Linux Audio

Check our new training course

Loading...
v3.15
   1/*
   2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_shared.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_bit.h"
 
  25#include "xfs_inum.h"
 
  26#include "xfs_sb.h"
  27#include "xfs_ag.h"
  28#include "xfs_mount.h"
  29#include "xfs_da_format.h"
  30#include "xfs_da_btree.h"
  31#include "xfs_dir2.h"
 
 
 
 
 
  32#include "xfs_inode.h"
  33#include "xfs_btree.h"
  34#include "xfs_trans.h"
 
  35#include "xfs_inode_item.h"
  36#include "xfs_extfree_item.h"
  37#include "xfs_alloc.h"
  38#include "xfs_bmap.h"
  39#include "xfs_bmap_util.h"
  40#include "xfs_bmap_btree.h"
  41#include "xfs_rtalloc.h"
  42#include "xfs_error.h"
 
  43#include "xfs_quota.h"
  44#include "xfs_trans_space.h"
  45#include "xfs_buf_item.h"
  46#include "xfs_trace.h"
  47#include "xfs_symlink.h"
  48#include "xfs_attr_leaf.h"
  49#include "xfs_dinode.h"
  50#include "xfs_filestream.h"
 
 
  51
  52
  53kmem_zone_t		*xfs_bmap_free_item_zone;
  54
  55/*
  56 * Miscellaneous helper functions
  57 */
  58
  59/*
  60 * Compute and fill in the value of the maximum depth of a bmap btree
  61 * in this filesystem.  Done once, during mount.
  62 */
  63void
  64xfs_bmap_compute_maxlevels(
  65	xfs_mount_t	*mp,		/* file system mount structure */
  66	int		whichfork)	/* data or attr fork */
  67{
  68	int		level;		/* btree level */
  69	uint		maxblocks;	/* max blocks at this level */
  70	uint		maxleafents;	/* max leaf entries possible */
  71	int		maxrootrecs;	/* max records in root block */
  72	int		minleafrecs;	/* min records in leaf block */
  73	int		minnoderecs;	/* min records in node block */
  74	int		sz;		/* root block size */
  75
  76	/*
  77	 * The maximum number of extents in a file, hence the maximum
  78	 * number of leaf entries, is controlled by the type of di_nextents
  79	 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
  80	 * (a signed 16-bit number, xfs_aextnum_t).
  81	 *
  82	 * Note that we can no longer assume that if we are in ATTR1 that
  83	 * the fork offset of all the inodes will be
  84	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
  85	 * with ATTR2 and then mounted back with ATTR1, keeping the
  86	 * di_forkoff's fixed but probably at various positions. Therefore,
  87	 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
  88	 * of a minimum size available.
  89	 */
  90	if (whichfork == XFS_DATA_FORK) {
  91		maxleafents = MAXEXTNUM;
  92		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
  93	} else {
  94		maxleafents = MAXAEXTNUM;
  95		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
  96	}
  97	maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
  98	minleafrecs = mp->m_bmap_dmnr[0];
  99	minnoderecs = mp->m_bmap_dmnr[1];
 100	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
 101	for (level = 1; maxblocks > 1; level++) {
 102		if (maxblocks <= maxrootrecs)
 103			maxblocks = 1;
 104		else
 105			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
 106	}
 107	mp->m_bm_maxlevels[whichfork] = level;
 108}
 109
 110STATIC int				/* error */
 111xfs_bmbt_lookup_eq(
 112	struct xfs_btree_cur	*cur,
 113	xfs_fileoff_t		off,
 114	xfs_fsblock_t		bno,
 115	xfs_filblks_t		len,
 116	int			*stat)	/* success/failure */
 117{
 118	cur->bc_rec.b.br_startoff = off;
 119	cur->bc_rec.b.br_startblock = bno;
 120	cur->bc_rec.b.br_blockcount = len;
 121	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 122}
 123
 124STATIC int				/* error */
 125xfs_bmbt_lookup_ge(
 
 126	struct xfs_btree_cur	*cur,
 127	xfs_fileoff_t		off,
 128	xfs_fsblock_t		bno,
 129	xfs_filblks_t		len,
 130	int			*stat)	/* success/failure */
 131{
 132	cur->bc_rec.b.br_startoff = off;
 133	cur->bc_rec.b.br_startblock = bno;
 134	cur->bc_rec.b.br_blockcount = len;
 135	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 136}
 137
 138/*
 139 * Check if the inode needs to be converted to btree format.
 140 */
 141static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 142{
 143	return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 144		XFS_IFORK_NEXTENTS(ip, whichfork) >
 145			XFS_IFORK_MAXEXT(ip, whichfork);
 146}
 147
 148/*
 149 * Check if the inode should be converted to extent format.
 150 */
 151static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 152{
 153	return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 154		XFS_IFORK_NEXTENTS(ip, whichfork) <=
 155			XFS_IFORK_MAXEXT(ip, whichfork);
 156}
 
 157
 158/*
 159 * Update the record referred to by cur to the value given
 160 * by [off, bno, len, state].
 161 * This either works (return 0) or gets an EFSCORRUPTED error.
 162 */
 163STATIC int
 164xfs_bmbt_update(
 165	struct xfs_btree_cur	*cur,
 166	xfs_fileoff_t		off,
 167	xfs_fsblock_t		bno,
 168	xfs_filblks_t		len,
 169	xfs_exntst_t		state)
 170{
 171	union xfs_btree_rec	rec;
 172
 173	xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 174	return xfs_btree_update(cur, &rec);
 175}
 176
 177/*
 178 * Compute the worst-case number of indirect blocks that will be used
 179 * for ip's delayed extent of length "len".
 180 */
 181STATIC xfs_filblks_t
 182xfs_bmap_worst_indlen(
 183	xfs_inode_t	*ip,		/* incore inode pointer */
 184	xfs_filblks_t	len)		/* delayed extent length */
 185{
 186	int		level;		/* btree level number */
 187	int		maxrecs;	/* maximum record count at this level */
 188	xfs_mount_t	*mp;		/* mount structure */
 189	xfs_filblks_t	rval;		/* return value */
 190
 191	mp = ip->i_mount;
 192	maxrecs = mp->m_bmap_dmxr[0];
 193	for (level = 0, rval = 0;
 194	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
 195	     level++) {
 196		len += maxrecs - 1;
 197		do_div(len, maxrecs);
 198		rval += len;
 199		if (len == 1)
 200			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
 201				level - 1;
 202		if (level == 0)
 203			maxrecs = mp->m_bmap_dmxr[1];
 204	}
 205	return rval;
 206}
 207
 208/*
 209 * Calculate the default attribute fork offset for newly created inodes.
 
 
 
 210 */
 211uint
 212xfs_default_attroffset(
 213	struct xfs_inode	*ip)
 214{
 215	struct xfs_mount	*mp = ip->i_mount;
 216	uint			offset;
 217
 218	if (mp->m_sb.sb_inodesize == 256) {
 219		offset = XFS_LITINO(mp, ip->i_d.di_version) -
 220				XFS_BMDR_SPACE_CALC(MINABTPTRS);
 221	} else {
 222		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
 223	}
 224
 225	ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
 226	return offset;
 227}
 228
 229/*
 230 * Helper routine to reset inode di_forkoff field when switching
 231 * attribute fork from local to extent format - we reset it where
 232 * possible to make space available for inline data fork extents.
 233 */
 234STATIC void
 235xfs_bmap_forkoff_reset(
 236	xfs_mount_t	*mp,
 237	xfs_inode_t	*ip,
 238	int		whichfork)
 239{
 240	if (whichfork == XFS_ATTR_FORK &&
 241	    ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
 242	    ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
 243	    ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
 244		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
 245
 246		if (dfl_forkoff > ip->i_d.di_forkoff)
 247			ip->i_d.di_forkoff = dfl_forkoff;
 248	}
 249}
 250
 251/*
 252 * Debug/sanity checking code
 
 253 */
 254
 255STATIC int
 256xfs_bmap_sanity_check(
 257	struct xfs_mount	*mp,
 258	struct xfs_buf		*bp,
 259	int			level)
 260{
 261	struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 262
 263	if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
 264	    block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
 265		return 0;
 266
 267	if (be16_to_cpu(block->bb_level) != level ||
 268	    be16_to_cpu(block->bb_numrecs) == 0 ||
 269	    be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
 270		return 0;
 271
 272	return 1;
 273}
 274
 275#ifdef DEBUG
 276STATIC struct xfs_buf *
 277xfs_bmap_get_bp(
 278	struct xfs_btree_cur	*cur,
 279	xfs_fsblock_t		bno)
 280{
 281	struct xfs_log_item_desc *lidp;
 282	int			i;
 283
 284	if (!cur)
 285		return NULL;
 286
 287	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
 288		if (!cur->bc_bufs[i])
 289			break;
 290		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
 291			return cur->bc_bufs[i];
 292	}
 293
 294	/* Chase down all the log items to see if the bp is there */
 295	list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
 296		struct xfs_buf_log_item	*bip;
 297		bip = (struct xfs_buf_log_item *)lidp->lid_item;
 298		if (bip->bli_item.li_type == XFS_LI_BUF &&
 299		    XFS_BUF_ADDR(bip->bli_buf) == bno)
 300			return bip->bli_buf;
 301	}
 302
 303	return NULL;
 304}
 305
 306STATIC void
 307xfs_check_block(
 308	struct xfs_btree_block	*block,
 309	xfs_mount_t		*mp,
 310	int			root,
 311	short			sz)
 312{
 313	int			i, j, dmxr;
 314	__be64			*pp, *thispa;	/* pointer to block address */
 315	xfs_bmbt_key_t		*prevp, *keyp;
 316
 317	ASSERT(be16_to_cpu(block->bb_level) > 0);
 318
 319	prevp = NULL;
 320	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
 321		dmxr = mp->m_bmap_dmxr[0];
 322		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
 323
 324		if (prevp) {
 325			ASSERT(be64_to_cpu(prevp->br_startoff) <
 326			       be64_to_cpu(keyp->br_startoff));
 327		}
 328		prevp = keyp;
 329
 330		/*
 331		 * Compare the block numbers to see if there are dups.
 332		 */
 333		if (root)
 334			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
 335		else
 336			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
 337
 338		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
 339			if (root)
 340				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
 341			else
 342				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
 343			if (*thispa == *pp) {
 344				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
 345					__func__, j, i,
 346					(unsigned long long)be64_to_cpu(*thispa));
 347				panic("%s: ptrs are equal in node\n",
 348					__func__);
 349			}
 350		}
 351	}
 352}
 353
 354/*
 355 * Check that the extents for the inode ip are in the right order in all
 356 * btree leaves.
 
 357 */
 358
 359STATIC void
 360xfs_bmap_check_leaf_extents(
 361	xfs_btree_cur_t		*cur,	/* btree cursor or null */
 362	xfs_inode_t		*ip,		/* incore inode pointer */
 363	int			whichfork)	/* data or attr fork */
 364{
 365	struct xfs_btree_block	*block;	/* current btree block */
 366	xfs_fsblock_t		bno;	/* block # of "block" */
 367	xfs_buf_t		*bp;	/* buffer for "block" */
 368	int			error;	/* error return value */
 369	xfs_extnum_t		i=0, j;	/* index into the extents list */
 370	xfs_ifork_t		*ifp;	/* fork structure */
 371	int			level;	/* btree level, for checking */
 372	xfs_mount_t		*mp;	/* file system mount structure */
 373	__be64			*pp;	/* pointer to block address */
 374	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
 375	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
 376	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
 377	int			bp_release = 0;
 378
 379	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
 380		return;
 381	}
 382
 383	bno = NULLFSBLOCK;
 384	mp = ip->i_mount;
 385	ifp = XFS_IFORK_PTR(ip, whichfork);
 386	block = ifp->if_broot;
 387	/*
 388	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 389	 */
 390	level = be16_to_cpu(block->bb_level);
 391	ASSERT(level > 0);
 392	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
 393	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
 394	bno = be64_to_cpu(*pp);
 395
 396	ASSERT(bno != NULLDFSBNO);
 397	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
 398	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
 399
 400	/*
 401	 * Go down the tree until leaf level is reached, following the first
 402	 * pointer (leftmost) at each level.
 403	 */
 404	while (level-- > 0) {
 405		/* See if buf is in cur first */
 406		bp_release = 0;
 407		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 408		if (!bp) {
 409			bp_release = 1;
 410			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 411						XFS_BMAP_BTREE_REF,
 412						&xfs_bmbt_buf_ops);
 413			if (error)
 414				goto error_norelse;
 415		}
 416		block = XFS_BUF_TO_BLOCK(bp);
 417		XFS_WANT_CORRUPTED_GOTO(
 418			xfs_bmap_sanity_check(mp, bp, level),
 419			error0);
 420		if (level == 0)
 421			break;
 422
 423		/*
 424		 * Check this block for basic sanity (increasing keys and
 425		 * no duplicate blocks).
 426		 */
 427
 428		xfs_check_block(block, mp, 0, 0);
 429		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
 430		bno = be64_to_cpu(*pp);
 431		XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
 432		if (bp_release) {
 433			bp_release = 0;
 434			xfs_trans_brelse(NULL, bp);
 435		}
 436	}
 437
 438	/*
 439	 * Here with bp and block set to the leftmost leaf node in the tree.
 440	 */
 441	i = 0;
 442
 443	/*
 444	 * Loop over all leaf nodes checking that all extents are in the right order.
 445	 */
 446	for (;;) {
 447		xfs_fsblock_t	nextbno;
 448		xfs_extnum_t	num_recs;
 449
 450
 451		num_recs = xfs_btree_get_numrecs(block);
 452
 453		/*
 454		 * Read-ahead the next leaf block, if any.
 455		 */
 456
 457		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
 458
 459		/*
 460		 * Check all the extents to make sure they are OK.
 461		 * If we had a previous block, the last entry should
 462		 * conform with the first entry in this one.
 463		 */
 464
 465		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
 466		if (i) {
 467			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
 468			       xfs_bmbt_disk_get_blockcount(&last) <=
 469			       xfs_bmbt_disk_get_startoff(ep));
 470		}
 471		for (j = 1; j < num_recs; j++) {
 472			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
 473			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
 474			       xfs_bmbt_disk_get_blockcount(ep) <=
 475			       xfs_bmbt_disk_get_startoff(nextp));
 476			ep = nextp;
 477		}
 478
 479		last = *ep;
 480		i += num_recs;
 481		if (bp_release) {
 482			bp_release = 0;
 483			xfs_trans_brelse(NULL, bp);
 484		}
 485		bno = nextbno;
 486		/*
 487		 * If we've reached the end, stop.
 488		 */
 489		if (bno == NULLFSBLOCK)
 490			break;
 491
 492		bp_release = 0;
 493		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
 494		if (!bp) {
 495			bp_release = 1;
 496			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
 497						XFS_BMAP_BTREE_REF,
 498						&xfs_bmbt_buf_ops);
 499			if (error)
 500				goto error_norelse;
 501		}
 502		block = XFS_BUF_TO_BLOCK(bp);
 503	}
 504	if (bp_release) {
 505		bp_release = 0;
 506		xfs_trans_brelse(NULL, bp);
 507	}
 508	return;
 509
 510error0:
 511	xfs_warn(mp, "%s: at error0", __func__);
 512	if (bp_release)
 513		xfs_trans_brelse(NULL, bp);
 514error_norelse:
 515	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
 516		__func__, i);
 517	panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
 518	return;
 519}
 520
 521/*
 522 * Add bmap trace insert entries for all the contents of the extent records.
 
 
 
 
 523 */
 524void
 525xfs_bmap_trace_exlist(
 526	xfs_inode_t	*ip,		/* incore inode pointer */
 527	xfs_extnum_t	cnt,		/* count of entries in the list */
 528	int		whichfork,	/* data or attr fork */
 529	unsigned long	caller_ip)
 530{
 531	xfs_extnum_t	idx;		/* extent record index */
 532	xfs_ifork_t	*ifp;		/* inode fork pointer */
 533	int		state = 0;
 534
 535	if (whichfork == XFS_ATTR_FORK)
 536		state |= BMAP_ATTRFORK;
 537
 538	ifp = XFS_IFORK_PTR(ip, whichfork);
 539	ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
 540	for (idx = 0; idx < cnt; idx++)
 541		trace_xfs_extlist(ip, idx, whichfork, caller_ip);
 542}
 
 
 
 543
 
 544/*
 545 * Validate that the bmbt_irecs being returned from bmapi are valid
 546 * given the caller's original parameters.  Specifically check the
 547 * ranges of the returned irecs to ensure that they only extend beyond
 548 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
 549 */
 550STATIC void
 551xfs_bmap_validate_ret(
 552	xfs_fileoff_t		bno,
 553	xfs_filblks_t		len,
 554	int			flags,
 555	xfs_bmbt_irec_t		*mval,
 556	int			nmap,
 557	int			ret_nmap)
 558{
 559	int			i;		/* index to map values */
 560
 561	ASSERT(ret_nmap <= nmap);
 562
 563	for (i = 0; i < ret_nmap; i++) {
 564		ASSERT(mval[i].br_blockcount > 0);
 565		if (!(flags & XFS_BMAPI_ENTIRE)) {
 566			ASSERT(mval[i].br_startoff >= bno);
 567			ASSERT(mval[i].br_blockcount <= len);
 568			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
 569			       bno + len);
 570		} else {
 571			ASSERT(mval[i].br_startoff < bno + len);
 572			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
 573			       bno);
 574		}
 575		ASSERT(i == 0 ||
 576		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
 577		       mval[i].br_startoff);
 578		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
 579		       mval[i].br_startblock != HOLESTARTBLOCK);
 580		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
 581		       mval[i].br_state == XFS_EXT_UNWRITTEN);
 582	}
 583}
 584
 585#else
 586#define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
 587#define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 588#endif /* DEBUG */
 589
 590/*
 591 * bmap free list manipulation functions
 592 */
 
 
 
 
 
 593
 594/*
 595 * Add the extent to the list of extents to be free at transaction end.
 596 * The list is maintained sorted (by block number).
 597 */
 598void
 599xfs_bmap_add_free(
 600	xfs_fsblock_t		bno,		/* fs block number of extent */
 601	xfs_filblks_t		len,		/* length of extent */
 602	xfs_bmap_free_t		*flist,		/* list of extents */
 603	xfs_mount_t		*mp)		/* mount point structure */
 604{
 605	xfs_bmap_free_item_t	*cur;		/* current (next) element */
 606	xfs_bmap_free_item_t	*new;		/* new element */
 607	xfs_bmap_free_item_t	*prev;		/* previous element */
 608#ifdef DEBUG
 609	xfs_agnumber_t		agno;
 610	xfs_agblock_t		agbno;
 611
 612	ASSERT(bno != NULLFSBLOCK);
 613	ASSERT(len > 0);
 614	ASSERT(len <= MAXEXTLEN);
 615	ASSERT(!isnullstartblock(bno));
 616	agno = XFS_FSB_TO_AGNO(mp, bno);
 617	agbno = XFS_FSB_TO_AGBNO(mp, bno);
 618	ASSERT(agno < mp->m_sb.sb_agcount);
 619	ASSERT(agbno < mp->m_sb.sb_agblocks);
 620	ASSERT(len < mp->m_sb.sb_agblocks);
 621	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
 622#endif
 623	ASSERT(xfs_bmap_free_item_zone != NULL);
 624	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
 625	new->xbfi_startblock = bno;
 626	new->xbfi_blockcount = (xfs_extlen_t)len;
 627	for (prev = NULL, cur = flist->xbf_first;
 628	     cur != NULL;
 629	     prev = cur, cur = cur->xbfi_next) {
 630		if (cur->xbfi_startblock >= bno)
 631			break;
 632	}
 633	if (prev)
 634		prev->xbfi_next = new;
 635	else
 636		flist->xbf_first = new;
 637	new->xbfi_next = cur;
 638	flist->xbf_count++;
 639}
 640
 641/*
 642 * Remove the entry "free" from the free item list.  Prev points to the
 643 * previous entry, unless "free" is the head of the list.
 644 */
 645void
 646xfs_bmap_del_free(
 647	xfs_bmap_free_t		*flist,	/* free item list header */
 648	xfs_bmap_free_item_t	*prev,	/* previous item on list, if any */
 649	xfs_bmap_free_item_t	*free)	/* list item to be freed */
 650{
 651	if (prev)
 652		prev->xbfi_next = free->xbfi_next;
 653	else
 654		flist->xbf_first = free->xbfi_next;
 655	flist->xbf_count--;
 656	kmem_zone_free(xfs_bmap_free_item_zone, free);
 657}
 658
 659/*
 660 * Free up any items left in the list.
 661 */
 662void
 663xfs_bmap_cancel(
 664	xfs_bmap_free_t		*flist)	/* list of bmap_free_items */
 
 665{
 666	xfs_bmap_free_item_t	*free;	/* free list item */
 667	xfs_bmap_free_item_t	*next;
 668
 669	if (flist->xbf_count == 0)
 670		return;
 671	ASSERT(flist->xbf_first != NULL);
 672	for (free = flist->xbf_first; free; free = next) {
 673		next = free->xbfi_next;
 674		xfs_bmap_del_free(flist, NULL, free);
 675	}
 676	ASSERT(flist->xbf_count == 0);
 677}
 678
 679/*
 680 * Inode fork format manipulation functions
 681 */
 682
 683/*
 684 * Transform a btree format file with only one leaf node, where the
 685 * extents list will fit in the inode, into an extents format file.
 686 * Since the file extents are already in-core, all we have to do is
 687 * give up the space for the btree root and pitch the leaf block.
 688 */
 689STATIC int				/* error */
 690xfs_bmap_btree_to_extents(
 691	xfs_trans_t		*tp,	/* transaction pointer */
 692	xfs_inode_t		*ip,	/* incore inode pointer */
 693	xfs_btree_cur_t		*cur,	/* btree cursor */
 694	int			*logflagsp, /* inode logging flags */
 695	int			whichfork)  /* data or attr fork */
 696{
 697	/* REFERENCED */
 698	struct xfs_btree_block	*cblock;/* child btree block */
 699	xfs_fsblock_t		cbno;	/* child block number */
 700	xfs_buf_t		*cbp;	/* child block's buffer */
 701	int			error;	/* error return value */
 702	xfs_ifork_t		*ifp;	/* inode fork data */
 703	xfs_mount_t		*mp;	/* mount point structure */
 704	__be64			*pp;	/* ptr to block address */
 705	struct xfs_btree_block	*rblock;/* root btree block */
 706
 707	mp = ip->i_mount;
 708	ifp = XFS_IFORK_PTR(ip, whichfork);
 709	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
 710	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
 711	rblock = ifp->if_broot;
 712	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
 713	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
 714	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
 715	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
 716	cbno = be64_to_cpu(*pp);
 717	*logflagsp = 0;
 718#ifdef DEBUG
 719	if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
 720		return error;
 721#endif
 722	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 723				&xfs_bmbt_buf_ops);
 724	if (error)
 725		return error;
 726	cblock = XFS_BUF_TO_BLOCK(cbp);
 727	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 728		return error;
 729	xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
 730	ip->i_d.di_nblocks--;
 731	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 732	xfs_trans_binval(tp, cbp);
 733	if (cur->bc_bufs[0] == cbp)
 734		cur->bc_bufs[0] = NULL;
 735	xfs_iroot_realloc(ip, -1, whichfork);
 736	ASSERT(ifp->if_broot == NULL);
 737	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
 738	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 739	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 740	return 0;
 741}
 742
 743/*
 744 * Convert an extents-format file into a btree-format file.
 745 * The new file will have a root block (in the inode) and a single child block.
 746 */
 747STATIC int					/* error */
 748xfs_bmap_extents_to_btree(
 749	xfs_trans_t		*tp,		/* transaction pointer */
 750	xfs_inode_t		*ip,		/* incore inode pointer */
 751	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
 752	xfs_bmap_free_t		*flist,		/* blocks freed in xaction */
 753	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
 754	int			wasdel,		/* converting a delayed alloc */
 755	int			*logflagsp,	/* inode logging flags */
 756	int			whichfork)	/* data or attr fork */
 757{
 758	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
 759	xfs_buf_t		*abp;		/* buffer for ablock */
 760	xfs_alloc_arg_t		args;		/* allocation arguments */
 761	xfs_bmbt_rec_t		*arp;		/* child record pointer */
 762	struct xfs_btree_block	*block;		/* btree root block */
 763	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
 764	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
 765	int			error;		/* error return value */
 766	xfs_extnum_t		i, cnt;		/* extent record index */
 767	xfs_ifork_t		*ifp;		/* inode fork pointer */
 768	xfs_bmbt_key_t		*kp;		/* root block key pointer */
 769	xfs_mount_t		*mp;		/* mount structure */
 770	xfs_extnum_t		nextents;	/* number of file extents */
 771	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
 772
 773	mp = ip->i_mount;
 774	ifp = XFS_IFORK_PTR(ip, whichfork);
 775	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
 776
 777	/*
 778	 * Make space in the inode incore.
 779	 */
 780	xfs_iroot_realloc(ip, 1, whichfork);
 781	ifp->if_flags |= XFS_IFBROOT;
 782
 783	/*
 784	 * Fill in the root.
 785	 */
 786	block = ifp->if_broot;
 787	if (xfs_sb_version_hascrc(&mp->m_sb))
 788		xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 789				 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
 790				 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 791	else
 792		xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
 793				 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
 794				 XFS_BTREE_LONG_PTRS);
 795
 796	/*
 797	 * Need a cursor.  Can't allocate until bb_level is filled in.
 798	 */
 799	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 800	cur->bc_private.b.firstblock = *firstblock;
 801	cur->bc_private.b.flist = flist;
 802	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
 803	/*
 804	 * Convert to a btree with two levels, one record in root.
 805	 */
 806	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
 807	memset(&args, 0, sizeof(args));
 808	args.tp = tp;
 809	args.mp = mp;
 810	args.firstblock = *firstblock;
 811	if (*firstblock == NULLFSBLOCK) {
 812		args.type = XFS_ALLOCTYPE_START_BNO;
 813		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
 814	} else if (flist->xbf_low) {
 815		args.type = XFS_ALLOCTYPE_START_BNO;
 816		args.fsbno = *firstblock;
 817	} else {
 818		args.type = XFS_ALLOCTYPE_NEAR_BNO;
 819		args.fsbno = *firstblock;
 820	}
 821	args.minlen = args.maxlen = args.prod = 1;
 822	args.wasdel = wasdel;
 823	*logflagsp = 0;
 824	if ((error = xfs_alloc_vextent(&args))) {
 825		xfs_iroot_realloc(ip, -1, whichfork);
 826		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 827		return error;
 828	}
 829	/*
 830	 * Allocation can't fail, the space was reserved.
 831	 */
 832	ASSERT(args.fsbno != NULLFSBLOCK);
 833	ASSERT(*firstblock == NULLFSBLOCK ||
 834	       args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
 835	       (flist->xbf_low &&
 836		args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
 837	*firstblock = cur->bc_private.b.firstblock = args.fsbno;
 838	cur->bc_private.b.allocated++;
 839	ip->i_d.di_nblocks++;
 840	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
 841	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
 842	/*
 843	 * Fill in the child block.
 844	 */
 845	abp->b_ops = &xfs_bmbt_buf_ops;
 846	ablock = XFS_BUF_TO_BLOCK(abp);
 847	if (xfs_sb_version_hascrc(&mp->m_sb))
 848		xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 849				XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
 850				XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
 851	else
 852		xfs_btree_init_block_int(mp, ablock, abp->b_bn,
 853				XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
 854				XFS_BTREE_LONG_PTRS);
 855
 856	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 857	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 858	for (cnt = i = 0; i < nextents; i++) {
 859		ep = xfs_iext_get_ext(ifp, i);
 860		if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
 861			arp->l0 = cpu_to_be64(ep->l0);
 862			arp->l1 = cpu_to_be64(ep->l1);
 863			arp++; cnt++;
 864		}
 865	}
 866	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
 867	xfs_btree_set_numrecs(ablock, cnt);
 868
 869	/*
 870	 * Fill in the root key and pointer.
 871	 */
 872	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
 873	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
 874	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
 875	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
 876						be16_to_cpu(block->bb_level)));
 877	*pp = cpu_to_be64(args.fsbno);
 878
 879	/*
 880	 * Do all this logging at the end so that
 881	 * the root is at the right level.
 882	 */
 883	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
 884	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
 885	ASSERT(*curp == NULL);
 886	*curp = cur;
 887	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
 888	return 0;
 889}
 890
 891/*
 892 * Convert a local file to an extents file.
 893 * This code is out of bounds for data forks of regular files,
 894 * since the file data needs to get logged so things will stay consistent.
 895 * (The bmap-level manipulations are ok, though).
 896 */
 897void
 898xfs_bmap_local_to_extents_empty(
 899	struct xfs_inode	*ip,
 900	int			whichfork)
 901{
 902	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
 903
 904	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 905	ASSERT(ifp->if_bytes == 0);
 906	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
 907
 908	xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
 909	ifp->if_flags &= ~XFS_IFINLINE;
 910	ifp->if_flags |= XFS_IFEXTENTS;
 911	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
 912}
 913
 914
 915STATIC int				/* error */
 916xfs_bmap_local_to_extents(
 917	xfs_trans_t	*tp,		/* transaction pointer */
 918	xfs_inode_t	*ip,		/* incore inode pointer */
 919	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
 920	xfs_extlen_t	total,		/* total blocks needed by transaction */
 921	int		*logflagsp,	/* inode logging flags */
 922	int		whichfork,
 923	void		(*init_fn)(struct xfs_trans *tp,
 924				   struct xfs_buf *bp,
 925				   struct xfs_inode *ip,
 926				   struct xfs_ifork *ifp))
 927{
 928	int		error = 0;
 929	int		flags;		/* logging flags returned */
 930	xfs_ifork_t	*ifp;		/* inode fork pointer */
 931	xfs_alloc_arg_t	args;		/* allocation arguments */
 932	xfs_buf_t	*bp;		/* buffer for extent block */
 933	xfs_bmbt_rec_host_t *ep;	/* extent record pointer */
 934
 935	/*
 936	 * We don't want to deal with the case of keeping inode data inline yet.
 937	 * So sending the data fork of a regular inode is invalid.
 938	 */
 939	ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
 940	ifp = XFS_IFORK_PTR(ip, whichfork);
 941	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
 942
 943	if (!ifp->if_bytes) {
 944		xfs_bmap_local_to_extents_empty(ip, whichfork);
 945		flags = XFS_ILOG_CORE;
 946		goto done;
 947	}
 948
 949	flags = 0;
 950	error = 0;
 951	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
 952								XFS_IFINLINE);
 953	memset(&args, 0, sizeof(args));
 954	args.tp = tp;
 955	args.mp = ip->i_mount;
 956	args.firstblock = *firstblock;
 957	/*
 958	 * Allocate a block.  We know we need only one, since the
 959	 * file currently fits in an inode.
 960	 */
 961	if (*firstblock == NULLFSBLOCK) {
 962		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
 963		args.type = XFS_ALLOCTYPE_START_BNO;
 964	} else {
 965		args.fsbno = *firstblock;
 966		args.type = XFS_ALLOCTYPE_NEAR_BNO;
 967	}
 968	args.total = total;
 969	args.minlen = args.maxlen = args.prod = 1;
 970	error = xfs_alloc_vextent(&args);
 971	if (error)
 972		goto done;
 973
 974	/* Can't fail, the space was reserved. */
 975	ASSERT(args.fsbno != NULLFSBLOCK);
 976	ASSERT(args.len == 1);
 977	*firstblock = args.fsbno;
 978	bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
 979
 980	/* initialise the block and copy the data */
 981	init_fn(tp, bp, ip, ifp);
 982
 983	/* account for the change in fork size and log everything */
 984	xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
 985	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
 986	xfs_bmap_local_to_extents_empty(ip, whichfork);
 987	flags |= XFS_ILOG_CORE;
 988
 989	xfs_iext_add(ifp, 0, 1);
 990	ep = xfs_iext_get_ext(ifp, 0);
 991	xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
 992	trace_xfs_bmap_post_update(ip, 0,
 993			whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
 994			_THIS_IP_);
 995	XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 996	ip->i_d.di_nblocks = 1;
 997	xfs_trans_mod_dquot_byino(tp, ip,
 998		XFS_TRANS_DQ_BCOUNT, 1L);
 999	flags |= xfs_ilog_fext(whichfork);
1000
1001done:
1002	*logflagsp = flags;
1003	return error;
1004}
1005
1006/*
1007 * Called from xfs_bmap_add_attrfork to handle btree format files.
1008 */
1009STATIC int					/* error */
1010xfs_bmap_add_attrfork_btree(
1011	xfs_trans_t		*tp,		/* transaction pointer */
1012	xfs_inode_t		*ip,		/* incore inode pointer */
1013	xfs_fsblock_t		*firstblock,	/* first block allocated */
1014	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
1015	int			*flags)		/* inode logging flags */
1016{
1017	xfs_btree_cur_t		*cur;		/* btree cursor */
1018	int			error;		/* error return value */
1019	xfs_mount_t		*mp;		/* file system mount struct */
1020	int			stat;		/* newroot status */
1021
1022	mp = ip->i_mount;
1023	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1024		*flags |= XFS_ILOG_DBROOT;
1025	else {
1026		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1027		cur->bc_private.b.flist = flist;
1028		cur->bc_private.b.firstblock = *firstblock;
1029		if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1030			goto error0;
1031		/* must be at least one entry */
1032		XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1033		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1034			goto error0;
1035		if (stat == 0) {
1036			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1037			return XFS_ERROR(ENOSPC);
1038		}
1039		*firstblock = cur->bc_private.b.firstblock;
1040		cur->bc_private.b.allocated = 0;
1041		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1042	}
1043	return 0;
1044error0:
1045	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1046	return error;
1047}
1048
1049/*
1050 * Called from xfs_bmap_add_attrfork to handle extents format files.
1051 */
1052STATIC int					/* error */
1053xfs_bmap_add_attrfork_extents(
1054	xfs_trans_t		*tp,		/* transaction pointer */
1055	xfs_inode_t		*ip,		/* incore inode pointer */
1056	xfs_fsblock_t		*firstblock,	/* first block allocated */
1057	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
1058	int			*flags)		/* inode logging flags */
1059{
1060	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
1061	int			error;		/* error return value */
1062
1063	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1064		return 0;
1065	cur = NULL;
1066	error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1067		flags, XFS_DATA_FORK);
1068	if (cur) {
1069		cur->bc_private.b.allocated = 0;
1070		xfs_btree_del_cursor(cur,
1071			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1072	}
1073	return error;
1074}
1075
1076/*
1077 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1078 * different data fork content type needs a different callout to do the
1079 * conversion. Some are basic and only require special block initialisation
1080 * callouts for the data formating, others (directories) are so specialised they
1081 * handle everything themselves.
1082 *
1083 * XXX (dgc): investigate whether directory conversion can use the generic
1084 * formatting callout. It should be possible - it's just a very complex
1085 * formatter.
1086 */
1087STATIC int					/* error */
1088xfs_bmap_add_attrfork_local(
1089	xfs_trans_t		*tp,		/* transaction pointer */
1090	xfs_inode_t		*ip,		/* incore inode pointer */
1091	xfs_fsblock_t		*firstblock,	/* first block allocated */
1092	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
1093	int			*flags)		/* inode logging flags */
1094{
1095	xfs_da_args_t		dargs;		/* args for dir/attr code */
 
