Linux Audio

Check our new training course

Loading...
v3.1
 
   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_types.h"
  23#include "xfs_bit.h"
  24#include "xfs_log.h"
  25#include "xfs_inum.h"
  26#include "xfs_trans.h"
  27#include "xfs_trans_priv.h"
  28#include "xfs_sb.h"
  29#include "xfs_ag.h"
  30#include "xfs_mount.h"
  31#include "xfs_bmap_btree.h"
  32#include "xfs_alloc_btree.h"
  33#include "xfs_ialloc_btree.h"
  34#include "xfs_attr_sf.h"
  35#include "xfs_dinode.h"
  36#include "xfs_inode.h"
 
 
 
 
 
  37#include "xfs_buf_item.h"
  38#include "xfs_inode_item.h"
  39#include "xfs_btree.h"
  40#include "xfs_alloc.h"
  41#include "xfs_ialloc.h"
  42#include "xfs_bmap.h"
 
 
  43#include "xfs_error.h"
  44#include "xfs_utils.h"
  45#include "xfs_quota.h"
  46#include "xfs_filestream.h"
  47#include "xfs_vnodeops.h"
  48#include "xfs_trace.h"
  49
  50kmem_zone_t *xfs_ifork_zone;
  51kmem_zone_t *xfs_inode_zone;
  52
  53/*
  54 * Used in xfs_itruncate_extents().  This is the maximum number of extents
  55 * freed from a file in a single transaction.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  56 */
  57#define	XFS_ITRUNC_MAX_EXTENTS	2
 
 
 
 
  58
  59STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
  60STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
  61STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
  62STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
 
  63
  64#ifdef DEBUG
  65/*
  66 * Make sure that the extents in the given memory buffer
  67 * are valid.
  68 */
  69STATIC void
  70xfs_validate_extents(
  71	xfs_ifork_t		*ifp,
  72	int			nrecs,
  73	xfs_exntfmt_t		fmt)
  74{
  75	xfs_bmbt_irec_t		irec;
  76	xfs_bmbt_rec_host_t	rec;
  77	int			i;
  78
  79	for (i = 0; i < nrecs; i++) {
  80		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
  81		rec.l0 = get_unaligned(&ep->l0);
  82		rec.l1 = get_unaligned(&ep->l1);
  83		xfs_bmbt_get_all(&rec, &irec);
  84		if (fmt == XFS_EXTFMT_NOSTATE)
  85			ASSERT(irec.br_state == XFS_EXT_NORM);
  86	}
  87}
  88#else /* DEBUG */
  89#define xfs_validate_extents(ifp, nrecs, fmt)
  90#endif /* DEBUG */
  91
  92/*
  93 * Check that none of the inode's in the buffer have a next
  94 * unlinked field of 0.
 
 
  95 */
  96#if defined(DEBUG)
  97void
  98xfs_inobp_check(
  99	xfs_mount_t	*mp,
 100	xfs_buf_t	*bp)
 101{
 102	int		i;
 103	int		j;
 104	xfs_dinode_t	*dip;
 105
 106	j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
 107
 108	for (i = 0; i < j; i++) {
 109		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 110					i * mp->m_sb.sb_inodesize);
 111		if (!dip->di_next_unlinked)  {
 112			xfs_alert(mp,
 113	"Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
 114				bp);
 115			ASSERT(dip->di_next_unlinked);
 116		}
 117	}
 118}
 119#endif
 120
 121/*
 122 * Find the buffer associated with the given inode map
 123 * We do basic validation checks on the buffer once it has been
 124 * retrieved from disk.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 125 */
 126STATIC int
 127xfs_imap_to_bp(
 128	xfs_mount_t	*mp,
 129	xfs_trans_t	*tp,
 130	struct xfs_imap	*imap,
 131	xfs_buf_t	**bpp,
 132	uint		buf_flags,
 133	uint		iget_flags)
 134{
 135	int		error;
 136	int		i;
 137	int		ni;
 138	xfs_buf_t	*bp;
 139
 140	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
 141				   (int)imap->im_len, buf_flags, &bp);
 142	if (error) {
 143		if (error != EAGAIN) {
 144			xfs_warn(mp,
 145				"%s: xfs_trans_read_buf() returned error %d.",
 146				__func__, error);
 147		} else {
 148			ASSERT(buf_flags & XBF_TRYLOCK);
 149		}
 150		return error;
 151	}
 152
 153	/*
 154	 * Validate the magic number and version of every inode in the buffer
 155	 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
 156	 */
 157#ifdef DEBUG
 158	ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
 159#else	/* usual case */
 160	ni = 1;
 161#endif
 162
 163	for (i = 0; i < ni; i++) {
 164		int		di_ok;
 165		xfs_dinode_t	*dip;
 166
 167		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
 168					(i << mp->m_sb.sb_inodelog));
 169		di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
 170			    XFS_DINODE_GOOD_VERSION(dip->di_version);
 171		if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
 172						XFS_ERRTAG_ITOBP_INOTOBP,
 173						XFS_RANDOM_ITOBP_INOTOBP))) {
 174			if (iget_flags & XFS_IGET_UNTRUSTED) {
 175				xfs_trans_brelse(tp, bp);
 176				return XFS_ERROR(EINVAL);
 177			}
 178			XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
 179						XFS_ERRLEVEL_HIGH, mp, dip);
 180#ifdef DEBUG
 181			xfs_emerg(mp,
 182				"bad inode magic/vsn daddr %lld #%d (magic=%x)",
 183				(unsigned long long)imap->im_blkno, i,
 184				be16_to_cpu(dip->di_magic));
 185			ASSERT(0);
 186#endif
 187			xfs_trans_brelse(tp, bp);
 188			return XFS_ERROR(EFSCORRUPTED);
 189		}
 190	}
 191
 192	xfs_inobp_check(mp, bp);
 193
 194	/*
 195	 * Mark the buffer as an inode buffer now that it looks good
 196	 */
 197	XFS_BUF_SET_VTYPE(bp, B_FS_INO);
 
 198
 199	*bpp = bp;
 200	return 0;
 
 
 201}
 202
 203/*
 204 * This routine is called to map an inode number within a file
 205 * system to the buffer containing the on-disk version of the
 206 * inode.  It returns a pointer to the buffer containing the
 207 * on-disk inode in the bpp parameter, and in the dip parameter
 208 * it returns a pointer to the on-disk inode within that buffer.
 209 *
 210 * If a non-zero error is returned, then the contents of bpp and
 211 * dipp are undefined.
 212 *
 213 * Use xfs_imap() to determine the size and location of the
 214 * buffer to read from disk.
 
 
 215 */
 216int
 217xfs_inotobp(
 218	xfs_mount_t	*mp,
 219	xfs_trans_t	*tp,
 220	xfs_ino_t	ino,
 221	xfs_dinode_t	**dipp,
 222	xfs_buf_t	**bpp,
 223	int		*offset,
 224	uint		imap_flags)
 225{
 226	struct xfs_imap	imap;
 227	xfs_buf_t	*bp;
 228	int		error;
 229
 230	imap.im_blkno = 0;
 231	error = xfs_imap(mp, tp, ino, &imap, imap_flags);
 232	if (error)
 233		return error;
 234
 235	error = xfs_imap_to_bp(mp, tp, &imap, &bp, XBF_LOCK, imap_flags);
 236	if (error)
 237		return error;
 
 
 
 
 238
 239	*dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
 240	*bpp = bp;
 241	*offset = imap.im_boffset;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 242	return 0;
 243}
 244
 245
 246/*
 247 * This routine is called to map an inode to the buffer containing
 248 * the on-disk version of the inode.  It returns a pointer to the
 249 * buffer containing the on-disk inode in the bpp parameter, and in
 250 * the dip parameter it returns a pointer to the on-disk inode within
 251 * that buffer.
 252 *
 253 * If a non-zero error is returned, then the contents of bpp and
 254 * dipp are undefined.
 255 *
 256 * The inode is expected to already been mapped to its buffer and read
 257 * in once, thus we can use the mapping information stored in the inode
 258 * rather than calling xfs_imap().  This allows us to avoid the overhead
 259 * of looking at the inode btree for small block file systems
 260 * (see xfs_imap()).
 261 */
 262int
 263xfs_itobp(
 264	xfs_mount_t	*mp,
 265	xfs_trans_t	*tp,
 266	xfs_inode_t	*ip,
 267	xfs_dinode_t	**dipp,
 268	xfs_buf_t	**bpp,
 269	uint		buf_flags)
 270{
 271	xfs_buf_t	*bp;
 272	int		error;
 273
 274	ASSERT(ip->i_imap.im_blkno != 0);
 
 
 
 275
 276	error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
 277	if (error)
 278		return error;
 
 279
 280	if (!bp) {
 281		ASSERT(buf_flags & XBF_TRYLOCK);
 282		ASSERT(tp == NULL);
 283		*bpp = NULL;
 284		return EAGAIN;
 285	}
 286
 287	*dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
 288	*bpp = bp;
 289	return 0;
 290}
 291
 292/*
 293 * Move inode type and inode format specific information from the
 294 * on-disk inode to the in-core inode.  For fifos, devs, and sockets
 295 * this means set if_rdev to the proper value.  For files, directories,
 296 * and symlinks this means to bring in the in-line data or extent
 297 * pointers.  For a file in B-tree format, only the root is immediately
 298 * brought in-core.  The rest will be in-lined in if_extents when it
 299 * is first referenced (see xfs_iread_extents()).
 300 */
 301STATIC int
 302xfs_iformat(
 303	xfs_inode_t		*ip,
 304	xfs_dinode_t		*dip)
 305{
 306	xfs_attr_shortform_t	*atp;
 307	int			size;
 308	int			error;
 309	xfs_fsize_t             di_size;
 310	ip->i_df.if_ext_max =
 311		XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
 312	error = 0;
 313
 314	if (unlikely(be32_to_cpu(dip->di_nextents) +
 315		     be16_to_cpu(dip->di_anextents) >
 316		     be64_to_cpu(dip->di_nblocks))) {
 317		xfs_warn(ip->i_mount,
 318			"corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
 319			(unsigned long long)ip->i_ino,
 320			(int)(be32_to_cpu(dip->di_nextents) +
 321			      be16_to_cpu(dip->di_anextents)),
 322			(unsigned long long)
 323				be64_to_cpu(dip->di_nblocks));
 324		XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
 325				     ip->i_mount, dip);
 326		return XFS_ERROR(EFSCORRUPTED);
 327	}
 328
 329	if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
 330		xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
 331			(unsigned long long)ip->i_ino,
 332			dip->di_forkoff);
 333		XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
 334				     ip->i_mount, dip);
 335		return XFS_ERROR(EFSCORRUPTED);
 336	}
 337
 338	if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
 339		     !ip->i_mount->m_rtdev_targp)) {
 340		xfs_warn(ip->i_mount,
 341			"corrupt dinode %Lu, has realtime flag set.",
 342			ip->i_ino);
 343		XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
 344				     XFS_ERRLEVEL_LOW, ip->i_mount, dip);
 345		return XFS_ERROR(EFSCORRUPTED);
 346	}
 347
 348	switch (ip->i_d.di_mode & S_IFMT) {
 349	case S_IFIFO:
 350	case S_IFCHR:
 351	case S_IFBLK:
 352	case S_IFSOCK:
 353		if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
 354			XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
 355					      ip->i_mount, dip);
 356			return XFS_ERROR(EFSCORRUPTED);
 357		}
 358		ip->i_d.di_size = 0;
 359		ip->i_size = 0;
 360		ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
 361		break;
 362
 363	case S_IFREG:
 364	case S_IFLNK:
 365	case S_IFDIR:
 366		switch (dip->di_format) {
 367		case XFS_DINODE_FMT_LOCAL:
 368			/*
 369			 * no local regular files yet
 370			 */
 371			if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
 372				xfs_warn(ip->i_mount,
 373			"corrupt inode %Lu (local format for regular file).",
 374					(unsigned long long) ip->i_ino);
 375				XFS_CORRUPTION_ERROR("xfs_iformat(4)",
 376						     XFS_ERRLEVEL_LOW,
 377						     ip->i_mount, dip);
 378				return XFS_ERROR(EFSCORRUPTED);
 379			}
 380
 381			di_size = be64_to_cpu(dip->di_size);
 382			if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
 383				xfs_warn(ip->i_mount,
 384			"corrupt inode %Lu (bad size %Ld for local inode).",
 385					(unsigned long long) ip->i_ino,
 386					(long long) di_size);
 387				XFS_CORRUPTION_ERROR("xfs_iformat(5)",
 388						     XFS_ERRLEVEL_LOW,
 389						     ip->i_mount, dip);
 390				return XFS_ERROR(EFSCORRUPTED);
 391			}
 392
 393			size = (int)di_size;
 394			error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
 395			break;
 396		case XFS_DINODE_FMT_EXTENTS:
 397			error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
 398			break;
 399		case XFS_DINODE_FMT_BTREE:
 400			error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
 401			break;
 402		default:
 403			XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
 404					 ip->i_mount);
 405			return XFS_ERROR(EFSCORRUPTED);
 406		}
 407		break;
 408
 409	default:
 410		XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
 411		return XFS_ERROR(EFSCORRUPTED);
 412	}
 413	if (error) {
 414		return error;
 415	}
 416	if (!XFS_DFORK_Q(dip))
 417		return 0;
 418	ASSERT(ip->i_afp == NULL);
 419	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
 420	ip->i_afp->if_ext_max =
 421		XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
 422	switch (dip->di_aformat) {
 423	case XFS_DINODE_FMT_LOCAL:
 424		atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
 425		size = be16_to_cpu(atp->hdr.totsize);
 426
 427		if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
 428			xfs_warn(ip->i_mount,
 429				"corrupt inode %Lu (bad attr fork size %Ld).",
 430				(unsigned long long) ip->i_ino,
 431				(long long) size);
 432			XFS_CORRUPTION_ERROR("xfs_iformat(8)",
 433					     XFS_ERRLEVEL_LOW,
 434					     ip->i_mount, dip);
 435			return XFS_ERROR(EFSCORRUPTED);
 436		}
 437
 438		error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
 439		break;
 440	case XFS_DINODE_FMT_EXTENTS:
 441		error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
 442		break;
 443	case XFS_DINODE_FMT_BTREE:
 444		error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
 445		break;
 446	default:
 447		error = XFS_ERROR(EFSCORRUPTED);
 448		break;
 449	}
 450	if (error) {
 451		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
 452		ip->i_afp = NULL;
 453		xfs_idestroy_fork(ip, XFS_DATA_FORK);
 454	}
 455	return error;
 456}
 457
 458/*
 459 * The file is in-lined in the on-disk inode.
 460 * If it fits into if_inline_data, then copy
 461 * it there, otherwise allocate a buffer for it
 462 * and copy the data there.  Either way, set
 463 * if_data to point at the data.
 464 * If we allocate a buffer for the data, make
 465 * sure that its size is a multiple of 4 and
 466 * record the real size in i_real_bytes.
 467 */
 468STATIC int
 469xfs_iformat_local(
 470	xfs_inode_t	*ip,
 471	xfs_dinode_t	*dip,
 472	int		whichfork,
 473	int		size)
 474{
 475	xfs_ifork_t	*ifp;
 476	int		real_size;
 477
 478	/*
 479	 * If the size is unreasonable, then something
 480	 * is wrong and we just bail out rather than crash in
 481	 * kmem_alloc() or memcpy() below.
 482	 */
 483	if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
 484		xfs_warn(ip->i_mount,
 485	"corrupt inode %Lu (bad size %d for local fork, size = %d).",
 486			(unsigned long long) ip->i_ino, size,
 487			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
 488		XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
 489				     ip->i_mount, dip);
 490		return XFS_ERROR(EFSCORRUPTED);
 491	}
 492	ifp = XFS_IFORK_PTR(ip, whichfork);
 493	real_size = 0;
 494	if (size == 0)
 495		ifp->if_u1.if_data = NULL;
 496	else if (size <= sizeof(ifp->if_u2.if_inline_data))
 497		ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
 498	else {
 499		real_size = roundup(size, 4);
 500		ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
 501	}
 502	ifp->if_bytes = size;
 503	ifp->if_real_bytes = real_size;
 504	if (size)
 505		memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
 506	ifp->if_flags &= ~XFS_IFEXTENTS;
 507	ifp->if_flags |= XFS_IFINLINE;
 508	return 0;
 509}
 
 
 
 510
 511/*
 512 * The file consists of a set of extents all
 513 * of which fit into the on-disk inode.
 514 * If there are few enough extents to fit into
 515 * the if_inline_ext, then copy them there.
 516 * Otherwise allocate a buffer for them and copy
 517 * them into it.  Either way, set if_extents
 518 * to point at the extents.
 519 */
 520STATIC int
 521xfs_iformat_extents(
 522	xfs_inode_t	*ip,
 523	xfs_dinode_t	*dip,
 524	int		whichfork)
 525{
 526	xfs_bmbt_rec_t	*dp;
 527	xfs_ifork_t	*ifp;
 528	int		nex;
 529	int		size;
 530	int		i;
 531
 532	ifp = XFS_IFORK_PTR(ip, whichfork);
 533	nex = XFS_DFORK_NEXTENTS(dip, whichfork);
 534	size = nex * (uint)sizeof(xfs_bmbt_rec_t);
 535
 536	/*
 537	 * If the number of extents is unreasonable, then something
 538	 * is wrong and we just bail out rather than crash in
 539	 * kmem_alloc() or memcpy() below.
 540	 */
 541	if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
 542		xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
 543			(unsigned long long) ip->i_ino, nex);
 544		XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
 545				     ip->i_mount, dip);
 546		return XFS_ERROR(EFSCORRUPTED);
 547	}
 548
 549	ifp->if_real_bytes = 0;
 550	if (nex == 0)
 551		ifp->if_u1.if_extents = NULL;
 552	else if (nex <= XFS_INLINE_EXTS)
 553		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
 554	else
 555		xfs_iext_add(ifp, 0, nex);
 556
 557	ifp->if_bytes = size;
 558	if (size) {
 559		dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
 560		xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
 561		for (i = 0; i < nex; i++, dp++) {
 562			xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
 563			ep->l0 = get_unaligned_be64(&dp->l0);
 564			ep->l1 = get_unaligned_be64(&dp->l1);
 565		}
 566		XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
 567		if (whichfork != XFS_DATA_FORK ||
 568			XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
 569				if (unlikely(xfs_check_nostate_extents(
 570				    ifp, 0, nex))) {
 571					XFS_ERROR_REPORT("xfs_iformat_extents(2)",
 572							 XFS_ERRLEVEL_LOW,
 573							 ip->i_mount);
 574					return XFS_ERROR(EFSCORRUPTED);
 575				}
 576	}
 577	ifp->if_flags |= XFS_IFEXTENTS;
 578	return 0;
 
 
 
 
 
 
 
 
 
 
 579}
 580
 581/*
 582 * The file has too many extents to fit into
 583 * the inode, so they are in B-tree format.
 584 * Allocate a buffer for the root of the B-tree
 585 * and copy the root into it.  The i_extents
 586 * field will remain NULL until all of the
 587 * extents are read in (when they are needed).
 
