Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   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_trans_space.h"
  20#include "xfs_trans.h"
  21#include "xfs_buf_item.h"
  22#include "xfs_inode_item.h"
  23#include "xfs_iunlink_item.h"
  24#include "xfs_ialloc.h"
  25#include "xfs_bmap.h"
  26#include "xfs_bmap_util.h"
  27#include "xfs_errortag.h"
  28#include "xfs_error.h"
  29#include "xfs_quota.h"
  30#include "xfs_filestream.h"
 
  31#include "xfs_trace.h"
  32#include "xfs_icache.h"
  33#include "xfs_symlink.h"
  34#include "xfs_trans_priv.h"
  35#include "xfs_log.h"
  36#include "xfs_bmap_btree.h"
  37#include "xfs_reflink.h"
  38#include "xfs_ag.h"
  39#include "xfs_log_priv.h"
  40
  41struct kmem_cache *xfs_inode_cache;
  42
  43/*
  44 * Used in xfs_itruncate_extents().  This is the maximum number of extents
  45 * freed from a file in a single transaction.
  46 */
  47#define	XFS_ITRUNC_MAX_EXTENTS	2
  48
  49STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
  50STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag,
  51	struct xfs_inode *);
  52
  53/*
  54 * helper function to extract extent size hint from inode
  55 */
  56xfs_extlen_t
  57xfs_get_extsz_hint(
  58	struct xfs_inode	*ip)
  59{
  60	/*
  61	 * No point in aligning allocations if we need to COW to actually
  62	 * write to them.
  63	 */
  64	if (xfs_is_always_cow_inode(ip))
  65		return 0;
  66	if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
  67		return ip->i_extsize;
  68	if (XFS_IS_REALTIME_INODE(ip))
  69		return ip->i_mount->m_sb.sb_rextsize;
  70	return 0;
  71}
  72
  73/*
  74 * Helper function to extract CoW extent size hint from inode.
  75 * Between the extent size hint and the CoW extent size hint, we
  76 * return the greater of the two.  If the value is zero (automatic),
  77 * use the default size.
  78 */
  79xfs_extlen_t
  80xfs_get_cowextsz_hint(
  81	struct xfs_inode	*ip)
  82{
  83	xfs_extlen_t		a, b;
  84
  85	a = 0;
  86	if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
  87		a = ip->i_cowextsize;
  88	b = xfs_get_extsz_hint(ip);
  89
  90	a = max(a, b);
  91	if (a == 0)
  92		return XFS_DEFAULT_COWEXTSZ_HINT;
  93	return a;
  94}
  95
  96/*
  97 * These two are wrapper routines around the xfs_ilock() routine used to
  98 * centralize some grungy code.  They are used in places that wish to lock the
  99 * inode solely for reading the extents.  The reason these places can't just
 100 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
 101 * bringing in of the extents from disk for a file in b-tree format.  If the
 102 * inode is in b-tree format, then we need to lock the inode exclusively until
 103 * the extents are read in.  Locking it exclusively all the time would limit
 104 * our parallelism unnecessarily, though.  What we do instead is check to see
 105 * if the extents have been read in yet, and only lock the inode exclusively
 106 * if they have not.
 107 *
 108 * The functions return a value which should be given to the corresponding
 109 * xfs_iunlock() call.
 110 */
 111uint
 112xfs_ilock_data_map_shared(
 113	struct xfs_inode	*ip)
 114{
 115	uint			lock_mode = XFS_ILOCK_SHARED;
 116
 117	if (xfs_need_iread_extents(&ip->i_df))
 
 118		lock_mode = XFS_ILOCK_EXCL;
 119	xfs_ilock(ip, lock_mode);
 120	return lock_mode;
 121}
 122
 123uint
 124xfs_ilock_attr_map_shared(
 125	struct xfs_inode	*ip)
 126{
 127	uint			lock_mode = XFS_ILOCK_SHARED;
 128
 129	if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
 
 130		lock_mode = XFS_ILOCK_EXCL;
 131	xfs_ilock(ip, lock_mode);
 132	return lock_mode;
 133}
 134
 135/*
 136 * You can't set both SHARED and EXCL for the same lock,
 137 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
 138 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
 139 * to set in lock_flags.
 140 */
 141static inline void
 142xfs_lock_flags_assert(
 143	uint		lock_flags)
 144{
 145	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 146		(XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 147	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 148		(XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 149	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 150		(XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 151	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 152	ASSERT(lock_flags != 0);
 153}
 154
 155/*
 156 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
 157 * multi-reader locks: invalidate_lock and the i_lock.  This routine allows
 158 * various combinations of the locks to be obtained.
 159 *
 160 * The 3 locks should always be ordered so that the IO lock is obtained first,
 161 * the mmap lock second and the ilock last in order to prevent deadlock.
 162 *
 163 * Basic locking order:
 164 *
 165 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
 166 *
 167 * mmap_lock locking order:
 168 *
 169 * i_rwsem -> page lock -> mmap_lock
 170 * mmap_lock -> invalidate_lock -> page_lock
 171 *
 172 * The difference in mmap_lock locking order mean that we cannot hold the
 173 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
 174 * can fault in pages during copy in/out (for buffered IO) or require the
 175 * mmap_lock in get_user_pages() to map the user pages into the kernel address
 176 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
 177 * fault because page faults already hold the mmap_lock.
 178 *
 179 * Hence to serialise fully against both syscall and mmap based IO, we need to
 180 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
 181 * both taken in places where we need to invalidate the page cache in a race
 182 * free manner (e.g. truncate, hole punch and other extent manipulation
 183 * functions).
 184 */
 185void
 186xfs_ilock(
 187	xfs_inode_t		*ip,
 188	uint			lock_flags)
 189{
 190	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
 191
 192	xfs_lock_flags_assert(lock_flags);
 193
 194	if (lock_flags & XFS_IOLOCK_EXCL) {
 195		down_write_nested(&VFS_I(ip)->i_rwsem,
 196				  XFS_IOLOCK_DEP(lock_flags));
 197	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 198		down_read_nested(&VFS_I(ip)->i_rwsem,
 199				 XFS_IOLOCK_DEP(lock_flags));
 200	}
 
 201
 202	if (lock_flags & XFS_MMAPLOCK_EXCL) {
 203		down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
 204				  XFS_MMAPLOCK_DEP(lock_flags));
 205	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
 206		down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
 207				 XFS_MMAPLOCK_DEP(lock_flags));
 208	}
 209
 210	if (lock_flags & XFS_ILOCK_EXCL)
 211		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 212	else if (lock_flags & XFS_ILOCK_SHARED)
 213		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 214}
 215
 216/*
 217 * This is just like xfs_ilock(), except that the caller
 218 * is guaranteed not to sleep.  It returns 1 if it gets
 219 * the requested locks and 0 otherwise.  If the IO lock is
 220 * obtained but the inode lock cannot be, then the IO lock
 221 * is dropped before returning.
 222 *
 223 * ip -- the inode being locked
 224 * lock_flags -- this parameter indicates the inode's locks to be
 225 *       to be locked.  See the comment for xfs_ilock() for a list
 226 *	 of valid values.
 227 */
 228int
 229xfs_ilock_nowait(
 230	xfs_inode_t		*ip,
 231	uint			lock_flags)
 232{
 233	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
 234
 235	xfs_lock_flags_assert(lock_flags);
 
 
 
 
 
 
 
 
 
 236
 237	if (lock_flags & XFS_IOLOCK_EXCL) {
 238		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
 239			goto out;
 240	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 241		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
 242			goto out;
 243	}
 244
 245	if (lock_flags & XFS_MMAPLOCK_EXCL) {
 246		if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
 247			goto out_undo_iolock;
 248	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
 249		if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
 250			goto out_undo_iolock;
 251	}
 252
 253	if (lock_flags & XFS_ILOCK_EXCL) {
 254		if (!mrtryupdate(&ip->i_lock))
 255			goto out_undo_mmaplock;
 256	} else if (lock_flags & XFS_ILOCK_SHARED) {
 257		if (!mrtryaccess(&ip->i_lock))
 258			goto out_undo_mmaplock;
 259	}
 260	return 1;
 261
 262out_undo_mmaplock:
 263	if (lock_flags & XFS_MMAPLOCK_EXCL)
 264		up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
 265	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 266		up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
 267out_undo_iolock:
 268	if (lock_flags & XFS_IOLOCK_EXCL)
 269		up_write(&VFS_I(ip)->i_rwsem);
 270	else if (lock_flags & XFS_IOLOCK_SHARED)
 271		up_read(&VFS_I(ip)->i_rwsem);
 272out:
 273	return 0;
 274}
 275
 276/*
 277 * xfs_iunlock() is used to drop the inode locks acquired with
 278 * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
 279 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
 280 * that we know which locks to drop.
 281 *
 282 * ip -- the inode being unlocked
 283 * lock_flags -- this parameter indicates the inode's locks to be
 284 *       to be unlocked.  See the comment for xfs_ilock() for a list
 285 *	 of valid values for this parameter.
 286 *
 287 */
 288void
 289xfs_iunlock(
 290	xfs_inode_t		*ip,
 291	uint			lock_flags)
 292{
 293	xfs_lock_flags_assert(lock_flags);
 
 
 
 
 
 
 
 
 
 
 294
 295	if (lock_flags & XFS_IOLOCK_EXCL)
 296		up_write(&VFS_I(ip)->i_rwsem);
 297	else if (lock_flags & XFS_IOLOCK_SHARED)
 298		up_read(&VFS_I(ip)->i_rwsem);
 299
 300	if (lock_flags & XFS_MMAPLOCK_EXCL)
 301		up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
 302	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 303		up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
 304
 305	if (lock_flags & XFS_ILOCK_EXCL)
 306		mrunlock_excl(&ip->i_lock);
 307	else if (lock_flags & XFS_ILOCK_SHARED)
 308		mrunlock_shared(&ip->i_lock);
 309
 310	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
 311}
 312
 313/*
 314 * give up write locks.  the i/o lock cannot be held nested
 315 * if it is being demoted.
 316 */
 317void
 318xfs_ilock_demote(
 319	xfs_inode_t		*ip,
 320	uint			lock_flags)
 321{
 322	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
 323	ASSERT((lock_flags &
 324		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
 325
 326	if (lock_flags & XFS_ILOCK_EXCL)
 327		mrdemote(&ip->i_lock);
 328	if (lock_flags & XFS_MMAPLOCK_EXCL)
 329		downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
 330	if (lock_flags & XFS_IOLOCK_EXCL)
 331		downgrade_write(&VFS_I(ip)->i_rwsem);
 332
 333	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
 334}
 335
 336#if defined(DEBUG) || defined(XFS_WARN)
 337static inline bool
 338__xfs_rwsem_islocked(
 339	struct rw_semaphore	*rwsem,
 340	bool			shared)
 341{
 342	if (!debug_locks)
 343		return rwsem_is_locked(rwsem);
 344
 345	if (!shared)
 346		return lockdep_is_held_type(rwsem, 0);
 347
 348	/*
 349	 * We are checking that the lock is held at least in shared
 350	 * mode but don't care that it might be held exclusively
 351	 * (i.e. shared | excl). Hence we check if the lock is held
 352	 * in any mode rather than an explicit shared mode.
 353	 */
 354	return lockdep_is_held_type(rwsem, -1);
 355}
 356
 357bool
 358xfs_isilocked(
 359	struct xfs_inode	*ip,
 360	uint			lock_flags)
 361{
 362	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
 363		if (!(lock_flags & XFS_ILOCK_SHARED))
 364			return !!ip->i_lock.mr_writer;
 365		return rwsem_is_locked(&ip->i_lock.mr_lock);
 366	}
 367
 368	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
 369		return __xfs_rwsem_islocked(&VFS_I(ip)->i_mapping->invalidate_lock,
 370				(lock_flags & XFS_MMAPLOCK_SHARED));
 371	}
 372
 373	if (lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) {
 374		return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem,
 375				(lock_flags & XFS_IOLOCK_SHARED));
 376	}
 377
 378	ASSERT(0);
 379	return false;
 380}
 381#endif
 382
 383/*
 384 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
 385 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
 386 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
 387 * errors and warnings.
 388 */
 389#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
 390static bool
 391xfs_lockdep_subclass_ok(
 392	int subclass)
 393{
 394	return subclass < MAX_LOCKDEP_SUBCLASSES;
 395}
 396#else
 397#define xfs_lockdep_subclass_ok(subclass)	(true)
 398#endif
 399
 400/*
 401 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
 402 * value. This can be called for any type of inode lock combination, including
 403 * parent locking. Care must be taken to ensure we don't overrun the subclass
 404 * storage fields in the class mask we build.
 405 */
 406static inline uint
 407xfs_lock_inumorder(
 408	uint	lock_mode,
 409	uint	subclass)
 410{
 411	uint	class = 0;
 412
 413	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
 414			      XFS_ILOCK_RTSUM)));
 415	ASSERT(xfs_lockdep_subclass_ok(subclass));
 416
 417	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
 418		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
 419		class += subclass << XFS_IOLOCK_SHIFT;
 420	}
 421
 422	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
 423		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
 424		class += subclass << XFS_MMAPLOCK_SHIFT;
 425	}
 426
 427	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
 428		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
 429		class += subclass << XFS_ILOCK_SHIFT;
 430	}
 431
 432	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
 433}
 434
 435/*
 436 * The following routine will lock n inodes in exclusive mode.  We assume the
 437 * caller calls us with the inodes in i_ino order.
 438 *
 439 * We need to detect deadlock where an inode that we lock is in the AIL and we
 440 * start waiting for another inode that is locked by a thread in a long running
 441 * transaction (such as truncate). This can result in deadlock since the long
 442 * running trans might need to wait for the inode we just locked in order to
 443 * push the tail and free space in the log.
 444 *
 445 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
 446 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
 447 * lock more than one at a time, lockdep will report false positives saying we
 448 * have violated locking orders.
 
 
 449 */
 450static void
 451xfs_lock_inodes(
 452	struct xfs_inode	**ips,
 453	int			inodes,
 454	uint			lock_mode)
 455{
 456	int			attempts = 0;
 457	uint			i;
 458	int			j;
 459	bool			try_lock;
 460	struct xfs_log_item	*lp;
 461
 462	/*
 463	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
 464	 * support an arbitrary depth of locking here, but absolute limits on
 465	 * inodes depend on the type of locking and the limits placed by
 466	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
 467	 * the asserts.
 468	 */
 469	ASSERT(ips && inodes >= 2 && inodes <= 5);
 470	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
 471			    XFS_ILOCK_EXCL));
 472	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
 473			      XFS_ILOCK_SHARED)));
 474	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
 475		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
 476	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
 477		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
 478
 479	if (lock_mode & XFS_IOLOCK_EXCL) {
 480		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
 481	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
 482		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
 483
 484again:
 485	try_lock = false;
 
 486	i = 0;
 
 
 487	for (; i < inodes; i++) {
 488		ASSERT(ips[i]);
 489
 490		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
 491			continue;
 492
 493		/*
 494		 * If try_lock is not set yet, make sure all locked inodes are
 495		 * not in the AIL.  If any are, set try_lock to be used later.
 
 496		 */
 
 497		if (!try_lock) {
 498			for (j = (i - 1); j >= 0 && !try_lock; j--) {
 499				lp = &ips[j]->i_itemp->ili_item;
 500				if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
 501					try_lock = true;
 
 502			}
 503		}
 504
 505		/*
 506		 * If any of the previous locks we have locked is in the AIL,
 507		 * we must TRY to get the second and subsequent locks. If
 508		 * we can't get any, we must release all we have
 509		 * and try again.
 510		 */
 511		if (!try_lock) {
 512			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
 513			continue;
 514		}
 515
 516		/* try_lock means we have an inode locked that is in the AIL. */
 517		ASSERT(i != 0);
 518		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
 519			continue;
 520
 521		/*
 522		 * Unlock all previous guys and try again.  xfs_iunlock will try
 523		 * to push the tail if the inode is in the AIL.
 524		 */
 525		attempts++;
 526		for (j = i - 1; j >= 0; j--) {
 527			/*
 528			 * Check to see if we've already unlocked this one.  Not
 529			 * the first one going back, and the inode ptr is the
 530			 * same.
 531			 */
 532			if (j != (i - 1) && ips[j] == ips[j + 1])
 533				continue;
 534
 535			xfs_iunlock(ips[j], lock_mode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 536		}
 
 537
 538		if ((attempts % 5) == 0) {
 539			delay(1); /* Don't just spin the CPU */
 540		}
 541		goto again;
 
 
 
 542	}
 
 543}
 544
 545/*
 546 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
 547 * mmaplock must be double-locked separately since we use i_rwsem and
 548 * invalidate_lock for that. We now support taking one lock EXCL and the
 549 * other SHARED.
 550 */
 551void
 552xfs_lock_two_inodes(
 553	struct xfs_inode	*ip0,
 554	uint			ip0_mode,
 555	struct xfs_inode	*ip1,
 556	uint			ip1_mode)
 557{
 
 558	int			attempts = 0;
 559	struct xfs_log_item	*lp;
 560
 561	ASSERT(hweight32(ip0_mode) == 1);
 562	ASSERT(hweight32(ip1_mode) == 1);
 563	ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
 564	ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
 565	ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
 566	ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
 567	ASSERT(ip0->i_ino != ip1->i_ino);
 568
 569	if (ip0->i_ino > ip1->i_ino) {
 570		swap(ip0, ip1);
 571		swap(ip0_mode, ip1_mode);
 
 572	}
 573
 574 again:
 575	xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
 576
 577	/*
 578	 * If the first lock we have locked is in the AIL, we must TRY to get
 579	 * the second lock. If we can't get it, we must release the first one
 580	 * and try again.
 581	 */
 582	lp = &ip0->i_itemp->ili_item;
 583	if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
 584		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
 585			xfs_iunlock(ip0, ip0_mode);
 586			if ((++attempts % 5) == 0)
 587				delay(1); /* Don't just spin the CPU */
 588			goto again;
 589		}
 590	} else {
 591		xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
 592	}
 593}
 594
 595uint
 596xfs_ip2xflags(
 
 597	struct xfs_inode	*ip)
 598{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 599	uint			flags = 0;
 600
 601	if (ip->i_diflags & XFS_DIFLAG_ANY) {
 602		if (ip->i_diflags & XFS_DIFLAG_REALTIME)
 603			flags |= FS_XFLAG_REALTIME;
 604		if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
 605			flags |= FS_XFLAG_PREALLOC;
 606		if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
 607			flags |= FS_XFLAG_IMMUTABLE;
 608		if (ip->i_diflags & XFS_DIFLAG_APPEND)
 609			flags |= FS_XFLAG_APPEND;
 610		if (ip->i_diflags & XFS_DIFLAG_SYNC)
 611			flags |= FS_XFLAG_SYNC;
 612		if (ip->i_diflags & XFS_DIFLAG_NOATIME)
 613			flags |= FS_XFLAG_NOATIME;
 614		if (ip->i_diflags & XFS_DIFLAG_NODUMP)
 615			flags |= FS_XFLAG_NODUMP;
 616		if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
 617			flags |= FS_XFLAG_RTINHERIT;
 618		if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
 619			flags |= FS_XFLAG_PROJINHERIT;
 620		if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
 621			flags |= FS_XFLAG_NOSYMLINKS;
 622		if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
 623			flags |= FS_XFLAG_EXTSIZE;
 624		if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
 625			flags |= FS_XFLAG_EXTSZINHERIT;
 626		if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
 627			flags |= FS_XFLAG_NODEFRAG;
 628		if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
 629			flags |= FS_XFLAG_FILESTREAM;
 630	}
 631
 632	if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
 633		if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
 634			flags |= FS_XFLAG_DAX;
 635		if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
 636			flags |= FS_XFLAG_COWEXTSIZE;
 637	}
 638
 639	if (xfs_inode_has_attr_fork(ip))
 640		flags |= FS_XFLAG_HASATTR;
 641	return flags;
 642}
 643
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 644/*
 645 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
 646 * is allowed, otherwise it has to be an exact match. If a CI match is found,
 647 * ci_name->name will point to a the actual name (caller must free) or
 648 * will be set to NULL if an exact match is found.
 649 */
 650int
 651xfs_lookup(
 652	struct xfs_inode	*dp,
 653	const struct xfs_name	*name,
 654	struct xfs_inode	**ipp,
 655	struct xfs_name		*ci_name)
 656{
 657	xfs_ino_t		inum;
 658	int			error;
 
 659
 660	trace_xfs_lookup(dp, name);
 661
 662	if (xfs_is_shutdown(dp->i_mount))
 663		return -EIO;
 664
 
 665	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
 
 
 666	if (error)
 667		goto out_unlock;
 668
 669	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
 670	if (error)
 671		goto out_free_name;
 672
 673	return 0;
 674
 675out_free_name:
 676	if (ci_name)
 677		kmem_free(ci_name->name);
 678out_unlock:
 679	*ipp = NULL;
 680	return error;
 681}
 682
 683/* Propagate di_flags from a parent inode to a child inode. */
 684static void
 685xfs_inode_inherit_flags(
 686	struct xfs_inode	*ip,
 687	const struct xfs_inode	*pip)
 688{
 689	unsigned int		di_flags = 0;
 690	xfs_failaddr_t		failaddr;
 691	umode_t			mode = VFS_I(ip)->i_mode;
 692
 693	if (S_ISDIR(mode)) {
 694		if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
 695			di_flags |= XFS_DIFLAG_RTINHERIT;
 696		if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
 697			di_flags |= XFS_DIFLAG_EXTSZINHERIT;
 698			ip->i_extsize = pip->i_extsize;
 699		}
 700		if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
 701			di_flags |= XFS_DIFLAG_PROJINHERIT;
 702	} else if (S_ISREG(mode)) {
 703		if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
 704		    xfs_has_realtime(ip->i_mount))
 705			di_flags |= XFS_DIFLAG_REALTIME;
 706		if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
 707			di_flags |= XFS_DIFLAG_EXTSIZE;
 708			ip->i_extsize = pip->i_extsize;
 709		}
 710	}
 711	if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
 712	    xfs_inherit_noatime)
 713		di_flags |= XFS_DIFLAG_NOATIME;
 714	if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
 715	    xfs_inherit_nodump)
 716		di_flags |= XFS_DIFLAG_NODUMP;
 717	if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
 718	    xfs_inherit_sync)
 719		di_flags |= XFS_DIFLAG_SYNC;
 720	if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
 721	    xfs_inherit_nosymlinks)
 722		di_flags |= XFS_DIFLAG_NOSYMLINKS;
 723	if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
 724	    xfs_inherit_nodefrag)
 725		di_flags |= XFS_DIFLAG_NODEFRAG;
 726	if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
 727		di_flags |= XFS_DIFLAG_FILESTREAM;
 728
 729	ip->i_diflags |= di_flags;
 730
 731	/*
 732	 * Inode verifiers on older kernels only check that the extent size
 733	 * hint is an integer multiple of the rt extent size on realtime files.
 734	 * They did not check the hint alignment on a directory with both
 735	 * rtinherit and extszinherit flags set.  If the misaligned hint is
 736	 * propagated from a directory into a new realtime file, new file
 737	 * allocations will fail due to math errors in the rt allocator and/or
 738	 * trip the verifiers.  Validate the hint settings in the new file so
 739	 * that we don't let broken hints propagate.
 740	 */
 741	failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
 742			VFS_I(ip)->i_mode, ip->i_diflags);
 743	if (failaddr) {
 744		ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
 745				   XFS_DIFLAG_EXTSZINHERIT);
 746		ip->i_extsize = 0;
 747	}
 748}
 749
 750/* Propagate di_flags2 from a parent inode to a child inode. */
 751static void
 752xfs_inode_inherit_flags2(
 753	struct xfs_inode	*ip,
 754	const struct xfs_inode	*pip)
 755{
 756	xfs_failaddr_t		failaddr;
 757
 758	if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
 759		ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
 760		ip->i_cowextsize = pip->i_cowextsize;
 761	}
 762	if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
 763		ip->i_diflags2 |= XFS_DIFLAG2_DAX;
 764
 765	/* Don't let invalid cowextsize hints propagate. */
 766	failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
 767			VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
 768	if (failaddr) {
 769		ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
 770		ip->i_cowextsize = 0;
 771	}
 772}
 773
 774/*
 775 * Initialise a newly allocated inode and return the in-core inode to the
 776 * caller locked exclusively.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 777 */
 778int
 779xfs_init_new_inode(
 780	struct user_namespace	*mnt_userns,
 781	struct xfs_trans	*tp,
 782	struct xfs_inode	*pip,
 783	xfs_ino_t		ino,
 784	umode_t			mode,
 785	xfs_nlink_t		nlink,
 786	dev_t			rdev,
 787	prid_t			prid,
 788	bool			init_xattrs,
 789	struct xfs_inode	**ipp)
 790{
 791	struct inode		*dir = pip ? VFS_I(pip) : NULL;
 792	struct xfs_mount	*mp = tp->t_mountp;
 793	struct xfs_inode	*ip;
 794	unsigned int		flags;
 795	int			error;
 796	struct timespec64	tv;
 797	struct inode		*inode;
 798
 799	/*
 800	 * Protect against obviously corrupt allocation btree records. Later
 801	 * xfs_iget checks will catch re-allocation of other active in-memory
 802	 * and on-disk inodes. If we don't catch reallocating the parent inode
 803	 * here we will deadlock in xfs_iget() so we have to do these checks
 804	 * first.
 805	 */
 806	if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
 807		xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
 808		return -EFSCORRUPTED;
 
 
 
 
 809	}
 
 810
 811	/*
 812	 * Get the in-core inode with the lock held exclusively to prevent
 813	 * others from looking at until we're done.
 