 
1096
1097	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1098		return 0;
1099
1100	if (S_ISDIR(ip->i_d.di_mode)) {
 
1101		memset(&dargs, 0, sizeof(dargs));
1102		dargs.dp = ip;
1103		dargs.firstblock = firstblock;
1104		dargs.flist = flist;
1105		dargs.total = ip->i_mount->m_dirblkfsbs;
1106		dargs.whichfork = XFS_DATA_FORK;
1107		dargs.trans = tp;
1108		return xfs_dir2_sf_to_block(&dargs);
1109	}
1110
1111	if (S_ISLNK(ip->i_d.di_mode))
1112		return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1113						 flags, XFS_DATA_FORK,
1114						 xfs_symlink_local_to_remote);
1115
1116	/* should only be called for types that support local format data */
1117	ASSERT(0);
1118	return EFSCORRUPTED;
1119}
1120
1121/*
1122 * Convert inode from non-attributed to attributed.
1123 * Must not be in a transaction, ip must not be locked.
1124 */
1125int						/* error code */
1126xfs_bmap_add_attrfork(
1127	xfs_inode_t		*ip,		/* incore inode pointer */
1128	int			size,		/* space new attribute needs */
1129	int			rsvd)		/* xact may use reserved blks */
1130{
1131	xfs_fsblock_t		firstblock;	/* 1st block/ag allocated */
1132	xfs_bmap_free_t		flist;		/* freed extent records */
1133	xfs_mount_t		*mp;		/* mount structure */
1134	xfs_trans_t		*tp;		/* transaction pointer */
1135	int			blks;		/* space reservation */
1136	int			version = 1;	/* superblock attr version */
1137	int			committed;	/* xaction was committed */
1138	int			logflags;	/* logging flags */
1139	int			error;		/* error return value */
1140	int			cancel_flags = 0;
1141
1142	ASSERT(XFS_IFORK_Q(ip) == 0);
1143
1144	mp = ip->i_mount;
1145	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1146	tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1147	blks = XFS_ADDAFORK_SPACE_RES(mp);
1148	if (rsvd)
1149		tp->t_flags |= XFS_TRANS_RESERVE;
1150	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1151	if (error) {
1152		xfs_trans_cancel(tp, 0);
1153		return error;
1154	}
1155	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1156	xfs_ilock(ip, XFS_ILOCK_EXCL);
1157	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1158			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1159			XFS_QMOPT_RES_REGBLKS);
1160	if (error)
1161		goto trans_cancel;
1162	cancel_flags |= XFS_TRANS_ABORT;
1163	if (XFS_IFORK_Q(ip))
1164		goto trans_cancel;
1165	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1166		/*
1167		 * For inodes coming from pre-6.2 filesystems.
1168		 */
1169		ASSERT(ip->i_d.di_aformat == 0);
1170		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1171	}
1172	ASSERT(ip->i_d.di_anextents == 0);
1173
1174	xfs_trans_ijoin(tp, ip, 0);
1175	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1176
1177	switch (ip->i_d.di_format) {
1178	case XFS_DINODE_FMT_DEV:
1179		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1180		break;
1181	case XFS_DINODE_FMT_UUID:
1182		ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1183		break;
1184	case XFS_DINODE_FMT_LOCAL:
1185	case XFS_DINODE_FMT_EXTENTS:
1186	case XFS_DINODE_FMT_BTREE:
1187		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1188		if (!ip->i_d.di_forkoff)
1189			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1190		else if (mp->m_flags & XFS_MOUNT_ATTR2)
1191			version = 2;
1192		break;
1193	default:
1194		ASSERT(0);
1195		error = XFS_ERROR(EINVAL);
1196		goto trans_cancel;
1197	}
1198
1199	ASSERT(ip->i_afp == NULL);
1200	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1201	ip->i_afp->if_flags = XFS_IFEXTENTS;
1202	logflags = 0;
1203	xfs_bmap_init(&flist, &firstblock);
1204	switch (ip->i_d.di_format) {
1205	case XFS_DINODE_FMT_LOCAL:
1206		error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1207			&logflags);
1208		break;
1209	case XFS_DINODE_FMT_EXTENTS:
1210		error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1211			&flist, &logflags);
1212		break;
1213	case XFS_DINODE_FMT_BTREE:
1214		error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1215			&logflags);
1216		break;
1217	default:
1218		error = 0;
1219		break;
1220	}
1221	if (logflags)
1222		xfs_trans_log_inode(tp, ip, logflags);
1223	if (error)
1224		goto bmap_cancel;
1225	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1226	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1227		__int64_t sbfields = 0;
1228
1229		spin_lock(&mp->m_sb_lock);
1230		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1231			xfs_sb_version_addattr(&mp->m_sb);
1232			sbfields |= XFS_SB_VERSIONNUM;
1233		}
1234		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1235			xfs_sb_version_addattr2(&mp->m_sb);
1236			sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1237		}
1238		if (sbfields) {
1239			spin_unlock(&mp->m_sb_lock);
1240			xfs_mod_sb(tp, sbfields);
1241		} else
1242			spin_unlock(&mp->m_sb_lock);
1243	}
1244
1245	error = xfs_bmap_finish(&tp, &flist, &committed);
1246	if (error)
1247		goto bmap_cancel;
1248	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1249	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1250	return error;
1251
1252bmap_cancel:
1253	xfs_bmap_cancel(&flist);
1254trans_cancel:
1255	xfs_trans_cancel(tp, cancel_flags);
1256	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1257	return error;
1258}
1259
1260/*
1261 * Internal and external extent tree search functions.
1262 */
1263
1264/*
1265 * Read in the extents to if_extents.
1266 * All inode fields are set up by caller, we just traverse the btree
1267 * and copy the records in. If the file system cannot contain unwritten
1268 * extents, the records are checked for no "state" flags.
1269 */
1270int					/* error */
1271xfs_bmap_read_extents(
1272	xfs_trans_t		*tp,	/* transaction pointer */
1273	xfs_inode_t		*ip,	/* incore inode */
1274	int			whichfork) /* data or attr fork */
1275{
1276	struct xfs_btree_block	*block;	/* current btree block */
1277	xfs_fsblock_t		bno;	/* block # of "block" */
1278	xfs_buf_t		*bp;	/* buffer for "block" */
1279	int			error;	/* error return value */
1280	xfs_exntfmt_t		exntf;	/* XFS_EXTFMT_NOSTATE, if checking */
1281	xfs_extnum_t		i, j;	/* index into the extents list */
1282	xfs_ifork_t		*ifp;	/* fork structure */
1283	int			level;	/* btree level, for checking */
1284	xfs_mount_t		*mp;	/* file system mount structure */
1285	__be64			*pp;	/* pointer to block address */
1286	/* REFERENCED */
1287	xfs_extnum_t		room;	/* number of entries there's room for */
1288
1289	bno = NULLFSBLOCK;
1290	mp = ip->i_mount;
1291	ifp = XFS_IFORK_PTR(ip, whichfork);
1292	exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1293					XFS_EXTFMT_INODE(ip);
1294	block = ifp->if_broot;
1295	/*
1296	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1297	 */
1298	level = be16_to_cpu(block->bb_level);
1299	ASSERT(level > 0);
1300	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1301	bno = be64_to_cpu(*pp);
1302	ASSERT(bno != NULLDFSBNO);
1303	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1304	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1305	/*
1306	 * Go down the tree until leaf level is reached, following the first
1307	 * pointer (leftmost) at each level.
1308	 */
1309	while (level-- > 0) {
1310		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1311				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1312		if (error)
1313			return error;
1314		block = XFS_BUF_TO_BLOCK(bp);
1315		XFS_WANT_CORRUPTED_GOTO(
1316			xfs_bmap_sanity_check(mp, bp, level),
1317			error0);
1318		if (level == 0)
1319			break;
1320		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1321		bno = be64_to_cpu(*pp);
1322		XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1323		xfs_trans_brelse(tp, bp);
1324	}
1325	/*
1326	 * Here with bp and block set to the leftmost leaf node in the tree.
1327	 */
1328	room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1329	i = 0;
1330	/*
1331	 * Loop over all leaf nodes.  Copy information to the extent records.
1332	 */
1333	for (;;) {
1334		xfs_bmbt_rec_t	*frp;
1335		xfs_fsblock_t	nextbno;
1336		xfs_extnum_t	num_recs;
1337		xfs_extnum_t	start;
1338
1339		num_recs = xfs_btree_get_numrecs(block);
1340		if (unlikely(i + num_recs > room)) {
1341			ASSERT(i + num_recs <= room);
1342			xfs_warn(ip->i_mount,
1343				"corrupt dinode %Lu, (btree extents).",
1344				(unsigned long long) ip->i_ino);
1345			XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1346				XFS_ERRLEVEL_LOW, ip->i_mount, block);
1347			goto error0;
1348		}
1349		XFS_WANT_CORRUPTED_GOTO(
1350			xfs_bmap_sanity_check(mp, bp, 0),
1351			error0);
1352		/*
1353		 * Read-ahead the next leaf block, if any.
1354		 */
1355		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1356		if (nextbno != NULLFSBLOCK)
1357			xfs_btree_reada_bufl(mp, nextbno, 1,
1358					     &xfs_bmbt_buf_ops);
1359		/*
1360		 * Copy records into the extent records.
1361		 */
1362		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1363		start = i;
1364		for (j = 0; j < num_recs; j++, i++, frp++) {
1365			xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1366			trp->l0 = be64_to_cpu(frp->l0);
1367			trp->l1 = be64_to_cpu(frp->l1);
1368		}
1369		if (exntf == XFS_EXTFMT_NOSTATE) {
1370			/*
1371			 * Check all attribute bmap btree records and
1372			 * any "older" data bmap btree records for a
1373			 * set bit in the "extent flag" position.
1374			 */
1375			if (unlikely(xfs_check_nostate_extents(ifp,
1376					start, num_recs))) {
1377				XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1378						 XFS_ERRLEVEL_LOW,
1379						 ip->i_mount);
1380				goto error0;
1381			}
1382		}
1383		xfs_trans_brelse(tp, bp);
1384		bno = nextbno;
1385		/*
1386		 * If we've reached the end, stop.
1387		 */
1388		if (bno == NULLFSBLOCK)
1389			break;
1390		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1391				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1392		if (error)
1393			return error;
1394		block = XFS_BUF_TO_BLOCK(bp);
1395	}
1396	ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1397	ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1398	XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1399	return 0;
1400error0:
1401	xfs_trans_brelse(tp, bp);
1402	return XFS_ERROR(EFSCORRUPTED);
1403}
1404
1405
1406/*
1407 * Search the extent records for the entry containing block bno.
1408 * If bno lies in a hole, point to the next entry.  If bno lies
1409 * past eof, *eofp will be set, and *prevp will contain the last
1410 * entry (null if none).  Else, *lastxp will be set to the index
1411 * of the found entry; *gotp will contain the entry.
1412 */
1413STATIC xfs_bmbt_rec_host_t *		/* pointer to found extent entry */
1414xfs_bmap_search_multi_extents(
1415	xfs_ifork_t	*ifp,		/* inode fork pointer */
1416	xfs_fileoff_t	bno,		/* block number searched for */
1417	int		*eofp,		/* out: end of file found */
1418	xfs_extnum_t	*lastxp,	/* out: last extent index */
1419	xfs_bmbt_irec_t	*gotp,		/* out: extent entry found */
1420	xfs_bmbt_irec_t	*prevp)		/* out: previous extent entry found */
1421{
1422	xfs_bmbt_rec_host_t *ep;		/* extent record pointer */
1423	xfs_extnum_t	lastx;		/* last extent index */
1424
1425	/*
1426	 * Initialize the extent entry structure to catch access to
1427	 * uninitialized br_startblock field.
1428	 */
1429	gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1430	gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1431	gotp->br_state = XFS_EXT_INVALID;
1432#if XFS_BIG_BLKNOS
1433	gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1434#else
1435	gotp->br_startblock = 0xffffa5a5;
1436#endif
1437	prevp->br_startoff = NULLFILEOFF;
1438
1439	ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1440	if (lastx > 0) {
1441		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1442	}
1443	if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1444		xfs_bmbt_get_all(ep, gotp);
1445		*eofp = 0;
1446	} else {
1447		if (lastx > 0) {
1448			*gotp = *prevp;
1449		}
1450		*eofp = 1;
1451		ep = NULL;
1452	}
1453	*lastxp = lastx;
1454	return ep;
1455}
1456
1457/*
1458 * Search the extents list for the inode, for the extent containing bno.
1459 * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1460 * *eofp will be set, and *prevp will contain the last entry (null if none).
1461 * Else, *lastxp will be set to the index of the found
1462 * entry; *gotp will contain the entry.
1463 */
1464STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
1465xfs_bmap_search_extents(
1466	xfs_inode_t     *ip,            /* incore inode pointer */
1467	xfs_fileoff_t   bno,            /* block number searched for */
1468	int             fork,      	/* data or attr fork */
1469	int             *eofp,          /* out: end of file found */
1470	xfs_extnum_t    *lastxp,        /* out: last extent index */
1471	xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1472	xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1473{
1474	xfs_ifork_t	*ifp;		/* inode fork pointer */
1475	xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1476
1477	XFS_STATS_INC(xs_look_exlist);
1478	ifp = XFS_IFORK_PTR(ip, fork);
1479
1480	ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1481
1482	if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1483		     !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1484		xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1485				"Access to block zero in inode %llu "
1486				"start_block: %llx start_off: %llx "
1487				"blkcnt: %llx extent-state: %x lastx: %x",
1488			(unsigned long long)ip->i_ino,
1489			(unsigned long long)gotp->br_startblock,
1490			(unsigned long long)gotp->br_startoff,
1491			(unsigned long long)gotp->br_blockcount,
1492			gotp->br_state, *lastxp);
1493		*lastxp = NULLEXTNUM;
1494		*eofp = 1;
1495		return NULL;
1496	}
1497	return ep;
1498}
1499
1500/*
1501 * Returns the file-relative block number of the first unused block(s)
1502 * in the file with at least "len" logically contiguous blocks free.
1503 * This is the lowest-address hole if the file has holes, else the first block
1504 * past the end of file.
1505 * Return 0 if the file is currently local (in-inode).
1506 */
1507int						/* error */
1508xfs_bmap_first_unused(
1509	xfs_trans_t	*tp,			/* transaction pointer */
1510	xfs_inode_t	*ip,			/* incore inode */
1511	xfs_extlen_t	len,			/* size of hole to find */
1512	xfs_fileoff_t	*first_unused,		/* unused block */
1513	int		whichfork)		/* data or attr fork */
1514{
1515	int		error;			/* error return value */
1516	int		idx;			/* extent record index */
1517	xfs_ifork_t	*ifp;			/* inode fork pointer */
1518	xfs_fileoff_t	lastaddr;		/* last block number seen */
1519	xfs_fileoff_t	lowest;			/* lowest useful block */
1520	xfs_fileoff_t	max;			/* starting useful block */
1521	xfs_fileoff_t	off;			/* offset for this block */
1522	xfs_extnum_t	nextents;		/* number of extent entries */
1523
1524	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1525	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1526	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1527	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1528		*first_unused = 0;
1529		return 0;
1530	}
1531	ifp = XFS_IFORK_PTR(ip, whichfork);
1532	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1533	    (error = xfs_iread_extents(tp, ip, whichfork)))
1534		return error;
1535	lowest = *first_unused;
1536	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1537	for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1538		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1539		off = xfs_bmbt_get_startoff(ep);
1540		/*
1541		 * See if the hole before this extent will work.
1542		 */
1543		if (off >= lowest + len && off - max >= len) {
1544			*first_unused = max;
1545			return 0;
1546		}
1547		lastaddr = off + xfs_bmbt_get_blockcount(ep);
1548		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1549	}
1550	*first_unused = max;
1551	return 0;
1552}
1553
1554/*
1555 * Returns the file-relative block number of the last block - 1 before
1556 * last_block (input value) in the file.
1557 * This is not based on i_size, it is based on the extent records.
1558 * Returns 0 for local files, as they do not have extent records.
1559 */
1560int						/* error */
1561xfs_bmap_last_before(
1562	xfs_trans_t	*tp,			/* transaction pointer */
1563	xfs_inode_t	*ip,			/* incore inode */
1564	xfs_fileoff_t	*last_block,		/* last block */
1565	int		whichfork)		/* data or attr fork */
1566{
1567	xfs_fileoff_t	bno;			/* input file offset */
1568	int		eof;			/* hit end of file */
1569	xfs_bmbt_rec_host_t *ep;		/* pointer to last extent */
1570	int		error;			/* error return value */
1571	xfs_bmbt_irec_t	got;			/* current extent value */
1572	xfs_ifork_t	*ifp;			/* inode fork pointer */
1573	xfs_extnum_t	lastx;			/* last extent used */
1574	xfs_bmbt_irec_t	prev;			/* previous extent value */
1575
1576	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1577	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1578	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1579	       return XFS_ERROR(EIO);
1580	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1581		*last_block = 0;
1582		return 0;
1583	}
1584	ifp = XFS_IFORK_PTR(ip, whichfork);
1585	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1586	    (error = xfs_iread_extents(tp, ip, whichfork)))
1587		return error;
1588	bno = *last_block - 1;
1589	ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1590		&prev);
1591	if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1592		if (prev.br_startoff == NULLFILEOFF)
1593			*last_block = 0;
1594		else
1595			*last_block = prev.br_startoff + prev.br_blockcount;
1596	}
1597	/*
1598	 * Otherwise *last_block is already the right answer.
1599	 */
1600	return 0;
1601}
1602
1603int
1604xfs_bmap_last_extent(
1605	struct xfs_trans	*tp,
1606	struct xfs_inode	*ip,
1607	int			whichfork,
1608	struct xfs_bmbt_irec	*rec,
1609	int			*is_empty)
1610{
1611	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1612	int			error;
1613	int			nextents;
1614
1615	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1616		error = xfs_iread_extents(tp, ip, whichfork);
1617		if (error)
1618			return error;
1619	}
1620
1621	nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1622	if (nextents == 0) {
1623		*is_empty = 1;
1624		return 0;
1625	}
1626
1627	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1628	*is_empty = 0;
1629	return 0;
1630}
1631
1632/*
1633 * Check the last inode extent to determine whether this allocation will result
1634 * in blocks being allocated at the end of the file. When we allocate new data
1635 * blocks at the end of the file which do not start at the previous data block,
1636 * we will try to align the new blocks at stripe unit boundaries.
1637 *
1638 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1639 * at, or past the EOF.
1640 */
1641STATIC int
1642xfs_bmap_isaeof(
1643	struct xfs_bmalloca	*bma,
1644	int			whichfork)
1645{
1646	struct xfs_bmbt_irec	rec;
1647	int			is_empty;
1648	int			error;
1649
1650	bma->aeof = 0;
1651	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1652				     &is_empty);
1653	if (error)
1654		return error;
1655
1656	if (is_empty) {
1657		bma->aeof = 1;
1658		return 0;
1659	}
1660
1661	/*
1662	 * Check if we are allocation or past the last extent, or at least into
1663	 * the last delayed allocated extent.
1664	 */
1665	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1666		(bma->offset >= rec.br_startoff &&
1667		 isnullstartblock(rec.br_startblock));
1668	return 0;
1669}
1670
1671/*
1672 * Returns the file-relative block number of the first block past eof in
1673 * the file.  This is not based on i_size, it is based on the extent records.
1674 * Returns 0 for local files, as they do not have extent records.
1675 */
1676int
1677xfs_bmap_last_offset(
1678	struct xfs_trans	*tp,
1679	struct xfs_inode	*ip,
1680	xfs_fileoff_t		*last_block,
1681	int			whichfork)
1682{
1683	struct xfs_bmbt_irec	rec;
1684	int			is_empty;
1685	int			error;
1686
1687	*last_block = 0;
1688
1689	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1690		return 0;
1691
1692	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1693	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1694	       return XFS_ERROR(EIO);
1695
1696	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1697	if (error || is_empty)
1698		return error;
1699
1700	*last_block = rec.br_startoff + rec.br_blockcount;
1701	return 0;
1702}
1703
1704/*
1705 * Returns whether the selected fork of the inode has exactly one
1706 * block or not.  For the data fork we check this matches di_size,
1707 * implying the file's range is 0..bsize-1.
1708 */
1709int					/* 1=>1 block, 0=>otherwise */
1710xfs_bmap_one_block(
1711	xfs_inode_t	*ip,		/* incore inode */
1712	int		whichfork)	/* data or attr fork */
1713{
1714	xfs_bmbt_rec_host_t *ep;	/* ptr to fork's extent */
1715	xfs_ifork_t	*ifp;		/* inode fork pointer */
1716	int		rval;		/* return value */
1717	xfs_bmbt_irec_t	s;		/* internal version of extent */
1718
1719#ifndef DEBUG
1720	if (whichfork == XFS_DATA_FORK)
1721		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1722#endif	/* !DEBUG */
1723	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1724		return 0;
1725	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1726		return 0;
1727	ifp = XFS_IFORK_PTR(ip, whichfork);
1728	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1729	ep = xfs_iext_get_ext(ifp, 0);
1730	xfs_bmbt_get_all(ep, &s);
1731	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1732	if (rval && whichfork == XFS_DATA_FORK)
1733		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1734	return rval;
1735}
1736
1737/*
1738 * Extent tree manipulation functions used during allocation.
1739 */
1740
1741/*
1742 * Convert a delayed allocation to a real allocation.
1743 */
1744STATIC int				/* error */
1745xfs_bmap_add_extent_delay_real(
1746	struct xfs_bmalloca	*bma)
1747{
1748	struct xfs_bmbt_irec	*new = &bma->got;
1749	int			diff;	/* temp value */
1750	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
1751	int			error;	/* error return value */
1752	int			i;	/* temp state */
1753	xfs_ifork_t		*ifp;	/* inode fork pointer */
1754	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1755	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1756					/* left is 0, right is 1, prev is 2 */
1757	int			rval=0;	/* return value (logging flags) */
1758	int			state = 0;/* state bits, accessed thru macros */
1759	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1760	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1761	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1762	xfs_filblks_t		temp2=0;/* value for da_new calculations */
1763	int			tmp_rval;	/* partial logging flags */
1764
1765	ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1766
1767	ASSERT(bma->idx >= 0);
1768	ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1769	ASSERT(!isnullstartblock(new->br_startblock));
1770	ASSERT(!bma->cur ||
1771	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1772
1773	XFS_STATS_INC(xs_add_exlist);
1774
1775#define	LEFT		r[0]
1776#define	RIGHT		r[1]
1777#define	PREV		r[2]
1778
1779	/*
1780	 * Set up a bunch of variables to make the tests simpler.
1781	 */
1782	ep = xfs_iext_get_ext(ifp, bma->idx);
1783	xfs_bmbt_get_all(ep, &PREV);
1784	new_endoff = new->br_startoff + new->br_blockcount;
1785	ASSERT(PREV.br_startoff <= new->br_startoff);
1786	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1787
1788	da_old = startblockval(PREV.br_startblock);
1789	da_new = 0;
1790
1791	/*
1792	 * Set flags determining what part of the previous delayed allocation
1793	 * extent is being replaced by a real allocation.
1794	 */
1795	if (PREV.br_startoff == new->br_startoff)
1796		state |= BMAP_LEFT_FILLING;
1797	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1798		state |= BMAP_RIGHT_FILLING;
1799
1800	/*
1801	 * Check and set flags if this segment has a left neighbor.
1802	 * Don't set contiguous if the combined extent would be too large.
1803	 */
1804	if (bma->idx > 0) {
1805		state |= BMAP_LEFT_VALID;
1806		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1807
1808		if (isnullstartblock(LEFT.br_startblock))
1809			state |= BMAP_LEFT_DELAY;
1810	}
1811
1812	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1813	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1814	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1815	    LEFT.br_state == new->br_state &&
1816	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1817		state |= BMAP_LEFT_CONTIG;
1818
1819	/*
1820	 * Check and set flags if this segment has a right neighbor.
1821	 * Don't set contiguous if the combined extent would be too large.
1822	 * Also check for all-three-contiguous being too large.
1823	 */
1824	if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1825		state |= BMAP_RIGHT_VALID;
1826		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1827
1828		if (isnullstartblock(RIGHT.br_startblock))
1829			state |= BMAP_RIGHT_DELAY;
1830	}
1831
1832	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1833	    new_endoff == RIGHT.br_startoff &&
1834	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1835	    new->br_state == RIGHT.br_state &&
1836	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1837	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1838		       BMAP_RIGHT_FILLING)) !=
1839		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1840		       BMAP_RIGHT_FILLING) ||
1841	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1842			<= MAXEXTLEN))
1843		state |= BMAP_RIGHT_CONTIG;
1844
1845	error = 0;
1846	/*
1847	 * Switch out based on the FILLING and CONTIG state bits.
1848	 */
1849	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1850			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1851	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1852	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1853		/*
1854		 * Filling in all of a previously delayed allocation extent.
1855		 * The left and right neighbors are both contiguous with new.
1856		 */
1857		bma->idx--;
1858		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1859		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1860			LEFT.br_blockcount + PREV.br_blockcount +
1861			RIGHT.br_blockcount);
1862		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1863
1864		xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1865		bma->ip->i_d.di_nextents--;
1866		if (bma->cur == NULL)
1867			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1868		else {
1869			rval = XFS_ILOG_CORE;
1870			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1871					RIGHT.br_startblock,
1872					RIGHT.br_blockcount, &i);
1873			if (error)
1874				goto done;
1875			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1876			error = xfs_btree_delete(bma->cur, &i);
1877			if (error)
1878				goto done;
1879			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1880			error = xfs_btree_decrement(bma->cur, 0, &i);
1881			if (error)
1882				goto done;
1883			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1884			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1885					LEFT.br_startblock,
1886					LEFT.br_blockcount +
1887					PREV.br_blockcount +
1888					RIGHT.br_blockcount, LEFT.br_state);
1889			if (error)
1890				goto done;
1891		}
1892		break;
1893
1894	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1895		/*
1896		 * Filling in all of a previously delayed allocation extent.
1897		 * The left neighbor is contiguous, the right is not.
1898		 */
1899		bma->idx--;
1900
1901		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1902		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1903			LEFT.br_blockcount + PREV.br_blockcount);
1904		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1905
1906		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1907		if (bma->cur == NULL)
1908			rval = XFS_ILOG_DEXT;
1909		else {
1910			rval = 0;
1911			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1912					LEFT.br_startblock, LEFT.br_blockcount,
1913					&i);
1914			if (error)
1915				goto done;
1916			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1917			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1918					LEFT.br_startblock,
1919					LEFT.br_blockcount +
1920					PREV.br_blockcount, LEFT.br_state);
1921			if (error)
1922				goto done;
1923		}
1924		break;
1925
1926	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1927		/*
1928		 * Filling in all of a previously delayed allocation extent.
1929		 * The right neighbor is contiguous, the left is not.
1930		 */
1931		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1932		xfs_bmbt_set_startblock(ep, new->br_startblock);
1933		xfs_bmbt_set_blockcount(ep,
1934			PREV.br_blockcount + RIGHT.br_blockcount);
1935		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1936
1937		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1938		if (bma->cur == NULL)
1939			rval = XFS_ILOG_DEXT;
1940		else {
1941			rval = 0;
1942			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1943					RIGHT.br_startblock,
1944					RIGHT.br_blockcount, &i);
1945			if (error)
1946				goto done;
1947			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1948			error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1949					new->br_startblock,
1950					PREV.br_blockcount +
1951					RIGHT.br_blockcount, PREV.br_state);
1952			if (error)
1953				goto done;
1954		}
1955		break;
1956
1957	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1958		/*
1959		 * Filling in all of a previously delayed allocation extent.
1960		 * Neither the left nor right neighbors are contiguous with
1961		 * the new one.
1962		 */
1963		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1964		xfs_bmbt_set_startblock(ep, new->br_startblock);
1965		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1966
1967		bma->ip->i_d.di_nextents++;
1968		if (bma->cur == NULL)
1969			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1970		else {
1971			rval = XFS_ILOG_CORE;
1972			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1973					new->br_startblock, new->br_blockcount,
1974					&i);
1975			if (error)
1976				goto done;
1977			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1978			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1979			error = xfs_btree_insert(bma->cur, &i);
1980			if (error)
1981				goto done;
1982			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1983		}
1984		break;
1985
1986	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1987		/*
1988		 * Filling in the first part of a previous delayed allocation.
1989		 * The left neighbor is contiguous.
1990		 */
1991		trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1992		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1993			LEFT.br_blockcount + new->br_blockcount);
1994		xfs_bmbt_set_startoff(ep,
1995			PREV.br_startoff + new->br_blockcount);
1996		trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1997
1998		temp = PREV.br_blockcount - new->br_blockcount;
1999		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2000		xfs_bmbt_set_blockcount(ep, temp);
2001		if (bma->cur == NULL)
2002			rval = XFS_ILOG_DEXT;
2003		else {
2004			rval = 0;
2005			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2006					LEFT.br_startblock, LEFT.br_blockcount,
2007					&i);
2008			if (error)
2009				goto done;
2010			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2011			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2012					LEFT.br_startblock,
2013					LEFT.br_blockcount +
2014					new->br_blockcount,
2015					LEFT.br_state);
2016			if (error)
2017				goto done;
2018		}
2019		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2020			startblockval(PREV.br_startblock));
2021		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2022		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2023
2024		bma->idx--;
2025		break;
2026
2027	case BMAP_LEFT_FILLING:
2028		/*
2029		 * Filling in the first part of a previous delayed allocation.
2030		 * The left neighbor is not contiguous.
2031		 */
2032		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2033		xfs_bmbt_set_startoff(ep, new_endoff);
2034		temp = PREV.br_blockcount - new->br_blockcount;
2035		xfs_bmbt_set_blockcount(ep, temp);
2036		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2037		bma->ip->i_d.di_nextents++;
2038		if (bma->cur == NULL)
2039			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2040		else {
2041			rval = XFS_ILOG_CORE;
2042			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2043					new->br_startblock, new->br_blockcount,
2044					&i);
2045			if (error)
2046				goto done;
2047			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2048			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2049			error = xfs_btree_insert(bma->cur, &i);
2050			if (error)
2051				goto done;
2052			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2053		}
2054
2055		if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2056			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2057					bma->firstblock, bma->flist,
2058					&bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2059			rval |= tmp_rval;
2060			if (error)
2061				goto done;
2062		}
2063		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2064			startblockval(PREV.br_startblock) -
2065			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
2066		ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2067		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2068		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2069		break;
2070
2071	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2072		/*
2073		 * Filling in the last part of a previous delayed allocation.
2074		 * The right neighbor is contiguous with the new allocation.
2075		 */
2076		temp = PREV.br_blockcount - new->br_blockcount;
2077		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2078		xfs_bmbt_set_blockcount(ep, temp);
2079		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2080			new->br_startoff, new->br_startblock,
2081			new->br_blockcount + RIGHT.br_blockcount,
2082			RIGHT.br_state);
2083		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2084		if (bma->cur == NULL)
2085			rval = XFS_ILOG_DEXT;
2086		else {
2087			rval = 0;
2088			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2089					RIGHT.br_startblock,
2090					RIGHT.br_blockcount, &i);
2091			if (error)
2092				goto done;
2093			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2094			error = xfs_bmbt_update(bma->cur, new->br_startoff,
2095					new->br_startblock,
2096					new->br_blockcount +
2097					RIGHT.br_blockcount,
2098					RIGHT.br_state);
2099			if (error)
2100				goto done;
2101		}
2102
2103		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2104			startblockval(PREV.br_startblock));
2105		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2106		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2107		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2108
2109		bma->idx++;
2110		break;
2111
2112	case BMAP_RIGHT_FILLING:
2113		/*
2114		 * Filling in the last part of a previous delayed allocation.
2115		 * The right neighbor is not contiguous.
2116		 */
2117		temp = PREV.br_blockcount - new->br_blockcount;
2118		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2119		xfs_bmbt_set_blockcount(ep, temp);
2120		xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2121		bma->ip->i_d.di_nextents++;
2122		if (bma->cur == NULL)
2123			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2124		else {
2125			rval = XFS_ILOG_CORE;
2126			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2127					new->br_startblock, new->br_blockcount,
2128					&i);
2129			if (error)
