Linux Audio

Check our new training course

Linux BSP development engineering services

Need help to port Linux and bootloaders to your hardware?
Loading...
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}
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4 * All Rights Reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 */
   6#include <linux/iversion.h>
   7
   8#include "xfs.h"
   9#include "xfs_fs.h"
  10#include "xfs_shared.h"
  11#include "xfs_format.h"
  12#include "xfs_log_format.h"
  13#include "xfs_trans_resv.h"
 
  14#include "xfs_mount.h"
  15#include "xfs_defer.h"
  16#include "xfs_inode.h"
 
 
  17#include "xfs_dir2.h"
 
  18#include "xfs_attr.h"
  19#include "xfs_bit.h"
  20#include "xfs_trans_space.h"
  21#include "xfs_trans.h"
  22#include "xfs_buf_item.h"
  23#include "xfs_inode_item.h"
  24#include "xfs_iunlink_item.h"
  25#include "xfs_ialloc.h"
  26#include "xfs_bmap.h"
  27#include "xfs_bmap_util.h"
  28#include "xfs_errortag.h"
  29#include "xfs_error.h"
  30#include "xfs_quota.h"
  31#include "xfs_filestream.h"
 
  32#include "xfs_trace.h"
  33#include "xfs_icache.h"
  34#include "xfs_symlink.h"
  35#include "xfs_trans_priv.h"
  36#include "xfs_log.h"
  37#include "xfs_bmap_btree.h"
  38#include "xfs_reflink.h"
  39#include "xfs_ag.h"
  40#include "xfs_log_priv.h"
  41#include "xfs_health.h"
  42#include "xfs_pnfs.h"
  43#include "xfs_parent.h"
  44#include "xfs_xattr.h"
  45#include "xfs_inode_util.h"
  46#include "xfs_metafile.h"
  47
  48struct kmem_cache *xfs_inode_cache;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  49
  50/*
  51 * These two are wrapper routines around the xfs_ilock() routine used to
  52 * centralize some grungy code.  They are used in places that wish to lock the
  53 * inode solely for reading the extents.  The reason these places can't just
  54 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
  55 * bringing in of the extents from disk for a file in b-tree format.  If the
  56 * inode is in b-tree format, then we need to lock the inode exclusively until
  57 * the extents are read in.  Locking it exclusively all the time would limit
  58 * our parallelism unnecessarily, though.  What we do instead is check to see
  59 * if the extents have been read in yet, and only lock the inode exclusively
  60 * if they have not.
  61 *
  62 * The functions return a value which should be given to the corresponding
  63 * xfs_iunlock() call.
  64 */
  65uint
  66xfs_ilock_data_map_shared(
  67	struct xfs_inode	*ip)
  68{
  69	uint			lock_mode = XFS_ILOCK_SHARED;
  70
  71	if (xfs_need_iread_extents(&ip->i_df))
 
  72		lock_mode = XFS_ILOCK_EXCL;
  73	xfs_ilock(ip, lock_mode);
  74	return lock_mode;
  75}
  76
  77uint
  78xfs_ilock_attr_map_shared(
  79	struct xfs_inode	*ip)
  80{
  81	uint			lock_mode = XFS_ILOCK_SHARED;
  82
  83	if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
 
  84		lock_mode = XFS_ILOCK_EXCL;
  85	xfs_ilock(ip, lock_mode);
  86	return lock_mode;
  87}
  88
  89/*
  90 * You can't set both SHARED and EXCL for the same lock,
  91 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
  92 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
  93 * to set in lock_flags.
  94 */
  95static inline void
  96xfs_lock_flags_assert(
  97	uint		lock_flags)
  98{
  99	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
 100		(XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
 101	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
 102		(XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
 103	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
 104		(XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 105	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
 106	ASSERT(lock_flags != 0);
 107}
 108
 109/*
 110 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
 111 * multi-reader locks: invalidate_lock and the i_lock.  This routine allows
 112 * various combinations of the locks to be obtained.
 113 *
 114 * The 3 locks should always be ordered so that the IO lock is obtained first,
 115 * the mmap lock second and the ilock last in order to prevent deadlock.
 116 *
 117 * Basic locking order:
 118 *
 119 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
 120 *
 121 * mmap_lock locking order:
 122 *
 123 * i_rwsem -> page lock -> mmap_lock
 124 * mmap_lock -> invalidate_lock -> page_lock
 125 *
 126 * The difference in mmap_lock locking order mean that we cannot hold the
 127 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
 128 * can fault in pages during copy in/out (for buffered IO) or require the
 129 * mmap_lock in get_user_pages() to map the user pages into the kernel address
 130 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
 131 * fault because page faults already hold the mmap_lock.
 132 *
 133 * Hence to serialise fully against both syscall and mmap based IO, we need to
 134 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
 135 * both taken in places where we need to invalidate the page cache in a race
 136 * free manner (e.g. truncate, hole punch and other extent manipulation
 137 * functions).
 138 */
 139void
 140xfs_ilock(
 141	xfs_inode_t		*ip,
 142	uint			lock_flags)
 143{
 144	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
 145
 146	xfs_lock_flags_assert(lock_flags);
 
 
 
 
 
 
 
 
 
 
 
 147
 148	if (lock_flags & XFS_IOLOCK_EXCL) {
 149		down_write_nested(&VFS_I(ip)->i_rwsem,
 150				  XFS_IOLOCK_DEP(lock_flags));
 151	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 152		down_read_nested(&VFS_I(ip)->i_rwsem,
 153				 XFS_IOLOCK_DEP(lock_flags));
 154	}
 155
 156	if (lock_flags & XFS_MMAPLOCK_EXCL) {
 157		down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
 158				  XFS_MMAPLOCK_DEP(lock_flags));
 159	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
 160		down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
 161				 XFS_MMAPLOCK_DEP(lock_flags));
 162	}
 163
 164	if (lock_flags & XFS_ILOCK_EXCL)
 165		down_write_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 166	else if (lock_flags & XFS_ILOCK_SHARED)
 167		down_read_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
 168}
 169
 170/*
 171 * This is just like xfs_ilock(), except that the caller
 172 * is guaranteed not to sleep.  It returns 1 if it gets
 173 * the requested locks and 0 otherwise.  If the IO lock is
 174 * obtained but the inode lock cannot be, then the IO lock
 175 * is dropped before returning.
 176 *
 177 * ip -- the inode being locked
 178 * lock_flags -- this parameter indicates the inode's locks to be
 179 *       to be locked.  See the comment for xfs_ilock() for a list
 180 *	 of valid values.
 181 */
 182int
 183xfs_ilock_nowait(
 184	xfs_inode_t		*ip,
 185	uint			lock_flags)
 186{
 187	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
 188
 189	xfs_lock_flags_assert(lock_flags);
 
 
 
 
 
 
 
 
 
 
 
 190
 191	if (lock_flags & XFS_IOLOCK_EXCL) {
 192		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
 193			goto out;
 194	} else if (lock_flags & XFS_IOLOCK_SHARED) {
 195		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
 196			goto out;
 197	}
 198
 199	if (lock_flags & XFS_MMAPLOCK_EXCL) {
 200		if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
 201			goto out_undo_iolock;
 202	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
 203		if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
 204			goto out_undo_iolock;
 205	}
 206
 207	if (lock_flags & XFS_ILOCK_EXCL) {
 208		if (!down_write_trylock(&ip->i_lock))
 209			goto out_undo_mmaplock;
 210	} else if (lock_flags & XFS_ILOCK_SHARED) {
 211		if (!down_read_trylock(&ip->i_lock))
 212			goto out_undo_mmaplock;
 213	}
 214	return 1;
 215
 216out_undo_mmaplock:
 217	if (lock_flags & XFS_MMAPLOCK_EXCL)
 218		up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
 219	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 220		up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
 221out_undo_iolock:
 222	if (lock_flags & XFS_IOLOCK_EXCL)
 223		up_write(&VFS_I(ip)->i_rwsem);
 224	else if (lock_flags & XFS_IOLOCK_SHARED)
 225		up_read(&VFS_I(ip)->i_rwsem);
 226out:
 227	return 0;
 228}
 229
 230/*
 231 * xfs_iunlock() is used to drop the inode locks acquired with
 232 * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
 233 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
 234 * that we know which locks to drop.
 235 *
 236 * ip -- the inode being unlocked
 237 * lock_flags -- this parameter indicates the inode's locks to be
 238 *       to be unlocked.  See the comment for xfs_ilock() for a list
 239 *	 of valid values for this parameter.
 240 *
 241 */
 242void
 243xfs_iunlock(
 244	xfs_inode_t		*ip,
 245	uint			lock_flags)
 246{
 247	xfs_lock_flags_assert(lock_flags);
 
 
 
 
 
 
 
 
 
 
 
 
 248
 249	if (lock_flags & XFS_IOLOCK_EXCL)
 250		up_write(&VFS_I(ip)->i_rwsem);
 251	else if (lock_flags & XFS_IOLOCK_SHARED)
 252		up_read(&VFS_I(ip)->i_rwsem);
 253
 254	if (lock_flags & XFS_MMAPLOCK_EXCL)
 255		up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
 256	else if (lock_flags & XFS_MMAPLOCK_SHARED)
 257		up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
 258
 259	if (lock_flags & XFS_ILOCK_EXCL)
 260		up_write(&ip->i_lock);
 261	else if (lock_flags & XFS_ILOCK_SHARED)
 262		up_read(&ip->i_lock);
 263
 264	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
 265}
 266
 267/*
 268 * give up write locks.  the i/o lock cannot be held nested
 269 * if it is being demoted.
 270 */
 271void
 272xfs_ilock_demote(
 273	xfs_inode_t		*ip,
 274	uint			lock_flags)
 275{
 276	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
 277	ASSERT((lock_flags &
 278		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
 279
 280	if (lock_flags & XFS_ILOCK_EXCL)
 281		downgrade_write(&ip->i_lock);
 282	if (lock_flags & XFS_MMAPLOCK_EXCL)
 283		downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
 284	if (lock_flags & XFS_IOLOCK_EXCL)
 285		downgrade_write(&VFS_I(ip)->i_rwsem);
 286
 287	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
 288}
 289
 290void
 291xfs_assert_ilocked(
 292	struct xfs_inode	*ip,
 
 293	uint			lock_flags)
 294{
 295	/*
 296	 * Sometimes we assert the ILOCK is held exclusively, but we're in
 297	 * a workqueue, so lockdep doesn't know we're the owner.
 298	 */
 299	if (lock_flags & XFS_ILOCK_SHARED)
 300		rwsem_assert_held(&ip->i_lock);
 301	else if (lock_flags & XFS_ILOCK_EXCL)
 302		rwsem_assert_held_write_nolockdep(&ip->i_lock);
 303
 304	if (lock_flags & XFS_MMAPLOCK_SHARED)
 305		rwsem_assert_held(&VFS_I(ip)->i_mapping->invalidate_lock);
 306	else if (lock_flags & XFS_MMAPLOCK_EXCL)
 307		rwsem_assert_held_write(&VFS_I(ip)->i_mapping->invalidate_lock);
 308
 309	if (lock_flags & XFS_IOLOCK_SHARED)
 310		rwsem_assert_held(&VFS_I(ip)->i_rwsem);
 311	else if (lock_flags & XFS_IOLOCK_EXCL)
 312		rwsem_assert_held_write(&VFS_I(ip)->i_rwsem);
 
 
 
 313}
 
 
 
 
 
 
 
 
 
 314
 315/*
 316 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
 317 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
 318 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
 319 * errors and warnings.
 320 */
 321#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
 322static bool
 323xfs_lockdep_subclass_ok(
 324	int subclass)
 325{
 326	return subclass < MAX_LOCKDEP_SUBCLASSES;
 327}
 328#else
 329#define xfs_lockdep_subclass_ok(subclass)	(true)
 330#endif
 331
 332/*
 333 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
 334 * value. This can be called for any type of inode lock combination, including
 335 * parent locking. Care must be taken to ensure we don't overrun the subclass
 336 * storage fields in the class mask we build.
 337 */
 338static inline uint
 339xfs_lock_inumorder(
 340	uint	lock_mode,
 341	uint	subclass)
 342{
 343	uint	class = 0;
 344
 345	ASSERT(!(lock_mode & XFS_ILOCK_PARENT));
 