 
 
 
 
 
 
 588 */
 589STATIC int
 590xfs_iformat_btree(
 591	xfs_inode_t		*ip,
 592	xfs_dinode_t		*dip,
 593	int			whichfork)
 594{
 595	xfs_bmdr_block_t	*dfp;
 596	xfs_ifork_t		*ifp;
 597	/* REFERENCED */
 598	int			nrecs;
 599	int			size;
 600
 601	ifp = XFS_IFORK_PTR(ip, whichfork);
 602	dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
 603	size = XFS_BMAP_BROOT_SPACE(dfp);
 604	nrecs = be16_to_cpu(dfp->bb_numrecs);
 605
 606	/*
 607	 * blow out if -- fork has less extents than can fit in
 608	 * fork (fork shouldn't be a btree format), root btree
 609	 * block has more records than can fit into the fork,
 610	 * or the number of extents is greater than the number of
 611	 * blocks.
 612	 */
 613	if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
 614	    || XFS_BMDR_SPACE_CALC(nrecs) >
 615			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
 616	    || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
 617		xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
 618			(unsigned long long) ip->i_ino);
 619		XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
 620				 ip->i_mount, dip);
 621		return XFS_ERROR(EFSCORRUPTED);
 622	}
 623
 624	ifp->if_broot_bytes = size;
 625	ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
 626	ASSERT(ifp->if_broot != NULL);
 627	/*
 628	 * Copy and convert from the on-disk structure
 629	 * to the in-memory structure.
 630	 */
 631	xfs_bmdr_to_bmbt(ip->i_mount, dfp,
 632			 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
 633			 ifp->if_broot, size);
 634	ifp->if_flags &= ~XFS_IFEXTENTS;
 635	ifp->if_flags |= XFS_IFBROOT;
 636
 637	return 0;
 638}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 639
 640STATIC void
 641xfs_dinode_from_disk(
 642	xfs_icdinode_t		*to,
 643	xfs_dinode_t		*from)
 644{
 645	to->di_magic = be16_to_cpu(from->di_magic);
 646	to->di_mode = be16_to_cpu(from->di_mode);
 647	to->di_version = from ->di_version;
 648	to->di_format = from->di_format;
 649	to->di_onlink = be16_to_cpu(from->di_onlink);
 650	to->di_uid = be32_to_cpu(from->di_uid);
 651	to->di_gid = be32_to_cpu(from->di_gid);
 652	to->di_nlink = be32_to_cpu(from->di_nlink);
 653	to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
 654	to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
 655	memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
 656	to->di_flushiter = be16_to_cpu(from->di_flushiter);
 657	to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
 658	to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
 659	to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
 660	to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
 661	to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
 662	to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
 663	to->di_size = be64_to_cpu(from->di_size);
 664	to->di_nblocks = be64_to_cpu(from->di_nblocks);
 665	to->di_extsize = be32_to_cpu(from->di_extsize);
 666	to->di_nextents = be32_to_cpu(from->di_nextents);
 667	to->di_anextents = be16_to_cpu(from->di_anextents);
 668	to->di_forkoff = from->di_forkoff;
 669	to->di_aformat	= from->di_aformat;
 670	to->di_dmevmask	= be32_to_cpu(from->di_dmevmask);
 671	to->di_dmstate	= be16_to_cpu(from->di_dmstate);
 672	to->di_flags	= be16_to_cpu(from->di_flags);
 673	to->di_gen	= be32_to_cpu(from->di_gen);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 674}
 675
 
 
 
 
 
 
 676void
 677xfs_dinode_to_disk(
 678	xfs_dinode_t		*to,
 679	xfs_icdinode_t		*from)
 680{
 681	to->di_magic = cpu_to_be16(from->di_magic);
 682	to->di_mode = cpu_to_be16(from->di_mode);
 683	to->di_version = from ->di_version;
 684	to->di_format = from->di_format;
 685	to->di_onlink = cpu_to_be16(from->di_onlink);
 686	to->di_uid = cpu_to_be32(from->di_uid);
 687	to->di_gid = cpu_to_be32(from->di_gid);
 688	to->di_nlink = cpu_to_be32(from->di_nlink);
 689	to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
 690	to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
 691	memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
 692	to->di_flushiter = cpu_to_be16(from->di_flushiter);
 693	to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
 694	to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
 695	to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
 696	to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
 697	to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
 698	to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
 699	to->di_size = cpu_to_be64(from->di_size);
 700	to->di_nblocks = cpu_to_be64(from->di_nblocks);
 701	to->di_extsize = cpu_to_be32(from->di_extsize);
 702	to->di_nextents = cpu_to_be32(from->di_nextents);
 703	to->di_anextents = cpu_to_be16(from->di_anextents);
 704	to->di_forkoff = from->di_forkoff;
 705	to->di_aformat = from->di_aformat;
 706	to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
 707	to->di_dmstate = cpu_to_be16(from->di_dmstate);
 708	to->di_flags = cpu_to_be16(from->di_flags);
 709	to->di_gen = cpu_to_be32(from->di_gen);
 710}
 711
 712STATIC uint
 713_xfs_dic2xflags(
 714	__uint16_t		di_flags)
 715{
 716	uint			flags = 0;
 717
 718	if (di_flags & XFS_DIFLAG_ANY) {
 719		if (di_flags & XFS_DIFLAG_REALTIME)
 720			flags |= XFS_XFLAG_REALTIME;
 721		if (di_flags & XFS_DIFLAG_PREALLOC)
 722			flags |= XFS_XFLAG_PREALLOC;
 723		if (di_flags & XFS_DIFLAG_IMMUTABLE)
 724			flags |= XFS_XFLAG_IMMUTABLE;
 725		if (di_flags & XFS_DIFLAG_APPEND)
 726			flags |= XFS_XFLAG_APPEND;
 727		if (di_flags & XFS_DIFLAG_SYNC)
 728			flags |= XFS_XFLAG_SYNC;
 729		if (di_flags & XFS_DIFLAG_NOATIME)
 730			flags |= XFS_XFLAG_NOATIME;
 731		if (di_flags & XFS_DIFLAG_NODUMP)
 732			flags |= XFS_XFLAG_NODUMP;
 733		if (di_flags & XFS_DIFLAG_RTINHERIT)
 734			flags |= XFS_XFLAG_RTINHERIT;
 735		if (di_flags & XFS_DIFLAG_PROJINHERIT)
 736			flags |= XFS_XFLAG_PROJINHERIT;
 737		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
 738			flags |= XFS_XFLAG_NOSYMLINKS;
 739		if (di_flags & XFS_DIFLAG_EXTSIZE)
 740			flags |= XFS_XFLAG_EXTSIZE;
 741		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
 742			flags |= XFS_XFLAG_EXTSZINHERIT;
 743		if (di_flags & XFS_DIFLAG_NODEFRAG)
 744			flags |= XFS_XFLAG_NODEFRAG;
 745		if (di_flags & XFS_DIFLAG_FILESTREAM)
 746			flags |= XFS_XFLAG_FILESTREAM;
 747	}
 748
 749	return flags;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 750}
 751
 752uint
 753xfs_ip2xflags(
 754	xfs_inode_t		*ip)
 
 
 
 
 
 
 
 
 
 755{
 756	xfs_icdinode_t		*dic = &ip->i_d;
 
 757
 758	return _xfs_dic2xflags(dic->di_flags) |
 759				(XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
 760}
 761
 762uint
 763xfs_dic2xflags(
 764	xfs_dinode_t		*dip)
 765{
 766	return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
 767				(XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 768}
 769
 770/*
 771 * Read the disk inode attributes into the in-core inode structure.
 
 
 
 772 */
 773int
 774xfs_iread(
 775	xfs_mount_t	*mp,
 776	xfs_trans_t	*tp,
 777	xfs_inode_t	*ip,
 778	uint		iget_flags)
 779{
 780	xfs_buf_t	*bp;
 781	xfs_dinode_t	*dip;
 782	int		error;
 783
 784	/*
 785	 * Fill in the location information in the in-core inode.
 
 786	 */
 787	error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
 788	if (error)
 789		return error;
 790
 791	/*
 792	 * Get pointers to the on-disk inode and the buffer containing it.
 793	 */
 794	error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
 795			       XBF_LOCK, iget_flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 796	if (error)
 797		return error;
 798	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
 799
 800	/*
 801	 * If we got something that isn't an inode it means someone
 802	 * (nfs or dmi) has a stale handle.
 803	 */
 804	if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) {
 805#ifdef DEBUG
 806		xfs_alert(mp,
 807			"%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
 808			__func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC);
 809#endif /* DEBUG */
 810		error = XFS_ERROR(EINVAL);
 811		goto out_brelse;
 812	}
 813
 
 
 
 
 814	/*
 815	 * If the on-disk inode is already linked to a directory
 816	 * entry, copy all of the inode into the in-core inode.
 817	 * xfs_iformat() handles copying in the inode format
 818	 * specific information.
 819	 * Otherwise, just get the truly permanent information.
 820	 */
 821	if (dip->di_mode) {
 822		xfs_dinode_from_disk(&ip->i_d, dip);
 823		error = xfs_iformat(ip, dip);
 824		if (error)  {
 825#ifdef DEBUG
 826			xfs_alert(mp, "%s: xfs_iformat() returned error %d",
 827				__func__, error);
 828#endif /* DEBUG */
 829			goto out_brelse;
 830		}
 831	} else {
 832		ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
 833		ip->i_d.di_version = dip->di_version;
 834		ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
 835		ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
 836		/*
 837		 * Make sure to pull in the mode here as well in
 838		 * case the inode is released without being used.
 839		 * This ensures that xfs_inactive() will see that
 840		 * the inode is already free and not try to mess
 841		 * with the uninitialized part of it.
 842		 */
 843		ip->i_d.di_mode = 0;
 844		/*
 845		 * Initialize the per-fork minima and maxima for a new
 846		 * inode here.  xfs_iformat will do it for old inodes.
 847		 */
 848		ip->i_df.if_ext_max =
 849			XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
 850	}
 
 
 
 
 
 851
 852	/*
 853	 * The inode format changed when we moved the link count and
 854	 * made it 32 bits long.  If this is an old format inode,
 855	 * convert it in memory to look like a new one.  If it gets
 856	 * flushed to disk we will convert back before flushing or
 857	 * logging it.  We zero out the new projid field and the old link
 858	 * count field.  We'll handle clearing the pad field (the remains
 859	 * of the old uuid field) when we actually convert the inode to
 860	 * the new format. We don't change the version number so that we
 861	 * can distinguish this from a real new format inode.
 862	 */
 863	if (ip->i_d.di_version == 1) {
 864		ip->i_d.di_nlink = ip->i_d.di_onlink;
 865		ip->i_d.di_onlink = 0;
 866		xfs_set_projid(ip, 0);
 867	}
 868
 869	ip->i_delayed_blks = 0;
 870	ip->i_size = ip->i_d.di_size;
 
 
 
 
 
 
 
 
 
 
 871
 872	/*
 873	 * Mark the buffer containing the inode as something to keep
 874	 * around for a while.  This helps to keep recently accessed
 875	 * meta-data in-core longer.
 876	 */
 877	xfs_buf_set_ref(bp, XFS_INO_REF);
 
 878
 879	/*
 880	 * Use xfs_trans_brelse() to release the buffer containing the
 881	 * on-disk inode, because it was acquired with xfs_trans_read_buf()
 882	 * in xfs_itobp() above.  If tp is NULL, this is just a normal
 883	 * brelse().  If we're within a transaction, then xfs_trans_brelse()
 884	 * will only release the buffer if it is not dirty within the
 885	 * transaction.  It will be OK to release the buffer in this case,
 886	 * because inodes on disk are never destroyed and we will be
 887	 * locking the new in-core inode before putting it in the hash
 888	 * table where other processes can find it.  Thus we don't have
 889	 * to worry about the inode being changed just because we released
 890	 * the buffer.
 891	 */
 892 out_brelse:
 893	xfs_trans_brelse(tp, bp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 894	return error;
 895}
 896
 897/*
 898 * Read in extents from a btree-format inode.
 899 * Allocate and fill in if_extents.  Real work is done in xfs_bmap.c.
 900 */
 901int
 902xfs_iread_extents(
 903	xfs_trans_t	*tp,
 904	xfs_inode_t	*ip,
 905	int		whichfork)
 906{
 907	int		error;
 908	xfs_ifork_t	*ifp;
 909	xfs_extnum_t	nextents;
 910
 911	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
 912		XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
 913				 ip->i_mount);
 914		return XFS_ERROR(EFSCORRUPTED);
 915	}
 916	nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
 917	ifp = XFS_IFORK_PTR(ip, whichfork);
 918
 919	/*
 920	 * We know that the size is valid (it's checked in iformat_btree)
 921	 */
 922	ifp->if_bytes = ifp->if_real_bytes = 0;
 923	ifp->if_flags |= XFS_IFEXTENTS;
 924	xfs_iext_add(ifp, 0, nextents);
 925	error = xfs_bmap_read_extents(tp, ip, whichfork);
 926	if (error) {
 927		xfs_iext_destroy(ifp);
 928		ifp->if_flags &= ~XFS_IFEXTENTS;
 929		return error;
 930	}
 931	xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 932	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 933}
 934
 935/*
 936 * Allocate an inode on disk and return a copy of its in-core version.
 937 * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
 938 * appropriately within the inode.  The uid and gid for the inode are
 939 * set according to the contents of the given cred structure.
 940 *
 941 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
 942 * has a free inode available, call xfs_iget()
 943 * to obtain the in-core version of the allocated inode.  Finally,
 944 * fill in the inode and log its initial contents.  In this case,
 945 * ialloc_context would be set to NULL and call_again set to false.
 946 *
 947 * If xfs_dialloc() does not have an available inode,
 948 * it will replenish its supply by doing an allocation. Since we can
 949 * only do one allocation within a transaction without deadlocks, we
 950 * must commit the current transaction before returning the inode itself.
 951 * In this case, therefore, we will set call_again to true and return.
 952 * The caller should then commit the current transaction, start a new
 953 * transaction, and call xfs_ialloc() again to actually get the inode.
 954 *
 955 * To ensure that some other process does not grab the inode that
 956 * was allocated during the first call to xfs_ialloc(), this routine
 957 * also returns the [locked] bp pointing to the head of the freelist
 958 * as ialloc_context.  The caller should hold this buffer across
 959 * the commit and pass it back into this routine on the second call.
 960 *
 961 * If we are allocating quota inodes, we do not have a parent inode
 962 * to attach to or associate with (i.e. pip == NULL) because they
 963 * are not linked into the directory structure - they are attached
 964 * directly to the superblock - and so have no parent.
 965 */
 966int
 967xfs_ialloc(
 968	xfs_trans_t	*tp,
 969	xfs_inode_t	*pip,
 970	mode_t		mode,
 971	xfs_nlink_t	nlink,
 972	xfs_dev_t	rdev,
 973	prid_t		prid,
 974	int		okalloc,
 975	xfs_buf_t	**ialloc_context,
 976	boolean_t	*call_again,
 977	xfs_inode_t	**ipp)
 978{
 979	xfs_ino_t	ino;
 980	xfs_inode_t	*ip;
 981	uint		flags;
 982	int		error;
 983	timespec_t	tv;
 984	int		filestreams = 0;
 
 
 
 
 
 985
 986	/*
 987	 * Call the space management code to pick
 988	 * the on-disk inode to be allocated.
 989	 */
 990	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
 991			    ialloc_context, call_again, &ino);
 992	if (error)
 993		return error;
 994	if (*call_again || ino == NULLFSINO) {
 995		*ipp = NULL;
 996		return 0;
 997	}
 998	ASSERT(*ialloc_context == NULL);
 999
1000	/*
1001	 * Get the in-core inode with the lock held exclusively.
1002	 * This is because we're setting fields here we need
1003	 * to prevent others from looking at until we're done.
1004	 */
1005	error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1006			 XFS_ILOCK_EXCL, &ip);
1007	if (error)
1008		return error;
1009	ASSERT(ip != NULL);
1010
1011	ip->i_d.di_mode = (__uint16_t)mode;
1012	ip->i_d.di_onlink = 0;
1013	ip->i_d.di_nlink = nlink;
1014	ASSERT(ip->i_d.di_nlink == nlink);
1015	ip->i_d.di_uid = current_fsuid();
1016	ip->i_d.di_gid = current_fsgid();
1017	xfs_set_projid(ip, prid);
1018	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1019
1020	/*
1021	 * If the superblock version is up to where we support new format
1022	 * inodes and this is currently an old format inode, then change
1023	 * the inode version number now.  This way we only do the conversion
1024	 * here rather than here and in the flush/logging code.
1025	 */
1026	if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1027	    ip->i_d.di_version == 1) {
1028		ip->i_d.di_version = 2;
1029		/*
1030		 * We've already zeroed the old link count, the projid field,
1031		 * and the pad field.
1032		 */
1033	}
1034
1035	/*
1036	 * Project ids won't be stored on disk if we are using a version 1 inode.
 