2130				goto done;
2131			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2132			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2133			error = xfs_btree_insert(bma->cur, &i);
2134			if (error)
2135				goto done;
2136			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2137		}
2138
2139		if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2140			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2141				bma->firstblock, bma->flist, &bma->cur, 1,
2142				&tmp_rval, XFS_DATA_FORK);
2143			rval |= tmp_rval;
2144			if (error)
2145				goto done;
2146		}
2147		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2148			startblockval(PREV.br_startblock) -
2149			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
2150		ep = xfs_iext_get_ext(ifp, bma->idx);
2151		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2152		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2153
2154		bma->idx++;
2155		break;
2156
2157	case 0:
2158		/*
2159		 * Filling in the middle part of a previous delayed allocation.
2160		 * Contiguity is impossible here.
2161		 * This case is avoided almost all the time.
2162		 *
2163		 * We start with a delayed allocation:
2164		 *
2165		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2166		 *  PREV @ idx
2167		 *
2168	         * and we are allocating:
2169		 *                     +rrrrrrrrrrrrrrrrr+
2170		 *			      new
2171		 *
2172		 * and we set it up for insertion as:
2173		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2174		 *                            new
2175		 *  PREV @ idx          LEFT              RIGHT
2176		 *                      inserted at idx + 1
2177		 */
2178		temp = new->br_startoff - PREV.br_startoff;
2179		temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2180		trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2181		xfs_bmbt_set_blockcount(ep, temp);	/* truncate PREV */
2182		LEFT = *new;
2183		RIGHT.br_state = PREV.br_state;
2184		RIGHT.br_startblock = nullstartblock(
2185				(int)xfs_bmap_worst_indlen(bma->ip, temp2));
2186		RIGHT.br_startoff = new_endoff;
2187		RIGHT.br_blockcount = temp2;
2188		/* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2189		xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2190		bma->ip->i_d.di_nextents++;
2191		if (bma->cur == NULL)
2192			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2193		else {
2194			rval = XFS_ILOG_CORE;
2195			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2196					new->br_startblock, new->br_blockcount,
2197					&i);
2198			if (error)
2199				goto done;
2200			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2201			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2202			error = xfs_btree_insert(bma->cur, &i);
2203			if (error)
2204				goto done;
2205			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2206		}
2207
2208		if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2209			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2210					bma->firstblock, bma->flist, &bma->cur,
2211					1, &tmp_rval, XFS_DATA_FORK);
2212			rval |= tmp_rval;
2213			if (error)
2214				goto done;
2215		}
2216		temp = xfs_bmap_worst_indlen(bma->ip, temp);
2217		temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2218		diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2219			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
2220		if (diff > 0) {
2221			error = xfs_icsb_modify_counters(bma->ip->i_mount,
2222					XFS_SBS_FDBLOCKS,
2223					-((int64_t)diff), 0);
2224			ASSERT(!error);
2225			if (error)
2226				goto done;
2227		}
2228
2229		ep = xfs_iext_get_ext(ifp, bma->idx);
2230		xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2231		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2232		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2233		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2234			nullstartblock((int)temp2));
2235		trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2236
2237		bma->idx++;
2238		da_new = temp + temp2;
2239		break;
2240
2241	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2242	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2243	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2244	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2245	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2246	case BMAP_LEFT_CONTIG:
2247	case BMAP_RIGHT_CONTIG:
2248		/*
2249		 * These cases are all impossible.
2250		 */
2251		ASSERT(0);
2252	}
2253
2254	/* convert to a btree if necessary */
2255	if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2256		int	tmp_logflags;	/* partial log flag return val */
2257
2258		ASSERT(bma->cur == NULL);
2259		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2260				bma->firstblock, bma->flist, &bma->cur,
2261				da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2262		bma->logflags |= tmp_logflags;
2263		if (error)
2264			goto done;
2265	}
2266
2267	/* adjust for changes in reserved delayed indirect blocks */
2268	if (da_old || da_new) {
2269		temp = da_new;
2270		if (bma->cur)
2271			temp += bma->cur->bc_private.b.allocated;
2272		ASSERT(temp <= da_old);
2273		if (temp < da_old)
2274			xfs_icsb_modify_counters(bma->ip->i_mount,
2275					XFS_SBS_FDBLOCKS,
2276					(int64_t)(da_old - temp), 0);
2277	}
2278
2279	/* clear out the allocated field, done with it now in any case. */
2280	if (bma->cur)
2281		bma->cur->bc_private.b.allocated = 0;
2282
2283	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2284done:
2285	bma->logflags |= rval;
2286	return error;
2287#undef	LEFT
2288#undef	RIGHT
2289#undef	PREV
2290}
2291
2292/*
2293 * Convert an unwritten allocation to a real allocation or vice versa.
2294 */
2295STATIC int				/* error */
2296xfs_bmap_add_extent_unwritten_real(
2297	struct xfs_trans	*tp,
2298	xfs_inode_t		*ip,	/* incore inode pointer */
2299	xfs_extnum_t		*idx,	/* extent number to update/insert */
2300	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2301	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2302	xfs_fsblock_t		*first,	/* pointer to firstblock variable */
2303	xfs_bmap_free_t		*flist,	/* list of extents to be freed */
2304	int			*logflagsp) /* inode logging flags */
2305{
2306	xfs_btree_cur_t		*cur;	/* btree cursor */
2307	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
2308	int			error;	/* error return value */
2309	int			i;	/* temp state */
2310	xfs_ifork_t		*ifp;	/* inode fork pointer */
2311	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2312	xfs_exntst_t		newext;	/* new extent state */
2313	xfs_exntst_t		oldext;	/* old extent state */
2314	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2315					/* left is 0, right is 1, prev is 2 */
2316	int			rval=0;	/* return value (logging flags) */
2317	int			state = 0;/* state bits, accessed thru macros */
2318
2319	*logflagsp = 0;
2320
2321	cur = *curp;
2322	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2323
2324	ASSERT(*idx >= 0);
2325	ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2326	ASSERT(!isnullstartblock(new->br_startblock));
2327
2328	XFS_STATS_INC(xs_add_exlist);
2329
2330#define	LEFT		r[0]
2331#define	RIGHT		r[1]
2332#define	PREV		r[2]
2333
2334	/*
2335	 * Set up a bunch of variables to make the tests simpler.
2336	 */
2337	error = 0;
2338	ep = xfs_iext_get_ext(ifp, *idx);
2339	xfs_bmbt_get_all(ep, &PREV);
2340	newext = new->br_state;
2341	oldext = (newext == XFS_EXT_UNWRITTEN) ?
2342		XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2343	ASSERT(PREV.br_state == oldext);
2344	new_endoff = new->br_startoff + new->br_blockcount;
2345	ASSERT(PREV.br_startoff <= new->br_startoff);
2346	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2347
2348	/*
2349	 * Set flags determining what part of the previous oldext allocation
2350	 * extent is being replaced by a newext allocation.
2351	 */
2352	if (PREV.br_startoff == new->br_startoff)
2353		state |= BMAP_LEFT_FILLING;
2354	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2355		state |= BMAP_RIGHT_FILLING;
2356
2357	/*
2358	 * Check and set flags if this segment has a left neighbor.
2359	 * Don't set contiguous if the combined extent would be too large.
2360	 */
2361	if (*idx > 0) {
2362		state |= BMAP_LEFT_VALID;
2363		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2364
2365		if (isnullstartblock(LEFT.br_startblock))
2366			state |= BMAP_LEFT_DELAY;
2367	}
2368
2369	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2370	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2371	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2372	    LEFT.br_state == newext &&
2373	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2374		state |= BMAP_LEFT_CONTIG;
2375
2376	/*
2377	 * Check and set flags if this segment has a right neighbor.
2378	 * Don't set contiguous if the combined extent would be too large.
2379	 * Also check for all-three-contiguous being too large.
2380	 */
2381	if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2382		state |= BMAP_RIGHT_VALID;
2383		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2384		if (isnullstartblock(RIGHT.br_startblock))
2385			state |= BMAP_RIGHT_DELAY;
2386	}
2387
2388	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2389	    new_endoff == RIGHT.br_startoff &&
2390	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2391	    newext == RIGHT.br_state &&
2392	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2393	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2394		       BMAP_RIGHT_FILLING)) !=
2395		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2396		       BMAP_RIGHT_FILLING) ||
2397	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2398			<= MAXEXTLEN))
2399		state |= BMAP_RIGHT_CONTIG;
2400
2401	/*
2402	 * Switch out based on the FILLING and CONTIG state bits.
2403	 */
2404	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2405			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2406	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2407	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2408		/*
2409		 * Setting all of a previous oldext extent to newext.
2410		 * The left and right neighbors are both contiguous with new.
2411		 */
2412		--*idx;
2413
2414		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2415		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2416			LEFT.br_blockcount + PREV.br_blockcount +
2417			RIGHT.br_blockcount);
2418		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2419
2420		xfs_iext_remove(ip, *idx + 1, 2, state);
2421		ip->i_d.di_nextents -= 2;
2422		if (cur == NULL)
2423			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2424		else {
2425			rval = XFS_ILOG_CORE;
2426			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2427					RIGHT.br_startblock,
2428					RIGHT.br_blockcount, &i)))
2429				goto done;
2430			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2431			if ((error = xfs_btree_delete(cur, &i)))
2432				goto done;
2433			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2434			if ((error = xfs_btree_decrement(cur, 0, &i)))
2435				goto done;
2436			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2437			if ((error = xfs_btree_delete(cur, &i)))
2438				goto done;
2439			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2440			if ((error = xfs_btree_decrement(cur, 0, &i)))
2441				goto done;
2442			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2443			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2444				LEFT.br_startblock,
2445				LEFT.br_blockcount + PREV.br_blockcount +
2446				RIGHT.br_blockcount, LEFT.br_state)))
2447				goto done;
2448		}
2449		break;
2450
2451	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2452		/*
2453		 * Setting all of a previous oldext extent to newext.
2454		 * The left neighbor is contiguous, the right is not.
2455		 */
2456		--*idx;
2457
2458		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2459		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2460			LEFT.br_blockcount + PREV.br_blockcount);
2461		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2462
2463		xfs_iext_remove(ip, *idx + 1, 1, state);
2464		ip->i_d.di_nextents--;
2465		if (cur == NULL)
2466			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2467		else {
2468			rval = XFS_ILOG_CORE;
2469			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2470					PREV.br_startblock, PREV.br_blockcount,
2471					&i)))
2472				goto done;
2473			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2474			if ((error = xfs_btree_delete(cur, &i)))
2475				goto done;
2476			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2477			if ((error = xfs_btree_decrement(cur, 0, &i)))
2478				goto done;
2479			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2480			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2481				LEFT.br_startblock,
2482				LEFT.br_blockcount + PREV.br_blockcount,
2483				LEFT.br_state)))
2484				goto done;
2485		}
2486		break;
2487
2488	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2489		/*
2490		 * Setting all of a previous oldext extent to newext.
2491		 * The right neighbor is contiguous, the left is not.
2492		 */
2493		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2494		xfs_bmbt_set_blockcount(ep,
2495			PREV.br_blockcount + RIGHT.br_blockcount);
2496		xfs_bmbt_set_state(ep, newext);
2497		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2498		xfs_iext_remove(ip, *idx + 1, 1, state);
2499		ip->i_d.di_nextents--;
2500		if (cur == NULL)
2501			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2502		else {
2503			rval = XFS_ILOG_CORE;
2504			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2505					RIGHT.br_startblock,
2506					RIGHT.br_blockcount, &i)))
2507				goto done;
2508			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2509			if ((error = xfs_btree_delete(cur, &i)))
2510				goto done;
2511			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2512			if ((error = xfs_btree_decrement(cur, 0, &i)))
2513				goto done;
2514			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2515			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2516				new->br_startblock,
2517				new->br_blockcount + RIGHT.br_blockcount,
2518				newext)))
2519				goto done;
2520		}
2521		break;
2522
2523	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2524		/*
2525		 * Setting all of a previous oldext extent to newext.
2526		 * Neither the left nor right neighbors are contiguous with
2527		 * the new one.
2528		 */
2529		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2530		xfs_bmbt_set_state(ep, newext);
2531		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2532
2533		if (cur == NULL)
2534			rval = XFS_ILOG_DEXT;
2535		else {
2536			rval = 0;
2537			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2538					new->br_startblock, new->br_blockcount,
2539					&i)))
2540				goto done;
2541			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2542			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2543				new->br_startblock, new->br_blockcount,
2544				newext)))
2545				goto done;
2546		}
2547		break;
2548
2549	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2550		/*
2551		 * Setting the first part of a previous oldext extent to newext.
2552		 * The left neighbor is contiguous.
2553		 */
2554		trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2555		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2556			LEFT.br_blockcount + new->br_blockcount);
2557		xfs_bmbt_set_startoff(ep,
2558			PREV.br_startoff + new->br_blockcount);
2559		trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2560
2561		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2562		xfs_bmbt_set_startblock(ep,
2563			new->br_startblock + new->br_blockcount);
2564		xfs_bmbt_set_blockcount(ep,
2565			PREV.br_blockcount - new->br_blockcount);
2566		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2567
2568		--*idx;
2569
2570		if (cur == NULL)
2571			rval = XFS_ILOG_DEXT;
2572		else {
2573			rval = 0;
2574			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2575					PREV.br_startblock, PREV.br_blockcount,
2576					&i)))
2577				goto done;
2578			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2579			if ((error = xfs_bmbt_update(cur,
2580				PREV.br_startoff + new->br_blockcount,
2581				PREV.br_startblock + new->br_blockcount,
2582				PREV.br_blockcount - new->br_blockcount,
2583				oldext)))
2584				goto done;
2585			if ((error = xfs_btree_decrement(cur, 0, &i)))
2586				goto done;
2587			error = xfs_bmbt_update(cur, LEFT.br_startoff,
2588				LEFT.br_startblock,
2589				LEFT.br_blockcount + new->br_blockcount,
2590				LEFT.br_state);
2591			if (error)
2592				goto done;
2593		}
2594		break;
2595
2596	case BMAP_LEFT_FILLING:
2597		/*
2598		 * Setting the first part of a previous oldext extent to newext.
2599		 * The left neighbor is not contiguous.
2600		 */
2601		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2602		ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2603		xfs_bmbt_set_startoff(ep, new_endoff);
2604		xfs_bmbt_set_blockcount(ep,
2605			PREV.br_blockcount - new->br_blockcount);
2606		xfs_bmbt_set_startblock(ep,
2607			new->br_startblock + new->br_blockcount);
2608		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2609
2610		xfs_iext_insert(ip, *idx, 1, new, state);
2611		ip->i_d.di_nextents++;
2612		if (cur == NULL)
2613			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2614		else {
2615			rval = XFS_ILOG_CORE;
2616			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2617					PREV.br_startblock, PREV.br_blockcount,
2618					&i)))
2619				goto done;
2620			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2621			if ((error = xfs_bmbt_update(cur,
2622				PREV.br_startoff + new->br_blockcount,
2623				PREV.br_startblock + new->br_blockcount,
2624				PREV.br_blockcount - new->br_blockcount,
2625				oldext)))
2626				goto done;
2627			cur->bc_rec.b = *new;
2628			if ((error = xfs_btree_insert(cur, &i)))
2629				goto done;
2630			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2631		}
2632		break;
2633
2634	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2635		/*
2636		 * Setting the last part of a previous oldext extent to newext.
2637		 * The right neighbor is contiguous with the new allocation.
2638		 */
2639		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2640		xfs_bmbt_set_blockcount(ep,
2641			PREV.br_blockcount - new->br_blockcount);
2642		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2643
2644		++*idx;
2645
2646		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2647		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2648			new->br_startoff, new->br_startblock,
2649			new->br_blockcount + RIGHT.br_blockcount, newext);
2650		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2651
2652		if (cur == NULL)
2653			rval = XFS_ILOG_DEXT;
2654		else {
2655			rval = 0;
2656			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2657					PREV.br_startblock,
2658					PREV.br_blockcount, &i)))
2659				goto done;
2660			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2661			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2662				PREV.br_startblock,
2663				PREV.br_blockcount - new->br_blockcount,
2664				oldext)))
2665				goto done;
2666			if ((error = xfs_btree_increment(cur, 0, &i)))
2667				goto done;
2668			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2669				new->br_startblock,
2670				new->br_blockcount + RIGHT.br_blockcount,
2671				newext)))
2672				goto done;
2673		}
2674		break;
2675
2676	case BMAP_RIGHT_FILLING:
2677		/*
2678		 * Setting the last part of a previous oldext extent to newext.
2679		 * The right neighbor is not contiguous.
2680		 */
2681		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2682		xfs_bmbt_set_blockcount(ep,
2683			PREV.br_blockcount - new->br_blockcount);
2684		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2685
2686		++*idx;
2687		xfs_iext_insert(ip, *idx, 1, new, state);
2688
2689		ip->i_d.di_nextents++;
2690		if (cur == NULL)
2691			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2692		else {
2693			rval = XFS_ILOG_CORE;
2694			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2695					PREV.br_startblock, PREV.br_blockcount,
2696					&i)))
2697				goto done;
2698			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2699			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2700				PREV.br_startblock,
2701				PREV.br_blockcount - new->br_blockcount,
2702				oldext)))
2703				goto done;
2704			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2705					new->br_startblock, new->br_blockcount,
2706					&i)))
2707				goto done;
2708			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2709			cur->bc_rec.b.br_state = XFS_EXT_NORM;
2710			if ((error = xfs_btree_insert(cur, &i)))
2711				goto done;
2712			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2713		}
2714		break;
2715
2716	case 0:
2717		/*
2718		 * Setting the middle part of a previous oldext extent to
2719		 * newext.  Contiguity is impossible here.
2720		 * One extent becomes three extents.
2721		 */
2722		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2723		xfs_bmbt_set_blockcount(ep,
2724			new->br_startoff - PREV.br_startoff);
2725		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2726
2727		r[0] = *new;
2728		r[1].br_startoff = new_endoff;
2729		r[1].br_blockcount =
2730			PREV.br_startoff + PREV.br_blockcount - new_endoff;
2731		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2732		r[1].br_state = oldext;
2733
2734		++*idx;
2735		xfs_iext_insert(ip, *idx, 2, &r[0], state);
2736
2737		ip->i_d.di_nextents += 2;
2738		if (cur == NULL)
2739			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2740		else {
2741			rval = XFS_ILOG_CORE;
2742			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2743					PREV.br_startblock, PREV.br_blockcount,
2744					&i)))
2745				goto done;
2746			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2747			/* new right extent - oldext */
2748			if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2749				r[1].br_startblock, r[1].br_blockcount,
2750				r[1].br_state)))
2751				goto done;
2752			/* new left extent - oldext */
2753			cur->bc_rec.b = PREV;
2754			cur->bc_rec.b.br_blockcount =
2755				new->br_startoff - PREV.br_startoff;
2756			if ((error = xfs_btree_insert(cur, &i)))
2757				goto done;
2758			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2759			/*
2760			 * Reset the cursor to the position of the new extent
2761			 * we are about to insert as we can't trust it after
2762			 * the previous insert.
2763			 */
2764			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2765					new->br_startblock, new->br_blockcount,
2766					&i)))
2767				goto done;
2768			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2769			/* new middle extent - newext */
2770			cur->bc_rec.b.br_state = new->br_state;
2771			if ((error = xfs_btree_insert(cur, &i)))
2772				goto done;
2773			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2774		}
2775		break;
2776
2777	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2778	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2779	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2780	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2781	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2782	case BMAP_LEFT_CONTIG:
2783	case BMAP_RIGHT_CONTIG:
2784		/*
2785		 * These cases are all impossible.
2786		 */
2787		ASSERT(0);
2788	}
2789
2790	/* convert to a btree if necessary */
2791	if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2792		int	tmp_logflags;	/* partial log flag return val */
2793
2794		ASSERT(cur == NULL);
2795		error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2796				0, &tmp_logflags, XFS_DATA_FORK);
2797		*logflagsp |= tmp_logflags;
2798		if (error)
2799			goto done;
2800	}
2801
2802	/* clear out the allocated field, done with it now in any case. */
2803	if (cur) {
2804		cur->bc_private.b.allocated = 0;
2805		*curp = cur;
2806	}
2807
2808	xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2809done:
2810	*logflagsp |= rval;
2811	return error;
2812#undef	LEFT
2813#undef	RIGHT
2814#undef	PREV
2815}
2816
2817/*
2818 * Convert a hole to a delayed allocation.
2819 */
2820STATIC void
2821xfs_bmap_add_extent_hole_delay(
2822	xfs_inode_t		*ip,	/* incore inode pointer */
2823	xfs_extnum_t		*idx,	/* extent number to update/insert */
2824	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2825{
2826	xfs_ifork_t		*ifp;	/* inode fork pointer */
2827	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2828	xfs_filblks_t		newlen=0;	/* new indirect size */
2829	xfs_filblks_t		oldlen=0;	/* old indirect size */
2830	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2831	int			state;  /* state bits, accessed thru macros */
2832	xfs_filblks_t		temp=0;	/* temp for indirect calculations */
2833
2834	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2835	state = 0;
2836	ASSERT(isnullstartblock(new->br_startblock));
2837
2838	/*
2839	 * Check and set flags if this segment has a left neighbor
2840	 */
2841	if (*idx > 0) {
2842		state |= BMAP_LEFT_VALID;
2843		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2844
2845		if (isnullstartblock(left.br_startblock))
2846			state |= BMAP_LEFT_DELAY;
2847	}
2848
2849	/*
2850	 * Check and set flags if the current (right) segment exists.
2851	 * If it doesn't exist, we're converting the hole at end-of-file.
2852	 */
2853	if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2854		state |= BMAP_RIGHT_VALID;
2855		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2856
2857		if (isnullstartblock(right.br_startblock))
2858			state |= BMAP_RIGHT_DELAY;
2859	}
2860
2861	/*
2862	 * Set contiguity flags on the left and right neighbors.
2863	 * Don't let extents get too large, even if the pieces are contiguous.
2864	 */
2865	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2866	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2867	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2868		state |= BMAP_LEFT_CONTIG;
2869
2870	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2871	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2872	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2873	    (!(state & BMAP_LEFT_CONTIG) ||
2874	     (left.br_blockcount + new->br_blockcount +
2875	      right.br_blockcount <= MAXEXTLEN)))
2876		state |= BMAP_RIGHT_CONTIG;
2877
2878	/*
2879	 * Switch out based on the contiguity flags.
2880	 */
2881	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2882	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2883		/*
2884		 * New allocation is contiguous with delayed allocations
2885		 * on the left and on the right.
2886		 * Merge all three into a single extent record.
2887		 */
2888		--*idx;
2889		temp = left.br_blockcount + new->br_blockcount +
2890			right.br_blockcount;
2891
2892		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2893		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2894		oldlen = startblockval(left.br_startblock) +
2895			startblockval(new->br_startblock) +
2896			startblockval(right.br_startblock);
2897		newlen = xfs_bmap_worst_indlen(ip, temp);
2898		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2899			nullstartblock((int)newlen));
2900		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2901
2902		xfs_iext_remove(ip, *idx + 1, 1, state);
2903		break;
2904
2905	case BMAP_LEFT_CONTIG:
2906		/*
2907		 * New allocation is contiguous with a delayed allocation
2908		 * on the left.
2909		 * Merge the new allocation with the left neighbor.
2910		 */
2911		--*idx;
2912		temp = left.br_blockcount + new->br_blockcount;
2913
2914		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2915		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2916		oldlen = startblockval(left.br_startblock) +
2917			startblockval(new->br_startblock);
2918		newlen = xfs_bmap_worst_indlen(ip, temp);
2919		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2920			nullstartblock((int)newlen));
2921		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2922		break;
2923
2924	case BMAP_RIGHT_CONTIG:
2925		/*
2926		 * New allocation is contiguous with a delayed allocation
2927		 * on the right.
2928		 * Merge the new allocation with the right neighbor.
2929		 */
2930		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2931		temp = new->br_blockcount + right.br_blockcount;
2932		oldlen = startblockval(new->br_startblock) +
2933			startblockval(right.br_startblock);
2934		newlen = xfs_bmap_worst_indlen(ip, temp);
2935		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2936			new->br_startoff,
2937			nullstartblock((int)newlen), temp, right.br_state);
2938		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2939		break;
2940
2941	case 0:
2942		/*
2943		 * New allocation is not contiguous with another
2944		 * delayed allocation.
2945		 * Insert a new entry.
2946		 */
2947		oldlen = newlen = 0;
2948		xfs_iext_insert(ip, *idx, 1, new, state);
2949		break;
2950	}
2951	if (oldlen != newlen) {
2952		ASSERT(oldlen > newlen);
2953		xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2954			(int64_t)(oldlen - newlen), 0);
2955		/*
2956		 * Nothing to do for disk quota accounting here.
2957		 */
2958	}
2959}
2960
2961/*
2962 * Convert a hole to a real allocation.
2963 */
2964STATIC int				/* error */
2965xfs_bmap_add_extent_hole_real(
2966	struct xfs_bmalloca	*bma,
2967	int			whichfork)
2968{
2969	struct xfs_bmbt_irec	*new = &bma->got;
2970	int			error;	/* error return value */
2971	int			i;	/* temp state */
2972	xfs_ifork_t		*ifp;	/* inode fork pointer */
2973	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2974	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2975	int			rval=0;	/* return value (logging flags) */
2976	int			state;	/* state bits, accessed thru macros */
2977
2978	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2979
2980	ASSERT(bma->idx >= 0);
2981	ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2982	ASSERT(!isnullstartblock(new->br_startblock));
2983	ASSERT(!bma->cur ||
2984	       !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2985
2986	XFS_STATS_INC(xs_add_exlist);
2987
2988	state = 0;
2989	if (whichfork == XFS_ATTR_FORK)
2990		state |= BMAP_ATTRFORK;
2991
2992	/*
2993	 * Check and set flags if this segment has a left neighbor.
2994	 */
2995	if (bma->idx > 0) {
2996		state |= BMAP_LEFT_VALID;
2997		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2998		if (isnullstartblock(left.br_startblock))
2999			state |= BMAP_LEFT_DELAY;
3000	}
3001
3002	/*
3003	 * Check and set flags if this segment has a current value.
3004	 * Not true if we're inserting into the "hole" at eof.
3005	 */
3006	if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3007		state |= BMAP_RIGHT_VALID;
3008		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3009		if (isnullstartblock(right.br_startblock))
3010			state |= BMAP_RIGHT_DELAY;
3011	}
3012
3013	/*
3014	 * We're inserting a real allocation between "left" and "right".
3015	 * Set the contiguity flags.  Don't let extents get too large.
3016	 */
3017	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3018	    left.br_startoff + left.br_blockcount == new->br_startoff &&
3019	    left.br_startblock + left.br_blockcount == new->br_startblock &&
3020	    left.br_state == new->br_state &&
3021	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3022		state |= BMAP_LEFT_CONTIG;
3023
3024	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3025	    new->br_startoff + new->br_blockcount == right.br_startoff &&
3026	    new->br_startblock + new->br_blockcount == right.br_startblock &&
3027	    new->br_state == right.br_state &&
3028	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3029	    (!(state & BMAP_LEFT_CONTIG) ||
3030	     left.br_blockcount + new->br_blockcount +
3031	     right.br_blockcount <= MAXEXTLEN))
3032		state |= BMAP_RIGHT_CONTIG;
3033
3034	error = 0;
3035	/*
3036	 * Select which case we're in here, and implement it.
3037	 */
3038	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3039	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3040		/*
3041		 * New allocation is contiguous with real allocations on the
3042		 * left and on the right.
3043		 * Merge all three into a single extent record.
3044		 */
3045		--bma->idx;
3046		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3047		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3048			left.br_blockcount + new->br_blockcount +
3049			right.br_blockcount);
3050		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3051
3052		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3053
3054		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3055			XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3056		if (bma->cur == NULL) {
3057			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3058		} else {
3059			rval = XFS_ILOG_CORE;
3060			error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3061					right.br_startblock, right.br_blockcount,
3062					&i);
3063			if (error)
3064				goto done;
3065			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3066			error = xfs_btree_delete(bma->cur, &i);
3067			if (error)
3068				goto done;
3069			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3070			error = xfs_btree_decrement(bma->cur, 0, &i);
3071			if (error)
3072				goto done;
3073			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3074			error = xfs_bmbt_update(bma->cur, left.br_startoff,
3075					left.br_startblock,
3076					left.br_blockcount +
3077						new->br_blockcount +
3078						right.br_blockcount,
3079					left.br_state);
3080			if (error)
3081				goto done;
3082		}
3083		break;
3084
3085	case BMAP_LEFT_CONTIG:
3086		/*
3087		 * New allocation is contiguous with a real allocation
3088		 * on the left.
3089		 * Merge the new allocation with the left neighbor.
3090		 */
3091		--bma->idx;
3092		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3093		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3094			left.br_blockcount + new->br_blockcount);
3095		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3096
3097		if (bma->cur == NULL) {
3098			rval = xfs_ilog_fext(whichfork);
3099		} else {
3100			rval = 0;
3101			error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3102					left.br_startblock, left.br_blockcount,
3103					&i);
3104			if (error)
3105				goto done;
3106			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3107			error = xfs_bmbt_update(bma->cur, left.br_startoff,
3108					left.br_startblock,
3109					left.br_blockcount +
3110						new->br_blockcount,
3111					left.br_state);
3112			if (error)
3113				goto done;
3114		}
3115		break;
3116
3117	case BMAP_RIGHT_CONTIG:
3118		/*
3119		 * New allocation is contiguous with a real allocation
3120		 * on the right.
3121		 * Merge the new allocation with the right neighbor.
3122		 */
3123		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3124		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3125			new->br_startoff, new->br_startblock,
3126			new->br_blockcount + right.br_blockcount,
3127			right.br_state);
3128		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3129
3130		if (bma->cur == NULL) {
3131			rval = xfs_ilog_fext(whichfork);
3132		} else {
3133			rval = 0;
3134			error = xfs_bmbt_lookup_eq(bma->cur,
3135					right.br_startoff,
3136					right.br_startblock,
3137					right.br_blockcount, &i);
3138			if (error)
3139				goto done;
3140			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3141			error = xfs_bmbt_update(bma->cur, new->br_startoff,
3142					new->br_startblock,
3143					new->br_blockcount +
3144						right.br_blockcount,
3145					right.br_state);
3146			if (error)
