Linux Audio

Check our new training course

Loading...
v5.9
   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_sb.h"
  15#include "xfs_mount.h"
  16#include "xfs_defer.h"
  17#include "xfs_inode.h"
 
 
  18#include "xfs_dir2.h"
 
  19#include "xfs_attr.h"
  20#include "xfs_trans_space.h"
  21#include "xfs_trans.h"
  22#include "xfs_buf_item.h"
  23#include "xfs_inode_item.h"
  24#include "xfs_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
  39kmem_zone_t *xfs_inode_zone;
  40
  41/*
  42 * Used in xfs_itruncate_extents().  This is the maximum number of extents
  43 * freed from a file in a single transaction.
  44 */
  45#define	XFS_ITRUNC_MAX_EXTENTS	2
  46
 
  47STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
  48STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
  49
  50/*
  51 * helper function to extract extent size hint from inode
  52 */
  53xfs_extlen_t
  54xfs_get_extsz_hint(
  55	struct xfs_inode	*ip)
  56{
  57	/*
  58	 * No point in aligning allocations if we need to COW to actually
  59	 * write to them.
  60	 */
  61	if (xfs_is_always_cow_inode(ip))
  62		return 0;
  63	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
  64		return ip->i_d.di_extsize;
  65	if (XFS_IS_REALTIME_INODE(ip))
  66		return ip->i_mount->m_sb.sb_rextsize;
  67	return 0;
  68}
  69
  70/*
  71 * Helper function to extract CoW extent size hint from inode.
  72 * Between the extent size hint and the CoW extent size hint, we
  73 * return the greater of the two.  If the value is zero (automatic),
  74 * use the default size.
  75 */
  76xfs_extlen_t
  77xfs_get_cowextsz_hint(
  78	struct xfs_inode	*ip)
  79{
  80	xfs_extlen_t		a, b;
  81
  82	a = 0;
  83	if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
  84		a = ip->i_d.di_cowextsize;
  85	b = xfs_get_extsz_hint(ip);
  86
  87	a = max(a, b);
  88	if (a == 0)
  89		return XFS_DEFAULT_COWEXTSZ_HINT;
  90	return a;
  91}
  92
  93/*
  94 * These two are wrapper routines around the xfs_ilock() routine used to
  95 * centralize some grungy code.  They are used in places that wish to lock the
  96 * inode solely for reading the extents.  The reason these places can't just
  97 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
  98 * bringing in of the extents from disk for a file in b-tree format.  If the
  99 * inode is in b-tree format, then we need to lock the inode exclusively until
 100 * the extents are read in.  Locking it exclusively all the time would limit
 101 * our parallelism unnecessarily, though.  What we do instead is check to see
 102 * if the extents have been read in yet, and only lock the inode exclusively
 103 * if they have not.
 104 *
 105 * The functions return a value which should be given to the corresponding
 106 * xfs_iunlock() call.
 107 */
 108uint
 109xfs_ilock_data_map_shared(
 110	struct xfs_inode	*ip)
 111{
 112	uint			lock_mode = XFS_ILOCK_SHARED;
 113
 114	if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE &&
 115	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
 116		lock_mode = XFS_ILOCK_EXCL;
 117	xfs_ilock(ip, lock_mode);
 118	return lock_mode;
 119}
 120
 121uint
 122xfs_ilock_attr_map_shared(
 123	struct xfs_inode	*ip)
 124{
 125	uint			lock_mode = XFS_ILOCK_SHARED;
 126
 127	if (ip->i_afp &&
 128	    ip->i_afp->if_format == XFS_DINODE_FMT_BTREE &&
 129	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
 130		lock_mode = XFS_ILOCK_EXCL;
 131	xfs_ilock(ip, lock_mode);
 132	return lock_mode;
 133}
 134
 135/*
 136 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
 137 * multi-reader locks: i_mmap_lock and the i_lock.  This routine allows
 138 * various combinations of the locks to be obtained.
 139 *
 140 * The 3 locks should always be ordered so that the IO lock is obtained first,
 141 * the mmap lock second and the ilock last in order to prevent deadlock.
 142 *
 143 * Basic locking order:
 144 *
 145 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
 146 *
 147 * mmap_lock locking order:
 148 *
 149 * i_rwsem -> page lock -> mmap_lock
 150 * mmap_lock -> i_mmap_lock -> page_lock
 151 *
 152 * The difference in mmap_lock locking order mean that we cannot hold the
 153 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
 154 * fault in pages during copy in/out (for buffered IO) or require the mmap_lock
 155 * in get_user_pages() to map the user pages into the kernel address space for
 156 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
 157 * page faults already hold the mmap_lock.
 158 *
 159 * Hence to serialise fully against both syscall and mmap based IO, we need to
 160 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
 161 * taken in places where we need to invalidate the page cache in a race
 162 * free manner (e.g. truncate, hole punch and other extent manipulation
 163 * functions).
 164 */
 165void
 166xfs_ilock(
 167	xfs_inode_t		*ip,
 168	uint			lock_flags)
 169{
 170	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
 171
 172	/*
 173	 * You can't set both SHARED and EXCL for the same lock,
 174	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 175	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 176	 */
 177	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 178	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 179	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 180	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 181	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 182	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 183	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 184
 185	if (lock_flags & XFS_IOLOCK_EXCL) {
 186		down_write_nested(&VFS_I(ip)->i_rwsem,
 187				  XFS_IOLOCK_DEP(lock_flags));
 188	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 189		down_read_nested(&VFS_I(ip)->i_rwsem,
 190				 XFS_IOLOCK_DEP(lock_flags));
 191	}
 192
 193	if (lock_flags & XFS_MMAPLOCK_EXCL)
 194		mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
 195	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 196		mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
 197
 198	if (lock_flags & XFS_ILOCK_EXCL)
 199		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 200	else if (lock_flags & XFS_ILOCK_SHARED)
 201		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 202}
 203
 204/*
 205 * This is just like xfs_ilock(), except that the caller
 206 * is guaranteed not to sleep.  It returns 1 if it gets
 207 * the requested locks and 0 otherwise.  If the IO lock is
 208 * obtained but the inode lock cannot be, then the IO lock
 209 * is dropped before returning.
 210 *
 211 * ip -- the inode being locked
 212 * lock_flags -- this parameter indicates the inode's locks to be
 213 *       to be locked.  See the comment for xfs_ilock() for a list
 214 *	 of valid values.
 215 */
 216int
 217xfs_ilock_nowait(
 218	xfs_inode_t		*ip,
 219	uint			lock_flags)
 220{
 221	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
 222
 223	/*
 224	 * You can't set both SHARED and EXCL for the same lock,
 225	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 226	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 227	 */
 228	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 229	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 230	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 231	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 232	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 233	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 234	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 235
 236	if (lock_flags & XFS_IOLOCK_EXCL) {
 237		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
 238			goto out;
 239	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 240		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
 241			goto out;
 242	}
 243
 244	if (lock_flags & XFS_MMAPLOCK_EXCL) {
 245		if (!mrtryupdate(&ip->i_mmaplock))
 246			goto out_undo_iolock;
 247	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
 248		if (!mrtryaccess(&ip->i_mmaplock))
 249			goto out_undo_iolock;
 250	}
 251
 252	if (lock_flags & XFS_ILOCK_EXCL) {
 253		if (!mrtryupdate(&ip->i_lock))
 254			goto out_undo_mmaplock;
 255	} else if (lock_flags & XFS_ILOCK_SHARED) {
 256		if (!mrtryaccess(&ip->i_lock))
 257			goto out_undo_mmaplock;
 258	}
 259	return 1;
 260
 261out_undo_mmaplock:
 262	if (lock_flags & XFS_MMAPLOCK_EXCL)
 263		mrunlock_excl(&ip->i_mmaplock);
 264	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 265		mrunlock_shared(&ip->i_mmaplock);
 266out_undo_iolock:
 267	if (lock_flags & XFS_IOLOCK_EXCL)
 268		up_write(&VFS_I(ip)->i_rwsem);
 269	else if (lock_flags & XFS_IOLOCK_SHARED)
 270		up_read(&VFS_I(ip)->i_rwsem);
 271out:
 272	return 0;
 273}
 274
 275/*
 276 * xfs_iunlock() is used to drop the inode locks acquired with
 277 * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
 278 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
 279 * that we know which locks to drop.
 280 *
 281 * ip -- the inode being unlocked
 282 * lock_flags -- this parameter indicates the inode's locks to be
 283 *       to be unlocked.  See the comment for xfs_ilock() for a list
 284 *	 of valid values for this parameter.
 285 *
 286 */
 287void
 288xfs_iunlock(
 289	xfs_inode_t		*ip,
 290	uint			lock_flags)
 291{
 292	/*
 293	 * You can't set both SHARED and EXCL for the same lock,
 294	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 295	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 296	 */
 297	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 298	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 299	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 300	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 301	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 302	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 303	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 304	ASSERT(lock_flags != 0);
 305
 306	if (lock_flags & XFS_IOLOCK_EXCL)
 307		up_write(&VFS_I(ip)->i_rwsem);
 308	else if (lock_flags & XFS_IOLOCK_SHARED)
 309		up_read(&VFS_I(ip)->i_rwsem);
 310
 311	if (lock_flags & XFS_MMAPLOCK_EXCL)
 312		mrunlock_excl(&ip->i_mmaplock);
 313	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 314		mrunlock_shared(&ip->i_mmaplock);
 315
 316	if (lock_flags & XFS_ILOCK_EXCL)
 317		mrunlock_excl(&ip->i_lock);
 318	else if (lock_flags & XFS_ILOCK_SHARED)
 319		mrunlock_shared(&ip->i_lock);
 320
 321	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
 322}
 323
 324/*
 325 * give up write locks.  the i/o lock cannot be held nested
 326 * if it is being demoted.
 327 */
 328void
 329xfs_ilock_demote(
 330	xfs_inode_t		*ip,
 331	uint			lock_flags)
 332{
 333	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
 334	ASSERT((lock_flags &
 335		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
 336
 337	if (lock_flags & XFS_ILOCK_EXCL)
 338		mrdemote(&ip->i_lock);
 339	if (lock_flags & XFS_MMAPLOCK_EXCL)
 340		mrdemote(&ip->i_mmaplock);
 341	if (lock_flags & XFS_IOLOCK_EXCL)
 342		downgrade_write(&VFS_I(ip)->i_rwsem);
 343
 344	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
 345}
 346
 347#if defined(DEBUG) || defined(XFS_WARN)
 348int
 349xfs_isilocked(
 350	xfs_inode_t		*ip,
 351	uint			lock_flags)
 352{
 353	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
 354		if (!(lock_flags & XFS_ILOCK_SHARED))
 355			return !!ip->i_lock.mr_writer;
 356		return rwsem_is_locked(&ip->i_lock.mr_lock);
 357	}
 358
 359	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
 360		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
 361			return !!ip->i_mmaplock.mr_writer;
 362		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
 363	}
 364
 365	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
 366		if (!(lock_flags & XFS_IOLOCK_SHARED))
 367			return !debug_locks ||
 368				lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
 369		return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
 370	}
 371
 372	ASSERT(0);
 373	return 0;
 374}
 375#endif
 376
 
 
 
 
 
 
 
 
 377/*
 378 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
 379 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
 380 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
 381 * errors and warnings.
 382 */
 383#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
 384static bool
 385xfs_lockdep_subclass_ok(
 386	int subclass)
 387{
 388	return subclass < MAX_LOCKDEP_SUBCLASSES;
 389}
 390#else
 391#define xfs_lockdep_subclass_ok(subclass)	(true)
 392#endif
 393
 394/*
 395 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
 396 * value. This can be called for any type of inode lock combination, including
 397 * parent locking. Care must be taken to ensure we don't overrun the subclass
 398 * storage fields in the class mask we build.
 399 */
 400static inline int
 401xfs_lock_inumorder(int lock_mode, int subclass)
 402{
 403	int	class = 0;
 404
 405	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
 406			      XFS_ILOCK_RTSUM)));
 407	ASSERT(xfs_lockdep_subclass_ok(subclass));
 408
 409	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
 410		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
 411		class += subclass << XFS_IOLOCK_SHIFT;
 412	}
 413
 414	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
 415		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
 416		class += subclass << XFS_MMAPLOCK_SHIFT;
 417	}
 418
 419	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
 420		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
 421		class += subclass << XFS_ILOCK_SHIFT;
 422	}
 423
 424	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
 425}
 426
 427/*
 428 * The following routine will lock n inodes in exclusive mode.  We assume the
 429 * caller calls us with the inodes in i_ino order.
 430 *
 431 * We need to detect deadlock where an inode that we lock is in the AIL and we
 432 * start waiting for another inode that is locked by a thread in a long running
 433 * transaction (such as truncate). This can result in deadlock since the long
 434 * running trans might need to wait for the inode we just locked in order to
 435 * push the tail and free space in the log.
 436 *
 437 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
 438 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
 439 * lock more than one at a time, lockdep will report false positives saying we
 440 * have violated locking orders.
 441 */
 442static void
 443xfs_lock_inodes(
 444	struct xfs_inode	**ips,
 445	int			inodes,
 446	uint			lock_mode)
 447{
 448	int			attempts = 0, i, j, try_lock;
 449	struct xfs_log_item	*lp;
 450
 451	/*
 452	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
 453	 * support an arbitrary depth of locking here, but absolute limits on
 454	 * inodes depend on the type of locking and the limits placed by
 455	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
 456	 * the asserts.
 457	 */
 458	ASSERT(ips && inodes >= 2 && inodes <= 5);
 459	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
 460			    XFS_ILOCK_EXCL));
 461	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
 462			      XFS_ILOCK_SHARED)));
 463	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
 464		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
 465	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
 466		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
 467
 468	if (lock_mode & XFS_IOLOCK_EXCL) {
 469		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
 470	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
 471		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
 472
 473	try_lock = 0;
 474	i = 0;
 475again:
 476	for (; i < inodes; i++) {
 477		ASSERT(ips[i]);
 478
 479		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
 480			continue;
 481
 482		/*
 483		 * If try_lock is not set yet, make sure all locked inodes are
 484		 * not in the AIL.  If any are, set try_lock to be used later.
 485		 */
 486		if (!try_lock) {
 487			for (j = (i - 1); j >= 0 && !try_lock; j--) {
 488				lp = &ips[j]->i_itemp->ili_item;
 489				if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
 490					try_lock++;
 491			}
 492		}
 493
 494		/*
 495		 * If any of the previous locks we have locked is in the AIL,
 496		 * we must TRY to get the second and subsequent locks. If
 497		 * we can't get any, we must release all we have
 498		 * and try again.
 499		 */
 500		if (!try_lock) {
 501			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
 502			continue;
 503		}
 504
 505		/* try_lock means we have an inode locked that is in the AIL. */
 506		ASSERT(i != 0);
 507		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
 508			continue;
 509
 510		/*
 511		 * Unlock all previous guys and try again.  xfs_iunlock will try
 512		 * to push the tail if the inode is in the AIL.
 513		 */
 514		attempts++;
 515		for (j = i - 1; j >= 0; j--) {
 516			/*
 517			 * Check to see if we've already unlocked this one.  Not
 518			 * the first one going back, and the inode ptr is the
 519			 * same.
 520			 */
 521			if (j != (i - 1) && ips[j] == ips[j + 1])
 522				continue;
 523
 524			xfs_iunlock(ips[j], lock_mode);
 525		}
 526
 527		if ((attempts % 5) == 0) {
 528			delay(1); /* Don't just spin the CPU */
 
 
 
 529		}
 530		i = 0;
 531		try_lock = 0;
 532		goto again;
 533	}
 
 
 
 
 
 
 
 
 
 
 534}
 535
 536/*
 537 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
 538 * the mmaplock or the ilock, but not more than one type at a time. If we lock
 539 * more than one at a time, lockdep will report false positives saying we have
 540 * violated locking orders.  The iolock must be double-locked separately since
 541 * we use i_rwsem for that.  We now support taking one lock EXCL and the other
 542 * SHARED.
 543 */
 544void
 545xfs_lock_two_inodes(
 546	struct xfs_inode	*ip0,
 547	uint			ip0_mode,
 548	struct xfs_inode	*ip1,
 549	uint			ip1_mode)
 550{
 551	struct xfs_inode	*temp;
 552	uint			mode_temp;
 553	int			attempts = 0;
 554	struct xfs_log_item	*lp;
 555
 556	ASSERT(hweight32(ip0_mode) == 1);
 557	ASSERT(hweight32(ip1_mode) == 1);
 558	ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
 559	ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
 560	ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
 561	       !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
 562	ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
 563	       !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
 564	ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
 565	       !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
 566	ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
 567	       !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
 568
 569	ASSERT(ip0->i_ino != ip1->i_ino);
 570
 571	if (ip0->i_ino > ip1->i_ino) {
 572		temp = ip0;
 573		ip0 = ip1;
 574		ip1 = temp;
 575		mode_temp = ip0_mode;
 576		ip0_mode = ip1_mode;
 577		ip1_mode = mode_temp;
 578	}
 579
 580 again:
 581	xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
 582
 583	/*
 584	 * If the first lock we have locked is in the AIL, we must TRY to get
 585	 * the second lock. If we can't get it, we must release the first one
 586	 * and try again.
 587	 */
 588	lp = &ip0->i_itemp->ili_item;
 589	if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
 590		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
 591			xfs_iunlock(ip0, ip0_mode);
 592			if ((++attempts % 5) == 0)
 593				delay(1); /* Don't just spin the CPU */
 594			goto again;
 595		}
 596	} else {
 597		xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
 598	}
 599}
 600
 
 601void
 602__xfs_iflock(
 603	struct xfs_inode	*ip)
 604{
 605	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
 606	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
 607
 608	do {
 609		prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
 610		if (xfs_isiflocked(ip))
 611			io_schedule();
 612	} while (!xfs_iflock_nowait(ip));
 613
 614	finish_wait(wq, &wait.wq_entry);
 615}
 616
 617STATIC uint
 618_xfs_dic2xflags(
 619	uint16_t		di_flags,
 620	uint64_t		di_flags2,
 621	bool			has_attr)
 622{
 623	uint			flags = 0;
 624
 625	if (di_flags & XFS_DIFLAG_ANY) {
 626		if (di_flags & XFS_DIFLAG_REALTIME)
 627			flags |= FS_XFLAG_REALTIME;
 628		if (di_flags & XFS_DIFLAG_PREALLOC)
 629			flags |= FS_XFLAG_PREALLOC;
 630		if (di_flags & XFS_DIFLAG_IMMUTABLE)
 631			flags |= FS_XFLAG_IMMUTABLE;
 632		if (di_flags & XFS_DIFLAG_APPEND)
 633			flags |= FS_XFLAG_APPEND;
 634		if (di_flags & XFS_DIFLAG_SYNC)
 635			flags |= FS_XFLAG_SYNC;
 636		if (di_flags & XFS_DIFLAG_NOATIME)
 637			flags |= FS_XFLAG_NOATIME;
 638		if (di_flags & XFS_DIFLAG_NODUMP)
 639			flags |= FS_XFLAG_NODUMP;
 640		if (di_flags & XFS_DIFLAG_RTINHERIT)
 641			flags |= FS_XFLAG_RTINHERIT;
 642		if (di_flags & XFS_DIFLAG_PROJINHERIT)
 643			flags |= FS_XFLAG_PROJINHERIT;
 644		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
 645			flags |= FS_XFLAG_NOSYMLINKS;
 646		if (di_flags & XFS_DIFLAG_EXTSIZE)
 647			flags |= FS_XFLAG_EXTSIZE;
 648		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
 649			flags |= FS_XFLAG_EXTSZINHERIT;
 650		if (di_flags & XFS_DIFLAG_NODEFRAG)
 651			flags |= FS_XFLAG_NODEFRAG;
 652		if (di_flags & XFS_DIFLAG_FILESTREAM)
 653			flags |= FS_XFLAG_FILESTREAM;
 654	}
 655
 656	if (di_flags2 & XFS_DIFLAG2_ANY) {
 657		if (di_flags2 & XFS_DIFLAG2_DAX)
 658			flags |= FS_XFLAG_DAX;
 659		if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
 660			flags |= FS_XFLAG_COWEXTSIZE;
 661	}
 662
 663	if (has_attr)
 664		flags |= FS_XFLAG_HASATTR;
 665
 666	return flags;
 667}
 668
 669uint
 670xfs_ip2xflags(
 671	struct xfs_inode	*ip)
 672{
 673	struct xfs_icdinode	*dic = &ip->i_d;
 674
 675	return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
 676}
 677
 678/*
 679 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
 680 * is allowed, otherwise it has to be an exact match. If a CI match is found,
 681 * ci_name->name will point to a the actual name (caller must free) or
 682 * will be set to NULL if an exact match is found.
 683 */
 684int
 685xfs_lookup(
 686	xfs_inode_t		*dp,
 687	struct xfs_name		*name,
 688	xfs_inode_t		**ipp,
 689	struct xfs_name		*ci_name)
 690{
 691	xfs_ino_t		inum;
 692	int			error;
 693
 694	trace_xfs_lookup(dp, name);
 695
 696	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 697		return -EIO;
 698
 699	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
 700	if (error)
 701		goto out_unlock;
 702
 703	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
 704	if (error)
 705		goto out_free_name;
 706
 707	return 0;
 708
 709out_free_name:
 710	if (ci_name)
 711		kmem_free(ci_name->name);
 712out_unlock:
 713	*ipp = NULL;
 714	return error;
 715}
 716
 717/*
 718 * Allocate an inode on disk and return a copy of its in-core version.
 719 * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
 720 * appropriately within the inode.  The uid and gid for the inode are
 721 * set according to the contents of the given cred structure.
 722 *
 723 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
 724 * has a free inode available, call xfs_iget() to obtain the in-core
 725 * version of the allocated inode.  Finally, fill in the inode and
 726 * log its initial contents.  In this case, ialloc_context would be
 727 * set to NULL.
 728 *
 729 * If xfs_dialloc() does not have an available inode, it will replenish
 730 * its supply by doing an allocation. Since we can only do one
 731 * allocation within a transaction without deadlocks, we must commit
 732 * the current transaction before returning the inode itself.
 733 * In this case, therefore, we will set ialloc_context and return.
 734 * The caller should then commit the current transaction, start a new
 735 * transaction, and call xfs_ialloc() again to actually get the inode.
 736 *
 737 * To ensure that some other process does not grab the inode that
 738 * was allocated during the first call to xfs_ialloc(), this routine
 739 * also returns the [locked] bp pointing to the head of the freelist
 740 * as ialloc_context.  The caller should hold this buffer across
 741 * the commit and pass it back into this routine on the second call.
 742 *
 743 * If we are allocating quota inodes, we do not have a parent inode
 744 * to attach to or associate with (i.e. pip == NULL) because they
 745 * are not linked into the directory structure - they are attached
 746 * directly to the superblock - and so have no parent.
 747 */
 748static int
 749xfs_ialloc(
 750	xfs_trans_t	*tp,
 751	xfs_inode_t	*pip,
 752	umode_t		mode,
 753	xfs_nlink_t	nlink,
 754	dev_t		rdev,
 755	prid_t		prid,
 
 756	xfs_buf_t	**ialloc_context,
 757	xfs_inode_t	**ipp)
 758{
 759	struct xfs_mount *mp = tp->t_mountp;
 760	xfs_ino_t	ino;
 761	xfs_inode_t	*ip;
 762	uint		flags;
 763	int		error;
 764	struct timespec64 tv;
 765	struct inode	*inode;
 766
 767	/*
 768	 * Call the space management code to pick
 769	 * the on-disk inode to be allocated.
 770	 */
 771	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode,
 772			    ialloc_context, &ino);
 773	if (error)
 774		return error;
 775	if (*ialloc_context || ino == NULLFSINO) {
 776		*ipp = NULL;
 777		return 0;
 778	}
 779	ASSERT(*ialloc_context == NULL);
 780
 781	/*
 782	 * Protect against obviously corrupt allocation btree records. Later
 783	 * xfs_iget checks will catch re-allocation of other active in-memory
 784	 * and on-disk inodes. If we don't catch reallocating the parent inode
 785	 * here we will deadlock in xfs_iget() so we have to do these checks
 786	 * first.
 787	 */
 788	if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
 789		xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
 790		return -EFSCORRUPTED;
 791	}
 792
 793	/*
 794	 * Get the in-core inode with the lock held exclusively.
 795	 * This is because we're setting fields here we need
 796	 * to prevent others from looking at until we're done.
 797	 */
 798	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
 799			 XFS_ILOCK_EXCL, &ip);
 800	if (error)
 801		return error;
 802	ASSERT(ip != NULL);
 803	inode = VFS_I(ip);
 
 
 
 
 
 
 
 
 
 804	inode->i_mode = mode;
 805	set_nlink(inode, nlink);
 806	inode->i_uid = current_fsuid();
 807	inode->i_rdev = rdev;
 808	ip->i_d.di_projid = prid;
 809
 810	if (pip && XFS_INHERIT_GID(pip)) {
 811		inode->i_gid = VFS_I(pip)->i_gid;
 812		if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
 813			inode->i_mode |= S_ISGID;
 814	} else {
 815		inode->i_gid = current_fsgid();
 816	}
 817
 818	/*
 819	 * If the group ID of the new file does not match the effective group
 820	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
 821	 * (and only if the irix_sgid_inherit compatibility variable is set).
 822	 */
 823	if (irix_sgid_inherit &&
 824	    (inode->i_mode & S_ISGID) && !in_group_p(inode->i_gid))
 
 825		inode->i_mode &= ~S_ISGID;
 826
 827	ip->i_d.di_size = 0;
 828	ip->i_df.if_nextents = 0;
 829	ASSERT(ip->i_d.di_nblocks == 0);
 830
 831	tv = current_time(inode);
 832	inode->i_mtime = tv;
 833	inode->i_atime = tv;
 834	inode->i_ctime = tv;
 835
 836	ip->i_d.di_extsize = 0;
 837	ip->i_d.di_dmevmask = 0;
 838	ip->i_d.di_dmstate = 0;
 839	ip->i_d.di_flags = 0;
 840
 841	if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
 842		inode_set_iversion(inode, 1);
 843		ip->i_d.di_flags2 = 0;
 844		ip->i_d.di_cowextsize = 0;
 845		ip->i_d.di_crtime = tv;
 
 846	}
 847
 
 848	flags = XFS_ILOG_CORE;
 849	switch (mode & S_IFMT) {
 850	case S_IFIFO:
 851	case S_IFCHR:
 852	case S_IFBLK:
 853	case S_IFSOCK:
 854		ip->i_df.if_format = XFS_DINODE_FMT_DEV;
 
 855		ip->i_df.if_flags = 0;
 856		flags |= XFS_ILOG_DEV;
 857		break;
 858	case S_IFREG:
 859	case S_IFDIR:
 860		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
 
 861			uint		di_flags = 0;
 862
 863			if (S_ISDIR(mode)) {
 864				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
 865					di_flags |= XFS_DIFLAG_RTINHERIT;
 866				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
 867					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
 868					ip->i_d.di_extsize = pip->i_d.di_extsize;
 869				}
 870				if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
 871					di_flags |= XFS_DIFLAG_PROJINHERIT;
 872			} else if (S_ISREG(mode)) {
 873				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
 874					di_flags |= XFS_DIFLAG_REALTIME;
 875				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
 876					di_flags |= XFS_DIFLAG_EXTSIZE;
 877					ip->i_d.di_extsize = pip->i_d.di_extsize;
 878				}
 879			}
 880			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
 881			    xfs_inherit_noatime)
 882				di_flags |= XFS_DIFLAG_NOATIME;
 883			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
 884			    xfs_inherit_nodump)
 885				di_flags |= XFS_DIFLAG_NODUMP;
 886			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
 887			    xfs_inherit_sync)
 888				di_flags |= XFS_DIFLAG_SYNC;
 889			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
 890			    xfs_inherit_nosymlinks)
 891				di_flags |= XFS_DIFLAG_NOSYMLINKS;
 892			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
 893			    xfs_inherit_nodefrag)
 894				di_flags |= XFS_DIFLAG_NODEFRAG;
 895			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
 896				di_flags |= XFS_DIFLAG_FILESTREAM;
 
 
 897
 898			ip->i_d.di_flags |= di_flags;
 
 899		}
 900		if (pip && (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY)) {
 
 
 
 901			if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
 902				ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
 903				ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
 904			}
 905			if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
 906				ip->i_d.di_flags2 |= XFS_DIFLAG2_DAX;
 907		}
 908		/* FALLTHROUGH */
 909	case S_IFLNK:
 910		ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
 911		ip->i_df.if_flags = XFS_IFEXTENTS;
 912		ip->i_df.if_bytes = 0;
 913		ip->i_df.if_u1.if_root = NULL;
 914		break;
 915	default:
 916		ASSERT(0);
 917	}
 
 
 
 
 