 346	ASSERT(xfs_lockdep_subclass_ok(subclass));
 347
 348	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
 349		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
 350		class += subclass << XFS_IOLOCK_SHIFT;
 351	}
 352
 353	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
 354		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
 355		class += subclass << XFS_MMAPLOCK_SHIFT;
 356	}
 357
 358	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
 359		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
 360		class += subclass << XFS_ILOCK_SHIFT;
 361	}
 362
 363	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
 364}
 365
 366/*
 367 * The following routine will lock n inodes in exclusive mode.  We assume the
 368 * caller calls us with the inodes in i_ino order.
 369 *
 370 * We need to detect deadlock where an inode that we lock is in the AIL and we
 371 * start waiting for another inode that is locked by a thread in a long running
 372 * transaction (such as truncate). This can result in deadlock since the long
 373 * running trans might need to wait for the inode we just locked in order to
 374 * push the tail and free space in the log.
 375 *
 376 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
 377 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
 378 * lock more than one at a time, lockdep will report false positives saying we
 379 * have violated locking orders.
 380 */
 381void
 382xfs_lock_inodes(
 383	struct xfs_inode	**ips,
 384	int			inodes,
 385	uint			lock_mode)
 386{
 387	int			attempts = 0;
 388	uint			i;
 389	int			j;
 390	bool			try_lock;
 391	struct xfs_log_item	*lp;
 392
 393	/*
 394	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
 395	 * support an arbitrary depth of locking here, but absolute limits on
 396	 * inodes depend on the type of locking and the limits placed by
 397	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
 398	 * the asserts.
 399	 */
 400	ASSERT(ips && inodes >= 2 && inodes <= 5);
 401	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
 402			    XFS_ILOCK_EXCL));
 403	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
 404			      XFS_ILOCK_SHARED)));
 405	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
 406		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
 407	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
 408		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
 409
 410	if (lock_mode & XFS_IOLOCK_EXCL) {
 411		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
 412	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
 413		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
 414
 
 
 415again:
 416	try_lock = false;
 417	i = 0;
 418	for (; i < inodes; i++) {
 419		ASSERT(ips[i]);
 420
 421		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
 422			continue;
 423
 424		/*
 425		 * If try_lock is not set yet, make sure all locked inodes are
 426		 * not in the AIL.  If any are, set try_lock to be used later.
 427		 */
 428		if (!try_lock) {
 429			for (j = (i - 1); j >= 0 && !try_lock; j--) {
 430				lp = &ips[j]->i_itemp->ili_item;
 431				if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
 432					try_lock = true;
 433			}
 434		}
 435
 436		/*
 437		 * If any of the previous locks we have locked is in the AIL,
 438		 * we must TRY to get the second and subsequent locks. If
 439		 * we can't get any, we must release all we have
 440		 * and try again.
 441		 */
 442		if (!try_lock) {
 443			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
 444			continue;
 445		}
 446
 447		/* try_lock means we have an inode locked that is in the AIL. */
 448		ASSERT(i != 0);
 449		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
 450			continue;
 451
 452		/*
 453		 * Unlock all previous guys and try again.  xfs_iunlock will try
 454		 * to push the tail if the inode is in the AIL.
 455		 */
 456		attempts++;
 457		for (j = i - 1; j >= 0; j--) {
 458			/*
 459			 * Check to see if we've already unlocked this one.  Not
 460			 * the first one going back, and the inode ptr is the
 461			 * same.
 462			 */
 463			if (j != (i - 1) && ips[j] == ips[j + 1])
 464				continue;
 465
 466			xfs_iunlock(ips[j], lock_mode);
 467		}
 468
 469		if ((attempts % 5) == 0) {
 470			delay(1); /* Don't just spin the CPU */
 
 
 
 471		}
 
 
 472		goto again;
 473	}
 
 
 
 
 
 
 
 
 
 
 474}
 475
 476/*
 477 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
 478 * mmaplock must be double-locked separately since we use i_rwsem and
 479 * invalidate_lock for that. We now support taking one lock EXCL and the
 480 * other SHARED.
 481 */
 482void
 483xfs_lock_two_inodes(
 484	struct xfs_inode	*ip0,
 485	uint			ip0_mode,
 486	struct xfs_inode	*ip1,
 487	uint			ip1_mode)
 488{
 
 489	int			attempts = 0;
 490	struct xfs_log_item	*lp;
 
 
 
 
 491
 492	ASSERT(hweight32(ip0_mode) == 1);
 493	ASSERT(hweight32(ip1_mode) == 1);
 494	ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
 495	ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
 496	ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
 497	ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
 498	ASSERT(ip0->i_ino != ip1->i_ino);
 499
 500	if (ip0->i_ino > ip1->i_ino) {
 501		swap(ip0, ip1);
 502		swap(ip0_mode, ip1_mode);
 
 503	}
 504
 505 again:
 506	xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
 507
 508	/*
 509	 * If the first lock we have locked is in the AIL, we must TRY to get
 510	 * the second lock. If we can't get it, we must release the first one
 511	 * and try again.
 512	 */
 513	lp = &ip0->i_itemp->ili_item;
 514	if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
 515		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
 516			xfs_iunlock(ip0, ip0_mode);
 517			if ((++attempts % 5) == 0)
 518				delay(1); /* Don't just spin the CPU */
 519			goto again;
 520		}
 521	} else {
 522		xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
 523	}
 524}
 525
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 526/*
 527 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
 528 * is allowed, otherwise it has to be an exact match. If a CI match is found,
 529 * ci_name->name will point to a the actual name (caller must free) or
 530 * will be set to NULL if an exact match is found.
 531 */
 532int
 533xfs_lookup(
 534	struct xfs_inode	*dp,
 535	const struct xfs_name	*name,
 536	struct xfs_inode	**ipp,
 537	struct xfs_name		*ci_name)
 538{
 539	xfs_ino_t		inum;
 540	int			error;
 541
 542	trace_xfs_lookup(dp, name);
 543
 544	if (xfs_is_shutdown(dp->i_mount))
 545		return -EIO;
 546	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
 547		return -EIO;
 548
 549	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
 550	if (error)
 551		goto out_unlock;
 552
 553	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
 554	if (error)
 555		goto out_free_name;
 556
 557	/*
 558	 * Fail if a directory entry in the regular directory tree points to
 559	 * a metadata file.
 560	 */
 561	if (XFS_IS_CORRUPT(dp->i_mount, xfs_is_metadir_inode(*ipp))) {
 562		xfs_fs_mark_sick(dp->i_mount, XFS_SICK_FS_METADIR);
 563		error = -EFSCORRUPTED;
 564		goto out_irele;
 565	}
 566
 567	return 0;
 568
 569out_irele:
 570	xfs_irele(*ipp);
 571out_free_name:
 572	if (ci_name)
 573		kfree(ci_name->name);
 574out_unlock:
 575	*ipp = NULL;
 576	return error;
 577}
 578
 579/*
 580 * Initialise a newly allocated inode and return the in-core inode to the
 581 * caller locked exclusively.
 582 *
 583 * Caller is responsible for unlocking the inode manually upon return
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 584 */
 585int
 586xfs_icreate(
 587	struct xfs_trans	*tp,
 588	xfs_ino_t		ino,
 589	const struct xfs_icreate_args *args,
 590	struct xfs_inode	**ipp)
 591{
 592	struct xfs_mount	*mp = tp->t_mountp;
 593	struct xfs_inode	*ip = NULL;
 594	int			error;
 
 
 
 
 
 
 
 
 
 595
 596	/*
 597	 * Get the in-core inode with the lock held exclusively to prevent
 598	 * others from looking at until we're done.
 599	 */
 600	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
 
 601	if (error)
 602		return error;
 
 
 
 
 
 603
 
 
 
 
 
 
 
 
 
 604	ASSERT(ip != NULL);
 605	xfs_trans_ijoin(tp, ip, 0);
 606	xfs_inode_init(tp, args, ip);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 607
 608	/* now that we have an i_mode we can setup the inode structure */
 609	xfs_setup_inode(ip);
 610
 611	*ipp = ip;
 612	return 0;
 613}
 614
 615/* Return dquots for the ids that will be assigned to a new file. */
 
 
 
 
 
 
 
 
 
 616int
 617xfs_icreate_dqalloc(
 618	const struct xfs_icreate_args	*args,
 619	struct xfs_dquot		**udqpp,
 620	struct xfs_dquot		**gdqpp,
 621	struct xfs_dquot		**pdqpp)
 622{
 623	struct inode			*dir = VFS_I(args->pip);
 624	kuid_t				uid = GLOBAL_ROOT_UID;
 625	kgid_t				gid = GLOBAL_ROOT_GID;
 626	prid_t				prid = 0;
 627	unsigned int			flags = XFS_QMOPT_QUOTALL;
 628
 629	if (args->idmap) {
 630		/*
 631		 * The uid/gid computation code must match what the VFS uses to
 632		 * assign i_[ug]id.  INHERIT adjusts the gid computation for
 633		 * setgid/grpid systems.
 634		 */
 635		uid = mapped_fsuid(args->idmap, i_user_ns(dir));
 636		gid = mapped_fsgid(args->idmap, i_user_ns(dir));
 637		prid = xfs_get_initial_prid(args->pip);
 638		flags |= XFS_QMOPT_INHERIT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 639	}
 640
 641	*udqpp = *gdqpp = *pdqpp = NULL;
 
 642
 643	return xfs_qm_vop_dqalloc(args->pip, uid, gid, prid, flags, udqpp,
 644			gdqpp, pdqpp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 645}
 646
 647int
 648xfs_create(
 649	const struct xfs_icreate_args *args,
 650	struct xfs_name		*name,
 651	struct xfs_inode	**ipp)
 
 
 652{
 653	struct xfs_inode	*dp = args->pip;
 654	struct xfs_dir_update	du = {
 655		.dp		= dp,
 656		.name		= name,
 657	};
 658	struct xfs_mount	*mp = dp->i_mount;
 
 659	struct xfs_trans	*tp = NULL;
 660	struct xfs_dquot	*udqp;
 661	struct xfs_dquot	*gdqp;
 662	struct xfs_dquot	*pdqp;
 
 
 
 
 
 663	struct xfs_trans_res	*tres;
 664	xfs_ino_t		ino;
 665	bool			unlock_dp_on_error = false;
 666	bool			is_dir = S_ISDIR(args->mode);
 667	uint			resblks;
 668	int			error;
 669
 670	trace_xfs_create(dp, name);
 671
 672	if (xfs_is_shutdown(mp))
 673		return -EIO;
 674	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
 675		return -EIO;
 676
 677	/* Make sure that we have allocated dquot(s) on disk. */
 678	error = xfs_icreate_dqalloc(args, &udqp, &gdqp, &pdqp);
 
 
 
 
 
 
 
 679	if (error)
 680		return error;
 681
 682	if (is_dir) {
 683		resblks = xfs_mkdir_space_res(mp, name->len);
 
 684		tres = &M_RES(mp)->tr_mkdir;
 685	} else {
 686		resblks = xfs_create_space_res(mp, name->len);
 687		tres = &M_RES(mp)->tr_create;
 688	}
 689
 690	error = xfs_parent_start(mp, &du.ppargs);
 691	if (error)
 692		goto out_release_dquots;
 693
 694	/*
 695	 * Initially assume that the file does not exist and
 696	 * reserve the resources for that case.  If that is not
 697	 * the case we'll drop the one we have and get a more
 698	 * appropriate transaction later.
 699	 */
 700	error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
 701			&tp);
 702	if (error == -ENOSPC) {
 703		/* flush outstanding delalloc blocks and retry */
 704		xfs_flush_inodes(mp);
 705		error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
 706				resblks, &tp);
 
 
 
 
 707	}
 708	if (error)
 709		goto out_parent;
 710
 711	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
 712	unlock_dp_on_error = true;
 713
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 714	/*
 715	 * A newly created regular or special file just has one directory
 716	 * entry pointing to them, but a directory also the "." entry
 717	 * pointing to itself.
 718	 */
 719	error = xfs_dialloc(&tp, args, &ino);
 720	if (!error)
 721		error = xfs_icreate(tp, ino, args, &du.ip);
 722	if (error)
 723		goto out_trans_cancel;
 724
 725	/*
 726	 * Now we join the directory inode to the transaction.  We do not do it
 727	 * earlier because xfs_dialloc might commit the previous transaction
 728	 * (and release all the locks).  An error from here on will result in
 729	 * the transaction cancel unlocking dp so don't do it explicitly in the
 730	 * error path.
 731	 */
 732	xfs_trans_ijoin(tp, dp, 0);
 
 733
 734	error = xfs_dir_create_child(tp, resblks, &du);
 735	if (error)
 
 
 
 736		goto out_trans_cancel;
 
 
 
 
 
 
 
 
 
 
 
 
 
 737
 738	/*
 739	 * If this is a synchronous mount, make sure that the
 740	 * create transaction goes to disk before returning to
 741	 * the user.
 742	 */
 743	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
 744		xfs_trans_set_sync(tp);
 745
 746	/*
 747	 * Attach the dquot(s) to the inodes and modify them incore.
 748	 * These ids of the inode couldn't have changed since the new
 749	 * inode has been locked ever since it was created.
 750	 */
 751	xfs_qm_vop_create_dqattach(tp, du.ip, udqp, gdqp, pdqp);
 
 
 
 
 752
 753	error = xfs_trans_commit(tp);
 754	if (error)
 755		goto out_release_inode;
 756
 757	xfs_qm_dqrele(udqp);
 758	xfs_qm_dqrele(gdqp);
 759	xfs_qm_dqrele(pdqp);
 760
 761	*ipp = du.ip;
 762	xfs_iunlock(du.ip, XFS_ILOCK_EXCL);
 763	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 764	xfs_parent_finish(mp, du.ppargs);
 765	return 0;
 766
 