3147				goto done;
3148		}
3149		break;
3150
3151	case 0:
3152		/*
3153		 * New allocation is not contiguous with another
3154		 * real allocation.
3155		 * Insert a new entry.
3156		 */
3157		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3158		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3159			XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3160		if (bma->cur == NULL) {
3161			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3162		} else {
3163			rval = XFS_ILOG_CORE;
3164			error = xfs_bmbt_lookup_eq(bma->cur,
3165					new->br_startoff,
3166					new->br_startblock,
3167					new->br_blockcount, &i);
3168			if (error)
3169				goto done;
3170			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3171			bma->cur->bc_rec.b.br_state = new->br_state;
3172			error = xfs_btree_insert(bma->cur, &i);
3173			if (error)
3174				goto done;
3175			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3176		}
3177		break;
3178	}
3179
3180	/* convert to a btree if necessary */
3181	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3182		int	tmp_logflags;	/* partial log flag return val */
3183
3184		ASSERT(bma->cur == NULL);
3185		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3186				bma->firstblock, bma->flist, &bma->cur,
3187				0, &tmp_logflags, whichfork);
3188		bma->logflags |= tmp_logflags;
3189		if (error)
3190			goto done;
3191	}
3192
3193	/* clear out the allocated field, done with it now in any case. */
3194	if (bma->cur)
3195		bma->cur->bc_private.b.allocated = 0;
3196
3197	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3198done:
3199	bma->logflags |= rval;
3200	return error;
3201}
3202
3203/*
3204 * Functions used in the extent read, allocate and remove paths
3205 */
3206
3207/*
3208 * Adjust the size of the new extent based on di_extsize and rt extsize.
3209 */
3210int
3211xfs_bmap_extsize_align(
3212	xfs_mount_t	*mp,
3213	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
3214	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
3215	xfs_extlen_t	extsz,		/* align to this extent size */
3216	int		rt,		/* is this a realtime inode? */
3217	int		eof,		/* is extent at end-of-file? */
3218	int		delay,		/* creating delalloc extent? */
3219	int		convert,	/* overwriting unwritten extent? */
3220	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
3221	xfs_extlen_t	*lenp)		/* in/out: aligned length */
3222{
3223	xfs_fileoff_t	orig_off;	/* original offset */
3224	xfs_extlen_t	orig_alen;	/* original length */
3225	xfs_fileoff_t	orig_end;	/* original off+len */
3226	xfs_fileoff_t	nexto;		/* next file offset */
3227	xfs_fileoff_t	prevo;		/* previous file offset */
3228	xfs_fileoff_t	align_off;	/* temp for offset */
3229	xfs_extlen_t	align_alen;	/* temp for length */
3230	xfs_extlen_t	temp;		/* temp for calculations */
3231
3232	if (convert)
3233		return 0;
3234
3235	orig_off = align_off = *offp;
3236	orig_alen = align_alen = *lenp;
3237	orig_end = orig_off + orig_alen;
3238
3239	/*
3240	 * If this request overlaps an existing extent, then don't
3241	 * attempt to perform any additional alignment.
3242	 */
3243	if (!delay && !eof &&
3244	    (orig_off >= gotp->br_startoff) &&
3245	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3246		return 0;
3247	}
3248
3249	/*
3250	 * If the file offset is unaligned vs. the extent size
3251	 * we need to align it.  This will be possible unless
3252	 * the file was previously written with a kernel that didn't
3253	 * perform this alignment, or if a truncate shot us in the
3254	 * foot.
3255	 */
3256	temp = do_mod(orig_off, extsz);
3257	if (temp) {
3258		align_alen += temp;
3259		align_off -= temp;
3260	}
3261	/*
3262	 * Same adjustment for the end of the requested area.
3263	 */
3264	if ((temp = (align_alen % extsz))) {
3265		align_alen += extsz - temp;
3266	}
3267	/*
3268	 * If the previous block overlaps with this proposed allocation
3269	 * then move the start forward without adjusting the length.
3270	 */
3271	if (prevp->br_startoff != NULLFILEOFF) {
3272		if (prevp->br_startblock == HOLESTARTBLOCK)
3273			prevo = prevp->br_startoff;
3274		else
3275			prevo = prevp->br_startoff + prevp->br_blockcount;
3276	} else
3277		prevo = 0;
3278	if (align_off != orig_off && align_off < prevo)
3279		align_off = prevo;
3280	/*
3281	 * If the next block overlaps with this proposed allocation
3282	 * then move the start back without adjusting the length,
3283	 * but not before offset 0.
3284	 * This may of course make the start overlap previous block,
3285	 * and if we hit the offset 0 limit then the next block
3286	 * can still overlap too.
3287	 */
3288	if (!eof && gotp->br_startoff != NULLFILEOFF) {
3289		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3290		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3291			nexto = gotp->br_startoff + gotp->br_blockcount;
3292		else
3293			nexto = gotp->br_startoff;
3294	} else
3295		nexto = NULLFILEOFF;
3296	if (!eof &&
3297	    align_off + align_alen != orig_end &&
3298	    align_off + align_alen > nexto)
3299		align_off = nexto > align_alen ? nexto - align_alen : 0;
3300	/*
3301	 * If we're now overlapping the next or previous extent that
3302	 * means we can't fit an extsz piece in this hole.  Just move
3303	 * the start forward to the first valid spot and set
3304	 * the length so we hit the end.
3305	 */
3306	if (align_off != orig_off && align_off < prevo)
3307		align_off = prevo;
3308	if (align_off + align_alen != orig_end &&
3309	    align_off + align_alen > nexto &&
3310	    nexto != NULLFILEOFF) {
3311		ASSERT(nexto > prevo);
3312		align_alen = nexto - align_off;
3313	}
3314
3315	/*
3316	 * If realtime, and the result isn't a multiple of the realtime
3317	 * extent size we need to remove blocks until it is.
3318	 */
3319	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3320		/*
3321		 * We're not covering the original request, or
3322		 * we won't be able to once we fix the length.
3323		 */
3324		if (orig_off < align_off ||
3325		    orig_end > align_off + align_alen ||
3326		    align_alen - temp < orig_alen)
3327			return XFS_ERROR(EINVAL);
3328		/*
3329		 * Try to fix it by moving the start up.
3330		 */
3331		if (align_off + temp <= orig_off) {
3332			align_alen -= temp;
3333			align_off += temp;
3334		}
3335		/*
3336		 * Try to fix it by moving the end in.
3337		 */
3338		else if (align_off + align_alen - temp >= orig_end)
3339			align_alen -= temp;
3340		/*
3341		 * Set the start to the minimum then trim the length.
3342		 */
3343		else {
3344			align_alen -= orig_off - align_off;
3345			align_off = orig_off;
3346			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3347		}
3348		/*
3349		 * Result doesn't cover the request, fail it.
3350		 */
3351		if (orig_off < align_off || orig_end > align_off + align_alen)
3352			return XFS_ERROR(EINVAL);
3353	} else {
3354		ASSERT(orig_off >= align_off);
3355		ASSERT(orig_end <= align_off + align_alen);
3356	}
3357
3358#ifdef DEBUG
3359	if (!eof && gotp->br_startoff != NULLFILEOFF)
3360		ASSERT(align_off + align_alen <= gotp->br_startoff);
3361	if (prevp->br_startoff != NULLFILEOFF)
3362		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3363#endif
3364
3365	*lenp = align_alen;
3366	*offp = align_off;
3367	return 0;
3368}
3369
3370#define XFS_ALLOC_GAP_UNITS	4
3371
3372void
3373xfs_bmap_adjacent(
3374	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3375{
3376	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3377	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3378	xfs_mount_t	*mp;		/* mount point structure */
3379	int		nullfb;		/* true if ap->firstblock isn't set */
3380	int		rt;		/* true if inode is realtime */
3381
3382#define	ISVALID(x,y)	\
3383	(rt ? \
3384		(x) < mp->m_sb.sb_rblocks : \
3385		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3386		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3387		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3388
3389	mp = ap->ip->i_mount;
3390	nullfb = *ap->firstblock == NULLFSBLOCK;
3391	rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3392	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3393	/*
3394	 * If allocating at eof, and there's a previous real block,
3395	 * try to use its last block as our starting point.
3396	 */
3397	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3398	    !isnullstartblock(ap->prev.br_startblock) &&
3399	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3400		    ap->prev.br_startblock)) {
3401		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3402		/*
3403		 * Adjust for the gap between prevp and us.
3404		 */
3405		adjust = ap->offset -
3406			(ap->prev.br_startoff + ap->prev.br_blockcount);
3407		if (adjust &&
3408		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3409			ap->blkno += adjust;
3410	}
3411	/*
3412	 * If not at eof, then compare the two neighbor blocks.
3413	 * Figure out whether either one gives us a good starting point,
3414	 * and pick the better one.
3415	 */
3416	else if (!ap->eof) {
3417		xfs_fsblock_t	gotbno;		/* right side block number */
3418		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3419		xfs_fsblock_t	prevbno;	/* left side block number */
3420		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3421
3422		/*
3423		 * If there's a previous (left) block, select a requested
3424		 * start block based on it.
3425		 */
3426		if (ap->prev.br_startoff != NULLFILEOFF &&
3427		    !isnullstartblock(ap->prev.br_startblock) &&
3428		    (prevbno = ap->prev.br_startblock +
3429			       ap->prev.br_blockcount) &&
3430		    ISVALID(prevbno, ap->prev.br_startblock)) {
3431			/*
3432			 * Calculate gap to end of previous block.
3433			 */
3434			adjust = prevdiff = ap->offset -
3435				(ap->prev.br_startoff +
3436				 ap->prev.br_blockcount);
3437			/*
3438			 * Figure the startblock based on the previous block's
3439			 * end and the gap size.
3440			 * Heuristic!
3441			 * If the gap is large relative to the piece we're
3442			 * allocating, or using it gives us an invalid block
3443			 * number, then just use the end of the previous block.
3444			 */
3445			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3446			    ISVALID(prevbno + prevdiff,
3447				    ap->prev.br_startblock))
3448				prevbno += adjust;
3449			else
3450				prevdiff += adjust;
3451			/*
3452			 * If the firstblock forbids it, can't use it,
3453			 * must use default.
3454			 */
3455			if (!rt && !nullfb &&
3456			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3457				prevbno = NULLFSBLOCK;
3458		}
3459		/*
3460		 * No previous block or can't follow it, just default.
3461		 */
3462		else
3463			prevbno = NULLFSBLOCK;
3464		/*
3465		 * If there's a following (right) block, select a requested
3466		 * start block based on it.
3467		 */
3468		if (!isnullstartblock(ap->got.br_startblock)) {
3469			/*
3470			 * Calculate gap to start of next block.
3471			 */
3472			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3473			/*
3474			 * Figure the startblock based on the next block's
3475			 * start and the gap size.
3476			 */
3477			gotbno = ap->got.br_startblock;
3478			/*
3479			 * Heuristic!
3480			 * If the gap is large relative to the piece we're
3481			 * allocating, or using it gives us an invalid block
3482			 * number, then just use the start of the next block
3483			 * offset by our length.
3484			 */
3485			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3486			    ISVALID(gotbno - gotdiff, gotbno))
3487				gotbno -= adjust;
3488			else if (ISVALID(gotbno - ap->length, gotbno)) {
3489				gotbno -= ap->length;
3490				gotdiff += adjust - ap->length;
3491			} else
3492				gotdiff += adjust;
3493			/*
3494			 * If the firstblock forbids it, can't use it,
3495			 * must use default.
3496			 */
3497			if (!rt && !nullfb &&
3498			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3499				gotbno = NULLFSBLOCK;
3500		}
3501		/*
3502		 * No next block, just default.
3503		 */
3504		else
3505			gotbno = NULLFSBLOCK;
3506		/*
3507		 * If both valid, pick the better one, else the only good
3508		 * one, else ap->blkno is already set (to 0 or the inode block).
3509		 */
3510		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3511			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3512		else if (prevbno != NULLFSBLOCK)
3513			ap->blkno = prevbno;
3514		else if (gotbno != NULLFSBLOCK)
3515			ap->blkno = gotbno;
3516	}
3517#undef ISVALID
3518}
3519
3520STATIC int
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3521xfs_bmap_btalloc_nullfb(
3522	struct xfs_bmalloca	*ap,
3523	struct xfs_alloc_arg	*args,
3524	xfs_extlen_t		*blen)
3525{
3526	struct xfs_mount	*mp = ap->ip->i_mount;
3527	struct xfs_perag	*pag;
3528	xfs_agnumber_t		ag, startag;
3529	int			notinit = 0;
3530	int			error;
3531
3532	if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3533		args->type = XFS_ALLOCTYPE_NEAR_BNO;
3534	else
3535		args->type = XFS_ALLOCTYPE_START_BNO;
3536	args->total = ap->total;
3537
3538	/*
3539	 * Search for an allocation group with a single extent large enough
3540	 * for the request.  If one isn't found, then adjust the minimum
3541	 * allocation size to the largest space found.
3542	 */
3543	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3544	if (startag == NULLAGNUMBER)
3545		startag = ag = 0;
3546
3547	pag = xfs_perag_get(mp, ag);
3548	while (*blen < args->maxlen) {
3549		if (!pag->pagf_init) {
3550			error = xfs_alloc_pagf_init(mp, args->tp, ag,
3551						    XFS_ALLOC_FLAG_TRYLOCK);
3552			if (error) {
3553				xfs_perag_put(pag);
3554				return error;
3555			}
3556		}
3557
3558		/*
3559		 * See xfs_alloc_fix_freelist...
3560		 */
3561		if (pag->pagf_init) {
3562			xfs_extlen_t	longest;
3563			longest = xfs_alloc_longest_free_extent(mp, pag);
3564			if (*blen < longest)
3565				*blen = longest;
3566		} else
3567			notinit = 1;
3568
3569		if (xfs_inode_is_filestream(ap->ip)) {
3570			if (*blen >= args->maxlen)
3571				break;
3572
3573			if (ap->userdata) {
3574				/*
3575				 * If startag is an invalid AG, we've
3576				 * come here once before and
3577				 * xfs_filestream_new_ag picked the
3578				 * best currently available.
3579				 *
3580				 * Don't continue looping, since we
3581				 * could loop forever.
3582				 */
3583				if (startag == NULLAGNUMBER)
3584					break;
3585
3586				error = xfs_filestream_new_ag(ap, &ag);
3587				xfs_perag_put(pag);
3588				if (error)
3589					return error;
3590
3591				/* loop again to set 'blen'*/
3592				startag = NULLAGNUMBER;
3593				pag = xfs_perag_get(mp, ag);
3594				continue;
3595			}
3596		}
3597		if (++ag == mp->m_sb.sb_agcount)
3598			ag = 0;
3599		if (ag == startag)
3600			break;
3601		xfs_perag_put(pag);
3602		pag = xfs_perag_get(mp, ag);
3603	}
3604	xfs_perag_put(pag);
3605
3606	/*
3607	 * Since the above loop did a BUF_TRYLOCK, it is
3608	 * possible that there is space for this request.
3609	 */
3610	if (notinit || *blen < ap->minlen)
3611		args->minlen = ap->minlen;
3612	/*
3613	 * If the best seen length is less than the request
3614	 * length, use the best as the minimum.
3615	 */
3616	else if (*blen < args->maxlen)
3617		args->minlen = *blen;
3618	/*
3619	 * Otherwise we've seen an extent as big as maxlen,
3620	 * use that as the minimum.
3621	 */
3622	else
3623		args->minlen = args->maxlen;
3624
3625	/*
3626	 * set the failure fallback case to look in the selected
3627	 * AG as the stream may have moved.
3628	 */
3629	if (xfs_inode_is_filestream(ap->ip))
3630		ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3631
3632	return 0;
3633}
3634
3635STATIC int
3636xfs_bmap_btalloc(
3637	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3638{
3639	xfs_mount_t	*mp;		/* mount point structure */
3640	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3641	xfs_extlen_t	align;		/* minimum allocation alignment */
3642	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3643	xfs_agnumber_t	ag;
3644	xfs_alloc_arg_t	args;
3645	xfs_extlen_t	blen;
3646	xfs_extlen_t	nextminlen = 0;
3647	int		nullfb;		/* true if ap->firstblock isn't set */
3648	int		isaligned;
3649	int		tryagain;
3650	int		error;
3651	int		stripe_align;
3652
3653	ASSERT(ap->length);
3654
3655	mp = ap->ip->i_mount;
3656
3657	/* stripe alignment for allocation is determined by mount parameters */
3658	stripe_align = 0;
3659	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3660		stripe_align = mp->m_swidth;
3661	else if (mp->m_dalign)
3662		stripe_align = mp->m_dalign;
3663
3664	align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3665	if (unlikely(align)) {
3666		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3667						align, 0, ap->eof, 0, ap->conv,
3668						&ap->offset, &ap->length);
3669		ASSERT(!error);
3670		ASSERT(ap->length);
3671	}
3672
3673
3674	nullfb = *ap->firstblock == NULLFSBLOCK;
3675	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3676	if (nullfb) {
3677		if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3678			ag = xfs_filestream_lookup_ag(ap->ip);
3679			ag = (ag != NULLAGNUMBER) ? ag : 0;
3680			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3681		} else {
3682			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3683		}
3684	} else
3685		ap->blkno = *ap->firstblock;
3686
3687	xfs_bmap_adjacent(ap);
3688
3689	/*
3690	 * If allowed, use ap->blkno; otherwise must use firstblock since
3691	 * it's in the right allocation group.
3692	 */
3693	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3694		;
3695	else
3696		ap->blkno = *ap->firstblock;
3697	/*
3698	 * Normal allocation, done through xfs_alloc_vextent.
3699	 */
3700	tryagain = isaligned = 0;
3701	memset(&args, 0, sizeof(args));
3702	args.tp = ap->tp;
3703	args.mp = mp;
3704	args.fsbno = ap->blkno;
3705
3706	/* Trim the allocation back to the maximum an AG can fit. */
3707	args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3708	args.firstblock = *ap->firstblock;
3709	blen = 0;
3710	if (nullfb) {
3711		error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3712		if (error)
3713			return error;
3714	} else if (ap->flist->xbf_low) {
3715		if (xfs_inode_is_filestream(ap->ip))
3716			args.type = XFS_ALLOCTYPE_FIRST_AG;
3717		else
3718			args.type = XFS_ALLOCTYPE_START_BNO;
3719		args.total = args.minlen = ap->minlen;
3720	} else {
3721		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3722		args.total = ap->total;
3723		args.minlen = ap->minlen;
3724	}
3725	/* apply extent size hints if obtained earlier */
3726	if (unlikely(align)) {
3727		args.prod = align;
3728		if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3729			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3730	} else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3731		args.prod = 1;
3732		args.mod = 0;
3733	} else {
3734		args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3735		if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3736			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3737	}
3738	/*
3739	 * If we are not low on available data blocks, and the
3740	 * underlying logical volume manager is a stripe, and
3741	 * the file offset is zero then try to allocate data
3742	 * blocks on stripe unit boundary.
3743	 * NOTE: ap->aeof is only set if the allocation length
3744	 * is >= the stripe unit and the allocation offset is
3745	 * at the end of file.
3746	 */
3747	if (!ap->flist->xbf_low && ap->aeof) {
3748		if (!ap->offset) {
3749			args.alignment = stripe_align;
3750			atype = args.type;
3751			isaligned = 1;
3752			/*
3753			 * Adjust for alignment
3754			 */
3755			if (blen > args.alignment && blen <= args.maxlen)
3756				args.minlen = blen - args.alignment;
3757			args.minalignslop = 0;
3758		} else {
3759			/*
3760			 * First try an exact bno allocation.
3761			 * If it fails then do a near or start bno
3762			 * allocation with alignment turned on.
3763			 */
3764			atype = args.type;
3765			tryagain = 1;
3766			args.type = XFS_ALLOCTYPE_THIS_BNO;
3767			args.alignment = 1;
3768			/*
3769			 * Compute the minlen+alignment for the
3770			 * next case.  Set slop so that the value
3771			 * of minlen+alignment+slop doesn't go up
3772			 * between the calls.
3773			 */
3774			if (blen > stripe_align && blen <= args.maxlen)
3775				nextminlen = blen - stripe_align;
3776			else
3777				nextminlen = args.minlen;
3778			if (nextminlen + stripe_align > args.minlen + 1)
3779				args.minalignslop =
3780					nextminlen + stripe_align -
3781					args.minlen - 1;
3782			else
3783				args.minalignslop = 0;
3784		}
3785	} else {
3786		args.alignment = 1;
3787		args.minalignslop = 0;
3788	}
3789	args.minleft = ap->minleft;
3790	args.wasdel = ap->wasdel;
3791	args.isfl = 0;
3792	args.userdata = ap->userdata;
3793	if ((error = xfs_alloc_vextent(&args)))
3794		return error;
3795	if (tryagain && args.fsbno == NULLFSBLOCK) {
3796		/*
3797		 * Exact allocation failed. Now try with alignment
3798		 * turned on.
3799		 */
3800		args.type = atype;
3801		args.fsbno = ap->blkno;
3802		args.alignment = stripe_align;
3803		args.minlen = nextminlen;
3804		args.minalignslop = 0;
3805		isaligned = 1;
3806		if ((error = xfs_alloc_vextent(&args)))
3807			return error;
3808	}
3809	if (isaligned && args.fsbno == NULLFSBLOCK) {
3810		/*
3811		 * allocation failed, so turn off alignment and
3812		 * try again.
3813		 */
3814		args.type = atype;
3815		args.fsbno = ap->blkno;
3816		args.alignment = 0;
3817		if ((error = xfs_alloc_vextent(&args)))
3818			return error;
3819	}
3820	if (args.fsbno == NULLFSBLOCK && nullfb &&
3821	    args.minlen > ap->minlen) {
3822		args.minlen = ap->minlen;
3823		args.type = XFS_ALLOCTYPE_START_BNO;
3824		args.fsbno = ap->blkno;
3825		if ((error = xfs_alloc_vextent(&args)))
3826			return error;
3827	}
3828	if (args.fsbno == NULLFSBLOCK && nullfb) {
3829		args.fsbno = 0;
3830		args.type = XFS_ALLOCTYPE_FIRST_AG;
3831		args.total = ap->minlen;
3832		args.minleft = 0;
3833		if ((error = xfs_alloc_vextent(&args)))
3834			return error;
3835		ap->flist->xbf_low = 1;
3836	}
3837	if (args.fsbno != NULLFSBLOCK) {
3838		/*
3839		 * check the allocation happened at the same or higher AG than
3840		 * the first block that was allocated.
3841		 */
3842		ASSERT(*ap->firstblock == NULLFSBLOCK ||
3843		       XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3844		       XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3845		       (ap->flist->xbf_low &&
3846			XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3847			XFS_FSB_TO_AGNO(mp, args.fsbno)));
3848
3849		ap->blkno = args.fsbno;
3850		if (*ap->firstblock == NULLFSBLOCK)
3851			*ap->firstblock = args.fsbno;
3852		ASSERT(nullfb || fb_agno == args.agno ||
3853		       (ap->flist->xbf_low && fb_agno < args.agno));
3854		ap->length = args.len;
3855		ap->ip->i_d.di_nblocks += args.len;
3856		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3857		if (ap->wasdel)
3858			ap->ip->i_delayed_blks -= args.len;
3859		/*
3860		 * Adjust the disk quota also. This was reserved
3861		 * earlier.
3862		 */
3863		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3864			ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3865					XFS_TRANS_DQ_BCOUNT,
3866			(long) args.len);
3867	} else {
3868		ap->blkno = NULLFSBLOCK;
3869		ap->length = 0;
3870	}
3871	return 0;
3872}
3873
3874/*
3875 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3876 * It figures out where to ask the underlying allocator to put the new extent.
3877 */
3878STATIC int
3879xfs_bmap_alloc(
3880	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3881{
3882	if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3883		return xfs_bmap_rtalloc(ap);
3884	return xfs_bmap_btalloc(ap);
3885}
3886
3887/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3888 * Trim the returned map to the required bounds
3889 */
3890STATIC void
3891xfs_bmapi_trim_map(
3892	struct xfs_bmbt_irec	*mval,
3893	struct xfs_bmbt_irec	*got,
3894	xfs_fileoff_t		*bno,
3895	xfs_filblks_t		len,
3896	xfs_fileoff_t		obno,
3897	xfs_fileoff_t		end,
3898	int			n,
3899	int			flags)
3900{
3901	if ((flags & XFS_BMAPI_ENTIRE) ||
3902	    got->br_startoff + got->br_blockcount <= obno) {
3903		*mval = *got;
3904		if (isnullstartblock(got->br_startblock))
3905			mval->br_startblock = DELAYSTARTBLOCK;
3906		return;
3907	}
3908
3909	if (obno > *bno)
3910		*bno = obno;
3911	ASSERT((*bno >= obno) || (n == 0));
3912	ASSERT(*bno < end);
3913	mval->br_startoff = *bno;
3914	if (isnullstartblock(got->br_startblock))
3915		mval->br_startblock = DELAYSTARTBLOCK;
3916	else
3917		mval->br_startblock = got->br_startblock +
3918					(*bno - got->br_startoff);
3919	/*
3920	 * Return the minimum of what we got and what we asked for for
3921	 * the length.  We can use the len variable here because it is
3922	 * modified below and we could have been there before coming
3923	 * here if the first part of the allocation didn't overlap what
3924	 * was asked for.
3925	 */
3926	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3927			got->br_blockcount - (*bno - got->br_startoff));
3928	mval->br_state = got->br_state;
3929	ASSERT(mval->br_blockcount <= len);
3930	return;
3931}
3932
3933/*
3934 * Update and validate the extent map to return
3935 */
3936STATIC void
3937xfs_bmapi_update_map(
3938	struct xfs_bmbt_irec	**map,
3939	xfs_fileoff_t		*bno,
3940	xfs_filblks_t		*len,
3941	xfs_fileoff_t		obno,
3942	xfs_fileoff_t		end,
3943	int			*n,
3944	int			flags)
3945{
3946	xfs_bmbt_irec_t	*mval = *map;
3947
3948	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3949	       ((mval->br_startoff + mval->br_blockcount) <= end));
3950	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3951	       (mval->br_startoff < obno));
3952
3953	*bno = mval->br_startoff + mval->br_blockcount;
3954	*len = end - *bno;
3955	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3956		/* update previous map with new information */
3957		ASSERT(mval->br_startblock == mval[-1].br_startblock);
3958		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3959		ASSERT(mval->br_state == mval[-1].br_state);
3960		mval[-1].br_blockcount = mval->br_blockcount;
3961		mval[-1].br_state = mval->br_state;
3962	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3963		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
3964		   mval[-1].br_startblock != HOLESTARTBLOCK &&
3965		   mval->br_startblock == mval[-1].br_startblock +
3966					  mval[-1].br_blockcount &&
3967		   ((flags & XFS_BMAPI_IGSTATE) ||
3968			mval[-1].br_state == mval->br_state)) {
3969		ASSERT(mval->br_startoff ==
3970		       mval[-1].br_startoff + mval[-1].br_blockcount);
3971		mval[-1].br_blockcount += mval->br_blockcount;
3972	} else if (*n > 0 &&
3973		   mval->br_startblock == DELAYSTARTBLOCK &&
3974		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
3975		   mval->br_startoff ==
3976		   mval[-1].br_startoff + mval[-1].br_blockcount) {
3977		mval[-1].br_blockcount += mval->br_blockcount;
3978		mval[-1].br_state = mval->br_state;
3979	} else if (!((*n == 0) &&
3980		     ((mval->br_startoff + mval->br_blockcount) <=
3981		      obno))) {
3982		mval++;
3983		(*n)++;
3984	}
3985	*map = mval;
3986}
3987
3988/*
3989 * Map file blocks to filesystem blocks without allocation.
3990 */
3991int
3992xfs_bmapi_read(
3993	struct xfs_inode	*ip,
3994	xfs_fileoff_t		bno,
3995	xfs_filblks_t		len,
3996	struct xfs_bmbt_irec	*mval,
3997	int			*nmap,
3998	int			flags)
3999{
4000	struct xfs_mount	*mp = ip->i_mount;
4001	struct xfs_ifork	*ifp;
4002	struct xfs_bmbt_irec	got;
4003	struct xfs_bmbt_irec	prev;
4004	xfs_fileoff_t		obno;
4005	xfs_fileoff_t		end;
4006	xfs_extnum_t		lastx;
4007	int			error;
4008	int			eof;
4009	int			n = 0;
4010	int			whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4011						XFS_ATTR_FORK : XFS_DATA_FORK;
4012
4013	ASSERT(*nmap >= 1);
4014	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4015			   XFS_BMAPI_IGSTATE)));
4016	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4017
4018	if (unlikely(XFS_TEST_ERROR(
4019	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4020	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4021	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4022		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4023		return XFS_ERROR(EFSCORRUPTED);
4024	}
4025
4026	if (XFS_FORCED_SHUTDOWN(mp))
4027		return XFS_ERROR(EIO);
4028
4029	XFS_STATS_INC(xs_blk_mapr);
4030
4031	ifp = XFS_IFORK_PTR(ip, whichfork);
4032
4033	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4034		error = xfs_iread_extents(NULL, ip, whichfork);
4035		if (error)
4036			return error;
4037	}
4038
4039	xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4040	end = bno + len;
4041	obno = bno;
4042
4043	while (bno < end && n < *nmap) {
4044		/* Reading past eof, act as though there's a hole up to end. */
4045		if (eof)
4046			got.br_startoff = end;
4047		if (got.br_startoff > bno) {
4048			/* Reading in a hole.  */
4049			mval->br_startoff = bno;
4050			mval->br_startblock = HOLESTARTBLOCK;
4051			mval->br_blockcount =
4052				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4053			mval->br_state = XFS_EXT_NORM;
4054			bno += mval->br_blockcount;
4055			len -= mval->br_blockcount;
4056			mval++;
4057			n++;
4058			continue;
4059		}
4060
4061		/* set up the extent map to return. */
4062		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4063		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4064
4065		/* If we're done, stop now. */
4066		if (bno >= end || n >= *nmap)
4067			break;
4068
4069		/* Else go on to the next record. */
4070		if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4071			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4072		else
4073			eof = 1;
4074	}
4075	*nmap = n;
4076	return 0;
4077}
4078
4079STATIC int
4080xfs_bmapi_reserve_delalloc(
4081	struct xfs_inode	*ip,
4082	xfs_fileoff_t		aoff,
4083	xfs_filblks_t		len,
4084	struct xfs_bmbt_irec	*got,
4085	struct xfs_bmbt_irec	*prev,
4086	xfs_extnum_t		*lastx,
4087	int			eof)
4088{
4089	struct xfs_mount	*mp = ip->i_mount;
4090	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4091	xfs_extlen_t		alen;
4092	xfs_extlen_t		indlen;
4093	char			rt = XFS_IS_REALTIME_INODE(ip);
4094	xfs_extlen_t		extsz;
4095	int			error;
4096
4097	alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4098	if (!eof)
4099		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4100
4101	/* Figure out the extent size, adjust alen */
4102	extsz = xfs_get_extsz_hint(ip);
4103	if (extsz) {
4104		/*
4105		 * Make sure we don't exceed a single extent length when we
4106		 * align the extent by reducing length we are going to
4107		 * allocate by the maximum amount extent size aligment may
4108		 * require.
4109		 */
4110		alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4111		error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4112					       1, 0, &aoff, &alen);
4113		ASSERT(!error);
4114	}
4115
4116	if (rt)
4117		extsz = alen / mp->m_sb.sb_rextsize;
4118
4119	/*
4120	 * Make a transaction-less quota reservation for delayed allocation
4121	 * blocks.  This number gets adjusted later.  We return if we haven't
4122	 * allocated blocks already inside this loop.
4123	 */
4124	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4125			rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4126	if (error)
4127		return error;
4128
4129	/*
4130	 * Split changing sb for alen and indlen since they could be coming
4131	 * from different places.
4132	 */
4133	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4134	ASSERT(indlen > 0);
4135
4136	if (rt) {
4137		error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4138					  -((int64_t)extsz), 0);
4139	} else {
4140		error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4141						 -((int64_t)alen), 0);
4142	}
4143
4144	if (error)
4145		goto out_unreserve_quota;
4146
4147	error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4148					 -((int64_t)indlen), 0);
4149	if (error)
4150		goto out_unreserve_blocks;
4151
4152
4153	ip->i_delayed_blks += alen;
4154
4155	got->br_startoff = aoff;
4156	got->br_startblock = nullstartblock(indlen);
4157	got->br_blockcount = alen;
4158	got->br_state = XFS_EXT_NORM;
4159	xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4160
4161	/*
4162	 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4163	 * might have merged it into one of the neighbouring ones.
4164	 */
4165	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4166
4167	ASSERT(got->br_startoff <= aoff);
4168	ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4169	ASSERT(isnullstartblock(got->br_startblock));
4170	ASSERT(got->br_state == XFS_EXT_NORM);
4171	return 0;
4172
4173out_unreserve_blocks:
4174	if (rt)
4175		xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4176	else
4177		xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4178out_unreserve_quota:
4179	if (XFS_IS_QUOTA_ON(mp))
4180		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4181				XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4182	return error;
4183}
4184
4185/*
4186 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4187 */
4188int
4189xfs_bmapi_delay(
4190	struct xfs_inode	*ip,	/* incore inode */
4191	xfs_fileoff_t		bno,	/* starting file offs. mapped */
4192	xfs_filblks_t		len,	/* length to map in file */
4193	struct xfs_bmbt_irec	*mval,	/* output: map values */
4194	int			*nmap,	/* i/o: mval size/count */
4195	int			flags)	/* XFS_BMAPI_... */
4196{
4197	struct xfs_mount	*mp = ip->i_mount;
4198	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4199	struct xfs_bmbt_irec	got;	/* current file extent record */
4200	struct xfs_bmbt_irec	prev;	/* previous file extent record */
4201	xfs_fileoff_t		obno;	/* old block number (offset) */
4202	xfs_fileoff_t		end;	/* end of mapped file region */
4203	xfs_extnum_t		lastx;	/* last useful extent number */
4204	int			eof;	/* we've hit the end of extents */
4205	int			n = 0;	/* current extent index */
4206	int			error = 0;
4207
4208	ASSERT(*nmap >= 1);
4209	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4210	ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4211	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4212
4213	if (unlikely(XFS_TEST_ERROR(
4214	    (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4215	     XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4216	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4217		XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4218		return XFS_ERROR(EFSCORRUPTED);
4219	}
4220
4221	if (XFS_FORCED_SHUTDOWN(mp))
4222		return XFS_ERROR(EIO);
4223
4224	XFS_STATS_INC(xs_blk_mapw);
4225
4226	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4227		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4228		if (error)
4229			return error;
4230	}
4231
4232	xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4233	end = bno + len;
4234	obno = bno;
4235
4236	while (bno < end && n < *nmap) {
4237		if (eof || got.br_startoff > bno) {
4238			error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4239							   &prev, &lastx, eof);
4240			if (error) {
4241				if (n == 0) {
4242					*nmap = 0;
4243					return error;
4244				}
4245				break;
4246			}
4247		}
4248
4249		/* set up the extent map to return. */
4250		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4251		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4252
4253		/* If we're done, stop now. */
4254		if (bno >= end || n >= *nmap)
4255			break;
4256
4257		/* Else go on to the next record. */
4258		prev = got;
4259		if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4260			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4261		else
4262			eof = 1;
4263	}
4264
4265	*nmap = n;
4266	return 0;
4267}
4268
4269
4270int
4271__xfs_bmapi_allocate(
4272	struct xfs_bmalloca	*bma)
 