 
1037	 */
1038	if ((prid != 0) && (ip->i_d.di_version == 1))
1039		xfs_bump_ino_vers2(tp, ip);
1040
1041	if (pip && XFS_INHERIT_GID(pip)) {
1042		ip->i_d.di_gid = pip->i_d.di_gid;
1043		if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
1044			ip->i_d.di_mode |= S_ISGID;
1045		}
1046	}
1047
1048	/*
1049	 * If the group ID of the new file does not match the effective group
1050	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1051	 * (and only if the irix_sgid_inherit compatibility variable is set).
1052	 */
1053	if ((irix_sgid_inherit) &&
1054	    (ip->i_d.di_mode & S_ISGID) &&
1055	    (!in_group_p((gid_t)ip->i_d.di_gid))) {
1056		ip->i_d.di_mode &= ~S_ISGID;
1057	}
1058
1059	ip->i_d.di_size = 0;
1060	ip->i_size = 0;
1061	ip->i_d.di_nextents = 0;
1062	ASSERT(ip->i_d.di_nblocks == 0);
1063
1064	nanotime(&tv);
1065	ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1066	ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1067	ip->i_d.di_atime = ip->i_d.di_mtime;
1068	ip->i_d.di_ctime = ip->i_d.di_mtime;
1069
1070	/*
1071	 * di_gen will have been taken care of in xfs_iread.
1072	 */
1073	ip->i_d.di_extsize = 0;
1074	ip->i_d.di_dmevmask = 0;
1075	ip->i_d.di_dmstate = 0;
1076	ip->i_d.di_flags = 0;
1077	flags = XFS_ILOG_CORE;
1078	switch (mode & S_IFMT) {
1079	case S_IFIFO:
1080	case S_IFCHR:
1081	case S_IFBLK:
1082	case S_IFSOCK:
1083		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1084		ip->i_df.if_u2.if_rdev = rdev;
1085		ip->i_df.if_flags = 0;
1086		flags |= XFS_ILOG_DEV;
1087		break;
1088	case S_IFREG:
1089		/*
1090		 * we can't set up filestreams until after the VFS inode
1091		 * is set up properly.
 
 
 
 
1092		 */
1093		if (pip && xfs_inode_is_filestream(pip))
1094			filestreams = 1;
1095		/* fall through */
1096	case S_IFDIR:
1097		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1098			uint	di_flags = 0;
1099
1100			if (S_ISDIR(mode)) {
1101				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1102					di_flags |= XFS_DIFLAG_RTINHERIT;
1103				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1104					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1105					ip->i_d.di_extsize = pip->i_d.di_extsize;
1106				}
1107			} else if (S_ISREG(mode)) {
1108				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1109					di_flags |= XFS_DIFLAG_REALTIME;
1110				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1111					di_flags |= XFS_DIFLAG_EXTSIZE;
1112					ip->i_d.di_extsize = pip->i_d.di_extsize;
1113				}
1114			}
1115			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1116			    xfs_inherit_noatime)
1117				di_flags |= XFS_DIFLAG_NOATIME;
1118			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1119			    xfs_inherit_nodump)
1120				di_flags |= XFS_DIFLAG_NODUMP;
1121			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1122			    xfs_inherit_sync)
1123				di_flags |= XFS_DIFLAG_SYNC;
1124			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1125			    xfs_inherit_nosymlinks)
1126				di_flags |= XFS_DIFLAG_NOSYMLINKS;
1127			if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1128				di_flags |= XFS_DIFLAG_PROJINHERIT;
1129			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1130			    xfs_inherit_nodefrag)
1131				di_flags |= XFS_DIFLAG_NODEFRAG;
1132			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1133				di_flags |= XFS_DIFLAG_FILESTREAM;
1134			ip->i_d.di_flags |= di_flags;
1135		}
1136		/* FALLTHROUGH */
1137	case S_IFLNK:
1138		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1139		ip->i_df.if_flags = XFS_IFEXTENTS;
1140		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1141		ip->i_df.if_u1.if_extents = NULL;
1142		break;
1143	default:
1144		ASSERT(0);
1145	}
1146	/*
1147	 * Attribute fork settings for new inode.
1148	 */
1149	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1150	ip->i_d.di_anextents = 0;
1151
1152	/*
1153	 * Log the new values stuffed into the inode.
 
 
1154	 */
1155	xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1156	xfs_trans_log_inode(tp, ip, flags);
1157
1158	/* now that we have an i_mode we can setup inode ops and unlock */
1159	xfs_setup_inode(ip);
1160
1161	/* now we have set up the vfs inode we can associate the filestream */
1162	if (filestreams) {
1163		error = xfs_filestream_associate(pip, ip);
1164		if (error < 0)
1165			return -error;
1166		if (!error)
1167			xfs_iflags_set(ip, XFS_IFILESTREAM);
1168	}
1169
1170	*ipp = ip;
1171	return 0;
 
 
 
 
 
 
 
 
1172}
1173
1174/*
1175 * Check to make sure that there are no blocks allocated to the
1176 * file beyond the size of the file.  We don't check this for
1177 * files with fixed size extents or real time extents, but we
1178 * at least do it for regular files.
1179 */
1180#ifdef DEBUG
1181STATIC void
1182xfs_isize_check(
1183	struct xfs_inode	*ip,
1184	xfs_fsize_t		isize)
1185{
1186	struct xfs_mount	*mp = ip->i_mount;
1187	xfs_fileoff_t		map_first;
1188	int			nimaps;
1189	xfs_bmbt_irec_t		imaps[2];
1190
1191	if (!S_ISREG(ip->i_d.di_mode))
1192		return;
1193
1194	if (XFS_IS_REALTIME_INODE(ip))
1195		return;
1196
1197	if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1198		return;
1199
1200	nimaps = 2;
1201	map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1202	/*
1203	 * The filesystem could be shutting down, so bmapi may return
1204	 * an error.
1205	 */
1206	if (xfs_bmapi(NULL, ip, map_first,
1207			 (XFS_B_TO_FSB(mp,
1208				       (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1209			  map_first),
1210			 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1211			 NULL))
1212	    return;
1213	ASSERT(nimaps == 1);
1214	ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1215}
1216#else	/* DEBUG */
1217#define xfs_isize_check(ip, isize)
1218#endif	/* DEBUG */
1219
1220/*
1221 * Free up the underlying blocks past new_size.  The new size must be smaller
1222 * than the current size.  This routine can be used both for the attribute and
1223 * data fork, and does not modify the inode size, which is left to the caller.
1224 *
1225 * The transaction passed to this routine must have made a permanent log
1226 * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1227 * given transaction and start new ones, so make sure everything involved in
1228 * the transaction is tidy before calling here.  Some transaction will be
1229 * returned to the caller to be committed.  The incoming transaction must
1230 * already include the inode, and both inode locks must be held exclusively.
1231 * The inode must also be "held" within the transaction.  On return the inode
1232 * will be "held" within the returned transaction.  This routine does NOT
1233 * require any disk space to be reserved for it within the transaction.
1234 *
1235 * If we get an error, we must return with the inode locked and linked into the
1236 * current transaction. This keeps things simple for the higher level code,
1237 * because it always knows that the inode is locked and held in the transaction
1238 * that returns to it whether errors occur or not.  We don't mark the inode
1239 * dirty on error so that transactions can be easily aborted if possible.
1240 */
1241int
1242xfs_itruncate_extents(
1243	struct xfs_trans	**tpp,
1244	struct xfs_inode	*ip,
1245	int			whichfork,
1246	xfs_fsize_t		new_size)
 
1247{
1248	struct xfs_mount	*mp = ip->i_mount;
1249	struct xfs_trans	*tp = *tpp;
1250	struct xfs_trans	*ntp;
1251	xfs_bmap_free_t		free_list;
1252	xfs_fsblock_t		first_block;
1253	xfs_fileoff_t		first_unmap_block;
1254	xfs_fileoff_t		last_block;
1255	xfs_filblks_t		unmap_len;
1256	int			committed;
1257	int			error = 0;
1258	int			done = 0;
1259
1260	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1261	ASSERT(new_size <= ip->i_size);
 
 
1262	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1263	ASSERT(ip->i_itemp != NULL);
1264	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1265	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1266
 
 
 
 
1267	/*
1268	 * Since it is possible for space to become allocated beyond
1269	 * the end of the file (in a crash where the space is allocated
1270	 * but the inode size is not yet updated), simply remove any
1271	 * blocks which show up between the new EOF and the maximum
1272	 * possible file size.  If the first block to be removed is
1273	 * beyond the maximum file size (ie it is the same as last_block),
1274	 * then there is nothing to do.
 
1275	 */
1276	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1277	last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1278	if (first_unmap_block == last_block)
1279		return 0;
 
1280
1281	ASSERT(first_unmap_block < last_block);
1282	unmap_len = last_block - first_unmap_block + 1;
1283	while (!done) {
1284		xfs_bmap_init(&free_list, &first_block);
1285		error = xfs_bunmapi(tp, ip,
1286				    first_unmap_block, unmap_len,
1287				    xfs_bmapi_aflag(whichfork),
1288				    XFS_ITRUNC_MAX_EXTENTS,
1289				    &first_block, &free_list,
1290				    &done);
1291		if (error)
1292			goto out_bmap_cancel;
1293
1294		/*
1295		 * Duplicate the transaction that has the permanent
1296		 * reservation and commit the old transaction.
1297		 */
1298		error = xfs_bmap_finish(&tp, &free_list, &committed);
1299		if (committed)
1300			xfs_trans_ijoin(tp, ip);
1301		if (error)
1302			goto out_bmap_cancel;
1303
1304		if (committed) {
1305			/*
1306			 * Mark the inode dirty so it will be logged and
1307			 * moved forward in the log as part of every commit.
1308			 */
1309			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1310		}
1311
1312		ntp = xfs_trans_dup(tp);
1313		error = xfs_trans_commit(tp, 0);
1314		tp = ntp;
1315
1316		xfs_trans_ijoin(tp, ip);
1317
 
 
 
 
1318		if (error)
1319			goto out;
1320
1321		/*
1322		 * Transaction commit worked ok so we can drop the extra ticket
1323		 * reference that we gained in xfs_trans_dup()
1324		 */
1325		xfs_log_ticket_put(tp->t_ticket);
1326		error = xfs_trans_reserve(tp, 0,
1327					XFS_ITRUNCATE_LOG_RES(mp), 0,
1328					XFS_TRANS_PERM_LOG_RES,
1329					XFS_ITRUNCATE_LOG_COUNT);
1330		if (error)
1331			goto out;
1332	}
1333
 
 
 
 
 
 
 
 
1334out:
1335	*tpp = tp;
1336	return error;
1337out_bmap_cancel:
1338	/*
1339	 * If the bunmapi call encounters an error, return to the caller where
1340	 * the transaction can be properly aborted.  We just need to make sure
1341	 * we're not holding any resources that we were not when we came in.
1342	 */
1343	xfs_bmap_cancel(&free_list);
1344	goto out;
1345}
1346
1347int
1348xfs_itruncate_data(
1349	struct xfs_trans	**tpp,
1350	struct xfs_inode	*ip,
1351	xfs_fsize_t		new_size)
 
 
 
 
1352{
1353	int			error;
1354
1355	trace_xfs_itruncate_data_start(ip, new_size);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1356
1357	/*
1358	 * The first thing we do is set the size to new_size permanently on
1359	 * disk.  This way we don't have to worry about anyone ever being able
1360	 * to look at the data being freed even in the face of a crash.
1361	 * What we're getting around here is the case where we free a block, it
1362	 * is allocated to another file, it is written to, and then we crash.
1363	 * If the new data gets written to the file but the log buffers
1364	 * containing the free and reallocation don't, then we'd end up with
1365	 * garbage in the blocks being freed.  As long as we make the new_size
1366	 * permanent before actually freeing any blocks it doesn't matter if
1367	 * they get written to.
1368	 */
1369	if (ip->i_d.di_nextents > 0) {
1370		/*
1371		 * If we are not changing the file size then do not update
1372		 * the on-disk file size - we may be called from
1373		 * xfs_inactive_free_eofblocks().  If we update the on-disk
1374		 * file size and then the system crashes before the contents
1375		 * of the file are flushed to disk then the files may be
1376		 * full of holes (ie NULL files bug).
1377		 */
1378		if (ip->i_size != new_size) {
1379			ip->i_d.di_size = new_size;
1380			ip->i_size = new_size;
1381			xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1382		}
1383	}
 
1384
1385	error = xfs_itruncate_extents(tpp, ip, XFS_DATA_FORK, new_size);
1386	if (error)
1387		return error;
 
 
 
 
 
 
 
 
 
1388
1389	/*
1390	 * If we are not changing the file size then do not update the on-disk
1391	 * file size - we may be called from xfs_inactive_free_eofblocks().
1392	 * If we update the on-disk file size and then the system crashes
1393	 * before the contents of the file are flushed to disk then the files
1394	 * may be full of holes (ie NULL files bug).
1395	 */
1396	xfs_isize_check(ip, new_size);
1397	if (ip->i_size != new_size) {
1398		ip->i_d.di_size = new_size;
1399		ip->i_size = new_size;
1400	}
1401
1402	ASSERT(new_size != 0 || ip->i_delayed_blks == 0);
1403	ASSERT(new_size != 0 || ip->i_d.di_nextents == 0);
1404
1405	/*
1406	 * Always re-log the inode so that our permanent transaction can keep
1407	 * on rolling it forward in the log.
 