 
 767 out_trans_cancel:
 768	xfs_trans_cancel(tp);
 769 out_release_inode:
 770	/*
 771	 * Wait until after the current transaction is aborted to finish the
 772	 * setup of the inode and release the inode.  This prevents recursive
 773	 * transactions and deadlocks from xfs_inactive.
 774	 */
 775	if (du.ip) {
 776		xfs_iunlock(du.ip, XFS_ILOCK_EXCL);
 777		xfs_finish_inode_setup(du.ip);
 778		xfs_irele(du.ip);
 779	}
 780 out_parent:
 781	xfs_parent_finish(mp, du.ppargs);
 782 out_release_dquots:
 783	xfs_qm_dqrele(udqp);
 784	xfs_qm_dqrele(gdqp);
 785	xfs_qm_dqrele(pdqp);
 786
 787	if (unlock_dp_on_error)
 788		xfs_iunlock(dp, XFS_ILOCK_EXCL);
 789	return error;
 790}
 791
 792int
 793xfs_create_tmpfile(
 794	const struct xfs_icreate_args *args,
 
 
 795	struct xfs_inode	**ipp)
 796{
 797	struct xfs_inode	*dp = args->pip;
 798	struct xfs_mount	*mp = dp->i_mount;
 799	struct xfs_inode	*ip = NULL;
 800	struct xfs_trans	*tp = NULL;
 801	struct xfs_dquot	*udqp;
 802	struct xfs_dquot	*gdqp;
 803	struct xfs_dquot	*pdqp;
 
 
 804	struct xfs_trans_res	*tres;
 805	xfs_ino_t		ino;
 806	uint			resblks;
 807	int			error;
 808
 809	ASSERT(args->flags & XFS_ICREATE_TMPFILE);
 
 810
 811	if (xfs_is_shutdown(mp))
 812		return -EIO;
 813
 814	/* Make sure that we have allocated dquot(s) on disk. */
 815	error = xfs_icreate_dqalloc(args, &udqp, &gdqp, &pdqp);
 
 
 
 
 
 816	if (error)
 817		return error;
 818
 819	resblks = XFS_IALLOC_SPACE_RES(mp);
 820	tres = &M_RES(mp)->tr_create_tmpfile;
 821
 822	error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
 823			&tp);
 
 
 
 
 824	if (error)
 825		goto out_release_dquots;
 826
 827	error = xfs_dialloc(&tp, args, &ino);
 828	if (!error)
 829		error = xfs_icreate(tp, ino, args, &ip);
 830	if (error)
 831		goto out_trans_cancel;
 832
 833	if (xfs_has_wsync(mp))
 
 
 
 
 
 834		xfs_trans_set_sync(tp);
 835
 836	/*
 837	 * Attach the dquot(s) to the inodes and modify them incore.
 838	 * These ids of the inode couldn't have changed since the new
 839	 * inode has been locked ever since it was created.
 840	 */
 841	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
 842
 843	error = xfs_iunlink(tp, ip);
 844	if (error)
 845		goto out_trans_cancel;
 846
 847	error = xfs_trans_commit(tp);
 848	if (error)
 849		goto out_release_inode;
 850
 851	xfs_qm_dqrele(udqp);
 852	xfs_qm_dqrele(gdqp);
 853	xfs_qm_dqrele(pdqp);
 854
 855	*ipp = ip;
 856	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 857	return 0;
 858
 859 out_trans_cancel:
 860	xfs_trans_cancel(tp);
 861 out_release_inode:
 862	/*
 863	 * Wait until after the current transaction is aborted to finish the
 864	 * setup of the inode and release the inode.  This prevents recursive
 865	 * transactions and deadlocks from xfs_inactive.
 866	 */
 867	if (ip) {
 868		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 869		xfs_finish_inode_setup(ip);
 870		xfs_irele(ip);
 871	}
 872 out_release_dquots:
 873	xfs_qm_dqrele(udqp);
 874	xfs_qm_dqrele(gdqp);
 875	xfs_qm_dqrele(pdqp);
 876
 877	return error;
 878}
 879
 880int
 881xfs_link(
 882	struct xfs_inode	*tdp,
 883	struct xfs_inode	*sip,
 884	struct xfs_name		*target_name)
 885{
 886	struct xfs_dir_update	du = {
 887		.dp		= tdp,
 888		.name		= target_name,
 889		.ip		= sip,
 890	};
 891	struct xfs_mount	*mp = tdp->i_mount;
 892	struct xfs_trans	*tp;
 893	int			error, nospace_error = 0;
 894	int			resblks;
 895
 896	trace_xfs_link(tdp, target_name);
 897
 898	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
 899
 900	if (xfs_is_shutdown(mp))
 901		return -EIO;
 902	if (xfs_ifork_zapped(tdp, XFS_DATA_FORK))
 903		return -EIO;
 904
 905	error = xfs_qm_dqattach(sip);
 906	if (error)
 907		goto std_return;
 908
 909	error = xfs_qm_dqattach(tdp);
 910	if (error)
 911		goto std_return;
 912
 913	error = xfs_parent_start(mp, &du.ppargs);
 
 
 
 
 
 914	if (error)
 915		goto std_return;
 916
 917	resblks = xfs_link_space_res(mp, target_name->len);
 918	error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
 919			&tp, &nospace_error);
 920	if (error)
 921		goto out_parent;
 922
 923	/*
 924	 * We don't allow reservationless or quotaless hardlinking when parent
 925	 * pointers are enabled because we can't back out if the xattrs must
 926	 * grow.
 927	 */
 928	if (du.ppargs && nospace_error) {
 929		error = nospace_error;
 
 930		goto error_return;
 931	}
 932
 
 
 
 
 
 
 
 
 933	/*
 934	 * If we are using project inheritance, we only allow hard link
 935	 * creation in our tree when the project IDs are the same; else
 936	 * the tree quota mechanism could be circumvented.
 937	 */
 938	if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 939		     tdp->i_projid != sip->i_projid)) {
 940		/*
 941		 * Project quota setup skips special files which can
 942		 * leave inodes in a PROJINHERIT directory without a
 943		 * project ID set. We need to allow links to be made
 944		 * to these "project-less" inodes because userspace
 945		 * expects them to succeed after project ID setup,
 946		 * but everything else should be rejected.
 947		 */
 948		if (!special_file(VFS_I(sip)->i_mode) ||
 949		    sip->i_projid != 0) {
 950			error = -EXDEV;
 951			goto error_return;
 952		}
 953	}
 954
 955	error = xfs_dir_add_child(tp, resblks, &du);
 
 
 
 
 
 
 
 956	if (error)
 957		goto error_return;
 958
 959	/*
 960	 * If this is a synchronous mount, make sure that the
 961	 * link transaction goes to disk before returning to
 962	 * the user.
 963	 */
 964	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
 965		xfs_trans_set_sync(tp);
 966
 967	error = xfs_trans_commit(tp);
 968	xfs_iunlock(tdp, XFS_ILOCK_EXCL);
 969	xfs_iunlock(sip, XFS_ILOCK_EXCL);
 970	xfs_parent_finish(mp, du.ppargs);
 971	return error;
 
 
 972
 973 error_return:
 974	xfs_trans_cancel(tp);
 975	xfs_iunlock(tdp, XFS_ILOCK_EXCL);
 976	xfs_iunlock(sip, XFS_ILOCK_EXCL);
 977 out_parent:
 978	xfs_parent_finish(mp, du.ppargs);
 979 std_return:
 980	if (error == -ENOSPC && nospace_error)
 981		error = nospace_error;
 982	return error;
 983}
 984
 985/* Clear the reflink flag and the cowblocks tag if possible. */
 986static void
 987xfs_itruncate_clear_reflink_flags(
 988	struct xfs_inode	*ip)
 989{
 990	struct xfs_ifork	*dfork;
 991	struct xfs_ifork	*cfork;
 992
 993	if (!xfs_is_reflink_inode(ip))
 994		return;
 995	dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
 996	cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
 997	if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
 998		ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
 999	if (cfork->if_bytes == 0)
1000		xfs_inode_clear_cowblocks_tag(ip);
1001}
1002
1003/*
1004 * Free up the underlying blocks past new_size.  The new size must be smaller
1005 * than the current size.  This routine can be used both for the attribute and
1006 * data fork, and does not modify the inode size, which is left to the caller.
1007 *
1008 * The transaction passed to this routine must have made a permanent log
1009 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1010 * given transaction and start new ones, so make sure everything involved in
1011 * the transaction is tidy before calling here.  Some transaction will be
1012 * returned to the caller to be committed.  The incoming transaction must
1013 * already include the inode, and both inode locks must be held exclusively.
1014 * The inode must also be "held" within the transaction.  On return the inode
1015 * will be "held" within the returned transaction.  This routine does NOT
1016 * require any disk space to be reserved for it within the transaction.
1017 *
1018 * If we get an error, we must return with the inode locked and linked into the
1019 * current transaction. This keeps things simple for the higher level code,
1020 * because it always knows that the inode is locked and held in the transaction
1021 * that returns to it whether errors occur or not.  We don't mark the inode
1022 * dirty on error so that transactions can be easily aborted if possible.
1023 */
1024int
1025xfs_itruncate_extents_flags(
1026	struct xfs_trans	**tpp,
1027	struct xfs_inode	*ip,
1028	int			whichfork,
1029	xfs_fsize_t		new_size,
1030	int			flags)
1031{
1032	struct xfs_mount	*mp = ip->i_mount;
1033	struct xfs_trans	*tp = *tpp;
 
 
1034	xfs_fileoff_t		first_unmap_block;
 
 
1035	int			error = 0;
 
1036
1037	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1038	if (atomic_read(&VFS_I(ip)->i_count))
1039		xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
1040	ASSERT(new_size <= XFS_ISIZE(ip));
1041	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1042	ASSERT(ip->i_itemp != NULL);
1043	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1044	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1045
1046	trace_xfs_itruncate_extents_start(ip, new_size);
1047
1048	flags |= xfs_bmapi_aflag(whichfork);
1049
1050	/*
1051	 * Since it is possible for space to become allocated beyond
1052	 * the end of the file (in a crash where the space is allocated
1053	 * but the inode size is not yet updated), simply remove any
1054	 * blocks which show up between the new EOF and the maximum
1055	 * possible file size.
1056	 *
1057	 * We have to free all the blocks to the bmbt maximum offset, even if
1058	 * the page cache can't scale that far.
1059	 */
1060	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1061	if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1062		WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1063		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1064	}
1065
1066	error = xfs_bunmapi_range(&tp, ip, flags, first_unmap_block,
1067			XFS_MAX_FILEOFF);
 
1068	if (error)
1069		goto out;
1070
1071	if (whichfork == XFS_DATA_FORK) {
1072		/* Remove all pending CoW reservations. */
1073		error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1074				first_unmap_block, XFS_MAX_FILEOFF, true);
1075		if (error)
1076			goto out;
1077
1078		xfs_itruncate_clear_reflink_flags(ip);
1079	}
1080
1081	/*
1082	 * Always re-log the inode so that our permanent transaction can keep
1083	 * on rolling it forward in the log.
1084	 */
1085	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1086
1087	trace_xfs_itruncate_extents_end(ip, new_size);
1088
1089out:
1090	*tpp = tp;
1091	return error;
 
 
 
 
 
 
 
 
1092}
1093
1094/*
1095 * Mark all the buffers attached to this directory stale.  In theory we should
1096 * never be freeing a directory with any blocks at all, but this covers the
1097 * case where we've recovered a directory swap with a "temporary" directory
1098 * created by online repair and now need to dump it.
1099 */
1100STATIC void
1101xfs_inactive_dir(
1102	struct xfs_inode	*dp)
1103{
1104	struct xfs_iext_cursor	icur;
1105	struct xfs_bmbt_irec	got;
1106	struct xfs_mount	*mp = dp->i_mount;
1107	struct xfs_da_geometry	*geo = mp->m_dir_geo;
1108	struct xfs_ifork	*ifp = xfs_ifork_ptr(dp, XFS_DATA_FORK);
1109	xfs_fileoff_t		off;
1110
1111	/*
1112	 * Invalidate each directory block.  All directory blocks are of
1113	 * fsbcount length and alignment, so we only need to walk those same
1114	 * offsets.  We hold the only reference to this inode, so we must wait
1115	 * for the buffer locks.
1116	 */
1117	for_each_xfs_iext(ifp, &icur, &got) {
1118		for (off = round_up(got.br_startoff, geo->fsbcount);
1119		     off < got.br_startoff + got.br_blockcount;
1120		     off += geo->fsbcount) {
1121			struct xfs_buf	*bp = NULL;
1122			xfs_fsblock_t	fsbno;
1123			int		error;
1124
1125			fsbno = (off - got.br_startoff) + got.br_startblock;
1126			error = xfs_buf_incore(mp->m_ddev_targp,
1127					XFS_FSB_TO_DADDR(mp, fsbno),
1128					XFS_FSB_TO_BB(mp, geo->fsbcount),
1129					XBF_LIVESCAN, &bp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1130			if (error)
1131				continue;
 
