xfs_inode	*ip)3045{3046	return ip->i_mount->m_always_cow && xfs_has_reflink(ip->i_mount);3047}3048