 918
 919	/*
 920	 * Log the new values stuffed into the inode.
 921	 */
 922	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 923	xfs_trans_log_inode(tp, ip, flags);
 924
 925	/* now that we have an i_mode we can setup the inode structure */
 926	xfs_setup_inode(ip);
 927
 928	*ipp = ip;
 929	return 0;
 930}
 931
 932/*
 933 * Allocates a new inode from disk and return a pointer to the
 934 * incore copy. This routine will internally commit the current
 935 * transaction and allocate a new one if the Space Manager needed
 936 * to do an allocation to replenish the inode free-list.
 937 *
 938 * This routine is designed to be called from xfs_create and
 939 * xfs_create_dir.
 940 *
 941 */
 942int
 943xfs_dir_ialloc(
 944	xfs_trans_t	**tpp,		/* input: current transaction;
 945					   output: may be a new transaction. */
 946	xfs_inode_t	*dp,		/* directory within whose allocate
 947					   the inode. */
 948	umode_t		mode,
 949	xfs_nlink_t	nlink,
 950	dev_t		rdev,
 951	prid_t		prid,		/* project id */
 952	xfs_inode_t	**ipp)		/* pointer to inode; it will be
 
 953					   locked. */
 
 
 954{
 955	xfs_trans_t	*tp;
 956	xfs_inode_t	*ip;
 957	xfs_buf_t	*ialloc_context = NULL;
 958	int		code;
 959	void		*dqinfo;
 960	uint		tflags;
 961
 962	tp = *tpp;
 963	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 964
 965	/*
 966	 * xfs_ialloc will return a pointer to an incore inode if
 967	 * the Space Manager has an available inode on the free
 968	 * list. Otherwise, it will do an allocation and replenish
 969	 * the freelist.  Since we can only do one allocation per
 970	 * transaction without deadlocks, we will need to commit the
 971	 * current transaction and start a new one.  We will then
 972	 * need to call xfs_ialloc again to get the inode.
 973	 *
 974	 * If xfs_ialloc did an allocation to replenish the freelist,
 975	 * it returns the bp containing the head of the freelist as
 976	 * ialloc_context. We will hold a lock on it across the
 977	 * transaction commit so that no other process can steal
 978	 * the inode(s) that we've just allocated.
 979	 */
 980	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
 981			&ip);
 982
 983	/*
 984	 * Return an error if we were unable to allocate a new inode.
 985	 * This should only happen if we run out of space on disk or
 986	 * encounter a disk error.
 987	 */
 988	if (code) {
 989		*ipp = NULL;
 990		return code;
 991	}
 992	if (!ialloc_context && !ip) {
 993		*ipp = NULL;
 994		return -ENOSPC;
 995	}
 996
 997	/*
 998	 * If the AGI buffer is non-NULL, then we were unable to get an
 999	 * inode in one operation.  We need to commit the current
1000	 * transaction and call xfs_ialloc() again.  It is guaranteed
1001	 * to succeed the second time.
1002	 */
1003	if (ialloc_context) {
1004		/*
1005		 * Normally, xfs_trans_commit releases all the locks.
1006		 * We call bhold to hang on to the ialloc_context across
1007		 * the commit.  Holding this buffer prevents any other
1008		 * processes from doing any allocations in this
1009		 * allocation group.
1010		 */
1011		xfs_trans_bhold(tp, ialloc_context);
1012
1013		/*
1014		 * We want the quota changes to be associated with the next
1015		 * transaction, NOT this one. So, detach the dqinfo from this
1016		 * and attach it to the next transaction.
1017		 */
1018		dqinfo = NULL;
1019		tflags = 0;
1020		if (tp->t_dqinfo) {
1021			dqinfo = (void *)tp->t_dqinfo;
1022			tp->t_dqinfo = NULL;
1023			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1024			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1025		}
1026
1027		code = xfs_trans_roll(&tp);
 
 
1028
1029		/*
1030		 * Re-attach the quota info that we detached from prev trx.
1031		 */
1032		if (dqinfo) {
1033			tp->t_dqinfo = dqinfo;
1034			tp->t_flags |= tflags;
1035		}
1036
1037		if (code) {
1038			xfs_buf_relse(ialloc_context);
1039			*tpp = tp;
1040			*ipp = NULL;
1041			return code;
1042		}
1043		xfs_trans_bjoin(tp, ialloc_context);
1044
1045		/*
1046		 * Call ialloc again. Since we've locked out all
1047		 * other allocations in this allocation group,
1048		 * this call should always succeed.
1049		 */
1050		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1051				  &ialloc_context, &ip);
1052
1053		/*
1054		 * If we get an error at this point, return to the caller
1055		 * so that the current transaction can be aborted.
1056		 */
1057		if (code) {
1058			*tpp = tp;
1059			*ipp = NULL;
1060			return code;
1061		}
1062		ASSERT(!ialloc_context && ip);
1063
 
 
 
1064	}
1065
1066	*ipp = ip;
1067	*tpp = tp;
1068
1069	return 0;
1070}
1071
1072/*
1073 * Decrement the link count on an inode & log the change.  If this causes the
1074 * link count to go to zero, move the inode to AGI unlinked list so that it can
1075 * be freed when the last active reference goes away via xfs_inactive().
1076 */
1077static int			/* error */
1078xfs_droplink(
1079	xfs_trans_t *tp,
1080	xfs_inode_t *ip)
1081{
1082	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1083
1084	drop_nlink(VFS_I(ip));
1085	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1086
1087	if (VFS_I(ip)->i_nlink)
1088		return 0;
1089
1090	return xfs_iunlink(tp, ip);
1091}
1092
1093/*
1094 * Increment the link count on an inode & log the change.
1095 */
1096static void
1097xfs_bumplink(
1098	xfs_trans_t *tp,
1099	xfs_inode_t *ip)
1100{
1101	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1102
 
1103	inc_nlink(VFS_I(ip));
1104	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 
1105}
1106
1107int
1108xfs_create(
1109	xfs_inode_t		*dp,
1110	struct xfs_name		*name,
1111	umode_t			mode,
1112	dev_t			rdev,
1113	xfs_inode_t		**ipp)
1114{
1115	int			is_dir = S_ISDIR(mode);
1116	struct xfs_mount	*mp = dp->i_mount;
1117	struct xfs_inode	*ip = NULL;
1118	struct xfs_trans	*tp = NULL;
1119	int			error;
 
 
1120	bool                    unlock_dp_on_error = false;
1121	prid_t			prid;
1122	struct xfs_dquot	*udqp = NULL;
1123	struct xfs_dquot	*gdqp = NULL;
1124	struct xfs_dquot	*pdqp = NULL;
1125	struct xfs_trans_res	*tres;
1126	uint			resblks;
1127
1128	trace_xfs_create(dp, name);
1129
1130	if (XFS_FORCED_SHUTDOWN(mp))
1131		return -EIO;
1132
1133	prid = xfs_get_initial_prid(dp);
1134
1135	/*
1136	 * Make sure that we have allocated dquot(s) on disk.
1137	 */
1138	error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
 
1139					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1140					&udqp, &gdqp, &pdqp);
1141	if (error)
1142		return error;
1143
1144	if (is_dir) {
 
1145		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1146		tres = &M_RES(mp)->tr_mkdir;
1147	} else {
1148		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1149		tres = &M_RES(mp)->tr_create;
1150	}
1151
1152	/*
1153	 * Initially assume that the file does not exist and
1154	 * reserve the resources for that case.  If that is not
1155	 * the case we'll drop the one we have and get a more
1156	 * appropriate transaction later.
1157	 */
1158	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1159	if (error == -ENOSPC) {
1160		/* flush outstanding delalloc blocks and retry */
1161		xfs_flush_inodes(mp);
1162		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1163	}
 
 
 
 
 
1164	if (error)
1165		goto out_release_inode;
1166
1167	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1168	unlock_dp_on_error = true;
1169
 
 
1170	/*
1171	 * Reserve disk quota and the inode.
1172	 */
1173	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1174						pdqp, resblks, 1, 0);
1175	if (error)
1176		goto out_trans_cancel;
1177
 
 
 
 
 
 
1178	/*
1179	 * A newly created regular or special file just has one directory
1180	 * entry pointing to them, but a directory also the "." entry
1181	 * pointing to itself.
1182	 */
1183	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip);
 
1184	if (error)
1185		goto out_trans_cancel;
1186
1187	/*
1188	 * Now we join the directory inode to the transaction.  We do not do it
1189	 * earlier because xfs_dir_ialloc might commit the previous transaction
1190	 * (and release all the locks).  An error from here on will result in
1191	 * the transaction cancel unlocking dp so don't do it explicitly in the
1192	 * error path.
1193	 */
1194	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1195	unlock_dp_on_error = false;
1196
1197	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1198					resblks - XFS_IALLOC_SPACE_RES(mp));
 
1199	if (error) {
1200		ASSERT(error != -ENOSPC);
1201		goto out_trans_cancel;
1202	}
1203	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1204	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1205
1206	if (is_dir) {
1207		error = xfs_dir_init(tp, ip, dp);
1208		if (error)
1209			goto out_trans_cancel;
1210
1211		xfs_bumplink(tp, dp);
 
 
1212	}
1213
1214	/*
1215	 * If this is a synchronous mount, make sure that the
1216	 * create transaction goes to disk before returning to
1217	 * the user.
1218	 */
1219	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1220		xfs_trans_set_sync(tp);
1221
1222	/*
1223	 * Attach the dquot(s) to the inodes and modify them incore.
1224	 * These ids of the inode couldn't have changed since the new
1225	 * inode has been locked ever since it was created.
1226	 */
1227	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1228
 
 
 
 
1229	error = xfs_trans_commit(tp);
1230	if (error)
1231		goto out_release_inode;
1232
1233	xfs_qm_dqrele(udqp);
1234	xfs_qm_dqrele(gdqp);
1235	xfs_qm_dqrele(pdqp);
1236
1237	*ipp = ip;
1238	return 0;
1239
 
 
1240 out_trans_cancel:
1241	xfs_trans_cancel(tp);
1242 out_release_inode:
1243	/*
1244	 * Wait until after the current transaction is aborted to finish the
1245	 * setup of the inode and release the inode.  This prevents recursive
1246	 * transactions and deadlocks from xfs_inactive.
1247	 */
1248	if (ip) {
1249		xfs_finish_inode_setup(ip);
1250		xfs_irele(ip);
1251	}
1252
1253	xfs_qm_dqrele(udqp);
1254	xfs_qm_dqrele(gdqp);
1255	xfs_qm_dqrele(pdqp);
1256
1257	if (unlock_dp_on_error)
1258		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1259	return error;
1260}
1261
1262int
1263xfs_create_tmpfile(
1264	struct xfs_inode	*dp,
 
1265	umode_t			mode,
1266	struct xfs_inode	**ipp)
1267{
1268	struct xfs_mount	*mp = dp->i_mount;
1269	struct xfs_inode	*ip = NULL;
1270	struct xfs_trans	*tp = NULL;
1271	int			error;
1272	prid_t                  prid;
1273	struct xfs_dquot	*udqp = NULL;
1274	struct xfs_dquot	*gdqp = NULL;
1275	struct xfs_dquot	*pdqp = NULL;
1276	struct xfs_trans_res	*tres;
1277	uint			resblks;
1278
1279	if (XFS_FORCED_SHUTDOWN(mp))
1280		return -EIO;
1281
1282	prid = xfs_get_initial_prid(dp);
1283
1284	/*
1285	 * Make sure that we have allocated dquot(s) on disk.
1286	 */
1287	error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
 
1288				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1289				&udqp, &gdqp, &pdqp);
1290	if (error)
1291		return error;
1292
1293	resblks = XFS_IALLOC_SPACE_RES(mp);
1294	tres = &M_RES(mp)->tr_create_tmpfile;
1295
1296	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
 
 
 
 
 
1297	if (error)
1298		goto out_release_inode;
1299
1300	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1301						pdqp, resblks, 1, 0);
1302	if (error)
1303		goto out_trans_cancel;
1304
1305	error = xfs_dir_ialloc(&tp, dp, mode, 0, 0, prid, &ip);
 
1306	if (error)
1307		goto out_trans_cancel;
1308
1309	if (mp->m_flags & XFS_MOUNT_WSYNC)
1310		xfs_trans_set_sync(tp);
1311
1312	/*
1313	 * Attach the dquot(s) to the inodes and modify them incore.
1314	 * These ids of the inode couldn't have changed since the new
1315	 * inode has been locked ever since it was created.
1316	 */
1317	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1318
1319	error = xfs_iunlink(tp, ip);
1320	if (error)
1321		goto out_trans_cancel;
1322
1323	error = xfs_trans_commit(tp);
1324	if (error)
1325		goto out_release_inode;
1326
1327	xfs_qm_dqrele(udqp);
1328	xfs_qm_dqrele(gdqp);
1329	xfs_qm_dqrele(pdqp);
1330
1331	*ipp = ip;
1332	return 0;
1333
1334 out_trans_cancel:
1335	xfs_trans_cancel(tp);
1336 out_release_inode:
1337	/*
1338	 * Wait until after the current transaction is aborted to finish the
1339	 * setup of the inode and release the inode.  This prevents recursive
1340	 * transactions and deadlocks from xfs_inactive.
1341	 */
1342	if (ip) {
1343		xfs_finish_inode_setup(ip);
1344		xfs_irele(ip);
1345	}
1346
1347	xfs_qm_dqrele(udqp);
1348	xfs_qm_dqrele(gdqp);
1349	xfs_qm_dqrele(pdqp);
1350
1351	return error;
1352}
1353
1354int
1355xfs_link(
1356	xfs_inode_t		*tdp,
1357	xfs_inode_t		*sip,
1358	struct xfs_name		*target_name)
1359{
1360	xfs_mount_t		*mp = tdp->i_mount;
1361	xfs_trans_t		*tp;
1362	int			error;
 
 
1363	int			resblks;
1364
1365	trace_xfs_link(tdp, target_name);
1366
1367	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1368
1369	if (XFS_FORCED_SHUTDOWN(mp))
1370		return -EIO;
1371
1372	error = xfs_qm_dqattach(sip);
1373	if (error)
1374		goto std_return;
1375
1376	error = xfs_qm_dqattach(tdp);
1377	if (error)
1378		goto std_return;
1379
1380	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1381	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1382	if (error == -ENOSPC) {
1383		resblks = 0;
1384		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1385	}
1386	if (error)
1387		goto std_return;
1388
1389	xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1390
1391	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1392	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1393
1394	/*
1395	 * If we are using project inheritance, we only allow hard link
1396	 * creation in our tree when the project IDs are the same; else
1397	 * the tree quota mechanism could be circumvented.
1398	 */
1399	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1400		     tdp->i_d.di_projid != sip->i_d.di_projid)) {
1401		error = -EXDEV;
1402		goto error_return;
1403	}
1404
1405	if (!resblks) {
1406		error = xfs_dir_canenter(tp, tdp, target_name);
1407		if (error)
1408			goto error_return;
1409	}
1410
 
 
1411	/*
1412	 * Handle initial link state of O_TMPFILE inode
1413	 */
1414	if (VFS_I(sip)->i_nlink == 0) {
1415		error = xfs_iunlink_remove(tp, sip);
1416		if (error)
1417			goto error_return;
1418	}
1419
1420	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1421				   resblks);
1422	if (error)
1423		goto error_return;
1424	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1425	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1426
1427	xfs_bumplink(tp, sip);
 
 
1428
1429	/*
1430	 * If this is a synchronous mount, make sure that the
1431	 * link transaction goes to disk before returning to
1432	 * the user.
1433	 */
1434	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1435		xfs_trans_set_sync(tp);
1436
 
 
 
 
 
 
1437	return xfs_trans_commit(tp);
1438
1439 error_return:
1440	xfs_trans_cancel(tp);
1441 std_return:
1442	return error;
1443}
1444
1445/* Clear the reflink flag and the cowblocks tag if possible. */
1446static void
1447xfs_itruncate_clear_reflink_flags(
1448	struct xfs_inode	*ip)
1449{
1450	struct xfs_ifork	*dfork;
1451	struct xfs_ifork	*cfork;
1452
1453	if (!xfs_is_reflink_inode(ip))
1454		return;
1455	dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1456	cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1457	if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1458		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1459	if (cfork->if_bytes == 0)
1460		xfs_inode_clear_cowblocks_tag(ip);
1461}
1462
1463/*
1464 * Free up the underlying blocks past new_size.  The new size must be smaller
1465 * than the current size.  This routine can be used both for the attribute and
1466 * data fork, and does not modify the inode size, which is left to the caller.
1467 *
1468 * The transaction passed to this routine must have made a permanent log
1469 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1470 * given transaction and start new ones, so make sure everything involved in
1471 * the transaction is tidy before calling here.  Some transaction will be
1472 * returned to the caller to be committed.  The incoming transaction must
1473 * already include the inode, and both inode locks must be held exclusively.
1474 * The inode must also be "held" within the transaction.  On return the inode
1475 * will be "held" within the returned transaction.  This routine does NOT
1476 * require any disk space to be reserved for it within the transaction.
1477 *
1478 * If we get an error, we must return with the inode locked and linked into the
1479 * current transaction. This keeps things simple for the higher level code,
1480 * because it always knows that the inode is locked and held in the transaction
1481 * that returns to it whether errors occur or not.  We don't mark the inode
1482 * dirty on error so that transactions can be easily aborted if possible.
1483 */
1484int
1485xfs_itruncate_extents_flags(
1486	struct xfs_trans	**tpp,
1487	struct xfs_inode	*ip,
1488	int			whichfork,
1489	xfs_fsize_t		new_size,
1490	int			flags)
1491{
1492	struct xfs_mount	*mp = ip->i_mount;
1493	struct xfs_trans	*tp = *tpp;
 
 
1494	xfs_fileoff_t		first_unmap_block;
 
1495	xfs_filblks_t		unmap_len;
1496	int			error = 0;
 
1497
1498	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1499	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1500	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1501	ASSERT(new_size <= XFS_ISIZE(ip));
1502	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1503	ASSERT(ip->i_itemp != NULL);
1504	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1505	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1506
1507	trace_xfs_itruncate_extents_start(ip, new_size);
1508
1509	flags |= xfs_bmapi_aflag(whichfork);
1510
1511	/*
1512	 * Since it is possible for space to become allocated beyond
1513	 * the end of the file (in a crash where the space is allocated
1514	 * but the inode size is not yet updated), simply remove any
1515	 * blocks which show up between the new EOF and the maximum
1516	 * possible file size.
1517	 *
1518	 * We have to free all the blocks to the bmbt maximum offset, even if
1519	 * the page cache can't scale that far.
1520	 */
1521	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1522	if (first_unmap_block >= XFS_MAX_FILEOFF) {
1523		WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1524		return 0;
1525	}
1526
1527	unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1528	while (unmap_len > 0) {
1529		ASSERT(tp->t_firstblock == NULLFSBLOCK);
1530		error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1531				flags, XFS_ITRUNC_MAX_EXTENTS);
 
 
 
 
 
1532		if (error)
1533			goto out;
1534
1535		/*
1536		 * Duplicate the transaction that has the permanent
1537		 * reservation and commit the old transaction.
1538		 */
1539		error = xfs_defer_finish(&tp);
1540		if (error)
1541			goto out;
1542
1543		error = xfs_trans_roll_inode(&tp, ip);
1544		if (error)
1545			goto out;
1546	}
1547
1548	if (whichfork == XFS_DATA_FORK) {
1549		/* Remove all pending CoW reservations. */
1550		error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1551				first_unmap_block, XFS_MAX_FILEOFF, true);
1552		if (error)
1553			goto out;
1554
1555		xfs_itruncate_clear_reflink_flags(ip);
 
 
 
 
 
1556	}
1557
1558	/*
1559	 * Always re-log the inode so that our permanent transaction can keep
1560	 * on rolling it forward in the log.
1561	 */
1562	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1563
1564	trace_xfs_itruncate_extents_end(ip, new_size);
1565
1566out:
1567	*tpp = tp;
1568	return error;
 
 
 
 
 
 
 
 
1569}
1570
1571int
1572xfs_release(
1573	xfs_inode_t	*ip)
1574{
1575	xfs_mount_t	*mp = ip->i_mount;
1576	int		error;
1577
1578	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1579		return 0;
1580
1581	/* If this is a read-only mount, don't do this (would generate I/O) */
1582	if (mp->m_flags & XFS_MOUNT_RDONLY)
1583		return 0;
1584
1585	if (!XFS_FORCED_SHUTDOWN(mp)) {
1586		int truncated;
1587
1588		/*
1589		 * If we previously truncated this file and removed old data
1590		 * in the process, we want to initiate "early" writeout on
1591		 * the last close.  This is an attempt to combat the notorious
1592		 * NULL files problem which is particularly noticeable from a
1593		 * truncate down, buffered (re-)write (delalloc), followed by
1594		 * a crash.  What we are effectively doing here is
1595		 * significantly reducing the time window where we'd otherwise
1596		 * be exposed to that problem.
1597		 */
1598		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1599		if (truncated) {
1600			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1601			if (ip->i_delayed_blks > 0) {
1602				error = filemap_flush(VFS_I(ip)->i_mapping);
1603				if (error)
1604					return error;
1605			}
1606		}
1607	}
1608
1609	if (VFS_I(ip)->i_nlink == 0)
1610		return 0;
1611
1612	if (xfs_can_free_eofblocks(ip, false)) {
1613
1614		/*
1615		 * Check if the inode is being opened, written and closed
1616		 * frequently and we have delayed allocation blocks outstanding
1617		 * (e.g. streaming writes from the NFS server), truncating the
1618		 * blocks past EOF will cause fragmentation to occur.
1619		 *
1620		 * In this case don't do the truncation, but we have to be
1621		 * careful how we detect this case. Blocks beyond EOF show up as
1622		 * i_delayed_blks even when the inode is clean, so we need to
1623		 * truncate them away first before checking for a dirty release.
1624		 * Hence on the first dirty close we will still remove the
1625		 * speculative allocation, but after that we will leave it in
1626		 * place.
1627		 */
1628		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1629			return 0;
1630		/*
1631		 * If we can't get the iolock just skip truncating the blocks
1632		 * past EOF because we could deadlock with the mmap_lock
1633		 * otherwise. We'll get another chance to drop them once the
1634		 * last reference to the inode is dropped, so we'll never leak
1635		 * blocks permanently.
1636		 */
1637		if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1638			error = xfs_free_eofblocks(ip);
1639			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1640			if (error)
1641				return error;
1642		}
1643
1644		/* delalloc blocks after truncation means it really is dirty */
1645		if (ip->i_delayed_blks)
1646			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1647	}
1648	return 0;
1649}
1650
1651/*
1652 * xfs_inactive_truncate
1653 *
1654 * Called to perform a truncate when an inode becomes unlinked.
1655 */
1656STATIC int
1657xfs_inactive_truncate(
1658	struct xfs_inode *ip)
1659{
1660	struct xfs_mount	*mp = ip->i_mount;
1661	struct xfs_trans	*tp;
1662	int			error;
1663
1664	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1665	if (error) {
1666		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1667		return error;
1668	}
 
1669	xfs_ilock(ip, XFS_ILOCK_EXCL);
1670	xfs_trans_ijoin(tp, ip, 0);
1671
1672	/*
1673	 * Log the inode size first to prevent stale data exposure in the event
1674	 * of a system crash before the truncate completes. See the related
1675	 * comment in xfs_vn_setattr_size() for details.
1676	 */
1677	ip->i_d.di_size = 0;
1678	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1679
1680	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1681	if (error)
1682		goto error_trans_cancel;
1683
1684	ASSERT(ip->i_df.if_nextents == 0);
1685
1686	error = xfs_trans_commit(tp);
1687	if (error)
1688		goto error_unlock;
1689
1690	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1691	return 0;
1692
1693error_trans_cancel:
1694	xfs_trans_cancel(tp);
1695error_unlock:
1696	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1697	return error;
1698}
1699
1700/*
1701 * xfs_inactive_ifree()
1702 *
1703 * Perform the inode free when an inode is unlinked.
1704 */
1705STATIC int
1706xfs_inactive_ifree(
1707	struct xfs_inode *ip)
1708{
 
 
1709	struct xfs_mount	*mp = ip->i_mount;
1710	struct xfs_trans	*tp;
1711	int			error;
1712
1713	/*
1714	 * We try to use a per-AG reservation for any block needed by the finobt
1715	 * tree, but as the finobt feature predates the per-AG reservation
1716	 * support a degraded file system might not have enough space for the
1717	 * reservation at mount time.  In that case try to dip into the reserved
1718	 * pool and pray.
1719	 *
1720	 * Send a warning if the reservation does happen to fail, as the inode
1721	 * now remains allocated and sits on the unlinked list until the fs is
1722	 * repaired.
1723	 */
1724	if (unlikely(mp->m_finobt_nores)) {
1725		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1726				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1727				&tp);
1728	} else {
1729		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1730	}
1731	if (error) {
1732		if (error == -ENOSPC) {
1733			xfs_warn_ratelimited(mp,
1734			"Failed to remove inode(s) from unlinked list. "
1735			"Please free space, unmount and run xfs_repair.");
1736		} else {
1737			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1738		}
1739		return error;
1740	}
1741
1742	/*
1743	 * We do not hold the inode locked across the entire rolling transaction
1744	 * here. We only need to hold it for the first transaction that
1745	 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1746	 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1747	 * here breaks the relationship between cluster buffer invalidation and
1748	 * stale inode invalidation on cluster buffer item journal commit
1749	 * completion, and can result in leaving dirty stale inodes hanging
1750	 * around in memory.
1751	 *
1752	 * We have no need for serialising this inode operation against other
1753	 * operations - we freed the inode and hence reallocation is required
1754	 * and that will serialise on reallocating the space the deferops need
1755	 * to free. Hence we can unlock the inode on the first commit of
1756	 * the transaction rather than roll it right through the deferops. This
1757	 * avoids relogging the XFS_ISTALE inode.
1758	 *
1759	 * We check that xfs_ifree() hasn't grown an internal transaction roll
1760	 * by asserting that the inode is still locked when it returns.
1761	 */
1762	xfs_ilock(ip, XFS_ILOCK_EXCL);
1763	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1764
1765	error = xfs_ifree(tp, ip);
1766	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1767	if (error) {
1768		/*
1769		 * If we fail to free the inode, shut down.  The cancel
1770		 * might do that, we need to make sure.  Otherwise the
1771		 * inode might be lost for a long time or forever.
1772		 */
1773		if (!XFS_FORCED_SHUTDOWN(mp)) {
1774			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1775				__func__, error);
1776			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1777		}
1778		xfs_trans_cancel(tp);
 
1779		return error;
1780	}
1781
1782	/*
1783	 * Credit the quota account(s). The inode is gone.
1784	 */
1785	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1786
1787	/*
1788	 * Just ignore errors at this point.  There is nothing we can do except
1789	 * to try to keep going. Make sure it's not a silent error.
1790	 */
 
 
 
 
 
 
1791	error = xfs_trans_commit(tp);
1792	if (error)
1793		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1794			__func__, error);
1795
 
1796	return 0;
1797}
1798
1799/*
1800 * xfs_inactive
1801 *
1802 * This is called when the vnode reference count for the vnode
1803 * goes to zero.  If the file has been unlinked, then it must
1804 * now be truncated.  Also, we clear all of the read-ahead state
1805 * kept for the inode here since the file is now closed.
1806 */
1807void
1808xfs_inactive(
1809	xfs_inode_t	*ip)
1810{
1811	struct xfs_mount	*mp;
1812	int			error;
1813	int			truncate = 0;
1814
1815	/*
1816	 * If the inode is already free, then there can be nothing
1817	 * to clean up here.
1818	 */
1819	if (VFS_I(ip)->i_mode == 0) {
 
1820		ASSERT(ip->i_df.if_broot_bytes == 0);
1821		return;
1822	}
1823
1824	mp = ip->i_mount;
1825	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1826
1827	/* If this is a read-only mount, don't do this (would generate I/O) */
1828	if (mp->m_flags & XFS_MOUNT_RDONLY)
1829		return;
1830
1831	/* Try to clean out the cow blocks if there are any. */
1832	if (xfs_inode_has_cow_data(ip))
1833		xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1834
1835	if (VFS_I(ip)->i_nlink != 0) {
1836		/*
1837		 * force is true because we are evicting an inode from the
1838		 * cache. Post-eof blocks must be freed, lest we end up with
1839		 * broken free space accounting.
1840		 *
1841		 * Note: don't bother with iolock here since lockdep complains
1842		 * about acquiring it in reclaim context. We have the only
1843		 * reference to the inode at this point anyways.
1844		 */
1845		if (xfs_can_free_eofblocks(ip, true))
 
1846			xfs_free_eofblocks(ip);
 
 
1847
1848		return;
1849	}
1850
1851	if (S_ISREG(VFS_I(ip)->i_mode) &&
1852	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1853	     ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1854		truncate = 1;
1855
1856	error = xfs_qm_dqattach(ip);
1857	if (error)
1858		return;
1859
1860	if (S_ISLNK(VFS_I(ip)->i_mode))
1861		error = xfs_inactive_symlink(ip);
1862	else if (truncate)
1863		error = xfs_inactive_truncate(ip);
1864	if (error)
1865		return;
1866
1867	/*
1868	 * If there are attributes associated with the file then blow them away
1869	 * now.  The code calls a routine that recursively deconstructs the
1870	 * attribute fork. If also blows away the in-core attribute fork.
1871	 */
1872	if (XFS_IFORK_Q(ip)) {
1873		error = xfs_attr_inactive(ip);
1874		if (error)
1875			return;
1876	}
1877
1878	ASSERT(!ip->i_afp);
 