1132
1133			xfs_buf_stale(bp);
1134			xfs_buf_relse(bp);
1135		}
1136	}
 
1137}
1138
1139/*
1140 * xfs_inactive_truncate
1141 *
1142 * Called to perform a truncate when an inode becomes unlinked.
1143 */
1144STATIC int
1145xfs_inactive_truncate(
1146	struct xfs_inode *ip)
1147{
1148	struct xfs_mount	*mp = ip->i_mount;
1149	struct xfs_trans	*tp;
1150	int			error;
1151
1152	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1153	if (error) {
1154		ASSERT(xfs_is_shutdown(mp));
1155		return error;
1156	}
 
1157	xfs_ilock(ip, XFS_ILOCK_EXCL);
1158	xfs_trans_ijoin(tp, ip, 0);
1159
1160	/*
1161	 * Log the inode size first to prevent stale data exposure in the event
1162	 * of a system crash before the truncate completes. See the related
1163	 * comment in xfs_vn_setattr_size() for details.
1164	 */
1165	ip->i_disk_size = 0;
1166	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1167
1168	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1169	if (error)
1170		goto error_trans_cancel;
1171
1172	ASSERT(ip->i_df.if_nextents == 0);
1173
1174	error = xfs_trans_commit(tp);
1175	if (error)
1176		goto error_unlock;
1177
1178	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1179	return 0;
1180
1181error_trans_cancel:
1182	xfs_trans_cancel(tp);
1183error_unlock:
1184	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1185	return error;
1186}
1187
1188/*
1189 * xfs_inactive_ifree()
1190 *
1191 * Perform the inode free when an inode is unlinked.
1192 */
1193STATIC int
1194xfs_inactive_ifree(
1195	struct xfs_inode *ip)
1196{
 
 
1197	struct xfs_mount	*mp = ip->i_mount;
1198	struct xfs_trans	*tp;
1199	int			error;
1200
1201	/*
1202	 * We try to use a per-AG reservation for any block needed by the finobt
1203	 * tree, but as the finobt feature predates the per-AG reservation
1204	 * support a degraded file system might not have enough space for the
1205	 * reservation at mount time.  In that case try to dip into the reserved
1206	 * pool and pray.
1207	 *
1208	 * Send a warning if the reservation does happen to fail, as the inode
1209	 * now remains allocated and sits on the unlinked list until the fs is
1210	 * repaired.
1211	 */
1212	if (unlikely(mp->m_finobt_nores)) {
1213		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1214				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1215				&tp);
1216	} else {
1217		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1218	}
1219	if (error) {
1220		if (error == -ENOSPC) {
1221			xfs_warn_ratelimited(mp,
1222			"Failed to remove inode(s) from unlinked list. "
1223			"Please free space, unmount and run xfs_repair.");
1224		} else {
1225			ASSERT(xfs_is_shutdown(mp));
1226		}
1227		return error;
1228	}
1229
1230	/*
1231	 * We do not hold the inode locked across the entire rolling transaction
1232	 * here. We only need to hold it for the first transaction that
1233	 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1234	 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1235	 * here breaks the relationship between cluster buffer invalidation and
1236	 * stale inode invalidation on cluster buffer item journal commit
1237	 * completion, and can result in leaving dirty stale inodes hanging
1238	 * around in memory.
1239	 *
1240	 * We have no need for serialising this inode operation against other
1241	 * operations - we freed the inode and hence reallocation is required
1242	 * and that will serialise on reallocating the space the deferops need
1243	 * to free. Hence we can unlock the inode on the first commit of
1244	 * the transaction rather than roll it right through the deferops. This
1245	 * avoids relogging the XFS_ISTALE inode.
1246	 *
1247	 * We check that xfs_ifree() hasn't grown an internal transaction roll
1248	 * by asserting that the inode is still locked when it returns.
1249	 */
1250	xfs_ilock(ip, XFS_ILOCK_EXCL);
1251	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1252
1253	error = xfs_ifree(tp, ip);
1254	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1255	if (error) {
1256		/*
1257		 * If we fail to free the inode, shut down.  The cancel
1258		 * might do that, we need to make sure.  Otherwise the
1259		 * inode might be lost for a long time or forever.
1260		 */
1261		if (!xfs_is_shutdown(mp)) {
1262			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1263				__func__, error);
1264			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1265		}
1266		xfs_trans_cancel(tp);
 
1267		return error;
1268	}
1269
1270	/*
1271	 * Credit the quota account(s). The inode is gone.
1272	 */
1273	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1274
1275	return xfs_trans_commit(tp);
1276}
1277
1278/*
1279 * Returns true if we need to update the on-disk metadata before we can free
1280 * the memory used by this inode.  Updates include freeing post-eof
1281 * preallocations; freeing COW staging extents; and marking the inode free in
1282 * the inobt if it is on the unlinked list.
1283 */
1284bool
1285xfs_inode_needs_inactive(
1286	struct xfs_inode	*ip)
1287{
1288	struct xfs_mount	*mp = ip->i_mount;
1289	struct xfs_ifork	*cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1290
1291	/*
1292	 * If the inode is already free, then there can be nothing
1293	 * to clean up here.
1294	 */
1295	if (VFS_I(ip)->i_mode == 0)
1296		return false;
1297
1298	/*
1299	 * If this is a read-only mount, don't do this (would generate I/O)
1300	 * unless we're in log recovery and cleaning the iunlinked list.
1301	 */
1302	if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1303		return false;
1304
1305	/* If the log isn't running, push inodes straight to reclaim. */
1306	if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1307		return false;
1308
1309	/* Metadata inodes require explicit resource cleanup. */
1310	if (xfs_is_internal_inode(ip))
1311		return false;
1312
1313	/* Want to clean out the cow blocks if there are any. */
1314	if (cow_ifp && cow_ifp->if_bytes > 0)
1315		return true;
1316
1317	/* Unlinked files must be freed. */
1318	if (VFS_I(ip)->i_nlink == 0)
1319		return true;
1320
1321	/*
1322	 * This file isn't being freed, so check if there are post-eof blocks
1323	 * to free.
1324	 *
1325	 * Note: don't bother with iolock here since lockdep complains about
1326	 * acquiring it in reclaim context. We have the only reference to the
1327	 * inode at this point anyways.
1328	 */
1329	return xfs_can_free_eofblocks(ip);
1330}
1331
1332/*
1333 * Save health status somewhere, if we're dumping an inode with uncorrected
1334 * errors and online repair isn't running.
1335 */
1336static inline void
1337xfs_inactive_health(
1338	struct xfs_inode	*ip)
1339{
1340	struct xfs_mount	*mp = ip->i_mount;
1341	struct xfs_perag	*pag;
1342	unsigned int		sick;
1343	unsigned int		checked;
1344
1345	xfs_inode_measure_sickness(ip, &sick, &checked);
1346	if (!sick)
1347		return;
1348
1349	trace_xfs_inode_unfixed_corruption(ip, sick);
1350
1351	if (sick & XFS_SICK_INO_FORGET)
1352		return;
1353
1354	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1355	if (!pag) {
1356		/* There had better still be a perag structure! */
1357		ASSERT(0);
1358		return;
1359	}
 
 
 
 
1360
1361	xfs_ag_mark_sick(pag, XFS_SICK_AG_INODES);
1362	xfs_perag_put(pag);
1363}
1364
1365/*
1366 * xfs_inactive
1367 *
1368 * This is called when the vnode reference count for the vnode
1369 * goes to zero.  If the file has been unlinked, then it must
1370 * now be truncated.  Also, we clear all of the read-ahead state
1371 * kept for the inode here since the file is now closed.
1372 */
1373int
1374xfs_inactive(
1375	xfs_inode_t	*ip)
1376{
1377	struct xfs_mount	*mp;
1378	int			error = 0;
1379	int			truncate = 0;
1380
1381	/*
1382	 * If the inode is already free, then there can be nothing
1383	 * to clean up here.
1384	 */
1385	if (VFS_I(ip)->i_mode == 0) {
 
1386		ASSERT(ip->i_df.if_broot_bytes == 0);
1387		goto out;
1388	}
1389
1390	mp = ip->i_mount;
1391	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1392
1393	xfs_inactive_health(ip);
1394
1395	/*
1396	 * If this is a read-only mount, don't do this (would generate I/O)
1397	 * unless we're in log recovery and cleaning the iunlinked list.
1398	 */
1399	if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1400		goto out;
1401
1402	/* Metadata inodes require explicit resource cleanup. */
1403	if (xfs_is_internal_inode(ip))
1404		goto out;
1405
1406	/* Try to clean out the cow blocks if there are any. */
1407	if (xfs_inode_has_cow_data(ip)) {
1408		error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1409		if (error)
1410			goto out;
1411	}
1412
1413	if (VFS_I(ip)->i_nlink != 0) {
1414		/*
1415		 * Note: don't bother with iolock here since lockdep complains
1416		 * about acquiring it in reclaim context. We have the only
1417		 * reference to the inode at this point anyways.
1418		 */
1419		if (xfs_can_free_eofblocks(ip))
1420			error = xfs_free_eofblocks(ip);
 
 
 
1421
1422		goto out;
1423	}
1424
1425	if (S_ISREG(VFS_I(ip)->i_mode) &&
1426	    (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1427	     xfs_inode_has_filedata(ip)))
1428		truncate = 1;
1429
1430	if (xfs_iflags_test(ip, XFS_IQUOTAUNCHECKED)) {
1431		/*
1432		 * If this inode is being inactivated during a quotacheck and
1433		 * has not yet been scanned by quotacheck, we /must/ remove
1434		 * the dquots from the inode before inactivation changes the
1435		 * block and inode counts.  Most probably this is a result of
1436		 * reloading the incore iunlinked list to purge unrecovered
1437		 * unlinked inodes.
1438		 */
1439		xfs_qm_dqdetach(ip);
1440	} else {
1441		error = xfs_qm_dqattach(ip);
1442		if (error)
1443			goto out;
1444	}
1445
1446	if (S_ISDIR(VFS_I(ip)->i_mode) && ip->i_df.if_nextents > 0) {
1447		xfs_inactive_dir(ip);
1448		truncate = 1;
1449	}
1450
1451	if (S_ISLNK(VFS_I(ip)->i_mode))
1452		error = xfs_inactive_symlink(ip);
1453	else if (truncate)
1454		error = xfs_inactive_truncate(ip);
1455	if (error)
1456		goto out;
1457
1458	/*
1459	 * If there are attributes associated with the file then blow them away
1460	 * now.  The code calls a routine that recursively deconstructs the
1461	 * attribute fork. If also blows away the in-core attribute fork.
1462	 */
1463	if (xfs_inode_has_attr_fork(ip)) {
1464		error = xfs_attr_inactive(ip);
1465		if (error)
1466			goto out;
1467	}
1468
1469	ASSERT(ip->i_forkoff == 0);
 
 
1470
1471	/*
1472	 * Free the inode.
1473	 */
1474	error = xfs_inactive_ifree(ip);
 
 
1475
1476out:
1477	/*
1478	 * We're done making metadata updates for this inode, so we can release
1479	 * the attached dquots.
1480	 */
1481	xfs_qm_dqdetach(ip);
1482	return error;
1483}
1484
1485/*
1486 * Find an inode on the unlinked list. This does not take references to the
1487 * inode as we have existence guarantees by holding the AGI buffer lock and that
1488 * only unlinked, referenced inodes can be on the unlinked inode list.  If we
1489 * don't find the inode in cache, then let the caller handle the situation.
1490 */
1491struct xfs_inode *
1492xfs_iunlink_lookup(
1493	struct xfs_perag	*pag,
1494	xfs_agino_t		agino)
 
 
 
 
1495{
1496	struct xfs_inode	*ip;
 
 
 
 
 
 
 
 
 
 
1497
1498	rcu_read_lock();
1499	ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1500	if (!ip) {
1501		/* Caller can handle inode not being in memory. */
1502		rcu_read_unlock();
1503		return NULL;
1504	}
 
1505
1506	/*
1507	 * Inode in RCU freeing limbo should not happen.  Warn about this and
1508	 * let the caller handle the failure.
1509	 */
1510	if (WARN_ON_ONCE(!ip->i_ino)) {
1511		rcu_read_unlock();
1512		return NULL;
1513	}
1514	ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1515	rcu_read_unlock();
1516	return ip;
1517}
1518
1519/*
1520 * Load the inode @next_agino into the cache and set its prev_unlinked pointer
1521 * to @prev_agino.  Caller must hold the AGI to synchronize with other changes
1522 * to the unlinked list.
1523 */
1524int
1525xfs_iunlink_reload_next(
1526	struct xfs_trans	*tp,
1527	struct xfs_buf		*agibp,
1528	xfs_agino_t		prev_agino,
1529	xfs_agino_t		next_agino)
1530{
1531	struct xfs_perag	*pag = agibp->b_pag;
1532	struct xfs_mount	*mp = pag_mount(pag);
1533	struct xfs_inode	*next_ip = NULL;
1534	int			error;
1535
1536	ASSERT(next_agino != NULLAGINO);
 