4273{
4274	struct xfs_mount	*mp = bma->ip->i_mount;
4275	int			whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4276						XFS_ATTR_FORK : XFS_DATA_FORK;
4277	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4278	int			tmp_logflags = 0;
4279	int			error;
 
4280
4281	ASSERT(bma->length > 0);
4282
 
 
4283	/*
4284	 * For the wasdelay case, we could also just allocate the stuff asked
4285	 * for in this bmap call but that wouldn't be as good.
4286	 */
4287	if (bma->wasdel) {
4288		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4289		bma->offset = bma->got.br_startoff;
4290		if (bma->idx != NULLEXTNUM && bma->idx) {
4291			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4292					 &bma->prev);
4293		}
4294	} else {
4295		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4296		if (!bma->eof)
4297			bma->length = XFS_FILBLKS_MIN(bma->length,
4298					bma->got.br_startoff - bma->offset);
4299	}
4300
4301	/*
4302	 * Indicate if this is the first user data in the file, or just any
4303	 * user data.
4304	 */
4305	if (!(bma->flags & XFS_BMAPI_METADATA)) {
4306		bma->userdata = (bma->offset == 0) ?
4307			XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4308	}
4309
4310	bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4311
4312	/*
4313	 * Only want to do the alignment at the eof if it is userdata and
4314	 * allocation length is larger than a stripe unit.
4315	 */
4316	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4317	    !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4318		error = xfs_bmap_isaeof(bma, whichfork);
4319		if (error)
4320			return error;
4321	}
4322
4323	error = xfs_bmap_alloc(bma);
4324	if (error)
4325		return error;
4326
4327	if (bma->flist->xbf_low)
4328		bma->minleft = 0;
4329	if (bma->cur)
4330		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4331	if (bma->blkno == NULLFSBLOCK)
4332		return 0;
4333	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4334		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4335		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4336		bma->cur->bc_private.b.flist = bma->flist;
4337	}
4338	/*
4339	 * Bump the number of extents we've allocated
4340	 * in this call.
4341	 */
4342	bma->nallocs++;
4343
4344	if (bma->cur)
4345		bma->cur->bc_private.b.flags =
4346			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4347
4348	bma->got.br_startoff = bma->offset;
4349	bma->got.br_startblock = bma->blkno;
4350	bma->got.br_blockcount = bma->length;
4351	bma->got.br_state = XFS_EXT_NORM;
4352
4353	/*
4354	 * A wasdelay extent has been initialized, so shouldn't be flagged
4355	 * as unwritten.
4356	 */
4357	if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4358	    xfs_sb_version_hasextflgbit(&mp->m_sb))
4359		bma->got.br_state = XFS_EXT_UNWRITTEN;
4360
4361	if (bma->wasdel)
4362		error = xfs_bmap_add_extent_delay_real(bma);
4363	else
4364		error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4365
4366	bma->logflags |= tmp_logflags;
4367	if (error)
4368		return error;
4369
4370	/*
4371	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4372	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4373	 * the neighbouring ones.
4374	 */
4375	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4376
4377	ASSERT(bma->got.br_startoff <= bma->offset);
4378	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4379	       bma->offset + bma->length);
4380	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4381	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4382	return 0;
4383}
4384
4385STATIC int
4386xfs_bmapi_convert_unwritten(
4387	struct xfs_bmalloca	*bma,
4388	struct xfs_bmbt_irec	*mval,
4389	xfs_filblks_t		len,
4390	int			flags)
4391{
4392	int			whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4393						XFS_ATTR_FORK : XFS_DATA_FORK;
4394	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4395	int			tmp_logflags = 0;
4396	int			error;
4397
4398	/* check if we need to do unwritten->real conversion */
4399	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4400	    (flags & XFS_BMAPI_PREALLOC))
4401		return 0;
4402
4403	/* check if we need to do real->unwritten conversion */
4404	if (mval->br_state == XFS_EXT_NORM &&
4405	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4406			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4407		return 0;
4408
4409	/*
4410	 * Modify (by adding) the state flag, if writing.
4411	 */
4412	ASSERT(mval->br_blockcount <= len);
4413	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4414		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4415					bma->ip, whichfork);
4416		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4417		bma->cur->bc_private.b.flist = bma->flist;
4418	}
4419	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4420				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4421
4422	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4423			&bma->cur, mval, bma->firstblock, bma->flist,
4424			&tmp_logflags);
4425	bma->logflags |= tmp_logflags;
4426	if (error)
4427		return error;
4428
4429	/*
4430	 * Update our extent pointer, given that
4431	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4432	 * of the neighbouring ones.
4433	 */
4434	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4435
4436	/*
4437	 * We may have combined previously unwritten space with written space,
4438	 * so generate another request.
4439	 */
4440	if (mval->br_blockcount < len)
4441		return EAGAIN;
4442	return 0;
4443}
4444
4445/*
4446 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4447 * extent state if necessary.  Details behaviour is controlled by the flags
4448 * parameter.  Only allocates blocks from a single allocation group, to avoid
4449 * locking problems.
4450 *
4451 * The returned value in "firstblock" from the first call in a transaction
4452 * must be remembered and presented to subsequent calls in "firstblock".
4453 * An upper bound for the number of blocks to be allocated is supplied to
4454 * the first call in "total"; if no allocation group has that many free
4455 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4456 */
4457int
4458xfs_bmapi_write(
4459	struct xfs_trans	*tp,		/* transaction pointer */
4460	struct xfs_inode	*ip,		/* incore inode */
4461	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4462	xfs_filblks_t		len,		/* length to map in file */
4463	int			flags,		/* XFS_BMAPI_... */
4464	xfs_fsblock_t		*firstblock,	/* first allocated block
4465						   controls a.g. for allocs */
4466	xfs_extlen_t		total,		/* total blocks needed */
4467	struct xfs_bmbt_irec	*mval,		/* output: map values */
4468	int			*nmap,		/* i/o: mval size/count */
4469	struct xfs_bmap_free	*flist)		/* i/o: list extents to free */
4470{
4471	struct xfs_mount	*mp = ip->i_mount;
4472	struct xfs_ifork	*ifp;
4473	struct xfs_bmalloca	bma = { NULL };	/* args for xfs_bmap_alloc */
4474	xfs_fileoff_t		end;		/* end of mapped file region */
4475	int			eof;		/* after the end of extents */
4476	int			error;		/* error return */
4477	int			n;		/* current extent index */
4478	xfs_fileoff_t		obno;		/* old block number (offset) */
4479	int			whichfork;	/* data or attr fork */
4480	char			inhole;		/* current location is hole in file */
4481	char			wasdelay;	/* old extent was delayed */
4482
4483#ifdef DEBUG
4484	xfs_fileoff_t		orig_bno;	/* original block number value */
4485	int			orig_flags;	/* original flags arg value */
4486	xfs_filblks_t		orig_len;	/* original value of len arg */
4487	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4488	int			orig_nmap;	/* original value of *nmap */
4489
4490	orig_bno = bno;
4491	orig_len = len;
4492	orig_flags = flags;
4493	orig_mval = mval;
4494	orig_nmap = *nmap;
4495#endif
4496	whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4497		XFS_ATTR_FORK : XFS_DATA_FORK;
4498
4499	ASSERT(*nmap >= 1);
4500	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4501	ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4502	ASSERT(tp != NULL);
4503	ASSERT(len > 0);
4504	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4505	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 
4506
4507	if (unlikely(XFS_TEST_ERROR(
4508	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4509	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
 
4510	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4511		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4512		return XFS_ERROR(EFSCORRUPTED);
4513	}
4514
4515	if (XFS_FORCED_SHUTDOWN(mp))
4516		return XFS_ERROR(EIO);
4517
4518	ifp = XFS_IFORK_PTR(ip, whichfork);
4519
4520	XFS_STATS_INC(xs_blk_mapw);
4521
 
 
 
 
 
 
 
4522	if (*firstblock == NULLFSBLOCK) {
4523		if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4524			bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4525		else
4526			bma.minleft = 1;
4527	} else {
4528		bma.minleft = 0;
4529	}
4530
4531	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4532		error = xfs_iread_extents(tp, ip, whichfork);
4533		if (error)
4534			goto error0;
4535	}
4536
4537	xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4538				&bma.prev);
4539	n = 0;
4540	end = bno + len;
4541	obno = bno;
4542
4543	bma.tp = tp;
4544	bma.ip = ip;
4545	bma.total = total;
4546	bma.userdata = 0;
4547	bma.flist = flist;
4548	bma.firstblock = firstblock;
4549
4550	if (flags & XFS_BMAPI_STACK_SWITCH)
4551		bma.stack_switch = 1;
4552
4553	while (bno < end && n < *nmap) {
4554		inhole = eof || bma.got.br_startoff > bno;
4555		wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4556
4557		/*
4558		 * First, deal with the hole before the allocated space
4559		 * that we found, if any.
4560		 */
4561		if (inhole || wasdelay) {
4562			bma.eof = eof;
4563			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4564			bma.wasdel = wasdelay;
4565			bma.offset = bno;
4566			bma.flags = flags;
4567
4568			/*
4569			 * There's a 32/64 bit type mismatch between the
4570			 * allocation length request (which can be 64 bits in
4571			 * length) and the bma length request, which is
4572			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4573			 * check for 32-bit overflows and handle them here.
4574			 */
4575			if (len > (xfs_filblks_t)MAXEXTLEN)
4576				bma.length = MAXEXTLEN;
4577			else
4578				bma.length = len;
4579
4580			ASSERT(len > 0);
4581			ASSERT(bma.length > 0);
4582			error = xfs_bmapi_allocate(&bma);
4583			if (error)
4584				goto error0;
4585			if (bma.blkno == NULLFSBLOCK)
4586				break;
4587		}
4588
4589		/* Deal with the allocated space we found.  */
4590		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4591							end, n, flags);
4592
4593		/* Execute unwritten extent conversion if necessary */
4594		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4595		if (error == EAGAIN)
4596			continue;
4597		if (error)
4598			goto error0;
4599
4600		/* update the extent map to return */
4601		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4602
4603		/*
4604		 * If we're done, stop now.  Stop when we've allocated
4605		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4606		 * the transaction may get too big.
4607		 */
4608		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4609			break;
4610
4611		/* Else go on to the next record. */
4612		bma.prev = bma.got;
4613		if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4614			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4615					 &bma.got);
4616		} else
4617			eof = 1;
4618	}
4619	*nmap = n;
4620
4621	/*
4622	 * Transform from btree to extents, give it cur.
4623	 */
4624	if (xfs_bmap_wants_extents(ip, whichfork)) {
4625		int		tmp_logflags = 0;
4626
4627		ASSERT(bma.cur);
4628		error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4629			&tmp_logflags, whichfork);
4630		bma.logflags |= tmp_logflags;
4631		if (error)
4632			goto error0;
4633	}
4634
4635	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4636	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4637		XFS_IFORK_MAXEXT(ip, whichfork));
4638	error = 0;
4639error0:
4640	/*
4641	 * Log everything.  Do this after conversion, there's no point in
4642	 * logging the extent records if we've converted to btree format.
4643	 */
4644	if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4645	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4646		bma.logflags &= ~xfs_ilog_fext(whichfork);
4647	else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4648		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4649		bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4650	/*
4651	 * Log whatever the flags say, even if error.  Otherwise we might miss
4652	 * detecting a case where the data is changed, there's an error,
4653	 * and it's not logged so we don't shutdown when we should.
4654	 */
4655	if (bma.logflags)
4656		xfs_trans_log_inode(tp, ip, bma.logflags);
4657
4658	if (bma.cur) {
4659		if (!error) {
4660			ASSERT(*firstblock == NULLFSBLOCK ||
4661			       XFS_FSB_TO_AGNO(mp, *firstblock) ==
4662			       XFS_FSB_TO_AGNO(mp,
4663				       bma.cur->bc_private.b.firstblock) ||
4664			       (flist->xbf_low &&
4665				XFS_FSB_TO_AGNO(mp, *firstblock) <
4666				XFS_FSB_TO_AGNO(mp,
4667					bma.cur->bc_private.b.firstblock)));
4668			*firstblock = bma.cur->bc_private.b.firstblock;
4669		}
4670		xfs_btree_del_cursor(bma.cur,
4671			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4672	}
4673	if (!error)
4674		xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4675			orig_nmap, *nmap);
4676	return error;
4677}
4678
4679/*
4680 * Called by xfs_bmapi to update file extent records and the btree
4681 * after removing space (or undoing a delayed allocation).
4682 */
4683STATIC int				/* error */
4684xfs_bmap_del_extent(
4685	xfs_inode_t		*ip,	/* incore inode pointer */
4686	xfs_trans_t		*tp,	/* current transaction pointer */
4687	xfs_extnum_t		*idx,	/* extent number to update/delete */
4688	xfs_bmap_free_t		*flist,	/* list of extents to be freed */
4689	xfs_btree_cur_t		*cur,	/* if null, not a btree */
4690	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
4691	int			*logflagsp, /* inode logging flags */
4692	int			whichfork) /* data or attr fork */
4693{
4694	xfs_filblks_t		da_new;	/* new delay-alloc indirect blocks */
4695	xfs_filblks_t		da_old;	/* old delay-alloc indirect blocks */
4696	xfs_fsblock_t		del_endblock=0;	/* first block past del */
4697	xfs_fileoff_t		del_endoff;	/* first offset past del */
4698	int			delay;	/* current block is delayed allocated */
4699	int			do_fx;	/* free extent at end of routine */
4700	xfs_bmbt_rec_host_t	*ep;	/* current extent entry pointer */
4701	int			error;	/* error return value */
4702	int			flags;	/* inode logging flags */
4703	xfs_bmbt_irec_t		got;	/* current extent entry */
4704	xfs_fileoff_t		got_endoff;	/* first offset past got */
4705	int			i;	/* temp state */
4706	xfs_ifork_t		*ifp;	/* inode fork pointer */
4707	xfs_mount_t		*mp;	/* mount structure */
4708	xfs_filblks_t		nblks;	/* quota/sb block count */
4709	xfs_bmbt_irec_t		new;	/* new record to be inserted */
4710	/* REFERENCED */
4711	uint			qfield;	/* quota field to update */
4712	xfs_filblks_t		temp;	/* for indirect length calculations */
4713	xfs_filblks_t		temp2;	/* for indirect length calculations */
4714	int			state = 0;
4715
4716	XFS_STATS_INC(xs_del_exlist);
4717
4718	if (whichfork == XFS_ATTR_FORK)
4719		state |= BMAP_ATTRFORK;
4720
4721	mp = ip->i_mount;
4722	ifp = XFS_IFORK_PTR(ip, whichfork);
4723	ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4724		(uint)sizeof(xfs_bmbt_rec_t)));
4725	ASSERT(del->br_blockcount > 0);
4726	ep = xfs_iext_get_ext(ifp, *idx);
4727	xfs_bmbt_get_all(ep, &got);
4728	ASSERT(got.br_startoff <= del->br_startoff);
4729	del_endoff = del->br_startoff + del->br_blockcount;
4730	got_endoff = got.br_startoff + got.br_blockcount;
4731	ASSERT(got_endoff >= del_endoff);
4732	delay = isnullstartblock(got.br_startblock);
4733	ASSERT(isnullstartblock(del->br_startblock) == delay);
4734	flags = 0;
4735	qfield = 0;
4736	error = 0;
4737	/*
4738	 * If deleting a real allocation, must free up the disk space.
4739	 */
4740	if (!delay) {
4741		flags = XFS_ILOG_CORE;
4742		/*
4743		 * Realtime allocation.  Free it and record di_nblocks update.
4744		 */
4745		if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4746			xfs_fsblock_t	bno;
4747			xfs_filblks_t	len;
4748
4749			ASSERT(do_mod(del->br_blockcount,
4750				      mp->m_sb.sb_rextsize) == 0);
4751			ASSERT(do_mod(del->br_startblock,
4752				      mp->m_sb.sb_rextsize) == 0);
4753			bno = del->br_startblock;
4754			len = del->br_blockcount;
4755			do_div(bno, mp->m_sb.sb_rextsize);
4756			do_div(len, mp->m_sb.sb_rextsize);
4757			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4758			if (error)
4759				goto done;
4760			do_fx = 0;
4761			nblks = len * mp->m_sb.sb_rextsize;
4762			qfield = XFS_TRANS_DQ_RTBCOUNT;
4763		}
4764		/*
4765		 * Ordinary allocation.
4766		 */
4767		else {
4768			do_fx = 1;
4769			nblks = del->br_blockcount;
4770			qfield = XFS_TRANS_DQ_BCOUNT;
4771		}
4772		/*
4773		 * Set up del_endblock and cur for later.
4774		 */
4775		del_endblock = del->br_startblock + del->br_blockcount;
4776		if (cur) {
4777			if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4778					got.br_startblock, got.br_blockcount,
4779					&i)))
4780				goto done;
4781			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4782		}
4783		da_old = da_new = 0;
4784	} else {
4785		da_old = startblockval(got.br_startblock);
4786		da_new = 0;
4787		nblks = 0;
4788		do_fx = 0;
4789	}
4790	/*
4791	 * Set flag value to use in switch statement.
4792	 * Left-contig is 2, right-contig is 1.
4793	 */
4794	switch (((got.br_startoff == del->br_startoff) << 1) |
4795		(got_endoff == del_endoff)) {
4796	case 3:
4797		/*
4798		 * Matches the whole extent.  Delete the entry.
4799		 */
4800		xfs_iext_remove(ip, *idx, 1,
4801				whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4802		--*idx;
4803		if (delay)
4804			break;
4805
4806		XFS_IFORK_NEXT_SET(ip, whichfork,
4807			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4808		flags |= XFS_ILOG_CORE;
4809		if (!cur) {
4810			flags |= xfs_ilog_fext(whichfork);
4811			break;
4812		}
4813		if ((error = xfs_btree_delete(cur, &i)))
4814			goto done;
4815		XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4816		break;
4817
4818	case 2:
4819		/*
4820		 * Deleting the first part of the extent.
4821		 */
4822		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4823		xfs_bmbt_set_startoff(ep, del_endoff);
4824		temp = got.br_blockcount - del->br_blockcount;
4825		xfs_bmbt_set_blockcount(ep, temp);
4826		if (delay) {
4827			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4828				da_old);
4829			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4830			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4831			da_new = temp;
4832			break;
4833		}
4834		xfs_bmbt_set_startblock(ep, del_endblock);
4835		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4836		if (!cur) {
4837			flags |= xfs_ilog_fext(whichfork);
4838			break;
4839		}
4840		if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4841				got.br_blockcount - del->br_blockcount,
4842				got.br_state)))
4843			goto done;
4844		break;
4845
4846	case 1:
4847		/*
4848		 * Deleting the last part of the extent.
4849		 */
4850		temp = got.br_blockcount - del->br_blockcount;
4851		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4852		xfs_bmbt_set_blockcount(ep, temp);
4853		if (delay) {
4854			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4855				da_old);
4856			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4857			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4858			da_new = temp;
4859			break;
4860		}
4861		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4862		if (!cur) {
4863			flags |= xfs_ilog_fext(whichfork);
4864			break;
4865		}
4866		if ((error = xfs_bmbt_update(cur, got.br_startoff,
4867				got.br_startblock,
4868				got.br_blockcount - del->br_blockcount,
4869				got.br_state)))
4870			goto done;
4871		break;
4872
4873	case 0:
4874		/*
4875		 * Deleting the middle of the extent.
4876		 */
4877		temp = del->br_startoff - got.br_startoff;
4878		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4879		xfs_bmbt_set_blockcount(ep, temp);
4880		new.br_startoff = del_endoff;
4881		temp2 = got_endoff - del_endoff;
4882		new.br_blockcount = temp2;
4883		new.br_state = got.br_state;
4884		if (!delay) {
4885			new.br_startblock = del_endblock;
4886			flags |= XFS_ILOG_CORE;
4887			if (cur) {
4888				if ((error = xfs_bmbt_update(cur,
4889						got.br_startoff,
4890						got.br_startblock, temp,
4891						got.br_state)))
4892					goto done;
4893				if ((error = xfs_btree_increment(cur, 0, &i)))
4894					goto done;
4895				cur->bc_rec.b = new;
4896				error = xfs_btree_insert(cur, &i);
4897				if (error && error != ENOSPC)
4898					goto done;
4899				/*
4900				 * If get no-space back from btree insert,
4901				 * it tried a split, and we have a zero
4902				 * block reservation.
4903				 * Fix up our state and return the error.
4904				 */
4905				if (error == ENOSPC) {
4906					/*
4907					 * Reset the cursor, don't trust
4908					 * it after any insert operation.
4909					 */
4910					if ((error = xfs_bmbt_lookup_eq(cur,
4911							got.br_startoff,
4912							got.br_startblock,
4913							temp, &i)))
4914						goto done;
4915					XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4916					/*
4917					 * Update the btree record back
4918					 * to the original value.
4919					 */
4920					if ((error = xfs_bmbt_update(cur,
4921							got.br_startoff,
4922							got.br_startblock,
4923							got.br_blockcount,
4924							got.br_state)))
4925						goto done;
4926					/*
4927					 * Reset the extent record back
4928					 * to the original value.
4929					 */
4930					xfs_bmbt_set_blockcount(ep,
4931						got.br_blockcount);
4932					flags = 0;
4933					error = XFS_ERROR(ENOSPC);
4934					goto done;
4935				}
4936				XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4937			} else
4938				flags |= xfs_ilog_fext(whichfork);
4939			XFS_IFORK_NEXT_SET(ip, whichfork,
4940				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4941		} else {
4942			ASSERT(whichfork == XFS_DATA_FORK);
4943			temp = xfs_bmap_worst_indlen(ip, temp);
4944			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4945			temp2 = xfs_bmap_worst_indlen(ip, temp2);
4946			new.br_startblock = nullstartblock((int)temp2);
4947			da_new = temp + temp2;
4948			while (da_new > da_old) {
4949				if (temp) {
4950					temp--;
4951					da_new--;
4952					xfs_bmbt_set_startblock(ep,
4953						nullstartblock((int)temp));
4954				}
4955				if (da_new == da_old)
4956					break;
4957				if (temp2) {
4958					temp2--;
4959					da_new--;
4960					new.br_startblock =
4961						nullstartblock((int)temp2);
4962				}
4963			}
4964		}
4965		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4966		xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4967		++*idx;
4968		break;
4969	}
4970	/*
4971	 * If we need to, add to list of extents to delete.
4972	 */
4973	if (do_fx)
4974		xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4975			mp);
4976	/*
4977	 * Adjust inode # blocks in the file.
4978	 */
4979	if (nblks)
4980		ip->i_d.di_nblocks -= nblks;
4981	/*
4982	 * Adjust quota data.
4983	 */
4984	if (qfield)
4985		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
4986
4987	/*
4988	 * Account for change in delayed indirect blocks.
4989	 * Nothing to do for disk quota accounting here.
4990	 */
4991	ASSERT(da_old >= da_new);
4992	if (da_old > da_new) {
4993		xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4994			(int64_t)(da_old - da_new), 0);
4995	}
4996done:
4997	*logflagsp = flags;
4998	return error;
4999}
5000
5001/*
5002 * Unmap (remove) blocks from a file.
5003 * If nexts is nonzero then the number of extents to remove is limited to
5004 * that value.  If not all extents in the block range can be removed then
5005 * *done is set.
5006 */
5007int						/* error */
5008xfs_bunmapi(
5009	xfs_trans_t		*tp,		/* transaction pointer */
5010	struct xfs_inode	*ip,		/* incore inode */
5011	xfs_fileoff_t		bno,		/* starting offset to unmap */
5012	xfs_filblks_t		len,		/* length to unmap in file */
5013	int			flags,		/* misc flags */
5014	xfs_extnum_t		nexts,		/* number of extents max */
5015	xfs_fsblock_t		*firstblock,	/* first allocated block
5016						   controls a.g. for allocs */
5017	xfs_bmap_free_t		*flist,		/* i/o: list extents to free */
5018	int			*done)		/* set if not done yet */
5019{
5020	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
5021	xfs_bmbt_irec_t		del;		/* extent being deleted */
5022	int			eof;		/* is deleting at eof */
5023	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
5024	int			error;		/* error return value */
5025	xfs_extnum_t		extno;		/* extent number in list */
5026	xfs_bmbt_irec_t		got;		/* current extent record */
5027	xfs_ifork_t		*ifp;		/* inode fork pointer */
5028	int			isrt;		/* freeing in rt area */
5029	xfs_extnum_t		lastx;		/* last extent index used */
5030	int			logflags;	/* transaction logging flags */
5031	xfs_extlen_t		mod;		/* rt extent offset */
5032	xfs_mount_t		*mp;		/* mount structure */
5033	xfs_extnum_t		nextents;	/* number of file extents */
5034	xfs_bmbt_irec_t		prev;		/* previous extent record */
5035	xfs_fileoff_t		start;		/* first file offset deleted */
5036	int			tmp_logflags;	/* partial logging flags */
5037	int			wasdel;		/* was a delayed alloc extent */
5038	int			whichfork;	/* data or attribute fork */
5039	xfs_fsblock_t		sum;
5040
5041	trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5042
5043	whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5044		XFS_ATTR_FORK : XFS_DATA_FORK;
5045	ifp = XFS_IFORK_PTR(ip, whichfork);
5046	if (unlikely(
5047	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5048	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5049		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5050				 ip->i_mount);
5051		return XFS_ERROR(EFSCORRUPTED);
5052	}
5053	mp = ip->i_mount;
5054	if (XFS_FORCED_SHUTDOWN(mp))
5055		return XFS_ERROR(EIO);
5056
5057	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5058	ASSERT(len > 0);
5059	ASSERT(nexts >= 0);
5060
5061	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5062	    (error = xfs_iread_extents(tp, ip, whichfork)))
5063		return error;
5064	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5065	if (nextents == 0) {
5066		*done = 1;
5067		return 0;
5068	}
5069	XFS_STATS_INC(xs_blk_unmap);
5070	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5071	start = bno;
5072	bno = start + len - 1;
5073	ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5074		&prev);
5075
5076	/*
5077	 * Check to see if the given block number is past the end of the
5078	 * file, back up to the last block if so...
5079	 */
5080	if (eof) {
5081		ep = xfs_iext_get_ext(ifp, --lastx);
5082		xfs_bmbt_get_all(ep, &got);
5083		bno = got.br_startoff + got.br_blockcount - 1;
5084	}
5085	logflags = 0;
5086	if (ifp->if_flags & XFS_IFBROOT) {
5087		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5088		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5089		cur->bc_private.b.firstblock = *firstblock;
5090		cur->bc_private.b.flist = flist;
5091		cur->bc_private.b.flags = 0;
5092	} else
5093		cur = NULL;
5094
5095	if (isrt) {
5096		/*
5097		 * Synchronize by locking the bitmap inode.
5098		 */
5099		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5100		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5101	}
5102
5103	extno = 0;
5104	while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5105	       (nexts == 0 || extno < nexts)) {
5106		/*
5107		 * Is the found extent after a hole in which bno lives?
5108		 * Just back up to the previous extent, if so.
5109		 */
5110		if (got.br_startoff > bno) {
5111			if (--lastx < 0)
5112				break;
5113			ep = xfs_iext_get_ext(ifp, lastx);
5114			xfs_bmbt_get_all(ep, &got);
5115		}
5116		/*
5117		 * Is the last block of this extent before the range
5118		 * we're supposed to delete?  If so, we're done.
5119		 */
5120		bno = XFS_FILEOFF_MIN(bno,
5121			got.br_startoff + got.br_blockcount - 1);
5122		if (bno < start)
5123			break;
5124		/*
5125		 * Then deal with the (possibly delayed) allocated space
5126		 * we found.
5127		 */
5128		ASSERT(ep != NULL);
5129		del = got;
5130		wasdel = isnullstartblock(del.br_startblock);
5131		if (got.br_startoff < start) {
5132			del.br_startoff = start;
5133			del.br_blockcount -= start - got.br_startoff;
5134			if (!wasdel)
5135				del.br_startblock += start - got.br_startoff;
5136		}
5137		if (del.br_startoff + del.br_blockcount > bno + 1)
5138			del.br_blockcount = bno + 1 - del.br_startoff;
5139		sum = del.br_startblock + del.br_blockcount;
5140		if (isrt &&
5141		    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5142			/*
5143			 * Realtime extent not lined up at the end.
5144			 * The extent could have been split into written
5145			 * and unwritten pieces, or we could just be
5146			 * unmapping part of it.  But we can't really
5147			 * get rid of part of a realtime extent.
5148			 */
5149			if (del.br_state == XFS_EXT_UNWRITTEN ||
5150			    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5151				/*
5152				 * This piece is unwritten, or we're not
5153				 * using unwritten extents.  Skip over it.
5154				 */
5155				ASSERT(bno >= mod);
5156				bno -= mod > del.br_blockcount ?
5157					del.br_blockcount : mod;
5158				if (bno < got.br_startoff) {
5159					if (--lastx >= 0)
5160						xfs_bmbt_get_all(xfs_iext_get_ext(
5161							ifp, lastx), &got);
5162				}
5163				continue;
5164			}
5165			/*
5166			 * It's written, turn it unwritten.
5167			 * This is better than zeroing it.
5168			 */
5169			ASSERT(del.br_state == XFS_EXT_NORM);
5170			ASSERT(xfs_trans_get_block_res(tp) > 0);
5171			/*
5172			 * If this spans a realtime extent boundary,
5173			 * chop it back to the start of the one we end at.
5174			 */
5175			if (del.br_blockcount > mod) {
5176				del.br_startoff += del.br_blockcount - mod;
5177				del.br_startblock += del.br_blockcount - mod;
5178				del.br_blockcount = mod;
5179			}
5180			del.br_state = XFS_EXT_UNWRITTEN;
5181			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5182					&lastx, &cur, &del, firstblock, flist,
5183					&logflags);
5184			if (error)
5185				goto error0;
5186			goto nodelete;
5187		}
5188		if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5189			/*
5190			 * Realtime extent is lined up at the end but not
5191			 * at the front.  We'll get rid of full extents if
5192			 * we can.
5193			 */
5194			mod = mp->m_sb.sb_rextsize - mod;
5195			if (del.br_blockcount > mod) {
5196				del.br_blockcount -= mod;
5197				del.br_startoff += mod;
5198				del.br_startblock += mod;
5199			} else if ((del.br_startoff == start &&
5200				    (del.br_state == XFS_EXT_UNWRITTEN ||
5201				     xfs_trans_get_block_res(tp) == 0)) ||
5202				   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5203				/*
5204				 * Can't make it unwritten.  There isn't
5205				 * a full extent here so just skip it.
5206				 */
5207				ASSERT(bno >= del.br_blockcount);
5208				bno -= del.br_blockcount;
5209				if (got.br_startoff > bno) {
5210					if (--lastx >= 0) {
5211						ep = xfs_iext_get_ext(ifp,
5212								      lastx);
5213						xfs_bmbt_get_all(ep, &got);
5214					}
5215				}
5216				continue;
5217			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5218				/*
5219				 * This one is already unwritten.
5220				 * It must have a written left neighbor.
5221				 * Unwrite the killed part of that one and
5222				 * try again.
5223				 */
5224				ASSERT(lastx > 0);
5225				xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5226						lastx - 1), &prev);
5227				ASSERT(prev.br_state == XFS_EXT_NORM);
5228				ASSERT(!isnullstartblock(prev.br_startblock));
5229				ASSERT(del.br_startblock ==
5230				       prev.br_startblock + prev.br_blockcount);
5231				if (prev.br_startoff < start) {
5232					mod = start - prev.br_startoff;
5233					prev.br_blockcount -= mod;
5234					prev.br_startblock += mod;
5235					prev.br_startoff = start;
5236				}
5237				prev.br_state = XFS_EXT_UNWRITTEN;
5238				lastx--;
5239				error = xfs_bmap_add_extent_unwritten_real(tp,
5240						ip, &lastx, &cur, &prev,
5241						firstblock, flist, &logflags);
5242				if (error)
5243					goto error0;
5244				goto nodelete;
5245			} else {
5246				ASSERT(del.br_state == XFS_EXT_NORM);
5247				del.br_state = XFS_EXT_UNWRITTEN;
5248				error = xfs_bmap_add_extent_unwritten_real(tp,
5249						ip, &lastx, &cur, &del,
5250						firstblock, flist, &logflags);
5251				if (error)
5252					goto error0;
5253				goto nodelete;
5254			}
5255		}
5256		if (wasdel) {
5257			ASSERT(startblockval(del.br_startblock) > 0);
5258			/* Update realtime/data freespace, unreserve quota */
5259			if (isrt) {
5260				xfs_filblks_t rtexts;
5261
5262				rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5263				do_div(rtexts, mp->m_sb.sb_rextsize);
5264				xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5265						(int64_t)rtexts, 0);
5266				(void)xfs_trans_reserve_quota_nblks(NULL,
5267					ip, -((long)del.br_blockcount), 0,
5268					XFS_QMOPT_RES_RTBLKS);
5269			} else {
5270				xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5271						(int64_t)del.br_blockcount, 0);
5272				(void)xfs_trans_reserve_quota_nblks(NULL,
5273					ip, -((long)del.br_blockcount), 0,
5274					XFS_QMOPT_RES_REGBLKS);
5275			}
5276			ip->i_delayed_blks -= del.br_blockcount;
5277			if (cur)
5278				cur->bc_private.b.flags |=
5279					XFS_BTCUR_BPRV_WASDEL;
5280		} else if (cur)
5281			cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5282		/*
5283		 * If it's the case where the directory code is running
5284		 * with no block reservation, and the deleted block is in
5285		 * the middle of its extent, and the resulting insert
5286		 * of an extent would cause transformation to btree format,
5287		 * then reject it.  The calling code will then swap
5288		 * blocks around instead.
5289		 * We have to do this now, rather than waiting for the
5290		 * conversion to btree format, since the transaction
5291		 * will be dirty.
5292		 */
5293		if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5294		    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5295		    XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5296			XFS_IFORK_MAXEXT(ip, whichfork) &&
5297		    del.br_startoff > got.br_startoff &&
5298		    del.br_startoff + del.br_blockcount <
5299		    got.br_startoff + got.br_blockcount) {
5300			error = XFS_ERROR(ENOSPC);
5301			goto error0;
5302		}
5303		error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5304				&tmp_logflags, whichfork);
5305		logflags |= tmp_logflags;
5306		if (error)
5307			goto error0;
5308		bno = del.br_startoff - 1;
5309nodelete:
5310		/*
5311		 * If not done go on to the next (previous) record.
5312		 */
5313		if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5314			if (lastx >= 0) {
5315				ep = xfs_iext_get_ext(ifp, lastx);
5316				if (xfs_bmbt_get_startoff(ep) > bno) {
5317					if (--lastx >= 0)
5318						ep = xfs_iext_get_ext(ifp,
5319								      lastx);
5320				}
5321				xfs_bmbt_get_all(ep, &got);
5322			}
5323			extno++;
5324		}
5325	}
5326	*done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5327
5328	/*
5329	 * Convert to a btree if necessary.
5330	 */
5331	if (xfs_bmap_needs_btree(ip, whichfork)) {
5332		ASSERT(cur == NULL);
5333		error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5334			&cur, 0, &tmp_logflags, whichfork);
5335		logflags |= tmp_logflags;
5336		if (error)
5337			goto error0;
5338	}
5339	/*
5340	 * transform from btree to extents, give it cur
5341	 */
5342	else if (xfs_bmap_wants_extents(ip, whichfork)) {
5343		ASSERT(cur != NULL);
5344		error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5345			whichfork);
5346		logflags |= tmp_logflags;
5347		if (error)
5348			goto error0;
5349	}
5350	/*
5351	 * transform from extents to local?
5352	 */
5353	error = 0;
5354error0:
5355	/*
5356	 * Log everything.  Do this after conversion, there's no point in
5357	 * logging the extent records if we've converted to btree format.
5358	 */
5359	if ((logflags & xfs_ilog_fext(whichfork)) &&
5360	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5361		logflags &= ~xfs_ilog_fext(whichfork);
5362	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5363		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5364		logflags &= ~xfs_ilog_fbroot(whichfork);
5365	/*
5366	 * Log inode even in the error case, if the transaction
5367	 * is dirty we'll need to shut down the filesystem.
5368	 */
5369	if (logflags)
5370		xfs_trans_log_inode(tp, ip, logflags);
5371	if (cur) {
5372		if (!error) {
5373			*firstblock = cur->bc_private.b.firstblock;
5374			cur->bc_private.b.allocated = 0;
5375		}
5376		xfs_btree_del_cursor(cur,
5377			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5378	}
5379	return error;
5380}
5381
5382/*
5383 * Shift extent records to the left to cover a hole.
5384 *
5385 * The maximum number of extents to be shifted in a single operation
5386 * is @num_exts, and @current_ext keeps track of the current extent
5387 * index we have shifted. @offset_shift_fsb is the length by which each
5388 * extent is shifted. If there is no hole to shift the extents
5389 * into, this will be considered invalid operation and we abort immediately.
5390 */
5391int
5392xfs_bmap_shift_extents(
5393	struct xfs_trans	*tp,
5394	struct xfs_inode	*ip,
5395	int			*done,
5396	xfs_fileoff_t		start_fsb,
5397	xfs_fileoff_t		offset_shift_fsb,
5398	xfs_extnum_t		*current_ext,
5399	xfs_fsblock_t		*firstblock,
5400	struct xfs_bmap_free	*flist,
5401	int			num_exts)
5402{
5403	struct xfs_btree_cur		*cur;
5404	struct xfs_bmbt_rec_host	*gotp;
5405	struct xfs_bmbt_irec            got;
5406	struct xfs_bmbt_irec		left;
5407	struct xfs_mount		*mp = ip->i_mount;
5408	struct xfs_ifork		*ifp;
5409	xfs_extnum_t			nexts = 0;
5410	xfs_fileoff_t			startoff;
5411	int				error = 0;
5412	int				i;
5413	int				whichfork = XFS_DATA_FORK;
5414	int				logflags;
5415	xfs_filblks_t			blockcount = 0;
5416	int				total_extents;
5417
5418	if (unlikely(XFS_TEST_ERROR(
5419	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5420	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5421	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5422		XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5423				 XFS_ERRLEVEL_LOW, mp);
5424		return XFS_ERROR(EFSCORRUPTED);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5425	}
5426
5427	if (XFS_FORCED_SHUTDOWN(mp))
5428		return XFS_ERROR(EIO);
5429
5430	ASSERT(current_ext != NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5431
5432	ifp = XFS_IFORK_PTR(ip, whichfork);
5433	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5434		/* Read in all the extents */
5435		error = xfs_iread_extents(tp, ip, whichfork);
5436		if (error)
5437			return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5438	}
5439
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5440	/*
5441	 * If *current_ext is 0, we would need to lookup the extent
5442	 * from where we would start shifting and store it in gotp.
5443	 */
5444	if (!*current_ext) {
5445		gotp = xfs_iext_bno_to_ext(ifp, start_fsb, current_ext);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5446		/*
5447		 * gotp can be null in 2 cases: 1) if there are no extents
5448		 * or 2) start_fsb lies in a hole beyond which there are
5449		 * no extents. Either way, we are done.
5450		 */
5451		if (!gotp) {
5452			*done = 1;
5453			return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5454		}
5455	}
 