1879	ASSERT(ip->i_d.di_forkoff == 0);
1880
1881	/*
1882	 * Free the inode.
1883	 */
1884	error = xfs_inactive_ifree(ip);
1885	if (error)
1886		return;
1887
1888	/*
1889	 * Release the dquots held by inode, if any.
1890	 */
1891	xfs_qm_dqdetach(ip);
1892}
1893
1894/*
1895 * In-Core Unlinked List Lookups
1896 * =============================
1897 *
1898 * Every inode is supposed to be reachable from some other piece of metadata
1899 * with the exception of the root directory.  Inodes with a connection to a
1900 * file descriptor but not linked from anywhere in the on-disk directory tree
1901 * are collectively known as unlinked inodes, though the filesystem itself
1902 * maintains links to these inodes so that on-disk metadata are consistent.
1903 *
1904 * XFS implements a per-AG on-disk hash table of unlinked inodes.  The AGI
1905 * header contains a number of buckets that point to an inode, and each inode
1906 * record has a pointer to the next inode in the hash chain.  This
1907 * singly-linked list causes scaling problems in the iunlink remove function
1908 * because we must walk that list to find the inode that points to the inode
1909 * being removed from the unlinked hash bucket list.
1910 *
1911 * What if we modelled the unlinked list as a collection of records capturing
1912 * "X.next_unlinked = Y" relations?  If we indexed those records on Y, we'd
1913 * have a fast way to look up unlinked list predecessors, which avoids the
1914 * slow list walk.  That's exactly what we do here (in-core) with a per-AG
1915 * rhashtable.
1916 *
1917 * Because this is a backref cache, we ignore operational failures since the
1918 * iunlink code can fall back to the slow bucket walk.  The only errors that
1919 * should bubble out are for obviously incorrect situations.
1920 *
1921 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1922 * access or have otherwise provided for concurrency control.
1923 */
1924
1925/* Capture a "X.next_unlinked = Y" relationship. */
1926struct xfs_iunlink {
1927	struct rhash_head	iu_rhash_head;
1928	xfs_agino_t		iu_agino;		/* X */
1929	xfs_agino_t		iu_next_unlinked;	/* Y */
1930};
1931
1932/* Unlinked list predecessor lookup hashtable construction */
1933static int
1934xfs_iunlink_obj_cmpfn(
1935	struct rhashtable_compare_arg	*arg,
1936	const void			*obj)
1937{
1938	const xfs_agino_t		*key = arg->key;
1939	const struct xfs_iunlink	*iu = obj;
1940
1941	if (iu->iu_next_unlinked != *key)
1942		return 1;
1943	return 0;
1944}
1945
1946static const struct rhashtable_params xfs_iunlink_hash_params = {
1947	.min_size		= XFS_AGI_UNLINKED_BUCKETS,
1948	.key_len		= sizeof(xfs_agino_t),
1949	.key_offset		= offsetof(struct xfs_iunlink,
1950					   iu_next_unlinked),
1951	.head_offset		= offsetof(struct xfs_iunlink, iu_rhash_head),
1952	.automatic_shrinking	= true,
1953	.obj_cmpfn		= xfs_iunlink_obj_cmpfn,
1954};
1955
1956/*
1957 * Return X, where X.next_unlinked == @agino.  Returns NULLAGINO if no such
1958 * relation is found.
1959 */
1960static xfs_agino_t
1961xfs_iunlink_lookup_backref(
1962	struct xfs_perag	*pag,
1963	xfs_agino_t		agino)
1964{
1965	struct xfs_iunlink	*iu;
1966
1967	iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1968			xfs_iunlink_hash_params);
1969	return iu ? iu->iu_agino : NULLAGINO;
1970}
1971
1972/*
1973 * Take ownership of an iunlink cache entry and insert it into the hash table.
1974 * If successful, the entry will be owned by the cache; if not, it is freed.
1975 * Either way, the caller does not own @iu after this call.
1976 */
1977static int
1978xfs_iunlink_insert_backref(
1979	struct xfs_perag	*pag,
1980	struct xfs_iunlink	*iu)
1981{
1982	int			error;
1983
1984	error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1985			&iu->iu_rhash_head, xfs_iunlink_hash_params);
1986	/*
1987	 * Fail loudly if there already was an entry because that's a sign of
1988	 * corruption of in-memory data.  Also fail loudly if we see an error
1989	 * code we didn't anticipate from the rhashtable code.  Currently we
1990	 * only anticipate ENOMEM.
1991	 */
1992	if (error) {
1993		WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1994		kmem_free(iu);
1995	}
1996	/*
1997	 * Absorb any runtime errors that aren't a result of corruption because
1998	 * this is a cache and we can always fall back to bucket list scanning.
1999	 */
2000	if (error != 0 && error != -EEXIST)
2001		error = 0;
2002	return error;
2003}
2004
2005/* Remember that @prev_agino.next_unlinked = @this_agino. */
2006static int
2007xfs_iunlink_add_backref(
2008	struct xfs_perag	*pag,
2009	xfs_agino_t		prev_agino,
2010	xfs_agino_t		this_agino)
2011{
2012	struct xfs_iunlink	*iu;
2013
2014	if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
2015		return 0;
2016
2017	iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
2018	iu->iu_agino = prev_agino;
2019	iu->iu_next_unlinked = this_agino;
2020
2021	return xfs_iunlink_insert_backref(pag, iu);
2022}
2023
2024/*
2025 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
2026 * If @next_unlinked is NULLAGINO, we drop the backref and exit.  If there
2027 * wasn't any such entry then we don't bother.
2028 */
2029static int
2030xfs_iunlink_change_backref(
2031	struct xfs_perag	*pag,
2032	xfs_agino_t		agino,
2033	xfs_agino_t		next_unlinked)
2034{
2035	struct xfs_iunlink	*iu;
2036	int			error;
2037
2038	/* Look up the old entry; if there wasn't one then exit. */
2039	iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
2040			xfs_iunlink_hash_params);
2041	if (!iu)
2042		return 0;
2043
2044	/*
2045	 * Remove the entry.  This shouldn't ever return an error, but if we
2046	 * couldn't remove the old entry we don't want to add it again to the
2047	 * hash table, and if the entry disappeared on us then someone's
2048	 * violated the locking rules and we need to fail loudly.  Either way
2049	 * we cannot remove the inode because internal state is or would have
2050	 * been corrupt.
2051	 */
2052	error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
2053			&iu->iu_rhash_head, xfs_iunlink_hash_params);
2054	if (error)
2055		return error;
2056
2057	/* If there is no new next entry just free our item and return. */
2058	if (next_unlinked == NULLAGINO) {
2059		kmem_free(iu);
2060		return 0;
2061	}
2062
2063	/* Update the entry and re-add it to the hash table. */
2064	iu->iu_next_unlinked = next_unlinked;
2065	return xfs_iunlink_insert_backref(pag, iu);
2066}
2067
2068/* Set up the in-core predecessor structures. */
2069int
2070xfs_iunlink_init(
2071	struct xfs_perag	*pag)
2072{
2073	return rhashtable_init(&pag->pagi_unlinked_hash,
2074			&xfs_iunlink_hash_params);
2075}
2076
2077/* Free the in-core predecessor structures. */
2078static void
2079xfs_iunlink_free_item(
2080	void			*ptr,
2081	void			*arg)
2082{
2083	struct xfs_iunlink	*iu = ptr;
2084	bool			*freed_anything = arg;
2085
2086	*freed_anything = true;
2087	kmem_free(iu);
2088}
2089
2090void
2091xfs_iunlink_destroy(
2092	struct xfs_perag	*pag)
2093{
2094	bool			freed_anything = false;
2095
2096	rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
2097			xfs_iunlink_free_item, &freed_anything);
2098
2099	ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
2100}
2101
2102/*
2103 * Point the AGI unlinked bucket at an inode and log the results.  The caller
2104 * is responsible for validating the old value.
2105 */
2106STATIC int
2107xfs_iunlink_update_bucket(
2108	struct xfs_trans	*tp,
2109	xfs_agnumber_t		agno,
2110	struct xfs_buf		*agibp,
2111	unsigned int		bucket_index,
2112	xfs_agino_t		new_agino)
2113{
2114	struct xfs_agi		*agi = agibp->b_addr;
2115	xfs_agino_t		old_value;
2116	int			offset;
2117
2118	ASSERT(xfs_verify_agino_or_null(tp->t_mountp, agno, new_agino));
2119
2120	old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2121	trace_xfs_iunlink_update_bucket(tp->t_mountp, agno, bucket_index,
2122			old_value, new_agino);
2123
2124	/*
2125	 * We should never find the head of the list already set to the value
2126	 * passed in because either we're adding or removing ourselves from the
2127	 * head of the list.
2128	 */
2129	if (old_value == new_agino) {
2130		xfs_buf_mark_corrupt(agibp);
2131		return -EFSCORRUPTED;
2132	}
2133
2134	agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2135	offset = offsetof(struct xfs_agi, agi_unlinked) +
2136			(sizeof(xfs_agino_t) * bucket_index);
2137	xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2138	return 0;
2139}
2140
2141/* Set an on-disk inode's next_unlinked pointer. */
2142STATIC void
2143xfs_iunlink_update_dinode(
2144	struct xfs_trans	*tp,
2145	xfs_agnumber_t		agno,
2146	xfs_agino_t		agino,
2147	struct xfs_buf		*ibp,
2148	struct xfs_dinode	*dip,
2149	struct xfs_imap		*imap,
2150	xfs_agino_t		next_agino)
2151{
2152	struct xfs_mount	*mp = tp->t_mountp;
2153	int			offset;
2154
2155	ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2156
2157	trace_xfs_iunlink_update_dinode(mp, agno, agino,
2158			be32_to_cpu(dip->di_next_unlinked), next_agino);
2159
2160	dip->di_next_unlinked = cpu_to_be32(next_agino);
2161	offset = imap->im_boffset +
2162			offsetof(struct xfs_dinode, di_next_unlinked);
2163
2164	/* need to recalc the inode CRC if appropriate */
2165	xfs_dinode_calc_crc(mp, dip);
2166	xfs_trans_inode_buf(tp, ibp);
2167	xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2168}
2169
2170/* Set an in-core inode's unlinked pointer and return the old value. */
2171STATIC int
2172xfs_iunlink_update_inode(
2173	struct xfs_trans	*tp,
2174	struct xfs_inode	*ip,
2175	xfs_agnumber_t		agno,
2176	xfs_agino_t		next_agino,
2177	xfs_agino_t		*old_next_agino)
2178{
2179	struct xfs_mount	*mp = tp->t_mountp;
2180	struct xfs_dinode	*dip;
2181	struct xfs_buf		*ibp;
2182	xfs_agino_t		old_value;
2183	int			error;
2184
2185	ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2186
2187	error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 0);
2188	if (error)
2189		return error;
2190
2191	/* Make sure the old pointer isn't garbage. */
2192	old_value = be32_to_cpu(dip->di_next_unlinked);
2193	if (!xfs_verify_agino_or_null(mp, agno, old_value)) {
2194		xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2195				sizeof(*dip), __this_address);
2196		error = -EFSCORRUPTED;
2197		goto out;
2198	}
2199
2200	/*
2201	 * Since we're updating a linked list, we should never find that the
2202	 * current pointer is the same as the new value, unless we're
2203	 * terminating the list.
2204	 */
2205	*old_next_agino = old_value;
2206	if (old_value == next_agino) {
2207		if (next_agino != NULLAGINO) {
2208			xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2209					dip, sizeof(*dip), __this_address);
2210			error = -EFSCORRUPTED;
2211		}
2212		goto out;
2213	}
2214
2215	/* Ok, update the new pointer. */
2216	xfs_iunlink_update_dinode(tp, agno, XFS_INO_TO_AGINO(mp, ip->i_ino),
2217			ibp, dip, &ip->i_imap, next_agino);
2218	return 0;
2219out:
2220	xfs_trans_brelse(tp, ibp);
2221	return error;
2222}
2223
2224/*
2225 * This is called when the inode's link count has gone to 0 or we are creating
2226 * a tmpfile via O_TMPFILE.  The inode @ip must have nlink == 0.
2227 *
2228 * We place the on-disk inode on a list in the AGI.  It will be pulled from this
2229 * list when the inode is freed.
2230 */
2231STATIC int
2232xfs_iunlink(
2233	struct xfs_trans	*tp,
2234	struct xfs_inode	*ip)
2235{
2236	struct xfs_mount	*mp = tp->t_mountp;
2237	struct xfs_agi		*agi;
2238	struct xfs_buf		*agibp;
2239	xfs_agino_t		next_agino;
2240	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2241	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2242	short			bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2243	int			error;
 
2244
2245	ASSERT(VFS_I(ip)->i_nlink == 0);
2246	ASSERT(VFS_I(ip)->i_mode != 0);
2247	trace_xfs_iunlink(ip);
2248
2249	/* Get the agi buffer first.  It ensures lock ordering on the list. */
2250	error = xfs_read_agi(mp, tp, agno, &agibp);
 
 
 
2251	if (error)
2252		return error;
2253	agi = agibp->b_addr;
2254
2255	/*
2256	 * Get the index into the agi hash table for the list this inode will
2257	 * go on.  Make sure the pointer isn't garbage and that this inode
2258	 * isn't already on the list.
2259	 */
2260	next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2261	if (next_agino == agino ||
2262	    !xfs_verify_agino_or_null(mp, agno, next_agino)) {
2263		xfs_buf_mark_corrupt(agibp);
2264		return -EFSCORRUPTED;
2265	}
2266
2267	if (next_agino != NULLAGINO) {
2268		xfs_agino_t		old_agino;
2269
 
2270		/*
2271		 * There is already another inode in the bucket, so point this
2272		 * inode to the current head of the list.
 
 
2273		 */
2274		error = xfs_iunlink_update_inode(tp, ip, agno, next_agino,
2275				&old_agino);
2276		if (error)
2277			return error;
2278		ASSERT(old_agino == NULLAGINO);
2279
2280		/*
2281		 * agino has been unlinked, add a backref from the next inode
2282		 * back to agino.
2283		 */
2284		error = xfs_iunlink_add_backref(agibp->b_pag, agino, next_agino);
2285		if (error)
2286			return error;
2287	}
2288
2289	/* Point the head of the list to point to this inode. */
2290	return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index, agino);
2291}
2292
2293/* Return the imap, dinode pointer, and buffer for an inode. */
2294STATIC int
2295xfs_iunlink_map_ino(
2296	struct xfs_trans	*tp,
2297	xfs_agnumber_t		agno,
2298	xfs_agino_t		agino,
2299	struct xfs_imap		*imap,
2300	struct xfs_dinode	**dipp,
2301	struct xfs_buf		**bpp)
2302{
2303	struct xfs_mount	*mp = tp->t_mountp;
2304	int			error;
2305
2306	imap->im_blkno = 0;
2307	error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2308	if (error) {
2309		xfs_warn(mp, "%s: xfs_imap returned error %d.",
2310				__func__, error);
2311		return error;
2312	}
2313
2314	error = xfs_imap_to_bp(mp, tp, imap, dipp, bpp, 0);
2315	if (error) {
2316		xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2317				__func__, error);
2318		return error;
2319	}
2320
2321	return 0;
2322}
2323
2324/*
2325 * Walk the unlinked chain from @head_agino until we find the inode that
2326 * points to @target_agino.  Return the inode number, map, dinode pointer,
2327 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2328 *
2329 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2330 * @agino, @imap, @dipp, and @bpp are all output parameters.
2331 *
2332 * Do not call this function if @target_agino is the head of the list.
2333 */
2334STATIC int
2335xfs_iunlink_map_prev(
2336	struct xfs_trans	*tp,
2337	xfs_agnumber_t		agno,
2338	xfs_agino_t		head_agino,
2339	xfs_agino_t		target_agino,
2340	xfs_agino_t		*agino,
2341	struct xfs_imap		*imap,
2342	struct xfs_dinode	**dipp,
2343	struct xfs_buf		**bpp,
2344	struct xfs_perag	*pag)
2345{
2346	struct xfs_mount	*mp = tp->t_mountp;
2347	xfs_agino_t		next_agino;
2348	int			error;
2349
2350	ASSERT(head_agino != target_agino);
2351	*bpp = NULL;
2352
2353	/* See if our backref cache can find it faster. */
2354	*agino = xfs_iunlink_lookup_backref(pag, target_agino);
2355	if (*agino != NULLAGINO) {
2356		error = xfs_iunlink_map_ino(tp, agno, *agino, imap, dipp, bpp);
2357		if (error)
2358			return error;
2359
2360		if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2361			return 0;
2362
2363		/*
2364		 * If we get here the cache contents were corrupt, so drop the
2365		 * buffer and fall back to walking the bucket list.
2366		 */
2367		xfs_trans_brelse(tp, *bpp);
2368		*bpp = NULL;
2369		WARN_ON_ONCE(1);
2370	}
2371
2372	trace_xfs_iunlink_map_prev_fallback(mp, agno);
2373
2374	/* Otherwise, walk the entire bucket until we find it. */
2375	next_agino = head_agino;
2376	while (next_agino != target_agino) {
2377		xfs_agino_t	unlinked_agino;
2378
2379		if (*bpp)
2380			xfs_trans_brelse(tp, *bpp);
 
 
2381
2382		*agino = next_agino;
2383		error = xfs_iunlink_map_ino(tp, agno, next_agino, imap, dipp,
2384				bpp);
2385		if (error)
2386			return error;
2387
2388		unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2389		/*
2390		 * Make sure this pointer is valid and isn't an obvious
2391		 * infinite loop.
2392		 */
2393		if (!xfs_verify_agino(mp, agno, unlinked_agino) ||
2394		    next_agino == unlinked_agino) {
2395			XFS_CORRUPTION_ERROR(__func__,
2396					XFS_ERRLEVEL_LOW, mp,
2397					*dipp, sizeof(**dipp));
2398			error = -EFSCORRUPTED;
2399			return error;
2400		}
2401		next_agino = unlinked_agino;
2402	}
2403
 
 
 
 
 
 
 
 
 
2404	return 0;
2405}
2406
2407/*
2408 * Pull the on-disk inode from the AGI unlinked list.
2409 */
2410STATIC int
2411xfs_iunlink_remove(
2412	struct xfs_trans	*tp,
2413	struct xfs_inode	*ip)
2414{
2415	struct xfs_mount	*mp = tp->t_mountp;
2416	struct xfs_agi		*agi;
2417	struct xfs_buf		*agibp;
2418	struct xfs_buf		*last_ibp;
2419	struct xfs_dinode	*last_dip = NULL;
2420	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2421	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2422	xfs_agino_t		next_agino;
2423	xfs_agino_t		head_agino;
2424	short			bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2425	int			error;
2426
2427	trace_xfs_iunlink_remove(ip);
2428
2429	/* Get the agi buffer first.  It ensures lock ordering on the list. */
2430	error = xfs_read_agi(mp, tp, agno, &agibp);
2431	if (error)
2432		return error;
2433	agi = agibp->b_addr;
2434
2435	/*
2436	 * Get the index into the agi hash table for the list this inode will
2437	 * go on.  Make sure the head pointer isn't garbage.
2438	 */
2439	head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2440	if (!xfs_verify_agino(mp, agno, head_agino)) {
2441		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2442				agi, sizeof(*agi));
2443		return -EFSCORRUPTED;
2444	}
2445
2446	/*
2447	 * Set our inode's next_unlinked pointer to NULL and then return
2448	 * the old pointer value so that we can update whatever was previous
2449	 * to us in the list to point to whatever was next in the list.
2450	 */
2451	error = xfs_iunlink_update_inode(tp, ip, agno, NULLAGINO, &next_agino);
2452	if (error)
2453		return error;
2454
 
 
2455	/*
2456	 * If there was a backref pointing from the next inode back to this
2457	 * one, remove it because we've removed this inode from the list.
2458	 *
2459	 * Later, if this inode was in the middle of the list we'll update
2460	 * this inode's backref to point from the next inode.
2461	 */
2462	if (next_agino != NULLAGINO) {
2463		error = xfs_iunlink_change_backref(agibp->b_pag, next_agino,
2464				NULLAGINO);
2465		if (error)
2466			return error;
2467	}
2468
2469	if (head_agino != agino) {
2470		struct xfs_imap	imap;
2471		xfs_agino_t	prev_agino;
2472
2473		/* We need to search the list for the inode being freed. */
2474		error = xfs_iunlink_map_prev(tp, agno, head_agino, agino,
2475				&prev_agino, &imap, &last_dip, &last_ibp,
2476				agibp->b_pag);
2477		if (error)
 
 
 
2478			return error;
2479
2480		/* Point the previous inode on the list to the next inode. */
2481		xfs_iunlink_update_dinode(tp, agno, prev_agino, last_ibp,
2482				last_dip, &imap, next_agino);
2483
 
 
 
 
 
 
 
 
 
 
 
 
 
2484		/*
2485		 * Now we deal with the backref for this inode.  If this inode
2486		 * pointed at a real inode, change the backref that pointed to
2487		 * us to point to our old next.  If this inode was the end of
2488		 * the list, delete the backref that pointed to us.  Note that
2489		 * change_backref takes care of deleting the backref if
2490		 * next_agino is NULLAGINO.
2491		 */
2492		return xfs_iunlink_change_backref(agibp->b_pag, agino,
2493				next_agino);
2494	}
2495
2496	/* Point the head of the list to the next unlinked inode. */
2497	return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index,
2498			next_agino);
2499}
2500
2501/*
2502 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2503 * mark it stale. We should only find clean inodes in this lookup that aren't
2504 * already stale.
2505 */
2506static void
2507xfs_ifree_mark_inode_stale(
2508	struct xfs_buf		*bp,
2509	struct xfs_inode	*free_ip,
2510	xfs_ino_t		inum)
2511{
2512	struct xfs_mount	*mp = bp->b_mount;
2513	struct xfs_perag	*pag = bp->b_pag;
2514	struct xfs_inode_log_item *iip;
2515	struct xfs_inode	*ip;
2516
2517retry:
2518	rcu_read_lock();
2519	ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
 
2520
2521	/* Inode not in memory, nothing to do */
2522	if (!ip) {
2523		rcu_read_unlock();
2524		return;
2525	}
 
 
 
2526
2527	/*
2528	 * because this is an RCU protected lookup, we could find a recently
2529	 * freed or even reallocated inode during the lookup. We need to check
2530	 * under the i_flags_lock for a valid inode here. Skip it if it is not
2531	 * valid, the wrong inode or stale.
2532	 */
2533	spin_lock(&ip->i_flags_lock);
2534	if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE)) {
2535		spin_unlock(&ip->i_flags_lock);
2536		rcu_read_unlock();
2537		return;
2538	}
2539
2540	/*
2541	 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2542	 * other inodes that we did not find in the list attached to the buffer
2543	 * and are not already marked stale. If we can't lock it, back off and
2544	 * retry.
2545	 */
2546	if (ip != free_ip) {
2547		if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2548			spin_unlock(&ip->i_flags_lock);
2549			rcu_read_unlock();
2550			delay(1);
2551			goto retry;
2552		}
2553	}
2554	ip->i_flags |= XFS_ISTALE;
2555	spin_unlock(&ip->i_flags_lock);
2556	rcu_read_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2557
2558	/*
2559	 * If we can't get the flush lock, the inode is already attached.  All
2560	 * we needed to do here is mark the inode stale so buffer IO completion
2561	 * will remove it from the AIL.
2562	 */
2563	iip = ip->i_itemp;
2564	if (!xfs_iflock_nowait(ip)) {
2565		ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2566		ASSERT(iip->ili_last_fields);
2567		goto out_iunlock;
2568	}
2569
2570	/*
2571	 * Inodes not attached to the buffer can be released immediately.
2572	 * Everything else has to go through xfs_iflush_abort() on journal
2573	 * commit as the flock synchronises removal of the inode from the
2574	 * cluster buffer against inode reclaim.
2575	 */
2576	if (!iip || list_empty(&iip->ili_item.li_bio_list)) {
2577		xfs_ifunlock(ip);
2578		goto out_iunlock;
2579	}
2580
2581	/* we have a dirty inode in memory that has not yet been flushed. */
2582	spin_lock(&iip->ili_lock);
2583	iip->ili_last_fields = iip->ili_fields;
2584	iip->ili_fields = 0;
2585	iip->ili_fsync_fields = 0;
2586	spin_unlock(&iip->ili_lock);
2587	ASSERT(iip->ili_last_fields);
2588
2589out_iunlock:
2590	if (ip != free_ip)
2591		xfs_iunlock(ip, XFS_ILOCK_EXCL);
2592}
2593
2594/*
2595 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2596 * inodes that are in memory - they all must be marked stale and attached to
2597 * the cluster buffer.
2598 */
2599STATIC int
2600xfs_ifree_cluster(
2601	struct xfs_inode	*free_ip,
2602	struct xfs_trans	*tp,
2603	struct xfs_icluster	*xic)
2604{
2605	struct xfs_mount	*mp = free_ip->i_mount;
2606	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
2607	struct xfs_buf		*bp;
2608	xfs_daddr_t		blkno;
2609	xfs_ino_t		inum = xic->first_ino;
2610	int			nbufs;
2611	int			i, j;
2612	int			ioffset;
2613	int			error;
 
 
 
 
 
 
2614
2615	nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
 
 
 
 
2616
2617	for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2618		/*
2619		 * The allocation bitmap tells us which inodes of the chunk were
2620		 * physically allocated. Skip the cluster if an inode falls into
2621		 * a sparse region.
2622		 */
2623		ioffset = inum - xic->first_ino;
2624		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2625			ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2626			continue;
2627		}
2628
2629		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2630					 XFS_INO_TO_AGBNO(mp, inum));
2631
2632		/*
2633		 * We obtain and lock the backing buffer first in the process
2634		 * here, as we have to ensure that any dirty inode that we
2635		 * can't get the flush lock on is attached to the buffer.
2636		 * If we scan the in-memory inodes first, then buffer IO can
2637		 * complete before we get a lock on it, and hence we may fail
2638		 * to mark all the active inodes on the buffer stale.
2639		 */
2640		error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2641				mp->m_bsize * igeo->blocks_per_cluster,
2642				XBF_UNMAPPED, &bp);
2643		if (error)
2644			return error;
 
2645
2646		/*
2647		 * This buffer may not have been correctly initialised as we
2648		 * didn't read it from disk. That's not important because we are
2649		 * only using to mark the buffer as stale in the log, and to
2650		 * attach stale cached inodes on it. That means it will never be
2651		 * dispatched for IO. If it is, we want to know about it, and we
2652		 * want it to fail. We can acheive this by adding a write
2653		 * verifier to the buffer.
2654		 */
2655		bp->b_ops = &xfs_inode_buf_ops;
2656
2657		/*
2658		 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2659		 * too. This requires lookups, and will skip inodes that we've
2660		 * already marked XFS_ISTALE.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2661		 */
2662		for (i = 0; i < igeo->inodes_per_cluster; i++)
2663			xfs_ifree_mark_inode_stale(bp, free_ip, inum + i);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2664
2665		xfs_trans_stale_inode_buf(tp, bp);
2666		xfs_trans_binval(tp, bp);
2667	}
 
 
2668	return 0;
2669}
2670
2671/*
2672 * This is called to return an inode to the inode free list.
2673 * The inode should already be truncated to 0 length and have
2674 * no pages associated with it.  This routine also assumes that
2675 * the inode is already a part of the transaction.
2676 *
2677 * The on-disk copy of the inode will have been added to the list
2678 * of unlinked inodes in the AGI. We need to remove the inode from
2679 * that list atomically with respect to freeing it here.
2680 */
2681int
2682xfs_ifree(
2683	struct xfs_trans	*tp,
2684	struct xfs_inode	*ip)
 
2685{
2686	int			error;
2687	struct xfs_icluster	xic = { 0 };
2688	struct xfs_inode_log_item *iip = ip->i_itemp;
2689
2690	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2691	ASSERT(VFS_I(ip)->i_nlink == 0);
2692	ASSERT(ip->i_df.if_nextents == 0);
 