 814	 */
 815	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
 
 816	if (error)
 817		return error;
 818
 819	ASSERT(ip != NULL);
 820	inode = VFS_I(ip);
 821	set_nlink(inode, nlink);
 822	inode->i_rdev = rdev;
 823	ip->i_projid = prid;
 824
 825	if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
 826		inode_fsuid_set(inode, mnt_userns);
 827		inode->i_gid = dir->i_gid;
 828		inode->i_mode = mode;
 829	} else {
 830		inode_init_owner(mnt_userns, inode, dir, mode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 831	}
 832
 833	/*
 834	 * If the group ID of the new file does not match the effective group
 835	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
 836	 * (and only if the irix_sgid_inherit compatibility variable is set).
 837	 */
 838	if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
 839	    !vfsgid_in_group_p(i_gid_into_vfsgid(mnt_userns, inode)))
 840		inode->i_mode &= ~S_ISGID;
 841
 842	ip->i_disk_size = 0;
 843	ip->i_df.if_nextents = 0;
 844	ASSERT(ip->i_nblocks == 0);
 845
 846	tv = current_time(inode);
 847	inode->i_mtime = tv;
 848	inode->i_atime = tv;
 849	inode->i_ctime = tv;
 850
 851	ip->i_extsize = 0;
 852	ip->i_diflags = 0;
 853
 854	if (xfs_has_v3inodes(mp)) {
 855		inode_set_iversion(inode, 1);
 856		ip->i_cowextsize = 0;
 857		ip->i_crtime = tv;
 
 
 
 
 
 
 
 
 
 
 
 
 
 858	}
 859
 
 860	flags = XFS_ILOG_CORE;
 861	switch (mode & S_IFMT) {
 862	case S_IFIFO:
 863	case S_IFCHR:
 864	case S_IFBLK:
 865	case S_IFSOCK:
 866		ip->i_df.if_format = XFS_DINODE_FMT_DEV;
 
 
 867		flags |= XFS_ILOG_DEV;
 868		break;
 869	case S_IFREG:
 
 
 
 
 
 
 
 870	case S_IFDIR:
 871		if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
 872			xfs_inode_inherit_flags(ip, pip);
 873		if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
 874			xfs_inode_inherit_flags2(ip, pip);
 875		fallthrough;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 876	case S_IFLNK:
 877		ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
 878		ip->i_df.if_bytes = 0;
 879		ip->i_df.if_u1.if_root = NULL;
 
 880		break;
 881	default:
 882		ASSERT(0);
 883	}
 884
 885	/*
 886	 * If we need to create attributes immediately after allocating the
 887	 * inode, initialise an empty attribute fork right now. We use the
 888	 * default fork offset for attributes here as we don't know exactly what
 889	 * size or how many attributes we might be adding. We can do this
 890	 * safely here because we know the data fork is completely empty and
 891	 * this saves us from needing to run a separate transaction to set the
 892	 * fork offset in the immediate future.
 893	 */
 894	if (init_xattrs && xfs_has_attr(mp)) {
 895		ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
 896		xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
 897	}
 898
 899	/*
 900	 * Log the new values stuffed into the inode.
 901	 */
 902	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 903	xfs_trans_log_inode(tp, ip, flags);
 904
 905	/* now that we have an i_mode we can setup the inode structure */
 906	xfs_setup_inode(ip);
 907
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 908	*ipp = ip;
 
 
 909	return 0;
 910}
 911
 912/*
 913 * Decrement the link count on an inode & log the change.  If this causes the
 914 * link count to go to zero, move the inode to AGI unlinked list so that it can
 915 * be freed when the last active reference goes away via xfs_inactive().
 916 */
 917static int			/* error */
 918xfs_droplink(
 919	xfs_trans_t *tp,
 920	xfs_inode_t *ip)
 921{
 
 
 922	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
 923
 
 
 924	drop_nlink(VFS_I(ip));
 925	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 926
 927	if (VFS_I(ip)->i_nlink)
 928		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 929
 930	return xfs_iunlink(tp, ip);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 931}
 932
 933/*
 934 * Increment the link count on an inode & log the change.
 935 */
 936static void
 937xfs_bumplink(
 938	xfs_trans_t *tp,
 939	xfs_inode_t *ip)
 940{
 941	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
 942
 
 
 943	inc_nlink(VFS_I(ip));
 
 
 
 
 
 
 
 
 
 
 
 
 
 944	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 
 945}
 946
 947int
 948xfs_create(
 949	struct user_namespace	*mnt_userns,
 950	xfs_inode_t		*dp,
 951	struct xfs_name		*name,
 952	umode_t			mode,
 953	dev_t			rdev,
 954	bool			init_xattrs,
 955	xfs_inode_t		**ipp)
 956{
 957	int			is_dir = S_ISDIR(mode);
 958	struct xfs_mount	*mp = dp->i_mount;
 959	struct xfs_inode	*ip = NULL;
 960	struct xfs_trans	*tp = NULL;
 961	int			error;
 
 
 962	bool                    unlock_dp_on_error = false;
 
 
 963	prid_t			prid;
 964	struct xfs_dquot	*udqp = NULL;
 965	struct xfs_dquot	*gdqp = NULL;
 966	struct xfs_dquot	*pdqp = NULL;
 967	struct xfs_trans_res	*tres;
 968	uint			resblks;
 969	xfs_ino_t		ino;
 970
 971	trace_xfs_create(dp, name);
 972
 973	if (xfs_is_shutdown(mp))
 974		return -EIO;
 975
 976	prid = xfs_get_initial_prid(dp);
 977
 978	/*
 979	 * Make sure that we have allocated dquot(s) on disk.
 980	 */
 981	error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns, &init_user_ns),
 982			mapped_fsgid(mnt_userns, &init_user_ns), prid,
 983			XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
 984			&udqp, &gdqp, &pdqp);
 985	if (error)
 986		return error;
 987
 988	if (is_dir) {
 
 989		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
 990		tres = &M_RES(mp)->tr_mkdir;
 
 
 991	} else {
 992		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
 993		tres = &M_RES(mp)->tr_create;
 
 
 994	}
 995
 
 
 996	/*
 997	 * Initially assume that the file does not exist and
 998	 * reserve the resources for that case.  If that is not
 999	 * the case we'll drop the one we have and get a more
1000	 * appropriate transaction later.
1001	 */
1002	error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1003			&tp);
1004	if (error == -ENOSPC) {
1005		/* flush outstanding delalloc blocks and retry */
1006		xfs_flush_inodes(mp);
1007		error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1008				resblks, &tp);
 
 
 
 
 
 
 
 
1009	}
1010	if (error)
1011		goto out_release_dquots;
1012
1013	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1014	unlock_dp_on_error = true;
1015
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1016	/*
1017	 * A newly created regular or special file just has one directory
1018	 * entry pointing to them, but a directory also the "." entry
1019	 * pointing to itself.
1020	 */
1021	error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1022	if (!error)
1023		error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1024				is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1025	if (error)
1026		goto out_trans_cancel;
 
1027
1028	/*
1029	 * Now we join the directory inode to the transaction.  We do not do it
1030	 * earlier because xfs_dialloc might commit the previous transaction
1031	 * (and release all the locks).  An error from here on will result in
1032	 * the transaction cancel unlocking dp so don't do it explicitly in the
1033	 * error path.
1034	 */
1035	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1036	unlock_dp_on_error = false;
1037
1038	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1039					resblks - XFS_IALLOC_SPACE_RES(mp));
 
1040	if (error) {
1041		ASSERT(error != -ENOSPC);
1042		goto out_trans_cancel;
1043	}
1044	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1045	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1046
1047	if (is_dir) {
1048		error = xfs_dir_init(tp, ip, dp);
1049		if (error)
1050			goto out_trans_cancel;
1051
1052		xfs_bumplink(tp, dp);
 
 
1053	}
1054
1055	/*
1056	 * If this is a synchronous mount, make sure that the
1057	 * create transaction goes to disk before returning to
1058	 * the user.
1059	 */
1060	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1061		xfs_trans_set_sync(tp);
1062
1063	/*
1064	 * Attach the dquot(s) to the inodes and modify them incore.
1065	 * These ids of the inode couldn't have changed since the new
1066	 * inode has been locked ever since it was created.
1067	 */
1068	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1069
1070	error = xfs_trans_commit(tp);
 
 
 
 
1071	if (error)
1072		goto out_release_inode;
1073
1074	xfs_qm_dqrele(udqp);
1075	xfs_qm_dqrele(gdqp);
1076	xfs_qm_dqrele(pdqp);
1077
1078	*ipp = ip;
1079	return 0;
1080
 
 
 
 
1081 out_trans_cancel:
1082	xfs_trans_cancel(tp);
1083 out_release_inode:
1084	/*
1085	 * Wait until after the current transaction is aborted to finish the
1086	 * setup of the inode and release the inode.  This prevents recursive
1087	 * transactions and deadlocks from xfs_inactive.
1088	 */
1089	if (ip) {
1090		xfs_finish_inode_setup(ip);
1091		xfs_irele(ip);
1092	}
1093 out_release_dquots:
1094	xfs_qm_dqrele(udqp);
1095	xfs_qm_dqrele(gdqp);
1096	xfs_qm_dqrele(pdqp);
1097
1098	if (unlock_dp_on_error)
1099		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1100	return error;
1101}
1102
1103int
1104xfs_create_tmpfile(
1105	struct user_namespace	*mnt_userns,
1106	struct xfs_inode	*dp,
 
1107	umode_t			mode,
1108	struct xfs_inode	**ipp)
1109{
1110	struct xfs_mount	*mp = dp->i_mount;
1111	struct xfs_inode	*ip = NULL;
1112	struct xfs_trans	*tp = NULL;
1113	int			error;
 
1114	prid_t                  prid;
1115	struct xfs_dquot	*udqp = NULL;
1116	struct xfs_dquot	*gdqp = NULL;
1117	struct xfs_dquot	*pdqp = NULL;
1118	struct xfs_trans_res	*tres;
1119	uint			resblks;
1120	xfs_ino_t		ino;
1121
1122	if (xfs_is_shutdown(mp))
1123		return -EIO;
1124
1125	prid = xfs_get_initial_prid(dp);
1126
1127	/*
1128	 * Make sure that we have allocated dquot(s) on disk.
1129	 */
1130	error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns, &init_user_ns),
1131			mapped_fsgid(mnt_userns, &init_user_ns), prid,
1132			XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1133			&udqp, &gdqp, &pdqp);
1134	if (error)
1135		return error;
1136
1137	resblks = XFS_IALLOC_SPACE_RES(mp);
1138	tres = &M_RES(mp)->tr_create_tmpfile;
1139
1140	error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1141			&tp);
1142	if (error)
1143		goto out_release_dquots;
 
 
 
 
 
 
 
1144
1145	error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1146	if (!error)
1147		error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1148				0, 0, prid, false, &ip);
1149	if (error)
1150		goto out_trans_cancel;
1151
1152	if (xfs_has_wsync(mp))
 
 
 
 
 
 
 
 
1153		xfs_trans_set_sync(tp);
1154
1155	/*
1156	 * Attach the dquot(s) to the inodes and modify them incore.
1157	 * These ids of the inode couldn't have changed since the new
1158	 * inode has been locked ever since it was created.
1159	 */
1160	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1161
 
1162	error = xfs_iunlink(tp, ip);
1163	if (error)
1164		goto out_trans_cancel;
1165
1166	error = xfs_trans_commit(tp);
1167	if (error)
1168		goto out_release_inode;
1169
1170	xfs_qm_dqrele(udqp);
1171	xfs_qm_dqrele(gdqp);
1172	xfs_qm_dqrele(pdqp);
1173
1174	*ipp = ip;
1175	return 0;
1176
 
 
1177 out_trans_cancel:
1178	xfs_trans_cancel(tp);
1179 out_release_inode:
1180	/*
1181	 * Wait until after the current transaction is aborted to finish the
1182	 * setup of the inode and release the inode.  This prevents recursive
1183	 * transactions and deadlocks from xfs_inactive.
1184	 */
1185	if (ip) {
1186		xfs_finish_inode_setup(ip);
1187		xfs_irele(ip);
1188	}
1189 out_release_dquots:
1190	xfs_qm_dqrele(udqp);
1191	xfs_qm_dqrele(gdqp);
1192	xfs_qm_dqrele(pdqp);
1193
1194	return error;
1195}
1196
1197int
1198xfs_link(
1199	xfs_inode_t		*tdp,
1200	xfs_inode_t		*sip,
1201	struct xfs_name		*target_name)
1202{
1203	xfs_mount_t		*mp = tdp->i_mount;
1204	xfs_trans_t		*tp;
1205	int			error, nospace_error = 0;
 
 
 
 
1206	int			resblks;
1207
1208	trace_xfs_link(tdp, target_name);
1209
1210	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1211
1212	if (xfs_is_shutdown(mp))
1213		return -EIO;
1214
1215	error = xfs_qm_dqattach(sip);
1216	if (error)
1217		goto std_return;
1218
1219	error = xfs_qm_dqattach(tdp);
1220	if (error)
1221		goto std_return;
1222
 
 
1223	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1224	error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
1225			&tp, &nospace_error);
1226	if (error)
1227		goto std_return;
 
 
 
 
 
 
 
 
 
 
1228
1229	/*
1230	 * If we are using project inheritance, we only allow hard link
1231	 * creation in our tree when the project IDs are the same; else
1232	 * the tree quota mechanism could be circumvented.
1233	 */
1234	if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1235		     tdp->i_projid != sip->i_projid)) {
1236		error = -EXDEV;
1237		goto error_return;
1238	}
1239
1240	if (!resblks) {
1241		error = xfs_dir_canenter(tp, tdp, target_name);
1242		if (error)
1243			goto error_return;
1244	}
1245
1246	/*
1247	 * Handle initial link state of O_TMPFILE inode
1248	 */
1249	if (VFS_I(sip)->i_nlink == 0) {
1250		struct xfs_perag	*pag;
1251
1252		pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1253		error = xfs_iunlink_remove(tp, pag, sip);
1254		xfs_perag_put(pag);
1255		if (error)
1256			goto error_return;
1257	}
1258
1259	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1260				   resblks);
1261	if (error)
1262		goto error_return;
1263	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1264	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1265
1266	xfs_bumplink(tp, sip);
 
 
1267
1268	/*
1269	 * If this is a synchronous mount, make sure that the
1270	 * link transaction goes to disk before returning to
1271	 * the user.
1272	 */
1273	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1274		xfs_trans_set_sync(tp);
 
 
 
 
 
 
 
1275
1276	return xfs_trans_commit(tp);
1277
 
 
1278 error_return:
1279	xfs_trans_cancel(tp);
1280 std_return:
1281	if (error == -ENOSPC && nospace_error)
1282		error = nospace_error;
1283	return error;
1284}
1285
1286/* Clear the reflink flag and the cowblocks tag if possible. */
1287static void
1288xfs_itruncate_clear_reflink_flags(
1289	struct xfs_inode	*ip)
1290{
1291	struct xfs_ifork	*dfork;
1292	struct xfs_ifork	*cfork;
1293
1294	if (!xfs_is_reflink_inode(ip))
1295		return;
1296	dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1297	cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
1298	if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1299		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1300	if (cfork->if_bytes == 0)
1301		xfs_inode_clear_cowblocks_tag(ip);
1302}
1303
1304/*
1305 * Free up the underlying blocks past new_size.  The new size must be smaller
1306 * than the current size.  This routine can be used both for the attribute and
1307 * data fork, and does not modify the inode size, which is left to the caller.
1308 *
1309 * The transaction passed to this routine must have made a permanent log
1310 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1311 * given transaction and start new ones, so make sure everything involved in
1312 * the transaction is tidy before calling here.  Some transaction will be
1313 * returned to the caller to be committed.  The incoming transaction must
1314 * already include the inode, and both inode locks must be held exclusively.
1315 * The inode must also be "held" within the transaction.  On return the inode
1316 * will be "held" within the returned transaction.  This routine does NOT
1317 * require any disk space to be reserved for it within the transaction.
1318 *
1319 * If we get an error, we must return with the inode locked and linked into the
1320 * current transaction. This keeps things simple for the higher level code,
1321 * because it always knows that the inode is locked and held in the transaction
1322 * that returns to it whether errors occur or not.  We don't mark the inode
1323 * dirty on error so that transactions can be easily aborted if possible.
1324 */
1325int
1326xfs_itruncate_extents_flags(
1327	struct xfs_trans	**tpp,
1328	struct xfs_inode	*ip,
1329	int			whichfork,
1330	xfs_fsize_t		new_size,
1331	int			flags)
1332{
1333	struct xfs_mount	*mp = ip->i_mount;
1334	struct xfs_trans	*tp = *tpp;
 
 
 
1335	xfs_fileoff_t		first_unmap_block;
 
1336	xfs_filblks_t		unmap_len;
 
1337	int			error = 0;
 
1338
1339	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1340	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1341	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1342	ASSERT(new_size <= XFS_ISIZE(ip));
1343	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1344	ASSERT(ip->i_itemp != NULL);
1345	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1346	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1347
1348	trace_xfs_itruncate_extents_start(ip, new_size);
1349
1350	flags |= xfs_bmapi_aflag(whichfork);
1351
1352	/*
1353	 * Since it is possible for space to become allocated beyond
1354	 * the end of the file (in a crash where the space is allocated
1355	 * but the inode size is not yet updated), simply remove any
1356	 * blocks which show up between the new EOF and the maximum
1357	 * possible file size.
1358	 *
1359	 * We have to free all the blocks to the bmbt maximum offset, even if
1360	 * the page cache can't scale that far.
1361	 */
1362	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1363	if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1364		WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1365		return 0;
1366	}
1367
1368	unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1369	while (unmap_len > 0) {
1370		ASSERT(tp->t_firstblock == NULLFSBLOCK);
1371		error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1372				flags, XFS_ITRUNC_MAX_EXTENTS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1373		if (error)
1374			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1375
1376		/* free the just unmapped extents */
1377		error = xfs_defer_finish(&tp);
1378		if (error)
1379			goto out;
1380	}
1381
1382	if (whichfork == XFS_DATA_FORK) {
1383		/* Remove all pending CoW reservations. */
1384		error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1385				first_unmap_block, XFS_MAX_FILEOFF, true);
 
 
1386		if (error)
1387			goto out;
1388
1389		xfs_itruncate_clear_reflink_flags(ip);
1390	}
1391
1392	/*
1393	 * Always re-log the inode so that our permanent transaction can keep
1394	 * on rolling it forward in the log.
1395	 */
1396	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1397
1398	trace_xfs_itruncate_extents_end(ip, new_size);
1399
1400out:
1401	*tpp = tp;
1402	return error;
 
 
 
 
 
 
 
 
1403}
1404
1405int
1406xfs_release(
1407	xfs_inode_t	*ip)
1408{
1409	xfs_mount_t	*mp = ip->i_mount;
1410	int		error = 0;
1411
1412	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1413		return 0;
1414
1415	/* If this is a read-only mount, don't do this (would generate I/O) */
1416	if (xfs_is_readonly(mp))
1417		return 0;
1418
1419	if (!xfs_is_shutdown(mp)) {
1420		int truncated;
1421
1422		/*
 
 
 
 
 
 
 
 
 
 
1423		 * If we previously truncated this file and removed old data
1424		 * in the process, we want to initiate "early" writeout on
1425		 * the last close.  This is an attempt to combat the notorious
1426		 * NULL files problem which is particularly noticeable from a
1427		 * truncate down, buffered (re-)write (delalloc), followed by
1428		 * a crash.  What we are effectively doing here is
1429		 * significantly reducing the time window where we'd otherwise
1430		 * be exposed to that problem.
1431		 */
1432		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1433		if (truncated) {
1434			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1435			if (ip->i_delayed_blks > 0) {
1436				error = filemap_flush(VFS_I(ip)->i_mapping);
1437				if (error)
1438					return error;
1439			}
1440		}
1441	}
1442
1443	if (VFS_I(ip)->i_nlink == 0)
1444		return 0;
1445
1446	/*
1447	 * If we can't get the iolock just skip truncating the blocks past EOF
1448	 * because we could deadlock with the mmap_lock otherwise. We'll get
1449	 * another chance to drop them once the last reference to the inode is
1450	 * dropped, so we'll never leak blocks permanently.
1451	 */
1452	if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1453		return 0;
1454
1455	if (xfs_can_free_eofblocks(ip, false)) {
 
1456		/*
1457		 * Check if the inode is being opened, written and closed
1458		 * frequently and we have delayed allocation blocks outstanding
1459		 * (e.g. streaming writes from the NFS server), truncating the
1460		 * blocks past EOF will cause fragmentation to occur.
 
1461		 *
1462		 * In this case don't do the truncation, but we have to be
1463		 * careful how we detect this case. Blocks beyond EOF show up as
1464		 * i_delayed_blks even when the inode is clean, so we need to
1465		 * truncate them away first before checking for a dirty release.
1466		 * Hence on the first dirty close we will still remove the
1467		 * speculative allocation, but after that we will leave it in
1468		 * place.
 
 
 
 
 
 
1469		 */
1470		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1471			goto out_unlock;
1472
1473		error = xfs_free_eofblocks(ip);
1474		if (error)
1475			goto out_unlock;
1476
1477		/* delalloc blocks after truncation means it really is dirty */
1478		if (ip->i_delayed_blks)
1479			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1480	}
1481
1482out_unlock:
1483	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1484	return error;
1485}
1486
1487/*
1488 * xfs_inactive_truncate
1489 *
1490 * Called to perform a truncate when an inode becomes unlinked.
1491 */
1492STATIC int
1493xfs_inactive_truncate(
1494	struct xfs_inode *ip)
1495{
1496	struct xfs_mount	*mp = ip->i_mount;
1497	struct xfs_trans	*tp;
1498	int			error;
1499
1500	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
 
1501	if (error) {
1502		ASSERT(xfs_is_shutdown(mp));
 
1503		return error;
1504	}
 
1505	xfs_ilock(ip, XFS_ILOCK_EXCL);
1506	xfs_trans_ijoin(tp, ip, 0);
1507
1508	/*
1509	 * Log the inode size first to prevent stale data exposure in the event
1510	 * of a system crash before the truncate completes. See the related
1511	 * comment in xfs_vn_setattr_size() for details.
1512	 */
1513	ip->i_disk_size = 0;
1514	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1515
1516	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1517	if (error)
1518		goto error_trans_cancel;
1519
1520	ASSERT(ip->i_df.if_nextents == 0);
1521
1522	error = xfs_trans_commit(tp);
1523	if (error)
1524		goto error_unlock;
1525
1526	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1527	return 0;
1528
1529error_trans_cancel:
1530	xfs_trans_cancel(tp);
1531error_unlock:
1532	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1533	return error;
1534}
1535
1536/*
1537 * xfs_inactive_ifree()
1538 *
1539 * Perform the inode free when an inode is unlinked.
1540 */
1541STATIC int
1542xfs_inactive_ifree(
1543	struct xfs_inode *ip)
1544{
 
 
 
1545	struct xfs_mount	*mp = ip->i_mount;
1546	struct xfs_trans	*tp;
1547	int			error;
1548
1549	/*
1550	 * We try to use a per-AG reservation for any block needed by the finobt
1551	 * tree, but as the finobt feature predates the per-AG reservation
1552	 * support a degraded file system might not have enough space for the
1553	 * reservation at mount time.  In that case try to dip into the reserved
1554	 * pool and pray.
1555	 *
1556	 * Send a warning if the reservation does happen to fail, as the inode
1557	 * now remains allocated and sits on the unlinked list until the fs is
1558	 * repaired.
1559	 */
1560	if (unlikely(mp->m_finobt_nores)) {
1561		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1562				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1563				&tp);
1564	} else {
1565		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1566	}
1567	if (error) {
1568		if (error == -ENOSPC) {
1569			xfs_warn_ratelimited(mp,
1570			"Failed to remove inode(s) from unlinked list. "
1571			"Please free space, unmount and run xfs_repair.");
1572		} else {
1573			ASSERT(xfs_is_shutdown(mp));
1574		}
1575		return error;
1576	}
1577
1578	/*
1579	 * We do not hold the inode locked across the entire rolling transaction
1580	 * here. We only need to hold it for the first transaction that
1581	 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1582	 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1583	 * here breaks the relationship between cluster buffer invalidation and
1584	 * stale inode invalidation on cluster buffer item journal commit
1585	 * completion, and can result in leaving dirty stale inodes hanging
1586	 * around in memory.
1587	 *
1588	 * We have no need for serialising this inode operation against other
1589	 * operations - we freed the inode and hence reallocation is required
1590	 * and that will serialise on reallocating the space the deferops need
1591	 * to free. Hence we can unlock the inode on the first commit of
1592	 * the transaction rather than roll it right through the deferops. This
1593	 * avoids relogging the XFS_ISTALE inode.
1594	 *
1595	 * We check that xfs_ifree() hasn't grown an internal transaction roll
1596	 * by asserting that the inode is still locked when it returns.
1597	 */
1598	xfs_ilock(ip, XFS_ILOCK_EXCL);
1599	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1600
1601	error = xfs_ifree(tp, ip);
1602	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1603	if (error) {
1604		/*
1605		 * If we fail to free the inode, shut down.  The cancel
1606		 * might do that, we need to make sure.  Otherwise the
1607		 * inode might be lost for a long time or forever.
1608		 */
1609		if (!xfs_is_shutdown(mp)) {
1610			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1611				__func__, error);
1612			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1613		}
1614		xfs_trans_cancel(tp);
 
1615		return error;
1616	}
1617
1618	/*
1619	 * Credit the quota account(s). The inode is gone.
1620	 */
1621	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1622
1623	/*
1624	 * Just ignore errors at this point.  There is nothing we can do except
1625	 * to try to keep going. Make sure it's not a silent error.
 
1626	 */
1627	error = xfs_trans_commit(tp);
 
 
 
 
1628	if (error)
1629		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1630			__func__, error);
1631
 
1632	return 0;
1633}
1634
1635/*
1636 * Returns true if we need to update the on-disk metadata before we can free
1637 * the memory used by this inode.  Updates include freeing post-eof
1638 * preallocations; freeing COW staging extents; and marking the inode free in
1639 * the inobt if it is on the unlinked list.
1640 */
1641bool
1642xfs_inode_needs_inactive(
1643	struct xfs_inode	*ip)
1644{
1645	struct xfs_mount	*mp = ip->i_mount;
1646	struct xfs_ifork	*cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1647
1648	/*
1649	 * If the inode is already free, then there can be nothing
1650	 * to clean up here.
1651	 */
1652	if (VFS_I(ip)->i_mode == 0)
1653		return false;
1654
1655	/* If this is a read-only mount, don't do this (would generate I/O) */
1656	if (xfs_is_readonly(mp))
1657		return false;
1658
1659	/* If the log isn't running, push inodes straight to reclaim. */
1660	if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1661		return false;
1662
1663	/* Metadata inodes require explicit resource cleanup. */
1664	if (xfs_is_metadata_inode(ip))
1665		return false;
1666
1667	/* Want to clean out the cow blocks if there are any. */
1668	if (cow_ifp && cow_ifp->if_bytes > 0)
1669		return true;
1670
1671	/* Unlinked files must be freed. */
1672	if (VFS_I(ip)->i_nlink == 0)
1673		return true;
1674
1675	/*
1676	 * This file isn't being freed, so check if there are post-eof blocks
1677	 * to free.  @force is true because we are evicting an inode from the
1678	 * cache.  Post-eof blocks must be freed, lest we end up with broken
1679	 * free space accounting.
1680	 *
1681	 * Note: don't bother with iolock here since lockdep complains about
1682	 * acquiring it in reclaim context. We have the only reference to the
1683	 * inode at this point anyways.
1684	 */
1685	return xfs_can_free_eofblocks(ip, true);
1686}
1687
1688/*
1689 * xfs_inactive
1690 *
1691 * This is called when the vnode reference count for the vnode
1692 * goes to zero.  If the file has been unlinked, then it must
1693 * now be truncated.  Also, we clear all of the read-ahead state
1694 * kept for the inode here since the file is now closed.
1695 */
1696void
1697xfs_inactive(
1698	xfs_inode_t	*ip)
1699{
1700	struct xfs_mount	*mp;
1701	int			error;
1702	int			truncate = 0;
1703
1704	/*
1705	 * If the inode is already free, then there can be nothing
1706	 * to clean up here.
1707	 */
1708	if (VFS_I(ip)->i_mode == 0) {
 
1709		ASSERT(ip->i_df.if_broot_bytes == 0);
1710		goto out;
1711	}
1712
1713	mp = ip->i_mount;
1714	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1715
1716	/* If this is a read-only mount, don't do this (would generate I/O) */
1717	if (xfs_is_readonly(mp))
1718		goto out;
1719
1720	/* Metadata inodes require explicit resource cleanup. */
1721	if (xfs_is_metadata_inode(ip))
1722		goto out;
1723
1724	/* Try to clean out the cow blocks if there are any. */
1725	if (xfs_inode_has_cow_data(ip))
1726		xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1727
1728	if (VFS_I(ip)->i_nlink != 0) {
1729		/*
1730		 * force is true because we are evicting an inode from the
1731		 * cache. Post-eof blocks must be freed, lest we end up with
1732		 * broken free space accounting.
1733		 *
1734		 * Note: don't bother with iolock here since lockdep complains
1735		 * about acquiring it in reclaim context. We have the only
1736		 * reference to the inode at this point anyways.
1737		 */
1738		if (xfs_can_free_eofblocks(ip, true))
1739			xfs_free_eofblocks(ip);
1740
1741		goto out;
1742	}
1743
1744	if (S_ISREG(VFS_I(ip)->i_mode) &&
1745	    (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1746	     ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1747		truncate = 1;
1748
1749	error = xfs_qm_dqattach(ip);
1750	if (error)
1751		goto out;
1752
1753	if (S_ISLNK(VFS_I(ip)->i_mode))
1754		error = xfs_inactive_symlink(ip);
1755	else if (truncate)
1756		error = xfs_inactive_truncate(ip);
1757	if (error)
1758		goto out;
1759
1760	/*
1761	 * If there are attributes associated with the file then blow them away
1762	 * now.  The code calls a routine that recursively deconstructs the
1763	 * attribute fork. If also blows away the in-core attribute fork.
 