1408	 */
1409	xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
 
1410
1411	trace_xfs_itruncate_data_end(ip, new_size);
 
 
 
 
 
 
 
 
 
 
1412	return 0;
 
 
 
 
 
 
1413}
1414
1415/*
1416 * This is called when the inode's link count goes to 0.
1417 * We place the on-disk inode on a list in the AGI.  It
1418 * will be pulled from this list when the inode is freed.
1419 */
1420int
1421xfs_iunlink(
1422	xfs_trans_t	*tp,
1423	xfs_inode_t	*ip)
1424{
1425	xfs_mount_t	*mp;
1426	xfs_agi_t	*agi;
1427	xfs_dinode_t	*dip;
1428	xfs_buf_t	*agibp;
1429	xfs_buf_t	*ibp;
1430	xfs_agino_t	agino;
1431	short		bucket_index;
1432	int		offset;
1433	int		error;
1434
1435	ASSERT(ip->i_d.di_nlink == 0);
1436	ASSERT(ip->i_d.di_mode != 0);
1437
1438	mp = tp->t_mountp;
1439
1440	/*
1441	 * Get the agi buffer first.  It ensures lock ordering
1442	 * on the list.
1443	 */
1444	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1445	if (error)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1446		return error;
1447	agi = XFS_BUF_TO_AGI(agibp);
1448
1449	/*
1450	 * Get the index into the agi hash table for the
1451	 * list this inode will go on.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1452	 */
1453	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1454	ASSERT(agino != 0);
1455	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1456	ASSERT(agi->agi_unlinked[bucket_index]);
1457	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1458
1459	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
 
 
1460		/*
1461		 * There is already another inode in the bucket we need
1462		 * to add ourselves to.  Add us at the front of the list.
1463		 * Here we put the head pointer into our next pointer,
1464		 * and then we fall through to point the head at us.
1465		 */
1466		error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1467		if (error)
1468			return error;
1469
1470		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1471		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1472		offset = ip->i_imap.im_boffset +
1473			offsetof(xfs_dinode_t, di_next_unlinked);
1474		xfs_trans_inode_buf(tp, ibp);
1475		xfs_trans_log_buf(tp, ibp, offset,
1476				  (offset + sizeof(xfs_agino_t) - 1));
1477		xfs_inobp_check(mp, ibp);
1478	}
1479
1480	/*
1481	 * Point the bucket head pointer at the inode being inserted.
1482	 */
1483	ASSERT(agino != 0);
1484	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1485	offset = offsetof(xfs_agi_t, agi_unlinked) +
1486		(sizeof(xfs_agino_t) * bucket_index);
1487	xfs_trans_log_buf(tp, agibp, offset,
1488			  (offset + sizeof(xfs_agino_t) - 1));
1489	return 0;
1490}
1491
1492/*
1493 * Pull the on-disk inode from the AGI unlinked list.
 
 
 
1494 */
1495STATIC int
1496xfs_iunlink_remove(
1497	xfs_trans_t	*tp,
1498	xfs_inode_t	*ip)
1499{
1500	xfs_ino_t	next_ino;
1501	xfs_mount_t	*mp;
1502	xfs_agi_t	*agi;
1503	xfs_dinode_t	*dip;
1504	xfs_buf_t	*agibp;
1505	xfs_buf_t	*ibp;
1506	xfs_agnumber_t	agno;
1507	xfs_agino_t	agino;
1508	xfs_agino_t	next_agino;
1509	xfs_buf_t	*last_ibp;
1510	xfs_dinode_t	*last_dip = NULL;
1511	short		bucket_index;
1512	int		offset, last_offset = 0;
1513	int		error;
1514
1515	mp = tp->t_mountp;
1516	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
 
 
 
 
1517
1518	/*
1519	 * Get the agi buffer first.  It ensures lock ordering
1520	 * on the list.
1521	 */
1522	error = xfs_read_agi(mp, tp, agno, &agibp);
1523	if (error)
1524		return error;
 
 
 
 
 
 
 
1525
1526	agi = XFS_BUF_TO_AGI(agibp);
 
 
 
 
 
 
1527
1528	/*
1529	 * Get the index into the agi hash table for the
1530	 * list this inode will go on.
 
 
 
 
1531	 */
1532	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1533	ASSERT(agino != 0);
1534	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1535	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1536	ASSERT(agi->agi_unlinked[bucket_index]);
1537
1538	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1539		/*
1540		 * We're at the head of the list.  Get the inode's
1541		 * on-disk buffer to see if there is anyone after us
1542		 * on the list.  Only modify our next pointer if it
1543		 * is not already NULLAGINO.  This saves us the overhead
1544		 * of dealing with the buffer when there is no need to
1545		 * change it.
1546		 */
1547		error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1548		if (error) {
1549			xfs_warn(mp, "%s: xfs_itobp() returned error %d.",
1550				__func__, error);
1551			return error;
1552		}
1553		next_agino = be32_to_cpu(dip->di_next_unlinked);
1554		ASSERT(next_agino != 0);
1555		if (next_agino != NULLAGINO) {
1556			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1557			offset = ip->i_imap.im_boffset +
1558				offsetof(xfs_dinode_t, di_next_unlinked);
1559			xfs_trans_inode_buf(tp, ibp);
1560			xfs_trans_log_buf(tp, ibp, offset,
1561					  (offset + sizeof(xfs_agino_t) - 1));
1562			xfs_inobp_check(mp, ibp);
1563		} else {
1564			xfs_trans_brelse(tp, ibp);
1565		}
1566		/*
1567		 * Point the bucket head pointer at the next inode.
1568		 */
1569		ASSERT(next_agino != 0);
1570		ASSERT(next_agino != agino);
1571		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1572		offset = offsetof(xfs_agi_t, agi_unlinked) +
1573			(sizeof(xfs_agino_t) * bucket_index);
1574		xfs_trans_log_buf(tp, agibp, offset,
1575				  (offset + sizeof(xfs_agino_t) - 1));
1576	} else {
1577		/*
1578		 * We need to search the list for the inode being freed.
1579		 */
1580		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1581		last_ibp = NULL;
1582		while (next_agino != agino) {
1583			/*
1584			 * If the last inode wasn't the one pointing to
1585			 * us, then release its buffer since we're not
1586			 * going to do anything with it.
1587			 */
1588			if (last_ibp != NULL) {
1589				xfs_trans_brelse(tp, last_ibp);
1590			}
1591			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1592			error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1593					    &last_ibp, &last_offset, 0);
1594			if (error) {
1595				xfs_warn(mp,
1596					"%s: xfs_inotobp() returned error %d.",
1597					__func__, error);
1598				return error;
1599			}
1600			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1601			ASSERT(next_agino != NULLAGINO);
1602			ASSERT(next_agino != 0);
1603		}
1604		/*
1605		 * Now last_ibp points to the buffer previous to us on
1606		 * the unlinked list.  Pull us from the list.
1607		 */
1608		error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1609		if (error) {
1610			xfs_warn(mp, "%s: xfs_itobp(2) returned error %d.",
1611				__func__, error);
1612			return error;
1613		}
1614		next_agino = be32_to_cpu(dip->di_next_unlinked);
1615		ASSERT(next_agino != 0);
1616		ASSERT(next_agino != agino);
1617		if (next_agino != NULLAGINO) {
1618			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1619			offset = ip->i_imap.im_boffset +
1620				offsetof(xfs_dinode_t, di_next_unlinked);
1621			xfs_trans_inode_buf(tp, ibp);
1622			xfs_trans_log_buf(tp, ibp, offset,
1623					  (offset + sizeof(xfs_agino_t) - 1));
1624			xfs_inobp_check(mp, ibp);
1625		} else {
1626			xfs_trans_brelse(tp, ibp);
1627		}
1628		/*
1629		 * Point the previous inode on the list to the next inode.
1630		 */
1631		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1632		ASSERT(next_agino != 0);
1633		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1634		xfs_trans_inode_buf(tp, last_ibp);
1635		xfs_trans_log_buf(tp, last_ibp, offset,
1636				  (offset + sizeof(xfs_agino_t) - 1));
1637		xfs_inobp_check(mp, last_ibp);
1638	}
1639	return 0;
 
 
1640}
1641
1642/*
1643 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1644 * inodes that are in memory - they all must be marked stale and attached to
1645 * the cluster buffer.
 
 
 
1646 */
1647STATIC void
1648xfs_ifree_cluster(
1649	xfs_inode_t	*free_ip,
1650	xfs_trans_t	*tp,
1651	xfs_ino_t	inum)
1652{
1653	xfs_mount_t		*mp = free_ip->i_mount;
1654	int			blks_per_cluster;
1655	int			nbufs;
1656	int			ninodes;
1657	int			i, j;
1658	xfs_daddr_t		blkno;
1659	xfs_buf_t		*bp;
1660	xfs_inode_t		*ip;
1661	xfs_inode_log_item_t	*iip;
1662	xfs_log_item_t		*lip;
1663	struct xfs_perag	*pag;
1664
1665	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1666	if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1667		blks_per_cluster = 1;
1668		ninodes = mp->m_sb.sb_inopblock;
1669		nbufs = XFS_IALLOC_BLOCKS(mp);
1670	} else {
1671		blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1672					mp->m_sb.sb_blocksize;
1673		ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1674		nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1675	}
1676
1677	for (j = 0; j < nbufs; j++, inum += ninodes) {
1678		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1679					 XFS_INO_TO_AGBNO(mp, inum));
1680
1681		/*
1682		 * We obtain and lock the backing buffer first in the process
1683		 * here, as we have to ensure that any dirty inode that we
1684		 * can't get the flush lock on is attached to the buffer.
1685		 * If we scan the in-memory inodes first, then buffer IO can
1686		 * complete before we get a lock on it, and hence we may fail
1687		 * to mark all the active inodes on the buffer stale.
1688		 */
1689		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1690					mp->m_bsize * blks_per_cluster,
1691					XBF_LOCK);
1692
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1693		/*
1694		 * Walk the inodes already attached to the buffer and mark them
1695		 * stale. These will all have the flush locks held, so an
1696		 * in-memory inode walk can't lock them. By marking them all
1697		 * stale first, we will not attempt to lock them in the loop
1698		 * below as the XFS_ISTALE flag will be set.
1699		 */
1700		lip = bp->b_fspriv;
1701		while (lip) {
1702			if (lip->li_type == XFS_LI_INODE) {
1703				iip = (xfs_inode_log_item_t *)lip;
1704				ASSERT(iip->ili_logged == 1);
1705				lip->li_cb = xfs_istale_done;
1706				xfs_trans_ail_copy_lsn(mp->m_ail,
1707							&iip->ili_flush_lsn,
1708							&iip->ili_item.li_lsn);
1709				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1710			}
1711			lip = lip->li_bio_list;
1712		}
1713
 
 
 
 
1714
 
1715		/*
1716		 * For each inode in memory attempt to add it to the inode
1717		 * buffer and set it up for being staled on buffer IO
1718		 * completion.  This is safe as we've locked out tail pushing
1719		 * and flushing by locking the buffer.
1720		 *
1721		 * We have already marked every inode that was part of a
1722		 * transaction stale above, which means there is no point in
1723		 * even trying to lock them.
1724		 */
1725		for (i = 0; i < ninodes; i++) {
1726retry:
1727			rcu_read_lock();
1728			ip = radix_tree_lookup(&pag->pag_ici_root,
1729					XFS_INO_TO_AGINO(mp, (inum + i)));
1730
1731			/* Inode not in memory, nothing to do */
1732			if (!ip) {
1733				rcu_read_unlock();
1734				continue;
1735			}
1736
1737			/*
1738			 * because this is an RCU protected lookup, we could
1739			 * find a recently freed or even reallocated inode
1740			 * during the lookup. We need to check under the
1741			 * i_flags_lock for a valid inode here. Skip it if it
1742			 * is not valid, the wrong inode or stale.
1743			 */
1744			spin_lock(&ip->i_flags_lock);
1745			if (ip->i_ino != inum + i ||
1746			    __xfs_iflags_test(ip, XFS_ISTALE)) {
1747				spin_unlock(&ip->i_flags_lock);
1748				rcu_read_unlock();
1749				continue;
1750			}
1751			spin_unlock(&ip->i_flags_lock);
1752
1753			/*
1754			 * Don't try to lock/unlock the current inode, but we
1755			 * _cannot_ skip the other inodes that we did not find
1756			 * in the list attached to the buffer and are not
1757			 * already marked stale. If we can't lock it, back off
1758			 * and retry.
1759			 */
1760			if (ip != free_ip &&
1761			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1762				rcu_read_unlock();
1763				delay(1);
1764				goto retry;
1765			}
1766			rcu_read_unlock();
1767
1768			xfs_iflock(ip);
1769			xfs_iflags_set(ip, XFS_ISTALE);
 
 
 
 
 
 
 
 
1770
1771			/*
1772			 * we don't need to attach clean inodes or those only
1773			 * with unlogged changes (which we throw away, anyway).
1774			 */
1775			iip = ip->i_itemp;
1776			if (!iip || xfs_inode_clean(ip)) {
1777				ASSERT(ip != free_ip);
1778				ip->i_update_core = 0;
1779				xfs_ifunlock(ip);
1780				xfs_iunlock(ip, XFS_ILOCK_EXCL);
1781				continue;
1782			}
1783
1784			iip->ili_last_fields = iip->ili_format.ilf_fields;
1785			iip->ili_format.ilf_fields = 0;
1786			iip->ili_logged = 1;
1787			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1788						&iip->ili_item.li_lsn);
1789
1790			xfs_buf_attach_iodone(bp, xfs_istale_done,
1791						  &iip->ili_item);
 
 
 
 
 
 
1792
1793			if (ip != free_ip)
1794				xfs_iunlock(ip, XFS_ILOCK_EXCL);
1795		}
 
 
 
 
 
 
 
 
 
1796
1797		xfs_trans_stale_inode_buf(tp, bp);
1798		xfs_trans_binval(tp, bp);
 
 
 
 
1799	}
1800
1801	xfs_perag_put(pag);
 
 
 
 
 
 
 
 
 
 
1802}
1803
1804/*
1805 * This is called to return an inode to the inode free list.
1806 * The inode should already be truncated to 0 length and have
1807 * no pages associated with it.  This routine also assumes that
1808 * the inode is already a part of the transaction.
1809 *
1810 * The on-disk copy of the inode will have been added to the list
1811 * of unlinked inodes in the AGI. We need to remove the inode from
1812 * that list atomically with respect to freeing it here.
1813 */
1814int
1815xfs_ifree(
1816	xfs_trans_t	*tp,
1817	xfs_inode_t	*ip,
1818	xfs_bmap_free_t	*flist)
1819{
 
 
 
 
1820	int			error;
1821	int			delete;
1822	xfs_ino_t		first_ino;
1823	xfs_dinode_t    	*dip;
1824	xfs_buf_t       	*ibp;
1825
1826	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1827	ASSERT(ip->i_d.di_nlink == 0);
1828	ASSERT(ip->i_d.di_nextents == 0);
1829	ASSERT(ip->i_d.di_anextents == 0);
1830	ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
1831	       (!S_ISREG(ip->i_d.di_mode)));
1832	ASSERT(ip->i_d.di_nblocks == 0);
1833
1834	/*
1835	 * Pull the on-disk inode from the AGI unlinked list.
1836	 */
1837	error = xfs_iunlink_remove(tp, ip);
1838	if (error != 0) {
1839		return error;
1840	}
1841
1842	error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1843	if (error != 0) {
1844		return error;
1845	}
1846	ip->i_d.di_mode = 0;		/* mark incore inode as free */
1847	ip->i_d.di_flags = 0;
1848	ip->i_d.di_dmevmask = 0;
1849	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
1850	ip->i_df.if_ext_max =
1851		XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
1852	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1853	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1854	/*
1855	 * Bump the generation count so no one will be confused
1856	 * by reincarnations of this inode.
1857	 */
1858	ip->i_d.di_gen++;
1859
1860	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 
 
1861
1862	error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
1863	if (error)
 
 
 
 
 
 
 
 
1864		return error;
 
1865
1866        /*
1867	* Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1868	* from picking up this inode when it is reclaimed (its incore state
1869	* initialzed but not flushed to disk yet). The in-core di_mode is
1870	* already cleared  and a corresponding transaction logged.
1871	* The hack here just synchronizes the in-core to on-disk
1872	* di_mode value in advance before the actual inode sync to disk.
1873	* This is OK because the inode is already unlinked and would never
1874	* change its di_mode again for this inode generation.
1875	* This is a temporary hack that would require a proper fix
1876	* in the future.
1877	*/
1878	dip->di_mode = 0;
1879
1880	if (delete) {
1881		xfs_ifree_cluster(ip, tp, first_ino);
1882	}
1883
1884	return 0;
 
 
 
 
 
 
 
1885}
1886
1887/*
1888 * Reallocate the space for if_broot based on the number of records
1889 * being added or deleted as indicated in rec_diff.  Move the records
1890 * and pointers in if_broot to fit the new size.  When shrinking this
1891 * will eliminate holes between the records and pointers created by
1892 * the caller.  When growing this will create holes to be filled in
1893 * by the caller.
1894 *
1895 * The caller must not request to add more records than would fit in
1896 * the on-disk inode root.  If the if_broot is currently NULL, then
1897 * if we adding records one will be allocated.  The caller must also
1898 * not request that the number of records go below zero, although
1899 * it can go to zero.
1900 *
1901 * ip -- the inode whose if_broot area is changing
1902 * ext_diff -- the change in the number of records, positive or negative,
1903 *	 requested for the if_broot array.
1904 */
1905void
1906xfs_iroot_realloc(
1907	xfs_inode_t		*ip,
1908	int			rec_diff,
1909	int			whichfork)
1910{
1911	struct xfs_mount	*mp = ip->i_mount;
1912	int			cur_max;
1913	xfs_ifork_t		*ifp;
1914	struct xfs_btree_block	*new_broot;
1915	int			new_max;
1916	size_t			new_size;
1917	char			*np;
1918	char			*op;
1919
1920	/*
1921	 * Handle the degenerate case quietly.
1922	 */
1923	if (rec_diff == 0) {
 
 
 
1924		return;
1925	}
1926
1927	ifp = XFS_IFORK_PTR(ip, whichfork);
1928	if (rec_diff > 0) {
1929		/*
1930		 * If there wasn't any memory allocated before, just
1931		 * allocate it now and get out.
1932		 */
1933		if (ifp->if_broot_bytes == 0) {
1934			new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
1935			ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1936			ifp->if_broot_bytes = (int)new_size;
1937			return;
 
 
 
 
 
 
 
 
 
 
 
1938		}
1939
1940		/*
1941		 * If there is already an existing if_broot, then we need
1942		 * to realloc() it and shift the pointers to their new
1943		 * location.  The records don't change location because
1944		 * they are kept butted up against the btree block header.
1945		 */
1946		cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1947		new_max = cur_max + rec_diff;
1948		new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1949		ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
1950				(size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
1951				KM_SLEEP | KM_NOFS);
1952		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1953						     ifp->if_broot_bytes);
1954		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1955						     (int)new_size);
1956		ifp->if_broot_bytes = (int)new_size;
1957		ASSERT(ifp->if_broot_bytes <=
1958			XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1959		memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
1960		return;
1961	}
 
1962
1963	/*
1964	 * rec_diff is less than 0.  In this case, we are shrinking the
1965	 * if_broot buffer.  It must already exist.  If we go to zero
1966	 * records, just get rid of the root and clear the status bit.
1967	 */
1968	ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
1969	cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1970	new_max = cur_max + rec_diff;
1971	ASSERT(new_max >= 0);
1972	if (new_max > 0)
1973		new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1974	else
1975		new_size = 0;
1976	if (new_size > 0) {
1977		new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1978		/*
1979		 * First copy over the btree block header.
1980		 */
1981		memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
1982	} else {
1983		new_broot = NULL;
1984		ifp->if_flags &= ~XFS_IFBROOT;
1985	}
1986
1987	/*
1988	 * Only copy the records and pointers if there are any.
 