2693	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2694	ASSERT(ip->i_d.di_nblocks == 0);
2695
2696	/*
2697	 * Pull the on-disk inode from the AGI unlinked list.
2698	 */
2699	error = xfs_iunlink_remove(tp, ip);
2700	if (error)
2701		return error;
2702
2703	error = xfs_difree(tp, ip->i_ino, &xic);
2704	if (error)
2705		return error;
2706
2707	/*
2708	 * Free any local-format data sitting around before we reset the
2709	 * data fork to extents format.  Note that the attr fork data has
2710	 * already been freed by xfs_attr_inactive.
2711	 */
2712	if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2713		kmem_free(ip->i_df.if_u1.if_data);
2714		ip->i_df.if_u1.if_data = NULL;
2715		ip->i_df.if_bytes = 0;
2716	}
2717
2718	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2719	ip->i_d.di_flags = 0;
2720	ip->i_d.di_flags2 = 0;
2721	ip->i_d.di_dmevmask = 0;
2722	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2723	ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2724
2725	/* Don't attempt to replay owner changes for a deleted inode */
2726	spin_lock(&iip->ili_lock);
2727	iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2728	spin_unlock(&iip->ili_lock);
2729
2730	/*
2731	 * Bump the generation count so no one will be confused
2732	 * by reincarnations of this inode.
2733	 */
2734	VFS_I(ip)->i_generation++;
2735	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2736
2737	if (xic.deleted)
2738		error = xfs_ifree_cluster(ip, tp, &xic);
2739
2740	return error;
2741}
2742
2743/*
2744 * This is called to unpin an inode.  The caller must have the inode locked
2745 * in at least shared mode so that the buffer cannot be subsequently pinned
2746 * once someone is waiting for it to be unpinned.
2747 */
2748static void
2749xfs_iunpin(
2750	struct xfs_inode	*ip)
2751{
2752	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2753
2754	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2755
2756	/* Give the log a push to start the unpinning I/O */
2757	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2758
2759}
2760
2761static void
2762__xfs_iunpin_wait(
2763	struct xfs_inode	*ip)
2764{
2765	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2766	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2767
2768	xfs_iunpin(ip);
2769
2770	do {
2771		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2772		if (xfs_ipincount(ip))
2773			io_schedule();
2774	} while (xfs_ipincount(ip));
2775	finish_wait(wq, &wait.wq_entry);
2776}
2777
2778void
2779xfs_iunpin_wait(
2780	struct xfs_inode	*ip)
2781{
2782	if (xfs_ipincount(ip))
2783		__xfs_iunpin_wait(ip);
2784}
2785
2786/*
2787 * Removing an inode from the namespace involves removing the directory entry
2788 * and dropping the link count on the inode. Removing the directory entry can
2789 * result in locking an AGF (directory blocks were freed) and removing a link
2790 * count can result in placing the inode on an unlinked list which results in
2791 * locking an AGI.
2792 *
2793 * The big problem here is that we have an ordering constraint on AGF and AGI
2794 * locking - inode allocation locks the AGI, then can allocate a new extent for
2795 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2796 * removes the inode from the unlinked list, requiring that we lock the AGI
2797 * first, and then freeing the inode can result in an inode chunk being freed
2798 * and hence freeing disk space requiring that we lock an AGF.
2799 *
2800 * Hence the ordering that is imposed by other parts of the code is AGI before
2801 * AGF. This means we cannot remove the directory entry before we drop the inode
2802 * reference count and put it on the unlinked list as this results in a lock
2803 * order of AGF then AGI, and this can deadlock against inode allocation and
2804 * freeing. Therefore we must drop the link counts before we remove the
2805 * directory entry.
2806 *
2807 * This is still safe from a transactional point of view - it is not until we
2808 * get to xfs_defer_finish() that we have the possibility of multiple
2809 * transactions in this operation. Hence as long as we remove the directory
2810 * entry and drop the link count in the first transaction of the remove
2811 * operation, there are no transactional constraints on the ordering here.
2812 */
2813int
2814xfs_remove(
2815	xfs_inode_t             *dp,
2816	struct xfs_name		*name,
2817	xfs_inode_t		*ip)
2818{
2819	xfs_mount_t		*mp = dp->i_mount;
2820	xfs_trans_t             *tp = NULL;
2821	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2822	int                     error = 0;
 
 
2823	uint			resblks;
2824
2825	trace_xfs_remove(dp, name);
2826
2827	if (XFS_FORCED_SHUTDOWN(mp))
2828		return -EIO;
2829
2830	error = xfs_qm_dqattach(dp);
2831	if (error)
2832		goto std_return;
2833
2834	error = xfs_qm_dqattach(ip);
2835	if (error)
2836		goto std_return;
2837
2838	/*
2839	 * We try to get the real space reservation first,
2840	 * allowing for directory btree deletion(s) implying
2841	 * possible bmap insert(s).  If we can't get the space
2842	 * reservation then we use 0 instead, and avoid the bmap
2843	 * btree insert(s) in the directory code by, if the bmap
2844	 * insert tries to happen, instead trimming the LAST
2845	 * block from the directory.
2846	 */
2847	resblks = XFS_REMOVE_SPACE_RES(mp);
2848	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2849	if (error == -ENOSPC) {
2850		resblks = 0;
2851		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2852				&tp);
2853	}
2854	if (error) {
2855		ASSERT(error != -ENOSPC);
2856		goto std_return;
2857	}
2858
2859	xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2860
2861	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2862	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2863
2864	/*
2865	 * If we're removing a directory perform some additional validation.
2866	 */
2867	if (is_dir) {
2868		ASSERT(VFS_I(ip)->i_nlink >= 2);
2869		if (VFS_I(ip)->i_nlink != 2) {
2870			error = -ENOTEMPTY;
2871			goto out_trans_cancel;
2872		}
2873		if (!xfs_dir_isempty(ip)) {
2874			error = -ENOTEMPTY;
2875			goto out_trans_cancel;
2876		}
2877
2878		/* Drop the link from ip's "..".  */
2879		error = xfs_droplink(tp, dp);
2880		if (error)
2881			goto out_trans_cancel;
2882
2883		/* Drop the "." link from ip to self.  */
2884		error = xfs_droplink(tp, ip);
2885		if (error)
2886			goto out_trans_cancel;
2887	} else {
2888		/*
2889		 * When removing a non-directory we need to log the parent
2890		 * inode here.  For a directory this is done implicitly
2891		 * by the xfs_droplink call for the ".." entry.
2892		 */
2893		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2894	}
2895	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2896
2897	/* Drop the link from dp to ip. */
2898	error = xfs_droplink(tp, ip);
2899	if (error)
2900		goto out_trans_cancel;
2901
2902	error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
 
 
2903	if (error) {
2904		ASSERT(error != -ENOENT);
2905		goto out_trans_cancel;
2906	}
2907
2908	/*
2909	 * If this is a synchronous mount, make sure that the
2910	 * remove transaction goes to disk before returning to
2911	 * the user.
2912	 */
2913	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2914		xfs_trans_set_sync(tp);
2915
 
 
 
 
2916	error = xfs_trans_commit(tp);
2917	if (error)
2918		goto std_return;
2919
2920	if (is_dir && xfs_inode_is_filestream(ip))
2921		xfs_filestream_deassociate(ip);
2922
2923	return 0;
2924
 
 
2925 out_trans_cancel:
2926	xfs_trans_cancel(tp);
2927 std_return:
2928	return error;
2929}
2930
2931/*
2932 * Enter all inodes for a rename transaction into a sorted array.
2933 */
2934#define __XFS_SORT_INODES	5
2935STATIC void
2936xfs_sort_for_rename(
2937	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2938	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2939	struct xfs_inode	*ip1,	/* in: inode of old entry */
2940	struct xfs_inode	*ip2,	/* in: inode of new entry */
2941	struct xfs_inode	*wip,	/* in: whiteout inode */
2942	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2943	int			*num_inodes)  /* in/out: inodes in array */
2944{
2945	int			i, j;
2946
2947	ASSERT(*num_inodes == __XFS_SORT_INODES);
2948	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2949
2950	/*
2951	 * i_tab contains a list of pointers to inodes.  We initialize
2952	 * the table here & we'll sort it.  We will then use it to
2953	 * order the acquisition of the inode locks.
2954	 *
2955	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2956	 */
2957	i = 0;
2958	i_tab[i++] = dp1;
2959	i_tab[i++] = dp2;
2960	i_tab[i++] = ip1;
2961	if (ip2)
2962		i_tab[i++] = ip2;
2963	if (wip)
2964		i_tab[i++] = wip;
2965	*num_inodes = i;
2966
2967	/*
2968	 * Sort the elements via bubble sort.  (Remember, there are at
2969	 * most 5 elements to sort, so this is adequate.)
2970	 */
2971	for (i = 0; i < *num_inodes; i++) {
2972		for (j = 1; j < *num_inodes; j++) {
2973			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2974				struct xfs_inode *temp = i_tab[j];
2975				i_tab[j] = i_tab[j-1];
2976				i_tab[j-1] = temp;
2977			}
2978		}
2979	}
2980}
2981
2982static int
2983xfs_finish_rename(
2984	struct xfs_trans	*tp)
 
2985{
 
 
2986	/*
2987	 * If this is a synchronous mount, make sure that the rename transaction
2988	 * goes to disk before returning to the user.
2989	 */
2990	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2991		xfs_trans_set_sync(tp);
2992
 
 
 
 
 
 
 
2993	return xfs_trans_commit(tp);
2994}
2995
2996/*
2997 * xfs_cross_rename()
2998 *
2999 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
3000 */
3001STATIC int
3002xfs_cross_rename(
3003	struct xfs_trans	*tp,
3004	struct xfs_inode	*dp1,
3005	struct xfs_name		*name1,
3006	struct xfs_inode	*ip1,
3007	struct xfs_inode	*dp2,
3008	struct xfs_name		*name2,
3009	struct xfs_inode	*ip2,
 
 
3010	int			spaceres)
3011{
3012	int		error = 0;
3013	int		ip1_flags = 0;
3014	int		ip2_flags = 0;
3015	int		dp2_flags = 0;
3016
3017	/* Swap inode number for dirent in first parent */
3018	error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
 
 
3019	if (error)
3020		goto out_trans_abort;
3021
3022	/* Swap inode number for dirent in second parent */
3023	error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
 
 
3024	if (error)
3025		goto out_trans_abort;
3026
3027	/*
3028	 * If we're renaming one or more directories across different parents,
3029	 * update the respective ".." entries (and link counts) to match the new
3030	 * parents.
3031	 */
3032	if (dp1 != dp2) {
3033		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3034
3035		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
3036			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
3037						dp1->i_ino, spaceres);
 
3038			if (error)
3039				goto out_trans_abort;
3040
3041			/* transfer ip2 ".." reference to dp1 */
3042			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
3043				error = xfs_droplink(tp, dp2);
3044				if (error)
3045					goto out_trans_abort;
3046				xfs_bumplink(tp, dp1);
 
 
3047			}
3048
3049			/*
3050			 * Although ip1 isn't changed here, userspace needs
3051			 * to be warned about the change, so that applications
3052			 * relying on it (like backup ones), will properly
3053			 * notify the change
3054			 */
3055			ip1_flags |= XFS_ICHGTIME_CHG;
3056			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3057		}
3058
3059		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
3060			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
3061						dp2->i_ino, spaceres);
 
3062			if (error)
3063				goto out_trans_abort;
3064
3065			/* transfer ip1 ".." reference to dp2 */
3066			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
3067				error = xfs_droplink(tp, dp1);
3068				if (error)
3069					goto out_trans_abort;
3070				xfs_bumplink(tp, dp2);
 
 
3071			}
3072
3073			/*
3074			 * Although ip2 isn't changed here, userspace needs
3075			 * to be warned about the change, so that applications
3076			 * relying on it (like backup ones), will properly
3077			 * notify the change
3078			 */
3079			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3080			ip2_flags |= XFS_ICHGTIME_CHG;
3081		}
3082	}
3083
3084	if (ip1_flags) {
3085		xfs_trans_ichgtime(tp, ip1, ip1_flags);
3086		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
3087	}
3088	if (ip2_flags) {
3089		xfs_trans_ichgtime(tp, ip2, ip2_flags);
3090		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
3091	}
3092	if (dp2_flags) {
3093		xfs_trans_ichgtime(tp, dp2, dp2_flags);
3094		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
3095	}
3096	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3097	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
3098	return xfs_finish_rename(tp);
3099
3100out_trans_abort:
 
3101	xfs_trans_cancel(tp);
3102	return error;
3103}
3104
3105/*
3106 * xfs_rename_alloc_whiteout()
3107 *
3108 * Return a referenced, unlinked, unlocked inode that can be used as a
3109 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3110 * crash between allocating the inode and linking it into the rename transaction
3111 * recovery will free the inode and we won't leak it.
3112 */
3113static int
3114xfs_rename_alloc_whiteout(
3115	struct xfs_inode	*dp,
3116	struct xfs_inode	**wip)
3117{
3118	struct xfs_inode	*tmpfile;
3119	int			error;
3120
3121	error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile);
3122	if (error)
3123		return error;
3124
3125	/*
3126	 * Prepare the tmpfile inode as if it were created through the VFS.
3127	 * Complete the inode setup and flag it as linkable.  nlink is already
3128	 * zero, so we can skip the drop_nlink.
 
3129	 */
 
3130	xfs_setup_iops(tmpfile);
3131	xfs_finish_inode_setup(tmpfile);
3132	VFS_I(tmpfile)->i_state |= I_LINKABLE;
3133
3134	*wip = tmpfile;
3135	return 0;
3136}
3137
3138/*
3139 * xfs_rename
3140 */
3141int
3142xfs_rename(
3143	struct xfs_inode	*src_dp,
3144	struct xfs_name		*src_name,
3145	struct xfs_inode	*src_ip,
3146	struct xfs_inode	*target_dp,
3147	struct xfs_name		*target_name,
3148	struct xfs_inode	*target_ip,
3149	unsigned int		flags)
3150{
3151	struct xfs_mount	*mp = src_dp->i_mount;
3152	struct xfs_trans	*tp;
 
 
3153	struct xfs_inode	*wip = NULL;		/* whiteout inode */
3154	struct xfs_inode	*inodes[__XFS_SORT_INODES];
3155	struct xfs_buf		*agibp;
3156	int			num_inodes = __XFS_SORT_INODES;
3157	bool			new_parent = (src_dp != target_dp);
3158	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3159	int			spaceres;
3160	int			error;
3161
3162	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3163
3164	if ((flags & RENAME_EXCHANGE) && !target_ip)
3165		return -EINVAL;
3166
3167	/*
3168	 * If we are doing a whiteout operation, allocate the whiteout inode
3169	 * we will be placing at the target and ensure the type is set
3170	 * appropriately.
3171	 */
3172	if (flags & RENAME_WHITEOUT) {
3173		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3174		error = xfs_rename_alloc_whiteout(target_dp, &wip);
3175		if (error)
3176			return error;
3177
3178		/* setup target dirent info as whiteout */
3179		src_name->type = XFS_DIR3_FT_CHRDEV;
3180	}
3181
3182	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3183				inodes, &num_inodes);
3184
3185	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3186	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3187	if (error == -ENOSPC) {
3188		spaceres = 0;
3189		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3190				&tp);
3191	}
3192	if (error)
3193		goto out_release_wip;
3194
3195	/*
3196	 * Attach the dquots to the inodes
3197	 */
3198	error = xfs_qm_vop_rename_dqattach(inodes);
3199	if (error)
3200		goto out_trans_cancel;
3201
3202	/*
3203	 * Lock all the participating inodes. Depending upon whether
3204	 * the target_name exists in the target directory, and
3205	 * whether the target directory is the same as the source
3206	 * directory, we can lock from 2 to 4 inodes.
3207	 */
3208	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3209
3210	/*
3211	 * Join all the inodes to the transaction. From this point on,
3212	 * we can rely on either trans_commit or trans_cancel to unlock
3213	 * them.
3214	 */
3215	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3216	if (new_parent)
3217		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3218	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3219	if (target_ip)
3220		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3221	if (wip)
3222		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3223
3224	/*
3225	 * If we are using project inheritance, we only allow renames
3226	 * into our tree when the project IDs are the same; else the
3227	 * tree quota mechanism would be circumvented.
3228	 */
3229	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3230		     target_dp->i_d.di_projid != src_ip->i_d.di_projid)) {
3231		error = -EXDEV;
3232		goto out_trans_cancel;
3233	}
3234
 
 
3235	/* RENAME_EXCHANGE is unique from here on. */
3236	if (flags & RENAME_EXCHANGE)
3237		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3238					target_dp, target_name, target_ip,
3239					spaceres);
3240
3241	/*
3242	 * Check for expected errors before we dirty the transaction
3243	 * so we can return an error without a transaction abort.
3244	 */
3245	if (target_ip == NULL) {
3246		/*
3247		 * If there's no space reservation, check the entry will
3248		 * fit before actually inserting it.
3249		 */
3250		if (!spaceres) {
3251			error = xfs_dir_canenter(tp, target_dp, target_name);
3252			if (error)
3253				goto out_trans_cancel;
3254		}
3255	} else {
3256		/*
3257		 * If target exists and it's a directory, check that whether
3258		 * it can be destroyed.
3259		 */
3260		if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3261		    (!xfs_dir_isempty(target_ip) ||
3262		     (VFS_I(target_ip)->i_nlink > 2))) {
3263			error = -EEXIST;
3264			goto out_trans_cancel;
3265		}
3266	}
3267
3268	/*
3269	 * Directory entry creation below may acquire the AGF. Remove
3270	 * the whiteout from the unlinked list first to preserve correct
3271	 * AGI/AGF locking order. This dirties the transaction so failures
3272	 * after this point will abort and log recovery will clean up the
3273	 * mess.
3274	 *
3275	 * For whiteouts, we need to bump the link count on the whiteout
3276	 * inode. After this point, we have a real link, clear the tmpfile
3277	 * state flag from the inode so it doesn't accidentally get misused
3278	 * in future.
3279	 */
3280	if (wip) {
3281		ASSERT(VFS_I(wip)->i_nlink == 0);
3282		error = xfs_iunlink_remove(tp, wip);
3283		if (error)
3284			goto out_trans_cancel;
3285
3286		xfs_bumplink(tp, wip);
3287		VFS_I(wip)->i_state &= ~I_LINKABLE;
3288	}
3289
3290	/*
3291	 * Set up the target.
3292	 */
3293	if (target_ip == NULL) {
3294		/*
3295		 * If target does not exist and the rename crosses
3296		 * directories, adjust the target directory link count
3297		 * to account for the ".." reference from the new entry.
3298		 */
3299		error = xfs_dir_createname(tp, target_dp, target_name,
3300					   src_ip->i_ino, spaceres);
 
3301		if (error)
3302			goto out_trans_cancel;
3303
3304		xfs_trans_ichgtime(tp, target_dp,
3305					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3306
3307		if (new_parent && src_is_directory) {
3308			xfs_bumplink(tp, target_dp);
 
 
3309		}
3310	} else { /* target_ip != NULL */
3311		/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3312		 * Link the source inode under the target name.
3313		 * If the source inode is a directory and we are moving
3314		 * it across directories, its ".." entry will be
3315		 * inconsistent until we replace that down below.
3316		 *
3317		 * In case there is already an entry with the same
3318		 * name at the destination directory, remove it first.
3319		 */
3320
3321		/*
3322		 * Check whether the replace operation will need to allocate
3323		 * blocks.  This happens when the shortform directory lacks
3324		 * space and we have to convert it to a block format directory.
3325		 * When more blocks are necessary, we must lock the AGI first
3326		 * to preserve locking order (AGI -> AGF).
3327		 */
3328		if (xfs_dir2_sf_replace_needblock(target_dp, src_ip->i_ino)) {
3329			error = xfs_read_agi(mp, tp,
3330					XFS_INO_TO_AGNO(mp, target_ip->i_ino),
3331					&agibp);
3332			if (error)
3333				goto out_trans_cancel;
3334		}
3335
3336		error = xfs_dir_replace(tp, target_dp, target_name,
3337					src_ip->i_ino, spaceres);
 
3338		if (error)
3339			goto out_trans_cancel;
3340
3341		xfs_trans_ichgtime(tp, target_dp,
3342					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3343
3344		/*
3345		 * Decrement the link count on the target since the target
3346		 * dir no longer points to it.
3347		 */
3348		error = xfs_droplink(tp, target_ip);
3349		if (error)
3350			goto out_trans_cancel;
3351
3352		if (src_is_directory) {
3353			/*
3354			 * Drop the link from the old "." entry.
3355			 */
3356			error = xfs_droplink(tp, target_ip);
3357			if (error)
3358				goto out_trans_cancel;
3359		}
3360	} /* target_ip != NULL */
3361
3362	/*
3363	 * Remove the source.
3364	 */
3365	if (new_parent && src_is_directory) {
3366		/*
3367		 * Rewrite the ".." entry to point to the new
3368		 * directory.
3369		 */
3370		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3371					target_dp->i_ino, spaceres);
 
3372		ASSERT(error != -EEXIST);
3373		if (error)
3374			goto out_trans_cancel;
3375	}
3376
3377	/*
3378	 * We always want to hit the ctime on the source inode.
3379	 *
3380	 * This isn't strictly required by the standards since the source
3381	 * inode isn't really being changed, but old unix file systems did
3382	 * it and some incremental backup programs won't work without it.
3383	 */
3384	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3385	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3386
3387	/*
3388	 * Adjust the link count on src_dp.  This is necessary when
3389	 * renaming a directory, either within one parent when
3390	 * the target existed, or across two parent directories.
3391	 */
3392	if (src_is_directory && (new_parent || target_ip != NULL)) {
3393
3394		/*
3395		 * Decrement link count on src_directory since the
3396		 * entry that's moved no longer points to it.
3397		 */
3398		error = xfs_droplink(tp, src_dp);
3399		if (error)
3400			goto out_trans_cancel;
3401	}
3402
3403	/*
3404	 * For whiteouts, we only need to update the source dirent with the
3405	 * inode number of the whiteout inode rather than removing it
3406	 * altogether.
3407	 */
3408	if (wip) {
3409		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3410					spaceres);
3411	} else
3412		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3413					   spaceres);
3414	if (error)
3415		goto out_trans_cancel;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3416
3417	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3418	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3419	if (new_parent)
3420		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3421
3422	error = xfs_finish_rename(tp);
3423	if (wip)
3424		xfs_irele(wip);
3425	return error;
3426
 
 
3427out_trans_cancel:
3428	xfs_trans_cancel(tp);
3429out_release_wip:
3430	if (wip)
3431		xfs_irele(wip);
3432	return error;
3433}
3434
3435static int
3436xfs_iflush(
3437	struct xfs_inode	*ip,
3438	struct xfs_buf		*bp)
3439{
3440	struct xfs_inode_log_item *iip = ip->i_itemp;
3441	struct xfs_dinode	*dip;
3442	struct xfs_mount	*mp = ip->i_mount;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3443	int			error;
3444
 
 
3445	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3446	ASSERT(xfs_isiflocked(ip));
3447	ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3448	       ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3449	ASSERT(iip->ili_item.li_buf == bp);
3450
3451	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3452
3453	/*
3454	 * We don't flush the inode if any of the following checks fail, but we
3455	 * do still update the log item and attach to the backing buffer as if
3456	 * the flush happened. This is a formality to facilitate predictable
3457	 * error handling as the caller will shutdown and fail the buffer.
 
 
3458	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3459	error = -EFSCORRUPTED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3460	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3461			       mp, XFS_ERRTAG_IFLUSH_1)) {
3462		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3463			"%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3464			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3465		goto flush_out;
3466	}
3467	if (S_ISREG(VFS_I(ip)->i_mode)) {
3468		if (XFS_TEST_ERROR(
3469		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3470		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3471		    mp, XFS_ERRTAG_IFLUSH_3)) {
3472			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3473				"%s: Bad regular inode %Lu, ptr "PTR_FMT,
3474				__func__, ip->i_ino, ip);
3475			goto flush_out;
3476		}
3477	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3478		if (XFS_TEST_ERROR(
3479		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3480		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3481		    ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3482		    mp, XFS_ERRTAG_IFLUSH_4)) {
3483			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3484				"%s: Bad directory inode %Lu, ptr "PTR_FMT,
3485				__func__, ip->i_ino, ip);
3486			goto flush_out;
3487		}
3488	}
3489	if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
3490				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
 
3491		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3492			"%s: detected corrupt incore inode %Lu, "
3493			"total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3494			__func__, ip->i_ino,
3495			ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
3496			ip->i_d.di_nblocks, ip);
3497		goto flush_out;
3498	}
3499	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3500				mp, XFS_ERRTAG_IFLUSH_6)) {
3501		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3502			"%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3503			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3504		goto flush_out;
3505	}
3506
3507	/*
3508	 * Inode item log recovery for v2 inodes are dependent on the
3509	 * di_flushiter count for correct sequencing. We bump the flush
3510	 * iteration count so we can detect flushes which postdate a log record
3511	 * during recovery. This is redundant as we now log every change and
3512	 * hence this can't happen but we need to still do it to ensure
3513	 * backwards compatibility with old kernels that predate logging all
3514	 * inode changes.
3515	 */
3516	if (!xfs_sb_version_has_v3inode(&mp->m_sb))
3517		ip->i_d.di_flushiter++;
3518
3519	/*
3520	 * If there are inline format data / attr forks attached to this inode,
3521	 * make sure they are not corrupt.
3522	 */
3523	if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3524	    xfs_ifork_verify_local_data(ip))
3525		goto flush_out;
3526	if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
3527	    xfs_ifork_verify_local_attr(ip))
3528		goto flush_out;
3529
3530	/*
3531	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3532	 * copy out the core of the inode, because if the inode is dirty at all
3533	 * the core must be.
3534	 */
3535	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3536
3537	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3538	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3539		ip->i_d.di_flushiter = 0;
3540
3541	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3542	if (XFS_IFORK_Q(ip))
3543		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
 
3544
3545	/*
3546	 * We've recorded everything logged in the inode, so we'd like to clear
3547	 * the ili_fields bits so we don't log and flush things unnecessarily.
3548	 * However, we can't stop logging all this information until the data
3549	 * we've copied into the disk buffer is written to disk.  If we did we
3550	 * might overwrite the copy of the inode in the log with all the data
3551	 * after re-logging only part of it, and in the face of a crash we
3552	 * wouldn't have all the data we need to recover.
3553	 *
3554	 * What we do is move the bits to the ili_last_fields field.  When
3555	 * logging the inode, these bits are moved back to the ili_fields field.
3556	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3557	 * know that the information those bits represent is permanently on
3558	 * disk.  As long as the flush completes before the inode is logged
3559	 * again, then both ili_fields and ili_last_fields will be cleared.
 
 
 
 
 
 
 
 
 
3560	 */
3561	error = 0;
3562flush_out:
3563	spin_lock(&iip->ili_lock);
3564	iip->ili_last_fields = iip->ili_fields;
3565	iip->ili_fields = 0;
3566	iip->ili_fsync_fields = 0;
3567	spin_unlock(&iip->ili_lock);
3568
3569	/*
3570	 * Store the current LSN of the inode so that we can tell whether the
3571	 * item has moved in the AIL from xfs_iflush_done().
3572	 */
3573	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3574				&iip->ili_item.li_lsn);
3575
3576	/* generate the checksum. */
3577	xfs_dinode_calc_crc(mp, dip);
3578	return error;
3579}
3580
3581/*
3582 * Non-blocking flush of dirty inode metadata into the backing buffer.
3583 *
3584 * The caller must have a reference to the inode and hold the cluster buffer
3585 * locked. The function will walk across all the inodes on the cluster buffer it
3586 * can find and lock without blocking, and flush them to the cluster buffer.
3587 *
3588 * On successful flushing of at least one inode, the caller must write out the
3589 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3590 * the caller needs to release the buffer. On failure, the filesystem will be
3591 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3592 * will be returned.
3593 */
3594int
3595xfs_iflush_cluster(
3596	struct xfs_buf		*bp)
3597{
3598	struct xfs_mount	*mp = bp->b_mount;
3599	struct xfs_log_item	*lip, *n;
3600	struct xfs_inode	*ip;
3601	struct xfs_inode_log_item *iip;
3602	int			clcount = 0;
3603	int			error = 0;
3604
3605	/*
3606	 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3607	 * can remove itself from the list.
 