1764	 */
1765	if (xfs_inode_has_attr_fork(ip)) {
 
 
1766		error = xfs_attr_inactive(ip);
1767		if (error)
1768			goto out;
1769	}
1770
1771	ASSERT(ip->i_forkoff == 0);
 
 
 
1772
1773	/*
1774	 * Free the inode.
1775	 */
1776	xfs_inactive_ifree(ip);
 
 
1777
1778out:
1779	/*
1780	 * We're done making metadata updates for this inode, so we can release
1781	 * the attached dquots.
1782	 */
1783	xfs_qm_dqdetach(ip);
1784}
1785
1786/*
1787 * In-Core Unlinked List Lookups
1788 * =============================
1789 *
1790 * Every inode is supposed to be reachable from some other piece of metadata
1791 * with the exception of the root directory.  Inodes with a connection to a
1792 * file descriptor but not linked from anywhere in the on-disk directory tree
1793 * are collectively known as unlinked inodes, though the filesystem itself
1794 * maintains links to these inodes so that on-disk metadata are consistent.
1795 *
1796 * XFS implements a per-AG on-disk hash table of unlinked inodes.  The AGI
1797 * header contains a number of buckets that point to an inode, and each inode
1798 * record has a pointer to the next inode in the hash chain.  This
1799 * singly-linked list causes scaling problems in the iunlink remove function
1800 * because we must walk that list to find the inode that points to the inode
1801 * being removed from the unlinked hash bucket list.
1802 *
1803 * Hence we keep an in-memory double linked list to link each inode on an
1804 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
1805 * based lists would require having 64 list heads in the perag, one for each
1806 * list. This is expensive in terms of memory (think millions of AGs) and cache
1807 * misses on lookups. Instead, use the fact that inodes on the unlinked list
1808 * must be referenced at the VFS level to keep them on the list and hence we
1809 * have an existence guarantee for inodes on the unlinked list.
1810 *
1811 * Given we have an existence guarantee, we can use lockless inode cache lookups
1812 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
1813 * for the double linked unlinked list, and we don't need any extra locking to
1814 * keep the list safe as all manipulations are done under the AGI buffer lock.
1815 * Keeping the list up to date does not require memory allocation, just finding
1816 * the XFS inode and updating the next/prev unlinked list aginos.
1817 */
1818
1819/*
1820 * Find an inode on the unlinked list. This does not take references to the
1821 * inode as we have existence guarantees by holding the AGI buffer lock and that
1822 * only unlinked, referenced inodes can be on the unlinked inode list.  If we
1823 * don't find the inode in cache, then let the caller handle the situation.
1824 */
1825static struct xfs_inode *
1826xfs_iunlink_lookup(
1827	struct xfs_perag	*pag,
1828	xfs_agino_t		agino)
1829{
1830	struct xfs_inode	*ip;
1831
1832	rcu_read_lock();
1833	ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1834
1835	/*
1836	 * Inode not in memory or in RCU freeing limbo should not happen.
1837	 * Warn about this and let the caller handle the failure.
1838	 */
1839	if (WARN_ON_ONCE(!ip || !ip->i_ino)) {
1840		rcu_read_unlock();
1841		return NULL;
1842	}
1843	ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1844	rcu_read_unlock();
1845	return ip;
1846}
1847
1848/* Update the prev pointer of the next agino. */
1849static int
1850xfs_iunlink_update_backref(
1851	struct xfs_perag	*pag,
1852	xfs_agino_t		prev_agino,
1853	xfs_agino_t		next_agino)
1854{
1855	struct xfs_inode	*ip;
1856
1857	/* No update necessary if we are at the end of the list. */
1858	if (next_agino == NULLAGINO)
1859		return 0;
1860
1861	ip = xfs_iunlink_lookup(pag, next_agino);
1862	if (!ip)
1863		return -EFSCORRUPTED;
1864	ip->i_prev_unlinked = prev_agino;
1865	return 0;
1866}
1867
1868/*
1869 * Point the AGI unlinked bucket at an inode and log the results.  The caller
1870 * is responsible for validating the old value.
1871 */
1872STATIC int
1873xfs_iunlink_update_bucket(
1874	struct xfs_trans	*tp,
1875	struct xfs_perag	*pag,
1876	struct xfs_buf		*agibp,
1877	unsigned int		bucket_index,
1878	xfs_agino_t		new_agino)
1879{
1880	struct xfs_agi		*agi = agibp->b_addr;
1881	xfs_agino_t		old_value;
1882	int			offset;
1883
1884	ASSERT(xfs_verify_agino_or_null(pag, new_agino));
1885
1886	old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1887	trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
1888			old_value, new_agino);
1889
1890	/*
1891	 * We should never find the head of the list already set to the value
1892	 * passed in because either we're adding or removing ourselves from the
1893	 * head of the list.
1894	 */
1895	if (old_value == new_agino) {
1896		xfs_buf_mark_corrupt(agibp);
1897		return -EFSCORRUPTED;
1898	}
1899
1900	agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
1901	offset = offsetof(struct xfs_agi, agi_unlinked) +
1902			(sizeof(xfs_agino_t) * bucket_index);
1903	xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
1904	return 0;
1905}
1906
1907static int
1908xfs_iunlink_insert_inode(
1909	struct xfs_trans	*tp,
1910	struct xfs_perag	*pag,
1911	struct xfs_buf		*agibp,
1912	struct xfs_inode	*ip)
1913{
1914	struct xfs_mount	*mp = tp->t_mountp;
1915	struct xfs_agi		*agi = agibp->b_addr;
1916	xfs_agino_t		next_agino;
1917	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1918	short			bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1919	int			error;
1920
1921	/*
1922	 * Get the index into the agi hash table for the list this inode will
1923	 * go on.  Make sure the pointer isn't garbage and that this inode
1924	 * isn't already on the list.
1925	 */
1926	next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1927	if (next_agino == agino ||
1928	    !xfs_verify_agino_or_null(pag, next_agino)) {
1929		xfs_buf_mark_corrupt(agibp);
1930		return -EFSCORRUPTED;
1931	}
1932
1933	/*
1934	 * Update the prev pointer in the next inode to point back to this
1935	 * inode.
1936	 */
1937	error = xfs_iunlink_update_backref(pag, agino, next_agino);
1938	if (error)
1939		return error;
 
 
1940
1941	if (next_agino != NULLAGINO) {
1942		/*
1943		 * There is already another inode in the bucket, so point this
1944		 * inode to the current head of the list.
 
 
1945		 */
1946		error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
 
1947		if (error)
1948			return error;
1949		ip->i_next_unlinked = next_agino;
1950	}
1951
1952	/* Point the head of the list to point to this inode. */
1953	return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
1954}
1955
1956/*
1957 * This is called when the inode's link count has gone to 0 or we are creating
1958 * a tmpfile via O_TMPFILE.  The inode @ip must have nlink == 0.
1959 *
1960 * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1961 * list when the inode is freed.
1962 */
1963STATIC int
1964xfs_iunlink(
1965	struct xfs_trans	*tp,
1966	struct xfs_inode	*ip)
1967{
1968	struct xfs_mount	*mp = tp->t_mountp;
1969	struct xfs_perag	*pag;
1970	struct xfs_buf		*agibp;
1971	int			error;
1972
1973	ASSERT(VFS_I(ip)->i_nlink == 0);
1974	ASSERT(VFS_I(ip)->i_mode != 0);
1975	trace_xfs_iunlink(ip);
1976
1977	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1978
1979	/* Get the agi buffer first.  It ensures lock ordering on the list. */
1980	error = xfs_read_agi(pag, tp, &agibp);
1981	if (error)
1982		goto out;
1983
1984	error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
1985out:
1986	xfs_perag_put(pag);
1987	return error;
1988}
1989
1990static int
1991xfs_iunlink_remove_inode(
1992	struct xfs_trans	*tp,
1993	struct xfs_perag	*pag,
1994	struct xfs_buf		*agibp,
1995	struct xfs_inode	*ip)
1996{
1997	struct xfs_mount	*mp = tp->t_mountp;
1998	struct xfs_agi		*agi = agibp->b_addr;
1999	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2000	xfs_agino_t		head_agino;
2001	short			bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2002	int			error;
2003
2004	trace_xfs_iunlink_remove(ip);
 
2005
2006	/*
2007	 * Get the index into the agi hash table for the list this inode will
2008	 * go on.  Make sure the head pointer isn't garbage.
2009	 */
2010	head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2011	if (!xfs_verify_agino(pag, head_agino)) {
2012		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2013				agi, sizeof(*agi));
2014		return -EFSCORRUPTED;
2015	}
2016
2017	/*
2018	 * Set our inode's next_unlinked pointer to NULL and then return
2019	 * the old pointer value so that we can update whatever was previous
2020	 * to us in the list to point to whatever was next in the list.
2021	 */
2022	error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
2023	if (error)
2024		return error;
2025
2026	/*
2027	 * Update the prev pointer in the next inode to point back to previous
2028	 * inode in the chain.
2029	 */
2030	error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
2031			ip->i_next_unlinked);
2032	if (error)
2033		return error;
2034
2035	if (head_agino != agino) {
2036		struct xfs_inode	*prev_ip;
2037
2038		prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
2039		if (!prev_ip)
2040			return -EFSCORRUPTED;
2041
2042		error = xfs_iunlink_log_inode(tp, prev_ip, pag,
2043				ip->i_next_unlinked);
2044		prev_ip->i_next_unlinked = ip->i_next_unlinked;
2045	} else {
2046		/* Point the head of the list to the next unlinked inode. */
2047		error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2048				ip->i_next_unlinked);
2049	}
2050
2051	ip->i_next_unlinked = NULLAGINO;
2052	ip->i_prev_unlinked = NULLAGINO;
2053	return error;
2054}
2055
2056/*
2057 * Pull the on-disk inode from the AGI unlinked list.
2058 */
2059STATIC int
2060xfs_iunlink_remove(
2061	struct xfs_trans	*tp,
2062	struct xfs_perag	*pag,
2063	struct xfs_inode	*ip)
2064{
2065	struct xfs_buf		*agibp;
2066	int			error;
 
 
 
 
 
 
 
 
 
 
 
 
2067
2068	trace_xfs_iunlink_remove(ip);
 
2069
2070	/* Get the agi buffer first.  It ensures lock ordering on the list. */
2071	error = xfs_read_agi(pag, tp, &agibp);
 
 
 
2072	if (error)
2073		return error;
2074
2075	return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
2076}
2077
2078/*
2079 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2080 * mark it stale. We should only find clean inodes in this lookup that aren't
2081 * already stale.
2082 */
2083static void
2084xfs_ifree_mark_inode_stale(
2085	struct xfs_perag	*pag,
2086	struct xfs_inode	*free_ip,
2087	xfs_ino_t		inum)
2088{
2089	struct xfs_mount	*mp = pag->pag_mount;
2090	struct xfs_inode_log_item *iip;
2091	struct xfs_inode	*ip;
2092
2093retry:
2094	rcu_read_lock();
2095	ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2096
2097	/* Inode not in memory, nothing to do */
2098	if (!ip) {
2099		rcu_read_unlock();
2100		return;
2101	}
2102
2103	/*
2104	 * because this is an RCU protected lookup, we could find a recently
2105	 * freed or even reallocated inode during the lookup. We need to check
2106	 * under the i_flags_lock for a valid inode here. Skip it if it is not
2107	 * valid, the wrong inode or stale.
2108	 */
2109	spin_lock(&ip->i_flags_lock);
2110	if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2111		goto out_iflags_unlock;
2112
2113	/*
2114	 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2115	 * other inodes that we did not find in the list attached to the buffer
2116	 * and are not already marked stale. If we can't lock it, back off and
2117	 * retry.
2118	 */
2119	if (ip != free_ip) {
2120		if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2121			spin_unlock(&ip->i_flags_lock);
2122			rcu_read_unlock();
2123			delay(1);
2124			goto retry;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2125		}
2126	}
2127	ip->i_flags |= XFS_ISTALE;
2128
2129	/*
2130	 * If the inode is flushing, it is already attached to the buffer.  All
2131	 * we needed to do here is mark the inode stale so buffer IO completion
2132	 * will remove it from the AIL.
2133	 */
2134	iip = ip->i_itemp;
2135	if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2136		ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2137		ASSERT(iip->ili_last_fields);
2138		goto out_iunlock;
2139	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2140
2141	/*
2142	 * Inodes not attached to the buffer can be released immediately.
2143	 * Everything else has to go through xfs_iflush_abort() on journal
2144	 * commit as the flock synchronises removal of the inode from the
2145	 * cluster buffer against inode reclaim.
2146	 */
2147	if (!iip || list_empty(&iip->ili_item.li_bio_list))
2148		goto out_iunlock;
2149
2150	__xfs_iflags_set(ip, XFS_IFLUSHING);
2151	spin_unlock(&ip->i_flags_lock);
2152	rcu_read_unlock();
 
 
2153
2154	/* we have a dirty inode in memory that has not yet been flushed. */
2155	spin_lock(&iip->ili_lock);
2156	iip->ili_last_fields = iip->ili_fields;
2157	iip->ili_fields = 0;
2158	iip->ili_fsync_fields = 0;
2159	spin_unlock(&iip->ili_lock);
2160	ASSERT(iip->ili_last_fields);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2161
2162	if (ip != free_ip)
2163		xfs_iunlock(ip, XFS_ILOCK_EXCL);
2164	return;
2165
2166out_iunlock:
2167	if (ip != free_ip)
2168		xfs_iunlock(ip, XFS_ILOCK_EXCL);
2169out_iflags_unlock:
2170	spin_unlock(&ip->i_flags_lock);
2171	rcu_read_unlock();
2172}
2173
2174/*
2175 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2176 * inodes that are in memory - they all must be marked stale and attached to
2177 * the cluster buffer.
2178 */
2179static int
2180xfs_ifree_cluster(
2181	struct xfs_trans	*tp,
2182	struct xfs_perag	*pag,
2183	struct xfs_inode	*free_ip,
2184	struct xfs_icluster	*xic)
2185{
2186	struct xfs_mount	*mp = free_ip->i_mount;
2187	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
2188	struct xfs_buf		*bp;
2189	xfs_daddr_t		blkno;
2190	xfs_ino_t		inum = xic->first_ino;
2191	int			nbufs;
2192	int			i, j;
2193	int			ioffset;
2194	int			error;
2195
2196	nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
 
 
2197
2198	for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2199		/*
2200		 * The allocation bitmap tells us which inodes of the chunk were
2201		 * physically allocated. Skip the cluster if an inode falls into
2202		 * a sparse region.
2203		 */
2204		ioffset = inum - xic->first_ino;
2205		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2206			ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2207			continue;
2208		}
2209
 
2210		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2211					 XFS_INO_TO_AGBNO(mp, inum));
2212
2213		/*
2214		 * We obtain and lock the backing buffer first in the process
2215		 * here to ensure dirty inodes attached to the buffer remain in
2216		 * the flushing state while we mark them stale.
2217		 *
2218		 * If we scan the in-memory inodes first, then buffer IO can
2219		 * complete before we get a lock on it, and hence we may fail
2220		 * to mark all the active inodes on the buffer stale.
2221		 */
2222		error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2223				mp->m_bsize * igeo->blocks_per_cluster,
2224				XBF_UNMAPPED, &bp);
2225		if (error)
2226			return error;
 
2227
2228		/*
2229		 * This buffer may not have been correctly initialised as we
2230		 * didn't read it from disk. That's not important because we are
2231		 * only using to mark the buffer as stale in the log, and to
2232		 * attach stale cached inodes on it. That means it will never be
2233		 * dispatched for IO. If it is, we want to know about it, and we
2234		 * want it to fail. We can acheive this by adding a write
2235		 * verifier to the buffer.
2236		 */
2237		bp->b_ops = &xfs_inode_buf_ops;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2238
2239		/*
2240		 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2241		 * too. This requires lookups, and will skip inodes that we've
2242		 * already marked XFS_ISTALE.
 
 
 
 
 
2243		 */
2244		for (i = 0; i < igeo->inodes_per_cluster; i++)
2245			xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2246
2247		xfs_trans_stale_inode_buf(tp, bp);
2248		xfs_trans_binval(tp, bp);
2249	}
 
 
2250	return 0;
2251}
2252
2253/*
2254 * This is called to return an inode to the inode free list.  The inode should
2255 * already be truncated to 0 length and have no pages associated with it.  This
2256 * routine also assumes that the inode is already a part of the transaction.
2257 *
2258 * The on-disk copy of the inode will have been added to the list of unlinked
2259 * inodes in the AGI. We need to remove the inode from that list atomically with
2260 * respect to freeing it here.
 
2261 */
2262int
2263xfs_ifree(
2264	struct xfs_trans	*tp,
2265	struct xfs_inode	*ip)
 
2266{
2267	struct xfs_mount	*mp = ip->i_mount;
2268	struct xfs_perag	*pag;
2269	struct xfs_icluster	xic = { 0 };
2270	struct xfs_inode_log_item *iip = ip->i_itemp;
2271	int			error;
 
 
2272
2273	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2274	ASSERT(VFS_I(ip)->i_nlink == 0);
2275	ASSERT(ip->i_df.if_nextents == 0);
2276	ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2277	ASSERT(ip->i_nblocks == 0);
2278
2279	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2280
2281	/*
2282	 * Free the inode first so that we guarantee that the AGI lock is going
2283	 * to be taken before we remove the inode from the unlinked list. This
2284	 * makes the AGI lock -> unlinked list modification order the same as
2285	 * used in O_TMPFILE creation.
2286	 */
2287	error = xfs_difree(tp, pag, ip->i_ino, &xic);
2288	if (error)
2289		goto out;
2290
2291	error = xfs_iunlink_remove(tp, pag, ip);
2292	if (error)
2293		goto out;
2294
2295	/*
2296	 * Free any local-format data sitting around before we reset the
2297	 * data fork to extents format.  Note that the attr fork data has
2298	 * already been freed by xfs_attr_inactive.
2299	 */
2300	if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2301		kmem_free(ip->i_df.if_u1.if_data);
2302		ip->i_df.if_u1.if_data = NULL;
2303		ip->i_df.if_bytes = 0;
2304	}
2305
2306	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2307	ip->i_diflags = 0;
2308	ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2309	ip->i_forkoff = 0;		/* mark the attr fork not in use */
2310	ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2311	if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2312		xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2313
2314	/* Don't attempt to replay owner changes for a deleted inode */
2315	spin_lock(&iip->ili_lock);
2316	iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2317	spin_unlock(&iip->ili_lock);
2318
 
 
 
 
 
 
2319	/*
2320	 * Bump the generation count so no one will be confused
2321	 * by reincarnations of this inode.
2322	 */
2323	VFS_I(ip)->i_generation++;
2324	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2325
2326	if (xic.deleted)
2327		error = xfs_ifree_cluster(tp, pag, ip, &xic);
2328out:
2329	xfs_perag_put(pag);
2330	return error;
2331}
2332
2333/*
2334 * This is called to unpin an inode.  The caller must have the inode locked
2335 * in at least shared mode so that the buffer cannot be subsequently pinned
2336 * once someone is waiting for it to be unpinned.
2337 */
2338static void
2339xfs_iunpin(
2340	struct xfs_inode	*ip)
2341{
2342	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2343
2344	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2345
2346	/* Give the log a push to start the unpinning I/O */
2347	xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2348
2349}
2350
2351static void
2352__xfs_iunpin_wait(
2353	struct xfs_inode	*ip)
2354{
2355	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2356	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2357
2358	xfs_iunpin(ip);
2359
2360	do {
2361		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2362		if (xfs_ipincount(ip))
2363			io_schedule();
2364	} while (xfs_ipincount(ip));
2365	finish_wait(wq, &wait.wq_entry);
2366}
2367
2368void
2369xfs_iunpin_wait(
2370	struct xfs_inode	*ip)
2371{
2372	if (xfs_ipincount(ip))
2373		__xfs_iunpin_wait(ip);
2374}
2375
2376/*
2377 * Removing an inode from the namespace involves removing the directory entry
2378 * and dropping the link count on the inode. Removing the directory entry can
2379 * result in locking an AGF (directory blocks were freed) and removing a link
2380 * count can result in placing the inode on an unlinked list which results in
2381 * locking an AGI.
2382 *
2383 * The big problem here is that we have an ordering constraint on AGF and AGI
2384 * locking - inode allocation locks the AGI, then can allocate a new extent for
2385 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2386 * removes the inode from the unlinked list, requiring that we lock the AGI
2387 * first, and then freeing the inode can result in an inode chunk being freed
2388 * and hence freeing disk space requiring that we lock an AGF.
2389 *
2390 * Hence the ordering that is imposed by other parts of the code is AGI before
2391 * AGF. This means we cannot remove the directory entry before we drop the inode
2392 * reference count and put it on the unlinked list as this results in a lock
2393 * order of AGF then AGI, and this can deadlock against inode allocation and
2394 * freeing. Therefore we must drop the link counts before we remove the
2395 * directory entry.
2396 *
2397 * This is still safe from a transactional point of view - it is not until we
2398 * get to xfs_defer_finish() that we have the possibility of multiple
2399 * transactions in this operation. Hence as long as we remove the directory
2400 * entry and drop the link count in the first transaction of the remove
2401 * operation, there are no transactional constraints on the ordering here.
2402 */
2403int
2404xfs_remove(
2405	xfs_inode_t             *dp,
2406	struct xfs_name		*name,
2407	xfs_inode_t		*ip)
2408{
2409	xfs_mount_t		*mp = dp->i_mount;
2410	xfs_trans_t             *tp = NULL;
2411	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2412	int			dontcare;
2413	int                     error = 0;
 
 
 
 
 
2414	uint			resblks;
 
2415
2416	trace_xfs_remove(dp, name);
2417
2418	if (xfs_is_shutdown(mp))
2419		return -EIO;
2420
2421	error = xfs_qm_dqattach(dp);
2422	if (error)
2423		goto std_return;
2424
2425	error = xfs_qm_dqattach(ip);
2426	if (error)
2427		goto std_return;
2428
 
 
 
 
 
 
 
 
 
2429	/*
2430	 * We try to get the real space reservation first, allowing for
2431	 * directory btree deletion(s) implying possible bmap insert(s).  If we
2432	 * can't get the space reservation then we use 0 instead, and avoid the
2433	 * bmap btree insert(s) in the directory code by, if the bmap insert
2434	 * tries to happen, instead trimming the LAST block from the directory.
2435	 *
2436	 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
2437	 * the directory code can handle a reservationless update and we don't
2438	 * want to prevent a user from trying to free space by deleting things.
2439	 */
2440	resblks = XFS_REMOVE_SPACE_RES(mp);
2441	error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
2442			&tp, &dontcare);
 
 
 
2443	if (error) {
2444		ASSERT(error != -ENOSPC);
2445		goto std_return;
 
2446	}
2447
 
 
 
 
 
2448	/*
2449	 * If we're removing a directory perform some additional validation.
2450	 */
 
2451	if (is_dir) {
2452		ASSERT(VFS_I(ip)->i_nlink >= 2);
2453		if (VFS_I(ip)->i_nlink != 2) {
2454			error = -ENOTEMPTY;
2455			goto out_trans_cancel;
2456		}
2457		if (!xfs_dir_isempty(ip)) {
2458			error = -ENOTEMPTY;
2459			goto out_trans_cancel;
2460		}
2461
2462		/* Drop the link from ip's "..".  */
2463		error = xfs_droplink(tp, dp);
2464		if (error)
2465			goto out_trans_cancel;
2466
2467		/* Drop the "." link from ip to self.  */
2468		error = xfs_droplink(tp, ip);
2469		if (error)
2470			goto out_trans_cancel;
2471
2472		/*
2473		 * Point the unlinked child directory's ".." entry to the root
2474		 * directory to eliminate back-references to inodes that may
2475		 * get freed before the child directory is closed.  If the fs
2476		 * gets shrunk, this can lead to dirent inode validation errors.
2477		 */
2478		if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2479			error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2480					tp->t_mountp->m_sb.sb_rootino, 0);
2481			if (error)
2482				goto out_trans_cancel;
2483		}
2484	} else {
2485		/*
2486		 * When removing a non-directory we need to log the parent
2487		 * inode here.  For a directory this is done implicitly
2488		 * by the xfs_droplink call for the ".." entry.
2489		 */
2490		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2491	}
2492	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2493
2494	/* Drop the link from dp to ip. */
2495	error = xfs_droplink(tp, ip);
2496	if (error)
2497		goto out_trans_cancel;
2498
2499	error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
 
 
 
 
 
2500	if (error) {
2501		ASSERT(error != -ENOENT);
2502		goto out_trans_cancel;
2503	}
2504
2505	/*
2506	 * If this is a synchronous mount, make sure that the
2507	 * remove transaction goes to disk before returning to
2508	 * the user.
2509	 */
2510	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2511		xfs_trans_set_sync(tp);
2512
2513	error = xfs_trans_commit(tp);
 
 
 
 
2514	if (error)
2515		goto std_return;
2516
2517	if (is_dir && xfs_inode_is_filestream(ip))
 
 
 
 
 
 
2518		xfs_filestream_deassociate(ip);
2519
2520	return 0;
2521
 
 
2522 out_trans_cancel:
2523	xfs_trans_cancel(tp);
2524 std_return:
2525	return error;
2526}
2527
2528/*
2529 * Enter all inodes for a rename transaction into a sorted array.
2530 */
2531#define __XFS_SORT_INODES	5
2532STATIC void
2533xfs_sort_for_rename(
2534	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2535	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2536	struct xfs_inode	*ip1,	/* in: inode of old entry */
2537	struct xfs_inode	*ip2,	/* in: inode of new entry */
2538	struct xfs_inode	*wip,	/* in: whiteout inode */
2539	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2540	int			*num_inodes)  /* in/out: inodes in array */
2541{
 
2542	int			i, j;
2543
2544	ASSERT(*num_inodes == __XFS_SORT_INODES);
2545	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2546
2547	/*
2548	 * i_tab contains a list of pointers to inodes.  We initialize
2549	 * the table here & we'll sort it.  We will then use it to
2550	 * order the acquisition of the inode locks.
2551	 *
2552	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2553	 */
2554	i = 0;
2555	i_tab[i++] = dp1;
2556	i_tab[i++] = dp2;
2557	i_tab[i++] = ip1;
2558	if (ip2)
2559		i_tab[i++] = ip2;
2560	if (wip)
2561		i_tab[i++] = wip;
2562	*num_inodes = i;
 