5456
5457	/* We are going to change core inode */
5458	logflags = XFS_ILOG_CORE;
5459	if (ifp->if_flags & XFS_IFBROOT) {
5460		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5461		cur->bc_private.b.firstblock = *firstblock;
5462		cur->bc_private.b.flist = flist;
5463		cur->bc_private.b.flags = 0;
5464	} else {
5465		cur = NULL;
5466		logflags |= XFS_ILOG_DEXT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5467	}
5468
 
 
 
 
5469	/*
5470	 * There may be delalloc extents in the data fork before the range we
5471	 * are collapsing out, so we cannot
5472	 * use the count of real extents here. Instead we have to calculate it
5473	 * from the incore fork.
5474	 */
5475	total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5476	while (nexts++ < num_exts && *current_ext < total_extents) {
 
 
 
5477
5478		gotp = xfs_iext_get_ext(ifp, *current_ext);
5479		xfs_bmbt_get_all(gotp, &got);
5480		startoff = got.br_startoff - offset_shift_fsb;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5481
5482		/*
5483		 * Before shifting extent into hole, make sure that the hole
5484		 * is large enough to accomodate the shift.
5485		 */
5486		if (*current_ext) {
5487			xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5488						*current_ext - 1), &left);
5489
5490			if (startoff < left.br_startoff + left.br_blockcount)
5491				error = XFS_ERROR(EINVAL);
5492		} else if (offset_shift_fsb > got.br_startoff) {
5493			/*
5494			 * When first extent is shifted, offset_shift_fsb
5495			 * should be less than the stating offset of
5496			 * the first extent.
5497			 */
5498			error = XFS_ERROR(EINVAL);
5499		}
 
5500
5501		if (error)
5502			goto del_cursor;
 
 
5503
5504		if (cur) {
5505			error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5506						   got.br_startblock,
5507						   got.br_blockcount,
5508						   &i);
5509			if (error)
5510				goto del_cursor;
5511			XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
5512		}
5513
5514		/* Check if we can merge 2 adjacent extents */
5515		if (*current_ext &&
5516		    left.br_startoff + left.br_blockcount == startoff &&
5517		    left.br_startblock + left.br_blockcount ==
5518				got.br_startblock &&
5519		    left.br_state == got.br_state &&
5520		    left.br_blockcount + got.br_blockcount <= MAXEXTLEN) {
5521			blockcount = left.br_blockcount +
5522				got.br_blockcount;
5523			xfs_iext_remove(ip, *current_ext, 1, 0);
5524			if (cur) {
5525				error = xfs_btree_delete(cur, &i);
5526				if (error)
5527					goto del_cursor;
5528				XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
5529			}
5530			XFS_IFORK_NEXT_SET(ip, whichfork,
5531				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5532			gotp = xfs_iext_get_ext(ifp, --*current_ext);
5533			xfs_bmbt_get_all(gotp, &got);
5534
5535			/* Make cursor point to the extent we will update */
5536			if (cur) {
5537				error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5538							   got.br_startblock,
5539							   got.br_blockcount,
5540							   &i);
5541				if (error)
5542					goto del_cursor;
5543				XFS_WANT_CORRUPTED_GOTO(i == 1, del_cursor);
5544			}
5545
5546			xfs_bmbt_set_blockcount(gotp, blockcount);
5547			got.br_blockcount = blockcount;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5548		} else {
5549			/* We have to update the startoff */
5550			xfs_bmbt_set_startoff(gotp, startoff);
5551			got.br_startoff = startoff;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5552		}
5553
5554		if (cur) {
5555			error = xfs_bmbt_update(cur, got.br_startoff,
5556						got.br_startblock,
5557						got.br_blockcount,
5558						got.br_state);
5559			if (error)
5560				goto del_cursor;
 
 
5561		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5562
5563		(*current_ext)++;
5564		total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
 
5565	}
 
5566
5567	/* Check if we are done */
5568	if (*current_ext == total_extents)
5569		*done = 1;
 
 
 
 
 
 
 
 
 
 
5570
5571del_cursor:
5572	if (cur)
5573		xfs_btree_del_cursor(cur,
5574			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5575
5576	xfs_trans_log_inode(tp, ip, logflags);
5577	return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
5578}
v3.5.6
   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 "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_types.h"
 
 
 
  21#include "xfs_bit.h"
  22#include "xfs_log.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_sb.h"
  26#include "xfs_ag.h"
 
 
 
  27#include "xfs_dir2.h"
  28#include "xfs_da_btree.h"
  29#include "xfs_bmap_btree.h"
  30#include "xfs_alloc_btree.h"
  31#include "xfs_ialloc_btree.h"
  32#include "xfs_dinode.h"
  33#include "xfs_inode.h"
  34#include "xfs_btree.h"
  35#include "xfs_mount.h"
  36#include "xfs_itable.h"
  37#include "xfs_inode_item.h"
  38#include "xfs_extfree_item.h"
  39#include "xfs_alloc.h"
  40#include "xfs_bmap.h"
 
 
  41#include "xfs_rtalloc.h"
  42#include "xfs_error.h"
  43#include "xfs_attr_leaf.h"
  44#include "xfs_quota.h"
  45#include "xfs_trans_space.h"
  46#include "xfs_buf_item.h"
 
 
 
 
  47#include "xfs_filestream.h"
  48#include "xfs_vnodeops.h"
  49#include "xfs_trace.h"
  50
  51
  52kmem_zone_t		*xfs_bmap_free_item_zone;
  53
  54/*
  55 * Prototypes for internal bmap routines.
 
 
 
 
 
  56 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  57
  58#ifdef DEBUG
  59STATIC void
  60xfs_bmap_check_leaf_extents(
  61	struct xfs_btree_cur	*cur,
  62	struct xfs_inode	*ip,
  63	int			whichfork);
  64#else
  65#define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
  66#endif
 
 
 
 
 
  67
 
 
 
 
 
 
 
 
 
  68
  69/*
  70 * Called from xfs_bmap_add_attrfork to handle extents format files.
  71 */
  72STATIC int					/* error */
  73xfs_bmap_add_attrfork_extents(
  74	xfs_trans_t		*tp,		/* transaction pointer */
  75	xfs_inode_t		*ip,		/* incore inode pointer */
  76	xfs_fsblock_t		*firstblock,	/* first block allocated */
  77	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
  78	int			*flags);	/* inode logging flags */
  79
  80/*
  81 * Called from xfs_bmap_add_attrfork to handle local format files.
 
 
  82 */
  83STATIC int					/* error */
  84xfs_bmap_add_attrfork_local(
  85	xfs_trans_t		*tp,		/* transaction pointer */
  86	xfs_inode_t		*ip,		/* incore inode pointer */
  87	xfs_fsblock_t		*firstblock,	/* first block allocated */
  88	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
  89	int			*flags);	/* inode logging flags */
 
 
 
 
 
 
  90
  91/*
  92 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
  93 * It figures out where to ask the underlying allocator to put the new extent.
  94 */
  95STATIC int				/* error */
  96xfs_bmap_alloc(
  97	xfs_bmalloca_t		*ap);	/* bmap alloc argument struct */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  98
  99/*
 100 * Transform a btree format file with only one leaf node, where the
 101 * extents list will fit in the inode, into an extents format file.
 102 * Since the file extents are already in-core, all we have to do is
 103 * give up the space for the btree root and pitch the leaf block.
 104 */
 105STATIC int				/* error */
 106xfs_bmap_btree_to_extents(
 107	xfs_trans_t		*tp,	/* transaction pointer */
 108	xfs_inode_t		*ip,	/* incore inode pointer */
 109	xfs_btree_cur_t		*cur,	/* btree cursor */
 110	int			*logflagsp, /* inode logging flags */
 111	int			whichfork); /* data or attr fork */
 
 
 
 
 
 
 
 
 
 
 112
 113/*
 114 * Remove the entry "free" from the free item list.  Prev points to the
 115 * previous entry, unless "free" is the head of the list.
 
 116 */
 117STATIC void
 118xfs_bmap_del_free(
 119	xfs_bmap_free_t		*flist,	/* free item list header */
 120	xfs_bmap_free_item_t	*prev,	/* previous item on list, if any */
 121	xfs_bmap_free_item_t	*free);	/* list item to be freed */
 
 
 
 
 
 
 
 
 
 
 
 122
 123/*
 124 * Convert an extents-format file into a btree-format file.
 125 * The new file will have a root block (in the inode) and a single child block.
 126 */
 127STATIC int					/* error */
 128xfs_bmap_extents_to_btree(
 129	xfs_trans_t		*tp,		/* transaction pointer */
 130	xfs_inode_t		*ip,		/* incore inode pointer */
 131	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
 132	xfs_bmap_free_t		*flist,		/* blocks freed in xaction */
 133	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
 134	int			wasdel,		/* converting a delayed alloc */
 135	int			*logflagsp,	/* inode logging flags */
 136	int			whichfork);	/* data or attr fork */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 137
 138/*
 139 * Convert a local file to an extents file.
 140 * This code is sort of bogus, since the file data needs to get
 141 * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 142 */
 143STATIC int				/* error */
 144xfs_bmap_local_to_extents(
 145	xfs_trans_t	*tp,		/* transaction pointer */
 146	xfs_inode_t	*ip,		/* incore inode pointer */
 147	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
 148	xfs_extlen_t	total,		/* total blocks needed by transaction */
 149	int		*logflagsp,	/* inode logging flags */
 150	int		whichfork);	/* data or attr fork */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 151
 152/*
 153 * Search the extents list for the inode, for the extent containing bno.
 154 * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 155 * *eofp will be set, and *prevp will contain the last entry (null if none).
 156 * Else, *lastxp will be set to the index of the found
 157 * entry; *gotp will contain the entry.
 158 */
 159STATIC xfs_bmbt_rec_host_t *		/* pointer to found extent entry */
 160xfs_bmap_search_extents(
 161	xfs_inode_t	*ip,		/* incore inode pointer */
 162	xfs_fileoff_t	bno,		/* block number searched for */
 163	int		whichfork,	/* data or attr fork */
 164	int		*eofp,		/* out: end of file found */
 165	xfs_extnum_t	*lastxp,	/* out: last extent index */
 166	xfs_bmbt_irec_t	*gotp,		/* out: extent entry found */
 167	xfs_bmbt_irec_t	*prevp);	/* out: previous extent entry found */
 
 
 
 
 168
 169/*
 170 * Compute the worst-case number of indirect blocks that will be used
 171 * for ip's delayed extent of length "len".
 172 */
 173STATIC xfs_filblks_t
 174xfs_bmap_worst_indlen(
 175	xfs_inode_t		*ip,	/* incore inode pointer */
 176	xfs_filblks_t		len);	/* delayed extent length */
 177
 178#ifdef DEBUG
 179/*
 180 * Perform various validation checks on the values being returned
 181 * from xfs_bmapi().
 
 
 182 */
 183STATIC void
 184xfs_bmap_validate_ret(
 185	xfs_fileoff_t		bno,
 186	xfs_filblks_t		len,
 187	int			flags,
 188	xfs_bmbt_irec_t		*mval,
 189	int			nmap,
 190	int			ret_nmap);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 191#else
 
 192#define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 193#endif /* DEBUG */
 194
 195STATIC int
 196xfs_bmap_count_tree(
 197	xfs_mount_t     *mp,
 198	xfs_trans_t     *tp,
 199	xfs_ifork_t	*ifp,
 200	xfs_fsblock_t   blockno,
 201	int             levelin,
 202	int		*count);
 203
 204STATIC void
 205xfs_bmap_count_leaves(
 206	xfs_ifork_t		*ifp,
 207	xfs_extnum_t		idx,
 208	int			numrecs,
 209	int			*count);
 
 
 
 
 
 
 
 
 
 
 
 210
 211STATIC void
 212xfs_bmap_disk_count_leaves(
 213	struct xfs_mount	*mp,
 214	struct xfs_btree_block	*block,
 215	int			numrecs,
 216	int			*count);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 217
 218/*
 219 * Bmap internal routines.
 
 220 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 221
 222STATIC int				/* error */
 223xfs_bmbt_lookup_eq(
 224	struct xfs_btree_cur	*cur,
 225	xfs_fileoff_t		off,
 226	xfs_fsblock_t		bno,
 227	xfs_filblks_t		len,
 228	int			*stat)	/* success/failure */
 229{
 230	cur->bc_rec.b.br_startoff = off;
 231	cur->bc_rec.b.br_startblock = bno;
 232	cur->bc_rec.b.br_blockcount = len;
 233	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 
 
 
 
 
 
 
 234}
 235
 
 
 
 
 
 
 
 
 
 
 236STATIC int				/* error */
 237xfs_bmbt_lookup_ge(
 238	struct xfs_btree_cur	*cur,
 239	xfs_fileoff_t		off,
 240	xfs_fsblock_t		bno,
 241	xfs_filblks_t		len,
 242	int			*stat)	/* success/failure */
 243{
 244	cur->bc_rec.b.br_startoff = off;
 245	cur->bc_rec.b.br_startblock = bno;
 246	cur->bc_rec.b.br_blockcount = len;
 247	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 248}
 249
 250/*
 251 * Check if the inode needs to be converted to btree format.
 
 252 */
 253static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 
 
 
 
 
 
 
 
 
 254{
 255	return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 256		XFS_IFORK_NEXTENTS(ip, whichfork) >
 257			XFS_IFORK_MAXEXT(ip, whichfork);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 258}
 259
 260/*
 261 * Check if the inode should be converted to extent format.
 
 
 
 262 */
 263static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 
 
 
 264{
 265	return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 266		XFS_IFORK_NEXTENTS(ip, whichfork) <=
 267			XFS_IFORK_MAXEXT(ip, whichfork);
 
 
 
 
 
 
 
 268}
 269
 270/*
 271 * Update the record referred to by cur to the value given
 272 * by [off, bno, len, state].
 273 * This either works (return 0) or gets an EFSCORRUPTED error.
 274 */
 275STATIC int
 276xfs_bmbt_update(
 277	struct xfs_btree_cur	*cur,
 278	xfs_fileoff_t		off,
 279	xfs_fsblock_t		bno,
 280	xfs_filblks_t		len,
 281	xfs_exntst_t		state)
 
 282{
 283	union xfs_btree_rec	rec;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 284
 285	xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 286	return xfs_btree_update(cur, &rec);
 
 287}
 288
 289/*
 290 * Called from xfs_bmap_add_attrfork to handle btree format files.
 291 */
 292STATIC int					/* error */
 293xfs_bmap_add_attrfork_btree(
 294	xfs_trans_t		*tp,		/* transaction pointer */
 295	xfs_inode_t		*ip,		/* incore inode pointer */
 296	xfs_fsblock_t		*firstblock,	/* first block allocated */
 297	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
 298	int			*flags)		/* inode logging flags */
 299{
 300	xfs_btree_cur_t		*cur;		/* btree cursor */
 301	int			error;		/* error return value */
 302	xfs_mount_t		*mp;		/* file system mount struct */
 303	int			stat;		/* newroot status */
 304
 305	mp = ip->i_mount;
 306	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 307		*flags |= XFS_ILOG_DBROOT;
 308	else {
 309		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 310		cur->bc_private.b.flist = flist;
 311		cur->bc_private.b.firstblock = *firstblock;
 312		if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 313			goto error0;
 314		/* must be at least one entry */
 315		XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 316		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 317			goto error0;
 318		if (stat == 0) {
 319			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 320			return XFS_ERROR(ENOSPC);
 321		}
 322		*firstblock = cur->bc_private.b.firstblock;
 323		cur->bc_private.b.allocated = 0;
 324		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 325	}
 326	return 0;
 327error0:
 328	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 329	return error;
 330}
 331
 332/*
 333 * Called from xfs_bmap_add_attrfork to handle extents format files.
 334 */
 335STATIC int					/* error */
 336xfs_bmap_add_attrfork_extents(
 337	xfs_trans_t		*tp,		/* transaction pointer */
 338	xfs_inode_t		*ip,		/* incore inode pointer */
 339	xfs_fsblock_t		*firstblock,	/* first block allocated */
 340	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
 341	int			*flags)		/* inode logging flags */
 342{
 343	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
 344	int			error;		/* error return value */
 345
 346	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 347		return 0;
 348	cur = NULL;
 349	error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 350		flags, XFS_DATA_FORK);
 351	if (cur) {
 352		cur->bc_private.b.allocated = 0;
 353		xfs_btree_del_cursor(cur,
 354			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 355	}
 356	return error;
 357}
 358
 359/*
 360 * Called from xfs_bmap_add_attrfork to handle local format files.
 
 
 
 
 
 
 
 
 361 */
 362STATIC int					/* error */
 363xfs_bmap_add_attrfork_local(
 364	xfs_trans_t		*tp,		/* transaction pointer */
 365	xfs_inode_t		*ip,		/* incore inode pointer */
 366	xfs_fsblock_t		*firstblock,	/* first block allocated */
 367	xfs_bmap_free_t		*flist,		/* blocks to free at commit */
 368	int			*flags)		/* inode logging flags */
 369{
 370	xfs_da_args_t		dargs;		/* args for dir/attr code */
 371	int			error;		/* error return value */
 372	xfs_mount_t		*mp;		/* mount structure pointer */
 373
 374	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 375		return 0;
 