 
 
1537
1538#ifdef DEBUG
1539	rcu_read_lock();
1540	next_ip = radix_tree_lookup(&pag->pag_ici_root, next_agino);
1541	ASSERT(next_ip == NULL);
1542	rcu_read_unlock();
1543#endif
1544
1545	xfs_info_ratelimited(mp,
1546 "Found unrecovered unlinked inode 0x%x in AG 0x%x.  Initiating recovery.",
1547			next_agino, pag_agno(pag));
 
 
1548
1549	/*
1550	 * Use an untrusted lookup just to be cautious in case the AGI has been
1551	 * corrupted and now points at a free inode.  That shouldn't happen,
1552	 * but we'd rather shut down now since we're already running in a weird
1553	 * situation.
1554	 */
1555	error = xfs_iget(mp, tp, xfs_agino_to_ino(pag, next_agino),
1556			XFS_IGET_UNTRUSTED, 0, &next_ip);
1557	if (error) {
1558		xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
1559		return error;
1560	}
1561
1562	/* If this is not an unlinked inode, something is very wrong. */
1563	if (VFS_I(next_ip)->i_nlink != 0) {
1564		xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
1565		error = -EFSCORRUPTED;
1566		goto rele;
1567	}
1568
1569	next_ip->i_prev_unlinked = prev_agino;
1570	trace_xfs_iunlink_reload_next(next_ip);
1571rele:
1572	ASSERT(!(VFS_I(next_ip)->i_state & I_DONTCACHE));
1573	if (xfs_is_quotacheck_running(mp) && next_ip)
1574		xfs_iflags_set(next_ip, XFS_IQUOTAUNCHECKED);
1575	xfs_irele(next_ip);
1576	return error;
1577}
1578
1579/*
1580 * Look up the inode number specified and if it is not already marked XFS_ISTALE
1581 * mark it stale. We should only find clean inodes in this lookup that aren't
1582 * already stale.
1583 */
1584static void
1585xfs_ifree_mark_inode_stale(
1586	struct xfs_perag	*pag,
1587	struct xfs_inode	*free_ip,
1588	xfs_ino_t		inum)
1589{
1590	struct xfs_mount	*mp = pag_mount(pag);
1591	struct xfs_inode_log_item *iip;
1592	struct xfs_inode	*ip;
 
 
 
 
 
 
 
 
 
 
1593
1594retry:
1595	rcu_read_lock();
1596	ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
1597
1598	/* Inode not in memory, nothing to do */
1599	if (!ip) {
1600		rcu_read_unlock();
1601		return;
1602	}
1603
1604	/*
1605	 * because this is an RCU protected lookup, we could find a recently
1606	 * freed or even reallocated inode during the lookup. We need to check
1607	 * under the i_flags_lock for a valid inode here. Skip it if it is not
1608	 * valid, the wrong inode or stale.
1609	 */
1610	spin_lock(&ip->i_flags_lock);
1611	if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
1612		goto out_iflags_unlock;
1613
1614	/*
1615	 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
1616	 * other inodes that we did not find in the list attached to the buffer
1617	 * and are not already marked stale. If we can't lock it, back off and
1618	 * retry.
1619	 */
1620	if (ip != free_ip) {
1621		if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1622			spin_unlock(&ip->i_flags_lock);
1623			rcu_read_unlock();
1624			delay(1);
1625			goto retry;
1626		}
1627	}
1628	ip->i_flags |= XFS_ISTALE;
1629
1630	/*
1631	 * If the inode is flushing, it is already attached to the buffer.  All
1632	 * we needed to do here is mark the inode stale so buffer IO completion
1633	 * will remove it from the AIL.
1634	 */
1635	iip = ip->i_itemp;
1636	if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
1637		ASSERT(!list_empty(&iip->ili_item.li_bio_list));
1638		ASSERT(iip->ili_last_fields);
1639		goto out_iunlock;
1640	}
1641
1642	/*
1643	 * Inodes not attached to the buffer can be released immediately.
1644	 * Everything else has to go through xfs_iflush_abort() on journal
1645	 * commit as the flock synchronises removal of the inode from the
1646	 * cluster buffer against inode reclaim.
1647	 */
1648	if (!iip || list_empty(&iip->ili_item.li_bio_list))
1649		goto out_iunlock;
 
 
 
1650
1651	__xfs_iflags_set(ip, XFS_IFLUSHING);
1652	spin_unlock(&ip->i_flags_lock);
1653	rcu_read_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1654
1655	/* we have a dirty inode in memory that has not yet been flushed. */
1656	spin_lock(&iip->ili_lock);
1657	iip->ili_last_fields = iip->ili_fields;
1658	iip->ili_fields = 0;
1659	iip->ili_fsync_fields = 0;
1660	spin_unlock(&iip->ili_lock);
1661	ASSERT(iip->ili_last_fields);
 
1662
1663	if (ip != free_ip)
1664		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1665	return;
 
 
1666
1667out_iunlock:
1668	if (ip != free_ip)
1669		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1670out_iflags_unlock:
1671	spin_unlock(&ip->i_flags_lock);
1672	rcu_read_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1673}
1674
1675/*
1676 * A big issue when freeing the inode cluster is that we _cannot_ skip any
1677 * inodes that are in memory - they all must be marked stale and attached to
1678 * the cluster buffer.
1679 */
1680static int
1681xfs_ifree_cluster(
1682	struct xfs_trans	*tp,
1683	struct xfs_perag	*pag,
1684	struct xfs_inode	*free_ip,
1685	struct xfs_icluster	*xic)
1686{
1687	struct xfs_mount	*mp = free_ip->i_mount;
1688	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
1689	struct xfs_buf		*bp;
1690	xfs_daddr_t		blkno;
1691	xfs_ino_t		inum = xic->first_ino;
1692	int			nbufs;
1693	int			i, j;
1694	int			ioffset;
1695	int			error;
 
 
 
 
 
 
1696
1697	nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
 
 
 
 
1698
1699	for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
1700		/*
1701		 * The allocation bitmap tells us which inodes of the chunk were
1702		 * physically allocated. Skip the cluster if an inode falls into
1703		 * a sparse region.
1704		 */
1705		ioffset = inum - xic->first_ino;
1706		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
1707			ASSERT(ioffset % igeo->inodes_per_cluster == 0);
1708			continue;
1709		}
1710
1711		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1712					 XFS_INO_TO_AGBNO(mp, inum));
1713
1714		/*
1715		 * We obtain and lock the backing buffer first in the process
1716		 * here to ensure dirty inodes attached to the buffer remain in
1717		 * the flushing state while we mark them stale.
1718		 *
1719		 * If we scan the in-memory inodes first, then buffer IO can
1720		 * complete before we get a lock on it, and hence we may fail
1721		 * to mark all the active inodes on the buffer stale.
1722		 */
1723		error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1724				mp->m_bsize * igeo->blocks_per_cluster,
1725				XBF_UNMAPPED, &bp);
1726		if (error)
1727			return error;
 
1728
1729		/*
1730		 * This buffer may not have been correctly initialised as we
1731		 * didn't read it from disk. That's not important because we are
1732		 * only using to mark the buffer as stale in the log, and to
1733		 * attach stale cached inodes on it.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1734		 *
1735		 * For the inode that triggered the cluster freeing, this
1736		 * attachment may occur in xfs_inode_item_precommit() after we
1737		 * have marked this buffer stale.  If this buffer was not in
1738		 * memory before xfs_ifree_cluster() started, it will not be
1739		 * marked XBF_DONE and this will cause problems later in
1740		 * xfs_inode_item_precommit() when we trip over a (stale, !done)
1741		 * buffer to attached to the transaction.
1742		 *
1743		 * Hence we have to mark the buffer as XFS_DONE here. This is
1744		 * safe because we are also marking the buffer as XBF_STALE and
1745		 * XFS_BLI_STALE. That means it will never be dispatched for
1746		 * IO and it won't be unlocked until the cluster freeing has
1747		 * been committed to the journal and the buffer unpinned. If it
1748		 * is written, we want to know about it, and we want it to
1749		 * fail. We can acheive this by adding a write verifier to the
1750		 * buffer.
1751		 */
1752		bp->b_flags |= XBF_DONE;
1753		bp->b_ops = &xfs_inode_buf_ops;
1754
1755		/*
1756		 * Now we need to set all the cached clean inodes as XFS_ISTALE,
1757		 * too. This requires lookups, and will skip inodes that we've
1758		 * already marked XFS_ISTALE.
1759		 */
1760		for (i = 0; i < igeo->inodes_per_cluster; i++)
1761			xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1762
1763		xfs_trans_stale_inode_buf(tp, bp);
1764		xfs_trans_binval(tp, bp);
1765	}
 
 
1766	return 0;
1767}
1768
1769/*
1770 * This is called to return an inode to the inode free list.  The inode should
1771 * already be truncated to 0 length and have no pages associated with it.  This
1772 * routine also assumes that the inode is already a part of the transaction.
1773 *
1774 * The on-disk copy of the inode will have been added to the list of unlinked
1775 * inodes in the AGI. We need to remove the inode from that list atomically with
1776 * respect to freeing it here.
 
1777 */
1778int
1779xfs_ifree(
1780	struct xfs_trans	*tp,
1781	struct xfs_inode	*ip)
 
1782{
1783	struct xfs_mount	*mp = ip->i_mount;
1784	struct xfs_perag	*pag;
1785	struct xfs_icluster	xic = { 0 };
1786	struct xfs_inode_log_item *iip = ip->i_itemp;
1787	int			error;
1788
1789	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1790	ASSERT(VFS_I(ip)->i_nlink == 0);
1791	ASSERT(ip->i_df.if_nextents == 0);
1792	ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
1793	ASSERT(ip->i_nblocks == 0);
 
1794
1795	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 
 
 
 
 
1796
1797	error = xfs_inode_uninit(tp, pag, ip, &xic);
1798	if (error)
1799		goto out;
1800
1801	if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
1802		xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
 
 
 
 
 
 
 
 
 
 
1803
1804	/* Don't attempt to replay owner changes for a deleted inode */
1805	spin_lock(&iip->ili_lock);
1806	iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
1807	spin_unlock(&iip->ili_lock);
1808
1809	if (xic.deleted)
1810		error = xfs_ifree_cluster(tp, pag, ip, &xic);
1811out:
1812	xfs_perag_put(pag);
1813	return error;
1814}
1815
1816/*
1817 * This is called to unpin an inode.  The caller must have the inode locked
1818 * in at least shared mode so that the buffer cannot be subsequently pinned
1819 * once someone is waiting for it to be unpinned.
1820 */
1821static void
1822xfs_iunpin(
1823	struct xfs_inode	*ip)
1824{
1825	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
1826
1827	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
1828
1829	/* Give the log a push to start the unpinning I/O */
1830	xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
1831
1832}
1833
1834static void
1835__xfs_iunpin_wait(
1836	struct xfs_inode	*ip)
1837{
1838	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
1839	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
1840
1841	xfs_iunpin(ip);
1842
1843	do {
1844		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1845		if (xfs_ipincount(ip))
1846			io_schedule();
1847	} while (xfs_ipincount(ip));
1848	finish_wait(wq, &wait.wq_entry);
1849}
1850
1851void
1852xfs_iunpin_wait(
1853	struct xfs_inode	*ip)
1854{
1855	if (xfs_ipincount(ip))
1856		__xfs_iunpin_wait(ip);
1857}
1858
1859/*
1860 * Removing an inode from the namespace involves removing the directory entry
1861 * and dropping the link count on the inode. Removing the directory entry can
1862 * result in locking an AGF (directory blocks were freed) and removing a link
1863 * count can result in placing the inode on an unlinked list which results in
1864 * locking an AGI.
1865 *
1866 * The big problem here is that we have an ordering constraint on AGF and AGI
1867 * locking - inode allocation locks the AGI, then can allocate a new extent for
1868 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
1869 * removes the inode from the unlinked list, requiring that we lock the AGI
1870 * first, and then freeing the inode can result in an inode chunk being freed
1871 * and hence freeing disk space requiring that we lock an AGF.
1872 *
1873 * Hence the ordering that is imposed by other parts of the code is AGI before
1874 * AGF. This means we cannot remove the directory entry before we drop the inode
1875 * reference count and put it on the unlinked list as this results in a lock
1876 * order of AGF then AGI, and this can deadlock against inode allocation and
1877 * freeing. Therefore we must drop the link counts before we remove the
1878 * directory entry.
1879 *
1880 * This is still safe from a transactional point of view - it is not until we
1881 * get to xfs_defer_finish() that we have the possibility of multiple
1882 * transactions in this operation. Hence as long as we remove the directory
1883 * entry and drop the link count in the first transaction of the remove
1884 * operation, there are no transactional constraints on the ordering here.
1885 */
1886int
1887xfs_remove(
1888	struct xfs_inode	*dp,
1889	struct xfs_name		*name,
1890	struct xfs_inode	*ip)
1891{
1892	struct xfs_dir_update	du = {
1893		.dp		= dp,
1894		.name		= name,
1895		.ip		= ip,
1896	};
1897	struct xfs_mount	*mp = dp->i_mount;
1898	struct xfs_trans	*tp = NULL;
1899	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
1900	int			dontcare;
1901	int                     error = 0;
 