2563
2564	/*
2565	 * Sort the elements via bubble sort.  (Remember, there are at
2566	 * most 5 elements to sort, so this is adequate.)
2567	 */
2568	for (i = 0; i < *num_inodes; i++) {
2569		for (j = 1; j < *num_inodes; j++) {
2570			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2571				struct xfs_inode *temp = i_tab[j];
2572				i_tab[j] = i_tab[j-1];
2573				i_tab[j-1] = temp;
2574			}
2575		}
2576	}
2577}
2578
2579static int
2580xfs_finish_rename(
2581	struct xfs_trans	*tp)
2582{
2583	/*
2584	 * If this is a synchronous mount, make sure that the rename transaction
2585	 * goes to disk before returning to the user.
2586	 */
2587	if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2588		xfs_trans_set_sync(tp);
2589
2590	return xfs_trans_commit(tp);
2591}
2592
2593/*
2594 * xfs_cross_rename()
2595 *
2596 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2597 */
2598STATIC int
2599xfs_cross_rename(
2600	struct xfs_trans	*tp,
2601	struct xfs_inode	*dp1,
2602	struct xfs_name		*name1,
2603	struct xfs_inode	*ip1,
2604	struct xfs_inode	*dp2,
2605	struct xfs_name		*name2,
2606	struct xfs_inode	*ip2,
2607	int			spaceres)
2608{
2609	int		error = 0;
2610	int		ip1_flags = 0;
2611	int		ip2_flags = 0;
2612	int		dp2_flags = 0;
2613
2614	/* Swap inode number for dirent in first parent */
2615	error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2616	if (error)
2617		goto out_trans_abort;
2618
2619	/* Swap inode number for dirent in second parent */
2620	error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2621	if (error)
2622		goto out_trans_abort;
2623
2624	/*
2625	 * If we're renaming one or more directories across different parents,
2626	 * update the respective ".." entries (and link counts) to match the new
2627	 * parents.
2628	 */
2629	if (dp1 != dp2) {
2630		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2631
2632		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2633			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2634						dp1->i_ino, spaceres);
2635			if (error)
2636				goto out_trans_abort;
2637
2638			/* transfer ip2 ".." reference to dp1 */
2639			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2640				error = xfs_droplink(tp, dp2);
2641				if (error)
2642					goto out_trans_abort;
2643				xfs_bumplink(tp, dp1);
2644			}
2645
2646			/*
2647			 * Although ip1 isn't changed here, userspace needs
2648			 * to be warned about the change, so that applications
2649			 * relying on it (like backup ones), will properly
2650			 * notify the change
2651			 */
2652			ip1_flags |= XFS_ICHGTIME_CHG;
2653			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2654		}
2655
2656		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2657			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2658						dp2->i_ino, spaceres);
2659			if (error)
2660				goto out_trans_abort;
2661
2662			/* transfer ip1 ".." reference to dp2 */
2663			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2664				error = xfs_droplink(tp, dp1);
2665				if (error)
2666					goto out_trans_abort;
2667				xfs_bumplink(tp, dp2);
2668			}
2669
2670			/*
2671			 * Although ip2 isn't changed here, userspace needs
2672			 * to be warned about the change, so that applications
2673			 * relying on it (like backup ones), will properly
2674			 * notify the change
2675			 */
2676			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2677			ip2_flags |= XFS_ICHGTIME_CHG;
2678		}
2679	}
2680
2681	if (ip1_flags) {
2682		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2683		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2684	}
2685	if (ip2_flags) {
2686		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2687		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2688	}
2689	if (dp2_flags) {
2690		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2691		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2692	}
2693	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2694	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2695	return xfs_finish_rename(tp);
2696
2697out_trans_abort:
2698	xfs_trans_cancel(tp);
2699	return error;
2700}
2701
2702/*
2703 * xfs_rename_alloc_whiteout()
2704 *
2705 * Return a referenced, unlinked, unlocked inode that can be used as a
2706 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2707 * crash between allocating the inode and linking it into the rename transaction
2708 * recovery will free the inode and we won't leak it.
2709 */
2710static int
2711xfs_rename_alloc_whiteout(
2712	struct user_namespace	*mnt_userns,
2713	struct xfs_name		*src_name,
2714	struct xfs_inode	*dp,
2715	struct xfs_inode	**wip)
2716{
2717	struct xfs_inode	*tmpfile;
2718	struct qstr		name;
2719	int			error;
2720
2721	error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
2722				   &tmpfile);
2723	if (error)
2724		return error;
2725
2726	name.name = src_name->name;
2727	name.len = src_name->len;
2728	error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2729	if (error) {
2730		xfs_finish_inode_setup(tmpfile);
2731		xfs_irele(tmpfile);
2732		return error;
2733	}
2734
2735	/*
2736	 * Prepare the tmpfile inode as if it were created through the VFS.
2737	 * Complete the inode setup and flag it as linkable.  nlink is already
2738	 * zero, so we can skip the drop_nlink.
2739	 */
2740	xfs_setup_iops(tmpfile);
2741	xfs_finish_inode_setup(tmpfile);
2742	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2743
2744	*wip = tmpfile;
2745	return 0;
2746}
2747
2748/*
2749 * xfs_rename
2750 */
2751int
2752xfs_rename(
2753	struct user_namespace	*mnt_userns,
2754	struct xfs_inode	*src_dp,
2755	struct xfs_name		*src_name,
2756	struct xfs_inode	*src_ip,
2757	struct xfs_inode	*target_dp,
2758	struct xfs_name		*target_name,
2759	struct xfs_inode	*target_ip,
2760	unsigned int		flags)
2761{
2762	struct xfs_mount	*mp = src_dp->i_mount;
2763	struct xfs_trans	*tp;
2764	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2765	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2766	int			i;
2767	int			num_inodes = __XFS_SORT_INODES;
2768	bool			new_parent = (src_dp != target_dp);
2769	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2770	int			spaceres;
2771	bool			retried = false;
2772	int			error, nospace_error = 0;
2773
2774	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2775
2776	if ((flags & RENAME_EXCHANGE) && !target_ip)
2777		return -EINVAL;
2778
2779	/*
2780	 * If we are doing a whiteout operation, allocate the whiteout inode
2781	 * we will be placing at the target and ensure the type is set
2782	 * appropriately.
2783	 */
2784	if (flags & RENAME_WHITEOUT) {
2785		error = xfs_rename_alloc_whiteout(mnt_userns, src_name,
2786						  target_dp, &wip);
2787		if (error)
2788			return error;
2789
2790		/* setup target dirent info as whiteout */
2791		src_name->type = XFS_DIR3_FT_CHRDEV;
2792	}
2793
2794	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2795				inodes, &num_inodes);
2796
2797retry:
2798	nospace_error = 0;
 
2799	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2800	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2801	if (error == -ENOSPC) {
2802		nospace_error = error;
2803		spaceres = 0;
2804		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2805				&tp);
 
 
 
2806	}
2807	if (error)
2808		goto out_release_wip;
2809
2810	/*
2811	 * Attach the dquots to the inodes
2812	 */
2813	error = xfs_qm_vop_rename_dqattach(inodes);
2814	if (error)
2815		goto out_trans_cancel;
 
 
2816
2817	/*
2818	 * Lock all the participating inodes. Depending upon whether
2819	 * the target_name exists in the target directory, and
2820	 * whether the target directory is the same as the source
2821	 * directory, we can lock from 2 to 5 inodes.
2822	 */
2823	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2824
2825	/*
2826	 * Join all the inodes to the transaction. From this point on,
2827	 * we can rely on either trans_commit or trans_cancel to unlock
2828	 * them.
2829	 */
2830	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2831	if (new_parent)
2832		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2833	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2834	if (target_ip)
2835		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2836	if (wip)
2837		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2838
2839	/*
2840	 * If we are using project inheritance, we only allow renames
2841	 * into our tree when the project IDs are the same; else the
2842	 * tree quota mechanism would be circumvented.
2843	 */
2844	if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2845		     target_dp->i_projid != src_ip->i_projid)) {
2846		error = -EXDEV;
2847		goto out_trans_cancel;
2848	}
2849
2850	/* RENAME_EXCHANGE is unique from here on. */
2851	if (flags & RENAME_EXCHANGE)
2852		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2853					target_dp, target_name, target_ip,
2854					spaceres);
2855
2856	/*
2857	 * Try to reserve quota to handle an expansion of the target directory.
2858	 * We'll allow the rename to continue in reservationless mode if we hit
2859	 * a space usage constraint.  If we trigger reservationless mode, save
2860	 * the errno if there isn't any free space in the target directory.
2861	 */
2862	if (spaceres != 0) {
2863		error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2864				0, false);
2865		if (error == -EDQUOT || error == -ENOSPC) {
2866			if (!retried) {
2867				xfs_trans_cancel(tp);
2868				xfs_blockgc_free_quota(target_dp, 0);
2869				retried = true;
2870				goto retry;
2871			}
2872
2873			nospace_error = error;
2874			spaceres = 0;
2875			error = 0;
2876		}
2877		if (error)
2878			goto out_trans_cancel;
2879	}
2880
2881	/*
2882	 * Check for expected errors before we dirty the transaction
2883	 * so we can return an error without a transaction abort.
2884	 */
2885	if (target_ip == NULL) {
2886		/*
2887		 * If there's no space reservation, check the entry will
2888		 * fit before actually inserting it.
2889		 */
2890		if (!spaceres) {
2891			error = xfs_dir_canenter(tp, target_dp, target_name);
2892			if (error)
2893				goto out_trans_cancel;
2894		}
2895	} else {
2896		/*
2897		 * If target exists and it's a directory, check that whether
2898		 * it can be destroyed.
2899		 */
2900		if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
2901		    (!xfs_dir_isempty(target_ip) ||
2902		     (VFS_I(target_ip)->i_nlink > 2))) {
2903			error = -EEXIST;
2904			goto out_trans_cancel;
2905		}
2906	}
2907
2908	/*
2909	 * Lock the AGI buffers we need to handle bumping the nlink of the
2910	 * whiteout inode off the unlinked list and to handle dropping the
2911	 * nlink of the target inode.  Per locking order rules, do this in
2912	 * increasing AG order and before directory block allocation tries to
2913	 * grab AGFs because we grab AGIs before AGFs.
2914	 *
2915	 * The (vfs) caller must ensure that if src is a directory then
2916	 * target_ip is either null or an empty directory.
2917	 */
2918	for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
2919		if (inodes[i] == wip ||
2920		    (inodes[i] == target_ip &&
2921		     (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
2922			struct xfs_perag	*pag;
2923			struct xfs_buf		*bp;
2924
2925			pag = xfs_perag_get(mp,
2926					XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
2927			error = xfs_read_agi(pag, tp, &bp);
2928			xfs_perag_put(pag);
2929			if (error)
2930				goto out_trans_cancel;
2931		}
2932	}
2933
2934	/*
2935	 * Directory entry creation below may acquire the AGF. Remove
2936	 * the whiteout from the unlinked list first to preserve correct
2937	 * AGI/AGF locking order. This dirties the transaction so failures
2938	 * after this point will abort and log recovery will clean up the
2939	 * mess.
2940	 *
2941	 * For whiteouts, we need to bump the link count on the whiteout
2942	 * inode. After this point, we have a real link, clear the tmpfile
2943	 * state flag from the inode so it doesn't accidentally get misused
2944	 * in future.
2945	 */
2946	if (wip) {
2947		struct xfs_perag	*pag;
2948
2949		ASSERT(VFS_I(wip)->i_nlink == 0);
2950
2951		pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
2952		error = xfs_iunlink_remove(tp, pag, wip);
2953		xfs_perag_put(pag);
2954		if (error)
2955			goto out_trans_cancel;
2956
2957		xfs_bumplink(tp, wip);
2958		VFS_I(wip)->i_state &= ~I_LINKABLE;
2959	}
2960
2961	/*
2962	 * Set up the target.
2963	 */
2964	if (target_ip == NULL) {
2965		/*
2966		 * If target does not exist and the rename crosses
2967		 * directories, adjust the target directory link count
2968		 * to account for the ".." reference from the new entry.
2969		 */
2970		error = xfs_dir_createname(tp, target_dp, target_name,
2971					   src_ip->i_ino, spaceres);
 
 
 
2972		if (error)
2973			goto out_trans_cancel;
2974
2975		xfs_trans_ichgtime(tp, target_dp,
2976					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2977
2978		if (new_parent && src_is_directory) {
2979			xfs_bumplink(tp, target_dp);
 
 
2980		}
2981	} else { /* target_ip != NULL */
2982		/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2983		 * Link the source inode under the target name.
2984		 * If the source inode is a directory and we are moving
2985		 * it across directories, its ".." entry will be
2986		 * inconsistent until we replace that down below.
2987		 *
2988		 * In case there is already an entry with the same
2989		 * name at the destination directory, remove it first.
2990		 */
2991		error = xfs_dir_replace(tp, target_dp, target_name,
2992					src_ip->i_ino, spaceres);
 
2993		if (error)
2994			goto out_trans_cancel;
2995
2996		xfs_trans_ichgtime(tp, target_dp,
2997					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2998
2999		/*
3000		 * Decrement the link count on the target since the target
3001		 * dir no longer points to it.
3002		 */
3003		error = xfs_droplink(tp, target_ip);
3004		if (error)
3005			goto out_trans_cancel;
3006
3007		if (src_is_directory) {
3008			/*
3009			 * Drop the link from the old "." entry.
3010			 */
3011			error = xfs_droplink(tp, target_ip);
3012			if (error)
3013				goto out_trans_cancel;
3014		}
3015	} /* target_ip != NULL */
3016
3017	/*
3018	 * Remove the source.
3019	 */
3020	if (new_parent && src_is_directory) {
3021		/*
3022		 * Rewrite the ".." entry to point to the new
3023		 * directory.
3024		 */
3025		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3026					target_dp->i_ino, spaceres);
3027		ASSERT(error != -EEXIST);
 
3028		if (error)
3029			goto out_trans_cancel;
3030	}
3031
3032	/*
3033	 * We always want to hit the ctime on the source inode.
3034	 *
3035	 * This isn't strictly required by the standards since the source
3036	 * inode isn't really being changed, but old unix file systems did
3037	 * it and some incremental backup programs won't work without it.
3038	 */
3039	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3040	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3041
3042	/*
3043	 * Adjust the link count on src_dp.  This is necessary when
3044	 * renaming a directory, either within one parent when
3045	 * the target existed, or across two parent directories.
3046	 */
3047	if (src_is_directory && (new_parent || target_ip != NULL)) {
3048
3049		/*
3050		 * Decrement link count on src_directory since the
3051		 * entry that's moved no longer points to it.
3052		 */
3053		error = xfs_droplink(tp, src_dp);
3054		if (error)
3055			goto out_trans_cancel;
3056	}
3057
3058	/*
3059	 * For whiteouts, we only need to update the source dirent with the
3060	 * inode number of the whiteout inode rather than removing it
3061	 * altogether.
3062	 */
3063	if (wip)
3064		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3065					spaceres);
3066	else
3067		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3068					   spaceres);
3069
3070	if (error)
3071		goto out_trans_cancel;
3072
3073	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3074	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3075	if (new_parent)
3076		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3077
3078	error = xfs_finish_rename(tp);
3079	if (wip)
3080		xfs_irele(wip);
3081	return error;
3082
3083out_trans_cancel:
3084	xfs_trans_cancel(tp);
3085out_release_wip:
3086	if (wip)
3087		xfs_irele(wip);
3088	if (error == -ENOSPC && nospace_error)
3089		error = nospace_error;
3090	return error;
3091}
3092
3093static int
3094xfs_iflush(
3095	struct xfs_inode	*ip,
3096	struct xfs_buf		*bp)
3097{
3098	struct xfs_inode_log_item *iip = ip->i_itemp;
3099	struct xfs_dinode	*dip;
3100	struct xfs_mount	*mp = ip->i_mount;
3101	int			error;
3102
3103	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3104	ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3105	ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3106	       ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3107	ASSERT(iip->ili_item.li_buf == bp);
3108
3109	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3110
3111	/*
3112	 * We don't flush the inode if any of the following checks fail, but we
3113	 * do still update the log item and attach to the backing buffer as if
3114	 * the flush happened. This is a formality to facilitate predictable
3115	 * error handling as the caller will shutdown and fail the buffer.
3116	 */
3117	error = -EFSCORRUPTED;
3118	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3119			       mp, XFS_ERRTAG_IFLUSH_1)) {
3120		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3121			"%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
3122			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3123		goto flush_out;
3124	}
3125	if (S_ISREG(VFS_I(ip)->i_mode)) {
3126		if (XFS_TEST_ERROR(
3127		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3128		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3129		    mp, XFS_ERRTAG_IFLUSH_3)) {
3130			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3131				"%s: Bad regular inode %llu, ptr "PTR_FMT,
3132				__func__, ip->i_ino, ip);
3133			goto flush_out;
3134		}
3135	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3136		if (XFS_TEST_ERROR(
3137		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3138		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3139		    ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3140		    mp, XFS_ERRTAG_IFLUSH_4)) {
3141			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3142				"%s: Bad directory inode %llu, ptr "PTR_FMT,
3143				__func__, ip->i_ino, ip);
3144			goto flush_out;
3145		}
3146	}
3147	if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
3148				ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3149		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3150			"%s: detected corrupt incore inode %llu, "
3151			"total extents = %llu nblocks = %lld, ptr "PTR_FMT,
3152			__func__, ip->i_ino,
3153			ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
3154			ip->i_nblocks, ip);
3155		goto flush_out;
3156	}
3157	if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3158				mp, XFS_ERRTAG_IFLUSH_6)) {
3159		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3160			"%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
3161			__func__, ip->i_ino, ip->i_forkoff, ip);
3162		goto flush_out;
3163	}
3164
3165	/*
3166	 * Inode item log recovery for v2 inodes are dependent on the flushiter
3167	 * count for correct sequencing.  We bump the flush iteration count so
3168	 * we can detect flushes which postdate a log record during recovery.
3169	 * This is redundant as we now log every change and hence this can't
3170	 * happen but we need to still do it to ensure backwards compatibility
3171	 * with old kernels that predate logging all inode changes.
3172	 */
3173	if (!xfs_has_v3inodes(mp))
3174		ip->i_flushiter++;
3175
3176	/*
3177	 * If there are inline format data / attr forks attached to this inode,
3178	 * make sure they are not corrupt.
3179	 */
3180	if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3181	    xfs_ifork_verify_local_data(ip))
3182		goto flush_out;
3183	if (xfs_inode_has_attr_fork(ip) &&
3184	    ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
3185	    xfs_ifork_verify_local_attr(ip))
3186		goto flush_out;
3187
3188	/*
3189	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3190	 * copy out the core of the inode, because if the inode is dirty at all
3191	 * the core must be.
3192	 */
3193	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3194
3195	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3196	if (!xfs_has_v3inodes(mp)) {
3197		if (ip->i_flushiter == DI_MAX_FLUSH)
3198			ip->i_flushiter = 0;
3199	}
3200
3201	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3202	if (xfs_inode_has_attr_fork(ip))
3203		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3204
3205	/*
3206	 * We've recorded everything logged in the inode, so we'd like to clear
3207	 * the ili_fields bits so we don't log and flush things unnecessarily.
3208	 * However, we can't stop logging all this information until the data
3209	 * we've copied into the disk buffer is written to disk.  If we did we
3210	 * might overwrite the copy of the inode in the log with all the data
3211	 * after re-logging only part of it, and in the face of a crash we
3212	 * wouldn't have all the data we need to recover.
3213	 *
3214	 * What we do is move the bits to the ili_last_fields field.  When
3215	 * logging the inode, these bits are moved back to the ili_fields field.
3216	 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3217	 * we know that the information those bits represent is permanently on
3218	 * disk.  As long as the flush completes before the inode is logged
3219	 * again, then both ili_fields and ili_last_fields will be cleared.
3220	 */
3221	error = 0;
3222flush_out:
3223	spin_lock(&iip->ili_lock);
3224	iip->ili_last_fields = iip->ili_fields;
3225	iip->ili_fields = 0;
3226	iip->ili_fsync_fields = 0;
3227	spin_unlock(&iip->ili_lock);
3228
3229	/*
3230	 * Store the current LSN of the inode so that we can tell whether the
3231	 * item has moved in the AIL from xfs_buf_inode_iodone().
3232	 */
3233	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3234				&iip->ili_item.li_lsn);
3235
3236	/* generate the checksum. */
3237	xfs_dinode_calc_crc(mp, dip);
 
 
 
 
3238	return error;
3239}
3240
3241/*
3242 * Non-blocking flush of dirty inode metadata into the backing buffer.
3243 *
3244 * The caller must have a reference to the inode and hold the cluster buffer
3245 * locked. The function will walk across all the inodes on the cluster buffer it
3246 * can find and lock without blocking, and flush them to the cluster buffer.
3247 *
3248 * On successful flushing of at least one inode, the caller must write out the
3249 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3250 * the caller needs to release the buffer. On failure, the filesystem will be
3251 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3252 * will be returned.
3253 */
3254int
3255xfs_iflush_cluster(
3256	struct xfs_buf		*bp)
 
3257{
3258	struct xfs_mount	*mp = bp->b_mount;
3259	struct xfs_log_item	*lip, *n;
3260	struct xfs_inode	*ip;
3261	struct xfs_inode_log_item *iip;
 
 
 
 
3262	int			clcount = 0;
3263	int			error = 0;
 
3264
3265	/*
3266	 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3267	 * will remove itself from the list.
3268	 */
3269	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3270		iip = (struct xfs_inode_log_item *)lip;
3271		ip = iip->ili_inode;
3272
3273		/*
3274		 * Quick and dirty check to avoid locks if possible.
3275		 */
3276		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3277			continue;
3278		if (xfs_ipincount(ip))
 
 
 
 
 
 
 
 
 
 
 
 
3279			continue;
3280
3281		/*
3282		 * The inode is still attached to the buffer, which means it is
3283		 * dirty but reclaim might try to grab it. Check carefully for
3284		 * that, and grab the ilock while still holding the i_flags_lock
3285		 * to guarantee reclaim will not be able to reclaim this inode
3286		 * once we drop the i_flags_lock.
3287		 */
3288		spin_lock(&ip->i_flags_lock);
3289		ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3290		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3291			spin_unlock(&ip->i_flags_lock);
3292			continue;
3293		}
 
3294
3295		/*
3296		 * ILOCK will pin the inode against reclaim and prevent
3297		 * concurrent transactions modifying the inode while we are
3298		 * flushing the inode. If we get the lock, set the flushing
3299		 * state before we drop the i_flags_lock.
3300		 */
3301		if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3302			spin_unlock(&ip->i_flags_lock);
3303			continue;
3304		}
3305		__xfs_iflags_set(ip, XFS_IFLUSHING);
3306		spin_unlock(&ip->i_flags_lock);
3307
3308		/*
3309		 * Abort flushing this inode if we are shut down because the
3310		 * inode may not currently be in the AIL. This can occur when
3311		 * log I/O failure unpins the inode without inserting into the
3312		 * AIL, leaving a dirty/unpinned inode attached to the buffer
3313		 * that otherwise looks like it should be flushed.
3314		 */
3315		if (xlog_is_shutdown(mp->m_log)) {
3316			xfs_iunpin_wait(ip);
3317			xfs_iflush_abort(ip);
3318			xfs_iunlock(ip, XFS_ILOCK_SHARED);
3319			error = -EIO;
3320			continue;
3321		}
3322
3323		/* don't block waiting on a log force to unpin dirty inodes */
3324		if (xfs_ipincount(ip)) {
3325			xfs_iflags_clear(ip, XFS_IFLUSHING);
3326			xfs_iunlock(ip, XFS_ILOCK_SHARED);
3327			continue;
3328		}
3329
3330		if (!xfs_inode_clean(ip))
3331			error = xfs_iflush(ip, bp);
3332		else
3333			xfs_iflags_clear(ip, XFS_IFLUSHING);
3334		xfs_iunlock(ip, XFS_ILOCK_SHARED);
3335		if (error)
3336			break;
3337		clcount++;
3338	}
3339
3340	if (error) {
3341		/*
3342		 * Shutdown first so we kill the log before we release this
3343		 * buffer. If it is an INODE_ALLOC buffer and pins the tail
3344		 * of the log, failing it before the _log_ is shut down can
3345		 * result in the log tail being moved forward in the journal
3346		 * on disk because log writes can still be taking place. Hence
3347		 * unpinning the tail will allow the ICREATE intent to be
3348		 * removed from the log an recovery will fail with uninitialised
3349		 * inode cluster buffers.
3350		 */
3351		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3352		bp->b_flags |= XBF_ASYNC;
3353		xfs_buf_ioend_fail(bp);
3354		return error;
 
 
3355	}
3356
3357	if (!clcount)
3358		return -EAGAIN;
 
 
3359
3360	XFS_STATS_INC(mp, xs_icluster_flushcnt);
3361	XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
 
 
 
3362	return 0;
3363
3364}
3365
3366/* Release an inode. */
3367void
3368xfs_irele(
3369	struct xfs_inode	*ip)
3370{
3371	trace_xfs_irele(ip, _RET_IP_);
3372	iput(VFS_I(ip));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3373}
3374
3375/*
3376 * Ensure all commited transactions touching the inode are written to the log.
 