 
3608	 */
3609	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3610		iip = (struct xfs_inode_log_item *)lip;
3611		ip = iip->ili_inode;
3612
3613		/*
3614		 * Quick and dirty check to avoid locks if possible.
3615		 */
3616		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLOCK))
3617			continue;
3618		if (xfs_ipincount(ip))
3619			continue;
3620
3621		/*
3622		 * The inode is still attached to the buffer, which means it is
3623		 * dirty but reclaim might try to grab it. Check carefully for
3624		 * that, and grab the ilock while still holding the i_flags_lock
3625		 * to guarantee reclaim will not be able to reclaim this inode
3626		 * once we drop the i_flags_lock.
3627		 */
3628		spin_lock(&ip->i_flags_lock);
3629		ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3630		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLOCK)) {
3631			spin_unlock(&ip->i_flags_lock);
3632			continue;
3633		}
3634
3635		/*
3636		 * ILOCK will pin the inode against reclaim and prevent
3637		 * concurrent transactions modifying the inode while we are
3638		 * flushing the inode.
3639		 */
3640		if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3641			spin_unlock(&ip->i_flags_lock);
3642			continue;
3643		}
3644		spin_unlock(&ip->i_flags_lock);
3645
3646		/*
3647		 * Skip inodes that are already flush locked as they have
3648		 * already been written to the buffer.
3649		 */
3650		if (!xfs_iflock_nowait(ip)) {
3651			xfs_iunlock(ip, XFS_ILOCK_SHARED);
3652			continue;
3653		}
3654
3655		/*
3656		 * Abort flushing this inode if we are shut down because the
3657		 * inode may not currently be in the AIL. This can occur when
3658		 * log I/O failure unpins the inode without inserting into the
3659		 * AIL, leaving a dirty/unpinned inode attached to the buffer
3660		 * that otherwise looks like it should be flushed.
3661		 */
3662		if (XFS_FORCED_SHUTDOWN(mp)) {
3663			xfs_iunpin_wait(ip);
3664			/* xfs_iflush_abort() drops the flush lock */
3665			xfs_iflush_abort(ip);
3666			xfs_iunlock(ip, XFS_ILOCK_SHARED);
3667			error = -EIO;
3668			continue;
3669		}
3670
3671		/* don't block waiting on a log force to unpin dirty inodes */
3672		if (xfs_ipincount(ip)) {
3673			xfs_ifunlock(ip);
3674			xfs_iunlock(ip, XFS_ILOCK_SHARED);
3675			continue;
3676		}
3677
3678		if (!xfs_inode_clean(ip))
3679			error = xfs_iflush(ip, bp);
3680		else
3681			xfs_ifunlock(ip);
3682		xfs_iunlock(ip, XFS_ILOCK_SHARED);
3683		if (error)
3684			break;
3685		clcount++;
3686	}
3687
3688	if (error) {
3689		bp->b_flags |= XBF_ASYNC;
3690		xfs_buf_ioend_fail(bp);
3691		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3692		return error;
3693	}
3694
3695	if (!clcount)
3696		return -EAGAIN;
3697
3698	XFS_STATS_INC(mp, xs_icluster_flushcnt);
3699	XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3700	return 0;
3701
3702}
3703
3704/* Release an inode. */
3705void
3706xfs_irele(
3707	struct xfs_inode	*ip)
3708{
3709	trace_xfs_irele(ip, _RET_IP_);
3710	iput(VFS_I(ip));
3711}
3712
3713/*
3714 * Ensure all commited transactions touching the inode are written to the log.
3715 */
3716int
3717xfs_log_force_inode(
3718	struct xfs_inode	*ip)
3719{
3720	xfs_lsn_t		lsn = 0;
3721
3722	xfs_ilock(ip, XFS_ILOCK_SHARED);
3723	if (xfs_ipincount(ip))
3724		lsn = ip->i_itemp->ili_last_lsn;
3725	xfs_iunlock(ip, XFS_ILOCK_SHARED);
3726
3727	if (!lsn)
3728		return 0;
3729	return xfs_log_force_lsn(ip->i_mount, lsn, XFS_LOG_SYNC, NULL);
3730}
3731
3732/*
3733 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3734 * abide vfs locking order (lowest pointer value goes first) and breaking the
3735 * layout leases before proceeding.  The loop is needed because we cannot call
3736 * the blocking break_layout() with the iolocks held, and therefore have to
3737 * back out both locks.
3738 */
3739static int
3740xfs_iolock_two_inodes_and_break_layout(
3741	struct inode		*src,
3742	struct inode		*dest)
3743{
3744	int			error;
3745
3746	if (src > dest)
3747		swap(src, dest);
3748
3749retry:
3750	/* Wait to break both inodes' layouts before we start locking. */
3751	error = break_layout(src, true);
3752	if (error)
3753		return error;
3754	if (src != dest) {
3755		error = break_layout(dest, true);
3756		if (error)
3757			return error;
3758	}
3759
3760	/* Lock one inode and make sure nobody got in and leased it. */
3761	inode_lock(src);
3762	error = break_layout(src, false);
3763	if (error) {
3764		inode_unlock(src);
3765		if (error == -EWOULDBLOCK)
3766			goto retry;
3767		return error;
3768	}
3769
3770	if (src == dest)
3771		return 0;
3772
3773	/* Lock the other inode and make sure nobody got in and leased it. */
3774	inode_lock_nested(dest, I_MUTEX_NONDIR2);
3775	error = break_layout(dest, false);
3776	if (error) {
3777		inode_unlock(src);
3778		inode_unlock(dest);
3779		if (error == -EWOULDBLOCK)
3780			goto retry;
3781		return error;
3782	}
3783
3784	return 0;
3785}
3786
3787/*
3788 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3789 * mmap activity.
3790 */
3791int
3792xfs_ilock2_io_mmap(
3793	struct xfs_inode	*ip1,
3794	struct xfs_inode	*ip2)
3795{
3796	int			ret;
3797
3798	ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3799	if (ret)
3800		return ret;
3801	if (ip1 == ip2)
3802		xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3803	else
3804		xfs_lock_two_inodes(ip1, XFS_MMAPLOCK_EXCL,
3805				    ip2, XFS_MMAPLOCK_EXCL);
3806	return 0;
3807}
3808
3809/* Unlock both inodes to allow IO and mmap activity. */
3810void
3811xfs_iunlock2_io_mmap(
3812	struct xfs_inode	*ip1,
3813	struct xfs_inode	*ip2)
3814{
3815	bool			same_inode = (ip1 == ip2);
3816
3817	xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3818	if (!same_inode)
3819		xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3820	inode_unlock(VFS_I(ip2));
3821	if (!same_inode)
3822		inode_unlock(VFS_I(ip1));
3823}
v4.10.11
 
   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_sb.h"
  27#include "xfs_mount.h"
  28#include "xfs_defer.h"
  29#include "xfs_inode.h"
  30#include "xfs_da_format.h"
  31#include "xfs_da_btree.h"
  32#include "xfs_dir2.h"
  33#include "xfs_attr_sf.h"
  34#include "xfs_attr.h"
  35#include "xfs_trans_space.h"
  36#include "xfs_trans.h"
  37#include "xfs_buf_item.h"
  38#include "xfs_inode_item.h"
  39#include "xfs_ialloc.h"
  40#include "xfs_bmap.h"
  41#include "xfs_bmap_util.h"
 
  42#include "xfs_error.h"
  43#include "xfs_quota.h"
  44#include "xfs_filestream.h"
  45#include "xfs_cksum.h"
  46#include "xfs_trace.h"
  47#include "xfs_icache.h"
  48#include "xfs_symlink.h"
  49#include "xfs_trans_priv.h"
  50#include "xfs_log.h"
  51#include "xfs_bmap_btree.h"
  52#include "xfs_reflink.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(struct xfs_inode *, struct xfs_buf *);
  63STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
  64STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
  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 * Helper function to extract CoW extent size hint from inode.
  82 * Between the extent size hint and the CoW extent size hint, we
  83 * return the greater of the two.  If the value is zero (automatic),
  84 * use the default size.
  85 */
  86xfs_extlen_t
  87xfs_get_cowextsz_hint(
  88	struct xfs_inode	*ip)
  89{
  90	xfs_extlen_t		a, b;
  91
  92	a = 0;
  93	if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
  94		a = ip->i_d.di_cowextsize;
  95	b = xfs_get_extsz_hint(ip);
  96
  97	a = max(a, b);
  98	if (a == 0)
  99		return XFS_DEFAULT_COWEXTSZ_HINT;
 100	return a;
 101}
 102
 103/*
 104 * These two are wrapper routines around the xfs_ilock() routine used to
 105 * centralize some grungy code.  They are used in places that wish to lock the
 106 * inode solely for reading the extents.  The reason these places can't just
 107 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
 108 * bringing in of the extents from disk for a file in b-tree format.  If the
 109 * inode is in b-tree format, then we need to lock the inode exclusively until
 110 * the extents are read in.  Locking it exclusively all the time would limit
 111 * our parallelism unnecessarily, though.  What we do instead is check to see
 112 * if the extents have been read in yet, and only lock the inode exclusively
 113 * if they have not.
 114 *
 115 * The functions return a value which should be given to the corresponding
 116 * xfs_iunlock() call.
 117 */
 118uint
 119xfs_ilock_data_map_shared(
 120	struct xfs_inode	*ip)
 121{
 122	uint			lock_mode = XFS_ILOCK_SHARED;
 123
 124	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
 125	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
 126		lock_mode = XFS_ILOCK_EXCL;
 127	xfs_ilock(ip, lock_mode);
 128	return lock_mode;
 129}
 130
 131uint
 132xfs_ilock_attr_map_shared(
 133	struct xfs_inode	*ip)
 134{
 135	uint			lock_mode = XFS_ILOCK_SHARED;
 136
 137	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
 
 138	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
 139		lock_mode = XFS_ILOCK_EXCL;
 140	xfs_ilock(ip, lock_mode);
 141	return lock_mode;
 142}
 143
 144/*
 145 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
 146 * multi-reader locks: i_mmap_lock and the i_lock.  This routine allows
 147 * various combinations of the locks to be obtained.
 148 *
 149 * The 3 locks should always be ordered so that the IO lock is obtained first,
 150 * the mmap lock second and the ilock last in order to prevent deadlock.
 151 *
 152 * Basic locking order:
 153 *
 154 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
 155 *
 156 * mmap_sem locking order:
 157 *
 158 * i_rwsem -> page lock -> mmap_sem
 159 * mmap_sem -> i_mmap_lock -> page_lock
 160 *
 161 * The difference in mmap_sem locking order mean that we cannot hold the
 162 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
 163 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
 164 * in get_user_pages() to map the user pages into the kernel address space for
 165 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
 166 * page faults already hold the mmap_sem.
 167 *
 168 * Hence to serialise fully against both syscall and mmap based IO, we need to
 169 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
 170 * taken in places where we need to invalidate the page cache in a race
 171 * free manner (e.g. truncate, hole punch and other extent manipulation
 172 * functions).
 173 */
 174void
 175xfs_ilock(
 176	xfs_inode_t		*ip,
 177	uint			lock_flags)
 178{
 179	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
 180
 181	/*
 182	 * You can't set both SHARED and EXCL for the same lock,
 183	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 184	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 185	 */
 186	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 187	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 188	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 189	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 190	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 191	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 192	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 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		mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
 204	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 205		mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
 206
 207	if (lock_flags & XFS_ILOCK_EXCL)
 208		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 209	else if (lock_flags & XFS_ILOCK_SHARED)
 210		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 211}
 212
 213/*
 214 * This is just like xfs_ilock(), except that the caller
 215 * is guaranteed not to sleep.  It returns 1 if it gets
 216 * the requested locks and 0 otherwise.  If the IO lock is
 217 * obtained but the inode lock cannot be, then the IO lock
 218 * is dropped before returning.
 219 *
 220 * ip -- the inode being locked
 221 * lock_flags -- this parameter indicates the inode's locks to be
 222 *       to be locked.  See the comment for xfs_ilock() for a list
 223 *	 of valid values.
 224 */
 225int
 226xfs_ilock_nowait(
 227	xfs_inode_t		*ip,
 228	uint			lock_flags)
 229{
 230	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
 231
 232	/*
 233	 * You can't set both SHARED and EXCL for the same lock,
 234	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 235	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 236	 */
 237	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 238	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 239	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 240	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 241	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 242	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 243	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 244
 245	if (lock_flags & XFS_IOLOCK_EXCL) {
 246		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
 247			goto out;
 248	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 249		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
 250			goto out;
 251	}
 252
 253	if (lock_flags & XFS_MMAPLOCK_EXCL) {
 254		if (!mrtryupdate(&ip->i_mmaplock))
 255			goto out_undo_iolock;
 256	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
 257		if (!mrtryaccess(&ip->i_mmaplock))
 258			goto out_undo_iolock;
 259	}
 260
 261	if (lock_flags & XFS_ILOCK_EXCL) {
 262		if (!mrtryupdate(&ip->i_lock))
 263			goto out_undo_mmaplock;
 264	} else if (lock_flags & XFS_ILOCK_SHARED) {
 265		if (!mrtryaccess(&ip->i_lock))
 266			goto out_undo_mmaplock;
 267	}
 268	return 1;
 269
 270out_undo_mmaplock:
 271	if (lock_flags & XFS_MMAPLOCK_EXCL)
 272		mrunlock_excl(&ip->i_mmaplock);
 273	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 274		mrunlock_shared(&ip->i_mmaplock);
 275out_undo_iolock:
 276	if (lock_flags & XFS_IOLOCK_EXCL)
 277		up_write(&VFS_I(ip)->i_rwsem);
 278	else if (lock_flags & XFS_IOLOCK_SHARED)
 279		up_read(&VFS_I(ip)->i_rwsem);
 280out:
 281	return 0;
 282}
 283
 284/*
 285 * xfs_iunlock() is used to drop the inode locks acquired with
 286 * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
 287 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
 288 * that we know which locks to drop.
 289 *
 290 * ip -- the inode being unlocked
 291 * lock_flags -- this parameter indicates the inode's locks to be
 292 *       to be unlocked.  See the comment for xfs_ilock() for a list
 293 *	 of valid values for this parameter.
 294 *
 295 */
 296void
 297xfs_iunlock(
 298	xfs_inode_t		*ip,
 299	uint			lock_flags)
 300{
 301	/*
 302	 * You can't set both SHARED and EXCL for the same lock,
 303	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
 304	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
 305	 */
 306	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 307	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 308	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 309	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 310	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 311	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 312	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 313	ASSERT(lock_flags != 0);
 314
 315	if (lock_flags & XFS_IOLOCK_EXCL)
 316		up_write(&VFS_I(ip)->i_rwsem);
 317	else if (lock_flags & XFS_IOLOCK_SHARED)
 318		up_read(&VFS_I(ip)->i_rwsem);
 319
 320	if (lock_flags & XFS_MMAPLOCK_EXCL)
 321		mrunlock_excl(&ip->i_mmaplock);
 322	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 323		mrunlock_shared(&ip->i_mmaplock);
 324
 325	if (lock_flags & XFS_ILOCK_EXCL)
 326		mrunlock_excl(&ip->i_lock);
 327	else if (lock_flags & XFS_ILOCK_SHARED)
 328		mrunlock_shared(&ip->i_lock);
 329
 330	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
 331}
 332
 333/*
 334 * give up write locks.  the i/o lock cannot be held nested
 335 * if it is being demoted.
 336 */
 337void
 338xfs_ilock_demote(
 339	xfs_inode_t		*ip,
 340	uint			lock_flags)
 341{
 342	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
 343	ASSERT((lock_flags &
 344		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
 345
 346	if (lock_flags & XFS_ILOCK_EXCL)
 347		mrdemote(&ip->i_lock);
 348	if (lock_flags & XFS_MMAPLOCK_EXCL)
 349		mrdemote(&ip->i_mmaplock);
 350	if (lock_flags & XFS_IOLOCK_EXCL)
 351		downgrade_write(&VFS_I(ip)->i_rwsem);
 352
 353	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
 354}
 355
 356#if defined(DEBUG) || defined(XFS_WARN)
 357int
 358xfs_isilocked(
 359	xfs_inode_t		*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		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
 370			return !!ip->i_mmaplock.mr_writer;
 371		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
 372	}
 373
 374	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
 375		if (!(lock_flags & XFS_IOLOCK_SHARED))
 376			return !debug_locks ||
 377				lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
 378		return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
 379	}
 380
 381	ASSERT(0);
 382	return 0;
 383}
 384#endif
 385
 386#ifdef DEBUG
 387int xfs_locked_n;
 388int xfs_small_retries;
 389int xfs_middle_retries;
 390int xfs_lots_retries;
 391int xfs_lock_delays;
 392#endif
 393
 394/*
 395 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
 396 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
 397 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
 398 * errors and warnings.
 399 */
 400#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
 401static bool
 402xfs_lockdep_subclass_ok(
 403	int subclass)
 404{
 405	return subclass < MAX_LOCKDEP_SUBCLASSES;
 406}
 407#else
 408#define xfs_lockdep_subclass_ok(subclass)	(true)
 409#endif
 410
 411/*
 412 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
 413 * value. This can be called for any type of inode lock combination, including
 414 * parent locking. Care must be taken to ensure we don't overrun the subclass
 415 * storage fields in the class mask we build.
 416 */
 417static inline int
 418xfs_lock_inumorder(int lock_mode, int subclass)
 419{
 420	int	class = 0;
 421
 422	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
 423			      XFS_ILOCK_RTSUM)));
 424	ASSERT(xfs_lockdep_subclass_ok(subclass));
 425
 426	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
 427		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
 428		class += subclass << XFS_IOLOCK_SHIFT;
 429	}
 430
 431	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
 432		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
 433		class += subclass << XFS_MMAPLOCK_SHIFT;
 434	}
 435
 436	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
 437		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
 438		class += subclass << XFS_ILOCK_SHIFT;
 439	}
 440
 441	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
 442}
 443
 444/*
 445 * The following routine will lock n inodes in exclusive mode.  We assume the
 446 * caller calls us with the inodes in i_ino order.
 447 *
 448 * We need to detect deadlock where an inode that we lock is in the AIL and we
 449 * start waiting for another inode that is locked by a thread in a long running
 450 * transaction (such as truncate). This can result in deadlock since the long
 451 * running trans might need to wait for the inode we just locked in order to
 452 * push the tail and free space in the log.
 453 *
 454 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
 455 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
 456 * lock more than one at a time, lockdep will report false positives saying we
 457 * have violated locking orders.
 458 */
 459static void
 460xfs_lock_inodes(
 461	xfs_inode_t	**ips,
 462	int		inodes,
 463	uint		lock_mode)
 464{
 465	int		attempts = 0, i, j, try_lock;
 466	xfs_log_item_t	*lp;
 467
 468	/*
 469	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
 470	 * support an arbitrary depth of locking here, but absolute limits on
 471	 * inodes depend on the the type of locking and the limits placed by
 472	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
 473	 * the asserts.
 474	 */
 475	ASSERT(ips && inodes >= 2 && inodes <= 5);
 476	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
 477			    XFS_ILOCK_EXCL));
 478	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
 479			      XFS_ILOCK_SHARED)));
 480	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
 481		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
 482	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
 483		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
 484
 485	if (lock_mode & XFS_IOLOCK_EXCL) {
 486		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
 487	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
 488		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
 489
 490	try_lock = 0;
 491	i = 0;
 492again:
 493	for (; i < inodes; i++) {
 494		ASSERT(ips[i]);
 495
 496		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
 497			continue;
 498
 499		/*
 500		 * If try_lock is not set yet, make sure all locked inodes are
 501		 * not in the AIL.  If any are, set try_lock to be used later.
 502		 */
 503		if (!try_lock) {
 504			for (j = (i - 1); j >= 0 && !try_lock; j--) {
 505				lp = (xfs_log_item_t *)ips[j]->i_itemp;
 506				if (lp && (lp->li_flags & XFS_LI_IN_AIL))
 507					try_lock++;
 508			}
 509		}
 510
 511		/*
 512		 * If any of the previous locks we have locked is in the AIL,
 513		 * we must TRY to get the second and subsequent locks. If
 514		 * we can't get any, we must release all we have
 515		 * and try again.
 516		 */
 517		if (!try_lock) {
 518			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
 519			continue;
 520		}
 521
 522		/* try_lock means we have an inode locked that is in the AIL. */
 523		ASSERT(i != 0);
 524		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
 525			continue;
 526
 527		/*
 528		 * Unlock all previous guys and try again.  xfs_iunlock will try
 529		 * to push the tail if the inode is in the AIL.
 530		 */
 531		attempts++;
 532		for (j = i - 1; j >= 0; j--) {
 533			/*
 534			 * Check to see if we've already unlocked this one.  Not
 535			 * the first one going back, and the inode ptr is the
 536			 * same.
 537			 */
 538			if (j != (i - 1) && ips[j] == ips[j + 1])
 539				continue;
 540
 541			xfs_iunlock(ips[j], lock_mode);
 542		}
 543
 544		if ((attempts % 5) == 0) {
 545			delay(1); /* Don't just spin the CPU */
 546#ifdef DEBUG
 547			xfs_lock_delays++;
 548#endif
 549		}
 550		i = 0;
 551		try_lock = 0;
 552		goto again;
 553	}
 554
 555#ifdef DEBUG
 556	if (attempts) {
 557		if (attempts < 5) xfs_small_retries++;
 558		else if (attempts < 100) xfs_middle_retries++;
 559		else xfs_lots_retries++;
 560	} else {
 561		xfs_locked_n++;
 562	}
 563#endif
 564}
 565
 566/*
 567 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
 568 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
 569 * lock more than one at a time, lockdep will report false positives saying we
 570 * have violated locking orders.
 
 
 571 */
 572void
 573xfs_lock_two_inodes(
 574	xfs_inode_t		*ip0,
 575	xfs_inode_t		*ip1,
 576	uint			lock_mode)
 
 577{
 578	xfs_inode_t		*temp;
 
 579	int			attempts = 0;
 580	xfs_log_item_t		*lp;
 581
 582	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
 583	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
 584		ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
 
 
 
 
 
 
 
 
 
 585
 586	ASSERT(ip0->i_ino != ip1->i_ino);
 587
 588	if (ip0->i_ino > ip1->i_ino) {
 589		temp = ip0;
 590		ip0 = ip1;
 591		ip1 = temp;
 
 
 
 592	}
 593
 594 again:
 595	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
 596
 597	/*
 598	 * If the first lock we have locked is in the AIL, we must TRY to get
 599	 * the second lock. If we can't get it, we must release the first one
 600	 * and try again.
 601	 */
 602	lp = (xfs_log_item_t *)ip0->i_itemp;
 603	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
 604		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
 605			xfs_iunlock(ip0, lock_mode);
 606			if ((++attempts % 5) == 0)
 607				delay(1); /* Don't just spin the CPU */
 608			goto again;
 609		}
 610	} else {
 611		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
 612	}
 613}
 614
 615
 616void
 617__xfs_iflock(
 618	struct xfs_inode	*ip)
 619{
 620	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
 621	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
 622
 623	do {
 624		prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
 625		if (xfs_isiflocked(ip))
 626			io_schedule();
 627	} while (!xfs_iflock_nowait(ip));
 628
 629	finish_wait(wq, &wait.wait);
 630}
 631
 632STATIC uint
 633_xfs_dic2xflags(
 634	__uint16_t		di_flags,
 635	uint64_t		di_flags2,
 636	bool			has_attr)
 637{
 638	uint			flags = 0;
 639
 640	if (di_flags & XFS_DIFLAG_ANY) {
 641		if (di_flags & XFS_DIFLAG_REALTIME)
 642			flags |= FS_XFLAG_REALTIME;
 643		if (di_flags & XFS_DIFLAG_PREALLOC)
 644			flags |= FS_XFLAG_PREALLOC;
 645		if (di_flags & XFS_DIFLAG_IMMUTABLE)
 646			flags |= FS_XFLAG_IMMUTABLE;
 647		if (di_flags & XFS_DIFLAG_APPEND)
 648			flags |= FS_XFLAG_APPEND;
 649		if (di_flags & XFS_DIFLAG_SYNC)
 650			flags |= FS_XFLAG_SYNC;
 651		if (di_flags & XFS_DIFLAG_NOATIME)
 652			flags |= FS_XFLAG_NOATIME;
 653		if (di_flags & XFS_DIFLAG_NODUMP)
 654			flags |= FS_XFLAG_NODUMP;
 655		if (di_flags & XFS_DIFLAG_RTINHERIT)
 656			flags |= FS_XFLAG_RTINHERIT;
 657		if (di_flags & XFS_DIFLAG_PROJINHERIT)
 658			flags |= FS_XFLAG_PROJINHERIT;
 659		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
 660			flags |= FS_XFLAG_NOSYMLINKS;
 661		if (di_flags & XFS_DIFLAG_EXTSIZE)
 662			flags |= FS_XFLAG_EXTSIZE;
 663		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
 664			flags |= FS_XFLAG_EXTSZINHERIT;
 665		if (di_flags & XFS_DIFLAG_NODEFRAG)
 666			flags |= FS_XFLAG_NODEFRAG;
 667		if (di_flags & XFS_DIFLAG_FILESTREAM)
 668			flags |= FS_XFLAG_FILESTREAM;
 669	}
 670
 671	if (di_flags2 & XFS_DIFLAG2_ANY) {
 672		if (di_flags2 & XFS_DIFLAG2_DAX)
 673			flags |= FS_XFLAG_DAX;
 674		if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
 675			flags |= FS_XFLAG_COWEXTSIZE;
 676	}
 677
 678	if (has_attr)
 679		flags |= FS_XFLAG_HASATTR;
 680
 681	return flags;
 682}
 683
 684uint
 685xfs_ip2xflags(
 686	struct xfs_inode	*ip)
 687{
 688	struct xfs_icdinode	*dic = &ip->i_d;
 689
 690	return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
 691}
 692
 693/*
 694 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
 695 * is allowed, otherwise it has to be an exact match. If a CI match is found,
 696 * ci_name->name will point to a the actual name (caller must free) or
 697 * will be set to NULL if an exact match is found.
 698 */
 699int
 700xfs_lookup(
 701	xfs_inode_t		*dp,
 702	struct xfs_name		*name,
 703	xfs_inode_t		**ipp,
 704	struct xfs_name		*ci_name)
 705{
 706	xfs_ino_t		inum;
 707	int			error;
 708
 709	trace_xfs_lookup(dp, name);
 710
 711	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 712		return -EIO;
 713
 714	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
 715	if (error)
 716		goto out_unlock;
 717
 718	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
 719	if (error)
 720		goto out_free_name;
 721
 722	return 0;
 723
 724out_free_name:
 725	if (ci_name)
 726		kmem_free(ci_name->name);
 727out_unlock:
 728	*ipp = NULL;
 729	return error;
 730}
 731
 732/*
 733 * Allocate an inode on disk and return a copy of its in-core version.
 734 * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
 735 * appropriately within the inode.  The uid and gid for the inode are
 736 * set according to the contents of the given cred structure.
 737 *
 738 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
 739 * has a free inode available, call xfs_iget() to obtain the in-core
 740 * version of the allocated inode.  Finally, fill in the inode and
 741 * log its initial contents.  In this case, ialloc_context would be
 742 * set to NULL.
 743 *
 744 * If xfs_dialloc() does not have an available inode, it will replenish
 745 * its supply by doing an allocation. Since we can only do one
 746 * allocation within a transaction without deadlocks, we must commit
 747 * the current transaction before returning the inode itself.
 748 * In this case, therefore, we will set ialloc_context and return.
 749 * The caller should then commit the current transaction, start a new
 750 * transaction, and call xfs_ialloc() again to actually get the inode.
 751 *
 752 * To ensure that some other process does not grab the inode that
 753 * was allocated during the first call to xfs_ialloc(), this routine
 754 * also returns the [locked] bp pointing to the head of the freelist
 755 * as ialloc_context.  The caller should hold this buffer across
 756 * the commit and pass it back into this routine on the second call.
 757 *
 758 * If we are allocating quota inodes, we do not have a parent inode
 759 * to attach to or associate with (i.e. pip == NULL) because they
 760 * are not linked into the directory structure - they are attached
 761 * directly to the superblock - and so have no parent.
 762 */
 763static int
 764xfs_ialloc(
 765	xfs_trans_t	*tp,
 766	xfs_inode_t	*pip,
 767	umode_t		mode,
 768	xfs_nlink_t	nlink,
 769	xfs_dev_t	rdev,
 770	prid_t		prid,
 771	int		okalloc,
 772	xfs_buf_t	**ialloc_context,
 773	xfs_inode_t	**ipp)
 774{
 775	struct xfs_mount *mp = tp->t_mountp;
 776	xfs_ino_t	ino;
 777	xfs_inode_t	*ip;
 778	uint		flags;
 779	int		error;
 780	struct timespec	tv;
 781	struct inode	*inode;
 782
 783	/*
 784	 * Call the space management code to pick
 785	 * the on-disk inode to be allocated.
 786	 */
 787	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
 788			    ialloc_context, &ino);
 789	if (error)
 790		return error;
 791	if (*ialloc_context || ino == NULLFSINO) {
 792		*ipp = NULL;
 793		return 0;
 794	}
 795	ASSERT(*ialloc_context == NULL);
 796
 797	/*
 
 
 
 
 
 
 
 
 
 
 
 
 798	 * Get the in-core inode with the lock held exclusively.
 799	 * This is because we're setting fields here we need
 800	 * to prevent others from looking at until we're done.
 801	 */
 802	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
 803			 XFS_ILOCK_EXCL, &ip);
 804	if (error)
 805		return error;
 806	ASSERT(ip != NULL);
 807	inode = VFS_I(ip);
 808
 809	/*
 810	 * We always convert v1 inodes to v2 now - we only support filesystems
 811	 * with >= v2 inode capability, so there is no reason for ever leaving
 812	 * an inode in v1 format.
 813	 */
 814	if (ip->i_d.di_version == 1)
 815		ip->i_d.di_version = 2;
 816
 817	inode->i_mode = mode;
 818	set_nlink(inode, nlink);
 819	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
 820	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
 821	xfs_set_projid(ip, prid);
 822
 823	if (pip && XFS_INHERIT_GID(pip)) {
 824		ip->i_d.di_gid = pip->i_d.di_gid;
 825		if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
 826			inode->i_mode |= S_ISGID;
 