 
 
1989	 */
1990	if (new_max > 0) {
1991		/*
1992		 * First copy the records.
1993		 */
1994		op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
1995		np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
1996		memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
1997
1998		/*
1999		 * Then copy the pointers.
2000		 */
2001		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2002						     ifp->if_broot_bytes);
2003		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2004						     (int)new_size);
2005		memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2006	}
2007	kmem_free(ifp->if_broot);
2008	ifp->if_broot = new_broot;
2009	ifp->if_broot_bytes = (int)new_size;
2010	ASSERT(ifp->if_broot_bytes <=
2011		XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2012	return;
2013}
2014
 
 
 
 
 
 
 
2015
2016/*
2017 * This is called when the amount of space needed for if_data
2018 * is increased or decreased.  The change in size is indicated by
2019 * the number of bytes that need to be added or deleted in the
2020 * byte_diff parameter.
2021 *
2022 * If the amount of space needed has decreased below the size of the
2023 * inline buffer, then switch to using the inline buffer.  Otherwise,
2024 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2025 * to what is needed.
2026 *
2027 * ip -- the inode whose if_data area is changing
2028 * byte_diff -- the change in the number of bytes, positive or negative,
2029 *	 requested for the if_data array.
2030 */
2031void
2032xfs_idata_realloc(
2033	xfs_inode_t	*ip,
2034	int		byte_diff,
2035	int		whichfork)
2036{
2037	xfs_ifork_t	*ifp;
2038	int		new_size;
2039	int		real_size;
 
 
 
 
 
 
 
2040
2041	if (byte_diff == 0) {
2042		return;
2043	}
2044
2045	ifp = XFS_IFORK_PTR(ip, whichfork);
2046	new_size = (int)ifp->if_bytes + byte_diff;
2047	ASSERT(new_size >= 0);
2048
2049	if (new_size == 0) {
2050		if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2051			kmem_free(ifp->if_u1.if_data);
2052		}
2053		ifp->if_u1.if_data = NULL;
2054		real_size = 0;
2055	} else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2056		/*
2057		 * If the valid extents/data can fit in if_inline_ext/data,
2058		 * copy them from the malloc'd vector and free it.
 
2059		 */
2060		if (ifp->if_u1.if_data == NULL) {
2061			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2062		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2063			ASSERT(ifp->if_real_bytes != 0);
2064			memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2065			      new_size);
2066			kmem_free(ifp->if_u1.if_data);
2067			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2068		}
2069		real_size = 0;
2070	} else {
 
 
2071		/*
2072		 * Stuck with malloc/realloc.
2073		 * For inline data, the underlying buffer must be
2074		 * a multiple of 4 bytes in size so that it can be
2075		 * logged and stay on word boundaries.  We enforce
2076		 * that here.
 
 
2077		 */
2078		real_size = roundup(new_size, 4);
2079		if (ifp->if_u1.if_data == NULL) {
2080			ASSERT(ifp->if_real_bytes == 0);
2081			ifp->if_u1.if_data = kmem_alloc(real_size,
2082							KM_SLEEP | KM_NOFS);
2083		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2084			/*
2085			 * Only do the realloc if the underlying size
2086			 * is really changing.
2087			 */
2088			if (ifp->if_real_bytes != real_size) {
2089				ifp->if_u1.if_data =
2090					kmem_realloc(ifp->if_u1.if_data,
2091							real_size,
2092							ifp->if_real_bytes,
2093							KM_SLEEP | KM_NOFS);
2094			}
2095		} else {
2096			ASSERT(ifp->if_real_bytes == 0);
2097			ifp->if_u1.if_data = kmem_alloc(real_size,
2098							KM_SLEEP | KM_NOFS);
2099			memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2100				ifp->if_bytes);
2101		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2102	}
2103	ifp->if_real_bytes = real_size;
2104	ifp->if_bytes = new_size;
2105	ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2106}
2107
2108void
2109xfs_idestroy_fork(
2110	xfs_inode_t	*ip,
2111	int		whichfork)
2112{
2113	xfs_ifork_t	*ifp;
2114
2115	ifp = XFS_IFORK_PTR(ip, whichfork);
2116	if (ifp->if_broot != NULL) {
2117		kmem_free(ifp->if_broot);
2118		ifp->if_broot = NULL;
2119	}
 
 
 
 
 
 
 
2120
2121	/*
2122	 * If the format is local, then we can't have an extents
2123	 * array so just look for an inline data array.  If we're
2124	 * not local then we may or may not have an extents list,
2125	 * so check and free it up if we do.
2126	 */
2127	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2128		if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2129		    (ifp->if_u1.if_data != NULL)) {
2130			ASSERT(ifp->if_real_bytes != 0);
2131			kmem_free(ifp->if_u1.if_data);
2132			ifp->if_u1.if_data = NULL;
2133			ifp->if_real_bytes = 0;
2134		}
2135	} else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2136		   ((ifp->if_flags & XFS_IFEXTIREC) ||
2137		    ((ifp->if_u1.if_extents != NULL) &&
2138		     (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2139		ASSERT(ifp->if_real_bytes != 0);
2140		xfs_iext_destroy(ifp);
2141	}
2142	ASSERT(ifp->if_u1.if_extents == NULL ||
2143	       ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2144	ASSERT(ifp->if_real_bytes == 0);
2145	if (whichfork == XFS_ATTR_FORK) {
2146		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2147		ip->i_afp = NULL;
2148	}
2149}
2150
2151/*
2152 * This is called to unpin an inode.  The caller must have the inode locked
2153 * in at least shared mode so that the buffer cannot be subsequently pinned
2154 * once someone is waiting for it to be unpinned.
2155 */
2156static void
2157xfs_iunpin_nowait(
2158	struct xfs_inode	*ip)
2159{
2160	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2161
2162	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2163
2164	/* Give the log a push to start the unpinning I/O */
2165	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
 
 
 
 
 
 
 
 
 
2166
 
 
 
 
 
 
 
 
2167}
2168
2169void
2170xfs_iunpin_wait(
2171	struct xfs_inode	*ip)
2172{
2173	if (xfs_ipincount(ip)) {
2174		xfs_iunpin_nowait(ip);
2175		wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2176	}
2177}
2178
2179/*
2180 * xfs_iextents_copy()
2181 *
2182 * This is called to copy the REAL extents (as opposed to the delayed
2183 * allocation extents) from the inode into the given buffer.  It
2184 * returns the number of bytes copied into the buffer.
2185 *
2186 * If there are no delayed allocation extents, then we can just
2187 * memcpy() the extents into the buffer.  Otherwise, we need to
2188 * examine each extent in turn and skip those which are delayed.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2189 */
2190int
2191xfs_iextents_copy(
2192	xfs_inode_t		*ip,
2193	xfs_bmbt_rec_t		*dp,
2194	int			whichfork)
2195{
2196	int			copied;
2197	int			i;
2198	xfs_ifork_t		*ifp;
2199	int			nrecs;
2200	xfs_fsblock_t		start_block;
2201
2202	ifp = XFS_IFORK_PTR(ip, whichfork);
2203	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2204	ASSERT(ifp->if_bytes > 0);
2205
2206	nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2207	XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2208	ASSERT(nrecs > 0);
2209
2210	/*
2211	 * There are some delayed allocation extents in the
2212	 * inode, so copy the extents one at a time and skip
2213	 * the delayed ones.  There must be at least one
2214	 * non-delayed extent.
2215	 */
2216	copied = 0;
2217	for (i = 0; i < nrecs; i++) {
2218		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2219		start_block = xfs_bmbt_get_startblock(ep);
2220		if (isnullstartblock(start_block)) {
2221			/*
2222			 * It's a delayed allocation extent, so skip it.
2223			 */
2224			continue;
2225		}
2226
2227		/* Translate to on disk format */
2228		put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2229		put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2230		dp++;
2231		copied++;
2232	}
2233	ASSERT(copied != 0);
2234	xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2235
2236	return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2237}
 
2238
2239/*
2240 * Each of the following cases stores data into the same region
2241 * of the on-disk inode, so only one of them can be valid at
2242 * any given time. While it is possible to have conflicting formats
2243 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2244 * in EXTENTS format, this can only happen when the fork has
2245 * changed formats after being modified but before being flushed.
2246 * In these cases, the format always takes precedence, because the
2247 * format indicates the current state of the fork.
2248 */
2249/*ARGSUSED*/
2250STATIC void
2251xfs_iflush_fork(
2252	xfs_inode_t		*ip,
2253	xfs_dinode_t		*dip,
2254	xfs_inode_log_item_t	*iip,
2255	int			whichfork,
2256	xfs_buf_t		*bp)
2257{
2258	char			*cp;
2259	xfs_ifork_t		*ifp;
2260	xfs_mount_t		*mp;
2261#ifdef XFS_TRANS_DEBUG
2262	int			first;
2263#endif
2264	static const short	brootflag[2] =
2265		{ XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2266	static const short	dataflag[2] =
2267		{ XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2268	static const short	extflag[2] =
2269		{ XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2270
2271	if (!iip)
2272		return;
2273	ifp = XFS_IFORK_PTR(ip, whichfork);
2274	/*
2275	 * This can happen if we gave up in iformat in an error path,
2276	 * for the attribute fork.
2277	 */
2278	if (!ifp) {
2279		ASSERT(whichfork == XFS_ATTR_FORK);
2280		return;
2281	}
2282	cp = XFS_DFORK_PTR(dip, whichfork);
2283	mp = ip->i_mount;
2284	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2285	case XFS_DINODE_FMT_LOCAL:
2286		if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2287		    (ifp->if_bytes > 0)) {
2288			ASSERT(ifp->if_u1.if_data != NULL);
2289			ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2290			memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2291		}
2292		break;
2293
2294	case XFS_DINODE_FMT_EXTENTS:
2295		ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2296		       !(iip->ili_format.ilf_fields & extflag[whichfork]));
2297		if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2298		    (ifp->if_bytes > 0)) {
2299			ASSERT(xfs_iext_get_ext(ifp, 0));
2300			ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2301			(void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2302				whichfork);
2303		}
2304		break;
2305
2306	case XFS_DINODE_FMT_BTREE:
2307		if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2308		    (ifp->if_broot_bytes > 0)) {
2309			ASSERT(ifp->if_broot != NULL);
2310			ASSERT(ifp->if_broot_bytes <=
2311			       (XFS_IFORK_SIZE(ip, whichfork) +
2312				XFS_BROOT_SIZE_ADJ));
2313			xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2314				(xfs_bmdr_block_t *)cp,
2315				XFS_DFORK_SIZE(dip, mp, whichfork));
2316		}
2317		break;
2318
2319	case XFS_DINODE_FMT_DEV:
2320		if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2321			ASSERT(whichfork == XFS_DATA_FORK);
2322			xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2323		}
2324		break;
2325
2326	case XFS_DINODE_FMT_UUID:
2327		if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2328			ASSERT(whichfork == XFS_DATA_FORK);
2329			memcpy(XFS_DFORK_DPTR(dip),
2330			       &ip->i_df.if_u2.if_uuid,
2331			       sizeof(uuid_t));
2332		}
2333		break;
2334
2335	default:
2336		ASSERT(0);
2337		break;
2338	}
2339}
2340
2341STATIC int
2342xfs_iflush_cluster(
2343	xfs_inode_t	*ip,
2344	xfs_buf_t	*bp)
2345{
2346	xfs_mount_t		*mp = ip->i_mount;
2347	struct xfs_perag	*pag;
2348	unsigned long		first_index, mask;
2349	unsigned long		inodes_per_cluster;
2350	int			ilist_size;
2351	xfs_inode_t		**ilist;
2352	xfs_inode_t		*iq;
2353	int			nr_found;
2354	int			clcount = 0;
2355	int			bufwasdelwri;
2356	int			i;
2357
2358	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 
 
 
 
 
 
2359
2360	inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2361	ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2362	ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2363	if (!ilist)
2364		goto out_put;
2365
2366	mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2367	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2368	rcu_read_lock();
2369	/* really need a gang lookup range call here */
2370	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2371					first_index, inodes_per_cluster);
2372	if (nr_found == 0)
2373		goto out_free;
2374
2375	for (i = 0; i < nr_found; i++) {
2376		iq = ilist[i];
2377		if (iq == ip)
2378			continue;
2379
2380		/*
2381		 * because this is an RCU protected lookup, we could find a
2382		 * recently freed or even reallocated inode during the lookup.
2383		 * We need to check under the i_flags_lock for a valid inode
2384		 * here. Skip it if it is not valid or the wrong inode.
2385		 */
2386		spin_lock(&ip->i_flags_lock);
2387		if (!ip->i_ino ||
2388		    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2389			spin_unlock(&ip->i_flags_lock);
2390			continue;
2391		}
2392		spin_unlock(&ip->i_flags_lock);
2393
2394		/*
2395		 * Do an un-protected check to see if the inode is dirty and
2396		 * is a candidate for flushing.  These checks will be repeated
2397		 * later after the appropriate locks are acquired.
2398		 */
2399		if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2400			continue;
 
 
 
2401
2402		/*
2403		 * Try to get locks.  If any are unavailable or it is pinned,
2404		 * then this inode cannot be flushed and is skipped.
2405		 */
 
 
2406
2407		if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
 
 
2408			continue;
2409		if (!xfs_iflock_nowait(iq)) {
2410			xfs_iunlock(iq, XFS_ILOCK_SHARED);
2411			continue;
2412		}
2413		if (xfs_ipincount(iq)) {
2414			xfs_ifunlock(iq);
2415			xfs_iunlock(iq, XFS_ILOCK_SHARED);
2416			continue;
2417		}
2418
2419		/*
2420		 * arriving here means that this inode can be flushed.  First
2421		 * re-check that it's dirty before flushing.
2422		 */
2423		if (!xfs_inode_clean(iq)) {
2424			int	error;
2425			error = xfs_iflush_int(iq, bp);
2426			if (error) {
2427				xfs_iunlock(iq, XFS_ILOCK_SHARED);
2428				goto cluster_corrupt_out;
2429			}
2430			clcount++;
2431		} else {
2432			xfs_ifunlock(iq);
2433		}
2434		xfs_iunlock(iq, XFS_ILOCK_SHARED);
2435	}
2436
2437	if (clcount) {
2438		XFS_STATS_INC(xs_icluster_flushcnt);
2439		XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2440	}
 
2441
2442out_free:
2443	rcu_read_unlock();
2444	kmem_free(ilist);
2445out_put:
2446	xfs_perag_put(pag);
2447	return 0;
 
 
 
 
 
 
 
 
 
2448
 
 
2449
2450cluster_corrupt_out:
2451	/*
2452	 * Corruption detected in the clustering loop.  Invalidate the
2453	 * inode buffer and shut down the filesystem.
 