 
 
 
 
 
3377 */
3378int
3379xfs_log_force_inode(
3380	struct xfs_inode	*ip)
 
3381{
3382	xfs_csn_t		seq = 0;
 
 
 
3383
3384	xfs_ilock(ip, XFS_ILOCK_SHARED);
3385	if (xfs_ipincount(ip))
3386		seq = ip->i_itemp->ili_commit_seq;
3387	xfs_iunlock(ip, XFS_ILOCK_SHARED);
3388
3389	if (!seq)
3390		return 0;
3391	return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3392}
3393
3394/*
3395 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3396 * abide vfs locking order (lowest pointer value goes first) and breaking the
3397 * layout leases before proceeding.  The loop is needed because we cannot call
3398 * the blocking break_layout() with the iolocks held, and therefore have to
3399 * back out both locks.
3400 */
3401static int
3402xfs_iolock_two_inodes_and_break_layout(
3403	struct inode		*src,
3404	struct inode		*dest)
3405{
3406	int			error;
3407
3408	if (src > dest)
3409		swap(src, dest);
3410
3411retry:
3412	/* Wait to break both inodes' layouts before we start locking. */
3413	error = break_layout(src, true);
3414	if (error)
3415		return error;
3416	if (src != dest) {
3417		error = break_layout(dest, true);
3418		if (error)
3419			return error;
 
 
3420	}
3421
3422	/* Lock one inode and make sure nobody got in and leased it. */
3423	inode_lock(src);
3424	error = break_layout(src, false);
3425	if (error) {
3426		inode_unlock(src);
3427		if (error == -EWOULDBLOCK)
3428			goto retry;
3429		return error;
 
 
 
3430	}
3431
3432	if (src == dest)
3433		return 0;
3434
3435	/* Lock the other inode and make sure nobody got in and leased it. */
3436	inode_lock_nested(dest, I_MUTEX_NONDIR2);
3437	error = break_layout(dest, false);
3438	if (error) {
3439		inode_unlock(src);
3440		inode_unlock(dest);
3441		if (error == -EWOULDBLOCK)
3442			goto retry;
3443		return error;
3444	}
3445
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3446	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
3447}
3448
3449static int
3450xfs_mmaplock_two_inodes_and_break_dax_layout(
3451	struct xfs_inode	*ip1,
3452	struct xfs_inode	*ip2)
3453{
3454	int			error;
3455	bool			retry;
3456	struct page		*page;
 
 
 
 
 
 
3457
3458	if (ip1->i_ino > ip2->i_ino)
3459		swap(ip1, ip2);
3460
3461again:
3462	retry = false;
3463	/* Lock the first inode */
3464	xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3465	error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
3466	if (error || retry) {
3467		xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3468		if (error == 0 && retry)
3469			goto again;
3470		return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3471	}
3472
3473	if (ip1 == ip2)
3474		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3475
3476	/* Nested lock the second inode */
3477	xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
3478	/*
3479	 * We cannot use xfs_break_dax_layouts() directly here because it may
3480	 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
3481	 * for this nested lock case.
3482	 */
3483	page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
3484	if (page && page_ref_count(page) != 1) {
3485		xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3486		xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3487		goto again;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3488	}
3489
3490	return 0;
3491}
 
 
3492
3493/*
3494 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3495 * mmap activity.
3496 */
3497int
3498xfs_ilock2_io_mmap(
3499	struct xfs_inode	*ip1,
3500	struct xfs_inode	*ip2)
3501{
3502	int			ret;
3503
3504	ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3505	if (ret)
3506		return ret;
3507
3508	if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3509		ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
3510		if (ret) {
3511			inode_unlock(VFS_I(ip2));
3512			if (ip1 != ip2)
3513				inode_unlock(VFS_I(ip1));
3514			return ret;
3515		}
3516	} else
3517		filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3518					    VFS_I(ip2)->i_mapping);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3519
 
 
3520	return 0;
3521}
3522
3523/* Unlock both inodes to allow IO and mmap activity. */
3524void
3525xfs_iunlock2_io_mmap(
3526	struct xfs_inode	*ip1,
3527	struct xfs_inode	*ip2)
3528{
3529	if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3530		xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3531		if (ip1 != ip2)
3532			xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3533	} else
3534		filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3535					      VFS_I(ip2)->i_mapping);
3536
3537	inode_unlock(VFS_I(ip2));
3538	if (ip1 != ip2)
3539		inode_unlock(VFS_I(ip1));
3540}
v3.15
 
   1/*
   2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include <linux/log2.h>
  19
  20#include "xfs.h"
  21#include "xfs_fs.h"
  22#include "xfs_shared.h"
  23#include "xfs_format.h"
  24#include "xfs_log_format.h"
  25#include "xfs_trans_resv.h"
  26#include "xfs_inum.h"
  27#include "xfs_sb.h"
  28#include "xfs_ag.h"
  29#include "xfs_mount.h"
 
  30#include "xfs_inode.h"
  31#include "xfs_da_format.h"
  32#include "xfs_da_btree.h"
  33#include "xfs_dir2.h"
  34#include "xfs_attr_sf.h"
  35#include "xfs_attr.h"
  36#include "xfs_trans_space.h"
  37#include "xfs_trans.h"
  38#include "xfs_buf_item.h"
  39#include "xfs_inode_item.h"
 
  40#include "xfs_ialloc.h"
  41#include "xfs_bmap.h"
  42#include "xfs_bmap_util.h"
 
  43#include "xfs_error.h"
  44#include "xfs_quota.h"
  45#include "xfs_filestream.h"
  46#include "xfs_cksum.h"
  47#include "xfs_trace.h"
  48#include "xfs_icache.h"
  49#include "xfs_symlink.h"
  50#include "xfs_trans_priv.h"
  51#include "xfs_log.h"
  52#include "xfs_bmap_btree.h"
 
 
 
  53
  54kmem_zone_t *xfs_inode_zone;
  55
  56/*
  57 * Used in xfs_itruncate_extents().  This is the maximum number of extents
  58 * freed from a file in a single transaction.
  59 */
  60#define	XFS_ITRUNC_MAX_EXTENTS	2
  61
  62STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
  63
  64STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
  65
  66/*
  67 * helper function to extract extent size hint from inode
  68 */
  69xfs_extlen_t
  70xfs_get_extsz_hint(
  71	struct xfs_inode	*ip)
  72{
  73	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
  74		return ip->i_d.di_extsize;
 
 
 
 
 
 
  75	if (XFS_IS_REALTIME_INODE(ip))
  76		return ip->i_mount->m_sb.sb_rextsize;
  77	return 0;
  78}
  79
  80/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  81 * These two are wrapper routines around the xfs_ilock() routine used to
  82 * centralize some grungy code.  They are used in places that wish to lock the
  83 * inode solely for reading the extents.  The reason these places can't just
  84 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
  85 * bringing in of the extents from disk for a file in b-tree format.  If the
  86 * inode is in b-tree format, then we need to lock the inode exclusively until
  87 * the extents are read in.  Locking it exclusively all the time would limit
  88 * our parallelism unnecessarily, though.  What we do instead is check to see
  89 * if the extents have been read in yet, and only lock the inode exclusively
  90 * if they have not.
  91 *
  92 * The functions return a value which should be given to the corresponding
  93 * xfs_iunlock() call.
  94 */
  95uint
  96xfs_ilock_data_map_shared(
  97	struct xfs_inode	*ip)
  98{
  99	uint			lock_mode = XFS_ILOCK_SHARED;
 100
 101	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 102	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
 103		lock_mode = XFS_ILOCK_EXCL;
 104	xfs_ilock(ip, lock_mode);
 105	return lock_mode;
 106}
 107
 108uint
 109xfs_ilock_attr_map_shared(
 110	struct xfs_inode	*ip)
 111{
 112	uint			lock_mode = XFS_ILOCK_SHARED;
 113
 114	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
 115	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
 116		lock_mode = XFS_ILOCK_EXCL;
 117	xfs_ilock(ip, lock_mode);
 118	return lock_mode;
 119}
 120
 121/*
 122 * The xfs inode contains 2 locks: a multi-reader lock called the
 123 * i_iolock and a multi-reader lock called the i_lock.  This routine
 124 * allows either or both of the locks to be obtained.
 125 *
 126 * The 2 locks should always be ordered so that the IO lock is
 127 * obtained first in order to prevent deadlock.
 128 *
 129 * ip -- the inode being locked
 130 * lock_flags -- this parameter indicates the inode's locks
 131 *       to be locked.  It can be:
 132 *		XFS_IOLOCK_SHARED,
 133 *		XFS_IOLOCK_EXCL,
 134 *		XFS_ILOCK_SHARED,
 135 *		XFS_ILOCK_EXCL,
 136 *		XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
 137 *		XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
 138 *		XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
 139 *		XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 140 */
 141void
 142xfs_ilock(
 143	xfs_inode_t		*ip,
 144	uint			lock_flags)
 145{
 146	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
 147
 148	/*
 149	 * You can't set both SHARED and EXCL for the same lock,
 150	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 151	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 152	 */
 153	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 154	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 155	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 156	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 157	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
 158
 159	if (lock_flags & XFS_IOLOCK_EXCL)
 160		mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
 161	else if (lock_flags & XFS_IOLOCK_SHARED)
 162		mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
 
 
 
 163
 164	if (lock_flags & XFS_ILOCK_EXCL)
 165		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 166	else if (lock_flags & XFS_ILOCK_SHARED)
 167		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 168}
 169
 170/*
 171 * This is just like xfs_ilock(), except that the caller
 172 * is guaranteed not to sleep.  It returns 1 if it gets
 173 * the requested locks and 0 otherwise.  If the IO lock is
 174 * obtained but the inode lock cannot be, then the IO lock
 175 * is dropped before returning.
 176 *
 177 * ip -- the inode being locked
 178 * lock_flags -- this parameter indicates the inode's locks to be
 179 *       to be locked.  See the comment for xfs_ilock() for a list
 180 *	 of valid values.
 181 */
 182int
 183xfs_ilock_nowait(
 184	xfs_inode_t		*ip,
 185	uint			lock_flags)
 186{
 187	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
 188
 189	/*
 190	 * You can't set both SHARED and EXCL for the same lock,
 191	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 192	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 193	 */
 194	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 195	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 196	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 197	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 198	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
 199
 200	if (lock_flags & XFS_IOLOCK_EXCL) {
 201		if (!mrtryupdate(&ip->i_iolock))
 202			goto out;
 203	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 204		if (!mrtryaccess(&ip->i_iolock))
 205			goto out;
 206	}
 
 
 
 
 
 
 
 
 
 207	if (lock_flags & XFS_ILOCK_EXCL) {
 208		if (!mrtryupdate(&ip->i_lock))
 209			goto out_undo_iolock;
 210	} else if (lock_flags & XFS_ILOCK_SHARED) {
 211		if (!mrtryaccess(&ip->i_lock))
 212			goto out_undo_iolock;
 213	}
 214	return 1;
 215
 216 out_undo_iolock:
 
 
 
 
 
 217	if (lock_flags & XFS_IOLOCK_EXCL)
 218		mrunlock_excl(&ip->i_iolock);
 219	else if (lock_flags & XFS_IOLOCK_SHARED)
 220		mrunlock_shared(&ip->i_iolock);
 221 out:
 222	return 0;
 223}
 224
 225/*
 226 * xfs_iunlock() is used to drop the inode locks acquired with
 227 * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
 228 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
 229 * that we know which locks to drop.
 230 *
 231 * ip -- the inode being unlocked
 232 * lock_flags -- this parameter indicates the inode's locks to be
 233 *       to be unlocked.  See the comment for xfs_ilock() for a list
 234 *	 of valid values for this parameter.
 235 *
 236 */
 237void
 238xfs_iunlock(
 239	xfs_inode_t		*ip,
 240	uint			lock_flags)
 241{
 242	/*
 243	 * You can't set both SHARED and EXCL for the same lock,
 244	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 245	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 246	 */
 247	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 248	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 249	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 250	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 251	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
 252	ASSERT(lock_flags != 0);
 253
 254	if (lock_flags & XFS_IOLOCK_EXCL)
 255		mrunlock_excl(&ip->i_iolock);
 256	else if (lock_flags & XFS_IOLOCK_SHARED)
 257		mrunlock_shared(&ip->i_iolock);
 
 
 
 
 
 258
 259	if (lock_flags & XFS_ILOCK_EXCL)
 260		mrunlock_excl(&ip->i_lock);
 261	else if (lock_flags & XFS_ILOCK_SHARED)
 262		mrunlock_shared(&ip->i_lock);
 263
 264	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
 265}
 266
 267/*
 268 * give up write locks.  the i/o lock cannot be held nested
 269 * if it is being demoted.
 270 */
 271void
 272xfs_ilock_demote(
 273	xfs_inode_t		*ip,
 274	uint			lock_flags)
 275{
 276	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
 277	ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
 
 278
 279	if (lock_flags & XFS_ILOCK_EXCL)
 280		mrdemote(&ip->i_lock);
 
 
 281	if (lock_flags & XFS_IOLOCK_EXCL)
 282		mrdemote(&ip->i_iolock);
 283
 284	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
 285}
 286
 287#if defined(DEBUG) || defined(XFS_WARN)
 288int
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 289xfs_isilocked(
 290	xfs_inode_t		*ip,
 291	uint			lock_flags)
 292{
 293	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
 294		if (!(lock_flags & XFS_ILOCK_SHARED))
 295			return !!ip->i_lock.mr_writer;
 296		return rwsem_is_locked(&ip->i_lock.mr_lock);
 297	}
 298
 299	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
 300		if (!(lock_flags & XFS_IOLOCK_SHARED))
 301			return !!ip->i_iolock.mr_writer;
 302		return rwsem_is_locked(&ip->i_iolock.mr_lock);
 
 
 
 
 303	}
 304
 305	ASSERT(0);
 306	return 0;
 307}
 308#endif
 309
 310#ifdef DEBUG
 311int xfs_locked_n;
 312int xfs_small_retries;
 313int xfs_middle_retries;
 314int xfs_lots_retries;
 315int xfs_lock_delays;
 
 
 
 
 
 
 
 
 
 316#endif
 317
 318/*
 319 * Bump the subclass so xfs_lock_inodes() acquires each lock with
 320 * a different value
 
 
 321 */
 322static inline int
 323xfs_lock_inumorder(int lock_mode, int subclass)
 
 
 324{
 325	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
 326		lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
 327	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
 328		lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
 329
 330	return lock_mode;
 
 
 
 
 
 331}
 332
 333/*
 334 * The following routine will lock n inodes in exclusive mode.
 335 * We assume the caller calls us with the inodes in i_ino order.
 
 
 
 
 
 
 336 *
 337 * We need to detect deadlock where an inode that we lock
 338 * is in the AIL and we start waiting for another inode that is locked
 339 * by a thread in a long running transaction (such as truncate). This can
 340 * result in deadlock since the long running trans might need to wait
 341 * for the inode we just locked in order to push the tail and free space
 342 * in the log.
 343 */
 344void
 345xfs_lock_inodes(
 346	xfs_inode_t	**ips,
 347	int		inodes,
 348	uint		lock_mode)
 349{
 350	int		attempts = 0, i, j, try_lock;
 351	xfs_log_item_t	*lp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 352
 353	ASSERT(ips && (inodes >= 2)); /* we need at least two */
 354
 355	try_lock = 0;
 356	i = 0;
 357
 358again:
 359	for (; i < inodes; i++) {
 360		ASSERT(ips[i]);
 361
 362		if (i && (ips[i] == ips[i-1]))	/* Already locked */
 363			continue;
 364
 365		/*
 366		 * If try_lock is not set yet, make sure all locked inodes
 367		 * are not in the AIL.
 368		 * If any are, set try_lock to be used later.
 369		 */
 370
 371		if (!try_lock) {
 372			for (j = (i - 1); j >= 0 && !try_lock; j--) {
 373				lp = (xfs_log_item_t *)ips[j]->i_itemp;
 374				if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
 375					try_lock++;
 376				}
 377			}
 378		}
 379
 380		/*
 381		 * If any of the previous locks we have locked is in the AIL,
 382		 * we must TRY to get the second and subsequent locks. If
 383		 * we can't get any, we must release all we have
 384		 * and try again.
 385		 */
 
 
 
 
 
 
 
 
 
 386
 387		if (try_lock) {
 388			/* try_lock must be 0 if i is 0. */
 
 
 
 
 389			/*
 390			 * try_lock means we have an inode locked
 391			 * that is in the AIL.
 
 392			 */
 393			ASSERT(i != 0);
 394			if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
 395				attempts++;
 396
 397				/*
 398				 * Unlock all previous guys and try again.
 399				 * xfs_iunlock will try to push the tail
 400				 * if the inode is in the AIL.
 401				 */
 402
 403				for(j = i - 1; j >= 0; j--) {
 404
 405					/*
 406					 * Check to see if we've already
 407					 * unlocked this one.
 408					 * Not the first one going back,
 409					 * and the inode ptr is the same.
 410					 */
 411					if ((j != (i - 1)) && ips[j] ==
 412								ips[j+1])
 413						continue;
 414
 415					xfs_iunlock(ips[j], lock_mode);
 416				}
 417
 418				if ((attempts % 5) == 0) {
 419					delay(1); /* Don't just spin the CPU */
 420#ifdef DEBUG
 421					xfs_lock_delays++;
 422#endif
 423				}
 424				i = 0;
 425				try_lock = 0;
 426				goto again;
 427			}
 428		} else {
 429			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
 430		}
 431	}
 432
 433#ifdef DEBUG
 434	if (attempts) {
 435		if (attempts < 5) xfs_small_retries++;
 436		else if (attempts < 100) xfs_middle_retries++;
 437		else xfs_lots_retries++;
 438	} else {
 439		xfs_locked_n++;
 440	}
 441#endif
 442}
 443
 444/*
 445 * xfs_lock_two_inodes() can only be used to lock one type of lock
 446 * at a time - the iolock or the ilock, but not both at once. If
 447 * we lock both at once, lockdep will report false positives saying
 448 * we have violated locking orders.
 449 */
 450void
 451xfs_lock_two_inodes(
 452	xfs_inode_t		*ip0,
 453	xfs_inode_t		*ip1,
 454	uint			lock_mode)
 
 455{
 456	xfs_inode_t		*temp;
 457	int			attempts = 0;
 458	xfs_log_item_t		*lp;
 459
 460	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
 461		ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
 
 
 
 
 462	ASSERT(ip0->i_ino != ip1->i_ino);
 463
 464	if (ip0->i_ino > ip1->i_ino) {
 465		temp = ip0;
 466		ip0 = ip1;
 467		ip1 = temp;
 468	}
 469
 470 again:
 471	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
 472
 473	/*
 474	 * If the first lock we have locked is in the AIL, we must TRY to get
 475	 * the second lock. If we can't get it, we must release the first one
 476	 * and try again.
 477	 */
 478	lp = (xfs_log_item_t *)ip0->i_itemp;
 479	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
 480		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
 481			xfs_iunlock(ip0, lock_mode);
 482			if ((++attempts % 5) == 0)
 483				delay(1); /* Don't just spin the CPU */
 484			goto again;
 485		}
 486	} else {
 487		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
 488	}
 489}
 490
 491
 492void
 493__xfs_iflock(
 494	struct xfs_inode	*ip)
 495{
 496	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
 497	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
 498
 499	do {
 500		prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
 501		if (xfs_isiflocked(ip))
 502			io_schedule();
 503	} while (!xfs_iflock_nowait(ip));
 504
 505	finish_wait(wq, &wait.wait);
 506}
 507
 508STATIC uint
 509_xfs_dic2xflags(
 510	__uint16_t		di_flags)
 511{
 512	uint			flags = 0;
 513
 514	if (di_flags & XFS_DIFLAG_ANY) {
 515		if (di_flags & XFS_DIFLAG_REALTIME)
 516			flags |= XFS_XFLAG_REALTIME;
 517		if (di_flags & XFS_DIFLAG_PREALLOC)
 518			flags |= XFS_XFLAG_PREALLOC;
 519		if (di_flags & XFS_DIFLAG_IMMUTABLE)
 520			flags |= XFS_XFLAG_IMMUTABLE;
 521		if (di_flags & XFS_DIFLAG_APPEND)
 522			flags |= XFS_XFLAG_APPEND;
 523		if (di_flags & XFS_DIFLAG_SYNC)
 524			flags |= XFS_XFLAG_SYNC;
 525		if (di_flags & XFS_DIFLAG_NOATIME)
 526			flags |= XFS_XFLAG_NOATIME;
 527		if (di_flags & XFS_DIFLAG_NODUMP)
 528			flags |= XFS_XFLAG_NODUMP;
 529		if (di_flags & XFS_DIFLAG_RTINHERIT)
 530			flags |= XFS_XFLAG_RTINHERIT;
 531		if (di_flags & XFS_DIFLAG_PROJINHERIT)
 532			flags |= XFS_XFLAG_PROJINHERIT;
 533		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
 534			flags |= XFS_XFLAG_NOSYMLINKS;
 535		if (di_flags & XFS_DIFLAG_EXTSIZE)
 536			flags |= XFS_XFLAG_EXTSIZE;
 537		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
 538			flags |= XFS_XFLAG_EXTSZINHERIT;
 539		if (di_flags & XFS_DIFLAG_NODEFRAG)
 540			flags |= XFS_XFLAG_NODEFRAG;
 541		if (di_flags & XFS_DIFLAG_FILESTREAM)
 542			flags |= XFS_XFLAG_FILESTREAM;
 
 
 
 
 
 
 
 543	}
 544
 
 
 545	return flags;
 546}
 547
 548uint
 549xfs_ip2xflags(
 550	xfs_inode_t		*ip)
 551{
 552	xfs_icdinode_t		*dic = &ip->i_d;
 553
 554	return _xfs_dic2xflags(dic->di_flags) |
 555				(XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
 556}
 557
 558uint
 559xfs_dic2xflags(
 560	xfs_dinode_t		*dip)
 561{
 562	return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
 563				(XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
 564}
 565
 566/*
 567 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
 568 * is allowed, otherwise it has to be an exact match. If a CI match is found,
 569 * ci_name->name will point to a the actual name (caller must free) or
 570 * will be set to NULL if an exact match is found.
 571 */
 572int
 573xfs_lookup(
 574	xfs_inode_t		*dp,
 575	struct xfs_name		*name,
 576	xfs_inode_t		**ipp,
 577	struct xfs_name		*ci_name)
 578{
 579	xfs_ino_t		inum;
 580	int			error;
 581	uint			lock_mode;
 582
 583	trace_xfs_lookup(dp, name);
 584
 585	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 586		return XFS_ERROR(EIO);
 587
 588	lock_mode = xfs_ilock_data_map_shared(dp);
 589	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
 590	xfs_iunlock(dp, lock_mode);
 591
 592	if (error)
 593		goto out;
 594
 595	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
 596	if (error)
 597		goto out_free_name;
 598
 599	return 0;
 600
 601out_free_name:
 602	if (ci_name)
 603		kmem_free(ci_name->name);
 604out:
 605	*ipp = NULL;
 606	return error;
 607}
 608
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 609/*
 610 * Allocate an inode on disk and return a copy of its in-core version.
 611 * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
 612 * appropriately within the inode.  The uid and gid for the inode are
 613 * set according to the contents of the given cred structure.
 614 *
 615 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
 616 * has a free inode available, call xfs_iget() to obtain the in-core
 617 * version of the allocated inode.  Finally, fill in the inode and
 618 * log its initial contents.  In this case, ialloc_context would be
 619 * set to NULL.
 620 *
 621 * If xfs_dialloc() does not have an available inode, it will replenish
 622 * its supply by doing an allocation. Since we can only do one
 623 * allocation within a transaction without deadlocks, we must commit
 624 * the current transaction before returning the inode itself.
 625 * In this case, therefore, we will set ialloc_context and return.
 626 * The caller should then commit the current transaction, start a new
 627 * transaction, and call xfs_ialloc() again to actually get the inode.
 628 *
 629 * To ensure that some other process does not grab the inode that
 630 * was allocated during the first call to xfs_ialloc(), this routine
 631 * also returns the [locked] bp pointing to the head of the freelist
 632 * as ialloc_context.  The caller should hold this buffer across
 633 * the commit and pass it back into this routine on the second call.
 634 *
 635 * If we are allocating quota inodes, we do not have a parent inode
 636 * to attach to or associate with (i.e. pip == NULL) because they
 637 * are not linked into the directory structure - they are attached
 638 * directly to the superblock - and so have no parent.
 639 */
 640int
 641xfs_ialloc(
 642	xfs_trans_t	*tp,
 643	xfs_inode_t	*pip,
 644	umode_t		mode,
 645	xfs_nlink_t	nlink,
 646	xfs_dev_t	rdev,
 647	prid_t		prid,
 648	int		okalloc,
 649	xfs_buf_t	**ialloc_context,
 650	xfs_inode_t	**ipp)
 651{
 652	struct xfs_mount *mp = tp->t_mountp;
 653	xfs_ino_t	ino;
 654	xfs_inode_t	*ip;
 655	uint		flags;
 656	int		error;
 657	timespec_t	tv;
 658	int		filestreams = 0;
 
 659
 660	/*
 661	 * Call the space management code to pick
 662	 * the on-disk inode to be allocated.
 
 
 
 663	 */
 664	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
 665			    ialloc_context, &ino);
 666	if (error)
 667		return error;
 668	if (*ialloc_context || ino == NULLFSINO) {
 669		*ipp = NULL;
 670		return 0;
 671	}
 672	ASSERT(*ialloc_context == NULL);
 673
 674	/*
 675	 * Get the in-core inode with the lock held exclusively.
 676	 * This is because we're setting fields here we need
 677	 * to prevent others from looking at until we're done.
 678	 */
 679	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
 680			 XFS_ILOCK_EXCL, &ip);
 681	if (error)
 682		return error;
 
 683	ASSERT(ip != NULL);
 684
 685	ip->i_d.di_mode = mode;
 686	ip->i_d.di_onlink = 0;
 687	ip->i_d.di_nlink = nlink;
 688	ASSERT(ip->i_d.di_nlink == nlink);
 689	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
 690	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
 691	xfs_set_projid(ip, prid);
 692	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
 693
 694	/*
 695	 * If the superblock version is up to where we support new format
 696	 * inodes and this is currently an old format inode, then change
 697	 * the inode version number now.  This way we only do the conversion
 698	 * here rather than here and in the flush/logging code.
 699	 */
 700	if (xfs_sb_version_hasnlink(&mp->m_sb) &&
 701	    ip->i_d.di_version == 1) {
 702		ip->i_d.di_version = 2;
 703		/*
 704		 * We've already zeroed the old link count, the projid field,
 705		 * and the pad field.
 706		 */
 707	}
 708
 709	/*
 710	 * Project ids won't be stored on disk if we are using a version 1 inode.
 711	 */
 712	if ((prid != 0) && (ip->i_d.di_version == 1))
 713		xfs_bump_ino_vers2(tp, ip);
 714
 715	if (pip && XFS_INHERIT_GID(pip)) {
 716		ip->i_d.di_gid = pip->i_d.di_gid;
 717		if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
 718			ip->i_d.di_mode |= S_ISGID;
 719		}
 720	}
 721
 722	/*
 723	 * If the group ID of the new file does not match the effective group
 724	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
 725	 * (and only if the irix_sgid_inherit compatibility variable is set).
 726	 */
 727	if ((irix_sgid_inherit) &&
 728	    (ip->i_d.di_mode & S_ISGID) &&
 729	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
 730		ip->i_d.di_mode &= ~S_ISGID;
 731	}
 732
 733	ip->i_d.di_size = 0;
 734	ip->i_d.di_nextents = 0;
 735	ASSERT(ip->i_d.di_nblocks == 0);
 736
 737	nanotime(&tv);
 738	ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
 739	ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
 740	ip->i_d.di_atime = ip->i_d.di_mtime;
 741	ip->i_d.di_ctime = ip->i_d.di_mtime;
 742
 743	/*
 744	 * di_gen will have been taken care of in xfs_iread.
 745	 */
 746	ip->i_d.di_extsize = 0;
 747	ip->i_d.di_dmevmask = 0;
 748	ip->i_d.di_dmstate = 0;
 749	ip->i_d.di_flags = 0;
 750
 751	if (ip->i_d.di_version == 3) {
 752		ASSERT(ip->i_d.di_ino == ino);
 753		ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
 754		ip->i_d.di_crc = 0;
 755		ip->i_d.di_changecount = 1;
 756		ip->i_d.di_lsn = 0;
 757		ip->i_d.di_flags2 = 0;
 758		memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
 759		ip->i_d.di_crtime = ip->i_d.di_mtime;
 760	}
 761
 762
 763	flags = XFS_ILOG_CORE;
 764	switch (mode & S_IFMT) {
 765	case S_IFIFO:
 766	case S_IFCHR:
 767	case S_IFBLK:
 768	case S_IFSOCK:
 769		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
 770		ip->i_df.if_u2.if_rdev = rdev;
 771		ip->i_df.if_flags = 0;
 772		flags |= XFS_ILOG_DEV;
 773		break;
 774	case S_IFREG:
 775		/*
 776		 * we can't set up filestreams until after the VFS inode
 777		 * is set up properly.
 778		 */
 779		if (pip && xfs_inode_is_filestream(pip))
 780			filestreams = 1;
 781		/* fall through */
 782	case S_IFDIR:
 783		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
 784			uint	di_flags = 0;
 785
 786			if (S_ISDIR(mode)) {
 787				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
 788					di_flags |= XFS_DIFLAG_RTINHERIT;
 789				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
 790					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
 791					ip->i_d.di_extsize = pip->i_d.di_extsize;
 792				}
 793			} else if (S_ISREG(mode)) {
 794				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
 795					di_flags |= XFS_DIFLAG_REALTIME;
 796				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
 797					di_flags |= XFS_DIFLAG_EXTSIZE;
 798					ip->i_d.di_extsize = pip->i_d.di_extsize;
 799				}
 800			}
 801			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
 802			    xfs_inherit_noatime)
 803				di_flags |= XFS_DIFLAG_NOATIME;
 804			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
 805			    xfs_inherit_nodump)
 806				di_flags |= XFS_DIFLAG_NODUMP;
 807			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
 808			    xfs_inherit_sync)
 809				di_flags |= XFS_DIFLAG_SYNC;
 810			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
 811			    xfs_inherit_nosymlinks)
 812				di_flags |= XFS_DIFLAG_NOSYMLINKS;
 813			if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
 814				di_flags |= XFS_DIFLAG_PROJINHERIT;
 815			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
 816			    xfs_inherit_nodefrag)
 817				di_flags |= XFS_DIFLAG_NODEFRAG;
 818			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
 819				di_flags |= XFS_DIFLAG_FILESTREAM;
 820			ip->i_d.di_flags |= di_flags;
 821		}
 822		/* FALLTHROUGH */
 823	case S_IFLNK:
 824		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
 825		ip->i_df.if_flags = XFS_IFEXTENTS;
 826		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
 827		ip->i_df.if_u1.if_extents = NULL;
 828		break;
 829	default:
 830		ASSERT(0);
 831	}
 