 
 827	}
 828
 829	/*
 830	 * If the group ID of the new file does not match the effective group
 831	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
 832	 * (and only if the irix_sgid_inherit compatibility variable is set).
 833	 */
 834	if ((irix_sgid_inherit) &&
 835	    (inode->i_mode & S_ISGID) &&
 836	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
 837		inode->i_mode &= ~S_ISGID;
 838
 839	ip->i_d.di_size = 0;
 840	ip->i_d.di_nextents = 0;
 841	ASSERT(ip->i_d.di_nblocks == 0);
 842
 843	tv = current_time(inode);
 844	inode->i_mtime = tv;
 845	inode->i_atime = tv;
 846	inode->i_ctime = tv;
 847
 848	ip->i_d.di_extsize = 0;
 849	ip->i_d.di_dmevmask = 0;
 850	ip->i_d.di_dmstate = 0;
 851	ip->i_d.di_flags = 0;
 852
 853	if (ip->i_d.di_version == 3) {
 854		inode->i_version = 1;
 855		ip->i_d.di_flags2 = 0;
 856		ip->i_d.di_cowextsize = 0;
 857		ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
 858		ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
 859	}
 860
 861
 862	flags = XFS_ILOG_CORE;
 863	switch (mode & S_IFMT) {
 864	case S_IFIFO:
 865	case S_IFCHR:
 866	case S_IFBLK:
 867	case S_IFSOCK:
 868		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
 869		ip->i_df.if_u2.if_rdev = rdev;
 870		ip->i_df.if_flags = 0;
 871		flags |= XFS_ILOG_DEV;
 872		break;
 873	case S_IFREG:
 874	case S_IFDIR:
 875		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
 876			uint64_t	di_flags2 = 0;
 877			uint		di_flags = 0;
 878
 879			if (S_ISDIR(mode)) {
 880				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
 881					di_flags |= XFS_DIFLAG_RTINHERIT;
 882				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
 883					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
 884					ip->i_d.di_extsize = pip->i_d.di_extsize;
 885				}
 886				if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
 887					di_flags |= XFS_DIFLAG_PROJINHERIT;
 888			} else if (S_ISREG(mode)) {
 889				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
 890					di_flags |= XFS_DIFLAG_REALTIME;
 891				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
 892					di_flags |= XFS_DIFLAG_EXTSIZE;
 893					ip->i_d.di_extsize = pip->i_d.di_extsize;
 894				}
 895			}
 896			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
 897			    xfs_inherit_noatime)
 898				di_flags |= XFS_DIFLAG_NOATIME;
 899			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
 900			    xfs_inherit_nodump)
 901				di_flags |= XFS_DIFLAG_NODUMP;
 902			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
 903			    xfs_inherit_sync)
 904				di_flags |= XFS_DIFLAG_SYNC;
 905			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
 906			    xfs_inherit_nosymlinks)
 907				di_flags |= XFS_DIFLAG_NOSYMLINKS;
 908			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
 909			    xfs_inherit_nodefrag)
 910				di_flags |= XFS_DIFLAG_NODEFRAG;
 911			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
 912				di_flags |= XFS_DIFLAG_FILESTREAM;
 913			if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
 914				di_flags2 |= XFS_DIFLAG2_DAX;
 915
 916			ip->i_d.di_flags |= di_flags;
 917			ip->i_d.di_flags2 |= di_flags2;
 918		}
 919		if (pip &&
 920		    (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
 921		    pip->i_d.di_version == 3 &&
 922		    ip->i_d.di_version == 3) {
 923			if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
 924				ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
 925				ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
 926			}
 
 
 927		}
 928		/* FALLTHROUGH */
 929	case S_IFLNK:
 930		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
 931		ip->i_df.if_flags = XFS_IFEXTENTS;
 932		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
 933		ip->i_df.if_u1.if_extents = NULL;
 934		break;
 935	default:
 936		ASSERT(0);
 937	}
 938	/*
 939	 * Attribute fork settings for new inode.
 940	 */
 941	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
 942	ip->i_d.di_anextents = 0;
 943
 944	/*
 945	 * Log the new values stuffed into the inode.
 946	 */
 947	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 948	xfs_trans_log_inode(tp, ip, flags);
 949
 950	/* now that we have an i_mode we can setup the inode structure */
 951	xfs_setup_inode(ip);
 952
 953	*ipp = ip;
 954	return 0;
 955}
 956
 957/*
 958 * Allocates a new inode from disk and return a pointer to the
 959 * incore copy. This routine will internally commit the current
 960 * transaction and allocate a new one if the Space Manager needed
 961 * to do an allocation to replenish the inode free-list.
 962 *
 963 * This routine is designed to be called from xfs_create and
 964 * xfs_create_dir.
 965 *
 966 */
 967int
 968xfs_dir_ialloc(
 969	xfs_trans_t	**tpp,		/* input: current transaction;
 970					   output: may be a new transaction. */
 971	xfs_inode_t	*dp,		/* directory within whose allocate
 972					   the inode. */
 973	umode_t		mode,
 974	xfs_nlink_t	nlink,
 975	xfs_dev_t	rdev,
 976	prid_t		prid,		/* project id */
 977	int		okalloc,	/* ok to allocate new space */
 978	xfs_inode_t	**ipp,		/* pointer to inode; it will be
 979					   locked. */
 980	int		*committed)
 981
 982{
 983	xfs_trans_t	*tp;
 984	xfs_inode_t	*ip;
 985	xfs_buf_t	*ialloc_context = NULL;
 986	int		code;
 987	void		*dqinfo;
 988	uint		tflags;
 989
 990	tp = *tpp;
 991	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
 992
 993	/*
 994	 * xfs_ialloc will return a pointer to an incore inode if
 995	 * the Space Manager has an available inode on the free
 996	 * list. Otherwise, it will do an allocation and replenish
 997	 * the freelist.  Since we can only do one allocation per
 998	 * transaction without deadlocks, we will need to commit the
 999	 * current transaction and start a new one.  We will then
1000	 * need to call xfs_ialloc again to get the inode.
1001	 *
1002	 * If xfs_ialloc did an allocation to replenish the freelist,
1003	 * it returns the bp containing the head of the freelist as
1004	 * ialloc_context. We will hold a lock on it across the
1005	 * transaction commit so that no other process can steal
1006	 * the inode(s) that we've just allocated.
1007	 */
1008	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1009			  &ialloc_context, &ip);
1010
1011	/*
1012	 * Return an error if we were unable to allocate a new inode.
1013	 * This should only happen if we run out of space on disk or
1014	 * encounter a disk error.
1015	 */
1016	if (code) {
1017		*ipp = NULL;
1018		return code;
1019	}
1020	if (!ialloc_context && !ip) {
1021		*ipp = NULL;
1022		return -ENOSPC;
1023	}
1024
1025	/*
1026	 * If the AGI buffer is non-NULL, then we were unable to get an
1027	 * inode in one operation.  We need to commit the current
1028	 * transaction and call xfs_ialloc() again.  It is guaranteed
1029	 * to succeed the second time.
1030	 */
1031	if (ialloc_context) {
1032		/*
1033		 * Normally, xfs_trans_commit releases all the locks.
1034		 * We call bhold to hang on to the ialloc_context across
1035		 * the commit.  Holding this buffer prevents any other
1036		 * processes from doing any allocations in this
1037		 * allocation group.
1038		 */
1039		xfs_trans_bhold(tp, ialloc_context);
1040
1041		/*
1042		 * We want the quota changes to be associated with the next
1043		 * transaction, NOT this one. So, detach the dqinfo from this
1044		 * and attach it to the next transaction.
1045		 */
1046		dqinfo = NULL;
1047		tflags = 0;
1048		if (tp->t_dqinfo) {
1049			dqinfo = (void *)tp->t_dqinfo;
1050			tp->t_dqinfo = NULL;
1051			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1052			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1053		}
1054
1055		code = xfs_trans_roll(&tp, NULL);
1056		if (committed != NULL)
1057			*committed = 1;
1058
1059		/*
1060		 * Re-attach the quota info that we detached from prev trx.
1061		 */
1062		if (dqinfo) {
1063			tp->t_dqinfo = dqinfo;
1064			tp->t_flags |= tflags;
1065		}
1066
1067		if (code) {
1068			xfs_buf_relse(ialloc_context);
1069			*tpp = tp;
1070			*ipp = NULL;
1071			return code;
1072		}
1073		xfs_trans_bjoin(tp, ialloc_context);
1074
1075		/*
1076		 * Call ialloc again. Since we've locked out all
1077		 * other allocations in this allocation group,
1078		 * this call should always succeed.
1079		 */
1080		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1081				  okalloc, &ialloc_context, &ip);
1082
1083		/*
1084		 * If we get an error at this point, return to the caller
1085		 * so that the current transaction can be aborted.
1086		 */
1087		if (code) {
1088			*tpp = tp;
1089			*ipp = NULL;
1090			return code;
1091		}
1092		ASSERT(!ialloc_context && ip);
1093
1094	} else {
1095		if (committed != NULL)
1096			*committed = 0;
1097	}
1098
1099	*ipp = ip;
1100	*tpp = tp;
1101
1102	return 0;
1103}
1104
1105/*
1106 * Decrement the link count on an inode & log the change.  If this causes the
1107 * link count to go to zero, move the inode to AGI unlinked list so that it can
1108 * be freed when the last active reference goes away via xfs_inactive().
1109 */
1110static int			/* error */
1111xfs_droplink(
1112	xfs_trans_t *tp,
1113	xfs_inode_t *ip)
1114{
1115	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1116
1117	drop_nlink(VFS_I(ip));
1118	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1119
1120	if (VFS_I(ip)->i_nlink)
1121		return 0;
1122
1123	return xfs_iunlink(tp, ip);
1124}
1125
1126/*
1127 * Increment the link count on an inode & log the change.
1128 */
1129static int
1130xfs_bumplink(
1131	xfs_trans_t *tp,
1132	xfs_inode_t *ip)
1133{
1134	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1135
1136	ASSERT(ip->i_d.di_version > 1);
1137	inc_nlink(VFS_I(ip));
1138	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1139	return 0;
1140}
1141
1142int
1143xfs_create(
1144	xfs_inode_t		*dp,
1145	struct xfs_name		*name,
1146	umode_t			mode,
1147	xfs_dev_t		rdev,
1148	xfs_inode_t		**ipp)
1149{
1150	int			is_dir = S_ISDIR(mode);
1151	struct xfs_mount	*mp = dp->i_mount;
1152	struct xfs_inode	*ip = NULL;
1153	struct xfs_trans	*tp = NULL;
1154	int			error;
1155	struct xfs_defer_ops	dfops;
1156	xfs_fsblock_t		first_block;
1157	bool                    unlock_dp_on_error = false;
1158	prid_t			prid;
1159	struct xfs_dquot	*udqp = NULL;
1160	struct xfs_dquot	*gdqp = NULL;
1161	struct xfs_dquot	*pdqp = NULL;
1162	struct xfs_trans_res	*tres;
1163	uint			resblks;
1164
1165	trace_xfs_create(dp, name);
1166
1167	if (XFS_FORCED_SHUTDOWN(mp))
1168		return -EIO;
1169
1170	prid = xfs_get_initial_prid(dp);
1171
1172	/*
1173	 * Make sure that we have allocated dquot(s) on disk.
1174	 */
1175	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1176					xfs_kgid_to_gid(current_fsgid()), prid,
1177					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1178					&udqp, &gdqp, &pdqp);
1179	if (error)
1180		return error;
1181
1182	if (is_dir) {
1183		rdev = 0;
1184		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1185		tres = &M_RES(mp)->tr_mkdir;
1186	} else {
1187		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1188		tres = &M_RES(mp)->tr_create;
1189	}
1190
1191	/*
1192	 * Initially assume that the file does not exist and
1193	 * reserve the resources for that case.  If that is not
1194	 * the case we'll drop the one we have and get a more
1195	 * appropriate transaction later.
1196	 */
1197	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1198	if (error == -ENOSPC) {
1199		/* flush outstanding delalloc blocks and retry */
1200		xfs_flush_inodes(mp);
1201		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1202	}
1203	if (error == -ENOSPC) {
1204		/* No space at all so try a "no-allocation" reservation */
1205		resblks = 0;
1206		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1207	}
1208	if (error)
1209		goto out_release_inode;
1210
1211	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1212	unlock_dp_on_error = true;
1213
1214	xfs_defer_init(&dfops, &first_block);
1215
1216	/*
1217	 * Reserve disk quota and the inode.
1218	 */
1219	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1220						pdqp, resblks, 1, 0);
1221	if (error)
1222		goto out_trans_cancel;
1223
1224	if (!resblks) {
1225		error = xfs_dir_canenter(tp, dp, name);
1226		if (error)
1227			goto out_trans_cancel;
1228	}
1229
1230	/*
1231	 * A newly created regular or special file just has one directory
1232	 * entry pointing to them, but a directory also the "." entry
1233	 * pointing to itself.
1234	 */
1235	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1236			       prid, resblks > 0, &ip, NULL);
1237	if (error)
1238		goto out_trans_cancel;
1239
1240	/*
1241	 * Now we join the directory inode to the transaction.  We do not do it
1242	 * earlier because xfs_dir_ialloc might commit the previous transaction
1243	 * (and release all the locks).  An error from here on will result in
1244	 * the transaction cancel unlocking dp so don't do it explicitly in the
1245	 * error path.
1246	 */
1247	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1248	unlock_dp_on_error = false;
1249
1250	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1251					&first_block, &dfops, resblks ?
1252					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1253	if (error) {
1254		ASSERT(error != -ENOSPC);
1255		goto out_trans_cancel;
1256	}
1257	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1258	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1259
1260	if (is_dir) {
1261		error = xfs_dir_init(tp, ip, dp);
1262		if (error)
1263			goto out_bmap_cancel;
1264
1265		error = xfs_bumplink(tp, dp);
1266		if (error)
1267			goto out_bmap_cancel;
1268	}
1269
1270	/*
1271	 * If this is a synchronous mount, make sure that the
1272	 * create transaction goes to disk before returning to
1273	 * the user.
1274	 */
1275	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1276		xfs_trans_set_sync(tp);
1277
1278	/*
1279	 * Attach the dquot(s) to the inodes and modify them incore.
1280	 * These ids of the inode couldn't have changed since the new
1281	 * inode has been locked ever since it was created.
1282	 */
1283	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1284
1285	error = xfs_defer_finish(&tp, &dfops, NULL);
1286	if (error)
1287		goto out_bmap_cancel;
1288
1289	error = xfs_trans_commit(tp);
1290	if (error)
1291		goto out_release_inode;
1292
1293	xfs_qm_dqrele(udqp);
1294	xfs_qm_dqrele(gdqp);
1295	xfs_qm_dqrele(pdqp);
1296
1297	*ipp = ip;
1298	return 0;
1299
1300 out_bmap_cancel:
1301	xfs_defer_cancel(&dfops);
1302 out_trans_cancel:
1303	xfs_trans_cancel(tp);
1304 out_release_inode:
1305	/*
1306	 * Wait until after the current transaction is aborted to finish the
1307	 * setup of the inode and release the inode.  This prevents recursive
1308	 * transactions and deadlocks from xfs_inactive.
1309	 */
1310	if (ip) {
1311		xfs_finish_inode_setup(ip);
1312		IRELE(ip);
1313	}
1314
1315	xfs_qm_dqrele(udqp);
1316	xfs_qm_dqrele(gdqp);
1317	xfs_qm_dqrele(pdqp);
1318
1319	if (unlock_dp_on_error)
1320		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1321	return error;
1322}
1323
1324int
1325xfs_create_tmpfile(
1326	struct xfs_inode	*dp,
1327	struct dentry		*dentry,
1328	umode_t			mode,
1329	struct xfs_inode	**ipp)
1330{
1331	struct xfs_mount	*mp = dp->i_mount;
1332	struct xfs_inode	*ip = NULL;
1333	struct xfs_trans	*tp = NULL;
1334	int			error;
1335	prid_t                  prid;
1336	struct xfs_dquot	*udqp = NULL;
1337	struct xfs_dquot	*gdqp = NULL;
1338	struct xfs_dquot	*pdqp = NULL;
1339	struct xfs_trans_res	*tres;
1340	uint			resblks;
1341
1342	if (XFS_FORCED_SHUTDOWN(mp))
1343		return -EIO;
1344
1345	prid = xfs_get_initial_prid(dp);
1346
1347	/*
1348	 * Make sure that we have allocated dquot(s) on disk.
1349	 */
1350	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1351				xfs_kgid_to_gid(current_fsgid()), prid,
1352				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1353				&udqp, &gdqp, &pdqp);
1354	if (error)
1355		return error;
1356
1357	resblks = XFS_IALLOC_SPACE_RES(mp);
1358	tres = &M_RES(mp)->tr_create_tmpfile;
1359
1360	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1361	if (error == -ENOSPC) {
1362		/* No space at all so try a "no-allocation" reservation */
1363		resblks = 0;
1364		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1365	}
1366	if (error)
1367		goto out_release_inode;
1368
1369	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1370						pdqp, resblks, 1, 0);
1371	if (error)
1372		goto out_trans_cancel;
1373
1374	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1375				prid, resblks > 0, &ip, NULL);
1376	if (error)
1377		goto out_trans_cancel;
1378
1379	if (mp->m_flags & XFS_MOUNT_WSYNC)
1380		xfs_trans_set_sync(tp);
1381
1382	/*
1383	 * Attach the dquot(s) to the inodes and modify them incore.
1384	 * These ids of the inode couldn't have changed since the new
1385	 * inode has been locked ever since it was created.
1386	 */
1387	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1388
1389	error = xfs_iunlink(tp, ip);
1390	if (error)
1391		goto out_trans_cancel;
1392
1393	error = xfs_trans_commit(tp);
1394	if (error)
1395		goto out_release_inode;
1396
1397	xfs_qm_dqrele(udqp);
1398	xfs_qm_dqrele(gdqp);
1399	xfs_qm_dqrele(pdqp);
1400
1401	*ipp = ip;
1402	return 0;
1403
1404 out_trans_cancel:
1405	xfs_trans_cancel(tp);
1406 out_release_inode:
1407	/*
1408	 * Wait until after the current transaction is aborted to finish the
1409	 * setup of the inode and release the inode.  This prevents recursive
1410	 * transactions and deadlocks from xfs_inactive.
1411	 */
1412	if (ip) {
1413		xfs_finish_inode_setup(ip);
1414		IRELE(ip);
1415	}
1416
1417	xfs_qm_dqrele(udqp);
1418	xfs_qm_dqrele(gdqp);
1419	xfs_qm_dqrele(pdqp);
1420
1421	return error;
1422}
1423
1424int
1425xfs_link(
1426	xfs_inode_t		*tdp,
1427	xfs_inode_t		*sip,
1428	struct xfs_name		*target_name)
1429{
1430	xfs_mount_t		*mp = tdp->i_mount;
1431	xfs_trans_t		*tp;
1432	int			error;
1433	struct xfs_defer_ops	dfops;
1434	xfs_fsblock_t           first_block;
1435	int			resblks;
1436
1437	trace_xfs_link(tdp, target_name);
1438
1439	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1440
1441	if (XFS_FORCED_SHUTDOWN(mp))
1442		return -EIO;
1443
1444	error = xfs_qm_dqattach(sip, 0);
1445	if (error)
1446		goto std_return;
1447
1448	error = xfs_qm_dqattach(tdp, 0);
1449	if (error)
1450		goto std_return;
1451
1452	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1453	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1454	if (error == -ENOSPC) {
1455		resblks = 0;
1456		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1457	}
1458	if (error)
1459		goto std_return;
1460
1461	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1462
1463	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1464	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1465
1466	/*
1467	 * If we are using project inheritance, we only allow hard link
1468	 * creation in our tree when the project IDs are the same; else
1469	 * the tree quota mechanism could be circumvented.
1470	 */
1471	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1472		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1473		error = -EXDEV;
1474		goto error_return;
1475	}
1476
1477	if (!resblks) {
1478		error = xfs_dir_canenter(tp, tdp, target_name);
1479		if (error)
1480			goto error_return;
1481	}
1482
1483	xfs_defer_init(&dfops, &first_block);
1484
1485	/*
1486	 * Handle initial link state of O_TMPFILE inode
1487	 */
1488	if (VFS_I(sip)->i_nlink == 0) {
1489		error = xfs_iunlink_remove(tp, sip);
1490		if (error)
1491			goto error_return;
1492	}
1493
1494	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1495					&first_block, &dfops, resblks);
1496	if (error)
1497		goto error_return;
1498	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1499	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1500
1501	error = xfs_bumplink(tp, sip);
1502	if (error)
1503		goto error_return;
1504
1505	/*
1506	 * If this is a synchronous mount, make sure that the
1507	 * link transaction goes to disk before returning to
1508	 * the user.
1509	 */
1510	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1511		xfs_trans_set_sync(tp);
1512
1513	error = xfs_defer_finish(&tp, &dfops, NULL);
1514	if (error) {
1515		xfs_defer_cancel(&dfops);
1516		goto error_return;
1517	}
1518
1519	return xfs_trans_commit(tp);
1520
1521 error_return:
1522	xfs_trans_cancel(tp);
1523 std_return:
1524	return error;
1525}
1526
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1527/*
1528 * Free up the underlying blocks past new_size.  The new size must be smaller
1529 * than the current size.  This routine can be used both for the attribute and
1530 * data fork, and does not modify the inode size, which is left to the caller.
1531 *
1532 * The transaction passed to this routine must have made a permanent log
1533 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1534 * given transaction and start new ones, so make sure everything involved in
1535 * the transaction is tidy before calling here.  Some transaction will be
1536 * returned to the caller to be committed.  The incoming transaction must
1537 * already include the inode, and both inode locks must be held exclusively.
1538 * The inode must also be "held" within the transaction.  On return the inode
1539 * will be "held" within the returned transaction.  This routine does NOT
1540 * require any disk space to be reserved for it within the transaction.
1541 *
1542 * If we get an error, we must return with the inode locked and linked into the
1543 * current transaction. This keeps things simple for the higher level code,
1544 * because it always knows that the inode is locked and held in the transaction
1545 * that returns to it whether errors occur or not.  We don't mark the inode
1546 * dirty on error so that transactions can be easily aborted if possible.
1547 */
1548int
1549xfs_itruncate_extents(
1550	struct xfs_trans	**tpp,
1551	struct xfs_inode	*ip,
1552	int			whichfork,
1553	xfs_fsize_t		new_size)
 
1554{
1555	struct xfs_mount	*mp = ip->i_mount;
1556	struct xfs_trans	*tp = *tpp;
1557	struct xfs_defer_ops	dfops;
1558	xfs_fsblock_t		first_block;
1559	xfs_fileoff_t		first_unmap_block;
1560	xfs_fileoff_t		last_block;
1561	xfs_filblks_t		unmap_len;
1562	int			error = 0;
1563	int			done = 0;
1564
1565	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1566	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1567	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1568	ASSERT(new_size <= XFS_ISIZE(ip));
1569	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1570	ASSERT(ip->i_itemp != NULL);
1571	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1572	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1573
1574	trace_xfs_itruncate_extents_start(ip, new_size);
1575
 
 
1576	/*
1577	 * Since it is possible for space to become allocated beyond
1578	 * the end of the file (in a crash where the space is allocated
1579	 * but the inode size is not yet updated), simply remove any
1580	 * blocks which show up between the new EOF and the maximum
1581	 * possible file size.  If the first block to be removed is
1582	 * beyond the maximum file size (ie it is the same as last_block),
1583	 * then there is nothing to do.
 
1584	 */
1585	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1586	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1587	if (first_unmap_block == last_block)
1588		return 0;
 
1589
1590	ASSERT(first_unmap_block < last_block);
1591	unmap_len = last_block - first_unmap_block + 1;
1592	while (!done) {
1593		xfs_defer_init(&dfops, &first_block);
1594		error = xfs_bunmapi(tp, ip,
1595				    first_unmap_block, unmap_len,
1596				    xfs_bmapi_aflag(whichfork),
1597				    XFS_ITRUNC_MAX_EXTENTS,
1598				    &first_block, &dfops,
1599				    &done);
1600		if (error)
1601			goto out_bmap_cancel;
1602
1603		/*
1604		 * Duplicate the transaction that has the permanent
1605		 * reservation and commit the old transaction.
1606		 */
1607		error = xfs_defer_finish(&tp, &dfops, ip);
1608		if (error)
1609			goto out_bmap_cancel;
1610
1611		error = xfs_trans_roll(&tp, ip);
1612		if (error)
1613			goto out;
1614	}
1615
1616	/* Remove all pending CoW reservations. */
1617	error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1618			last_block, true);
1619	if (error)
1620		goto out;
 
1621
1622	/*
1623	 * Clear the reflink flag if we truncated everything.
1624	 */
1625	if (ip->i_d.di_nblocks == 0 && xfs_is_reflink_inode(ip)) {
1626		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1627		xfs_inode_clear_cowblocks_tag(ip);
1628	}
1629
1630	/*
1631	 * Always re-log the inode so that our permanent transaction can keep
1632	 * on rolling it forward in the log.
1633	 */
1634	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1635
1636	trace_xfs_itruncate_extents_end(ip, new_size);
1637
1638out:
1639	*tpp = tp;
1640	return error;
1641out_bmap_cancel:
1642	/*
1643	 * If the bunmapi call encounters an error, return to the caller where
1644	 * the transaction can be properly aborted.  We just need to make sure
1645	 * we're not holding any resources that we were not when we came in.
1646	 */
1647	xfs_defer_cancel(&dfops);
1648	goto out;
1649}
1650
1651int
1652xfs_release(
1653	xfs_inode_t	*ip)
1654{
1655	xfs_mount_t	*mp = ip->i_mount;
1656	int		error;
1657
1658	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1659		return 0;
1660
1661	/* If this is a read-only mount, don't do this (would generate I/O) */
1662	if (mp->m_flags & XFS_MOUNT_RDONLY)
1663		return 0;
1664
1665	if (!XFS_FORCED_SHUTDOWN(mp)) {
1666		int truncated;
1667
1668		/*
1669		 * If we previously truncated this file and removed old data
1670		 * in the process, we want to initiate "early" writeout on
1671		 * the last close.  This is an attempt to combat the notorious
1672		 * NULL files problem which is particularly noticeable from a
1673		 * truncate down, buffered (re-)write (delalloc), followed by
1674		 * a crash.  What we are effectively doing here is
1675		 * significantly reducing the time window where we'd otherwise
1676		 * be exposed to that problem.
1677		 */
1678		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1679		if (truncated) {
1680			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1681			if (ip->i_delayed_blks > 0) {
1682				error = filemap_flush(VFS_I(ip)->i_mapping);
1683				if (error)
1684					return error;
1685			}
1686		}
1687	}
1688
1689	if (VFS_I(ip)->i_nlink == 0)
1690		return 0;
1691
1692	if (xfs_can_free_eofblocks(ip, false)) {
1693
1694		/*
1695		 * Check if the inode is being opened, written and closed
1696		 * frequently and we have delayed allocation blocks outstanding
1697		 * (e.g. streaming writes from the NFS server), truncating the
1698		 * blocks past EOF will cause fragmentation to occur.
1699		 *
1700		 * In this case don't do the truncation, but we have to be
1701		 * careful how we detect this case. Blocks beyond EOF show up as
1702		 * i_delayed_blks even when the inode is clean, so we need to
1703		 * truncate them away first before checking for a dirty release.
1704		 * Hence on the first dirty close we will still remove the
1705		 * speculative allocation, but after that we will leave it in
1706		 * place.
1707		 */
1708		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1709			return 0;
1710		/*
1711		 * If we can't get the iolock just skip truncating the blocks
1712		 * past EOF because we could deadlock with the mmap_sem
1713		 * otherwise. We'll get another chance to drop them once the
1714		 * last reference to the inode is dropped, so we'll never leak
1715		 * blocks permanently.
1716		 */
1717		if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1718			error = xfs_free_eofblocks(ip);
1719			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1720			if (error)
1721				return error;
1722		}
1723
1724		/* delalloc blocks after truncation means it really is dirty */
1725		if (ip->i_delayed_blks)
1726			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1727	}
1728	return 0;
1729}
1730
1731/*
1732 * xfs_inactive_truncate
1733 *
1734 * Called to perform a truncate when an inode becomes unlinked.
1735 */
1736STATIC int
1737xfs_inactive_truncate(
1738	struct xfs_inode *ip)
1739{
1740	struct xfs_mount	*mp = ip->i_mount;
1741	struct xfs_trans	*tp;
1742	int			error;
1743
1744	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1745	if (error) {
1746		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1747		return error;
1748	}
1749
1750	xfs_ilock(ip, XFS_ILOCK_EXCL);
1751	xfs_trans_ijoin(tp, ip, 0);
1752
1753	/*
1754	 * Log the inode size first to prevent stale data exposure in the event
1755	 * of a system crash before the truncate completes. See the related
1756	 * comment in xfs_vn_setattr_size() for details.
1757	 */
1758	ip->i_d.di_size = 0;
1759	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1760
1761	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1762	if (error)
1763		goto error_trans_cancel;
1764
1765	ASSERT(ip->i_d.di_nextents == 0);
1766
1767	error = xfs_trans_commit(tp);
1768	if (error)
1769		goto error_unlock;
1770
1771	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1772	return 0;
1773
1774error_trans_cancel:
1775	xfs_trans_cancel(tp);
1776error_unlock:
1777	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1778	return error;
1779}
1780
1781/*
1782 * xfs_inactive_ifree()
1783 *
1784 * Perform the inode free when an inode is unlinked.
1785 */
1786STATIC int
1787xfs_inactive_ifree(
1788	struct xfs_inode *ip)
1789{
1790	struct xfs_defer_ops	dfops;
1791	xfs_fsblock_t		first_block;
1792	struct xfs_mount	*mp = ip->i_mount;
1793	struct xfs_trans	*tp;
1794	int			error;
1795
1796	/*
1797	 * We try to use a per-AG reservation for any block needed by the finobt
1798	 * tree, but as the finobt feature predates the per-AG reservation
1799	 * support a degraded file system might not have enough space for the
1800	 * reservation at mount time.  In that case try to dip into the reserved
1801	 * pool and pray.
1802	 *
1803	 * Send a warning if the reservation does happen to fail, as the inode
1804	 * now remains allocated and sits on the unlinked list until the fs is
1805	 * repaired.
1806	 */
1807	if (unlikely(mp->m_inotbt_nores)) {
1808		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1809				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1810				&tp);
1811	} else {
1812		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1813	}
1814	if (error) {
1815		if (error == -ENOSPC) {
1816			xfs_warn_ratelimited(mp,
1817			"Failed to remove inode(s) from unlinked list. "
1818			"Please free space, unmount and run xfs_repair.");
1819		} else {
1820			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1821		}
1822		return error;
1823	}
1824
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1825	xfs_ilock(ip, XFS_ILOCK_EXCL);
1826	xfs_trans_ijoin(tp, ip, 0);
1827
1828	xfs_defer_init(&dfops, &first_block);
1829	error = xfs_ifree(tp, ip, &dfops);
1830	if (error) {
1831		/*
1832		 * If we fail to free the inode, shut down.  The cancel
1833		 * might do that, we need to make sure.  Otherwise the
1834		 * inode might be lost for a long time or forever.
1835		 */
1836		if (!XFS_FORCED_SHUTDOWN(mp)) {
1837			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1838				__func__, error);
1839			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1840		}
1841		xfs_trans_cancel(tp);
1842		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1843		return error;
1844	}
1845
1846	/*
1847	 * Credit the quota account(s). The inode is gone.
1848	 */
1849	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1850
1851	/*
1852	 * Just ignore errors at this point.  There is nothing we can do except
1853	 * to try to keep going. Make sure it's not a silent error.
1854	 */
1855	error = xfs_defer_finish(&tp, &dfops, NULL);
1856	if (error) {
1857		xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1858			__func__, error);
1859		xfs_defer_cancel(&dfops);
1860	}
1861	error = xfs_trans_commit(tp);
1862	if (error)
1863		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1864			__func__, error);
1865
1866	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1867	return 0;
1868}
1869
1870/*
1871 * xfs_inactive
1872 *
1873 * This is called when the vnode reference count for the vnode
1874 * goes to zero.  If the file has been unlinked, then it must
1875 * now be truncated.  Also, we clear all of the read-ahead state
1876 * kept for the inode here since the file is now closed.
1877 */
1878void
1879xfs_inactive(
1880	xfs_inode_t	*ip)
1881{
1882	struct xfs_mount	*mp;
1883	int			error;
1884	int			truncate = 0;
1885
1886	/*
1887	 * If the inode is already free, then there can be nothing
1888	 * to clean up here.
1889	 */
1890	if (VFS_I(ip)->i_mode == 0) {
1891		ASSERT(ip->i_df.if_real_bytes == 0);
1892		ASSERT(ip->i_df.if_broot_bytes == 0);
1893		return;
1894	}
1895
1896	mp = ip->i_mount;
1897	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1898
1899	/* If this is a read-only mount, don't do this (would generate I/O) */
1900	if (mp->m_flags & XFS_MOUNT_RDONLY)
1901		return;
1902
 