 
 
2454	 */
2455	rcu_read_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2456	/*
2457	 * Clean up the buffer.  If it was B_DELWRI, just release it --
2458	 * brelse can handle it with no problems.  If not, shut down the
2459	 * filesystem before releasing the buffer.
2460	 */
2461	bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2462	if (bufwasdelwri)
2463		xfs_buf_relse(bp);
 
 
 
 
2464
2465	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2466
2467	if (!bufwasdelwri) {
2468		/*
2469		 * Just like incore_relse: if we have b_iodone functions,
2470		 * mark the buffer as an error and call them.  Otherwise
2471		 * mark it as stale and brelse.
2472		 */
2473		if (bp->b_iodone) {
2474			XFS_BUF_UNDONE(bp);
2475			XFS_BUF_STALE(bp);
2476			xfs_buf_ioerror(bp, EIO);
2477			xfs_buf_ioend(bp, 0);
2478		} else {
2479			XFS_BUF_STALE(bp);
2480			xfs_buf_relse(bp);
2481		}
2482	}
2483
2484	/*
2485	 * Unlocks the flush lock
2486	 */
2487	xfs_iflush_abort(iq);
2488	kmem_free(ilist);
2489	xfs_perag_put(pag);
2490	return XFS_ERROR(EFSCORRUPTED);
 
 
 
 
2491}
2492
2493/*
2494 * xfs_iflush() will write a modified inode's changes out to the
2495 * inode's on disk home.  The caller must have the inode lock held
2496 * in at least shared mode and the inode flush completion must be
2497 * active as well.  The inode lock will still be held upon return from
2498 * the call and the caller is free to unlock it.
2499 * The inode flush will be completed when the inode reaches the disk.
2500 * The flags indicate how the inode's buffer should be written out.
2501 */
2502int
2503xfs_iflush(
2504	xfs_inode_t		*ip,
2505	uint			flags)
2506{
2507	xfs_inode_log_item_t	*iip;
2508	xfs_buf_t		*bp;
2509	xfs_dinode_t		*dip;
2510	xfs_mount_t		*mp;
2511	int			error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2512
2513	XFS_STATS_INC(xs_iflush_count);
 
 
2514
2515	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2516	ASSERT(!completion_done(&ip->i_flush));
2517	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2518	       ip->i_d.di_nextents > ip->i_df.if_ext_max);
2519
2520	iip = ip->i_itemp;
2521	mp = ip->i_mount;
 
2522
2523	/*
2524	 * We can't flush the inode until it is unpinned, so wait for it if we
2525	 * are allowed to block.  We know no one new can pin it, because we are
2526	 * holding the inode lock shared and you need to hold it exclusively to
2527	 * pin the inode.
2528	 *
2529	 * If we are not allowed to block, force the log out asynchronously so
2530	 * that when we come back the inode will be unpinned. If other inodes
2531	 * in the same cluster are dirty, they will probably write the inode
2532	 * out for us if they occur after the log force completes.
2533	 */
2534	if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
2535		xfs_iunpin_nowait(ip);
2536		xfs_ifunlock(ip);
2537		return EAGAIN;
2538	}
2539	xfs_iunpin_wait(ip);
2540
2541	/*
2542	 * For stale inodes we cannot rely on the backing buffer remaining
2543	 * stale in cache for the remaining life of the stale inode and so
2544	 * xfs_itobp() below may give us a buffer that no longer contains
2545	 * inodes below. We have to check this after ensuring the inode is
2546	 * unpinned so that it is safe to reclaim the stale inode after the
2547	 * flush call.
2548	 */
2549	if (xfs_iflags_test(ip, XFS_ISTALE)) {
2550		xfs_ifunlock(ip);
2551		return 0;
2552	}
2553
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2554	/*
2555	 * This may have been unpinned because the filesystem is shutting
2556	 * down forcibly. If that's the case we must not write this inode
2557	 * to disk, because the log record didn't make it to disk!
2558	 */
2559	if (XFS_FORCED_SHUTDOWN(mp)) {
2560		ip->i_update_core = 0;
2561		if (iip)
2562			iip->ili_format.ilf_fields = 0;
2563		xfs_ifunlock(ip);
2564		return XFS_ERROR(EIO);
2565	}
2566
2567	/*
2568	 * Get the buffer containing the on-disk inode.
2569	 */
2570	error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2571				(flags & SYNC_TRYLOCK) ? XBF_TRYLOCK : XBF_LOCK);
2572	if (error || !bp) {
2573		xfs_ifunlock(ip);
2574		return error;
2575	}
2576
2577	/*
2578	 * First flush out the inode that xfs_iflush was called with.
 
 
 
2579	 */
2580	error = xfs_iflush_int(ip, bp);
2581	if (error)
2582		goto corrupt_out;
2583
2584	/*
2585	 * If the buffer is pinned then push on the log now so we won't
2586	 * get stuck waiting in the write for too long.
2587	 */
2588	if (xfs_buf_ispinned(bp))
2589		xfs_log_force(mp, 0);
 
 
 
 
 
 
2590
2591	/*
2592	 * inode clustering:
2593	 * see if other inodes can be gathered into this write
 
2594	 */
2595	error = xfs_iflush_cluster(ip, bp);
2596	if (error)
2597		goto cluster_corrupt_out;
 
 
2598
2599	if (flags & SYNC_WAIT)
2600		error = xfs_bwrite(mp, bp);
2601	else
2602		xfs_bdwrite(mp, bp);
2603	return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2604
2605corrupt_out:
2606	xfs_buf_relse(bp);
2607	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2608cluster_corrupt_out:
2609	/*
2610	 * Unlocks the flush lock
 
2611	 */
2612	xfs_iflush_abort(ip);
2613	return XFS_ERROR(EFSCORRUPTED);
2614}
 
2615
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2616
2617STATIC int
2618xfs_iflush_int(
2619	xfs_inode_t		*ip,
2620	xfs_buf_t		*bp)
2621{
2622	xfs_inode_log_item_t	*iip;
2623	xfs_dinode_t		*dip;
2624	xfs_mount_t		*mp;
2625#ifdef XFS_TRANS_DEBUG
2626	int			first;
2627#endif
2628
2629	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2630	ASSERT(!completion_done(&ip->i_flush));
2631	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2632	       ip->i_d.di_nextents > ip->i_df.if_ext_max);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2633
2634	iip = ip->i_itemp;
2635	mp = ip->i_mount;
 
 
 
 
 
 
 
2636
2637	/* set *dip = inode's place in the buffer */
2638	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
 
 
 
2639
2640	/*
2641	 * Clear i_update_core before copying out the data.
2642	 * This is for coordination with our timestamp updates
2643	 * that don't hold the inode lock. They will always
2644	 * update the timestamps BEFORE setting i_update_core,
2645	 * so if we clear i_update_core after they set it we
2646	 * are guaranteed to see their updates to the timestamps.
2647	 * I believe that this depends on strongly ordered memory
2648	 * semantics, but we have that.  We use the SYNCHRONIZE
2649	 * macro to make sure that the compiler does not reorder
2650	 * the i_update_core access below the data copy below.
2651	 */
2652	ip->i_update_core = 0;
2653	SYNCHRONIZE();
2654
2655	/*
2656	 * Make sure to get the latest timestamps from the Linux inode.
 
 
 
2657	 */
2658	xfs_synchronize_times(ip);
2659
2660	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2661			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2662		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2663			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2664			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2665		goto corrupt_out;
2666	}
2667	if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2668				mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2669		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2670			"%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2671			__func__, ip->i_ino, ip, ip->i_d.di_magic);
2672		goto corrupt_out;
2673	}
2674	if (S_ISREG(ip->i_d.di_mode)) {
2675		if (XFS_TEST_ERROR(
2676		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2677		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2678		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2679			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2680				"%s: Bad regular inode %Lu, ptr 0x%p",
2681				__func__, ip->i_ino, ip);
2682			goto corrupt_out;
2683		}
2684	} else if (S_ISDIR(ip->i_d.di_mode)) {
2685		if (XFS_TEST_ERROR(
2686		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2687		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2688		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2689		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2690			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2691				"%s: Bad directory inode %Lu, ptr 0x%p",
2692				__func__, ip->i_ino, ip);
2693			goto corrupt_out;
2694		}
2695	}
2696	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2697				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2698				XFS_RANDOM_IFLUSH_5)) {
2699		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2700			"%s: detected corrupt incore inode %Lu, "
2701			"total extents = %d, nblocks = %Ld, ptr 0x%p",
2702			__func__, ip->i_ino,
2703			ip->i_d.di_nextents + ip->i_d.di_anextents,
2704			ip->i_d.di_nblocks, ip);
2705		goto corrupt_out;
2706	}
2707	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2708				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2709		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2710			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2711			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2712		goto corrupt_out;
2713	}
 
2714	/*
2715	 * bump the flush iteration count, used to detect flushes which
2716	 * postdate a log record during recovery.
 
 
 
 
2717	 */
 
 
2718
2719	ip->i_d.di_flushiter++;
 
 
 
 
 
 
 
 
 
 
2720
2721	/*
2722	 * Copy the dirty parts of the inode into the on-disk
2723	 * inode.  We always copy out the core of the inode,
2724	 * because if the inode is dirty at all the core must
2725	 * be.
2726	 */
2727	xfs_dinode_to_disk(dip, &ip->i_d);
2728
2729	/* Wrap, we never let the log put out DI_MAX_FLUSH */
2730	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2731		ip->i_d.di_flushiter = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2732
2733	/*
2734	 * If this is really an old format inode and the superblock version
2735	 * has not been updated to support only new format inodes, then
2736	 * convert back to the old inode format.  If the superblock version
2737	 * has been updated, then make the conversion permanent.
2738	 */
2739	ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2740	if (ip->i_d.di_version == 1) {
2741		if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2742			/*
2743			 * Convert it back.
2744			 */
2745			ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2746			dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2747		} else {
2748			/*
2749			 * The superblock version has already been bumped,
2750			 * so just make the conversion to the new inode
2751			 * format permanent.
2752			 */
2753			ip->i_d.di_version = 2;
2754			dip->di_version = 2;
2755			ip->i_d.di_onlink = 0;
2756			dip->di_onlink = 0;
2757			memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2758			memset(&(dip->di_pad[0]), 0,
2759			      sizeof(dip->di_pad));
2760			ASSERT(xfs_get_projid(ip) == 0);
2761		}
2762	}
2763
2764	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2765	if (XFS_IFORK_Q(ip))
2766		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2767	xfs_inobp_check(mp, bp);
2768
2769	/*
2770	 * We've recorded everything logged in the inode, so we'd
2771	 * like to clear the ilf_fields bits so we don't log and
2772	 * flush things unnecessarily.  However, we can't stop
2773	 * logging all this information until the data we've copied
2774	 * into the disk buffer is written to disk.  If we did we might
2775	 * overwrite the copy of the inode in the log with all the
2776	 * data after re-logging only part of it, and in the face of
2777	 * a crash we wouldn't have all the data we need to recover.
2778	 *
2779	 * What we do is move the bits to the ili_last_fields field.
2780	 * When logging the inode, these bits are moved back to the
2781	 * ilf_fields field.  In the xfs_iflush_done() routine we
2782	 * clear ili_last_fields, since we know that the information
2783	 * those bits represent is permanently on disk.  As long as
2784	 * the flush completes before the inode is logged again, then
2785	 * both ilf_fields and ili_last_fields will be cleared.
2786	 *
2787	 * We can play with the ilf_fields bits here, because the inode
2788	 * lock must be held exclusively in order to set bits there
2789	 * and the flush lock protects the ili_last_fields bits.
2790	 * Set ili_logged so the flush done
2791	 * routine can tell whether or not to look in the AIL.
2792	 * Also, store the current LSN of the inode so that we can tell
2793	 * whether the item has moved in the AIL from xfs_iflush_done().
2794	 * In order to read the lsn we need the AIL lock, because
2795	 * it is a 64 bit value that cannot be read atomically.
2796	 */
2797	if (iip != NULL && iip->ili_format.ilf_fields != 0) {
2798		iip->ili_last_fields = iip->ili_format.ilf_fields;
2799		iip->ili_format.ilf_fields = 0;
2800		iip->ili_logged = 1;
2801
2802		xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2803					&iip->ili_item.li_lsn);
 
 
 
 
 
2804
2805		/*
2806		 * Attach the function xfs_iflush_done to the inode's
2807		 * buffer.  This will remove the inode from the AIL
2808		 * and unlock the inode's flush lock when the inode is
2809		 * completely written to disk.
2810		 */
2811		xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
 
 
 
2812
2813		ASSERT(bp->b_fspriv != NULL);
2814		ASSERT(bp->b_iodone != NULL);
2815	} else {
2816		/*
2817		 * We're flushing an inode which is not in the AIL and has
2818		 * not been logged but has i_update_core set.  For this
2819		 * case we can use a B_DELWRI flush and immediately drop
2820		 * the inode flush lock because we can avoid the whole
2821		 * AIL state thing.  It's OK to drop the flush lock now,
2822		 * because we've already locked the buffer and to do anything
2823		 * you really need both.
2824		 */
2825		if (iip != NULL) {
2826			ASSERT(iip->ili_logged == 0);
2827			ASSERT(iip->ili_last_fields == 0);
2828			ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
 
2829		}
2830		xfs_ifunlock(ip);
2831	}
2832
2833	return 0;
 
 
 
 
 
 
 
 
 
 
 
2834
2835corrupt_out:
2836	return XFS_ERROR(EFSCORRUPTED);
2837}
 
 
 
 
 
 
 
 
 
 
 
2838
2839/*
2840 * Return a pointer to the extent record at file index idx.
2841 */
2842xfs_bmbt_rec_host_t *
2843xfs_iext_get_ext(
2844	xfs_ifork_t	*ifp,		/* inode fork pointer */
2845	xfs_extnum_t	idx)		/* index of target extent */
2846{
2847	ASSERT(idx >= 0);
2848	ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2849
2850	if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2851		return ifp->if_u1.if_ext_irec->er_extbuf;
2852	} else if (ifp->if_flags & XFS_IFEXTIREC) {
2853		xfs_ext_irec_t	*erp;		/* irec pointer */
2854		int		erp_idx = 0;	/* irec index */
2855		xfs_extnum_t	page_idx = idx;	/* ext index in target list */
 
 
 
2856
2857		erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
2858		return &erp->er_extbuf[page_idx];
2859	} else if (ifp->if_bytes) {
2860		return &ifp->if_u1.if_extents[idx];
2861	} else {
2862		return NULL;
 
 
 
 
 
 
 
 
 
2863	}
 
 
 
 
 
 
 