 832	/*
 833	 * Attribute fork settings for new inode.
 834	 */
 835	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
 836	ip->i_d.di_anextents = 0;
 
 
 
 
 
 
 
 
 837
 838	/*
 839	 * Log the new values stuffed into the inode.
 840	 */
 841	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 842	xfs_trans_log_inode(tp, ip, flags);
 843
 844	/* now that we have an i_mode we can setup inode ops and unlock */
 845	xfs_setup_inode(ip);
 846
 847	/* now we have set up the vfs inode we can associate the filestream */
 848	if (filestreams) {
 849		error = xfs_filestream_associate(pip, ip);
 850		if (error < 0)
 851			return -error;
 852		if (!error)
 853			xfs_iflags_set(ip, XFS_IFILESTREAM);
 854	}
 855
 856	*ipp = ip;
 857	return 0;
 858}
 859
 860/*
 861 * Allocates a new inode from disk and return a pointer to the
 862 * incore copy. This routine will internally commit the current
 863 * transaction and allocate a new one if the Space Manager needed
 864 * to do an allocation to replenish the inode free-list.
 865 *
 866 * This routine is designed to be called from xfs_create and
 867 * xfs_create_dir.
 868 *
 869 */
 870int
 871xfs_dir_ialloc(
 872	xfs_trans_t	**tpp,		/* input: current transaction;
 873					   output: may be a new transaction. */
 874	xfs_inode_t	*dp,		/* directory within whose allocate
 875					   the inode. */
 876	umode_t		mode,
 877	xfs_nlink_t	nlink,
 878	xfs_dev_t	rdev,
 879	prid_t		prid,		/* project id */
 880	int		okalloc,	/* ok to allocate new space */
 881	xfs_inode_t	**ipp,		/* pointer to inode; it will be
 882					   locked. */
 883	int		*committed)
 884
 885{
 886	xfs_trans_t	*tp;
 887	xfs_trans_t	*ntp;
 888	xfs_inode_t	*ip;
 889	xfs_buf_t	*ialloc_context = NULL;
 890	int		code;
 891	void		*dqinfo;
 892	uint		tflags;
 893
 894	tp = *tpp;
 895	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 896
 897	/*
 898	 * xfs_ialloc will return a pointer to an incore inode if
 899	 * the Space Manager has an available inode on the free
 900	 * list. Otherwise, it will do an allocation and replenish
 901	 * the freelist.  Since we can only do one allocation per
 902	 * transaction without deadlocks, we will need to commit the
 903	 * current transaction and start a new one.  We will then
 904	 * need to call xfs_ialloc again to get the inode.
 905	 *
 906	 * If xfs_ialloc did an allocation to replenish the freelist,
 907	 * it returns the bp containing the head of the freelist as
 908	 * ialloc_context. We will hold a lock on it across the
 909	 * transaction commit so that no other process can steal
 910	 * the inode(s) that we've just allocated.
 911	 */
 912	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
 913			  &ialloc_context, &ip);
 914
 915	/*
 916	 * Return an error if we were unable to allocate a new inode.
 917	 * This should only happen if we run out of space on disk or
 918	 * encounter a disk error.
 919	 */
 920	if (code) {
 921		*ipp = NULL;
 922		return code;
 923	}
 924	if (!ialloc_context && !ip) {
 925		*ipp = NULL;
 926		return XFS_ERROR(ENOSPC);
 927	}
 928
 929	/*
 930	 * If the AGI buffer is non-NULL, then we were unable to get an
 931	 * inode in one operation.  We need to commit the current
 932	 * transaction and call xfs_ialloc() again.  It is guaranteed
 933	 * to succeed the second time.
 934	 */
 935	if (ialloc_context) {
 936		struct xfs_trans_res tres;
 937
 938		/*
 939		 * Normally, xfs_trans_commit releases all the locks.
 940		 * We call bhold to hang on to the ialloc_context across
 941		 * the commit.  Holding this buffer prevents any other
 942		 * processes from doing any allocations in this
 943		 * allocation group.
 944		 */
 945		xfs_trans_bhold(tp, ialloc_context);
 946		/*
 947		 * Save the log reservation so we can use
 948		 * them in the next transaction.
 949		 */
 950		tres.tr_logres = xfs_trans_get_log_res(tp);
 951		tres.tr_logcount = xfs_trans_get_log_count(tp);
 952
 953		/*
 954		 * We want the quota changes to be associated with the next
 955		 * transaction, NOT this one. So, detach the dqinfo from this
 956		 * and attach it to the next transaction.
 957		 */
 958		dqinfo = NULL;
 959		tflags = 0;
 960		if (tp->t_dqinfo) {
 961			dqinfo = (void *)tp->t_dqinfo;
 962			tp->t_dqinfo = NULL;
 963			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
 964			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
 965		}
 966
 967		ntp = xfs_trans_dup(tp);
 968		code = xfs_trans_commit(tp, 0);
 969		tp = ntp;
 970		if (committed != NULL) {
 971			*committed = 1;
 972		}
 973		/*
 974		 * If we get an error during the commit processing,
 975		 * release the buffer that is still held and return
 976		 * to the caller.
 977		 */
 978		if (code) {
 979			xfs_buf_relse(ialloc_context);
 980			if (dqinfo) {
 981				tp->t_dqinfo = dqinfo;
 982				xfs_trans_free_dqinfo(tp);
 983			}
 984			*tpp = ntp;
 985			*ipp = NULL;
 986			return code;
 987		}
 988
 989		/*
 990		 * transaction commit worked ok so we can drop the extra ticket
 991		 * reference that we gained in xfs_trans_dup()
 992		 */
 993		xfs_log_ticket_put(tp->t_ticket);
 994		tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
 995		code = xfs_trans_reserve(tp, &tres, 0, 0);
 996
 997		/*
 998		 * Re-attach the quota info that we detached from prev trx.
 999		 */
1000		if (dqinfo) {
1001			tp->t_dqinfo = dqinfo;
1002			tp->t_flags |= tflags;
1003		}
1004
1005		if (code) {
1006			xfs_buf_relse(ialloc_context);
1007			*tpp = ntp;
1008			*ipp = NULL;
1009			return code;
1010		}
1011		xfs_trans_bjoin(tp, ialloc_context);
1012
1013		/*
1014		 * Call ialloc again. Since we've locked out all
1015		 * other allocations in this allocation group,
1016		 * this call should always succeed.
1017		 */
1018		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1019				  okalloc, &ialloc_context, &ip);
1020
1021		/*
1022		 * If we get an error at this point, return to the caller
1023		 * so that the current transaction can be aborted.
1024		 */
1025		if (code) {
1026			*tpp = tp;
1027			*ipp = NULL;
1028			return code;
1029		}
1030		ASSERT(!ialloc_context && ip);
1031
1032	} else {
1033		if (committed != NULL)
1034			*committed = 0;
1035	}
1036
1037	*ipp = ip;
1038	*tpp = tp;
1039
1040	return 0;
1041}
1042
1043/*
1044 * Decrement the link count on an inode & log the change.
1045 * If this causes the link count to go to zero, initiate the
1046 * logging activity required to truncate a file.
1047 */
1048int				/* error */
1049xfs_droplink(
1050	xfs_trans_t *tp,
1051	xfs_inode_t *ip)
1052{
1053	int	error;
1054
1055	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1056
1057	ASSERT (ip->i_d.di_nlink > 0);
1058	ip->i_d.di_nlink--;
1059	drop_nlink(VFS_I(ip));
1060	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1061
1062	error = 0;
1063	if (ip->i_d.di_nlink == 0) {
1064		/*
1065		 * We're dropping the last link to this file.
1066		 * Move the on-disk inode to the AGI unlinked list.
1067		 * From xfs_inactive() we will pull the inode from
1068		 * the list and free it.
1069		 */
1070		error = xfs_iunlink(tp, ip);
1071	}
1072	return error;
1073}
1074
1075/*
1076 * This gets called when the inode's version needs to be changed from 1 to 2.
1077 * Currently this happens when the nlink field overflows the old 16-bit value
1078 * or when chproj is called to change the project for the first time.
1079 * As a side effect the superblock version will also get rev'd
1080 * to contain the NLINK bit.
1081 */
1082void
1083xfs_bump_ino_vers2(
1084	xfs_trans_t	*tp,
1085	xfs_inode_t	*ip)
1086{
1087	xfs_mount_t	*mp;
1088
1089	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1090	ASSERT(ip->i_d.di_version == 1);
1091
1092	ip->i_d.di_version = 2;
1093	ip->i_d.di_onlink = 0;
1094	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1095	mp = tp->t_mountp;
1096	if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1097		spin_lock(&mp->m_sb_lock);
1098		if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1099			xfs_sb_version_addnlink(&mp->m_sb);
1100			spin_unlock(&mp->m_sb_lock);
1101			xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
1102		} else {
1103			spin_unlock(&mp->m_sb_lock);
1104		}
1105	}
1106	/* Caller must log the inode */
1107}
1108
1109/*
1110 * Increment the link count on an inode & log the change.
1111 */
1112int
1113xfs_bumplink(
1114	xfs_trans_t *tp,
1115	xfs_inode_t *ip)
1116{
1117	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1118
1119	ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
1120	ip->i_d.di_nlink++;
1121	inc_nlink(VFS_I(ip));
1122	if ((ip->i_d.di_version == 1) &&
1123	    (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
1124		/*
1125		 * The inode has increased its number of links beyond
1126		 * what can fit in an old format inode.  It now needs
1127		 * to be converted to a version 2 inode with a 32 bit
1128		 * link count.  If this is the first inode in the file
1129		 * system to do this, then we need to bump the superblock
1130		 * version number as well.
1131		 */
1132		xfs_bump_ino_vers2(tp, ip);
1133	}
1134
1135	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1136	return 0;
1137}
1138
1139int
1140xfs_create(
 
1141	xfs_inode_t		*dp,
1142	struct xfs_name		*name,
1143	umode_t			mode,
1144	xfs_dev_t		rdev,
 
1145	xfs_inode_t		**ipp)
1146{
1147	int			is_dir = S_ISDIR(mode);
1148	struct xfs_mount	*mp = dp->i_mount;
1149	struct xfs_inode	*ip = NULL;
1150	struct xfs_trans	*tp = NULL;
1151	int			error;
1152	xfs_bmap_free_t		free_list;
1153	xfs_fsblock_t		first_block;
1154	bool                    unlock_dp_on_error = false;
1155	uint			cancel_flags;
1156	int			committed;
1157	prid_t			prid;
1158	struct xfs_dquot	*udqp = NULL;
1159	struct xfs_dquot	*gdqp = NULL;
1160	struct xfs_dquot	*pdqp = NULL;
1161	struct xfs_trans_res	tres;
1162	uint			resblks;
 
1163
1164	trace_xfs_create(dp, name);
1165
1166	if (XFS_FORCED_SHUTDOWN(mp))
1167		return XFS_ERROR(EIO);
1168
1169	prid = xfs_get_initial_prid(dp);
1170
1171	/*
1172	 * Make sure that we have allocated dquot(s) on disk.
1173	 */
1174	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1175					xfs_kgid_to_gid(current_fsgid()), prid,
1176					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1177					&udqp, &gdqp, &pdqp);
1178	if (error)
1179		return error;
1180
1181	if (is_dir) {
1182		rdev = 0;
1183		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1184		tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1185		tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
1186		tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1187	} else {
1188		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1189		tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1190		tres.tr_logcount = XFS_CREATE_LOG_COUNT;
1191		tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1192	}
1193
1194	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1195
1196	/*
1197	 * Initially assume that the file does not exist and
1198	 * reserve the resources for that case.  If that is not
1199	 * the case we'll drop the one we have and get a more
1200	 * appropriate transaction later.
1201	 */
1202	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1203	error = xfs_trans_reserve(tp, &tres, resblks, 0);
1204	if (error == ENOSPC) {
1205		/* flush outstanding delalloc blocks and retry */
1206		xfs_flush_inodes(mp);
1207		error = xfs_trans_reserve(tp, &tres, resblks, 0);
1208	}
1209	if (error == ENOSPC) {
1210		/* No space at all so try a "no-allocation" reservation */
1211		resblks = 0;
1212		error = xfs_trans_reserve(tp, &tres, 0, 0);
1213	}
1214	if (error) {
1215		cancel_flags = 0;
1216		goto out_trans_cancel;
1217	}
 
 
1218
1219	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1220	unlock_dp_on_error = true;
1221
1222	xfs_bmap_init(&free_list, &first_block);
1223
1224	/*
1225	 * Reserve disk quota and the inode.
1226	 */
1227	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1228						pdqp, resblks, 1, 0);
1229	if (error)
1230		goto out_trans_cancel;
1231
1232	error = xfs_dir_canenter(tp, dp, name, resblks);
1233	if (error)
1234		goto out_trans_cancel;
1235
1236	/*
1237	 * A newly created regular or special file just has one directory
1238	 * entry pointing to them, but a directory also the "." entry
1239	 * pointing to itself.
1240	 */
1241	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1242			       prid, resblks > 0, &ip, &committed);
1243	if (error) {
1244		if (error == ENOSPC)
1245			goto out_trans_cancel;
1246		goto out_trans_abort;
1247	}
1248
1249	/*
1250	 * Now we join the directory inode to the transaction.  We do not do it
1251	 * earlier because xfs_dir_ialloc might commit the previous transaction
1252	 * (and release all the locks).  An error from here on will result in
1253	 * the transaction cancel unlocking dp so don't do it explicitly in the
1254	 * error path.
1255	 */
1256	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1257	unlock_dp_on_error = false;
1258
1259	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1260					&first_block, &free_list, resblks ?
1261					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1262	if (error) {
1263		ASSERT(error != ENOSPC);
1264		goto out_trans_abort;
1265	}
1266	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1267	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1268
1269	if (is_dir) {
1270		error = xfs_dir_init(tp, ip, dp);
1271		if (error)
1272			goto out_bmap_cancel;
1273
1274		error = xfs_bumplink(tp, dp);
1275		if (error)
1276			goto out_bmap_cancel;
1277	}
1278
1279	/*
1280	 * If this is a synchronous mount, make sure that the
1281	 * create transaction goes to disk before returning to
1282	 * the user.
1283	 */
1284	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1285		xfs_trans_set_sync(tp);
1286
1287	/*
1288	 * Attach the dquot(s) to the inodes and modify them incore.
1289	 * These ids of the inode couldn't have changed since the new
1290	 * inode has been locked ever since it was created.
1291	 */
1292	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1293
1294	error = xfs_bmap_finish(&tp, &free_list, &committed);
1295	if (error)
1296		goto out_bmap_cancel;
1297
1298	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1299	if (error)
1300		goto out_release_inode;
1301
1302	xfs_qm_dqrele(udqp);
1303	xfs_qm_dqrele(gdqp);
1304	xfs_qm_dqrele(pdqp);
1305
1306	*ipp = ip;
1307	return 0;
1308
1309 out_bmap_cancel:
1310	xfs_bmap_cancel(&free_list);
1311 out_trans_abort:
1312	cancel_flags |= XFS_TRANS_ABORT;
1313 out_trans_cancel:
1314	xfs_trans_cancel(tp, cancel_flags);
1315 out_release_inode:
1316	/*
1317	 * Wait until after the current transaction is aborted to
1318	 * release the inode.  This prevents recursive transactions
1319	 * and deadlocks from xfs_inactive.
1320	 */
1321	if (ip)
1322		IRELE(ip);
1323
 
 
1324	xfs_qm_dqrele(udqp);
1325	xfs_qm_dqrele(gdqp);
1326	xfs_qm_dqrele(pdqp);
1327
1328	if (unlock_dp_on_error)
1329		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1330	return error;
1331}
1332
1333int
1334xfs_create_tmpfile(
 
1335	struct xfs_inode	*dp,
1336	struct dentry		*dentry,
1337	umode_t			mode,
1338	struct xfs_inode	**ipp)
1339{
1340	struct xfs_mount	*mp = dp->i_mount;
1341	struct xfs_inode	*ip = NULL;
1342	struct xfs_trans	*tp = NULL;
1343	int			error;
1344	uint			cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1345	prid_t                  prid;
1346	struct xfs_dquot	*udqp = NULL;
1347	struct xfs_dquot	*gdqp = NULL;
1348	struct xfs_dquot	*pdqp = NULL;
1349	struct xfs_trans_res	*tres;
1350	uint			resblks;
 
1351
1352	if (XFS_FORCED_SHUTDOWN(mp))
1353		return XFS_ERROR(EIO);
1354
1355	prid = xfs_get_initial_prid(dp);
1356
1357	/*
1358	 * Make sure that we have allocated dquot(s) on disk.
1359	 */
1360	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1361				xfs_kgid_to_gid(current_fsgid()), prid,
1362				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1363				&udqp, &gdqp, &pdqp);
1364	if (error)
1365		return error;
1366
1367	resblks = XFS_IALLOC_SPACE_RES(mp);
1368	tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1369
1370	tres = &M_RES(mp)->tr_create_tmpfile;
1371	error = xfs_trans_reserve(tp, tres, resblks, 0);
1372	if (error == ENOSPC) {
1373		/* No space at all so try a "no-allocation" reservation */
1374		resblks = 0;
1375		error = xfs_trans_reserve(tp, tres, 0, 0);
1376	}
1377	if (error) {
1378		cancel_flags = 0;
1379		goto out_trans_cancel;
1380	}
1381
1382	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1383						pdqp, resblks, 1, 0);
 
 
1384	if (error)
1385		goto out_trans_cancel;
1386
1387	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1388				prid, resblks > 0, &ip, NULL);
1389	if (error) {
1390		if (error == ENOSPC)
1391			goto out_trans_cancel;
1392		goto out_trans_abort;
1393	}
1394
1395	if (mp->m_flags & XFS_MOUNT_WSYNC)
1396		xfs_trans_set_sync(tp);
1397
1398	/*
1399	 * Attach the dquot(s) to the inodes and modify them incore.
1400	 * These ids of the inode couldn't have changed since the new
1401	 * inode has been locked ever since it was created.
1402	 */
1403	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1404
1405	ip->i_d.di_nlink--;
1406	error = xfs_iunlink(tp, ip);
1407	if (error)
1408		goto out_trans_abort;
1409
1410	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1411	if (error)
1412		goto out_release_inode;
1413
1414	xfs_qm_dqrele(udqp);
1415	xfs_qm_dqrele(gdqp);
1416	xfs_qm_dqrele(pdqp);
1417
1418	*ipp = ip;
1419	return 0;
1420
1421 out_trans_abort:
1422	cancel_flags |= XFS_TRANS_ABORT;
1423 out_trans_cancel:
1424	xfs_trans_cancel(tp, cancel_flags);
1425 out_release_inode:
1426	/*
1427	 * Wait until after the current transaction is aborted to
1428	 * release the inode.  This prevents recursive transactions
1429	 * and deadlocks from xfs_inactive.
1430	 */
1431	if (ip)
1432		IRELE(ip);
1433
 
 
1434	xfs_qm_dqrele(udqp);
1435	xfs_qm_dqrele(gdqp);
1436	xfs_qm_dqrele(pdqp);
1437
1438	return error;
1439}
1440
1441int
1442xfs_link(
1443	xfs_inode_t		*tdp,
1444	xfs_inode_t		*sip,
1445	struct xfs_name		*target_name)
1446{
1447	xfs_mount_t		*mp = tdp->i_mount;
1448	xfs_trans_t		*tp;
1449	int			error;
1450	xfs_bmap_free_t         free_list;
1451	xfs_fsblock_t           first_block;
1452	int			cancel_flags;
1453	int			committed;
1454	int			resblks;
1455
1456	trace_xfs_link(tdp, target_name);
1457
1458	ASSERT(!S_ISDIR(sip->i_d.di_mode));
1459
1460	if (XFS_FORCED_SHUTDOWN(mp))
1461		return XFS_ERROR(EIO);
1462
1463	error = xfs_qm_dqattach(sip, 0);
1464	if (error)
1465		goto std_return;
1466
1467	error = xfs_qm_dqattach(tdp, 0);
1468	if (error)
1469		goto std_return;
1470
1471	tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1472	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1473	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1474	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1475	if (error == ENOSPC) {
1476		resblks = 0;
1477		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1478	}
1479	if (error) {
1480		cancel_flags = 0;
1481		goto error_return;
1482	}
1483
1484	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1485
1486	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1487	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1488
1489	/*
1490	 * If we are using project inheritance, we only allow hard link
1491	 * creation in our tree when the project IDs are the same; else
1492	 * the tree quota mechanism could be circumvented.
1493	 */
1494	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1495		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1496		error = XFS_ERROR(EXDEV);
1497		goto error_return;
1498	}
1499
1500	error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1501	if (error)
1502		goto error_return;
 
 
1503
1504	xfs_bmap_init(&free_list, &first_block);
 
 
 
 
1505
1506	if (sip->i_d.di_nlink == 0) {
1507		error = xfs_iunlink_remove(tp, sip);
 
1508		if (error)
1509			goto abort_return;
1510	}
1511
1512	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1513					&first_block, &free_list, resblks);
1514	if (error)
1515		goto abort_return;
1516	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1517	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1518
1519	error = xfs_bumplink(tp, sip);
1520	if (error)
1521		goto abort_return;
1522
1523	/*
1524	 * If this is a synchronous mount, make sure that the
1525	 * link transaction goes to disk before returning to
1526	 * the user.
1527	 */
1528	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1529		xfs_trans_set_sync(tp);
1530	}
1531
1532	error = xfs_bmap_finish (&tp, &free_list, &committed);
1533	if (error) {
1534		xfs_bmap_cancel(&free_list);
1535		goto abort_return;
1536	}
1537
1538	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1539
1540 abort_return:
1541	cancel_flags |= XFS_TRANS_ABORT;
1542 error_return:
1543	xfs_trans_cancel(tp, cancel_flags);
1544 std_return:
 
 
1545	return error;
1546}
1547
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1548/*
1549 * Free up the underlying blocks past new_size.  The new size must be smaller
1550 * than the current size.  This routine can be used both for the attribute and
1551 * data fork, and does not modify the inode size, which is left to the caller.
1552 *
1553 * The transaction passed to this routine must have made a permanent log
1554 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1555 * given transaction and start new ones, so make sure everything involved in
1556 * the transaction is tidy before calling here.  Some transaction will be
1557 * returned to the caller to be committed.  The incoming transaction must
1558 * already include the inode, and both inode locks must be held exclusively.
1559 * The inode must also be "held" within the transaction.  On return the inode
1560 * will be "held" within the returned transaction.  This routine does NOT
1561 * require any disk space to be reserved for it within the transaction.
1562 *
1563 * If we get an error, we must return with the inode locked and linked into the
1564 * current transaction. This keeps things simple for the higher level code,
1565 * because it always knows that the inode is locked and held in the transaction
1566 * that returns to it whether errors occur or not.  We don't mark the inode
1567 * dirty on error so that transactions can be easily aborted if possible.
1568 */
1569int
1570xfs_itruncate_extents(
1571	struct xfs_trans	**tpp,
1572	struct xfs_inode	*ip,
1573	int			whichfork,
1574	xfs_fsize_t		new_size)
 
1575{
1576	struct xfs_mount	*mp = ip->i_mount;
1577	struct xfs_trans	*tp = *tpp;
1578	struct xfs_trans	*ntp;
1579	xfs_bmap_free_t		free_list;
1580	xfs_fsblock_t		first_block;
1581	xfs_fileoff_t		first_unmap_block;
1582	xfs_fileoff_t		last_block;
1583	xfs_filblks_t		unmap_len;
1584	int			committed;
1585	int			error = 0;
1586	int			done = 0;
1587
1588	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1589	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1590	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1591	ASSERT(new_size <= XFS_ISIZE(ip));
1592	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1593	ASSERT(ip->i_itemp != NULL);
1594	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1595	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1596
1597	trace_xfs_itruncate_extents_start(ip, new_size);
1598
 
 
1599	/*
1600	 * Since it is possible for space to become allocated beyond
1601	 * the end of the file (in a crash where the space is allocated
1602	 * but the inode size is not yet updated), simply remove any
1603	 * blocks which show up between the new EOF and the maximum
1604	 * possible file size.  If the first block to be removed is
1605	 * beyond the maximum file size (ie it is the same as last_block),
1606	 * then there is nothing to do.
 