 
 
 
1903	if (VFS_I(ip)->i_nlink != 0) {
1904		/*
1905		 * force is true because we are evicting an inode from the
1906		 * cache. Post-eof blocks must be freed, lest we end up with
1907		 * broken free space accounting.
 
 
 
 
1908		 */
1909		if (xfs_can_free_eofblocks(ip, true)) {
1910			xfs_ilock(ip, XFS_IOLOCK_EXCL);
1911			xfs_free_eofblocks(ip);
1912			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1913		}
1914
1915		return;
1916	}
1917
1918	if (S_ISREG(VFS_I(ip)->i_mode) &&
1919	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1920	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1921		truncate = 1;
1922
1923	error = xfs_qm_dqattach(ip, 0);
1924	if (error)
1925		return;
1926
1927	if (S_ISLNK(VFS_I(ip)->i_mode))
1928		error = xfs_inactive_symlink(ip);
1929	else if (truncate)
1930		error = xfs_inactive_truncate(ip);
1931	if (error)
1932		return;
1933
1934	/*
1935	 * If there are attributes associated with the file then blow them away
1936	 * now.  The code calls a routine that recursively deconstructs the
1937	 * attribute fork. If also blows away the in-core attribute fork.
1938	 */
1939	if (XFS_IFORK_Q(ip)) {
1940		error = xfs_attr_inactive(ip);
1941		if (error)
1942			return;
1943	}
1944
1945	ASSERT(!ip->i_afp);
1946	ASSERT(ip->i_d.di_anextents == 0);
1947	ASSERT(ip->i_d.di_forkoff == 0);
1948
1949	/*
1950	 * Free the inode.
1951	 */
1952	error = xfs_inactive_ifree(ip);
1953	if (error)
1954		return;
1955
1956	/*
1957	 * Release the dquots held by inode, if any.
1958	 */
1959	xfs_qm_dqdetach(ip);
1960}
1961
1962/*
1963 * This is called when the inode's link count goes to 0 or we are creating a
1964 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1965 * set to true as the link count is dropped to zero by the VFS after we've
1966 * created the file successfully, so we have to add it to the unlinked list
1967 * while the link count is non-zero.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1968 *
1969 * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1970 * list when the inode is freed.
1971 */
1972STATIC int
1973xfs_iunlink(
1974	struct xfs_trans *tp,
1975	struct xfs_inode *ip)
1976{
1977	xfs_mount_t	*mp = tp->t_mountp;
1978	xfs_agi_t	*agi;
1979	xfs_dinode_t	*dip;
1980	xfs_buf_t	*agibp;
1981	xfs_buf_t	*ibp;
1982	xfs_agino_t	agino;
1983	short		bucket_index;
1984	int		offset;
1985	int		error;
1986
 
1987	ASSERT(VFS_I(ip)->i_mode != 0);
 
1988
1989	/*
1990	 * Get the agi buffer first.  It ensures lock ordering
1991	 * on the list.
1992	 */
1993	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1994	if (error)
1995		return error;
1996	agi = XFS_BUF_TO_AGI(agibp);
1997
1998	/*
1999	 * Get the index into the agi hash table for the
2000	 * list this inode will go on.
 
2001	 */
2002	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2003	ASSERT(agino != 0);
2004	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2005	ASSERT(agi->agi_unlinked[bucket_index]);
2006	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
 
 
 
 
2007
2008	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2009		/*
2010		 * There is already another inode in the bucket we need
2011		 * to add ourselves to.  Add us at the front of the list.
2012		 * Here we put the head pointer into our next pointer,
2013		 * and then we fall through to point the head at us.
2014		 */
2015		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2016				       0, 0);
2017		if (error)
2018			return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2019
2020		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2021		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2022		offset = ip->i_imap.im_boffset +
2023			offsetof(xfs_dinode_t, di_next_unlinked);
2024
2025		/* need to recalc the inode CRC if appropriate */
2026		xfs_dinode_calc_crc(mp, dip);
 
 
 
2027
2028		xfs_trans_inode_buf(tp, ibp);
2029		xfs_trans_log_buf(tp, ibp, offset,
2030				  (offset + sizeof(xfs_agino_t) - 1));
2031		xfs_inobp_check(mp, ibp);
 
 
 
 
 
 
 
 
 
 
2032	}
2033
2034	/*
2035	 * Point the bucket head pointer at the inode being inserted.
2036	 */
2037	ASSERT(agino != 0);
2038	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2039	offset = offsetof(xfs_agi_t, agi_unlinked) +
2040		(sizeof(xfs_agino_t) * bucket_index);
2041	xfs_trans_log_buf(tp, agibp, offset,
2042			  (offset + sizeof(xfs_agino_t) - 1));
2043	return 0;
2044}
2045
2046/*
2047 * Pull the on-disk inode from the AGI unlinked list.
2048 */
2049STATIC int
2050xfs_iunlink_remove(
2051	xfs_trans_t	*tp,
2052	xfs_inode_t	*ip)
2053{
2054	xfs_ino_t	next_ino;
2055	xfs_mount_t	*mp;
2056	xfs_agi_t	*agi;
2057	xfs_dinode_t	*dip;
2058	xfs_buf_t	*agibp;
2059	xfs_buf_t	*ibp;
2060	xfs_agnumber_t	agno;
2061	xfs_agino_t	agino;
2062	xfs_agino_t	next_agino;
2063	xfs_buf_t	*last_ibp;
2064	xfs_dinode_t	*last_dip = NULL;
2065	short		bucket_index;
2066	int		offset, last_offset = 0;
2067	int		error;
 
 
 
 
 
2068
2069	mp = tp->t_mountp;
2070	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
 
 
 
 
 
 
 
 
2071
2072	/*
2073	 * Get the agi buffer first.  It ensures lock ordering
2074	 * on the list.
 
2075	 */
2076	error = xfs_read_agi(mp, tp, agno, &agibp);
2077	if (error)
2078		return error;
2079
2080	agi = XFS_BUF_TO_AGI(agibp);
2081
2082	/*
2083	 * Get the index into the agi hash table for the
2084	 * list this inode will go on.
 
 
 
2085	 */
2086	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2087	ASSERT(agino != 0);
2088	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2089	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2090	ASSERT(agi->agi_unlinked[bucket_index]);
2091
2092	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2093		/*
2094		 * We're at the head of the list.  Get the inode's on-disk
2095		 * buffer to see if there is anyone after us on the list.
2096		 * Only modify our next pointer if it is not already NULLAGINO.
2097		 * This saves us the overhead of dealing with the buffer when
2098		 * there is no need to change it.
2099		 */
2100		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2101				       0, 0);
2102		if (error) {
2103			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2104				__func__, error);
2105			return error;
2106		}
2107		next_agino = be32_to_cpu(dip->di_next_unlinked);
2108		ASSERT(next_agino != 0);
2109		if (next_agino != NULLAGINO) {
2110			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2111			offset = ip->i_imap.im_boffset +
2112				offsetof(xfs_dinode_t, di_next_unlinked);
2113
2114			/* need to recalc the inode CRC if appropriate */
2115			xfs_dinode_calc_crc(mp, dip);
2116
2117			xfs_trans_inode_buf(tp, ibp);
2118			xfs_trans_log_buf(tp, ibp, offset,
2119					  (offset + sizeof(xfs_agino_t) - 1));
2120			xfs_inobp_check(mp, ibp);
2121		} else {
2122			xfs_trans_brelse(tp, ibp);
2123		}
2124		/*
2125		 * Point the bucket head pointer at the next inode.
 
 
 
 
 
2126		 */
2127		ASSERT(next_agino != 0);
2128		ASSERT(next_agino != agino);
2129		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2130		offset = offsetof(xfs_agi_t, agi_unlinked) +
2131			(sizeof(xfs_agino_t) * bucket_index);
2132		xfs_trans_log_buf(tp, agibp, offset,
2133				  (offset + sizeof(xfs_agino_t) - 1));
2134	} else {
2135		/*
2136		 * We need to search the list for the inode being freed.
2137		 */
2138		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2139		last_ibp = NULL;
2140		while (next_agino != agino) {
2141			struct xfs_imap	imap;
2142
2143			if (last_ibp)
2144				xfs_trans_brelse(tp, last_ibp);
2145
2146			imap.im_blkno = 0;
2147			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2148
2149			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2150			if (error) {
2151				xfs_warn(mp,
2152	"%s: xfs_imap returned error %d.",
2153					 __func__, error);
2154				return error;
2155			}
2156
2157			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2158					       &last_ibp, 0, 0);
2159			if (error) {
2160				xfs_warn(mp,
2161	"%s: xfs_imap_to_bp returned error %d.",
2162					__func__, error);
2163				return error;
2164			}
2165
2166			last_offset = imap.im_boffset;
2167			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2168			ASSERT(next_agino != NULLAGINO);
2169			ASSERT(next_agino != 0);
2170		}
 
 
 
 
 
 
 
2171
2172		/*
2173		 * Now last_ibp points to the buffer previous to us on the
2174		 * unlinked list.  Pull us from the list.
2175		 */
2176		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2177				       0, 0);
2178		if (error) {
2179			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2180				__func__, error);
2181			return error;
 
 
2182		}
2183		next_agino = be32_to_cpu(dip->di_next_unlinked);
2184		ASSERT(next_agino != 0);
2185		ASSERT(next_agino != agino);
2186		if (next_agino != NULLAGINO) {
2187			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2188			offset = ip->i_imap.im_boffset +
2189				offsetof(xfs_dinode_t, di_next_unlinked);
2190
2191			/* need to recalc the inode CRC if appropriate */
2192			xfs_dinode_calc_crc(mp, dip);
2193
2194			xfs_trans_inode_buf(tp, ibp);
2195			xfs_trans_log_buf(tp, ibp, offset,
2196					  (offset + sizeof(xfs_agino_t) - 1));
2197			xfs_inobp_check(mp, ibp);
2198		} else {
2199			xfs_trans_brelse(tp, ibp);
2200		}
2201		/*
2202		 * Point the previous inode on the list to the next inode.
2203		 */
2204		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2205		ASSERT(next_agino != 0);
2206		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2207
2208		/* need to recalc the inode CRC if appropriate */
2209		xfs_dinode_calc_crc(mp, last_dip);
 
 
 
 
 
 
 
 
 
2210
2211		xfs_trans_inode_buf(tp, last_ibp);
2212		xfs_trans_log_buf(tp, last_ibp, offset,
2213				  (offset + sizeof(xfs_agino_t) - 1));
2214		xfs_inobp_check(mp, last_ibp);
 
 
 
 
 
2215	}
2216	return 0;
 
 
 
 
 
 
 
 
 
 
 
2217}
2218
2219/*
2220 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2221 * inodes that are in memory - they all must be marked stale and attached to
2222 * the cluster buffer.
2223 */
2224STATIC int
2225xfs_ifree_cluster(
2226	xfs_inode_t		*free_ip,
2227	xfs_trans_t		*tp,
2228	struct xfs_icluster	*xic)
2229{
2230	xfs_mount_t		*mp = free_ip->i_mount;
2231	int			blks_per_cluster;
2232	int			inodes_per_cluster;
 
 
2233	int			nbufs;
2234	int			i, j;
2235	int			ioffset;
2236	xfs_daddr_t		blkno;
2237	xfs_buf_t		*bp;
2238	xfs_inode_t		*ip;
2239	xfs_inode_log_item_t	*iip;
2240	xfs_log_item_t		*lip;
2241	struct xfs_perag	*pag;
2242	xfs_ino_t		inum;
2243
2244	inum = xic->first_ino;
2245	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2246	blks_per_cluster = xfs_icluster_size_fsb(mp);
2247	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2248	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2249
2250	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2251		/*
2252		 * The allocation bitmap tells us which inodes of the chunk were
2253		 * physically allocated. Skip the cluster if an inode falls into
2254		 * a sparse region.
2255		 */
2256		ioffset = inum - xic->first_ino;
2257		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2258			ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2259			continue;
2260		}
2261
2262		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2263					 XFS_INO_TO_AGBNO(mp, inum));
2264
2265		/*
2266		 * We obtain and lock the backing buffer first in the process
2267		 * here, as we have to ensure that any dirty inode that we
2268		 * can't get the flush lock on is attached to the buffer.
2269		 * If we scan the in-memory inodes first, then buffer IO can
2270		 * complete before we get a lock on it, and hence we may fail
2271		 * to mark all the active inodes on the buffer stale.
2272		 */
2273		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2274					mp->m_bsize * blks_per_cluster,
2275					XBF_UNMAPPED);
2276
2277		if (!bp)
2278			return -ENOMEM;
2279
2280		/*
2281		 * This buffer may not have been correctly initialised as we
2282		 * didn't read it from disk. That's not important because we are
2283		 * only using to mark the buffer as stale in the log, and to
2284		 * attach stale cached inodes on it. That means it will never be
2285		 * dispatched for IO. If it is, we want to know about it, and we
2286		 * want it to fail. We can acheive this by adding a write
2287		 * verifier to the buffer.
2288		 */
2289		 bp->b_ops = &xfs_inode_buf_ops;
2290
2291		/*
2292		 * Walk the inodes already attached to the buffer and mark them
2293		 * stale. These will all have the flush locks held, so an
2294		 * in-memory inode walk can't lock them. By marking them all
2295		 * stale first, we will not attempt to lock them in the loop
2296		 * below as the XFS_ISTALE flag will be set.
2297		 */
2298		lip = bp->b_fspriv;
2299		while (lip) {
2300			if (lip->li_type == XFS_LI_INODE) {
2301				iip = (xfs_inode_log_item_t *)lip;
2302				ASSERT(iip->ili_logged == 1);
2303				lip->li_cb = xfs_istale_done;
2304				xfs_trans_ail_copy_lsn(mp->m_ail,
2305							&iip->ili_flush_lsn,
2306							&iip->ili_item.li_lsn);
2307				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2308			}
2309			lip = lip->li_bio_list;
2310		}
2311
2312
2313		/*
2314		 * For each inode in memory attempt to add it to the inode
2315		 * buffer and set it up for being staled on buffer IO
2316		 * completion.  This is safe as we've locked out tail pushing
2317		 * and flushing by locking the buffer.
2318		 *
2319		 * We have already marked every inode that was part of a
2320		 * transaction stale above, which means there is no point in
2321		 * even trying to lock them.
2322		 */
2323		for (i = 0; i < inodes_per_cluster; i++) {
2324retry:
2325			rcu_read_lock();
2326			ip = radix_tree_lookup(&pag->pag_ici_root,
2327					XFS_INO_TO_AGINO(mp, (inum + i)));
2328
2329			/* Inode not in memory, nothing to do */
2330			if (!ip) {
2331				rcu_read_unlock();
2332				continue;
2333			}
2334
2335			/*
2336			 * because this is an RCU protected lookup, we could
2337			 * find a recently freed or even reallocated inode
2338			 * during the lookup. We need to check under the
2339			 * i_flags_lock for a valid inode here. Skip it if it
2340			 * is not valid, the wrong inode or stale.
2341			 */
2342			spin_lock(&ip->i_flags_lock);
2343			if (ip->i_ino != inum + i ||
2344			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2345				spin_unlock(&ip->i_flags_lock);
2346				rcu_read_unlock();
2347				continue;
2348			}
2349			spin_unlock(&ip->i_flags_lock);
2350
2351			/*
2352			 * Don't try to lock/unlock the current inode, but we
2353			 * _cannot_ skip the other inodes that we did not find
2354			 * in the list attached to the buffer and are not
2355			 * already marked stale. If we can't lock it, back off
2356			 * and retry.
2357			 */
2358			if (ip != free_ip &&
2359			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2360				rcu_read_unlock();
2361				delay(1);
2362				goto retry;
2363			}
2364			rcu_read_unlock();
2365
2366			xfs_iflock(ip);
2367			xfs_iflags_set(ip, XFS_ISTALE);
2368
2369			/*
2370			 * we don't need to attach clean inodes or those only
2371			 * with unlogged changes (which we throw away, anyway).
2372			 */
2373			iip = ip->i_itemp;
2374			if (!iip || xfs_inode_clean(ip)) {
2375				ASSERT(ip != free_ip);
2376				xfs_ifunlock(ip);
2377				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2378				continue;
2379			}
2380
2381			iip->ili_last_fields = iip->ili_fields;
2382			iip->ili_fields = 0;
2383			iip->ili_fsync_fields = 0;
2384			iip->ili_logged = 1;
2385			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2386						&iip->ili_item.li_lsn);
2387
2388			xfs_buf_attach_iodone(bp, xfs_istale_done,
2389						  &iip->ili_item);
2390
2391			if (ip != free_ip)
2392				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2393		}
2394
2395		xfs_trans_stale_inode_buf(tp, bp);
2396		xfs_trans_binval(tp, bp);
2397	}
2398
2399	xfs_perag_put(pag);
2400	return 0;
2401}
2402
2403/*
2404 * This is called to return an inode to the inode free list.
2405 * The inode should already be truncated to 0 length and have
2406 * no pages associated with it.  This routine also assumes that
2407 * the inode is already a part of the transaction.
2408 *
2409 * The on-disk copy of the inode will have been added to the list
2410 * of unlinked inodes in the AGI. We need to remove the inode from
2411 * that list atomically with respect to freeing it here.
2412 */
2413int
2414xfs_ifree(
2415	xfs_trans_t	*tp,
2416	xfs_inode_t	*ip,
2417	struct xfs_defer_ops	*dfops)
2418{
2419	int			error;
2420	struct xfs_icluster	xic = { 0 };
 
2421
2422	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2423	ASSERT(VFS_I(ip)->i_nlink == 0);
2424	ASSERT(ip->i_d.di_nextents == 0);
2425	ASSERT(ip->i_d.di_anextents == 0);
2426	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2427	ASSERT(ip->i_d.di_nblocks == 0);
2428
2429	/*
2430	 * Pull the on-disk inode from the AGI unlinked list.
2431	 */
2432	error = xfs_iunlink_remove(tp, ip);
2433	if (error)
2434		return error;
2435
2436	error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2437	if (error)
2438		return error;
2439
 
 
 
 
 
 
 
 
 
 
 
2440	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2441	ip->i_d.di_flags = 0;
 
2442	ip->i_d.di_dmevmask = 0;
2443	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2444	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2445	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
 
 
 
 
 