 
2864}
2865
2866/*
2867 * Insert new item(s) into the extent records for incore inode
2868 * fork 'ifp'.  'count' new items are inserted at index 'idx'.
2869 */
2870void
2871xfs_iext_insert(
2872	xfs_inode_t	*ip,		/* incore inode pointer */
2873	xfs_extnum_t	idx,		/* starting index of new items */
2874	xfs_extnum_t	count,		/* number of inserted items */
2875	xfs_bmbt_irec_t	*new,		/* items to insert */
2876	int		state)		/* type of extent conversion */
2877{
2878	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2879	xfs_extnum_t	i;		/* extent record index */
2880
2881	trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
2882
2883	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2884	xfs_iext_add(ifp, idx, count);
2885	for (i = idx; i < idx + count; i++, new++)
2886		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
2887}
2888
2889/*
2890 * This is called when the amount of space required for incore file
2891 * extents needs to be increased. The ext_diff parameter stores the
2892 * number of new extents being added and the idx parameter contains
2893 * the extent index where the new extents will be added. If the new
2894 * extents are being appended, then we just need to (re)allocate and
2895 * initialize the space. Otherwise, if the new extents are being
2896 * inserted into the middle of the existing entries, a bit more work
2897 * is required to make room for the new extents to be inserted. The
2898 * caller is responsible for filling in the new extent entries upon
2899 * return.
2900 */
2901void
2902xfs_iext_add(
2903	xfs_ifork_t	*ifp,		/* inode fork pointer */
2904	xfs_extnum_t	idx,		/* index to begin adding exts */
2905	int		ext_diff)	/* number of extents to add */
2906{
2907	int		byte_diff;	/* new bytes being added */
2908	int		new_size;	/* size of extents after adding */
2909	xfs_extnum_t	nextents;	/* number of extents in file */
2910
2911	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2912	ASSERT((idx >= 0) && (idx <= nextents));
2913	byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
2914	new_size = ifp->if_bytes + byte_diff;
2915	/*
2916	 * If the new number of extents (nextents + ext_diff)
2917	 * fits inside the inode, then continue to use the inline
2918	 * extent buffer.
2919	 */
2920	if (nextents + ext_diff <= XFS_INLINE_EXTS) {
2921		if (idx < nextents) {
2922			memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
2923				&ifp->if_u2.if_inline_ext[idx],
2924				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
2925			memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
2926		}
2927		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2928		ifp->if_real_bytes = 0;
2929	}
2930	/*
2931	 * Otherwise use a linear (direct) extent list.
2932	 * If the extents are currently inside the inode,
2933	 * xfs_iext_realloc_direct will switch us from
2934	 * inline to direct extent allocation mode.
2935	 */
2936	else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
2937		xfs_iext_realloc_direct(ifp, new_size);
2938		if (idx < nextents) {
2939			memmove(&ifp->if_u1.if_extents[idx + ext_diff],
2940				&ifp->if_u1.if_extents[idx],
2941				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
2942			memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
2943		}
2944	}
2945	/* Indirection array */
2946	else {
2947		xfs_ext_irec_t	*erp;
2948		int		erp_idx = 0;
2949		int		page_idx = idx;
2950
2951		ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
2952		if (ifp->if_flags & XFS_IFEXTIREC) {
2953			erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
2954		} else {
2955			xfs_iext_irec_init(ifp);
2956			ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2957			erp = ifp->if_u1.if_ext_irec;
2958		}
2959		/* Extents fit in target extent page */
2960		if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
2961			if (page_idx < erp->er_extcount) {
2962				memmove(&erp->er_extbuf[page_idx + ext_diff],
2963					&erp->er_extbuf[page_idx],
2964					(erp->er_extcount - page_idx) *
2965					sizeof(xfs_bmbt_rec_t));
2966				memset(&erp->er_extbuf[page_idx], 0, byte_diff);
2967			}
2968			erp->er_extcount += ext_diff;
2969			xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2970		}
2971		/* Insert a new extent page */
2972		else if (erp) {
2973			xfs_iext_add_indirect_multi(ifp,
2974				erp_idx, page_idx, ext_diff);
2975		}
2976		/*
2977		 * If extent(s) are being appended to the last page in
2978		 * the indirection array and the new extent(s) don't fit
2979		 * in the page, then erp is NULL and erp_idx is set to
2980		 * the next index needed in the indirection array.
2981		 */
2982		else {
2983			int	count = ext_diff;
2984
2985			while (count) {
2986				erp = xfs_iext_irec_new(ifp, erp_idx);
2987				erp->er_extcount = count;
2988				count -= MIN(count, (int)XFS_LINEAR_EXTS);
2989				if (count) {
2990					erp_idx++;
2991				}
2992			}
2993		}
2994	}
2995	ifp->if_bytes = new_size;
2996}
2997
2998/*
2999 * This is called when incore extents are being added to the indirection
3000 * array and the new extents do not fit in the target extent list. The
3001 * erp_idx parameter contains the irec index for the target extent list
3002 * in the indirection array, and the idx parameter contains the extent
3003 * index within the list. The number of extents being added is stored
3004 * in the count parameter.
3005 *
3006 *    |-------|   |-------|
3007 *    |       |   |       |    idx - number of extents before idx
3008 *    |  idx  |   | count |
3009 *    |       |   |       |    count - number of extents being inserted at idx
3010 *    |-------|   |-------|
3011 *    | count |   | nex2  |    nex2 - number of extents after idx + count
3012 *    |-------|   |-------|
3013 */
3014void
3015xfs_iext_add_indirect_multi(
3016	xfs_ifork_t	*ifp,			/* inode fork pointer */
3017	int		erp_idx,		/* target extent irec index */
3018	xfs_extnum_t	idx,			/* index within target list */
3019	int		count)			/* new extents being added */
3020{
3021	int		byte_diff;		/* new bytes being added */
3022	xfs_ext_irec_t	*erp;			/* pointer to irec entry */
3023	xfs_extnum_t	ext_diff;		/* number of extents to add */
3024	xfs_extnum_t	ext_cnt;		/* new extents still needed */
3025	xfs_extnum_t	nex2;			/* extents after idx + count */
3026	xfs_bmbt_rec_t	*nex2_ep = NULL;	/* temp list for nex2 extents */
3027	int		nlists;			/* number of irec's (lists) */
3028
3029	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3030	erp = &ifp->if_u1.if_ext_irec[erp_idx];
3031	nex2 = erp->er_extcount - idx;
3032	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3033
3034	/*
3035	 * Save second part of target extent list
3036	 * (all extents past */
3037	if (nex2) {
3038		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3039		nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3040		memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3041		erp->er_extcount -= nex2;
3042		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3043		memset(&erp->er_extbuf[idx], 0, byte_diff);
3044	}
3045
3046	/*
3047	 * Add the new extents to the end of the target
3048	 * list, then allocate new irec record(s) and
3049	 * extent buffer(s) as needed to store the rest
3050	 * of the new extents.
3051	 */
3052	ext_cnt = count;
3053	ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3054	if (ext_diff) {
3055		erp->er_extcount += ext_diff;
3056		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3057		ext_cnt -= ext_diff;
3058	}
3059	while (ext_cnt) {
3060		erp_idx++;
3061		erp = xfs_iext_irec_new(ifp, erp_idx);
3062		ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3063		erp->er_extcount = ext_diff;
3064		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3065		ext_cnt -= ext_diff;
3066	}
3067
3068	/* Add nex2 extents back to indirection array */
3069	if (nex2) {
3070		xfs_extnum_t	ext_avail;
3071		int		i;
3072
3073		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3074		ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3075		i = 0;
3076		/*
3077		 * If nex2 extents fit in the current page, append
3078		 * nex2_ep after the new extents.
3079		 */
3080		if (nex2 <= ext_avail) {
3081			i = erp->er_extcount;
3082		}
3083		/*
3084		 * Otherwise, check if space is available in the
3085		 * next page.
3086		 */
3087		else if ((erp_idx < nlists - 1) &&
3088			 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3089			  ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3090			erp_idx++;
3091			erp++;
3092			/* Create a hole for nex2 extents */
3093			memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3094				erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3095		}
3096		/*
3097		 * Final choice, create a new extent page for
3098		 * nex2 extents.
3099		 */
3100		else {
3101			erp_idx++;
3102			erp = xfs_iext_irec_new(ifp, erp_idx);
3103		}
3104		memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3105		kmem_free(nex2_ep);
3106		erp->er_extcount += nex2;
3107		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3108	}
3109}
3110
3111/*
3112 * This is called when the amount of space required for incore file
3113 * extents needs to be decreased. The ext_diff parameter stores the
3114 * number of extents to be removed and the idx parameter contains
3115 * the extent index where the extents will be removed from.
3116 *
3117 * If the amount of space needed has decreased below the linear
3118 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3119 * extent array.  Otherwise, use kmem_realloc() to adjust the
3120 * size to what is needed.
3121 */
3122void
3123xfs_iext_remove(
3124	xfs_inode_t	*ip,		/* incore inode pointer */
3125	xfs_extnum_t	idx,		/* index to begin removing exts */
3126	int		ext_diff,	/* number of extents to remove */
3127	int		state)		/* type of extent conversion */
3128{
3129	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3130	xfs_extnum_t	nextents;	/* number of extents in file */
3131	int		new_size;	/* size of extents after removal */
3132
3133	trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3134
3135	ASSERT(ext_diff > 0);
3136	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3137	new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3138
3139	if (new_size == 0) {
3140		xfs_iext_destroy(ifp);
3141	} else if (ifp->if_flags & XFS_IFEXTIREC) {
3142		xfs_iext_remove_indirect(ifp, idx, ext_diff);
3143	} else if (ifp->if_real_bytes) {
3144		xfs_iext_remove_direct(ifp, idx, ext_diff);
3145	} else {
3146		xfs_iext_remove_inline(ifp, idx, ext_diff);
3147	}
3148	ifp->if_bytes = new_size;
3149}
3150
3151/*
3152 * This removes ext_diff extents from the inline buffer, beginning
3153 * at extent index idx.
3154 */
3155void
3156xfs_iext_remove_inline(
3157	xfs_ifork_t	*ifp,		/* inode fork pointer */
3158	xfs_extnum_t	idx,		/* index to begin removing exts */
3159	int		ext_diff)	/* number of extents to remove */
3160{
3161	int		nextents;	/* number of extents in file */
3162
3163	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3164	ASSERT(idx < XFS_INLINE_EXTS);
3165	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3166	ASSERT(((nextents - ext_diff) > 0) &&
3167		(nextents - ext_diff) < XFS_INLINE_EXTS);
3168
3169	if (idx + ext_diff < nextents) {
3170		memmove(&ifp->if_u2.if_inline_ext[idx],
3171			&ifp->if_u2.if_inline_ext[idx + ext_diff],
3172			(nextents - (idx + ext_diff)) *
3173			 sizeof(xfs_bmbt_rec_t));
3174		memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3175			0, ext_diff * sizeof(xfs_bmbt_rec_t));
3176	} else {
3177		memset(&ifp->if_u2.if_inline_ext[idx], 0,
3178			ext_diff * sizeof(xfs_bmbt_rec_t));
3179	}
3180}
3181
3182/*
3183 * This removes ext_diff extents from a linear (direct) extent list,
3184 * beginning at extent index idx. If the extents are being removed
3185 * from the end of the list (ie. truncate) then we just need to re-
3186 * allocate the list to remove the extra space. Otherwise, if the
3187 * extents are being removed from the middle of the existing extent
3188 * entries, then we first need to move the extent records beginning
3189 * at idx + ext_diff up in the list to overwrite the records being
3190 * removed, then remove the extra space via kmem_realloc.
3191 */
3192void
3193xfs_iext_remove_direct(
3194	xfs_ifork_t	*ifp,		/* inode fork pointer */
3195	xfs_extnum_t	idx,		/* index to begin removing exts */
3196	int		ext_diff)	/* number of extents to remove */
3197{
3198	xfs_extnum_t	nextents;	/* number of extents in file */
3199	int		new_size;	/* size of extents after removal */
3200
3201	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3202	new_size = ifp->if_bytes -
3203		(ext_diff * sizeof(xfs_bmbt_rec_t));
3204	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3205
3206	if (new_size == 0) {
3207		xfs_iext_destroy(ifp);
3208		return;
 
 
 
 
 
 
3209	}
3210	/* Move extents up in the list (if needed) */
3211	if (idx + ext_diff < nextents) {
3212		memmove(&ifp->if_u1.if_extents[idx],
3213			&ifp->if_u1.if_extents[idx + ext_diff],
3214			(nextents - (idx + ext_diff)) *
3215			 sizeof(xfs_bmbt_rec_t));
3216	}
3217	memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3218		0, ext_diff * sizeof(xfs_bmbt_rec_t));
3219	/*
3220	 * Reallocate the direct extent list. If the extents
3221	 * will fit inside the inode then xfs_iext_realloc_direct
3222	 * will switch from direct to inline extent allocation
3223	 * mode for us.
3224	 */
3225	xfs_iext_realloc_direct(ifp, new_size);
3226	ifp->if_bytes = new_size;
3227}
3228
3229/*
3230 * This is called when incore extents are being removed from the
3231 * indirection array and the extents being removed span multiple extent
3232 * buffers. The idx parameter contains the file extent index where we
3233 * want to begin removing extents, and the count parameter contains
3234 * how many extents need to be removed.
3235 *
3236 *    |-------|   |-------|
3237 *    | nex1  |   |       |    nex1 - number of extents before idx
3238 *    |-------|   | count |
3239 *    |       |   |       |    count - number of extents being removed at idx
3240 *    | count |   |-------|
3241 *    |       |   | nex2  |    nex2 - number of extents after idx + count
3242 *    |-------|   |-------|
3243 */
3244void
3245xfs_iext_remove_indirect(
3246	xfs_ifork_t	*ifp,		/* inode fork pointer */
3247	xfs_extnum_t	idx,		/* index to begin removing extents */
3248	int		count)		/* number of extents to remove */
3249{
3250	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3251	int		erp_idx = 0;	/* indirection array index */
3252	xfs_extnum_t	ext_cnt;	/* extents left to remove */
3253	xfs_extnum_t	ext_diff;	/* extents to remove in current list */
3254	xfs_extnum_t	nex1;		/* number of extents before idx */
3255	xfs_extnum_t	nex2;		/* extents after idx + count */
3256	int		page_idx = idx;	/* index in target extent list */
3257
3258	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3259	erp = xfs_iext_idx_to_irec(ifp,  &page_idx, &erp_idx, 0);
3260	ASSERT(erp != NULL);
3261	nex1 = page_idx;
3262	ext_cnt = count;
3263	while (ext_cnt) {
3264		nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3265		ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3266		/*
3267		 * Check for deletion of entire list;
3268		 * xfs_iext_irec_remove() updates extent offsets.
3269		 */
3270		if (ext_diff == erp->er_extcount) {
3271			xfs_iext_irec_remove(ifp, erp_idx);
3272			ext_cnt -= ext_diff;
3273			nex1 = 0;
3274			if (ext_cnt) {
3275				ASSERT(erp_idx < ifp->if_real_bytes /
3276					XFS_IEXT_BUFSZ);
3277				erp = &ifp->if_u1.if_ext_irec[erp_idx];
3278				nex1 = 0;
3279				continue;
3280			} else {
3281				break;
3282			}
3283		}
3284		/* Move extents up (if needed) */
3285		if (nex2) {
3286			memmove(&erp->er_extbuf[nex1],
3287				&erp->er_extbuf[nex1 + ext_diff],
3288				nex2 * sizeof(xfs_bmbt_rec_t));
3289		}
3290		/* Zero out rest of page */
3291		memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3292			((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3293		/* Update remaining counters */
3294		erp->er_extcount -= ext_diff;
3295		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3296		ext_cnt -= ext_diff;
3297		nex1 = 0;
3298		erp_idx++;
3299		erp++;
3300	}
3301	ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3302	xfs_iext_irec_compact(ifp);
3303}
3304
3305/*
3306 * Create, destroy, or resize a linear (direct) block of extents.
3307 */
3308void
3309xfs_iext_realloc_direct(
3310	xfs_ifork_t	*ifp,		/* inode fork pointer */
3311	int		new_size)	/* new size of extents */
3312{
3313	int		rnew_size;	/* real new size of extents */
3314
3315	rnew_size = new_size;
3316
3317	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3318		((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3319		 (new_size != ifp->if_real_bytes)));
3320
3321	/* Free extent records */
3322	if (new_size == 0) {
3323		xfs_iext_destroy(ifp);
3324	}
3325	/* Resize direct extent list and zero any new bytes */
3326	else if (ifp->if_real_bytes) {
3327		/* Check if extents will fit inside the inode */
3328		if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3329			xfs_iext_direct_to_inline(ifp, new_size /
3330				(uint)sizeof(xfs_bmbt_rec_t));
3331			ifp->if_bytes = new_size;
3332			return;
3333		}
3334		if (!is_power_of_2(new_size)){
3335			rnew_size = roundup_pow_of_two(new_size);
3336		}
3337		if (rnew_size != ifp->if_real_bytes) {
3338			ifp->if_u1.if_extents =
3339				kmem_realloc(ifp->if_u1.if_extents,
3340						rnew_size,
3341						ifp->if_real_bytes, KM_NOFS);
3342		}
3343		if (rnew_size > ifp->if_real_bytes) {
3344			memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3345				(uint)sizeof(xfs_bmbt_rec_t)], 0,
3346				rnew_size - ifp->if_real_bytes);
3347		}
3348	}
 
 
 
 
 
 
3349	/*
3350	 * Switch from the inline extent buffer to a direct
3351	 * extent list. Be sure to include the inline extent
3352	 * bytes in new_size.
3353	 */
3354	else {
3355		new_size += ifp->if_bytes;
3356		if (!is_power_of_2(new_size)) {
3357			rnew_size = roundup_pow_of_two(new_size);
3358		}
3359		xfs_iext_inline_to_direct(ifp, rnew_size);
3360	}
3361	ifp->if_real_bytes = rnew_size;
3362	ifp->if_bytes = new_size;
3363}
3364
3365/*
3366 * Switch from linear (direct) extent records to inline buffer.
 