 
1902	uint			resblks;
1903
1904	trace_xfs_remove(dp, name);
1905
1906	if (xfs_is_shutdown(mp))
1907		return -EIO;
1908	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
1909		return -EIO;
1910
1911	error = xfs_qm_dqattach(dp);
1912	if (error)
1913		goto std_return;
1914
1915	error = xfs_qm_dqattach(ip);
1916	if (error)
1917		goto std_return;
1918
1919	error = xfs_parent_start(mp, &du.ppargs);
1920	if (error)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1921		goto std_return;
 
 
 
 
 
 
1922
1923	/*
1924	 * We try to get the real space reservation first, allowing for
1925	 * directory btree deletion(s) implying possible bmap insert(s).  If we
1926	 * can't get the space reservation then we use 0 instead, and avoid the
1927	 * bmap btree insert(s) in the directory code by, if the bmap insert
1928	 * tries to happen, instead trimming the LAST block from the directory.
1929	 *
1930	 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
1931	 * the directory code can handle a reservationless update and we don't
1932	 * want to prevent a user from trying to free space by deleting things.
1933	 */
1934	resblks = xfs_remove_space_res(mp, name->len);
1935	error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
1936			&tp, &dontcare);
1937	if (error) {
1938		ASSERT(error != -ENOSPC);
1939		goto out_parent;
 
 
 
 
 
 
 
 
 
 
 
 
 
1940	}
 
1941
1942	error = xfs_dir_remove_child(tp, resblks, &du);
 
1943	if (error)
1944		goto out_trans_cancel;
1945
 
 
 
 
 
 
 
 
1946	/*
1947	 * If this is a synchronous mount, make sure that the
1948	 * remove transaction goes to disk before returning to
1949	 * the user.
1950	 */
1951	if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1952		xfs_trans_set_sync(tp);
1953
 
 
 
 
1954	error = xfs_trans_commit(tp);
1955	if (error)
1956		goto out_unlock;
1957
1958	if (is_dir && xfs_inode_is_filestream(ip))
1959		xfs_filestream_deassociate(ip);
1960
1961	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1962	xfs_iunlock(dp, XFS_ILOCK_EXCL);
1963	xfs_parent_finish(mp, du.ppargs);
1964	return 0;
1965
 
 
1966 out_trans_cancel:
1967	xfs_trans_cancel(tp);
1968 out_unlock:
1969	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1970	xfs_iunlock(dp, XFS_ILOCK_EXCL);
1971 out_parent:
1972	xfs_parent_finish(mp, du.ppargs);
1973 std_return:
1974	return error;
1975}
1976
1977static inline void
1978xfs_iunlock_rename(
1979	struct xfs_inode	**i_tab,
1980	int			num_inodes)
1981{
1982	int			i;
1983
1984	for (i = num_inodes - 1; i >= 0; i--) {
1985		/* Skip duplicate inodes if src and target dps are the same */
1986		if (!i_tab[i] || (i > 0 && i_tab[i] == i_tab[i - 1]))
1987			continue;
1988		xfs_iunlock(i_tab[i], XFS_ILOCK_EXCL);
1989	}
1990}
1991
1992/*
1993 * Enter all inodes for a rename transaction into a sorted array.
1994 */
1995#define __XFS_SORT_INODES	5
1996STATIC void
1997xfs_sort_for_rename(
1998	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
1999	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2000	struct xfs_inode	*ip1,	/* in: inode of old entry */
2001	struct xfs_inode	*ip2,	/* in: inode of new entry */
2002	struct xfs_inode	*wip,	/* in: whiteout inode */
2003	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2004	int			*num_inodes)  /* in/out: inodes in array */
2005{
2006	int			i;
2007
2008	ASSERT(*num_inodes == __XFS_SORT_INODES);
2009	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2010
2011	/*
2012	 * i_tab contains a list of pointers to inodes.  We initialize
2013	 * the table here & we'll sort it.  We will then use it to
2014	 * order the acquisition of the inode locks.
2015	 *
2016	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2017	 */
2018	i = 0;
2019	i_tab[i++] = dp1;
2020	i_tab[i++] = dp2;
2021	i_tab[i++] = ip1;
2022	if (ip2)
2023		i_tab[i++] = ip2;
2024	if (wip)
2025		i_tab[i++] = wip;
2026	*num_inodes = i;
2027
2028	xfs_sort_inodes(i_tab, *num_inodes);
 
 
 
 
 
 
 
 
 
 
 
 
2029}
2030
2031void
2032xfs_sort_inodes(
2033	struct xfs_inode	**i_tab,
2034	unsigned int		num_inodes)
2035{
2036	int			i, j;
2037
2038	ASSERT(num_inodes <= __XFS_SORT_INODES);
2039
2040	/*
2041	 * Sort the elements via bubble sort.  (Remember, there are at
2042	 * most 5 elements to sort, so this is adequate.)
2043	 */
2044	for (i = 0; i < num_inodes; i++) {
2045		for (j = 1; j < num_inodes; j++) {
2046			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino)
2047				swap(i_tab[j], i_tab[j - 1]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2048		}
2049	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2050}
2051
2052/*
2053 * xfs_rename_alloc_whiteout()
2054 *
2055 * Return a referenced, unlinked, unlocked inode that can be used as a
2056 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2057 * crash between allocating the inode and linking it into the rename transaction
2058 * recovery will free the inode and we won't leak it.
2059 */
2060static int
2061xfs_rename_alloc_whiteout(
2062	struct mnt_idmap	*idmap,
2063	struct xfs_name		*src_name,
2064	struct xfs_inode	*dp,
2065	struct xfs_inode	**wip)
2066{
2067	struct xfs_icreate_args	args = {
2068		.idmap		= idmap,
2069		.pip		= dp,
2070		.mode		= S_IFCHR | WHITEOUT_MODE,
2071		.flags		= XFS_ICREATE_TMPFILE,
2072	};
2073	struct xfs_inode	*tmpfile;
2074	struct qstr		name;
2075	int			error;
2076
2077	error = xfs_create_tmpfile(&args, &tmpfile);
2078	if (error)
2079		return error;
2080
2081	name.name = src_name->name;
2082	name.len = src_name->len;
2083	error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2084	if (error) {
2085		xfs_finish_inode_setup(tmpfile);
2086		xfs_irele(tmpfile);
2087		return error;
2088	}
2089
2090	/*
2091	 * Prepare the tmpfile inode as if it were created through the VFS.
2092	 * Complete the inode setup and flag it as linkable.  nlink is already
2093	 * zero, so we can skip the drop_nlink.
 
2094	 */
 
2095	xfs_setup_iops(tmpfile);
2096	xfs_finish_inode_setup(tmpfile);
2097	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2098
2099	*wip = tmpfile;
2100	return 0;
2101}
2102
2103/*
2104 * xfs_rename
2105 */
2106int
2107xfs_rename(
2108	struct mnt_idmap	*idmap,
2109	struct xfs_inode	*src_dp,
2110	struct xfs_name		*src_name,
2111	struct xfs_inode	*src_ip,
2112	struct xfs_inode	*target_dp,
2113	struct xfs_name		*target_name,
2114	struct xfs_inode	*target_ip,
2115	unsigned int		flags)
2116{
2117	struct xfs_dir_update	du_src = {
2118		.dp		= src_dp,
2119		.name		= src_name,
2120		.ip		= src_ip,
2121	};
2122	struct xfs_dir_update	du_tgt = {
2123		.dp		= target_dp,
2124		.name		= target_name,
2125		.ip		= target_ip,
2126	};
2127	struct xfs_dir_update	du_wip = { };
2128	struct xfs_mount	*mp = src_dp->i_mount;
2129	struct xfs_trans	*tp;
 
 
 
2130	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2131	int			i;
2132	int			num_inodes = __XFS_SORT_INODES;
2133	bool			new_parent = (src_dp != target_dp);
2134	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2135	int			spaceres;
2136	bool			retried = false;
2137	int			error, nospace_error = 0;
2138
2139	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2140
2141	if ((flags & RENAME_EXCHANGE) && !target_ip)
2142		return -EINVAL;
2143
2144	/*
2145	 * If we are doing a whiteout operation, allocate the whiteout inode
2146	 * we will be placing at the target and ensure the type is set
2147	 * appropriately.
2148	 */
2149	if (flags & RENAME_WHITEOUT) {
2150		error = xfs_rename_alloc_whiteout(idmap, src_name, target_dp,
2151				&du_wip.ip);
2152		if (error)
2153			return error;
2154
2155		/* setup target dirent info as whiteout */
2156		src_name->type = XFS_DIR3_FT_CHRDEV;
2157	}
2158
2159	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, du_wip.ip,
2160			inodes, &num_inodes);
2161
2162	error = xfs_parent_start(mp, &du_src.ppargs);
2163	if (error)
2164		goto out_release_wip;
2165
2166	if (du_wip.ip) {
2167		error = xfs_parent_start(mp, &du_wip.ppargs);
2168		if (error)
2169			goto out_src_ppargs;
2170	}
2171
2172	if (target_ip) {
2173		error = xfs_parent_start(mp, &du_tgt.ppargs);
2174		if (error)
2175			goto out_wip_ppargs;
2176	}
2177
2178retry:
2179	nospace_error = 0;
2180	spaceres = xfs_rename_space_res(mp, src_name->len, target_ip != NULL,
2181			target_name->len, du_wip.ip != NULL);
2182	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2183	if (error == -ENOSPC) {
2184		nospace_error = error;
2185		spaceres = 0;
2186		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2187				&tp);
2188	}
2189	if (error)
2190		goto out_tgt_ppargs;
2191
2192	/*
2193	 * We don't allow reservationless renaming when parent pointers are
2194	 * enabled because we can't back out if the xattrs must grow.
2195	 */
2196	if (du_src.ppargs && nospace_error) {
2197		error = nospace_error;
2198		xfs_trans_cancel(tp);
2199		goto out_tgt_ppargs;
2200	}
2201
2202	/*
2203	 * Attach the dquots to the inodes
2204	 */
2205	error = xfs_qm_vop_rename_dqattach(inodes);
2206	if (error) {
2207		xfs_trans_cancel(tp);
2208		goto out_tgt_ppargs;
2209	}
2210
2211	/*
2212	 * Lock all the participating inodes. Depending upon whether
2213	 * the target_name exists in the target directory, and
2214	 * whether the target directory is the same as the source
2215	 * directory, we can lock from 2 to 5 inodes.
2216	 */
2217	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2218
2219	/*
2220	 * Join all the inodes to the transaction.
 
 
2221	 */
2222	xfs_trans_ijoin(tp, src_dp, 0);
2223	if (new_parent)
2224		xfs_trans_ijoin(tp, target_dp, 0);
2225	xfs_trans_ijoin(tp, src_ip, 0);
2226	if (target_ip)
2227		xfs_trans_ijoin(tp, target_ip, 0);
2228	if (du_wip.ip)
2229		xfs_trans_ijoin(tp, du_wip.ip, 0);
2230
2231	/*
2232	 * If we are using project inheritance, we only allow renames
2233	 * into our tree when the project IDs are the same; else the
2234	 * tree quota mechanism would be circumvented.
2235	 */
2236	if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2237		     target_dp->i_projid != src_ip->i_projid)) {
2238		error = -EXDEV;
2239		goto out_trans_cancel;
2240	}
2241
 
 
2242	/* RENAME_EXCHANGE is unique from here on. */
2243	if (flags & RENAME_EXCHANGE) {
2244		error = xfs_dir_exchange_children(tp, &du_src, &du_tgt,
2245				spaceres);
2246		if (error)
2247			goto out_trans_cancel;
2248		goto out_commit;
2249	}
2250
2251	/*
2252	 * Try to reserve quota to handle an expansion of the target directory.
2253	 * We'll allow the rename to continue in reservationless mode if we hit
2254	 * a space usage constraint.  If we trigger reservationless mode, save
2255	 * the errno if there isn't any free space in the target directory.
2256	 */
2257	if (spaceres != 0) {
2258		error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2259				0, false);
2260		if (error == -EDQUOT || error == -ENOSPC) {
2261			if (!retried) {
2262				xfs_trans_cancel(tp);
2263				xfs_iunlock_rename(inodes, num_inodes);
2264				xfs_blockgc_free_quota(target_dp, 0);
2265				retried = true;
2266				goto retry;
2267			}
2268
2269			nospace_error = error;
2270			spaceres = 0;
2271			error = 0;
2272		}
 
 
 
 
 
 
 
 
2273		if (error)
2274			goto out_trans_cancel;
2275	}
2276
2277	/*
2278	 * We don't allow quotaless renaming when parent pointers are enabled
2279	 * because we can't back out if the xattrs must grow.
2280	 */
2281	if (du_src.ppargs && nospace_error) {
2282		error = nospace_error;
2283		goto out_trans_cancel;
2284	}
2285
2286	/*
2287	 * Lock the AGI buffers we need to handle bumping the nlink of the
2288	 * whiteout inode off the unlinked list and to handle dropping the
2289	 * nlink of the target inode.  Per locking order rules, do this in
2290	 * increasing AG order and before directory block allocation tries to
2291	 * grab AGFs because we grab AGIs before AGFs.
2292	 *
2293	 * The (vfs) caller must ensure that if src is a directory then
2294	 * target_ip is either null or an empty directory.
2295	 */
2296	for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
2297		if (inodes[i] == du_wip.ip ||
2298		    (inodes[i] == target_ip &&
2299		     (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
2300			struct xfs_perag	*pag;
2301			struct xfs_buf		*bp;
2302
2303			pag = xfs_perag_get(mp,
2304					XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
2305			error = xfs_read_agi(pag, tp, 0, &bp);
2306			xfs_perag_put(pag);
2307			if (error)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2308				goto out_trans_cancel;
 
2309		}
2310	}
2311
2312	error = xfs_dir_rename_children(tp, &du_src, &du_tgt, spaceres,
2313			&du_wip);
2314	if (error)
2315		goto out_trans_cancel;
2316
2317	if (du_wip.ip) {
2318		/*
2319		 * Now we have a real link, clear the "I'm a tmpfile" state
2320		 * flag from the inode so it doesn't accidentally get misused in
2321		 * future.
 