1607	 */
1608	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1609	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1610	if (first_unmap_block == last_block)
1611		return 0;
 
1612
1613	ASSERT(first_unmap_block < last_block);
1614	unmap_len = last_block - first_unmap_block + 1;
1615	while (!done) {
1616		xfs_bmap_init(&free_list, &first_block);
1617		error = xfs_bunmapi(tp, ip,
1618				    first_unmap_block, unmap_len,
1619				    xfs_bmapi_aflag(whichfork),
1620				    XFS_ITRUNC_MAX_EXTENTS,
1621				    &first_block, &free_list,
1622				    &done);
1623		if (error)
1624			goto out_bmap_cancel;
1625
1626		/*
1627		 * Duplicate the transaction that has the permanent
1628		 * reservation and commit the old transaction.
1629		 */
1630		error = xfs_bmap_finish(&tp, &free_list, &committed);
1631		if (committed)
1632			xfs_trans_ijoin(tp, ip, 0);
1633		if (error)
1634			goto out_bmap_cancel;
1635
1636		if (committed) {
1637			/*
1638			 * Mark the inode dirty so it will be logged and
1639			 * moved forward in the log as part of every commit.
1640			 */
1641			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1642		}
1643
1644		ntp = xfs_trans_dup(tp);
1645		error = xfs_trans_commit(tp, 0);
1646		tp = ntp;
1647
1648		xfs_trans_ijoin(tp, ip, 0);
1649
 
 
1650		if (error)
1651			goto out;
 
1652
1653		/*
1654		 * Transaction commit worked ok so we can drop the extra ticket
1655		 * reference that we gained in xfs_trans_dup()
1656		 */
1657		xfs_log_ticket_put(tp->t_ticket);
1658		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1659		if (error)
1660			goto out;
 
 
1661	}
1662
1663	/*
1664	 * Always re-log the inode so that our permanent transaction can keep
1665	 * on rolling it forward in the log.
1666	 */
1667	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1668
1669	trace_xfs_itruncate_extents_end(ip, new_size);
1670
1671out:
1672	*tpp = tp;
1673	return error;
1674out_bmap_cancel:
1675	/*
1676	 * If the bunmapi call encounters an error, return to the caller where
1677	 * the transaction can be properly aborted.  We just need to make sure
1678	 * we're not holding any resources that we were not when we came in.
1679	 */
1680	xfs_bmap_cancel(&free_list);
1681	goto out;
1682}
1683
1684int
1685xfs_release(
1686	xfs_inode_t	*ip)
1687{
1688	xfs_mount_t	*mp = ip->i_mount;
1689	int		error;
1690
1691	if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1692		return 0;
1693
1694	/* If this is a read-only mount, don't do this (would generate I/O) */
1695	if (mp->m_flags & XFS_MOUNT_RDONLY)
1696		return 0;
1697
1698	if (!XFS_FORCED_SHUTDOWN(mp)) {
1699		int truncated;
1700
1701		/*
1702		 * If we are using filestreams, and we have an unlinked
1703		 * file that we are processing the last close on, then nothing
1704		 * will be able to reopen and write to this file. Purge this
1705		 * inode from the filestreams cache so that it doesn't delay
1706		 * teardown of the inode.
1707		 */
1708		if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1709			xfs_filestream_deassociate(ip);
1710
1711		/*
1712		 * If we previously truncated this file and removed old data
1713		 * in the process, we want to initiate "early" writeout on
1714		 * the last close.  This is an attempt to combat the notorious
1715		 * NULL files problem which is particularly noticeable from a
1716		 * truncate down, buffered (re-)write (delalloc), followed by
1717		 * a crash.  What we are effectively doing here is
1718		 * significantly reducing the time window where we'd otherwise
1719		 * be exposed to that problem.
1720		 */
1721		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1722		if (truncated) {
1723			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1724			if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1725				error = -filemap_flush(VFS_I(ip)->i_mapping);
1726				if (error)
1727					return error;
1728			}
1729		}
1730	}
1731
1732	if (ip->i_d.di_nlink == 0)
 
 
 
 
 
 
 
 
 
1733		return 0;
1734
1735	if (xfs_can_free_eofblocks(ip, false)) {
1736
1737		/*
1738		 * If we can't get the iolock just skip truncating the blocks
1739		 * past EOF because we could deadlock with the mmap_sem
1740		 * otherwise.  We'll get another chance to drop them once the
1741		 * last reference to the inode is dropped, so we'll never leak
1742		 * blocks permanently.
1743		 *
1744		 * Further, check if the inode is being opened, written and
1745		 * closed frequently and we have delayed allocation blocks
1746		 * outstanding (e.g. streaming writes from the NFS server),
1747		 * truncating the blocks past EOF will cause fragmentation to
1748		 * occur.
1749		 *
1750		 * In this case don't do the truncation, either, but we have to
1751		 * be careful how we detect this case. Blocks beyond EOF show
1752		 * up as i_delayed_blks even when the inode is clean, so we
1753		 * need to truncate them away first before checking for a dirty
1754		 * release. Hence on the first dirty close we will still remove
1755		 * the speculative allocation, but after that we will leave it
1756		 * in place.
1757		 */
1758		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1759			return 0;
1760
1761		error = xfs_free_eofblocks(mp, ip, true);
1762		if (error && error != EAGAIN)
1763			return error;
1764
1765		/* delalloc blocks after truncation means it really is dirty */
1766		if (ip->i_delayed_blks)
1767			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1768	}
1769	return 0;
 
 
 
1770}
1771
1772/*
1773 * xfs_inactive_truncate
1774 *
1775 * Called to perform a truncate when an inode becomes unlinked.
1776 */
1777STATIC int
1778xfs_inactive_truncate(
1779	struct xfs_inode *ip)
1780{
1781	struct xfs_mount	*mp = ip->i_mount;
1782	struct xfs_trans	*tp;
1783	int			error;
1784
1785	tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1786	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1787	if (error) {
1788		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1789		xfs_trans_cancel(tp, 0);
1790		return error;
1791	}
1792
1793	xfs_ilock(ip, XFS_ILOCK_EXCL);
1794	xfs_trans_ijoin(tp, ip, 0);
1795
1796	/*
1797	 * Log the inode size first to prevent stale data exposure in the event
1798	 * of a system crash before the truncate completes. See the related
1799	 * comment in xfs_setattr_size() for details.
1800	 */
1801	ip->i_d.di_size = 0;
1802	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1803
1804	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1805	if (error)
1806		goto error_trans_cancel;
1807
1808	ASSERT(ip->i_d.di_nextents == 0);
1809
1810	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1811	if (error)
1812		goto error_unlock;
1813
1814	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1815	return 0;
1816
1817error_trans_cancel:
1818	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1819error_unlock:
1820	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1821	return error;
1822}
1823
1824/*
1825 * xfs_inactive_ifree()
1826 *
1827 * Perform the inode free when an inode is unlinked.
1828 */
1829STATIC int
1830xfs_inactive_ifree(
1831	struct xfs_inode *ip)
1832{
1833	xfs_bmap_free_t		free_list;
1834	xfs_fsblock_t		first_block;
1835	int			committed;
1836	struct xfs_mount	*mp = ip->i_mount;
1837	struct xfs_trans	*tp;
1838	int			error;
1839
1840	tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1841	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1842	if (error) {
1843		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1844		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
 
 
 
 
 
1845		return error;
1846	}
1847
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1848	xfs_ilock(ip, XFS_ILOCK_EXCL);
1849	xfs_trans_ijoin(tp, ip, 0);
1850
1851	xfs_bmap_init(&free_list, &first_block);
1852	error = xfs_ifree(tp, ip, &free_list);
1853	if (error) {
1854		/*
1855		 * If we fail to free the inode, shut down.  The cancel
1856		 * might do that, we need to make sure.  Otherwise the
1857		 * inode might be lost for a long time or forever.
1858		 */
1859		if (!XFS_FORCED_SHUTDOWN(mp)) {
1860			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1861				__func__, error);
1862			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1863		}
1864		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1865		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1866		return error;
1867	}
1868
1869	/*
1870	 * Credit the quota account(s). The inode is gone.
1871	 */
1872	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1873
1874	/*
1875	 * Just ignore errors at this point.  There is nothing we can
1876	 * do except to try to keep going. Make sure it's not a silent
1877	 * error.
1878	 */
1879	error = xfs_bmap_finish(&tp,  &free_list, &committed);
1880	if (error)
1881		xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1882			__func__, error);
1883	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1884	if (error)
1885		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1886			__func__, error);
1887
1888	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1889	return 0;
1890}
1891
1892/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1893 * xfs_inactive
1894 *
1895 * This is called when the vnode reference count for the vnode
1896 * goes to zero.  If the file has been unlinked, then it must
1897 * now be truncated.  Also, we clear all of the read-ahead state
1898 * kept for the inode here since the file is now closed.
1899 */
1900void
1901xfs_inactive(
1902	xfs_inode_t	*ip)
1903{
1904	struct xfs_mount	*mp;
1905	int			error;
1906	int			truncate = 0;
1907
1908	/*
1909	 * If the inode is already free, then there can be nothing
1910	 * to clean up here.
1911	 */
1912	if (ip->i_d.di_mode == 0) {
1913		ASSERT(ip->i_df.if_real_bytes == 0);
1914		ASSERT(ip->i_df.if_broot_bytes == 0);
1915		return;
1916	}
1917
1918	mp = ip->i_mount;
 
1919
1920	/* If this is a read-only mount, don't do this (would generate I/O) */
1921	if (mp->m_flags & XFS_MOUNT_RDONLY)
1922		return;
 
 
 
 
 
 
 
 
1923
1924	if (ip->i_d.di_nlink != 0) {
1925		/*
1926		 * force is true because we are evicting an inode from the
1927		 * cache. Post-eof blocks must be freed, lest we end up with
1928		 * broken free space accounting.
 
 
 
 
1929		 */
1930		if (xfs_can_free_eofblocks(ip, true))
1931			xfs_free_eofblocks(mp, ip, false);
1932
1933		return;
1934	}
1935
1936	if (S_ISREG(ip->i_d.di_mode) &&
1937	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1938	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1939		truncate = 1;
1940
1941	error = xfs_qm_dqattach(ip, 0);
1942	if (error)
1943		return;
1944
1945	if (S_ISLNK(ip->i_d.di_mode))
1946		error = xfs_inactive_symlink(ip);
1947	else if (truncate)
1948		error = xfs_inactive_truncate(ip);
1949	if (error)
1950		return;
1951
1952	/*
1953	 * If there are attributes associated with the file then blow them away
1954	 * now.  The code calls a routine that recursively deconstructs the
1955	 * attribute fork.  We need to just commit the current transaction
1956	 * because we can't use it for xfs_attr_inactive().
1957	 */
1958	if (ip->i_d.di_anextents > 0) {
1959		ASSERT(ip->i_d.di_forkoff != 0);
1960
1961		error = xfs_attr_inactive(ip);
1962		if (error)
1963			return;
1964	}
1965
1966	if (ip->i_afp)
1967		xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1968
1969	ASSERT(ip->i_d.di_anextents == 0);
1970
1971	/*
1972	 * Free the inode.
1973	 */
1974	error = xfs_inactive_ifree(ip);
1975	if (error)
1976		return;
1977
 
1978	/*
1979	 * Release the dquots held by inode, if any.
 
1980	 */
1981	xfs_qm_dqdetach(ip);
1982}
1983
1984/*
1985 * This is called when the inode's link count goes to 0.
1986 * We place the on-disk inode on a list in the AGI.  It
1987 * will be pulled from this list when the inode is freed.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1988 */
1989int
1990xfs_iunlink(
1991	xfs_trans_t	*tp,
1992	xfs_inode_t	*ip)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1993{
1994	xfs_mount_t	*mp;
1995	xfs_agi_t	*agi;
1996	xfs_dinode_t	*dip;
1997	xfs_buf_t	*agibp;
1998	xfs_buf_t	*ibp;
1999	xfs_agino_t	agino;
2000	short		bucket_index;
2001	int		offset;
2002	int		error;
2003
2004	ASSERT(ip->i_d.di_nlink == 0);
2005	ASSERT(ip->i_d.di_mode != 0);
 
 
 
 
 
 
 
2006
2007	mp = tp->t_mountp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2008
2009	/*
2010	 * Get the agi buffer first.  It ensures lock ordering
2011	 * on the list.
 
2012	 */
2013	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
2014	if (error)
2015		return error;
2016	agi = XFS_BUF_TO_AGI(agibp);
 
 
2017
2018	/*
2019	 * Get the index into the agi hash table for the
2020	 * list this inode will go on.
2021	 */
2022	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2023	ASSERT(agino != 0);
2024	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2025	ASSERT(agi->agi_unlinked[bucket_index]);
2026	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2027
2028	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2029		/*
2030		 * There is already another inode in the bucket we need
2031		 * to add ourselves to.  Add us at the front of the list.
2032		 * Here we put the head pointer into our next pointer,
2033		 * and then we fall through to point the head at us.
2034		 */
2035		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2036				       0, 0);
2037		if (error)
2038			return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2039
2040		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2041		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2042		offset = ip->i_imap.im_boffset +
2043			offsetof(xfs_dinode_t, di_next_unlinked);
 
 
 
 
 
 
 
 
 
2044
2045		/* need to recalc the inode CRC if appropriate */
2046		xfs_dinode_calc_crc(mp, dip);
2047
2048		xfs_trans_inode_buf(tp, ibp);
2049		xfs_trans_log_buf(tp, ibp, offset,
2050				  (offset + sizeof(xfs_agino_t) - 1));
2051		xfs_inobp_check(mp, ibp);
 
 
 
 
 
2052	}
2053
2054	/*
2055	 * Point the bucket head pointer at the inode being inserted.
 
 
2056	 */
2057	ASSERT(agino != 0);
2058	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2059	offset = offsetof(xfs_agi_t, agi_unlinked) +
2060		(sizeof(xfs_agino_t) * bucket_index);
2061	xfs_trans_log_buf(tp, agibp, offset,
2062			  (offset + sizeof(xfs_agino_t) - 1));
2063	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2064}
2065
2066/*
2067 * Pull the on-disk inode from the AGI unlinked list.
2068 */
2069STATIC int
2070xfs_iunlink_remove(
2071	xfs_trans_t	*tp,
2072	xfs_inode_t	*ip)
 
2073{
2074	xfs_ino_t	next_ino;
2075	xfs_mount_t	*mp;
2076	xfs_agi_t	*agi;
2077	xfs_dinode_t	*dip;
2078	xfs_buf_t	*agibp;
2079	xfs_buf_t	*ibp;
2080	xfs_agnumber_t	agno;
2081	xfs_agino_t	agino;
2082	xfs_agino_t	next_agino;
2083	xfs_buf_t	*last_ibp;
2084	xfs_dinode_t	*last_dip = NULL;
2085	short		bucket_index;
2086	int		offset, last_offset = 0;
2087	int		error;
2088
2089	mp = tp->t_mountp;
2090	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2091
2092	/*
2093	 * Get the agi buffer first.  It ensures lock ordering
2094	 * on the list.
2095	 */
2096	error = xfs_read_agi(mp, tp, agno, &agibp);
2097	if (error)
2098		return error;
2099
2100	agi = XFS_BUF_TO_AGI(agibp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2101
2102	/*
2103	 * Get the index into the agi hash table for the
2104	 * list this inode will go on.
 
 
 
 
 
 
 
 
 
 
 
 
2105	 */
2106	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2107	ASSERT(agino != 0);
2108	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2109	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2110	ASSERT(agi->agi_unlinked[bucket_index]);
2111
2112	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2113		/*
2114		 * We're at the head of the list.  Get the inode's on-disk
2115		 * buffer to see if there is anyone after us on the list.
2116		 * Only modify our next pointer if it is not already NULLAGINO.
2117		 * This saves us the overhead of dealing with the buffer when
2118		 * there is no need to change it.
2119		 */
2120		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2121				       0, 0);
2122		if (error) {
2123			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2124				__func__, error);
2125			return error;
2126		}
2127		next_agino = be32_to_cpu(dip->di_next_unlinked);
2128		ASSERT(next_agino != 0);
2129		if (next_agino != NULLAGINO) {
2130			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2131			offset = ip->i_imap.im_boffset +
2132				offsetof(xfs_dinode_t, di_next_unlinked);
2133
2134			/* need to recalc the inode CRC if appropriate */
2135			xfs_dinode_calc_crc(mp, dip);
2136
2137			xfs_trans_inode_buf(tp, ibp);
2138			xfs_trans_log_buf(tp, ibp, offset,
2139					  (offset + sizeof(xfs_agino_t) - 1));
2140			xfs_inobp_check(mp, ibp);
2141		} else {
2142			xfs_trans_brelse(tp, ibp);
2143		}
2144		/*
2145		 * Point the bucket head pointer at the next inode.
2146		 */
2147		ASSERT(next_agino != 0);
2148		ASSERT(next_agino != agino);
2149		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2150		offset = offsetof(xfs_agi_t, agi_unlinked) +
2151			(sizeof(xfs_agino_t) * bucket_index);
2152		xfs_trans_log_buf(tp, agibp, offset,
2153				  (offset + sizeof(xfs_agino_t) - 1));
2154	} else {
2155		/*
2156		 * We need to search the list for the inode being freed.
2157		 */
2158		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2159		last_ibp = NULL;
2160		while (next_agino != agino) {
2161			struct xfs_imap	imap;
2162
2163			if (last_ibp)
2164				xfs_trans_brelse(tp, last_ibp);
2165
2166			imap.im_blkno = 0;
2167			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2168
2169			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2170			if (error) {
2171				xfs_warn(mp,
2172	"%s: xfs_imap returned error %d.",
2173					 __func__, error);
2174				return error;
2175			}
2176
2177			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2178					       &last_ibp, 0, 0);
2179			if (error) {
2180				xfs_warn(mp,
2181	"%s: xfs_imap_to_bp returned error %d.",
2182					__func__, error);
2183				return error;
2184			}
2185
2186			last_offset = imap.im_boffset;
2187			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2188			ASSERT(next_agino != NULLAGINO);
2189			ASSERT(next_agino != 0);
2190		}
2191
2192		/*
2193		 * Now last_ibp points to the buffer previous to us on the
2194		 * unlinked list.  Pull us from the list.
2195		 */
2196		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2197				       0, 0);
2198		if (error) {
2199			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2200				__func__, error);
2201			return error;
2202		}
2203		next_agino = be32_to_cpu(dip->di_next_unlinked);
2204		ASSERT(next_agino != 0);
2205		ASSERT(next_agino != agino);
2206		if (next_agino != NULLAGINO) {
2207			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2208			offset = ip->i_imap.im_boffset +
2209				offsetof(xfs_dinode_t, di_next_unlinked);
2210
2211			/* need to recalc the inode CRC if appropriate */
2212			xfs_dinode_calc_crc(mp, dip);
2213
2214			xfs_trans_inode_buf(tp, ibp);
2215			xfs_trans_log_buf(tp, ibp, offset,
2216					  (offset + sizeof(xfs_agino_t) - 1));
2217			xfs_inobp_check(mp, ibp);
2218		} else {
2219			xfs_trans_brelse(tp, ibp);
2220		}
2221		/*
2222		 * Point the previous inode on the list to the next inode.
2223		 */
2224		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2225		ASSERT(next_agino != 0);
2226		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2227
2228		/* need to recalc the inode CRC if appropriate */
2229		xfs_dinode_calc_crc(mp, last_dip);
 
2230
2231		xfs_trans_inode_buf(tp, last_ibp);
2232		xfs_trans_log_buf(tp, last_ibp, offset,
2233				  (offset + sizeof(xfs_agino_t) - 1));
2234		xfs_inobp_check(mp, last_ibp);
2235	}
2236	return 0;
2237}
2238
2239/*
2240 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2241 * inodes that are in memory - they all must be marked stale and attached to
2242 * the cluster buffer.
2243 */
2244STATIC int
2245xfs_ifree_cluster(
2246	xfs_inode_t	*free_ip,
2247	xfs_trans_t	*tp,
2248	xfs_ino_t	inum)
2249{
2250	xfs_mount_t		*mp = free_ip->i_mount;
2251	int			blks_per_cluster;
2252	int			inodes_per_cluster;
 
 
 
2253	int			nbufs;
2254	int			i, j;
2255	xfs_daddr_t		blkno;
2256	xfs_buf_t		*bp;
2257	xfs_inode_t		*ip;
2258	xfs_inode_log_item_t	*iip;
2259	xfs_log_item_t		*lip;
2260	struct xfs_perag	*pag;
2261
2262	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2263	blks_per_cluster = xfs_icluster_size_fsb(mp);
2264	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2265	nbufs = mp->m_ialloc_blks / blks_per_cluster;
 
 
 
 
 
 
 
2266
2267	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2268		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2269					 XFS_INO_TO_AGBNO(mp, inum));
2270
2271		/*
2272		 * We obtain and lock the backing buffer first in the process
2273		 * here, as we have to ensure that any dirty inode that we
2274		 * can't get the flush lock on is attached to the buffer.
 
2275		 * If we scan the in-memory inodes first, then buffer IO can
2276		 * complete before we get a lock on it, and hence we may fail
2277		 * to mark all the active inodes on the buffer stale.
2278		 */
2279		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2280					mp->m_bsize * blks_per_cluster,
2281					XBF_UNMAPPED);
2282
2283		if (!bp)
2284			return ENOMEM;
2285
2286		/*
2287		 * This buffer may not have been correctly initialised as we
2288		 * didn't read it from disk. That's not important because we are
2289		 * only using to mark the buffer as stale in the log, and to
2290		 * attach stale cached inodes on it. That means it will never be
2291		 * dispatched for IO. If it is, we want to know about it, and we
2292		 * want it to fail. We can acheive this by adding a write
2293		 * verifier to the buffer.
2294		 */
2295		 bp->b_ops = &xfs_inode_buf_ops;
2296
2297		/*
2298		 * Walk the inodes already attached to the buffer and mark them
2299		 * stale. These will all have the flush locks held, so an
2300		 * in-memory inode walk can't lock them. By marking them all
2301		 * stale first, we will not attempt to lock them in the loop
2302		 * below as the XFS_ISTALE flag will be set.
2303		 */
2304		lip = bp->b_fspriv;
2305		while (lip) {
2306			if (lip->li_type == XFS_LI_INODE) {
2307				iip = (xfs_inode_log_item_t *)lip;
2308				ASSERT(iip->ili_logged == 1);
2309				lip->li_cb = xfs_istale_done;
2310				xfs_trans_ail_copy_lsn(mp->m_ail,
2311							&iip->ili_flush_lsn,
2312							&iip->ili_item.li_lsn);
2313				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2314			}
2315			lip = lip->li_bio_list;
2316		}
2317
2318
2319		/*
2320		 * For each inode in memory attempt to add it to the inode
2321		 * buffer and set it up for being staled on buffer IO
2322		 * completion.  This is safe as we've locked out tail pushing
2323		 * and flushing by locking the buffer.
2324		 *
2325		 * We have already marked every inode that was part of a
2326		 * transaction stale above, which means there is no point in
2327		 * even trying to lock them.
2328		 */
2329		for (i = 0; i < inodes_per_cluster; i++) {
2330retry:
2331			rcu_read_lock();
2332			ip = radix_tree_lookup(&pag->pag_ici_root,
2333					XFS_INO_TO_AGINO(mp, (inum + i)));
2334
2335			/* Inode not in memory, nothing to do */
2336			if (!ip) {
2337				rcu_read_unlock();
2338				continue;
2339			}
2340
2341			/*
2342			 * because this is an RCU protected lookup, we could
2343			 * find a recently freed or even reallocated inode
2344			 * during the lookup. We need to check under the
2345			 * i_flags_lock for a valid inode here. Skip it if it
2346			 * is not valid, the wrong inode or stale.
2347			 */
2348			spin_lock(&ip->i_flags_lock);
2349			if (ip->i_ino != inum + i ||
2350			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2351				spin_unlock(&ip->i_flags_lock);
2352				rcu_read_unlock();
2353				continue;
2354			}
2355			spin_unlock(&ip->i_flags_lock);
2356
2357			/*
2358			 * Don't try to lock/unlock the current inode, but we
2359			 * _cannot_ skip the other inodes that we did not find
2360			 * in the list attached to the buffer and are not
2361			 * already marked stale. If we can't lock it, back off
2362			 * and retry.
2363			 */
2364			if (ip != free_ip &&
2365			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2366				rcu_read_unlock();
2367				delay(1);
2368				goto retry;
2369			}
2370			rcu_read_unlock();
2371
2372			xfs_iflock(ip);
2373			xfs_iflags_set(ip, XFS_ISTALE);
2374
2375			/*
2376			 * we don't need to attach clean inodes or those only
2377			 * with unlogged changes (which we throw away, anyway).
2378			 */
2379			iip = ip->i_itemp;
2380			if (!iip || xfs_inode_clean(ip)) {
2381				ASSERT(ip != free_ip);
2382				xfs_ifunlock(ip);
2383				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2384				continue;
2385			}
2386
2387			iip->ili_last_fields = iip->ili_fields;
2388			iip->ili_fields = 0;
2389			iip->ili_logged = 1;
2390			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2391						&iip->ili_item.li_lsn);
2392
2393			xfs_buf_attach_iodone(bp, xfs_istale_done,
2394						  &iip->ili_item);
2395
2396			if (ip != free_ip)
2397				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2398		}
2399
2400		xfs_trans_stale_inode_buf(tp, bp);
2401		xfs_trans_binval(tp, bp);
2402	}
2403
2404	xfs_perag_put(pag);
2405	return 0;
2406}
2407
2408/*
2409 * This is called to return an inode to the inode free list.
2410 * The inode should already be truncated to 0 length and have
2411 * no pages associated with it.  This routine also assumes that
2412 * the inode is already a part of the transaction.
2413 *
2414 * The on-disk copy of the inode will have been added to the list
2415 * of unlinked inodes in the AGI. We need to remove the inode from
2416 * that list atomically with respect to freeing it here.
2417 */
2418int
2419xfs_ifree(
2420	xfs_trans_t	*tp,
2421	xfs_inode_t	*ip,
2422	xfs_bmap_free_t	*flist)
2423{
 
 
 
 
2424	int			error;
2425	int			delete;
2426	xfs_ino_t		first_ino;
2427
2428	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2429	ASSERT(ip->i_d.di_nlink == 0);
2430	ASSERT(ip->i_d.di_nextents == 0);
2431	ASSERT(ip->i_d.di_anextents == 0);
2432	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2433	ASSERT(ip->i_d.di_nblocks == 0);
 
2434
2435	/*
2436	 * Pull the on-disk inode from the AGI unlinked list.
 
 
 
2437	 */
2438	error = xfs_iunlink_remove(tp, ip);
2439	if (error)
2440		return error;
2441
2442	error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2443	if (error)
2444		return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2445
2446	ip->i_d.di_mode = 0;		/* mark incore inode as free */
2447	ip->i_d.di_flags = 0;
2448	ip->i_d.di_dmevmask = 0;
2449	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2450	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2451	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2452	/*
2453	 * Bump the generation count so no one will be confused
2454	 * by reincarnations of this inode.
2455	 */
2456	ip->i_d.di_gen++;
2457	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2458
2459	if (delete)
2460		error = xfs_ifree_cluster(ip, tp, first_ino);
2461
 
2462	return error;
2463}
2464
2465/*
2466 * This is called to unpin an inode.  The caller must have the inode locked
2467 * in at least shared mode so that the buffer cannot be subsequently pinned
2468 * once someone is waiting for it to be unpinned.
2469 */
2470static void
2471xfs_iunpin(
2472	struct xfs_inode	*ip)
2473{
2474	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2475
2476	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2477
2478	/* Give the log a push to start the unpinning I/O */
2479	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2480
2481}
2482
2483static void
2484__xfs_iunpin_wait(
2485	struct xfs_inode	*ip)
2486{
2487	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2488	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2489
2490	xfs_iunpin(ip);
2491
2492	do {
2493		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2494		if (xfs_ipincount(ip))
2495			io_schedule();
2496	} while (xfs_ipincount(ip));
2497	finish_wait(wq, &wait.wait);
2498}
2499
2500void
2501xfs_iunpin_wait(
2502	struct xfs_inode	*ip)
2503{
2504	if (xfs_ipincount(ip))
2505		__xfs_iunpin_wait(ip);
2506}
2507
2508/*
2509 * Removing an inode from the namespace involves removing the directory entry
2510 * and dropping the link count on the inode. Removing the directory entry can
2511 * result in locking an AGF (directory blocks were freed) and removing a link
2512 * count can result in placing the inode on an unlinked list which results in
2513 * locking an AGI.
2514 *
2515 * The big problem here is that we have an ordering constraint on AGF and AGI
2516 * locking - inode allocation locks the AGI, then can allocate a new extent for
2517 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2518 * removes the inode from the unlinked list, requiring that we lock the AGI
2519 * first, and then freeing the inode can result in an inode chunk being freed
2520 * and hence freeing disk space requiring that we lock an AGF.
2521 *
2522 * Hence the ordering that is imposed by other parts of the code is AGI before
2523 * AGF. This means we cannot remove the directory entry before we drop the inode
2524 * reference count and put it on the unlinked list as this results in a lock
2525 * order of AGF then AGI, and this can deadlock against inode allocation and
2526 * freeing. Therefore we must drop the link counts before we remove the
2527 * directory entry.
2528 *
2529 * This is still safe from a transactional point of view - it is not until we
2530 * get to xfs_bmap_finish() that we have the possibility of multiple
2531 * transactions in this operation. Hence as long as we remove the directory
2532 * entry and drop the link count in the first transaction of the remove
2533 * operation, there are no transactional constraints on the ordering here.
2534 */
2535int
2536xfs_remove(
2537	xfs_inode_t             *dp,
2538	struct xfs_name		*name,
2539	xfs_inode_t		*ip)
2540{
2541	xfs_mount_t		*mp = dp->i_mount;
2542	xfs_trans_t             *tp = NULL;
2543	int			is_dir = S_ISDIR(ip->i_d.di_mode);
 
2544	int                     error = 0;
2545	xfs_bmap_free_t         free_list;
2546	xfs_fsblock_t           first_block;
2547	int			cancel_flags;
2548	int			committed;
2549	int			link_zero;
2550	uint			resblks;
2551	uint			log_count;
2552
2553	trace_xfs_remove(dp, name);
2554
2555	if (XFS_FORCED_SHUTDOWN(mp))
2556		return XFS_ERROR(EIO);
2557
2558	error = xfs_qm_dqattach(dp, 0);
2559	if (error)
2560		goto std_return;
2561
2562	error = xfs_qm_dqattach(ip, 0);
2563	if (error)
2564		goto std_return;
2565
2566	if (is_dir) {
2567		tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2568		log_count = XFS_DEFAULT_LOG_COUNT;
2569	} else {
2570		tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2571		log_count = XFS_REMOVE_LOG_COUNT;
2572	}
2573	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2574
2575	/*
2576	 * We try to get the real space reservation first,
2577	 * allowing for directory btree deletion(s) implying
2578	 * possible bmap insert(s).  If we can't get the space
2579	 * reservation then we use 0 instead, and avoid the bmap
2580	 * btree insert(s) in the directory code by, if the bmap
2581	 * insert tries to happen, instead trimming the LAST
2582	 * block from the directory.
 