3367 */
3368void
3369xfs_iext_direct_to_inline(
3370	xfs_ifork_t	*ifp,		/* inode fork pointer */
3371	xfs_extnum_t	nextents)	/* number of extents in file */
3372{
3373	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3374	ASSERT(nextents <= XFS_INLINE_EXTS);
3375	/*
3376	 * The inline buffer was zeroed when we switched
3377	 * from inline to direct extent allocation mode,
3378	 * so we don't need to clear it here.
3379	 */
3380	memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3381		nextents * sizeof(xfs_bmbt_rec_t));
3382	kmem_free(ifp->if_u1.if_extents);
3383	ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3384	ifp->if_real_bytes = 0;
 
 
 
 
 
 
 
3385}
3386
3387/*
3388 * Switch from inline buffer to linear (direct) extent records.
3389 * new_size should already be rounded up to the next power of 2
3390 * by the caller (when appropriate), so use new_size as it is.
3391 * However, since new_size may be rounded up, we can't update
3392 * if_bytes here. It is the caller's responsibility to update
3393 * if_bytes upon return.
3394 */
3395void
3396xfs_iext_inline_to_direct(
3397	xfs_ifork_t	*ifp,		/* inode fork pointer */
3398	int		new_size)	/* number of extents in file */
3399{
3400	ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3401	memset(ifp->if_u1.if_extents, 0, new_size);
3402	if (ifp->if_bytes) {
3403		memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3404			ifp->if_bytes);
3405		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3406			sizeof(xfs_bmbt_rec_t));
3407	}
3408	ifp->if_real_bytes = new_size;
3409}
3410
3411/*
3412 * Resize an extent indirection array to new_size bytes.
3413 */
3414STATIC void
3415xfs_iext_realloc_indirect(
3416	xfs_ifork_t	*ifp,		/* inode fork pointer */
3417	int		new_size)	/* new indirection array size */
3418{
3419	int		nlists;		/* number of irec's (ex lists) */
3420	int		size;		/* current indirection array size */
3421
3422	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3423	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3424	size = nlists * sizeof(xfs_ext_irec_t);
3425	ASSERT(ifp->if_real_bytes);
3426	ASSERT((new_size >= 0) && (new_size != size));
3427	if (new_size == 0) {
3428		xfs_iext_destroy(ifp);
3429	} else {
3430		ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3431			kmem_realloc(ifp->if_u1.if_ext_irec,
3432				new_size, size, KM_NOFS);
3433	}
3434}
3435
3436/*
3437 * Switch from indirection array to linear (direct) extent allocations.
3438 */
3439STATIC void
3440xfs_iext_indirect_to_direct(
3441	 xfs_ifork_t	*ifp)		/* inode fork pointer */
3442{
3443	xfs_bmbt_rec_host_t *ep;	/* extent record pointer */
3444	xfs_extnum_t	nextents;	/* number of extents in file */
3445	int		size;		/* size of file extents */
3446
3447	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3448	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3449	ASSERT(nextents <= XFS_LINEAR_EXTS);
3450	size = nextents * sizeof(xfs_bmbt_rec_t);
3451
3452	xfs_iext_irec_compact_pages(ifp);
3453	ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
 
3454
3455	ep = ifp->if_u1.if_ext_irec->er_extbuf;
3456	kmem_free(ifp->if_u1.if_ext_irec);
3457	ifp->if_flags &= ~XFS_IFEXTIREC;
3458	ifp->if_u1.if_extents = ep;
3459	ifp->if_bytes = size;
3460	if (nextents < XFS_LINEAR_EXTS) {
3461		xfs_iext_realloc_direct(ifp, size);
3462	}
3463}
3464
3465/*
3466 * Free incore file extents.
 
 
3467 */
3468void
3469xfs_iext_destroy(
3470	xfs_ifork_t	*ifp)		/* inode fork pointer */
 
3471{
3472	if (ifp->if_flags & XFS_IFEXTIREC) {
3473		int	erp_idx;
3474		int	nlists;
3475
3476		nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3477		for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3478			xfs_iext_irec_remove(ifp, erp_idx);
3479		}
3480		ifp->if_flags &= ~XFS_IFEXTIREC;
3481	} else if (ifp->if_real_bytes) {
3482		kmem_free(ifp->if_u1.if_extents);
3483	} else if (ifp->if_bytes) {
3484		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3485			sizeof(xfs_bmbt_rec_t));
3486	}
3487	ifp->if_u1.if_extents = NULL;
3488	ifp->if_real_bytes = 0;
3489	ifp->if_bytes = 0;
3490}
3491
3492/*
3493 * Return a pointer to the extent record for file system block bno.
3494 */
3495xfs_bmbt_rec_host_t *			/* pointer to found extent record */
3496xfs_iext_bno_to_ext(
3497	xfs_ifork_t	*ifp,		/* inode fork pointer */
3498	xfs_fileoff_t	bno,		/* block number to search for */
3499	xfs_extnum_t	*idxp)		/* index of target extent */
3500{
3501	xfs_bmbt_rec_host_t *base;	/* pointer to first extent */
3502	xfs_filblks_t	blockcount = 0;	/* number of blocks in extent */
3503	xfs_bmbt_rec_host_t *ep = NULL;	/* pointer to target extent */
3504	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
3505	int		high;		/* upper boundary in search */
3506	xfs_extnum_t	idx = 0;	/* index of target extent */
3507	int		low;		/* lower boundary in search */
3508	xfs_extnum_t	nextents;	/* number of file extents */
3509	xfs_fileoff_t	startoff = 0;	/* start offset of extent */
3510
3511	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3512	if (nextents == 0) {
3513		*idxp = 0;
3514		return NULL;
3515	}
3516	low = 0;
3517	if (ifp->if_flags & XFS_IFEXTIREC) {
3518		/* Find target extent list */
3519		int	erp_idx = 0;
3520		erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3521		base = erp->er_extbuf;
3522		high = erp->er_extcount - 1;
3523	} else {
3524		base = ifp->if_u1.if_extents;
3525		high = nextents - 1;
3526	}
3527	/* Binary search extent records */
3528	while (low <= high) {
3529		idx = (low + high) >> 1;
3530		ep = base + idx;
3531		startoff = xfs_bmbt_get_startoff(ep);
3532		blockcount = xfs_bmbt_get_blockcount(ep);
3533		if (bno < startoff) {
3534			high = idx - 1;
3535		} else if (bno >= startoff + blockcount) {
3536			low = idx + 1;
3537		} else {
3538			/* Convert back to file-based extent index */
3539			if (ifp->if_flags & XFS_IFEXTIREC) {
3540				idx += erp->er_extoff;
3541			}
3542			*idxp = idx;
3543			return ep;
3544		}
3545	}
3546	/* Convert back to file-based extent index */
3547	if (ifp->if_flags & XFS_IFEXTIREC) {
3548		idx += erp->er_extoff;
3549	}
3550	if (bno >= startoff + blockcount) {
3551		if (++idx == nextents) {
3552			ep = NULL;
3553		} else {
3554			ep = xfs_iext_get_ext(ifp, idx);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3555		}
3556	}
3557	*idxp = idx;
3558	return ep;
3559}
3560
3561/*
3562 * Return a pointer to the indirection array entry containing the
3563 * extent record for filesystem block bno. Store the index of the
3564 * target irec in *erp_idxp.
3565 */
3566xfs_ext_irec_t *			/* pointer to found extent record */
3567xfs_iext_bno_to_irec(
3568	xfs_ifork_t	*ifp,		/* inode fork pointer */
3569	xfs_fileoff_t	bno,		/* block number to search for */
3570	int		*erp_idxp)	/* irec index of target ext list */
3571{
3572	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
3573	xfs_ext_irec_t	*erp_next;	/* next indirection array entry */
3574	int		erp_idx;	/* indirection array index */
3575	int		nlists;		/* number of extent irec's (lists) */
3576	int		high;		/* binary search upper limit */
3577	int		low;		/* binary search lower limit */
3578
3579	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3580	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3581	erp_idx = 0;
3582	low = 0;
3583	high = nlists - 1;
3584	while (low <= high) {
3585		erp_idx = (low + high) >> 1;
3586		erp = &ifp->if_u1.if_ext_irec[erp_idx];
3587		erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3588		if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3589			high = erp_idx - 1;
3590		} else if (erp_next && bno >=
3591			   xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3592			low = erp_idx + 1;
3593		} else {
3594			break;
3595		}
 
 
 
 
3596	}
3597	*erp_idxp = erp_idx;
3598	return erp;
 
 
 
 
 
3599}
3600
3601/*
3602 * Return a pointer to the indirection array entry containing the
3603 * extent record at file extent index *idxp. Store the index of the
3604 * target irec in *erp_idxp and store the page index of the target
3605 * extent record in *idxp.
3606 */
3607xfs_ext_irec_t *
3608xfs_iext_idx_to_irec(
3609	xfs_ifork_t	*ifp,		/* inode fork pointer */
3610	xfs_extnum_t	*idxp,		/* extent index (file -> page) */
3611	int		*erp_idxp,	/* pointer to target irec */
3612	int		realloc)	/* new bytes were just added */
3613{
3614	xfs_ext_irec_t	*prev;		/* pointer to previous irec */
3615	xfs_ext_irec_t	*erp = NULL;	/* pointer to current irec */
3616	int		erp_idx;	/* indirection array index */
3617	int		nlists;		/* number of irec's (ex lists) */
3618	int		high;		/* binary search upper limit */
3619	int		low;		/* binary search lower limit */
3620	xfs_extnum_t	page_idx = *idxp; /* extent index in target list */
3621
3622	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3623	ASSERT(page_idx >= 0);
3624	ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3625	ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3626
3627	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3628	erp_idx = 0;
3629	low = 0;
3630	high = nlists - 1;
3631
3632	/* Binary search extent irec's */
3633	while (low <= high) {
3634		erp_idx = (low + high) >> 1;
3635		erp = &ifp->if_u1.if_ext_irec[erp_idx];
3636		prev = erp_idx > 0 ? erp - 1 : NULL;
3637		if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3638		     realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3639			high = erp_idx - 1;
3640		} else if (page_idx > erp->er_extoff + erp->er_extcount ||
3641			   (page_idx == erp->er_extoff + erp->er_extcount &&
3642			    !realloc)) {
3643			low = erp_idx + 1;
3644		} else if (page_idx == erp->er_extoff + erp->er_extcount &&
3645			   erp->er_extcount == XFS_LINEAR_EXTS) {
3646			ASSERT(realloc);
3647			page_idx = 0;
3648			erp_idx++;
3649			erp = erp_idx < nlists ? erp + 1 : NULL;
3650			break;
3651		} else {
3652			page_idx -= erp->er_extoff;
3653			break;
3654		}
 
 
 
 
 
3655	}
3656	*idxp = page_idx;
3657	*erp_idxp = erp_idx;
3658	return(erp);
3659}
3660
3661/*
3662 * Allocate and initialize an indirection array once the space needed
3663 * for incore extents increases above XFS_IEXT_BUFSZ.
3664 */
3665void
3666xfs_iext_irec_init(
3667	xfs_ifork_t	*ifp)		/* inode fork pointer */
 
 
 
3668{
3669	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3670	xfs_extnum_t	nextents;	/* number of extents in file */
3671
3672	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3673	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3674	ASSERT(nextents <= XFS_LINEAR_EXTS);
3675
3676	erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3677
3678	if (nextents == 0) {
3679		ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3680	} else if (!ifp->if_real_bytes) {
3681		xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3682	} else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3683		xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3684	}
3685	erp->er_extbuf = ifp->if_u1.if_extents;
3686	erp->er_extcount = nextents;
3687	erp->er_extoff = 0;
3688
3689	ifp->if_flags |= XFS_IFEXTIREC;
3690	ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3691	ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3692	ifp->if_u1.if_ext_irec = erp;
3693
3694	return;
 
 
 
3695}
3696
3697/*
3698 * Allocate and initialize a new entry in the indirection array.
3699 */
3700xfs_ext_irec_t *
3701xfs_iext_irec_new(
3702	xfs_ifork_t	*ifp,		/* inode fork pointer */
3703	int		erp_idx)	/* index for new irec */
3704{
3705	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3706	int		i;		/* loop counter */
3707	int		nlists;		/* number of irec's (ex lists) */
3708
3709	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3710	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
 
 
3711
3712	/* Resize indirection array */
3713	xfs_iext_realloc_indirect(ifp, ++nlists *
3714				  sizeof(xfs_ext_irec_t));
3715	/*
3716	 * Move records down in the array so the
3717	 * new page can use erp_idx.
3718	 */
3719	erp = ifp->if_u1.if_ext_irec;
3720	for (i = nlists - 1; i > erp_idx; i--) {
3721		memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3722	}
3723	ASSERT(i == erp_idx);
3724
3725	/* Initialize new extent record */
3726	erp = ifp->if_u1.if_ext_irec;
3727	erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3728	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3729	memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3730	erp[erp_idx].er_extcount = 0;
3731	erp[erp_idx].er_extoff = erp_idx > 0 ?
3732		erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3733	return (&erp[erp_idx]);
3734}
3735
3736/*
3737 * Remove a record from the indirection array.
3738 */
3739void
3740xfs_iext_irec_remove(
3741	xfs_ifork_t	*ifp,		/* inode fork pointer */
3742	int		erp_idx)	/* irec index to remove */
3743{
3744	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3745	int		i;		/* loop counter */
3746	int		nlists;		/* number of irec's (ex lists) */
3747
3748	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3749	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3750	erp = &ifp->if_u1.if_ext_irec[erp_idx];
3751	if (erp->er_extbuf) {
3752		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3753			-erp->er_extcount);
3754		kmem_free(erp->er_extbuf);
3755	}
3756	/* Compact extent records */
3757	erp = ifp->if_u1.if_ext_irec;
3758	for (i = erp_idx; i < nlists - 1; i++) {
3759		memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3760	}
3761	/*
3762	 * Manually free the last extent record from the indirection
3763	 * array.  A call to xfs_iext_realloc_indirect() with a size
3764	 * of zero would result in a call to xfs_iext_destroy() which
3765	 * would in turn call this function again, creating a nasty
3766	 * infinite loop.
3767	 */
3768	if (--nlists) {
3769		xfs_iext_realloc_indirect(ifp,
3770			nlists * sizeof(xfs_ext_irec_t));
3771	} else {
3772		kmem_free(ifp->if_u1.if_ext_irec);
3773	}
3774	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3775}
3776
3777/*
3778 * This is called to clean up large amounts of unused memory allocated
3779 * by the indirection array.  Before compacting anything though, verify
3780 * that the indirection array is still needed and switch back to the
3781 * linear extent list (or even the inline buffer) if possible.  The
3782 * compaction policy is as follows:
3783 *
3784 *    Full Compaction: Extents fit into a single page (or inline buffer)
3785 * Partial Compaction: Extents occupy less than 50% of allocated space
3786 *      No Compaction: Extents occupy at least 50% of allocated space
3787 */
3788void
3789xfs_iext_irec_compact(
3790	xfs_ifork_t	*ifp)		/* inode fork pointer */
3791{
3792	xfs_extnum_t	nextents;	/* number of extents in file */
3793	int		nlists;		/* number of irec's (ex lists) */
3794
3795	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3796	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3797	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3798
3799	if (nextents == 0) {
3800		xfs_iext_destroy(ifp);
3801	} else if (nextents <= XFS_INLINE_EXTS) {
3802		xfs_iext_indirect_to_direct(ifp);
3803		xfs_iext_direct_to_inline(ifp, nextents);
3804	} else if (nextents <= XFS_LINEAR_EXTS) {
3805		xfs_iext_indirect_to_direct(ifp);
3806	} else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3807		xfs_iext_irec_compact_pages(ifp);
3808	}
 
 
 
 
 
 
 
 
3809}
3810
3811/*
3812 * Combine extents from neighboring extent pages.
3813 */
3814void
3815xfs_iext_irec_compact_pages(
3816	xfs_ifork_t	*ifp)		/* inode fork pointer */
3817{
3818	xfs_ext_irec_t	*erp, *erp_next;/* pointers to irec entries */
3819	int		erp_idx = 0;	/* indirection array index */
3820	int		nlists;		/* number of irec's (ex lists) */
3821
3822	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3823	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3824	while (erp_idx < nlists - 1) {
3825		erp = &ifp->if_u1.if_ext_irec[erp_idx];
3826		erp_next = erp + 1;
3827		if (erp_next->er_extcount <=
3828		    (XFS_LINEAR_EXTS - erp->er_extcount)) {
3829			memcpy(&erp->er_extbuf[erp->er_extcount],
3830				erp_next->er_extbuf, erp_next->er_extcount *
3831				sizeof(xfs_bmbt_rec_t));
3832			erp->er_extcount += erp_next->er_extcount;
3833			/*
3834			 * Free page before removing extent record
3835			 * so er_extoffs don't get modified in
3836			 * xfs_iext_irec_remove.
3837			 */
3838			kmem_free(erp_next->er_extbuf);
3839			erp_next->er_extbuf = NULL;
3840			xfs_iext_irec_remove(ifp, erp_idx + 1);
3841			nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3842		} else {
3843			erp_idx++;
3844		}
3845	}
3846}
3847
3848/*
3849 * This is called to update the er_extoff field in the indirection
3850 * array when extents have been added or removed from one of the
3851 * extent lists. erp_idx contains the irec index to begin updating
3852 * at and ext_diff contains the number of extents that were added
3853 * or removed.
3854 */
3855void
3856xfs_iext_irec_update_extoffs(
3857	xfs_ifork_t	*ifp,		/* inode fork pointer */
3858	int		erp_idx,	/* irec index to update */
3859	int		ext_diff)	/* number of new extents */
3860{
3861	int		i;		/* loop counter */
3862	int		nlists;		/* number of irec's (ex lists */
3863
3864	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3865	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3866	for (i = erp_idx; i < nlists; i++) {
3867		ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
3868	}
3869}
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}