 
 
 
2322		 */
2323		VFS_I(du_wip.ip)->i_state &= ~I_LINKABLE;
2324	}
 
 
 
2325
2326out_commit:
2327	/*
2328	 * If this is a synchronous mount, make sure that the rename
2329	 * transaction goes to disk before returning to the user.
2330	 */
2331	if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2332		xfs_trans_set_sync(tp);
2333
2334	error = xfs_trans_commit(tp);
2335	nospace_error = 0;
2336	goto out_unlock;
 
 
 
 
2337
2338out_trans_cancel:
2339	xfs_trans_cancel(tp);
2340out_unlock:
2341	xfs_iunlock_rename(inodes, num_inodes);
2342out_tgt_ppargs:
2343	xfs_parent_finish(mp, du_tgt.ppargs);
2344out_wip_ppargs:
2345	xfs_parent_finish(mp, du_wip.ppargs);
2346out_src_ppargs:
2347	xfs_parent_finish(mp, du_src.ppargs);
2348out_release_wip:
2349	if (du_wip.ip)
2350		xfs_irele(du_wip.ip);
2351	if (error == -ENOSPC && nospace_error)
2352		error = nospace_error;
2353	return error;
2354}
2355
2356static int
2357xfs_iflush(
2358	struct xfs_inode	*ip,
2359	struct xfs_buf		*bp)
2360{
2361	struct xfs_inode_log_item *iip = ip->i_itemp;
2362	struct xfs_dinode	*dip;
2363	struct xfs_mount	*mp = ip->i_mount;
2364	int			error;
2365
2366	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
2367	ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
2368	ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
2369	       ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2370	ASSERT(iip->ili_item.li_buf == bp);
2371
2372	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
2373
2374	/*
2375	 * We don't flush the inode if any of the following checks fail, but we
2376	 * do still update the log item and attach to the backing buffer as if
2377	 * the flush happened. This is a formality to facilitate predictable
2378	 * error handling as the caller will shutdown and fail the buffer.
2379	 */
2380	error = -EFSCORRUPTED;
2381	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2382			       mp, XFS_ERRTAG_IFLUSH_1)) {
2383		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2384			"%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
2385			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2386		goto flush_out;
2387	}
2388	if (S_ISREG(VFS_I(ip)->i_mode)) {
2389		if (XFS_TEST_ERROR(
2390		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
2391		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
2392		    mp, XFS_ERRTAG_IFLUSH_3)) {
2393			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2394				"%s: Bad regular inode %llu, ptr "PTR_FMT,
2395				__func__, ip->i_ino, ip);
2396			goto flush_out;
2397		}
2398	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
2399		if (XFS_TEST_ERROR(
2400		    ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
2401		    ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
2402		    ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
2403		    mp, XFS_ERRTAG_IFLUSH_4)) {
2404			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2405				"%s: Bad directory inode %llu, ptr "PTR_FMT,
2406				__func__, ip->i_ino, ip);
2407			goto flush_out;
2408		}
2409	}
2410	if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
2411				ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
2412		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2413			"%s: detected corrupt incore inode %llu, "
2414			"total extents = %llu nblocks = %lld, ptr "PTR_FMT,
2415			__func__, ip->i_ino,
2416			ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
2417			ip->i_nblocks, ip);
2418		goto flush_out;
2419	}
2420	if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
2421				mp, XFS_ERRTAG_IFLUSH_6)) {
2422		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2423			"%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
2424			__func__, ip->i_ino, ip->i_forkoff, ip);
2425		goto flush_out;
2426	}
2427
2428	/*
2429	 * Inode item log recovery for v2 inodes are dependent on the flushiter
2430	 * count for correct sequencing.  We bump the flush iteration count so
2431	 * we can detect flushes which postdate a log record during recovery.
2432	 * This is redundant as we now log every change and hence this can't
2433	 * happen but we need to still do it to ensure backwards compatibility
2434	 * with old kernels that predate logging all inode changes.
2435	 */
2436	if (!xfs_has_v3inodes(mp))
2437		ip->i_flushiter++;
2438
2439	/*
2440	 * If there are inline format data / attr forks attached to this inode,
2441	 * make sure they are not corrupt.
 
2442	 */
2443	if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
2444	    xfs_ifork_verify_local_data(ip))
2445		goto flush_out;
2446	if (xfs_inode_has_attr_fork(ip) &&
2447	    ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
2448	    xfs_ifork_verify_local_attr(ip))
2449		goto flush_out;
 
 
 
2450
2451	/*
2452	 * Copy the dirty parts of the inode into the on-disk inode.  We always
2453	 * copy out the core of the inode, because if the inode is dirty at all
2454	 * the core must be.
2455	 */
2456	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2457
2458	/* Wrap, we never let the log put out DI_MAX_FLUSH */
2459	if (!xfs_has_v3inodes(mp)) {
2460		if (ip->i_flushiter == DI_MAX_FLUSH)
2461			ip->i_flushiter = 0;
 
 
2462	}
2463
2464	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
2465	if (xfs_inode_has_attr_fork(ip))
2466		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
 
2467
2468	/*
2469	 * We've recorded everything logged in the inode, so we'd like to clear
2470	 * the ili_fields bits so we don't log and flush things unnecessarily.
2471	 * However, we can't stop logging all this information until the data
2472	 * we've copied into the disk buffer is written to disk.  If we did we
2473	 * might overwrite the copy of the inode in the log with all the data
2474	 * after re-logging only part of it, and in the face of a crash we
2475	 * wouldn't have all the data we need to recover.
2476	 *
2477	 * What we do is move the bits to the ili_last_fields field.  When
2478	 * logging the inode, these bits are moved back to the ili_fields field.
2479	 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
2480	 * we know that the information those bits represent is permanently on
2481	 * disk.  As long as the flush completes before the inode is logged
2482	 * again, then both ili_fields and ili_last_fields will be cleared.
2483	 */
2484	error = 0;
2485flush_out:
2486	spin_lock(&iip->ili_lock);
2487	iip->ili_last_fields = iip->ili_fields;
2488	iip->ili_fields = 0;
2489	iip->ili_fsync_fields = 0;
2490	set_bit(XFS_LI_FLUSHING, &iip->ili_item.li_flags);
2491	spin_unlock(&iip->ili_lock);
2492
2493	/*
2494	 * Store the current LSN of the inode so that we can tell whether the
2495	 * item has moved in the AIL from xfs_buf_inode_iodone().
2496	 */
2497	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2498				&iip->ili_item.li_lsn);
2499
2500	/* generate the checksum. */
2501	xfs_dinode_calc_crc(mp, dip);
2502	if (error)
2503		xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
2504	return error;
2505}
2506
2507/*
2508 * Non-blocking flush of dirty inode metadata into the backing buffer.
2509 *
2510 * The caller must have a reference to the inode and hold the cluster buffer
2511 * locked. The function will walk across all the inodes on the cluster buffer it
2512 * can find and lock without blocking, and flush them to the cluster buffer.
2513 *
2514 * On successful flushing of at least one inode, the caller must write out the
2515 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
2516 * the caller needs to release the buffer. On failure, the filesystem will be
2517 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
2518 * will be returned.
2519 */
2520int
2521xfs_iflush_cluster(
 
2522	struct xfs_buf		*bp)
2523{
2524	struct xfs_mount	*mp = bp->b_mount;
2525	struct xfs_log_item	*lip, *n;
2526	struct xfs_inode	*ip;
2527	struct xfs_inode_log_item *iip;
 
 
 
 
2528	int			clcount = 0;
2529	int			error = 0;
 
 
 
 
 
 
 
 
 
2530
2531	/*
2532	 * We must use the safe variant here as on shutdown xfs_iflush_abort()
2533	 * will remove itself from the list.
2534	 */
2535	list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
2536		iip = (struct xfs_inode_log_item *)lip;
2537		ip = iip->ili_inode;
 
 
 
 
 
 
2538
2539		/*
2540		 * Quick and dirty check to avoid locks if possible.
 
 
 
2541		 */
2542		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
2543			continue;
2544		if (xfs_ipincount(ip))
 
2545			continue;
 
2546
2547		/*
2548		 * The inode is still attached to the buffer, which means it is
2549		 * dirty but reclaim might try to grab it. Check carefully for
2550		 * that, and grab the ilock while still holding the i_flags_lock
2551		 * to guarantee reclaim will not be able to reclaim this inode
2552		 * once we drop the i_flags_lock.
2553		 */
2554		spin_lock(&ip->i_flags_lock);
2555		ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
2556		if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
2557			spin_unlock(&ip->i_flags_lock);
2558			continue;
2559		}
 
2560
2561		/*
2562		 * ILOCK will pin the inode against reclaim and prevent
2563		 * concurrent transactions modifying the inode while we are
2564		 * flushing the inode. If we get the lock, set the flushing
2565		 * state before we drop the i_flags_lock.
2566		 */
2567		if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
2568			spin_unlock(&ip->i_flags_lock);
2569			continue;
2570		}
2571		__xfs_iflags_set(ip, XFS_IFLUSHING);
2572		spin_unlock(&ip->i_flags_lock);
2573
2574		/*
2575		 * Abort flushing this inode if we are shut down because the
2576		 * inode may not currently be in the AIL. This can occur when
2577		 * log I/O failure unpins the inode without inserting into the
2578		 * AIL, leaving a dirty/unpinned inode attached to the buffer
2579		 * that otherwise looks like it should be flushed.
2580		 */
2581		if (xlog_is_shutdown(mp->m_log)) {
2582			xfs_iunpin_wait(ip);
2583			xfs_iflush_abort(ip);
2584			xfs_iunlock(ip, XFS_ILOCK_SHARED);
2585			error = -EIO;
 
 
2586			continue;
2587		}
2588
2589		/* don't block waiting on a log force to unpin dirty inodes */
2590		if (xfs_ipincount(ip)) {
2591			xfs_iflags_clear(ip, XFS_IFLUSHING);
2592			xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
 
 
 
 
 
2593			continue;
2594		}
2595
2596		if (!xfs_inode_clean(ip))
2597			error = xfs_iflush(ip, bp);
2598		else
2599			xfs_iflags_clear(ip, XFS_IFLUSHING);
2600		xfs_iunlock(ip, XFS_ILOCK_SHARED);
2601		if (error)
2602			break;
2603		clcount++;
 
 
 
 
 
 
 
 
2604	}
2605
2606	if (error) {
2607		/*
2608		 * Shutdown first so we kill the log before we release this
2609		 * buffer. If it is an INODE_ALLOC buffer and pins the tail
2610		 * of the log, failing it before the _log_ is shut down can
2611		 * result in the log tail being moved forward in the journal
2612		 * on disk because log writes can still be taking place. Hence
2613		 * unpinning the tail will allow the ICREATE intent to be
2614		 * removed from the log an recovery will fail with uninitialised
2615		 * inode cluster buffers.
2616		 */
2617		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2618		bp->b_flags |= XBF_ASYNC;
2619		xfs_buf_ioend_fail(bp);
2620		return error;
2621	}
2622
2623	if (!clcount)
2624		return -EAGAIN;
2625
2626	XFS_STATS_INC(mp, xs_icluster_flushcnt);
2627	XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
2628	return 0;
2629
2630}
2631
2632/* Release an inode. */
2633void
2634xfs_irele(
2635	struct xfs_inode	*ip)
2636{
2637	trace_xfs_irele(ip, _RET_IP_);
2638	iput(VFS_I(ip));
2639}
 
 
 
 
 
 
2640
2641/*
2642 * Ensure all commited transactions touching the inode are written to the log.
2643 */
2644int
2645xfs_log_force_inode(
2646	struct xfs_inode	*ip)
2647{
2648	xfs_csn_t		seq = 0;
2649
2650	xfs_ilock(ip, XFS_ILOCK_SHARED);
2651	if (xfs_ipincount(ip))
2652		seq = ip->i_itemp->ili_commit_seq;
2653	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
 
 
 
 
 
 
 
 
 
 
 
2654
2655	if (!seq)
2656		return 0;
2657	return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
 
 
 
 
2658}
2659
2660/*
2661 * Grab the exclusive iolock for a data copy from src to dest, making sure to
2662 * abide vfs locking order (lowest pointer value goes first) and breaking the
2663 * layout leases before proceeding.  The loop is needed because we cannot call
2664 * the blocking break_layout() with the iolocks held, and therefore have to
2665 * back out both locks.
 