 376	if (S_ISDIR(ip->i_d.di_mode)) {
 377		mp = ip->i_mount;
 378		memset(&dargs, 0, sizeof(dargs));
 379		dargs.dp = ip;
 380		dargs.firstblock = firstblock;
 381		dargs.flist = flist;
 382		dargs.total = mp->m_dirblkfsbs;
 383		dargs.whichfork = XFS_DATA_FORK;
 384		dargs.trans = tp;
 385		error = xfs_dir2_sf_to_block(&dargs);
 386	} else
 387		error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 388			XFS_DATA_FORK);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 389	return error;
 390}
 391
 392/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 393 * Convert a delayed allocation to a real allocation.
 394 */
 395STATIC int				/* error */
 396xfs_bmap_add_extent_delay_real(
 397	struct xfs_bmalloca	*bma)
 398{
 399	struct xfs_bmbt_irec	*new = &bma->got;
 400	int			diff;	/* temp value */
 401	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
 402	int			error;	/* error return value */
 403	int			i;	/* temp state */
 404	xfs_ifork_t		*ifp;	/* inode fork pointer */
 405	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
 406	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
 407					/* left is 0, right is 1, prev is 2 */
 408	int			rval=0;	/* return value (logging flags) */
 409	int			state = 0;/* state bits, accessed thru macros */
 410	xfs_filblks_t		da_new; /* new count del alloc blocks used */
 411	xfs_filblks_t		da_old; /* old count del alloc blocks used */
 412	xfs_filblks_t		temp=0;	/* value for da_new calculations */
 413	xfs_filblks_t		temp2=0;/* value for da_new calculations */
 414	int			tmp_rval;	/* partial logging flags */
 415
 416	ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
 417
 418	ASSERT(bma->idx >= 0);
 419	ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 420	ASSERT(!isnullstartblock(new->br_startblock));
 421	ASSERT(!bma->cur ||
 422	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
 423
 424	XFS_STATS_INC(xs_add_exlist);
 425
 426#define	LEFT		r[0]
 427#define	RIGHT		r[1]
 428#define	PREV		r[2]
 429
 430	/*
 431	 * Set up a bunch of variables to make the tests simpler.
 432	 */
 433	ep = xfs_iext_get_ext(ifp, bma->idx);
 434	xfs_bmbt_get_all(ep, &PREV);
 435	new_endoff = new->br_startoff + new->br_blockcount;
 436	ASSERT(PREV.br_startoff <= new->br_startoff);
 437	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 438
 439	da_old = startblockval(PREV.br_startblock);
 440	da_new = 0;
 441
 442	/*
 443	 * Set flags determining what part of the previous delayed allocation
 444	 * extent is being replaced by a real allocation.
 445	 */
 446	if (PREV.br_startoff == new->br_startoff)
 447		state |= BMAP_LEFT_FILLING;
 448	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 449		state |= BMAP_RIGHT_FILLING;
 450
 451	/*
 452	 * Check and set flags if this segment has a left neighbor.
 453	 * Don't set contiguous if the combined extent would be too large.
 454	 */
 455	if (bma->idx > 0) {
 456		state |= BMAP_LEFT_VALID;
 457		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
 458
 459		if (isnullstartblock(LEFT.br_startblock))
 460			state |= BMAP_LEFT_DELAY;
 461	}
 462
 463	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 464	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 465	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 466	    LEFT.br_state == new->br_state &&
 467	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 468		state |= BMAP_LEFT_CONTIG;
 469
 470	/*
 471	 * Check and set flags if this segment has a right neighbor.
 472	 * Don't set contiguous if the combined extent would be too large.
 473	 * Also check for all-three-contiguous being too large.
 474	 */
 475	if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 476		state |= BMAP_RIGHT_VALID;
 477		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
 478
 479		if (isnullstartblock(RIGHT.br_startblock))
 480			state |= BMAP_RIGHT_DELAY;
 481	}
 482
 483	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 484	    new_endoff == RIGHT.br_startoff &&
 485	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 486	    new->br_state == RIGHT.br_state &&
 487	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 488	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 489		       BMAP_RIGHT_FILLING)) !=
 490		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 491		       BMAP_RIGHT_FILLING) ||
 492	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 493			<= MAXEXTLEN))
 494		state |= BMAP_RIGHT_CONTIG;
 495
 496	error = 0;
 497	/*
 498	 * Switch out based on the FILLING and CONTIG state bits.
 499	 */
 500	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 501			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 502	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 503	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 504		/*
 505		 * Filling in all of a previously delayed allocation extent.
 506		 * The left and right neighbors are both contiguous with new.
 507		 */
 508		bma->idx--;
 509		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 510		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 511			LEFT.br_blockcount + PREV.br_blockcount +
 512			RIGHT.br_blockcount);
 513		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 514
 515		xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
 516		bma->ip->i_d.di_nextents--;
 517		if (bma->cur == NULL)
 518			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 519		else {
 520			rval = XFS_ILOG_CORE;
 521			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 522					RIGHT.br_startblock,
 523					RIGHT.br_blockcount, &i);
 524			if (error)
 525				goto done;
 526			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 527			error = xfs_btree_delete(bma->cur, &i);
 528			if (error)
 529				goto done;
 530			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 531			error = xfs_btree_decrement(bma->cur, 0, &i);
 532			if (error)
 533				goto done;
 534			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 535			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 536					LEFT.br_startblock,
 537					LEFT.br_blockcount +
 538					PREV.br_blockcount +
 539					RIGHT.br_blockcount, LEFT.br_state);
 540			if (error)
 541				goto done;
 542		}
 543		break;
 544
 545	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 546		/*
 547		 * Filling in all of a previously delayed allocation extent.
 548		 * The left neighbor is contiguous, the right is not.
 549		 */
 550		bma->idx--;
 551
 552		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 553		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 554			LEFT.br_blockcount + PREV.br_blockcount);
 555		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 556
 557		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 558		if (bma->cur == NULL)
 559			rval = XFS_ILOG_DEXT;
 560		else {
 561			rval = 0;
 562			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 563					LEFT.br_startblock, LEFT.br_blockcount,
 564					&i);
 565			if (error)
 566				goto done;
 567			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 568			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 569					LEFT.br_startblock,
 570					LEFT.br_blockcount +
 571					PREV.br_blockcount, LEFT.br_state);
 572			if (error)
 573				goto done;
 574		}
 575		break;
 576
 577	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 578		/*
 579		 * Filling in all of a previously delayed allocation extent.
 580		 * The right neighbor is contiguous, the left is not.
 581		 */
 582		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 583		xfs_bmbt_set_startblock(ep, new->br_startblock);
 584		xfs_bmbt_set_blockcount(ep,
 585			PREV.br_blockcount + RIGHT.br_blockcount);
 586		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 587
 588		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 589		if (bma->cur == NULL)
 590			rval = XFS_ILOG_DEXT;
 591		else {
 592			rval = 0;
 593			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 594					RIGHT.br_startblock,
 595					RIGHT.br_blockcount, &i);
 596			if (error)
 597				goto done;
 598			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 599			error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
 600					new->br_startblock,
 601					PREV.br_blockcount +
 602					RIGHT.br_blockcount, PREV.br_state);
 603			if (error)
 604				goto done;
 605		}
 606		break;
 607
 608	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 609		/*
 610		 * Filling in all of a previously delayed allocation extent.
 611		 * Neither the left nor right neighbors are contiguous with
 612		 * the new one.
 613		 */
 614		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 615		xfs_bmbt_set_startblock(ep, new->br_startblock);
 616		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 617
 618		bma->ip->i_d.di_nextents++;
 619		if (bma->cur == NULL)
 620			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 621		else {
 622			rval = XFS_ILOG_CORE;
 623			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 624					new->br_startblock, new->br_blockcount,
 625					&i);
 626			if (error)
 627				goto done;
 628			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 629			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 630			error = xfs_btree_insert(bma->cur, &i);
 631			if (error)
 632				goto done;
 633			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 634		}
 635		break;
 636
 637	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 638		/*
 639		 * Filling in the first part of a previous delayed allocation.
 640		 * The left neighbor is contiguous.
 641		 */
 642		trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 643		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
 644			LEFT.br_blockcount + new->br_blockcount);
 645		xfs_bmbt_set_startoff(ep,
 646			PREV.br_startoff + new->br_blockcount);
 647		trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 648
 649		temp = PREV.br_blockcount - new->br_blockcount;
 650		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 651		xfs_bmbt_set_blockcount(ep, temp);
 652		if (bma->cur == NULL)
 653			rval = XFS_ILOG_DEXT;
 654		else {
 655			rval = 0;
 656			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 657					LEFT.br_startblock, LEFT.br_blockcount,
 658					&i);
 659			if (error)
 660				goto done;
 661			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 662			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 663					LEFT.br_startblock,
 664					LEFT.br_blockcount +
 665					new->br_blockcount,
 666					LEFT.br_state);
 667			if (error)
 668				goto done;
 669		}
 670		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 671			startblockval(PREV.br_startblock));
 672		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 673		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 674
 675		bma->idx--;
 676		break;
 677
 678	case BMAP_LEFT_FILLING:
 679		/*
 680		 * Filling in the first part of a previous delayed allocation.
 681		 * The left neighbor is not contiguous.
 682		 */
 683		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 684		xfs_bmbt_set_startoff(ep, new_endoff);
 685		temp = PREV.br_blockcount - new->br_blockcount;
 686		xfs_bmbt_set_blockcount(ep, temp);
 687		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
 688		bma->ip->i_d.di_nextents++;
 689		if (bma->cur == NULL)
 690			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 691		else {
 692			rval = XFS_ILOG_CORE;
 693			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 694					new->br_startblock, new->br_blockcount,
 695					&i);
 696			if (error)
 697				goto done;
 698			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 699			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 700			error = xfs_btree_insert(bma->cur, &i);
 701			if (error)
 702				goto done;
 703			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 704		}
 705
 706		if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 707			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 708					bma->firstblock, bma->flist,
 709					&bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
 710			rval |= tmp_rval;
 711			if (error)
 712				goto done;
 713		}
 714		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 715			startblockval(PREV.br_startblock) -
 716			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
 717		ep = xfs_iext_get_ext(ifp, bma->idx + 1);
 718		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 719		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 720		break;
 721
 722	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 723		/*
 724		 * Filling in the last part of a previous delayed allocation.
 725		 * The right neighbor is contiguous with the new allocation.
 726		 */
 727		temp = PREV.br_blockcount - new->br_blockcount;
 728		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 729		xfs_bmbt_set_blockcount(ep, temp);
 730		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
 731			new->br_startoff, new->br_startblock,
 732			new->br_blockcount + RIGHT.br_blockcount,
 733			RIGHT.br_state);
 734		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 735		if (bma->cur == NULL)
 736			rval = XFS_ILOG_DEXT;
 737		else {
 738			rval = 0;
 739			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 740					RIGHT.br_startblock,
 741					RIGHT.br_blockcount, &i);
 742			if (error)
 743				goto done;
 744			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 745			error = xfs_bmbt_update(bma->cur, new->br_startoff,
 746					new->br_startblock,
 747					new->br_blockcount +
 748					RIGHT.br_blockcount,
 749					RIGHT.br_state);
 750			if (error)
 751				goto done;
 752		}
 753
 754		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 755			startblockval(PREV.br_startblock));
 756		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 757		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 758		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 759
 760		bma->idx++;
 761		break;
 762
 763	case BMAP_RIGHT_FILLING:
 764		/*
 765		 * Filling in the last part of a previous delayed allocation.
 766		 * The right neighbor is not contiguous.
 767		 */
 768		temp = PREV.br_blockcount - new->br_blockcount;
 769		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 770		xfs_bmbt_set_blockcount(ep, temp);
 771		xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
 772		bma->ip->i_d.di_nextents++;
 773		if (bma->cur == NULL)
 774			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 775		else {
 776			rval = XFS_ILOG_CORE;
 777			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 778					new->br_startblock, new->br_blockcount,
 779					&i);
 780			if (error)
 781				goto done;
 782			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 783			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 784			error = xfs_btree_insert(bma->cur, &i);
 785			if (error)
 786				goto done;
 787			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 788		}
 789
 790		if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 791			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 792				bma->firstblock, bma->flist, &bma->cur, 1,
 793				&tmp_rval, XFS_DATA_FORK);
 794			rval |= tmp_rval;
 795			if (error)
 796				goto done;
 797		}
 798		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 799			startblockval(PREV.br_startblock) -
 800			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
 801		ep = xfs_iext_get_ext(ifp, bma->idx);
 802		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 803		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 804
 805		bma->idx++;
 806		break;
 807
 808	case 0:
 809		/*
 810		 * Filling in the middle part of a previous delayed allocation.
 811		 * Contiguity is impossible here.
 812		 * This case is avoided almost all the time.
 813		 *
 814		 * We start with a delayed allocation:
 815		 *
 816		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 817		 *  PREV @ idx
 818		 *
 819	         * and we are allocating:
 820		 *                     +rrrrrrrrrrrrrrrrr+
 821		 *			      new
 822		 *
 823		 * and we set it up for insertion as:
 824		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 825		 *                            new
 826		 *  PREV @ idx          LEFT              RIGHT
 827		 *                      inserted at idx + 1
 828		 */
 829		temp = new->br_startoff - PREV.br_startoff;
 830		temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 831		trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
 832		xfs_bmbt_set_blockcount(ep, temp);	/* truncate PREV */
 833		LEFT = *new;
 834		RIGHT.br_state = PREV.br_state;
 835		RIGHT.br_startblock = nullstartblock(
 836				(int)xfs_bmap_worst_indlen(bma->ip, temp2));
 837		RIGHT.br_startoff = new_endoff;
 838		RIGHT.br_blockcount = temp2;
 839		/* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 840		xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
 841		bma->ip->i_d.di_nextents++;
 842		if (bma->cur == NULL)
 843			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 844		else {
 845			rval = XFS_ILOG_CORE;
 846			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 847					new->br_startblock, new->br_blockcount,
 848					&i);
 849			if (error)
 850				goto done;
 851			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 852			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 853			error = xfs_btree_insert(bma->cur, &i);
 854			if (error)
 855				goto done;
 856			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 857		}
 858
 859		if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 860			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 861					bma->firstblock, bma->flist, &bma->cur,
 862					1, &tmp_rval, XFS_DATA_FORK);
 863			rval |= tmp_rval;
 864			if (error)
 865				goto done;
 866		}
 867		temp = xfs_bmap_worst_indlen(bma->ip, temp);
 868		temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
 869		diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
 870			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
 871		if (diff > 0) {
 872			error = xfs_icsb_modify_counters(bma->ip->i_mount,
 873					XFS_SBS_FDBLOCKS,
 874					-((int64_t)diff), 0);
 875			ASSERT(!error);
 876			if (error)
 877				goto done;
 878		}
 879
 880		ep = xfs_iext_get_ext(ifp, bma->idx);
 881		xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 882		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 883		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 884		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
 885			nullstartblock((int)temp2));
 886		trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 887
 888		bma->idx++;
 889		da_new = temp + temp2;
 890		break;
 891
 892	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 893	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 894	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
 895	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 896	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 897	case BMAP_LEFT_CONTIG:
 898	case BMAP_RIGHT_CONTIG:
 899		/*
 900		 * These cases are all impossible.
 901		 */
 902		ASSERT(0);
 903	}
 904
 905	/* convert to a btree if necessary */
 906	if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 907		int	tmp_logflags;	/* partial log flag return val */
 908
 909		ASSERT(bma->cur == NULL);
 910		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 911				bma->firstblock, bma->flist, &bma->cur,
 912				da_old > 0, &tmp_logflags, XFS_DATA_FORK);
 913		bma->logflags |= tmp_logflags;
 914		if (error)
 915			goto done;
 916	}
 917
 918	/* adjust for changes in reserved delayed indirect blocks */
 919	if (da_old || da_new) {
 920		temp = da_new;
 921		if (bma->cur)
 922			temp += bma->cur->bc_private.b.allocated;
 923		ASSERT(temp <= da_old);
 924		if (temp < da_old)
 925			xfs_icsb_modify_counters(bma->ip->i_mount,
 926					XFS_SBS_FDBLOCKS,
 927					(int64_t)(da_old - temp), 0);
 928	}
 929
 930	/* clear out the allocated field, done with it now in any case. */
 931	if (bma->cur)
 932		bma->cur->bc_private.b.allocated = 0;
 933
 934	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
 935done:
 936	bma->logflags |= rval;
 937	return error;
 938#undef	LEFT
 939#undef	RIGHT
 940#undef	PREV
 941}
 942
 943/*
 944 * Convert an unwritten allocation to a real allocation or vice versa.
 945 */
 946STATIC int				/* error */
 947xfs_bmap_add_extent_unwritten_real(
 948	struct xfs_trans	*tp,
 949	xfs_inode_t		*ip,	/* incore inode pointer */
 950	xfs_extnum_t		*idx,	/* extent number to update/insert */
 951	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
 952	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
 953	xfs_fsblock_t		*first,	/* pointer to firstblock variable */
 954	xfs_bmap_free_t		*flist,	/* list of extents to be freed */
 955	int			*logflagsp) /* inode logging flags */
 956{
 957	xfs_btree_cur_t		*cur;	/* btree cursor */
 958	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
 959	int			error;	/* error return value */
 960	int			i;	/* temp state */
 961	xfs_ifork_t		*ifp;	/* inode fork pointer */
 962	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
 963	xfs_exntst_t		newext;	/* new extent state */
 964	xfs_exntst_t		oldext;	/* old extent state */
 965	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
 966					/* left is 0, right is 1, prev is 2 */
 967	int			rval=0;	/* return value (logging flags) */
 968	int			state = 0;/* state bits, accessed thru macros */
 969
 970	*logflagsp = 0;
 971
 972	cur = *curp;
 973	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 974
 975	ASSERT(*idx >= 0);
 976	ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 977	ASSERT(!isnullstartblock(new->br_startblock));
 978
 979	XFS_STATS_INC(xs_add_exlist);
 980
 981#define	LEFT		r[0]
 982#define	RIGHT		r[1]
 983#define	PREV		r[2]
 984
 985	/*
 986	 * Set up a bunch of variables to make the tests simpler.
 987	 */
 988	error = 0;
 989	ep = xfs_iext_get_ext(ifp, *idx);
 990	xfs_bmbt_get_all(ep, &PREV);
 991	newext = new->br_state;
 992	oldext = (newext == XFS_EXT_UNWRITTEN) ?
 993		XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
 994	ASSERT(PREV.br_state == oldext);
 995	new_endoff = new->br_startoff + new->br_blockcount;
 996	ASSERT(PREV.br_startoff <= new->br_startoff);
 997	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 998
 999	/*
1000	 * Set flags determining what part of the previous oldext allocation
1001	 * extent is being replaced by a newext allocation.
1002	 */
1003	if (PREV.br_startoff == new->br_startoff)
1004		state |= BMAP_LEFT_FILLING;
1005	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1006		state |= BMAP_RIGHT_FILLING;
1007
1008	/*
1009	 * Check and set flags if this segment has a left neighbor.
1010	 * Don't set contiguous if the combined extent would be too large.
1011	 */
1012	if (*idx > 0) {
1013		state |= BMAP_LEFT_VALID;
1014		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1015
1016		if (isnullstartblock(LEFT.br_startblock))
1017			state |= BMAP_LEFT_DELAY;
1018	}
1019
1020	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1021	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1022	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1023	    LEFT.br_state == newext &&
1024	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1025		state |= BMAP_LEFT_CONTIG;
1026
1027	/*
1028	 * Check and set flags if this segment has a right neighbor.
1029	 * Don't set contiguous if the combined extent would be too large.
1030	 * Also check for all-three-contiguous being too large.
1031	 */
1032	if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1033		state |= BMAP_RIGHT_VALID;
1034		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1035		if (isnullstartblock(RIGHT.br_startblock))
1036			state |= BMAP_RIGHT_DELAY;
1037	}
1038
1039	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1040	    new_endoff == RIGHT.br_startoff &&
1041	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1042	    newext == RIGHT.br_state &&
1043	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1044	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1045		       BMAP_RIGHT_FILLING)) !=
1046		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1047		       BMAP_RIGHT_FILLING) ||
1048	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1049			<= MAXEXTLEN))
1050		state |= BMAP_RIGHT_CONTIG;
1051
1052	/*
1053	 * Switch out based on the FILLING and CONTIG state bits.
1054	 */
1055	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1056			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1057	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1058	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1059		/*
1060		 * Setting all of a previous oldext extent to newext.
1061		 * The left and right neighbors are both contiguous with new.
1062		 */
1063		--*idx;
1064
1065		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1066		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1067			LEFT.br_blockcount + PREV.br_blockcount +
1068			RIGHT.br_blockcount);
1069		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1070
1071		xfs_iext_remove(ip, *idx + 1, 2, state);
1072		ip->i_d.di_nextents -= 2;
1073		if (cur == NULL)
1074			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1075		else {
1076			rval = XFS_ILOG_CORE;
1077			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1078					RIGHT.br_startblock,
1079					RIGHT.br_blockcount, &i)))
1080				goto done;
1081			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1082			if ((error = xfs_btree_delete(cur, &i)))
1083				goto done;
1084			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1085			if ((error = xfs_btree_decrement(cur, 0, &i)))
1086				goto done;
1087			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1088			if ((error = xfs_btree_delete(cur, &i)))
1089				goto done;
1090			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1091			if ((error = xfs_btree_decrement(cur, 0, &i)))
1092				goto done;
1093			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1094			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1095				LEFT.br_startblock,
1096				LEFT.br_blockcount + PREV.br_blockcount +
1097				RIGHT.br_blockcount, LEFT.br_state)))
1098				goto done;
1099		}
1100		break;
1101
1102	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1103		/*
1104		 * Setting all of a previous oldext extent to newext.
1105		 * The left neighbor is contiguous, the right is not.
1106		 */
1107		--*idx;
1108
1109		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1110		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1111			LEFT.br_blockcount + PREV.br_blockcount);
1112		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1113
1114		xfs_iext_remove(ip, *idx + 1, 1, state);
1115		ip->i_d.di_nextents--;
1116		if (cur == NULL)
1117			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1118		else {
1119			rval = XFS_ILOG_CORE;
1120			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1121					PREV.br_startblock, PREV.br_blockcount,
1122					&i)))
1123				goto done;
1124			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1125			if ((error = xfs_btree_delete(cur, &i)))
1126				goto done;
1127			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1128			if ((error = xfs_btree_decrement(cur, 0, &i)))
1129				goto done;
1130			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1131			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1132				LEFT.br_startblock,
1133				LEFT.br_blockcount + PREV.br_blockcount,
1134				LEFT.br_state)))
1135				goto done;
1136		}
1137		break;
1138
1139	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1140		/*
1141		 * Setting all of a previous oldext extent to newext.
1142		 * The right neighbor is contiguous, the left is not.
1143		 */
1144		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1145		xfs_bmbt_set_blockcount(ep,
1146			PREV.br_blockcount + RIGHT.br_blockcount);
1147		xfs_bmbt_set_state(ep, newext);
1148		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1149		xfs_iext_remove(ip, *idx + 1, 1, state);
1150		ip->i_d.di_nextents--;
1151		if (cur == NULL)
1152			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1153		else {
1154			rval = XFS_ILOG_CORE;
1155			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1156					RIGHT.br_startblock,
1157					RIGHT.br_blockcount, &i)))
1158				goto done;
1159			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1160			if ((error = xfs_btree_delete(cur, &i)))
1161				goto done;
1162			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1163			if ((error = xfs_btree_decrement(cur, 0, &i)))
1164				goto done;
1165			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1166			if ((error = xfs_bmbt_update(cur, new->br_startoff,
1167				new->br_startblock,
1168				new->br_blockcount + RIGHT.br_blockcount,
1169				newext)))
1170				goto done;
1171		}
1172		break;
1173
1174	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1175		/*
1176		 * Setting all of a previous oldext extent to newext.
1177		 * Neither the left nor right neighbors are contiguous with
1178		 * the new one.
1179		 */
1180		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1181		xfs_bmbt_set_state(ep, newext);
1182		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1183
1184		if (cur == NULL)
1185			rval = XFS_ILOG_DEXT;
1186		else {
1187			rval = 0;
1188			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1189					new->br_startblock, new->br_blockcount,
1190					&i)))
1191				goto done;
1192			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1193			if ((error = xfs_bmbt_update(cur, new->br_startoff,
1194				new->br_startblock, new->br_blockcount,
1195				newext)))
1196				goto done;
1197		}
1198		break;
1199
1200	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1201		/*
1202		 * Setting the first part of a previous oldext extent to newext.
1203		 * The left neighbor is contiguous.
1204		 */
1205		trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1206		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1207			LEFT.br_blockcount + new->br_blockcount);
1208		xfs_bmbt_set_startoff(ep,
1209			PREV.br_startoff + new->br_blockcount);
1210		trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1211
1212		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1213		xfs_bmbt_set_startblock(ep,
1214			new->br_startblock + new->br_blockcount);
1215		xfs_bmbt_set_blockcount(ep,
1216			PREV.br_blockcount - new->br_blockcount);
1217		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1218
1219		--*idx;
1220
1221		if (cur == NULL)
1222			rval = XFS_ILOG_DEXT;
1223		else {
1224			rval = 0;
1225			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1226					PREV.br_startblock, PREV.br_blockcount,
1227					&i)))
1228				goto done;
1229			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1230			if ((error = xfs_bmbt_update(cur,
1231				PREV.br_startoff + new->br_blockcount,
1232				PREV.br_startblock + new->br_blockcount,
1233				PREV.br_blockcount - new->br_blockcount,
1234				oldext)))
1235				goto done;
1236			if ((error = xfs_btree_decrement(cur, 0, &i)))
1237				goto done;
1238			error = xfs_bmbt_update(cur, LEFT.br_startoff,
1239				LEFT.br_startblock,
1240				LEFT.br_blockcount + new->br_blockcount,
1241				LEFT.br_state);
1242			if (error)
1243				goto done;
1244		}
1245		break;
1246
1247	case BMAP_LEFT_FILLING:
1248		/*
1249		 * Setting the first part of a previous oldext extent to newext.
1250		 * The left neighbor is not contiguous.
1251		 */
1252		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1253		ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1254		xfs_bmbt_set_startoff(ep, new_endoff);
1255		xfs_bmbt_set_blockcount(ep,
1256			PREV.br_blockcount - new->br_blockcount);
1257		xfs_bmbt_set_startblock(ep,
1258			new->br_startblock + new->br_blockcount);
1259		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1260
1261		xfs_iext_insert(ip, *idx, 1, new, state);
1262		ip->i_d.di_nextents++;
1263		if (cur == NULL)
1264			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1265		else {
1266			rval = XFS_ILOG_CORE;
1267			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1268					PREV.br_startblock, PREV.br_blockcount,
1269					&i)))
1270				goto done;
1271			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1272			if ((error = xfs_bmbt_update(cur,
1273				PREV.br_startoff + new->br_blockcount,
1274				PREV.br_startblock + new->br_blockcount,
1275				PREV.br_blockcount - new->br_blockcount,
1276				oldext)))
1277				goto done;
1278			cur->bc_rec.b = *new;
1279			if ((error = xfs_btree_insert(cur, &i)))
1280				goto done;
1281			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1282		}
1283		break;
1284
1285	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1286		/*
1287		 * Setting the last part of a previous oldext extent to newext.
1288		 * The right neighbor is contiguous with the new allocation.
1289		 */
1290		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1291		xfs_bmbt_set_blockcount(ep,
1292			PREV.br_blockcount - new->br_blockcount);
1293		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1294
1295		++*idx;
1296
1297		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1298		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1299			new->br_startoff, new->br_startblock,
1300			new->br_blockcount + RIGHT.br_blockcount, newext);
1301		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1302
1303		if (cur == NULL)
1304			rval = XFS_ILOG_DEXT;
1305		else {
1306			rval = 0;
1307			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1308					PREV.br_startblock,
1309					PREV.br_blockcount, &i)))
1310				goto done;
1311			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1312			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1313				PREV.br_startblock,
1314				PREV.br_blockcount - new->br_blockcount,
1315				oldext)))
1316				goto done;
1317			if ((error = xfs_btree_increment(cur, 0, &i)))
1318				goto done;
1319			if ((error = xfs_bmbt_update(cur, new->br_startoff,
1320				new->br_startblock,
1321				new->br_blockcount + RIGHT.br_blockcount,
1322				newext)))
1323				goto done;
1324		}
1325		break;
1326
1327	case BMAP_RIGHT_FILLING:
1328		/*
1329		 * Setting the last part of a previous oldext extent to newext.
1330		 * The right neighbor is not contiguous.
1331		 */
1332		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1333		xfs_bmbt_set_blockcount(ep,
1334			PREV.br_blockcount - new->br_blockcount);
1335		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1336
1337		++*idx;
1338		xfs_iext_insert(ip, *idx, 1, new, state);
1339
1340		ip->i_d.di_nextents++;
1341		if (cur == NULL)
1342			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1343		else {
1344			rval = XFS_ILOG_CORE;
1345			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1346					PREV.br_startblock, PREV.br_blockcount,
1347					&i)))
1348				goto done;
1349			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1350			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1351				PREV.br_startblock,
1352				PREV.br_blockcount - new->br_blockcount,
1353				oldext)))
1354				goto done;
1355			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1356					new->br_startblock, new->br_blockcount,
1357					&i)))
1358				goto done;
1359			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1360			cur->bc_rec.b.br_state = XFS_EXT_NORM;
1361			if ((error = xfs_btree_insert(cur, &i)))
1362				goto done;
1363			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1364		}
1365		break;
1366
1367	case 0:
1368		/*
1369		 * Setting the middle part of a previous oldext extent to
1370		 * newext.  Contiguity is impossible here.
1371		 * One extent becomes three extents.
1372		 */
1373		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1374		xfs_bmbt_set_blockcount(ep,
1375			new->br_startoff - PREV.br_startoff);
1376		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1377
1378		r[0] = *new;
1379		r[1].br_startoff = new_endoff;
1380		r[1].br_blockcount =
1381			PREV.br_startoff + PREV.br_blockcount - new_endoff;
1382		r[1].br_startblock = new->br_startblock + new->br_blockcount;
1383		r[1].br_state = oldext;
1384
1385		++*idx;
1386		xfs_iext_insert(ip, *idx, 2, &r[0], state);
1387
1388		ip->i_d.di_nextents += 2;
1389		if (cur == NULL)
1390			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1391		else {
1392			rval = XFS_ILOG_CORE;
1393			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1394					PREV.br_startblock, PREV.br_blockcount,
1395					&i)))
1396				goto done;
1397			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1398			/* new right extent - oldext */
1399			if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1400				r[1].br_startblock, r[1].br_blockcount,
1401				r[1].br_state)))
1402				goto done;
1403			/* new left extent - oldext */
1404			cur->bc_rec.b = PREV;
1405			cur->bc_rec.b.br_blockcount =
1406				new->br_startoff - PREV.br_startoff;
1407			if ((error = xfs_btree_insert(cur, &i)))
1408				goto done;
1409			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1410			/*
1411			 * Reset the cursor to the position of the new extent
1412			 * we are about to insert as we can't trust it after
1413			 * the previous insert.
1414			 */
1415			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1416					new->br_startblock, new->br_blockcount,
1417					&i)))
1418				goto done;
1419			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1420			/* new middle extent - newext */
1421			cur->bc_rec.b.br_state = new->br_state;
1422			if ((error = xfs_btree_insert(cur, &i)))
1423				goto done;
1424			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1425		}
1426		break;
1427
1428	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1429	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1430	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1431	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1432	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1433	case BMAP_LEFT_CONTIG:
1434	case BMAP_RIGHT_CONTIG:
1435		/*
1436		 * These cases are all impossible.
1437		 */
1438		ASSERT(0);
1439	}
1440
1441	/* convert to a btree if necessary */
1442	if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
1443		int	tmp_logflags;	/* partial log flag return val */
1444
1445		ASSERT(cur == NULL);
1446		error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1447				0, &tmp_logflags, XFS_DATA_FORK);
1448		*logflagsp |= tmp_logflags;
1449		if (error)
1450			goto done;
1451	}
1452
1453	/* clear out the allocated field, done with it now in any case. */
1454	if (cur) {
1455		cur->bc_private.b.allocated = 0;
1456		*curp = cur;
1457	}
1458
1459	xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1460done:
1461	*logflagsp |= rval;
1462	return error;
1463#undef	LEFT
1464#undef	RIGHT
1465#undef	PREV
1466}
1467
1468/*
1469 * Convert a hole to a delayed allocation.
1470 */
1471STATIC void
1472xfs_bmap_add_extent_hole_delay(
1473	xfs_inode_t		*ip,	/* incore inode pointer */
1474	xfs_extnum_t		*idx,	/* extent number to update/insert */
1475	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
1476{
1477	xfs_ifork_t		*ifp;	/* inode fork pointer */
1478	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
1479	xfs_filblks_t		newlen=0;	/* new indirect size */
1480	xfs_filblks_t		oldlen=0;	/* old indirect size */
1481	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
1482	int			state;  /* state bits, accessed thru macros */
1483	xfs_filblks_t		temp=0;	/* temp for indirect calculations */
1484
1485	ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1486	state = 0;
1487	ASSERT(isnullstartblock(new->br_startblock));
1488
1489	/*
1490	 * Check and set flags if this segment has a left neighbor
1491	 */
1492	if (*idx > 0) {
1493		state |= BMAP_LEFT_VALID;
1494		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1495
1496		if (isnullstartblock(left.br_startblock))
1497			state |= BMAP_LEFT_DELAY;
1498	}
1499
1500	/*
1501	 * Check and set flags if the current (right) segment exists.
1502	 * If it doesn't exist, we're converting the hole at end-of-file.
1503	 */
1504	if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1505		state |= BMAP_RIGHT_VALID;
1506		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1507
1508		if (isnullstartblock(right.br_startblock))
1509			state |= BMAP_RIGHT_DELAY;
1510	}
1511
1512	/*
1513	 * Set contiguity flags on the left and right neighbors.
1514	 * Don't let extents get too large, even if the pieces are contiguous.
1515	 */
1516	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1517	    left.br_startoff + left.br_blockcount == new->br_startoff &&
1518	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1519		state |= BMAP_LEFT_CONTIG;
1520
1521	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1522	    new->br_startoff + new->br_blockcount == right.br_startoff &&
1523	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1524	    (!(state & BMAP_LEFT_CONTIG) ||
1525	     (left.br_blockcount + new->br_blockcount +
1526	      right.br_blockcount <= MAXEXTLEN)))
1527		state |= BMAP_RIGHT_CONTIG;
1528
1529	/*
1530	 * Switch out based on the contiguity flags.
1531	 */
1532	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1533	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1534		/*
1535		 * New allocation is contiguous with delayed allocations
1536		 * on the left and on the right.
1537		 * Merge all three into a single extent record.
1538		 */
1539		--*idx;
1540		temp = left.br_blockcount + new->br_blockcount +
1541			right.br_blockcount;
1542
1543		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1544		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1545		oldlen = startblockval(left.br_startblock) +
1546			startblockval(new->br_startblock) +
1547			startblockval(right.br_startblock);
1548		newlen = xfs_bmap_worst_indlen(ip, temp);
1549		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1550			nullstartblock((int)newlen));
1551		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1552
1553		xfs_iext_remove(ip, *idx + 1, 1, state);
1554		break;
1555
1556	case BMAP_LEFT_CONTIG:
1557		/*
1558		 * New allocation is contiguous with a delayed allocation
1559		 * on the left.
1560		 * Merge the new allocation with the left neighbor.
1561		 */
1562		--*idx;
1563		temp = left.br_blockcount + new->br_blockcount;
1564
1565		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1566		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1567		oldlen = startblockval(left.br_startblock) +
1568			startblockval(new->br_startblock);
1569		newlen = xfs_bmap_worst_indlen(ip, temp);
1570		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1571			nullstartblock((int)newlen));
1572		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1573		break;
1574
1575	case BMAP_RIGHT_CONTIG:
1576		/*
1577		 * New allocation is contiguous with a delayed allocation
1578		 * on the right.
1579		 * Merge the new allocation with the right neighbor.
1580		 */
1581		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1582		temp = new->br_blockcount + right.br_blockcount;
1583		oldlen = startblockval(new->br_startblock) +
1584			startblockval(right.br_startblock);
1585		newlen = xfs_bmap_worst_indlen(ip, temp);
1586		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1587			new->br_startoff,
1588			nullstartblock((int)newlen), temp, right.br_state);
1589		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1590		break;
1591
1592	case 0:
1593		/*
1594		 * New allocation is not contiguous with another
1595		 * delayed allocation.
1596		 * Insert a new entry.
1597		 */
1598		oldlen = newlen = 0;
1599		xfs_iext_insert(ip, *idx, 1, new, state);
1600		break;
1601	}
1602	if (oldlen != newlen) {
1603		ASSERT(oldlen > newlen);
1604		xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1605			(int64_t)(oldlen - newlen), 0);
1606		/*
1607		 * Nothing to do for disk quota accounting here.
1608		 */
1609	}
1610}
1611
1612/*
1613 * Convert a hole to a real allocation.
1614 */
1615STATIC int				/* error */
1616xfs_bmap_add_extent_hole_real(
1617	struct xfs_bmalloca	*bma,
1618	int			whichfork)
1619{
1620	struct xfs_bmbt_irec	*new = &bma->got;
1621	int			error;	/* error return value */
1622	int			i;	/* temp state */
1623	xfs_ifork_t		*ifp;	/* inode fork pointer */
1624	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
1625	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
1626	int			rval=0;	/* return value (logging flags) */
1627	int			state;	/* state bits, accessed thru macros */
1628
1629	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1630
1631	ASSERT(bma->idx >= 0);
1632	ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1633	ASSERT(!isnullstartblock(new->br_startblock));
1634	ASSERT(!bma->cur ||
1635	       !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1636
1637	XFS_STATS_INC(xs_add_exlist);
1638
1639	state = 0;
1640	if (whichfork == XFS_ATTR_FORK)
1641		state |= BMAP_ATTRFORK;
1642
1643	/*
1644	 * Check and set flags if this segment has a left neighbor.
1645	 */
1646	if (bma->idx > 0) {
1647		state |= BMAP_LEFT_VALID;
1648		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1649		if (isnullstartblock(left.br_startblock))
1650			state |= BMAP_LEFT_DELAY;
1651	}
1652
1653	/*
1654	 * Check and set flags if this segment has a current value.
1655	 * Not true if we're inserting into the "hole" at eof.
1656	 */
1657	if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1658		state |= BMAP_RIGHT_VALID;
1659		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1660		if (isnullstartblock(right.br_startblock))
1661			state |= BMAP_RIGHT_DELAY;
1662	}
1663
1664	/*
1665	 * We're inserting a real allocation between "left" and "right".
1666	 * Set the contiguity flags.  Don't let extents get too large.
1667	 */
1668	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1669	    left.br_startoff + left.br_blockcount == new->br_startoff &&
1670	    left.br_startblock + left.br_blockcount == new->br_startblock &&
1671	    left.br_state == new->br_state &&
1672	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1673		state |= BMAP_LEFT_CONTIG;
1674
1675	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1676	    new->br_startoff + new->br_blockcount == right.br_startoff &&
1677	    new->br_startblock + new->br_blockcount == right.br_startblock &&
1678	    new->br_state == right.br_state &&
1679	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1680	    (!(state & BMAP_LEFT_CONTIG) ||
1681	     left.br_blockcount + new->br_blockcount +
1682	     right.br_blockcount <= MAXEXTLEN))
1683		state |= BMAP_RIGHT_CONTIG;
1684
1685	error = 0;
1686	/*
1687	 * Select which case we're in here, and implement it.
1688	 */
1689	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1690	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1691		/*
1692		 * New allocation is contiguous with real allocations on the
1693		 * left and on the right.
1694		 * Merge all three into a single extent record.
1695		 */
1696		--bma->idx;
1697		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1698		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1699			left.br_blockcount + new->br_blockcount +
1700			right.br_blockcount);
1701		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1702
1703		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1704
1705		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1706			XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1707		if (bma->cur == NULL) {
1708			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1709		} else {
1710			rval = XFS_ILOG_CORE;
1711			error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1712					right.br_startblock, right.br_blockcount,
1713					&i);
1714			if (error)
1715				goto done;
1716			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1717			error = xfs_btree_delete(bma->cur, &i);
1718			if (error)
1719				goto done;
1720			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1721			error = xfs_btree_decrement(bma->cur, 0, &i);
1722			if (error)
1723				goto done;
1724			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1725			error = xfs_bmbt_update(bma->cur, left.br_startoff,
1726					left.br_startblock,
1727					left.br_blockcount +
1728						new->br_blockcount +
1729						right.br_blockcount,
1730					left.br_state);
1731			if (error)
1732				goto done;
1733		}
1734		break;
1735
1736	case BMAP_LEFT_CONTIG:
1737		/*
1738		 * New allocation is contiguous with a real allocation
1739		 * on the left.
1740		 * Merge the new allocation with the left neighbor.
1741		 */
1742		--bma->idx;
1743		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1744		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1745			left.br_blockcount + new->br_blockcount);
1746		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1747
1748		if (bma->cur == NULL) {
1749			rval = xfs_ilog_fext(whichfork);
1750		} else {
1751			rval = 0;
1752			error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1753					left.br_startblock, left.br_blockcount,
1754					&i);
1755			if (error)
1756				goto done;
1757			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1758			error = xfs_bmbt_update(bma->cur, left.br_startoff,
1759					left.br_startblock,
1760					left.br_blockcount +
1761						new->br_blockcount,
1762					left.br_state);
1763			if (error)
1764				goto done;
1765		}
1766		break;
1767
1768	case BMAP_RIGHT_CONTIG:
1769		/*
1770		 * New allocation is contiguous with a real allocation
1771		 * on the right.
1772		 * Merge the new allocation with the right neighbor.
1773		 */
1774		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1775		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1776			new->br_startoff, new->br_startblock,
1777			new->br_blockcount + right.br_blockcount,
1778			right.br_state);
1779		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1780
1781		if (bma->cur == NULL) {
1782			rval = xfs_ilog_fext(whichfork);
1783		} else {
1784			rval = 0;
1785			error = xfs_bmbt_lookup_eq(bma->cur,
1786					right.br_startoff,
1787					right.br_startblock,
1788					right.br_blockcount, &i);
1789			if (error)
1790				goto done;
1791			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1792			error = xfs_bmbt_update(bma->cur, new->br_startoff,
1793					new->br_startblock,
1794					new->br_blockcount +
1795						right.br_blockcount,
1796					right.br_state);
1797			if (error)
1798				goto done;
1799		}
1800		break;
1801
1802	case 0:
1803		/*
1804		 * New allocation is not contiguous with another
1805		 * real allocation.
1806		 * Insert a new entry.
1807		 */
1808		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1809		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1810			XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1811		if (bma->cur == NULL) {
1812			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1813		} else {
1814			rval = XFS_ILOG_CORE;
1815			error = xfs_bmbt_lookup_eq(bma->cur,
1816					new->br_startoff,
1817					new->br_startblock,
1818					new->br_blockcount, &i);
1819			if (error)
1820				goto done;
1821			XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1822			bma->cur->bc_rec.b.br_state = new->br_state;
1823			error = xfs_btree_insert(bma->cur, &i);
1824			if (error)
1825				goto done;
1826			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1827		}
1828		break;
1829	}
1830
1831	/* convert to a btree if necessary */
1832	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1833		int	tmp_logflags;	/* partial log flag return val */
1834
1835		ASSERT(bma->cur == NULL);
1836		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1837				bma->firstblock, bma->flist, &bma->cur,
1838				0, &tmp_logflags, whichfork);
1839		bma->logflags |= tmp_logflags;
1840		if (error)
1841			goto done;
1842	}
1843
1844	/* clear out the allocated field, done with it now in any case. */
1845	if (bma->cur)
1846		bma->cur->bc_private.b.allocated = 0;
1847
1848	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1849done:
1850	bma->logflags |= rval;
1851	return error;
1852}
1853
1854/*
 
 
 