2446	/*
2447	 * Bump the generation count so no one will be confused
2448	 * by reincarnations of this inode.
2449	 */
2450	VFS_I(ip)->i_generation++;
2451	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2452
2453	if (xic.deleted)
2454		error = xfs_ifree_cluster(ip, tp, &xic);
2455
2456	return error;
2457}
2458
2459/*
2460 * This is called to unpin an inode.  The caller must have the inode locked
2461 * in at least shared mode so that the buffer cannot be subsequently pinned
2462 * once someone is waiting for it to be unpinned.
2463 */
2464static void
2465xfs_iunpin(
2466	struct xfs_inode	*ip)
2467{
2468	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2469
2470	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2471
2472	/* Give the log a push to start the unpinning I/O */
2473	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2474
2475}
2476
2477static void
2478__xfs_iunpin_wait(
2479	struct xfs_inode	*ip)
2480{
2481	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2482	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2483
2484	xfs_iunpin(ip);
2485
2486	do {
2487		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2488		if (xfs_ipincount(ip))
2489			io_schedule();
2490	} while (xfs_ipincount(ip));
2491	finish_wait(wq, &wait.wait);
2492}
2493
2494void
2495xfs_iunpin_wait(
2496	struct xfs_inode	*ip)
2497{
2498	if (xfs_ipincount(ip))
2499		__xfs_iunpin_wait(ip);
2500}
2501
2502/*
2503 * Removing an inode from the namespace involves removing the directory entry
2504 * and dropping the link count on the inode. Removing the directory entry can
2505 * result in locking an AGF (directory blocks were freed) and removing a link
2506 * count can result in placing the inode on an unlinked list which results in
2507 * locking an AGI.
2508 *
2509 * The big problem here is that we have an ordering constraint on AGF and AGI
2510 * locking - inode allocation locks the AGI, then can allocate a new extent for
2511 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2512 * removes the inode from the unlinked list, requiring that we lock the AGI
2513 * first, and then freeing the inode can result in an inode chunk being freed
2514 * and hence freeing disk space requiring that we lock an AGF.
2515 *
2516 * Hence the ordering that is imposed by other parts of the code is AGI before
2517 * AGF. This means we cannot remove the directory entry before we drop the inode
2518 * reference count and put it on the unlinked list as this results in a lock
2519 * order of AGF then AGI, and this can deadlock against inode allocation and
2520 * freeing. Therefore we must drop the link counts before we remove the
2521 * directory entry.
2522 *
2523 * This is still safe from a transactional point of view - it is not until we
2524 * get to xfs_defer_finish() that we have the possibility of multiple
2525 * transactions in this operation. Hence as long as we remove the directory
2526 * entry and drop the link count in the first transaction of the remove
2527 * operation, there are no transactional constraints on the ordering here.
2528 */
2529int
2530xfs_remove(
2531	xfs_inode_t             *dp,
2532	struct xfs_name		*name,
2533	xfs_inode_t		*ip)
2534{
2535	xfs_mount_t		*mp = dp->i_mount;
2536	xfs_trans_t             *tp = NULL;
2537	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2538	int                     error = 0;
2539	struct xfs_defer_ops	dfops;
2540	xfs_fsblock_t           first_block;
2541	uint			resblks;
2542
2543	trace_xfs_remove(dp, name);
2544
2545	if (XFS_FORCED_SHUTDOWN(mp))
2546		return -EIO;
2547
2548	error = xfs_qm_dqattach(dp, 0);
2549	if (error)
2550		goto std_return;
2551
2552	error = xfs_qm_dqattach(ip, 0);
2553	if (error)
2554		goto std_return;
2555
2556	/*
2557	 * We try to get the real space reservation first,
2558	 * allowing for directory btree deletion(s) implying
2559	 * possible bmap insert(s).  If we can't get the space
2560	 * reservation then we use 0 instead, and avoid the bmap
2561	 * btree insert(s) in the directory code by, if the bmap
2562	 * insert tries to happen, instead trimming the LAST
2563	 * block from the directory.
2564	 */
2565	resblks = XFS_REMOVE_SPACE_RES(mp);
2566	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2567	if (error == -ENOSPC) {
2568		resblks = 0;
2569		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2570				&tp);
2571	}
2572	if (error) {
2573		ASSERT(error != -ENOSPC);
2574		goto std_return;
2575	}
2576
2577	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2578
2579	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2580	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2581
2582	/*
2583	 * If we're removing a directory perform some additional validation.
2584	 */
2585	if (is_dir) {
2586		ASSERT(VFS_I(ip)->i_nlink >= 2);
2587		if (VFS_I(ip)->i_nlink != 2) {
2588			error = -ENOTEMPTY;
2589			goto out_trans_cancel;
2590		}
2591		if (!xfs_dir_isempty(ip)) {
2592			error = -ENOTEMPTY;
2593			goto out_trans_cancel;
2594		}
2595
2596		/* Drop the link from ip's "..".  */
2597		error = xfs_droplink(tp, dp);
2598		if (error)
2599			goto out_trans_cancel;
2600
2601		/* Drop the "." link from ip to self.  */
2602		error = xfs_droplink(tp, ip);
2603		if (error)
2604			goto out_trans_cancel;
2605	} else {
2606		/*
2607		 * When removing a non-directory we need to log the parent
2608		 * inode here.  For a directory this is done implicitly
2609		 * by the xfs_droplink call for the ".." entry.
2610		 */
2611		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2612	}
2613	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2614
2615	/* Drop the link from dp to ip. */
2616	error = xfs_droplink(tp, ip);
2617	if (error)
2618		goto out_trans_cancel;
2619
2620	xfs_defer_init(&dfops, &first_block);
2621	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2622					&first_block, &dfops, resblks);
2623	if (error) {
2624		ASSERT(error != -ENOENT);
2625		goto out_bmap_cancel;
2626	}
2627
2628	/*
2629	 * If this is a synchronous mount, make sure that the
2630	 * remove transaction goes to disk before returning to
2631	 * the user.
2632	 */
2633	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2634		xfs_trans_set_sync(tp);
2635
2636	error = xfs_defer_finish(&tp, &dfops, NULL);
2637	if (error)
2638		goto out_bmap_cancel;
2639
2640	error = xfs_trans_commit(tp);
2641	if (error)
2642		goto std_return;
2643
2644	if (is_dir && xfs_inode_is_filestream(ip))
2645		xfs_filestream_deassociate(ip);
2646
2647	return 0;
2648
2649 out_bmap_cancel:
2650	xfs_defer_cancel(&dfops);
2651 out_trans_cancel:
2652	xfs_trans_cancel(tp);
2653 std_return:
2654	return error;
2655}
2656
2657/*
2658 * Enter all inodes for a rename transaction into a sorted array.
2659 */
2660#define __XFS_SORT_INODES	5
2661STATIC void
2662xfs_sort_for_rename(
2663	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2664	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2665	struct xfs_inode	*ip1,	/* in: inode of old entry */
2666	struct xfs_inode	*ip2,	/* in: inode of new entry */
2667	struct xfs_inode	*wip,	/* in: whiteout inode */
2668	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2669	int			*num_inodes)  /* in/out: inodes in array */
2670{
2671	int			i, j;
2672
2673	ASSERT(*num_inodes == __XFS_SORT_INODES);
2674	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2675
2676	/*
2677	 * i_tab contains a list of pointers to inodes.  We initialize
2678	 * the table here & we'll sort it.  We will then use it to
2679	 * order the acquisition of the inode locks.
2680	 *
2681	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2682	 */
2683	i = 0;
2684	i_tab[i++] = dp1;
2685	i_tab[i++] = dp2;
2686	i_tab[i++] = ip1;
2687	if (ip2)
2688		i_tab[i++] = ip2;
2689	if (wip)
2690		i_tab[i++] = wip;
2691	*num_inodes = i;
2692
2693	/*
2694	 * Sort the elements via bubble sort.  (Remember, there are at
2695	 * most 5 elements to sort, so this is adequate.)
2696	 */
2697	for (i = 0; i < *num_inodes; i++) {
2698		for (j = 1; j < *num_inodes; j++) {
2699			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2700				struct xfs_inode *temp = i_tab[j];
2701				i_tab[j] = i_tab[j-1];
2702				i_tab[j-1] = temp;
2703			}
2704		}
2705	}
2706}
2707
2708static int
2709xfs_finish_rename(
2710	struct xfs_trans	*tp,
2711	struct xfs_defer_ops	*dfops)
2712{
2713	int			error;
2714
2715	/*
2716	 * If this is a synchronous mount, make sure that the rename transaction
2717	 * goes to disk before returning to the user.
2718	 */
2719	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2720		xfs_trans_set_sync(tp);
2721
2722	error = xfs_defer_finish(&tp, dfops, NULL);
2723	if (error) {
2724		xfs_defer_cancel(dfops);
2725		xfs_trans_cancel(tp);
2726		return error;
2727	}
2728
2729	return xfs_trans_commit(tp);
2730}
2731
2732/*
2733 * xfs_cross_rename()
2734 *
2735 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2736 */
2737STATIC int
2738xfs_cross_rename(
2739	struct xfs_trans	*tp,
2740	struct xfs_inode	*dp1,
2741	struct xfs_name		*name1,
2742	struct xfs_inode	*ip1,
2743	struct xfs_inode	*dp2,
2744	struct xfs_name		*name2,
2745	struct xfs_inode	*ip2,
2746	struct xfs_defer_ops	*dfops,
2747	xfs_fsblock_t		*first_block,
2748	int			spaceres)
2749{
2750	int		error = 0;
2751	int		ip1_flags = 0;
2752	int		ip2_flags = 0;
2753	int		dp2_flags = 0;
2754
2755	/* Swap inode number for dirent in first parent */
2756	error = xfs_dir_replace(tp, dp1, name1,
2757				ip2->i_ino,
2758				first_block, dfops, spaceres);
2759	if (error)
2760		goto out_trans_abort;
2761
2762	/* Swap inode number for dirent in second parent */
2763	error = xfs_dir_replace(tp, dp2, name2,
2764				ip1->i_ino,
2765				first_block, dfops, spaceres);
2766	if (error)
2767		goto out_trans_abort;
2768
2769	/*
2770	 * If we're renaming one or more directories across different parents,
2771	 * update the respective ".." entries (and link counts) to match the new
2772	 * parents.
2773	 */
2774	if (dp1 != dp2) {
2775		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2776
2777		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2778			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2779						dp1->i_ino, first_block,
2780						dfops, spaceres);
2781			if (error)
2782				goto out_trans_abort;
2783
2784			/* transfer ip2 ".." reference to dp1 */
2785			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2786				error = xfs_droplink(tp, dp2);
2787				if (error)
2788					goto out_trans_abort;
2789				error = xfs_bumplink(tp, dp1);
2790				if (error)
2791					goto out_trans_abort;
2792			}
2793
2794			/*
2795			 * Although ip1 isn't changed here, userspace needs
2796			 * to be warned about the change, so that applications
2797			 * relying on it (like backup ones), will properly
2798			 * notify the change
2799			 */
2800			ip1_flags |= XFS_ICHGTIME_CHG;
2801			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2802		}
2803
2804		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2805			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2806						dp2->i_ino, first_block,
2807						dfops, spaceres);
2808			if (error)
2809				goto out_trans_abort;
2810
2811			/* transfer ip1 ".." reference to dp2 */
2812			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2813				error = xfs_droplink(tp, dp1);
2814				if (error)
2815					goto out_trans_abort;
2816				error = xfs_bumplink(tp, dp2);
2817				if (error)
2818					goto out_trans_abort;
2819			}
2820
2821			/*
2822			 * Although ip2 isn't changed here, userspace needs
2823			 * to be warned about the change, so that applications
2824			 * relying on it (like backup ones), will properly
2825			 * notify the change
2826			 */
2827			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2828			ip2_flags |= XFS_ICHGTIME_CHG;
2829		}
2830	}
2831
2832	if (ip1_flags) {
2833		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2834		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2835	}
2836	if (ip2_flags) {
2837		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2838		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2839	}
2840	if (dp2_flags) {
2841		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2842		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2843	}
2844	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2845	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2846	return xfs_finish_rename(tp, dfops);
2847
2848out_trans_abort:
2849	xfs_defer_cancel(dfops);
2850	xfs_trans_cancel(tp);
2851	return error;
2852}
2853
2854/*
2855 * xfs_rename_alloc_whiteout()
2856 *
2857 * Return a referenced, unlinked, unlocked inode that that can be used as a
2858 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2859 * crash between allocating the inode and linking it into the rename transaction
2860 * recovery will free the inode and we won't leak it.
2861 */
2862static int
2863xfs_rename_alloc_whiteout(
2864	struct xfs_inode	*dp,
2865	struct xfs_inode	**wip)
2866{
2867	struct xfs_inode	*tmpfile;
2868	int			error;
2869
2870	error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2871	if (error)
2872		return error;
2873
2874	/*
2875	 * Prepare the tmpfile inode as if it were created through the VFS.
2876	 * Otherwise, the link increment paths will complain about nlink 0->1.
2877	 * Drop the link count as done by d_tmpfile(), complete the inode setup
2878	 * and flag it as linkable.
2879	 */
2880	drop_nlink(VFS_I(tmpfile));
2881	xfs_setup_iops(tmpfile);
2882	xfs_finish_inode_setup(tmpfile);
2883	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2884
2885	*wip = tmpfile;
2886	return 0;
2887}
2888
2889/*
2890 * xfs_rename
2891 */
2892int
2893xfs_rename(
2894	struct xfs_inode	*src_dp,
2895	struct xfs_name		*src_name,
2896	struct xfs_inode	*src_ip,
2897	struct xfs_inode	*target_dp,
2898	struct xfs_name		*target_name,
2899	struct xfs_inode	*target_ip,
2900	unsigned int		flags)
2901{
2902	struct xfs_mount	*mp = src_dp->i_mount;
2903	struct xfs_trans	*tp;
2904	struct xfs_defer_ops	dfops;
2905	xfs_fsblock_t		first_block;
2906	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2907	struct xfs_inode	*inodes[__XFS_SORT_INODES];
 
2908	int			num_inodes = __XFS_SORT_INODES;
2909	bool			new_parent = (src_dp != target_dp);
2910	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2911	int			spaceres;
2912	int			error;
2913
2914	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2915
2916	if ((flags & RENAME_EXCHANGE) && !target_ip)
2917		return -EINVAL;
2918
2919	/*
2920	 * If we are doing a whiteout operation, allocate the whiteout inode
2921	 * we will be placing at the target and ensure the type is set
2922	 * appropriately.
2923	 */
2924	if (flags & RENAME_WHITEOUT) {
2925		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2926		error = xfs_rename_alloc_whiteout(target_dp, &wip);
2927		if (error)
2928			return error;
2929
2930		/* setup target dirent info as whiteout */
2931		src_name->type = XFS_DIR3_FT_CHRDEV;
2932	}
2933
2934	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2935				inodes, &num_inodes);
2936
2937	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2938	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2939	if (error == -ENOSPC) {
2940		spaceres = 0;
2941		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2942				&tp);
2943	}
2944	if (error)
2945		goto out_release_wip;
2946
2947	/*
2948	 * Attach the dquots to the inodes
2949	 */
2950	error = xfs_qm_vop_rename_dqattach(inodes);
2951	if (error)
2952		goto out_trans_cancel;
2953
2954	/*
2955	 * Lock all the participating inodes. Depending upon whether
2956	 * the target_name exists in the target directory, and
2957	 * whether the target directory is the same as the source
2958	 * directory, we can lock from 2 to 4 inodes.
2959	 */
2960	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2961
2962	/*
2963	 * Join all the inodes to the transaction. From this point on,
2964	 * we can rely on either trans_commit or trans_cancel to unlock
2965	 * them.
2966	 */
2967	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2968	if (new_parent)
2969		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2970	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2971	if (target_ip)
2972		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2973	if (wip)
2974		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2975
2976	/*
2977	 * If we are using project inheritance, we only allow renames
2978	 * into our tree when the project IDs are the same; else the
2979	 * tree quota mechanism would be circumvented.
2980	 */
2981	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2982		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2983		error = -EXDEV;
2984		goto out_trans_cancel;
2985	}
2986
2987	xfs_defer_init(&dfops, &first_block);
2988
2989	/* RENAME_EXCHANGE is unique from here on. */
2990	if (flags & RENAME_EXCHANGE)
2991		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2992					target_dp, target_name, target_ip,
2993					&dfops, &first_block, spaceres);
2994
2995	/*
2996	 * Set up the target.
 
2997	 */
2998	if (target_ip == NULL) {
2999		/*
3000		 * If there's no space reservation, check the entry will
3001		 * fit before actually inserting it.
3002		 */
3003		if (!spaceres) {
3004			error = xfs_dir_canenter(tp, target_dp, target_name);
3005			if (error)
3006				goto out_trans_cancel;
3007		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3008		/*
3009		 * If target does not exist and the rename crosses
3010		 * directories, adjust the target directory link count
3011		 * to account for the ".." reference from the new entry.
3012		 */
3013		error = xfs_dir_createname(tp, target_dp, target_name,
3014						src_ip->i_ino, &first_block,
3015						&dfops, spaceres);
3016		if (error)
3017			goto out_bmap_cancel;
3018
3019		xfs_trans_ichgtime(tp, target_dp,
3020					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3021
3022		if (new_parent && src_is_directory) {
3023			error = xfs_bumplink(tp, target_dp);
3024			if (error)
3025				goto out_bmap_cancel;
3026		}
3027	} else { /* target_ip != NULL */
3028		/*
3029		 * If target exists and it's a directory, check that both
3030		 * target and source are directories and that target can be
3031		 * destroyed, or that neither is a directory.
3032		 */
3033		if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3034			/*
3035			 * Make sure target dir is empty.
3036			 */
3037			if (!(xfs_dir_isempty(target_ip)) ||
3038			    (VFS_I(target_ip)->i_nlink > 2)) {
3039				error = -EEXIST;
3040				goto out_trans_cancel;
3041			}
3042		}
3043
3044		/*
3045		 * Link the source inode under the target name.
3046		 * If the source inode is a directory and we are moving
3047		 * it across directories, its ".." entry will be
3048		 * inconsistent until we replace that down below.
3049		 *
3050		 * In case there is already an entry with the same
3051		 * name at the destination directory, remove it first.
3052		 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3053		error = xfs_dir_replace(tp, target_dp, target_name,
3054					src_ip->i_ino,
3055					&first_block, &dfops, spaceres);
3056		if (error)
3057			goto out_bmap_cancel;
3058
3059		xfs_trans_ichgtime(tp, target_dp,
3060					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3061
3062		/*
3063		 * Decrement the link count on the target since the target
3064		 * dir no longer points to it.
3065		 */
3066		error = xfs_droplink(tp, target_ip);
3067		if (error)
3068			goto out_bmap_cancel;
3069
3070		if (src_is_directory) {
3071			/*
3072			 * Drop the link from the old "." entry.
3073			 */
3074			error = xfs_droplink(tp, target_ip);
3075			if (error)
3076				goto out_bmap_cancel;
3077		}
3078	} /* target_ip != NULL */
3079
3080	/*
3081	 * Remove the source.
3082	 */
3083	if (new_parent && src_is_directory) {
3084		/*
3085		 * Rewrite the ".." entry to point to the new
3086		 * directory.
3087		 */
3088		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3089					target_dp->i_ino,
3090					&first_block, &dfops, spaceres);
3091		ASSERT(error != -EEXIST);
3092		if (error)
3093			goto out_bmap_cancel;
3094	}
3095
3096	/*
3097	 * We always want to hit the ctime on the source inode.
3098	 *
3099	 * This isn't strictly required by the standards since the source
3100	 * inode isn't really being changed, but old unix file systems did
3101	 * it and some incremental backup programs won't work without it.
3102	 */
3103	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3104	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3105
3106	/*
3107	 * Adjust the link count on src_dp.  This is necessary when
3108	 * renaming a directory, either within one parent when
3109	 * the target existed, or across two parent directories.
3110	 */
3111	if (src_is_directory && (new_parent || target_ip != NULL)) {
3112
3113		/*
3114		 * Decrement link count on src_directory since the
3115		 * entry that's moved no longer points to it.
3116		 */
3117		error = xfs_droplink(tp, src_dp);
3118		if (error)
3119			goto out_bmap_cancel;
3120	}
3121
3122	/*
3123	 * For whiteouts, we only need to update the source dirent with the
3124	 * inode number of the whiteout inode rather than removing it
3125	 * altogether.
3126	 */
3127	if (wip) {
3128		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3129					&first_block, &dfops, spaceres);
3130	} else
3131		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3132					   &first_block, &dfops, spaceres);
3133	if (error)
3134		goto out_bmap_cancel;
3135
3136	/*
3137	 * For whiteouts, we need to bump the link count on the whiteout inode.
3138	 * This means that failures all the way up to this point leave the inode
3139	 * on the unlinked list and so cleanup is a simple matter of dropping
3140	 * the remaining reference to it. If we fail here after bumping the link
3141	 * count, we're shutting down the filesystem so we'll never see the
3142	 * intermediate state on disk.
3143	 */
3144	if (wip) {
3145		ASSERT(VFS_I(wip)->i_nlink == 0);
3146		error = xfs_bumplink(tp, wip);
3147		if (error)
3148			goto out_bmap_cancel;
3149		error = xfs_iunlink_remove(tp, wip);
3150		if (error)
3151			goto out_bmap_cancel;
3152		xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3153
3154		/*
3155		 * Now we have a real link, clear the "I'm a tmpfile" state
3156		 * flag from the inode so it doesn't accidentally get misused in
3157		 * future.
3158		 */
3159		VFS_I(wip)->i_state &= ~I_LINKABLE;
3160	}
3161
3162	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3163	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3164	if (new_parent)
3165		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3166
3167	error = xfs_finish_rename(tp, &dfops);
3168	if (wip)
3169		IRELE(wip);
3170	return error;
3171
3172out_bmap_cancel:
3173	xfs_defer_cancel(&dfops);
3174out_trans_cancel:
3175	xfs_trans_cancel(tp);
3176out_release_wip:
3177	if (wip)
3178		IRELE(wip);
3179	return error;
3180}
3181
3182STATIC int
3183xfs_iflush_cluster(
3184	struct xfs_inode	*ip,
3185	struct xfs_buf		*bp)
3186{
 
 
3187	struct xfs_mount	*mp = ip->i_mount;
3188	struct xfs_perag	*pag;
3189	unsigned long		first_index, mask;
3190	unsigned long		inodes_per_cluster;
3191	int			cilist_size;
3192	struct xfs_inode	**cilist;
3193	struct xfs_inode	*cip;
3194	int			nr_found;
3195	int			clcount = 0;
3196	int			bufwasdelwri;
3197	int			i;
3198
3199	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3200
3201	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3202	cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3203	cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3204	if (!cilist)
3205		goto out_put;
3206
3207	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3208	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3209	rcu_read_lock();
3210	/* really need a gang lookup range call here */
3211	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3212					first_index, inodes_per_cluster);
3213	if (nr_found == 0)
3214		goto out_free;
3215
3216	for (i = 0; i < nr_found; i++) {
3217		cip = cilist[i];
3218		if (cip == ip)
3219			continue;
3220
3221		/*
3222		 * because this is an RCU protected lookup, we could find a
3223		 * recently freed or even reallocated inode during the lookup.
3224		 * We need to check under the i_flags_lock for a valid inode
3225		 * here. Skip it if it is not valid or the wrong inode.
3226		 */
3227		spin_lock(&cip->i_flags_lock);
3228		if (!cip->i_ino ||
3229		    __xfs_iflags_test(cip, XFS_ISTALE)) {
3230			spin_unlock(&cip->i_flags_lock);
3231			continue;
3232		}
3233
3234		/*
3235		 * Once we fall off the end of the cluster, no point checking
3236		 * any more inodes in the list because they will also all be
3237		 * outside the cluster.
3238		 */
3239		if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3240			spin_unlock(&cip->i_flags_lock);
3241			break;
3242		}
3243		spin_unlock(&cip->i_flags_lock);
3244
3245		/*
3246		 * Do an un-protected check to see if the inode is dirty and
3247		 * is a candidate for flushing.  These checks will be repeated
3248		 * later after the appropriate locks are acquired.
3249		 */
3250		if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3251			continue;
3252
3253		/*
3254		 * Try to get locks.  If any are unavailable or it is pinned,
3255		 * then this inode cannot be flushed and is skipped.
3256		 */
3257
3258		if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3259			continue;
3260		if (!xfs_iflock_nowait(cip)) {
3261			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3262			continue;
3263		}
3264		if (xfs_ipincount(cip)) {
3265			xfs_ifunlock(cip);
3266			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3267			continue;
3268		}
3269
3270
3271		/*
3272		 * Check the inode number again, just to be certain we are not
3273		 * racing with freeing in xfs_reclaim_inode(). See the comments
3274		 * in that function for more information as to why the initial
3275		 * check is not sufficient.
3276		 */
3277		if (!cip->i_ino) {
3278			xfs_ifunlock(cip);
3279			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3280			continue;
3281		}
3282
3283		/*
3284		 * arriving here means that this inode can be flushed.  First
3285		 * re-check that it's dirty before flushing.
3286		 */
3287		if (!xfs_inode_clean(cip)) {
3288			int	error;
3289			error = xfs_iflush_int(cip, bp);
3290			if (error) {
3291				xfs_iunlock(cip, XFS_ILOCK_SHARED);
3292				goto cluster_corrupt_out;
3293			}
3294			clcount++;
3295		} else {
3296			xfs_ifunlock(cip);
3297		}
3298		xfs_iunlock(cip, XFS_ILOCK_SHARED);
3299	}
3300
3301	if (clcount) {
3302		XFS_STATS_INC(mp, xs_icluster_flushcnt);
3303		XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3304	}
3305
3306out_free:
3307	rcu_read_unlock();
3308	kmem_free(cilist);
3309out_put:
3310	xfs_perag_put(pag);
3311	return 0;
3312
3313
3314cluster_corrupt_out:
3315	/*
3316	 * Corruption detected in the clustering loop.  Invalidate the
3317	 * inode buffer and shut down the filesystem.
3318	 */
3319	rcu_read_unlock();
3320	/*
3321	 * Clean up the buffer.  If it was delwri, just release it --
3322	 * brelse can handle it with no problems.  If not, shut down the
3323	 * filesystem before releasing the buffer.
3324	 */
3325	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3326	if (bufwasdelwri)
3327		xfs_buf_relse(bp);
3328
3329	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3330
3331	if (!bufwasdelwri) {
3332		/*
3333		 * Just like incore_relse: if we have b_iodone functions,
3334		 * mark the buffer as an error and call them.  Otherwise
3335		 * mark it as stale and brelse.
3336		 */
3337		if (bp->b_iodone) {
3338			bp->b_flags &= ~XBF_DONE;
3339			xfs_buf_stale(bp);
3340			xfs_buf_ioerror(bp, -EIO);
3341			xfs_buf_ioend(bp);
3342		} else {
3343			xfs_buf_stale(bp);
3344			xfs_buf_relse(bp);
3345		}
3346	}
3347
3348	/*
3349	 * Unlocks the flush lock
3350	 */
3351	xfs_iflush_abort(cip, false);
3352	kmem_free(cilist);
3353	xfs_perag_put(pag);
3354	return -EFSCORRUPTED;
3355}
3356
3357/*
3358 * Flush dirty inode metadata into the backing buffer.
3359 *
3360 * The caller must have the inode lock and the inode flush lock held.  The
3361 * inode lock will still be held upon return to the caller, and the inode
3362 * flush lock will be released after the inode has reached the disk.
3363 *
3364 * The caller must write out the buffer returned in *bpp and release it.
3365 */
3366int
3367xfs_iflush(
3368	struct xfs_inode	*ip,
3369	struct xfs_buf		**bpp)
3370{
3371	struct xfs_mount	*mp = ip->i_mount;
3372	struct xfs_buf		*bp = NULL;
3373	struct xfs_dinode	*dip;
3374	int			error;
3375
3376	XFS_STATS_INC(mp, xs_iflush_count);
3377
3378	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3379	ASSERT(xfs_isiflocked(ip));
3380	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3381	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
 
3382
3383	*bpp = NULL;
3384
3385	xfs_iunpin_wait(ip);
3386
3387	/*
3388	 * For stale inodes we cannot rely on the backing buffer remaining
3389	 * stale in cache for the remaining life of the stale inode and so
3390	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3391	 * inodes below. We have to check this after ensuring the inode is
3392	 * unpinned so that it is safe to reclaim the stale inode after the
3393	 * flush call.
3394	 */
3395	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3396		xfs_ifunlock(ip);
3397		return 0;
3398	}
3399
3400	/*
3401	 * This may have been unpinned because the filesystem is shutting
3402	 * down forcibly. If that's the case we must not write this inode
3403	 * to disk, because the log record didn't make it to disk.
3404	 *
3405	 * We also have to remove the log item from the AIL in this case,
3406	 * as we wait for an empty AIL as part of the unmount process.
3407	 */
3408	if (XFS_FORCED_SHUTDOWN(mp)) {
3409		error = -EIO;
3410		goto abort_out;
3411	}
3412
3413	/*
3414	 * Get the buffer containing the on-disk inode. We are doing a try-lock
3415	 * operation here, so we may get  an EAGAIN error. In that case, we
3416	 * simply want to return with the inode still dirty.
3417	 *
3418	 * If we get any other error, we effectively have a corruption situation
3419	 * and we cannot flush the inode, so we treat it the same as failing
3420	 * xfs_iflush_int().
3421	 */
3422	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3423			       0);
3424	if (error == -EAGAIN) {
3425		xfs_ifunlock(ip);
3426		return error;
3427	}
3428	if (error)
3429		goto corrupt_out;
3430
3431	/*
3432	 * First flush out the inode that xfs_iflush was called with.
3433	 */
3434	error = xfs_iflush_int(ip, bp);
3435	if (error)
3436		goto corrupt_out;
3437
3438	/*
3439	 * If the buffer is pinned then push on the log now so we won't
3440	 * get stuck waiting in the write for too long.
3441	 */
3442	if (xfs_buf_ispinned(bp))
3443		xfs_log_force(mp, 0);
3444
3445	/*
3446	 * inode clustering:
3447	 * see if other inodes can be gathered into this write
3448	 */
3449	error = xfs_iflush_cluster(ip, bp);
3450	if (error)
3451		goto cluster_corrupt_out;
3452
3453	*bpp = bp;
3454	return 0;
3455
3456corrupt_out:
3457	if (bp)
3458		xfs_buf_relse(bp);
3459	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3460cluster_corrupt_out:
3461	error = -EFSCORRUPTED;
3462abort_out:
3463	/*
3464	 * Unlocks the flush lock
3465	 */
3466	xfs_iflush_abort(ip, false);
3467	return error;
3468}
3469
3470STATIC int
3471xfs_iflush_int(
3472	struct xfs_inode	*ip,
3473	struct xfs_buf		*bp)
3474{
3475	struct xfs_inode_log_item *iip = ip->i_itemp;
3476	struct xfs_dinode	*dip;
3477	struct xfs_mount	*mp = ip->i_mount;
3478
3479	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3480	ASSERT(xfs_isiflocked(ip));
3481	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3482	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3483	ASSERT(iip != NULL && iip->ili_fields != 0);
3484	ASSERT(ip->i_d.di_version > 1);
3485
3486	/* set *dip = inode's place in the buffer */
3487	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3488
3489	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3490			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3491		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3492			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3493			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3494		goto corrupt_out;
3495	}
3496	if (S_ISREG(VFS_I(ip)->i_mode)) {
3497		if (XFS_TEST_ERROR(
3498		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3499		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3500		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3501			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3502				"%s: Bad regular inode %Lu, ptr 0x%p",
3503				__func__, ip->i_ino, ip);
3504			goto corrupt_out;
3505		}
3506	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3507		if (XFS_TEST_ERROR(
3508		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3509		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3510		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3511		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3512			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3513				"%s: Bad directory inode %Lu, ptr 0x%p",
3514				__func__, ip->i_ino, ip);
3515			goto corrupt_out;
3516		}
3517	}
3518	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3519				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3520				XFS_RANDOM_IFLUSH_5)) {
3521		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3522			"%s: detected corrupt incore inode %Lu, "
3523			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3524			__func__, ip->i_ino,
3525			ip->i_d.di_nextents + ip->i_d.di_anextents,
3526			ip->i_d.di_nblocks, ip);
3527		goto corrupt_out;
3528	}
3529	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3530				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3531		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3532			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3533			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3534		goto corrupt_out;
3535	}
3536
3537	/*
3538	 * Inode item log recovery for v2 inodes are dependent on the
3539	 * di_flushiter count for correct sequencing. We bump the flush
3540	 * iteration count so we can detect flushes which postdate a log record
3541	 * during recovery. This is redundant as we now log every change and
3542	 * hence this can't happen but we need to still do it to ensure
3543	 * backwards compatibility with old kernels that predate logging all
3544	 * inode changes.
3545	 */
3546	if (ip->i_d.di_version < 3)
3547		ip->i_d.di_flushiter++;
3548
3549	/*
 
 
 
 
 
 
 
 
 
 
 
3550	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3551	 * copy out the core of the inode, because if the inode is dirty at all
3552	 * the core must be.
3553	 */
3554	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3555
3556	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3557	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3558		ip->i_d.di_flushiter = 0;
3559
3560	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3561	if (XFS_IFORK_Q(ip))
3562		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3563	xfs_inobp_check(mp, bp);
3564
3565	/*
3566	 * We've recorded everything logged in the inode, so we'd like to clear
3567	 * the ili_fields bits so we don't log and flush things unnecessarily.
3568	 * However, we can't stop logging all this information until the data
3569	 * we've copied into the disk buffer is written to disk.  If we did we
3570	 * might overwrite the copy of the inode in the log with all the data
3571	 * after re-logging only part of it, and in the face of a crash we
3572	 * wouldn't have all the data we need to recover.
3573	 *
3574	 * What we do is move the bits to the ili_last_fields field.  When
3575	 * logging the inode, these bits are moved back to the ili_fields field.
3576	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3577	 * know that the information those bits represent is permanently on
3578	 * disk.  As long as the flush completes before the inode is logged
3579	 * again, then both ili_fields and ili_last_fields will be cleared.
3580	 *
3581	 * We can play with the ili_fields bits here, because the inode lock
3582	 * must be held exclusively in order to set bits there and the flush
3583	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3584	 * done routine can tell whether or not to look in the AIL.  Also, store
3585	 * the current LSN of the inode so that we can tell whether the item has
3586	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3587	 * need the AIL lock, because it is a 64 bit value that cannot be read
3588	 * atomically.
3589	 */
 
 
 
3590	iip->ili_last_fields = iip->ili_fields;
3591	iip->ili_fields = 0;
3592	iip->ili_fsync_fields = 0;
3593	iip->ili_logged = 1;
3594
 
 
 
 
3595	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3596				&iip->ili_item.li_lsn);
3597
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3598	/*
3599	 * Attach the function xfs_iflush_done to the inode's
3600	 * buffer.  This will remove the inode from the AIL
3601	 * and unlock the inode's flush lock when the inode is
3602	 * completely written to disk.
3603	 */
3604	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3605
3606	/* generate the checksum. */
3607	xfs_dinode_calc_crc(mp, dip);
 
 
 
 
 
 
 
 
3608
3609	ASSERT(bp->b_fspriv != NULL);
3610	ASSERT(bp->b_iodone != NULL);
 
 
 
 
 
 
3611	return 0;
 
 
 
 
 
 
 
 
 
3612
3613corrupt_out:
3614	return -EFSCORRUPTED;
 
 
 
 
3615}