 
2583	 */
2584	resblks = XFS_REMOVE_SPACE_RES(mp);
2585	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2586	if (error == ENOSPC) {
2587		resblks = 0;
2588		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2589	}
2590	if (error) {
2591		ASSERT(error != ENOSPC);
2592		cancel_flags = 0;
2593		goto out_trans_cancel;
2594	}
2595
2596	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2597
2598	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2599	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2600
2601	/*
2602	 * If we're removing a directory perform some additional validation.
2603	 */
2604	cancel_flags |= XFS_TRANS_ABORT;
2605	if (is_dir) {
2606		ASSERT(ip->i_d.di_nlink >= 2);
2607		if (ip->i_d.di_nlink != 2) {
2608			error = XFS_ERROR(ENOTEMPTY);
2609			goto out_trans_cancel;
2610		}
2611		if (!xfs_dir_isempty(ip)) {
2612			error = XFS_ERROR(ENOTEMPTY);
2613			goto out_trans_cancel;
2614		}
2615
2616		/* Drop the link from ip's "..".  */
2617		error = xfs_droplink(tp, dp);
2618		if (error)
2619			goto out_trans_cancel;
2620
2621		/* Drop the "." link from ip to self.  */
2622		error = xfs_droplink(tp, ip);
2623		if (error)
2624			goto out_trans_cancel;
 
 
 
 
 
 
 
 
 
 
 
 
 
2625	} else {
2626		/*
2627		 * When removing a non-directory we need to log the parent
2628		 * inode here.  For a directory this is done implicitly
2629		 * by the xfs_droplink call for the ".." entry.
2630		 */
2631		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2632	}
2633	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2634
2635	/* Drop the link from dp to ip. */
2636	error = xfs_droplink(tp, ip);
2637	if (error)
2638		goto out_trans_cancel;
2639
2640	/* Determine if this is the last link while the inode is locked */
2641	link_zero = (ip->i_d.di_nlink == 0);
2642
2643	xfs_bmap_init(&free_list, &first_block);
2644	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2645					&first_block, &free_list, resblks);
2646	if (error) {
2647		ASSERT(error != ENOENT);
2648		goto out_bmap_cancel;
2649	}
2650
2651	/*
2652	 * If this is a synchronous mount, make sure that the
2653	 * remove transaction goes to disk before returning to
2654	 * the user.
2655	 */
2656	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2657		xfs_trans_set_sync(tp);
2658
2659	error = xfs_bmap_finish(&tp, &free_list, &committed);
2660	if (error)
2661		goto out_bmap_cancel;
2662
2663	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2664	if (error)
2665		goto std_return;
2666
2667	/*
2668	 * If we are using filestreams, kill the stream association.
2669	 * If the file is still open it may get a new one but that
2670	 * will get killed on last close in xfs_close() so we don't
2671	 * have to worry about that.
2672	 */
2673	if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2674		xfs_filestream_deassociate(ip);
2675
2676	return 0;
2677
2678 out_bmap_cancel:
2679	xfs_bmap_cancel(&free_list);
2680 out_trans_cancel:
2681	xfs_trans_cancel(tp, cancel_flags);
2682 std_return:
2683	return error;
2684}
2685
2686/*
2687 * Enter all inodes for a rename transaction into a sorted array.
2688 */
 
2689STATIC void
2690xfs_sort_for_rename(
2691	xfs_inode_t	*dp1,	/* in: old (source) directory inode */
2692	xfs_inode_t	*dp2,	/* in: new (target) directory inode */
2693	xfs_inode_t	*ip1,	/* in: inode of old entry */
2694	xfs_inode_t	*ip2,	/* in: inode of new entry, if it
2695				   already exists, NULL otherwise. */
2696	xfs_inode_t	**i_tab,/* out: array of inode returned, sorted */
2697	int		*num_inodes)  /* out: number of inodes in array */
2698{
2699	xfs_inode_t		*temp;
2700	int			i, j;
2701
 
 
 
2702	/*
2703	 * i_tab contains a list of pointers to inodes.  We initialize
2704	 * the table here & we'll sort it.  We will then use it to
2705	 * order the acquisition of the inode locks.
2706	 *
2707	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2708	 */
2709	i_tab[0] = dp1;
2710	i_tab[1] = dp2;
2711	i_tab[2] = ip1;
2712	if (ip2) {
2713		*num_inodes = 4;
2714		i_tab[3] = ip2;
2715	} else {
2716		*num_inodes = 3;
2717		i_tab[3] = NULL;
2718	}
2719
2720	/*
2721	 * Sort the elements via bubble sort.  (Remember, there are at
2722	 * most 4 elements to sort, so this is adequate.)
2723	 */
2724	for (i = 0; i < *num_inodes; i++) {
2725		for (j = 1; j < *num_inodes; j++) {
2726			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2727				temp = i_tab[j];
2728				i_tab[j] = i_tab[j-1];
2729				i_tab[j-1] = temp;
2730			}
2731		}
2732	}
2733}
2734
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2735/*
2736 * xfs_rename
2737 */
2738int
2739xfs_rename(
2740	xfs_inode_t	*src_dp,
2741	struct xfs_name	*src_name,
2742	xfs_inode_t	*src_ip,
2743	xfs_inode_t	*target_dp,
2744	struct xfs_name	*target_name,
2745	xfs_inode_t	*target_ip)
2746{
2747	xfs_trans_t	*tp = NULL;
2748	xfs_mount_t	*mp = src_dp->i_mount;
2749	int		new_parent;		/* moving to a new dir */
2750	int		src_is_directory;	/* src_name is a directory */
2751	int		error;
2752	xfs_bmap_free_t free_list;
2753	xfs_fsblock_t   first_block;
2754	int		cancel_flags;
2755	int		committed;
2756	xfs_inode_t	*inodes[4];
2757	int		spaceres;
2758	int		num_inodes;
 
2759
2760	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2761
2762	new_parent = (src_dp != target_dp);
2763	src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2764
2765	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2766				inodes, &num_inodes);
2767
2768	xfs_bmap_init(&free_list, &first_block);
2769	tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2770	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2771	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2772	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2773	if (error == ENOSPC) {
 
2774		spaceres = 0;
2775		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2776	}
2777	if (error) {
2778		xfs_trans_cancel(tp, 0);
2779		goto std_return;
2780	}
 
 
2781
2782	/*
2783	 * Attach the dquots to the inodes
2784	 */
2785	error = xfs_qm_vop_rename_dqattach(inodes);
2786	if (error) {
2787		xfs_trans_cancel(tp, cancel_flags);
2788		goto std_return;
2789	}
2790
2791	/*
2792	 * Lock all the participating inodes. Depending upon whether
2793	 * the target_name exists in the target directory, and
2794	 * whether the target directory is the same as the source
2795	 * directory, we can lock from 2 to 4 inodes.
2796	 */
2797	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2798
2799	/*
2800	 * Join all the inodes to the transaction. From this point on,
2801	 * we can rely on either trans_commit or trans_cancel to unlock
2802	 * them.
2803	 */
2804	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2805	if (new_parent)
2806		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2807	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2808	if (target_ip)
2809		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
 
 
2810
2811	/*
2812	 * If we are using project inheritance, we only allow renames
2813	 * into our tree when the project IDs are the same; else the
2814	 * tree quota mechanism would be circumvented.
2815	 */
2816	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2817		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2818		error = XFS_ERROR(EXDEV);
2819		goto error_return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2820	}
2821
2822	/*
2823	 * Set up the target.
 
2824	 */
2825	if (target_ip == NULL) {
2826		/*
2827		 * If there's no space reservation, check the entry will
2828		 * fit before actually inserting it.
2829		 */
2830		error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2831		if (error)
2832			goto error_return;
 
 
 
 
 
 
 
 
 
2833		/*
2834		 * If target does not exist and the rename crosses
2835		 * directories, adjust the target directory link count
2836		 * to account for the ".." reference from the new entry.
2837		 */
2838		error = xfs_dir_createname(tp, target_dp, target_name,
2839						src_ip->i_ino, &first_block,
2840						&free_list, spaceres);
2841		if (error == ENOSPC)
2842			goto error_return;
2843		if (error)
2844			goto abort_return;
2845
2846		xfs_trans_ichgtime(tp, target_dp,
2847					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2848
2849		if (new_parent && src_is_directory) {
2850			error = xfs_bumplink(tp, target_dp);
2851			if (error)
2852				goto abort_return;
2853		}
2854	} else { /* target_ip != NULL */
2855		/*
2856		 * If target exists and it's a directory, check that both
2857		 * target and source are directories and that target can be
2858		 * destroyed, or that neither is a directory.
2859		 */
2860		if (S_ISDIR(target_ip->i_d.di_mode)) {
2861			/*
2862			 * Make sure target dir is empty.
2863			 */
2864			if (!(xfs_dir_isempty(target_ip)) ||
2865			    (target_ip->i_d.di_nlink > 2)) {
2866				error = XFS_ERROR(EEXIST);
2867				goto error_return;
2868			}
2869		}
2870
2871		/*
2872		 * Link the source inode under the target name.
2873		 * If the source inode is a directory and we are moving
2874		 * it across directories, its ".." entry will be
2875		 * inconsistent until we replace that down below.
2876		 *
2877		 * In case there is already an entry with the same
2878		 * name at the destination directory, remove it first.
2879		 */
2880		error = xfs_dir_replace(tp, target_dp, target_name,
2881					src_ip->i_ino,
2882					&first_block, &free_list, spaceres);
2883		if (error)
2884			goto abort_return;
2885
2886		xfs_trans_ichgtime(tp, target_dp,
2887					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2888
2889		/*
2890		 * Decrement the link count on the target since the target
2891		 * dir no longer points to it.
2892		 */
2893		error = xfs_droplink(tp, target_ip);
2894		if (error)
2895			goto abort_return;
2896
2897		if (src_is_directory) {
2898			/*
2899			 * Drop the link from the old "." entry.
2900			 */
2901			error = xfs_droplink(tp, target_ip);
2902			if (error)
2903				goto abort_return;
2904		}
2905	} /* target_ip != NULL */
2906
2907	/*
2908	 * Remove the source.
2909	 */
2910	if (new_parent && src_is_directory) {
2911		/*
2912		 * Rewrite the ".." entry to point to the new
2913		 * directory.
2914		 */
2915		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2916					target_dp->i_ino,
2917					&first_block, &free_list, spaceres);
2918		ASSERT(error != EEXIST);
2919		if (error)
2920			goto abort_return;
2921	}
2922
2923	/*
2924	 * We always want to hit the ctime on the source inode.
2925	 *
2926	 * This isn't strictly required by the standards since the source
2927	 * inode isn't really being changed, but old unix file systems did
2928	 * it and some incremental backup programs won't work without it.
2929	 */
2930	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2931	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2932
2933	/*
2934	 * Adjust the link count on src_dp.  This is necessary when
2935	 * renaming a directory, either within one parent when
2936	 * the target existed, or across two parent directories.
2937	 */
2938	if (src_is_directory && (new_parent || target_ip != NULL)) {
2939
2940		/*
2941		 * Decrement link count on src_directory since the
2942		 * entry that's moved no longer points to it.
2943		 */
2944		error = xfs_droplink(tp, src_dp);
2945		if (error)
2946			goto abort_return;
2947	}
2948
2949	error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2950					&first_block, &free_list, spaceres);
 
 
 
 
 
 
 
 
 
 
2951	if (error)
2952		goto abort_return;
2953
2954	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2955	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2956	if (new_parent)
2957		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2958
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2959	/*
2960	 * If this is a synchronous mount, make sure that the
2961	 * rename transaction goes to disk before returning to
2962	 * the user.
 
2963	 */
2964	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2965		xfs_trans_set_sync(tp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2966	}
2967
2968	error = xfs_bmap_finish(&tp, &free_list, &committed);
2969	if (error) {
2970		xfs_bmap_cancel(&free_list);
2971		xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2972				 XFS_TRANS_ABORT));
2973		goto std_return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2974	}
2975
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2976	/*
2977	 * trans_commit will unlock src_ip, target_ip & decrement
2978	 * the vnode references.
2979	 */
2980	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 
2981
2982 abort_return:
2983	cancel_flags |= XFS_TRANS_ABORT;
2984 error_return:
2985	xfs_bmap_cancel(&free_list);
2986	xfs_trans_cancel(tp, cancel_flags);
2987 std_return:
2988	return error;
2989}
2990
2991STATIC int
 
 
 
 
 
 
 
 
 
 
 
 
 
2992xfs_iflush_cluster(
2993	xfs_inode_t	*ip,
2994	xfs_buf_t	*bp)
2995{
2996	xfs_mount_t		*mp = ip->i_mount;
2997	struct xfs_perag	*pag;
2998	unsigned long		first_index, mask;
2999	unsigned long		inodes_per_cluster;
3000	int			ilist_size;
3001	xfs_inode_t		**ilist;
3002	xfs_inode_t		*iq;
3003	int			nr_found;
3004	int			clcount = 0;
3005	int			bufwasdelwri;
3006	int			i;
3007
3008	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 
 
 
 
 
 
3009
3010	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3011	ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3012	ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
3013	if (!ilist)
3014		goto out_put;
3015
3016	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3017	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3018	rcu_read_lock();
3019	/* really need a gang lookup range call here */
3020	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
3021					first_index, inodes_per_cluster);
3022	if (nr_found == 0)
3023		goto out_free;
3024
3025	for (i = 0; i < nr_found; i++) {
3026		iq = ilist[i];
3027		if (iq == ip)
3028			continue;
3029
3030		/*
3031		 * because this is an RCU protected lookup, we could find a
3032		 * recently freed or even reallocated inode during the lookup.
3033		 * We need to check under the i_flags_lock for a valid inode
3034		 * here. Skip it if it is not valid or the wrong inode.
 
3035		 */
3036		spin_lock(&ip->i_flags_lock);
3037		if (!ip->i_ino ||
3038		    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3039			spin_unlock(&ip->i_flags_lock);
3040			continue;
3041		}
3042		spin_unlock(&ip->i_flags_lock);
3043
3044		/*
3045		 * Do an un-protected check to see if the inode is dirty and
3046		 * is a candidate for flushing.  These checks will be repeated
3047		 * later after the appropriate locks are acquired.
 
3048		 */
3049		if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
 
3050			continue;
 
 
 
3051
3052		/*
3053		 * Try to get locks.  If any are unavailable or it is pinned,
3054		 * then this inode cannot be flushed and is skipped.
3055		 */
3056
3057		if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3058			continue;
3059		if (!xfs_iflock_nowait(iq)) {
3060			xfs_iunlock(iq, XFS_ILOCK_SHARED);
 
 
 
3061			continue;
3062		}
3063		if (xfs_ipincount(iq)) {
3064			xfs_ifunlock(iq);
3065			xfs_iunlock(iq, XFS_ILOCK_SHARED);
 
 
3066			continue;
3067		}
3068
 
 
 
 
 
 
 
 
 
 
 
3069		/*
3070		 * arriving here means that this inode can be flushed.  First
3071		 * re-check that it's dirty before flushing.
3072		 */
3073		if (!xfs_inode_clean(iq)) {
3074			int	error;
3075			error = xfs_iflush_int(iq, bp);
3076			if (error) {
3077				xfs_iunlock(iq, XFS_ILOCK_SHARED);
3078				goto cluster_corrupt_out;
3079			}
3080			clcount++;
3081		} else {
3082			xfs_ifunlock(iq);
3083		}
3084		xfs_iunlock(iq, XFS_ILOCK_SHARED);
3085	}
3086
3087	if (clcount) {
3088		XFS_STATS_INC(xs_icluster_flushcnt);
3089		XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3090	}
3091
3092out_free:
3093	rcu_read_unlock();
3094	kmem_free(ilist);
3095out_put:
3096	xfs_perag_put(pag);
3097	return 0;
3098
 
3099
3100cluster_corrupt_out:
3101	/*
3102	 * Corruption detected in the clustering loop.  Invalidate the
3103	 * inode buffer and shut down the filesystem.
3104	 */
3105	rcu_read_unlock();
3106	/*
3107	 * Clean up the buffer.  If it was delwri, just release it --
3108	 * brelse can handle it with no problems.  If not, shut down the
3109	 * filesystem before releasing the buffer.
3110	 */
3111	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3112	if (bufwasdelwri)
3113		xfs_buf_relse(bp);
3114
3115	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3116
3117	if (!bufwasdelwri) {
3118		/*
3119		 * Just like incore_relse: if we have b_iodone functions,
3120		 * mark the buffer as an error and call them.  Otherwise
3121		 * mark it as stale and brelse.
3122		 */
3123		if (bp->b_iodone) {
3124			XFS_BUF_UNDONE(bp);
3125			xfs_buf_stale(bp);
3126			xfs_buf_ioerror(bp, EIO);
3127			xfs_buf_ioend(bp, 0);
3128		} else {
3129			xfs_buf_stale(bp);
3130			xfs_buf_relse(bp);
3131		}
3132	}
3133
3134	/*
3135	 * Unlocks the flush lock
3136	 */
3137	xfs_iflush_abort(iq, false);
3138	kmem_free(ilist);
3139	xfs_perag_put(pag);
3140	return XFS_ERROR(EFSCORRUPTED);
3141}
3142
3143/*
3144 * Flush dirty inode metadata into the backing buffer.
3145 *
3146 * The caller must have the inode lock and the inode flush lock held.  The
3147 * inode lock will still be held upon return to the caller, and the inode
3148 * flush lock will be released after the inode has reached the disk.
3149 *
3150 * The caller must write out the buffer returned in *bpp and release it.
3151 */
3152int
3153xfs_iflush(
3154	struct xfs_inode	*ip,
3155	struct xfs_buf		**bpp)
3156{
3157	struct xfs_mount	*mp = ip->i_mount;
3158	struct xfs_buf		*bp;
3159	struct xfs_dinode	*dip;
3160	int			error;
3161
3162	XFS_STATS_INC(xs_iflush_count);
 
 
 
3163
3164	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3165	ASSERT(xfs_isiflocked(ip));
3166	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3167	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3168
3169	*bpp = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
3170
3171	xfs_iunpin_wait(ip);
 
3172
3173	/*
3174	 * For stale inodes we cannot rely on the backing buffer remaining
3175	 * stale in cache for the remaining life of the stale inode and so
3176	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3177	 * inodes below. We have to check this after ensuring the inode is
3178	 * unpinned so that it is safe to reclaim the stale inode after the
3179	 * flush call.
3180	 */
3181	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3182		xfs_ifunlock(ip);
3183		return 0;
3184	}
3185
3186	/*
3187	 * This may have been unpinned because the filesystem is shutting
3188	 * down forcibly. If that's the case we must not write this inode
3189	 * to disk, because the log record didn't make it to disk.
3190	 *
3191	 * We also have to remove the log item from the AIL in this case,
3192	 * as we wait for an empty AIL as part of the unmount process.
3193	 */
3194	if (XFS_FORCED_SHUTDOWN(mp)) {
3195		error = XFS_ERROR(EIO);
3196		goto abort_out;
3197	}
3198
3199	/*
3200	 * Get the buffer containing the on-disk inode.
3201	 */
3202	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3203			       0);
3204	if (error || !bp) {
3205		xfs_ifunlock(ip);
 
 
 
 
3206		return error;
3207	}
3208
3209	/*
3210	 * First flush out the inode that xfs_iflush was called with.
3211	 */
3212	error = xfs_iflush_int(ip, bp);
3213	if (error)
3214		goto corrupt_out;
3215
3216	/*
3217	 * If the buffer is pinned then push on the log now so we won't
3218	 * get stuck waiting in the write for too long.
3219	 */
3220	if (xfs_buf_ispinned(bp))
3221		xfs_log_force(mp, 0);
3222
3223	/*
3224	 * inode clustering:
3225	 * see if other inodes can be gathered into this write
3226	 */
3227	error = xfs_iflush_cluster(ip, bp);
3228	if (error)
3229		goto cluster_corrupt_out;
3230
3231	*bpp = bp;
3232	return 0;
3233
3234corrupt_out:
3235	xfs_buf_relse(bp);
3236	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3237cluster_corrupt_out:
3238	error = XFS_ERROR(EFSCORRUPTED);
3239abort_out:
3240	/*
3241	 * Unlocks the flush lock
3242	 */
3243	xfs_iflush_abort(ip, false);
3244	return error;
3245}
3246
3247STATIC int
3248xfs_iflush_int(
3249	struct xfs_inode	*ip,
3250	struct xfs_buf		*bp)
3251{
3252	struct xfs_inode_log_item *iip = ip->i_itemp;
3253	struct xfs_dinode	*dip;
3254	struct xfs_mount	*mp = ip->i_mount;
3255
3256	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3257	ASSERT(xfs_isiflocked(ip));
3258	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3259	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3260	ASSERT(iip != NULL && iip->ili_fields != 0);
3261
3262	/* set *dip = inode's place in the buffer */
3263	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3264
3265	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3266			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3267		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3268			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3269			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3270		goto corrupt_out;
3271	}
3272	if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3273				mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3274		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3275			"%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3276			__func__, ip->i_ino, ip, ip->i_d.di_magic);
3277		goto corrupt_out;
3278	}
3279	if (S_ISREG(ip->i_d.di_mode)) {
3280		if (XFS_TEST_ERROR(
3281		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3282		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3283		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3284			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3285				"%s: Bad regular inode %Lu, ptr 0x%p",
3286				__func__, ip->i_ino, ip);
3287			goto corrupt_out;
3288		}
3289	} else if (S_ISDIR(ip->i_d.di_mode)) {
3290		if (XFS_TEST_ERROR(
3291		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3292		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3293		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3294		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3295			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3296				"%s: Bad directory inode %Lu, ptr 0x%p",
3297				__func__, ip->i_ino, ip);
3298			goto corrupt_out;
3299		}
3300	}
3301	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3302				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3303				XFS_RANDOM_IFLUSH_5)) {
3304		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3305			"%s: detected corrupt incore inode %Lu, "
3306			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3307			__func__, ip->i_ino,
3308			ip->i_d.di_nextents + ip->i_d.di_anextents,
3309			ip->i_d.di_nblocks, ip);
3310		goto corrupt_out;
3311	}
3312	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3313				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3314		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3315			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3316			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3317		goto corrupt_out;
3318	}
3319
3320	/*
3321	 * Inode item log recovery for v1/v2 inodes are dependent on the
3322	 * di_flushiter count for correct sequencing. We bump the flush
3323	 * iteration count so we can detect flushes which postdate a log record
3324	 * during recovery. This is redundant as we now log every change and
3325	 * hence this can't happen but we need to still do it to ensure
3326	 * backwards compatibility with old kernels that predate logging all
3327	 * inode changes.
3328	 */
3329	if (ip->i_d.di_version < 3)
3330		ip->i_d.di_flushiter++;
3331
3332	/*
3333	 * Copy the dirty parts of the inode into the on-disk
3334	 * inode.  We always copy out the core of the inode,
3335	 * because if the inode is dirty at all the core must
3336	 * be.
3337	 */
3338	xfs_dinode_to_disk(dip, &ip->i_d);
3339
3340	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3341	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3342		ip->i_d.di_flushiter = 0;
3343
 
 
3344	/*
3345	 * If this is really an old format inode and the superblock version
3346	 * has not been updated to support only new format inodes, then
3347	 * convert back to the old inode format.  If the superblock version
3348	 * has been updated, then make the conversion permanent.
3349	 */
3350	ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3351	if (ip->i_d.di_version == 1) {
3352		if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3353			/*
3354			 * Convert it back.
3355			 */
3356			ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3357			dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3358		} else {
3359			/*
3360			 * The superblock version has already been bumped,
3361			 * so just make the conversion to the new inode
3362			 * format permanent.
3363			 */
3364			ip->i_d.di_version = 2;
3365			dip->di_version = 2;
3366			ip->i_d.di_onlink = 0;
3367			dip->di_onlink = 0;
3368			memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3369			memset(&(dip->di_pad[0]), 0,
3370			      sizeof(dip->di_pad));
3371			ASSERT(xfs_get_projid(ip) == 0);
3372		}
3373	}
3374
3375	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3376	if (XFS_IFORK_Q(ip))
3377		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3378	xfs_inobp_check(mp, bp);
3379
3380	/*
3381	 * We've recorded everything logged in the inode, so we'd like to clear
3382	 * the ili_fields bits so we don't log and flush things unnecessarily.
3383	 * However, we can't stop logging all this information until the data
3384	 * we've copied into the disk buffer is written to disk.  If we did we
3385	 * might overwrite the copy of the inode in the log with all the data
3386	 * after re-logging only part of it, and in the face of a crash we
3387	 * wouldn't have all the data we need to recover.
3388	 *
3389	 * What we do is move the bits to the ili_last_fields field.  When
3390	 * logging the inode, these bits are moved back to the ili_fields field.
3391	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3392	 * know that the information those bits represent is permanently on
3393	 * disk.  As long as the flush completes before the inode is logged
3394	 * again, then both ili_fields and ili_last_fields will be cleared.
3395	 *
3396	 * We can play with the ili_fields bits here, because the inode lock
3397	 * must be held exclusively in order to set bits there and the flush
3398	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3399	 * done routine can tell whether or not to look in the AIL.  Also, store
3400	 * the current LSN of the inode so that we can tell whether the item has
3401	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3402	 * need the AIL lock, because it is a 64 bit value that cannot be read
3403	 * atomically.
3404	 */
3405	iip->ili_last_fields = iip->ili_fields;
3406	iip->ili_fields = 0;
3407	iip->ili_logged = 1;
3408
3409	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3410				&iip->ili_item.li_lsn);
3411
3412	/*
3413	 * Attach the function xfs_iflush_done to the inode's
3414	 * buffer.  This will remove the inode from the AIL
3415	 * and unlock the inode's flush lock when the inode is
3416	 * completely written to disk.
3417	 */
3418	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3419
3420	/* update the lsn in the on disk inode if required */
3421	if (ip->i_d.di_version == 3)
3422		dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3423
3424	/* generate the checksum. */
3425	xfs_dinode_calc_crc(mp, dip);
3426
3427	ASSERT(bp->b_fspriv != NULL);
3428	ASSERT(bp->b_iodone != NULL);
3429	return 0;
 
3430
3431corrupt_out:
3432	return XFS_ERROR(EFSCORRUPTED);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3433}