 
2666 */
2667static int
2668xfs_iolock_two_inodes_and_break_layout(
2669	struct inode		*src,
2670	struct inode		*dest)
2671{
 
 
 
2672	int			error;
2673
2674	if (src > dest)
2675		swap(src, dest);
 
 
 
 
2676
2677retry:
2678	/* Wait to break both inodes' layouts before we start locking. */
2679	error = break_layout(src, true);
2680	if (error)
2681		return error;
2682	if (src != dest) {
2683		error = break_layout(dest, true);
2684		if (error)
2685			return error;
2686	}
2687
2688	/* Lock one inode and make sure nobody got in and leased it. */
2689	inode_lock(src);
2690	error = break_layout(src, false);
2691	if (error) {
2692		inode_unlock(src);
2693		if (error == -EWOULDBLOCK)
2694			goto retry;
2695		return error;
2696	}
2697
2698	if (src == dest)
 
 
 
 
 
 
 
 
 
2699		return 0;
 
2700
2701	/* Lock the other inode and make sure nobody got in and leased it. */
2702	inode_lock_nested(dest, I_MUTEX_NONDIR2);
2703	error = break_layout(dest, false);
2704	if (error) {
2705		inode_unlock(src);
2706		inode_unlock(dest);
2707		if (error == -EWOULDBLOCK)
2708			goto retry;
2709		return error;
 
 
2710	}
2711
2712	return 0;
2713}
2714
2715static int
2716xfs_mmaplock_two_inodes_and_break_dax_layout(
2717	struct xfs_inode	*ip1,
2718	struct xfs_inode	*ip2)
2719{
2720	int			error;
2721	bool			retry;
2722	struct page		*page;
2723
2724	if (ip1->i_ino > ip2->i_ino)
2725		swap(ip1, ip2);
2726
2727again:
2728	retry = false;
2729	/* Lock the first inode */
2730	xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
2731	error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
2732	if (error || retry) {
2733		xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
2734		if (error == 0 && retry)
2735			goto again;
2736		return error;
2737	}
 
 
2738
2739	if (ip1 == ip2)
2740		return 0;
 
 
 
 
2741
2742	/* Nested lock the second inode */
2743	xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
2744	/*
2745	 * We cannot use xfs_break_dax_layouts() directly here because it may
2746	 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
2747	 * for this nested lock case.
2748	 */
2749	page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
2750	if (page && page_ref_count(page) != 1) {
2751		xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
2752		xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
2753		goto again;
2754	}
2755
2756	return 0;
2757}
2758
2759/*
2760 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
2761 * mmap activity.
2762 */
2763int
2764xfs_ilock2_io_mmap(
2765	struct xfs_inode	*ip1,
2766	struct xfs_inode	*ip2)
2767{
2768	int			ret;
2769
2770	ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
2771	if (ret)
2772		return ret;
2773
2774	if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
2775		ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
2776		if (ret) {
2777			inode_unlock(VFS_I(ip2));
2778			if (ip1 != ip2)
2779				inode_unlock(VFS_I(ip1));
2780			return ret;
2781		}
2782	} else
2783		filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
2784					    VFS_I(ip2)->i_mapping);
2785
 
2786	return 0;
2787}
2788
2789/* Unlock both inodes to allow IO and mmap activity. */
2790void
2791xfs_iunlock2_io_mmap(
2792	struct xfs_inode	*ip1,
2793	struct xfs_inode	*ip2)
2794{
2795	if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
2796		xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
2797		if (ip1 != ip2)
2798			xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
2799	} else
2800		filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
2801					      VFS_I(ip2)->i_mapping);
2802
2803	inode_unlock(VFS_I(ip2));
2804	if (ip1 != ip2)
2805		inode_unlock(VFS_I(ip1));
2806}
2807
2808/* Drop the MMAPLOCK and the IOLOCK after a remap completes. */
2809void
2810xfs_iunlock2_remapping(
2811	struct xfs_inode	*ip1,
2812	struct xfs_inode	*ip2)
2813{
2814	xfs_iflags_clear(ip1, XFS_IREMAPPING);
 
 
2815
2816	if (ip1 != ip2)
2817		xfs_iunlock(ip1, XFS_MMAPLOCK_SHARED);
2818	xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
 
 
 
2819
2820	if (ip1 != ip2)
2821		inode_unlock_shared(VFS_I(ip1));
2822	inode_unlock(VFS_I(ip2));
2823}
2824
2825/*
2826 * Reload the incore inode list for this inode.  Caller should ensure that
2827 * the link count cannot change, either by taking ILOCK_SHARED or otherwise
2828 * preventing other threads from executing.
2829 */
2830int
2831xfs_inode_reload_unlinked_bucket(
2832	struct xfs_trans	*tp,
2833	struct xfs_inode	*ip)
2834{
2835	struct xfs_mount	*mp = tp->t_mountp;
2836	struct xfs_buf		*agibp;
2837	struct xfs_agi		*agi;
2838	struct xfs_perag	*pag;
2839	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2840	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2841	xfs_agino_t		prev_agino, next_agino;
2842	unsigned int		bucket;
2843	bool			foundit = false;
2844	int			error;
2845
2846	/* Grab the first inode in the list */
2847	pag = xfs_perag_get(mp, agno);
2848	error = xfs_ialloc_read_agi(pag, tp, 0, &agibp);
2849	xfs_perag_put(pag);
2850	if (error)
2851		return error;
2852
2853	/*
2854	 * We've taken ILOCK_SHARED and the AGI buffer lock to stabilize the
2855	 * incore unlinked list pointers for this inode.  Check once more to
2856	 * see if we raced with anyone else to reload the unlinked list.
2857	 */
2858	if (!xfs_inode_unlinked_incomplete(ip)) {
2859		foundit = true;
2860		goto out_agibp;
2861	}
2862
2863	bucket = agino % XFS_AGI_UNLINKED_BUCKETS;
2864	agi = agibp->b_addr;
2865
2866	trace_xfs_inode_reload_unlinked_bucket(ip);
2867
2868	xfs_info_ratelimited(mp,
2869 "Found unrecovered unlinked inode 0x%x in AG 0x%x.  Initiating list recovery.",
2870			agino, agno);
2871
2872	prev_agino = NULLAGINO;
2873	next_agino = be32_to_cpu(agi->agi_unlinked[bucket]);
2874	while (next_agino != NULLAGINO) {
2875		struct xfs_inode	*next_ip = NULL;
2876
2877		/* Found this caller's inode, set its backlink. */
2878		if (next_agino == agino) {
2879			next_ip = ip;
2880			next_ip->i_prev_unlinked = prev_agino;
2881			foundit = true;
2882			goto next_inode;
2883		}
2884
2885		/* Try in-memory lookup first. */
2886		next_ip = xfs_iunlink_lookup(pag, next_agino);
2887		if (next_ip)
2888			goto next_inode;
2889
2890		/* Inode not in memory, try reloading it. */
2891		error = xfs_iunlink_reload_next(tp, agibp, prev_agino,
2892				next_agino);
2893		if (error)
2894			break;
2895
2896		/* Grab the reloaded inode. */
2897		next_ip = xfs_iunlink_lookup(pag, next_agino);
2898		if (!next_ip) {
2899			/* No incore inode at all?  We reloaded it... */
2900			ASSERT(next_ip != NULL);
2901			error = -EFSCORRUPTED;
2902			break;
2903		}
2904
2905next_inode:
2906		prev_agino = next_agino;
2907		next_agino = next_ip->i_next_unlinked;
2908	}
2909
2910out_agibp:
2911	xfs_trans_brelse(tp, agibp);
2912	/* Should have found this inode somewhere in the iunlinked bucket. */
2913	if (!error && !foundit)
2914		error = -EFSCORRUPTED;
2915	return error;
2916}
2917
2918/* Decide if this inode is missing its unlinked list and reload it. */
2919int
2920xfs_inode_reload_unlinked(
2921	struct xfs_inode	*ip)
2922{
2923	struct xfs_trans	*tp;
2924	int			error;
2925
2926	error = xfs_trans_alloc_empty(ip->i_mount, &tp);
2927	if (error)
2928		return error;
2929
2930	xfs_ilock(ip, XFS_ILOCK_SHARED);
2931	if (xfs_inode_unlinked_incomplete(ip))
2932		error = xfs_inode_reload_unlinked_bucket(tp, ip);
2933	xfs_iunlock(ip, XFS_ILOCK_SHARED);
2934	xfs_trans_cancel(tp);
2935
2936	return error;
2937}
2938
2939/* Has this inode fork been zapped by repair? */
2940bool
2941xfs_ifork_zapped(
2942	const struct xfs_inode	*ip,
2943	int			whichfork)
2944{
2945	unsigned int		datamask = 0;
2946
2947	switch (whichfork) {
2948	case XFS_DATA_FORK:
2949		switch (ip->i_vnode.i_mode & S_IFMT) {
2950		case S_IFDIR:
2951			datamask = XFS_SICK_INO_DIR_ZAPPED;
2952			break;
2953		case S_IFLNK:
2954			datamask = XFS_SICK_INO_SYMLINK_ZAPPED;
2955			break;
2956		}
2957		return ip->i_sick & (XFS_SICK_INO_BMBTD_ZAPPED | datamask);
2958	case XFS_ATTR_FORK:
2959		return ip->i_sick & XFS_SICK_INO_BMBTA_ZAPPED;
2960	default:
2961		return false;
2962	}
2963}
2964
2965/* Compute the number of data and realtime blocks used by a file. */
2966void
2967xfs_inode_count_blocks(
2968	struct xfs_trans	*tp,
2969	struct xfs_inode	*ip,
2970	xfs_filblks_t		*dblocks,
2971	xfs_filblks_t		*rblocks)
2972{
2973	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
 
 
2974
2975	*rblocks = 0;
2976	if (XFS_IS_REALTIME_INODE(ip))
2977		xfs_bmap_count_leaves(ifp, rblocks);
2978	*dblocks = ip->i_nblocks - *rblocks;
2979}
 
2980
2981static void
2982xfs_wait_dax_page(
2983	struct inode		*inode)
2984{
2985	struct xfs_inode        *ip = XFS_I(inode);
2986
2987	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
2988	schedule();
2989	xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
2990}
2991
2992int
2993xfs_break_dax_layouts(
2994	struct inode		*inode,
2995	bool			*retry)
2996{
2997	struct page		*page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2998
2999	xfs_assert_ilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL);
 
3000
3001	page = dax_layout_busy_page(inode->i_mapping);
3002	if (!page)
3003		return 0;
 
 
 
 
3004
3005	*retry = true;
3006	return ___wait_var_event(&page->_refcount,
3007			atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
3008			0, 0, xfs_wait_dax_page(inode));
3009}
3010
3011int
3012xfs_break_layouts(
3013	struct inode		*inode,
3014	uint			*iolock,
3015	enum layout_break_reason reason)
3016{
3017	bool			retry;
3018	int			error;
3019
3020	xfs_assert_ilocked(XFS_I(inode), XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL);
3021
3022	do {
3023		retry = false;
3024		switch (reason) {
3025		case BREAK_UNMAP:
3026			error = xfs_break_dax_layouts(inode, &retry);
3027			if (error || retry)
3028				break;
3029			fallthrough;
3030		case BREAK_WRITE:
3031			error = xfs_break_leased_layouts(inode, iolock, &retry);
3032			break;
3033		default:
3034			WARN_ON_ONCE(1);
3035			error = -EINVAL;
3036		}
3037	} while (error == 0 && retry);
3038
3039	return error;
3040}
3041
3042/* Returns the size of fundamental allocation unit for a file, in bytes. */
3043unsigned int
3044xfs_inode_alloc_unitsize(
3045	struct xfs_inode	*ip)
3046{
3047	unsigned int		blocks = 1;
3048
3049	if (XFS_IS_REALTIME_INODE(ip))
3050		blocks = ip->i_mount->m_sb.sb_rextsize;
3051
3052	return XFS_FSB_TO_B(ip->i_mount, blocks);
3053}
3054
3055/* Should we always be using copy on write for file writes? */
3056bool
3057xfs_is_always_cow_inode(
3058	const struct xfs_inode	*ip)
3059{
3060	return ip->i_mount->m_always_cow && xfs_has_reflink(ip->i_mount);
3061}