 
1855 * Adjust the size of the new extent based on di_extsize and rt extsize.
1856 */
1857STATIC int
1858xfs_bmap_extsize_align(
1859	xfs_mount_t	*mp,
1860	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
1861	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
1862	xfs_extlen_t	extsz,		/* align to this extent size */
1863	int		rt,		/* is this a realtime inode? */
1864	int		eof,		/* is extent at end-of-file? */
1865	int		delay,		/* creating delalloc extent? */
1866	int		convert,	/* overwriting unwritten extent? */
1867	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
1868	xfs_extlen_t	*lenp)		/* in/out: aligned length */
1869{
1870	xfs_fileoff_t	orig_off;	/* original offset */
1871	xfs_extlen_t	orig_alen;	/* original length */
1872	xfs_fileoff_t	orig_end;	/* original off+len */
1873	xfs_fileoff_t	nexto;		/* next file offset */
1874	xfs_fileoff_t	prevo;		/* previous file offset */
1875	xfs_fileoff_t	align_off;	/* temp for offset */
1876	xfs_extlen_t	align_alen;	/* temp for length */
1877	xfs_extlen_t	temp;		/* temp for calculations */
1878
1879	if (convert)
1880		return 0;
1881
1882	orig_off = align_off = *offp;
1883	orig_alen = align_alen = *lenp;
1884	orig_end = orig_off + orig_alen;
1885
1886	/*
1887	 * If this request overlaps an existing extent, then don't
1888	 * attempt to perform any additional alignment.
1889	 */
1890	if (!delay && !eof &&
1891	    (orig_off >= gotp->br_startoff) &&
1892	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1893		return 0;
1894	}
1895
1896	/*
1897	 * If the file offset is unaligned vs. the extent size
1898	 * we need to align it.  This will be possible unless
1899	 * the file was previously written with a kernel that didn't
1900	 * perform this alignment, or if a truncate shot us in the
1901	 * foot.
1902	 */
1903	temp = do_mod(orig_off, extsz);
1904	if (temp) {
1905		align_alen += temp;
1906		align_off -= temp;
1907	}
1908	/*
1909	 * Same adjustment for the end of the requested area.
1910	 */
1911	if ((temp = (align_alen % extsz))) {
1912		align_alen += extsz - temp;
1913	}
1914	/*
1915	 * If the previous block overlaps with this proposed allocation
1916	 * then move the start forward without adjusting the length.
1917	 */
1918	if (prevp->br_startoff != NULLFILEOFF) {
1919		if (prevp->br_startblock == HOLESTARTBLOCK)
1920			prevo = prevp->br_startoff;
1921		else
1922			prevo = prevp->br_startoff + prevp->br_blockcount;
1923	} else
1924		prevo = 0;
1925	if (align_off != orig_off && align_off < prevo)
1926		align_off = prevo;
1927	/*
1928	 * If the next block overlaps with this proposed allocation
1929	 * then move the start back without adjusting the length,
1930	 * but not before offset 0.
1931	 * This may of course make the start overlap previous block,
1932	 * and if we hit the offset 0 limit then the next block
1933	 * can still overlap too.
1934	 */
1935	if (!eof && gotp->br_startoff != NULLFILEOFF) {
1936		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1937		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1938			nexto = gotp->br_startoff + gotp->br_blockcount;
1939		else
1940			nexto = gotp->br_startoff;
1941	} else
1942		nexto = NULLFILEOFF;
1943	if (!eof &&
1944	    align_off + align_alen != orig_end &&
1945	    align_off + align_alen > nexto)
1946		align_off = nexto > align_alen ? nexto - align_alen : 0;
1947	/*
1948	 * If we're now overlapping the next or previous extent that
1949	 * means we can't fit an extsz piece in this hole.  Just move
1950	 * the start forward to the first valid spot and set
1951	 * the length so we hit the end.
1952	 */
1953	if (align_off != orig_off && align_off < prevo)
1954		align_off = prevo;
1955	if (align_off + align_alen != orig_end &&
1956	    align_off + align_alen > nexto &&
1957	    nexto != NULLFILEOFF) {
1958		ASSERT(nexto > prevo);
1959		align_alen = nexto - align_off;
1960	}
1961
1962	/*
1963	 * If realtime, and the result isn't a multiple of the realtime
1964	 * extent size we need to remove blocks until it is.
1965	 */
1966	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1967		/*
1968		 * We're not covering the original request, or
1969		 * we won't be able to once we fix the length.
1970		 */
1971		if (orig_off < align_off ||
1972		    orig_end > align_off + align_alen ||
1973		    align_alen - temp < orig_alen)
1974			return XFS_ERROR(EINVAL);
1975		/*
1976		 * Try to fix it by moving the start up.
1977		 */
1978		if (align_off + temp <= orig_off) {
1979			align_alen -= temp;
1980			align_off += temp;
1981		}
1982		/*
1983		 * Try to fix it by moving the end in.
1984		 */
1985		else if (align_off + align_alen - temp >= orig_end)
1986			align_alen -= temp;
1987		/*
1988		 * Set the start to the minimum then trim the length.
1989		 */
1990		else {
1991			align_alen -= orig_off - align_off;
1992			align_off = orig_off;
1993			align_alen -= align_alen % mp->m_sb.sb_rextsize;
1994		}
1995		/*
1996		 * Result doesn't cover the request, fail it.
1997		 */
1998		if (orig_off < align_off || orig_end > align_off + align_alen)
1999			return XFS_ERROR(EINVAL);
2000	} else {
2001		ASSERT(orig_off >= align_off);
2002		ASSERT(orig_end <= align_off + align_alen);
2003	}
2004
2005#ifdef DEBUG
2006	if (!eof && gotp->br_startoff != NULLFILEOFF)
2007		ASSERT(align_off + align_alen <= gotp->br_startoff);
2008	if (prevp->br_startoff != NULLFILEOFF)
2009		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2010#endif
2011
2012	*lenp = align_alen;
2013	*offp = align_off;
2014	return 0;
2015}
2016
2017#define XFS_ALLOC_GAP_UNITS	4
2018
2019STATIC void
2020xfs_bmap_adjacent(
2021	xfs_bmalloca_t	*ap)		/* bmap alloc argument struct */
2022{
2023	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
2024	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
2025	xfs_mount_t	*mp;		/* mount point structure */
2026	int		nullfb;		/* true if ap->firstblock isn't set */
2027	int		rt;		/* true if inode is realtime */
2028
2029#define	ISVALID(x,y)	\
2030	(rt ? \
2031		(x) < mp->m_sb.sb_rblocks : \
2032		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2033		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2034		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2035
2036	mp = ap->ip->i_mount;
2037	nullfb = *ap->firstblock == NULLFSBLOCK;
2038	rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2039	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2040	/*
2041	 * If allocating at eof, and there's a previous real block,
2042	 * try to use its last block as our starting point.
2043	 */
2044	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2045	    !isnullstartblock(ap->prev.br_startblock) &&
2046	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2047		    ap->prev.br_startblock)) {
2048		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2049		/*
2050		 * Adjust for the gap between prevp and us.
2051		 */
2052		adjust = ap->offset -
2053			(ap->prev.br_startoff + ap->prev.br_blockcount);
2054		if (adjust &&
2055		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2056			ap->blkno += adjust;
2057	}
2058	/*
2059	 * If not at eof, then compare the two neighbor blocks.
2060	 * Figure out whether either one gives us a good starting point,
2061	 * and pick the better one.
2062	 */
2063	else if (!ap->eof) {
2064		xfs_fsblock_t	gotbno;		/* right side block number */
2065		xfs_fsblock_t	gotdiff=0;	/* right side difference */
2066		xfs_fsblock_t	prevbno;	/* left side block number */
2067		xfs_fsblock_t	prevdiff=0;	/* left side difference */
2068
2069		/*
2070		 * If there's a previous (left) block, select a requested
2071		 * start block based on it.
2072		 */
2073		if (ap->prev.br_startoff != NULLFILEOFF &&
2074		    !isnullstartblock(ap->prev.br_startblock) &&
2075		    (prevbno = ap->prev.br_startblock +
2076			       ap->prev.br_blockcount) &&
2077		    ISVALID(prevbno, ap->prev.br_startblock)) {
2078			/*
2079			 * Calculate gap to end of previous block.
2080			 */
2081			adjust = prevdiff = ap->offset -
2082				(ap->prev.br_startoff +
2083				 ap->prev.br_blockcount);
2084			/*
2085			 * Figure the startblock based on the previous block's
2086			 * end and the gap size.
2087			 * Heuristic!
2088			 * If the gap is large relative to the piece we're
2089			 * allocating, or using it gives us an invalid block
2090			 * number, then just use the end of the previous block.
2091			 */
2092			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2093			    ISVALID(prevbno + prevdiff,
2094				    ap->prev.br_startblock))
2095				prevbno += adjust;
2096			else
2097				prevdiff += adjust;
2098			/*
2099			 * If the firstblock forbids it, can't use it,
2100			 * must use default.
2101			 */
2102			if (!rt && !nullfb &&
2103			    XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2104				prevbno = NULLFSBLOCK;
2105		}
2106		/*
2107		 * No previous block or can't follow it, just default.
2108		 */
2109		else
2110			prevbno = NULLFSBLOCK;
2111		/*
2112		 * If there's a following (right) block, select a requested
2113		 * start block based on it.
2114		 */
2115		if (!isnullstartblock(ap->got.br_startblock)) {
2116			/*
2117			 * Calculate gap to start of next block.
2118			 */
2119			adjust = gotdiff = ap->got.br_startoff - ap->offset;
2120			/*
2121			 * Figure the startblock based on the next block's
2122			 * start and the gap size.
2123			 */
2124			gotbno = ap->got.br_startblock;
2125			/*
2126			 * Heuristic!
2127			 * If the gap is large relative to the piece we're
2128			 * allocating, or using it gives us an invalid block
2129			 * number, then just use the start of the next block
2130			 * offset by our length.
2131			 */
2132			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2133			    ISVALID(gotbno - gotdiff, gotbno))
2134				gotbno -= adjust;
2135			else if (ISVALID(gotbno - ap->length, gotbno)) {
2136				gotbno -= ap->length;
2137				gotdiff += adjust - ap->length;
2138			} else
2139				gotdiff += adjust;
2140			/*
2141			 * If the firstblock forbids it, can't use it,
2142			 * must use default.
2143			 */
2144			if (!rt && !nullfb &&
2145			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2146				gotbno = NULLFSBLOCK;
2147		}
2148		/*
2149		 * No next block, just default.
2150		 */
2151		else
2152			gotbno = NULLFSBLOCK;
2153		/*
2154		 * If both valid, pick the better one, else the only good
2155		 * one, else ap->blkno is already set (to 0 or the inode block).
2156		 */
2157		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2158			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2159		else if (prevbno != NULLFSBLOCK)
2160			ap->blkno = prevbno;
2161		else if (gotbno != NULLFSBLOCK)
2162			ap->blkno = gotbno;
2163	}
2164#undef ISVALID
2165}
2166
2167STATIC int
2168xfs_bmap_rtalloc(
2169	xfs_bmalloca_t	*ap)		/* bmap alloc argument struct */
2170{
2171	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
2172	int		error;		/* error return value */
2173	xfs_mount_t	*mp;		/* mount point structure */
2174	xfs_extlen_t	prod = 0;	/* product factor for allocators */
2175	xfs_extlen_t	ralen = 0;	/* realtime allocation length */
2176	xfs_extlen_t	align;		/* minimum allocation alignment */
2177	xfs_rtblock_t	rtb;
2178
2179	mp = ap->ip->i_mount;
2180	align = xfs_get_extsz_hint(ap->ip);
2181	prod = align / mp->m_sb.sb_rextsize;
2182	error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2183					align, 1, ap->eof, 0,
2184					ap->conv, &ap->offset, &ap->length);
2185	if (error)
2186		return error;
2187	ASSERT(ap->length);
2188	ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2189
2190	/*
2191	 * If the offset & length are not perfectly aligned
2192	 * then kill prod, it will just get us in trouble.
2193	 */
2194	if (do_mod(ap->offset, align) || ap->length % align)
2195		prod = 1;
2196	/*
2197	 * Set ralen to be the actual requested length in rtextents.
2198	 */
2199	ralen = ap->length / mp->m_sb.sb_rextsize;
2200	/*
2201	 * If the old value was close enough to MAXEXTLEN that
2202	 * we rounded up to it, cut it back so it's valid again.
2203	 * Note that if it's a really large request (bigger than
2204	 * MAXEXTLEN), we don't hear about that number, and can't
2205	 * adjust the starting point to match it.
2206	 */
2207	if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2208		ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2209
2210	/*
2211	 * Lock out other modifications to the RT bitmap inode.
2212	 */
2213	xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2214	xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2215
2216	/*
2217	 * If it's an allocation to an empty file at offset 0,
2218	 * pick an extent that will space things out in the rt area.
2219	 */
2220	if (ap->eof && ap->offset == 0) {
2221		xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2222
2223		error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2224		if (error)
2225			return error;
2226		ap->blkno = rtx * mp->m_sb.sb_rextsize;
2227	} else {
2228		ap->blkno = 0;
2229	}
2230
2231	xfs_bmap_adjacent(ap);
2232
2233	/*
2234	 * Realtime allocation, done through xfs_rtallocate_extent.
2235	 */
2236	atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2237	do_div(ap->blkno, mp->m_sb.sb_rextsize);
2238	rtb = ap->blkno;
2239	ap->length = ralen;
2240	if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2241				&ralen, atype, ap->wasdel, prod, &rtb)))
2242		return error;
2243	if (rtb == NULLFSBLOCK && prod > 1 &&
2244	    (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2245					   ap->length, &ralen, atype,
2246					   ap->wasdel, 1, &rtb)))
2247		return error;
2248	ap->blkno = rtb;
2249	if (ap->blkno != NULLFSBLOCK) {
2250		ap->blkno *= mp->m_sb.sb_rextsize;
2251		ralen *= mp->m_sb.sb_rextsize;
2252		ap->length = ralen;
2253		ap->ip->i_d.di_nblocks += ralen;
2254		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2255		if (ap->wasdel)
2256			ap->ip->i_delayed_blks -= ralen;
2257		/*
2258		 * Adjust the disk quota also. This was reserved
2259		 * earlier.
2260		 */
2261		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2262			ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2263					XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2264	} else {
2265		ap->length = 0;
2266	}
2267	return 0;
2268}
2269
2270STATIC int
2271xfs_bmap_btalloc_nullfb(
2272	struct xfs_bmalloca	*ap,
2273	struct xfs_alloc_arg	*args,
2274	xfs_extlen_t		*blen)
2275{
2276	struct xfs_mount	*mp = ap->ip->i_mount;
2277	struct xfs_perag	*pag;
2278	xfs_agnumber_t		ag, startag;
2279	int			notinit = 0;
2280	int			error;
2281
2282	if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2283		args->type = XFS_ALLOCTYPE_NEAR_BNO;
2284	else
2285		args->type = XFS_ALLOCTYPE_START_BNO;
2286	args->total = ap->total;
2287
2288	/*
2289	 * Search for an allocation group with a single extent large enough
2290	 * for the request.  If one isn't found, then adjust the minimum
2291	 * allocation size to the largest space found.
2292	 */
2293	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2294	if (startag == NULLAGNUMBER)
2295		startag = ag = 0;
2296
2297	pag = xfs_perag_get(mp, ag);
2298	while (*blen < args->maxlen) {
2299		if (!pag->pagf_init) {
2300			error = xfs_alloc_pagf_init(mp, args->tp, ag,
2301						    XFS_ALLOC_FLAG_TRYLOCK);
2302			if (error) {
2303				xfs_perag_put(pag);
2304				return error;
2305			}
2306		}
2307
2308		/*
2309		 * See xfs_alloc_fix_freelist...
2310		 */
2311		if (pag->pagf_init) {
2312			xfs_extlen_t	longest;
2313			longest = xfs_alloc_longest_free_extent(mp, pag);
2314			if (*blen < longest)
2315				*blen = longest;
2316		} else
2317			notinit = 1;
2318
2319		if (xfs_inode_is_filestream(ap->ip)) {
2320			if (*blen >= args->maxlen)
2321				break;
2322
2323			if (ap->userdata) {
2324				/*
2325				 * If startag is an invalid AG, we've
2326				 * come here once before and
2327				 * xfs_filestream_new_ag picked the
2328				 * best currently available.
2329				 *
2330				 * Don't continue looping, since we
2331				 * could loop forever.
2332				 */
2333				if (startag == NULLAGNUMBER)
2334					break;
2335
2336				error = xfs_filestream_new_ag(ap, &ag);
2337				xfs_perag_put(pag);
2338				if (error)
2339					return error;
2340
2341				/* loop again to set 'blen'*/
2342				startag = NULLAGNUMBER;
2343				pag = xfs_perag_get(mp, ag);
2344				continue;
2345			}
2346		}
2347		if (++ag == mp->m_sb.sb_agcount)
2348			ag = 0;
2349		if (ag == startag)
2350			break;
2351		xfs_perag_put(pag);
2352		pag = xfs_perag_get(mp, ag);
2353	}
2354	xfs_perag_put(pag);
2355
2356	/*
2357	 * Since the above loop did a BUF_TRYLOCK, it is
2358	 * possible that there is space for this request.
2359	 */
2360	if (notinit || *blen < ap->minlen)
2361		args->minlen = ap->minlen;
2362	/*
2363	 * If the best seen length is less than the request
2364	 * length, use the best as the minimum.
2365	 */
2366	else if (*blen < args->maxlen)
2367		args->minlen = *blen;
2368	/*
2369	 * Otherwise we've seen an extent as big as maxlen,
2370	 * use that as the minimum.
2371	 */
2372	else
2373		args->minlen = args->maxlen;
2374
2375	/*
2376	 * set the failure fallback case to look in the selected
2377	 * AG as the stream may have moved.
2378	 */
2379	if (xfs_inode_is_filestream(ap->ip))
2380		ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2381
2382	return 0;
2383}
2384
2385STATIC int
2386xfs_bmap_btalloc(
2387	xfs_bmalloca_t	*ap)		/* bmap alloc argument struct */
2388{
2389	xfs_mount_t	*mp;		/* mount point structure */
2390	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
2391	xfs_extlen_t	align;		/* minimum allocation alignment */
2392	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
2393	xfs_agnumber_t	ag;
2394	xfs_alloc_arg_t	args;
2395	xfs_extlen_t	blen;
2396	xfs_extlen_t	nextminlen = 0;
2397	int		nullfb;		/* true if ap->firstblock isn't set */
2398	int		isaligned;
2399	int		tryagain;
2400	int		error;
 
2401
2402	ASSERT(ap->length);
2403
2404	mp = ap->ip->i_mount;
 
 
 
 
 
 
 
 
2405	align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2406	if (unlikely(align)) {
2407		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2408						align, 0, ap->eof, 0, ap->conv,
2409						&ap->offset, &ap->length);
2410		ASSERT(!error);
2411		ASSERT(ap->length);
2412	}
 
 
2413	nullfb = *ap->firstblock == NULLFSBLOCK;
2414	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2415	if (nullfb) {
2416		if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2417			ag = xfs_filestream_lookup_ag(ap->ip);
2418			ag = (ag != NULLAGNUMBER) ? ag : 0;
2419			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2420		} else {
2421			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2422		}
2423	} else
2424		ap->blkno = *ap->firstblock;
2425
2426	xfs_bmap_adjacent(ap);
2427
2428	/*
2429	 * If allowed, use ap->blkno; otherwise must use firstblock since
2430	 * it's in the right allocation group.
2431	 */
2432	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2433		;
2434	else
2435		ap->blkno = *ap->firstblock;
2436	/*
2437	 * Normal allocation, done through xfs_alloc_vextent.
2438	 */
2439	tryagain = isaligned = 0;
 
2440	args.tp = ap->tp;
2441	args.mp = mp;
2442	args.fsbno = ap->blkno;
2443
2444	/* Trim the allocation back to the maximum an AG can fit. */
2445	args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2446	args.firstblock = *ap->firstblock;
2447	blen = 0;
2448	if (nullfb) {
2449		error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2450		if (error)
2451			return error;
2452	} else if (ap->flist->xbf_low) {
2453		if (xfs_inode_is_filestream(ap->ip))
2454			args.type = XFS_ALLOCTYPE_FIRST_AG;
2455		else
2456			args.type = XFS_ALLOCTYPE_START_BNO;
2457		args.total = args.minlen = ap->minlen;
2458	} else {
2459		args.type = XFS_ALLOCTYPE_NEAR_BNO;
2460		args.total = ap->total;
2461		args.minlen = ap->minlen;
2462	}
2463	/* apply extent size hints if obtained earlier */
2464	if (unlikely(align)) {
2465		args.prod = align;
2466		if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2467			args.mod = (xfs_extlen_t)(args.prod - args.mod);
2468	} else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2469		args.prod = 1;
2470		args.mod = 0;
2471	} else {
2472		args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2473		if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2474			args.mod = (xfs_extlen_t)(args.prod - args.mod);
2475	}
2476	/*
2477	 * If we are not low on available data blocks, and the
2478	 * underlying logical volume manager is a stripe, and
2479	 * the file offset is zero then try to allocate data
2480	 * blocks on stripe unit boundary.
2481	 * NOTE: ap->aeof is only set if the allocation length
2482	 * is >= the stripe unit and the allocation offset is
2483	 * at the end of file.
2484	 */
2485	if (!ap->flist->xbf_low && ap->aeof) {
2486		if (!ap->offset) {
2487			args.alignment = mp->m_dalign;
2488			atype = args.type;
2489			isaligned = 1;
2490			/*
2491			 * Adjust for alignment
2492			 */
2493			if (blen > args.alignment && blen <= args.maxlen)
2494				args.minlen = blen - args.alignment;
2495			args.minalignslop = 0;
2496		} else {
2497			/*
2498			 * First try an exact bno allocation.
2499			 * If it fails then do a near or start bno
2500			 * allocation with alignment turned on.
2501			 */
2502			atype = args.type;
2503			tryagain = 1;
2504			args.type = XFS_ALLOCTYPE_THIS_BNO;
2505			args.alignment = 1;
2506			/*
2507			 * Compute the minlen+alignment for the
2508			 * next case.  Set slop so that the value
2509			 * of minlen+alignment+slop doesn't go up
2510			 * between the calls.
2511			 */
2512			if (blen > mp->m_dalign && blen <= args.maxlen)
2513				nextminlen = blen - mp->m_dalign;
2514			else
2515				nextminlen = args.minlen;
2516			if (nextminlen + mp->m_dalign > args.minlen + 1)
2517				args.minalignslop =
2518					nextminlen + mp->m_dalign -
2519					args.minlen - 1;
2520			else
2521				args.minalignslop = 0;
2522		}
2523	} else {
2524		args.alignment = 1;
2525		args.minalignslop = 0;
2526	}
2527	args.minleft = ap->minleft;
2528	args.wasdel = ap->wasdel;
2529	args.isfl = 0;
2530	args.userdata = ap->userdata;
2531	if ((error = xfs_alloc_vextent(&args)))
2532		return error;
2533	if (tryagain && args.fsbno == NULLFSBLOCK) {
2534		/*
2535		 * Exact allocation failed. Now try with alignment
2536		 * turned on.
2537		 */
2538		args.type = atype;
2539		args.fsbno = ap->blkno;
2540		args.alignment = mp->m_dalign;
2541		args.minlen = nextminlen;
2542		args.minalignslop = 0;
2543		isaligned = 1;
2544		if ((error = xfs_alloc_vextent(&args)))
2545			return error;
2546	}
2547	if (isaligned && args.fsbno == NULLFSBLOCK) {
2548		/*
2549		 * allocation failed, so turn off alignment and
2550		 * try again.
2551		 */
2552		args.type = atype;
2553		args.fsbno = ap->blkno;
2554		args.alignment = 0;
2555		if ((error = xfs_alloc_vextent(&args)))
2556			return error;
2557	}
2558	if (args.fsbno == NULLFSBLOCK && nullfb &&
2559	    args.minlen > ap->minlen) {
2560		args.minlen = ap->minlen;
2561		args.type = XFS_ALLOCTYPE_START_BNO;
2562		args.fsbno = ap->blkno;
2563		if ((error = xfs_alloc_vextent(&args)))
2564			return error;
2565	}
2566	if (args.fsbno == NULLFSBLOCK && nullfb) {
2567		args.fsbno = 0;
2568		args.type = XFS_ALLOCTYPE_FIRST_AG;
2569		args.total = ap->minlen;
2570		args.minleft = 0;
2571		if ((error = xfs_alloc_vextent(&args)))
2572			return error;
2573		ap->flist->xbf_low = 1;
2574	}
2575	if (args.fsbno != NULLFSBLOCK) {
2576		/*
2577		 * check the allocation happened at the same or higher AG than
2578		 * the first block that was allocated.
2579		 */
2580		ASSERT(*ap->firstblock == NULLFSBLOCK ||
2581		       XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2582		       XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2583		       (ap->flist->xbf_low &&
2584			XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2585			XFS_FSB_TO_AGNO(mp, args.fsbno)));
2586
2587		ap->blkno = args.fsbno;
2588		if (*ap->firstblock == NULLFSBLOCK)
2589			*ap->firstblock = args.fsbno;
2590		ASSERT(nullfb || fb_agno == args.agno ||
2591		       (ap->flist->xbf_low && fb_agno < args.agno));
2592		ap->length = args.len;
2593		ap->ip->i_d.di_nblocks += args.len;
2594		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2595		if (ap->wasdel)
2596			ap->ip->i_delayed_blks -= args.len;
2597		/*
2598		 * Adjust the disk quota also. This was reserved
2599		 * earlier.
2600		 */
2601		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2602			ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2603					XFS_TRANS_DQ_BCOUNT,
2604			(long) args.len);
2605	} else {
2606		ap->blkno = NULLFSBLOCK;
2607		ap->length = 0;
2608	}
2609	return 0;
2610}
2611
2612/*
2613 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2614 * It figures out where to ask the underlying allocator to put the new extent.
2615 */
2616STATIC int
2617xfs_bmap_alloc(
2618	xfs_bmalloca_t	*ap)		/* bmap alloc argument struct */
2619{
2620	if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2621		return xfs_bmap_rtalloc(ap);
2622	return xfs_bmap_btalloc(ap);
2623}
2624
2625/*
2626 * Transform a btree format file with only one leaf node, where the
2627 * extents list will fit in the inode, into an extents format file.
2628 * Since the file extents are already in-core, all we have to do is
2629 * give up the space for the btree root and pitch the leaf block.
2630 */
2631STATIC int				/* error */
2632xfs_bmap_btree_to_extents(
2633	xfs_trans_t		*tp,	/* transaction pointer */
2634	xfs_inode_t		*ip,	/* incore inode pointer */
2635	xfs_btree_cur_t		*cur,	/* btree cursor */
2636	int			*logflagsp, /* inode logging flags */
2637	int			whichfork)  /* data or attr fork */
2638{
2639	/* REFERENCED */
2640	struct xfs_btree_block	*cblock;/* child btree block */
2641	xfs_fsblock_t		cbno;	/* child block number */
2642	xfs_buf_t		*cbp;	/* child block's buffer */
2643	int			error;	/* error return value */
2644	xfs_ifork_t		*ifp;	/* inode fork data */
2645	xfs_mount_t		*mp;	/* mount point structure */
2646	__be64			*pp;	/* ptr to block address */
2647	struct xfs_btree_block	*rblock;/* root btree block */
2648
2649	mp = ip->i_mount;
2650	ifp = XFS_IFORK_PTR(ip, whichfork);
2651	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2652	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2653	rblock = ifp->if_broot;
2654	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2655	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2656	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2657	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2658	cbno = be64_to_cpu(*pp);
2659	*logflagsp = 0;
2660#ifdef DEBUG
2661	if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2662		return error;
2663#endif
2664	if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2665			XFS_BMAP_BTREE_REF)))
2666		return error;
2667	cblock = XFS_BUF_TO_BLOCK(cbp);
2668	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2669		return error;
2670	xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2671	ip->i_d.di_nblocks--;
2672	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2673	xfs_trans_binval(tp, cbp);
2674	if (cur->bc_bufs[0] == cbp)
2675		cur->bc_bufs[0] = NULL;
2676	xfs_iroot_realloc(ip, -1, whichfork);
2677	ASSERT(ifp->if_broot == NULL);
2678	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2679	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2680	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2681	return 0;
2682}
2683
2684/*
2685 * Called by xfs_bmapi to update file extent records and the btree
2686 * after removing space (or undoing a delayed allocation).
2687 */
2688STATIC int				/* error */
2689xfs_bmap_del_extent(
2690	xfs_inode_t		*ip,	/* incore inode pointer */
2691	xfs_trans_t		*tp,	/* current transaction pointer */
2692	xfs_extnum_t		*idx,	/* extent number to update/delete */
2693	xfs_bmap_free_t		*flist,	/* list of extents to be freed */
2694	xfs_btree_cur_t		*cur,	/* if null, not a btree */
2695	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
2696	int			*logflagsp, /* inode logging flags */
2697	int			whichfork) /* data or attr fork */
2698{
2699	xfs_filblks_t		da_new;	/* new delay-alloc indirect blocks */
2700	xfs_filblks_t		da_old;	/* old delay-alloc indirect blocks */
2701	xfs_fsblock_t		del_endblock=0;	/* first block past del */
2702	xfs_fileoff_t		del_endoff;	/* first offset past del */
2703	int			delay;	/* current block is delayed allocated */
2704	int			do_fx;	/* free extent at end of routine */
2705	xfs_bmbt_rec_host_t	*ep;	/* current extent entry pointer */
2706	int			error;	/* error return value */
2707	int			flags;	/* inode logging flags */
2708	xfs_bmbt_irec_t		got;	/* current extent entry */
2709	xfs_fileoff_t		got_endoff;	/* first offset past got */
2710	int			i;	/* temp state */
2711	xfs_ifork_t		*ifp;	/* inode fork pointer */
2712	xfs_mount_t		*mp;	/* mount structure */
2713	xfs_filblks_t		nblks;	/* quota/sb block count */
2714	xfs_bmbt_irec_t		new;	/* new record to be inserted */
2715	/* REFERENCED */
2716	uint			qfield;	/* quota field to update */
2717	xfs_filblks_t		temp;	/* for indirect length calculations */
2718	xfs_filblks_t		temp2;	/* for indirect length calculations */
2719	int			state = 0;
2720
2721	XFS_STATS_INC(xs_del_exlist);
2722
2723	if (whichfork == XFS_ATTR_FORK)
2724		state |= BMAP_ATTRFORK;
2725
2726	mp = ip->i_mount;
2727	ifp = XFS_IFORK_PTR(ip, whichfork);
2728	ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2729		(uint)sizeof(xfs_bmbt_rec_t)));
2730	ASSERT(del->br_blockcount > 0);
2731	ep = xfs_iext_get_ext(ifp, *idx);
2732	xfs_bmbt_get_all(ep, &got);
2733	ASSERT(got.br_startoff <= del->br_startoff);
2734	del_endoff = del->br_startoff + del->br_blockcount;
2735	got_endoff = got.br_startoff + got.br_blockcount;
2736	ASSERT(got_endoff >= del_endoff);
2737	delay = isnullstartblock(got.br_startblock);
2738	ASSERT(isnullstartblock(del->br_startblock) == delay);
2739	flags = 0;
2740	qfield = 0;
2741	error = 0;
2742	/*
2743	 * If deleting a real allocation, must free up the disk space.
2744	 */
2745	if (!delay) {
2746		flags = XFS_ILOG_CORE;
2747		/*
2748		 * Realtime allocation.  Free it and record di_nblocks update.
2749		 */
2750		if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2751			xfs_fsblock_t	bno;
2752			xfs_filblks_t	len;
2753
2754			ASSERT(do_mod(del->br_blockcount,
2755				      mp->m_sb.sb_rextsize) == 0);
2756			ASSERT(do_mod(del->br_startblock,
2757				      mp->m_sb.sb_rextsize) == 0);
2758			bno = del->br_startblock;
2759			len = del->br_blockcount;
2760			do_div(bno, mp->m_sb.sb_rextsize);
2761			do_div(len, mp->m_sb.sb_rextsize);
2762			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2763			if (error)
2764				goto done;
2765			do_fx = 0;
2766			nblks = len * mp->m_sb.sb_rextsize;
2767			qfield = XFS_TRANS_DQ_RTBCOUNT;
2768		}
2769		/*
2770		 * Ordinary allocation.
2771		 */
2772		else {
2773			do_fx = 1;
2774			nblks = del->br_blockcount;
2775			qfield = XFS_TRANS_DQ_BCOUNT;
2776		}
2777		/*
2778		 * Set up del_endblock and cur for later.
2779		 */
2780		del_endblock = del->br_startblock + del->br_blockcount;
2781		if (cur) {
2782			if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2783					got.br_startblock, got.br_blockcount,
2784					&i)))
2785				goto done;
2786			XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2787		}
2788		da_old = da_new = 0;
2789	} else {
2790		da_old = startblockval(got.br_startblock);
2791		da_new = 0;
2792		nblks = 0;
2793		do_fx = 0;
2794	}
2795	/*
2796	 * Set flag value to use in switch statement.
2797	 * Left-contig is 2, right-contig is 1.
2798	 */
2799	switch (((got.br_startoff == del->br_startoff) << 1) |
2800		(got_endoff == del_endoff)) {
2801	case 3:
2802		/*
2803		 * Matches the whole extent.  Delete the entry.
2804		 */
2805		xfs_iext_remove(ip, *idx, 1,
2806				whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2807		--*idx;
2808		if (delay)
2809			break;
2810
2811		XFS_IFORK_NEXT_SET(ip, whichfork,
2812			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2813		flags |= XFS_ILOG_CORE;
2814		if (!cur) {
2815			flags |= xfs_ilog_fext(whichfork);
2816			break;
2817		}
2818		if ((error = xfs_btree_delete(cur, &i)))
2819			goto done;
2820		XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2821		break;
2822
2823	case 2:
2824		/*
2825		 * Deleting the first part of the extent.
2826		 */
2827		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2828		xfs_bmbt_set_startoff(ep, del_endoff);
2829		temp = got.br_blockcount - del->br_blockcount;
2830		xfs_bmbt_set_blockcount(ep, temp);
2831		if (delay) {
2832			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2833				da_old);
2834			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2835			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2836			da_new = temp;
2837			break;
2838		}
2839		xfs_bmbt_set_startblock(ep, del_endblock);
2840		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2841		if (!cur) {
2842			flags |= xfs_ilog_fext(whichfork);
2843			break;
2844		}
2845		if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2846				got.br_blockcount - del->br_blockcount,
2847				got.br_state)))
2848			goto done;
2849		break;
2850
2851	case 1:
2852		/*
2853		 * Deleting the last part of the extent.
2854		 */
2855		temp = got.br_blockcount - del->br_blockcount;
2856		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2857		xfs_bmbt_set_blockcount(ep, temp);
2858		if (delay) {
2859			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2860				da_old);
2861			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2862			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2863			da_new = temp;
2864			break;
2865		}
2866		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2867		if (!cur) {
2868			flags |= xfs_ilog_fext(whichfork);
2869			break;
2870		}
2871		if ((error = xfs_bmbt_update(cur, got.br_startoff,
2872				got.br_startblock,
2873				got.br_blockcount - del->br_blockcount,
2874				got.br_state)))
2875			goto done;
2876		break;
2877
2878	case 0:
2879		/*
2880		 * Deleting the middle of the extent.
2881		 */
2882		temp = del->br_startoff - got.br_startoff;
2883		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2884		xfs_bmbt_set_blockcount(ep, temp);
2885		new.br_startoff = del_endoff;
2886		temp2 = got_endoff - del_endoff;
2887		new.br_blockcount = temp2;
2888		new.br_state = got.br_state;
2889		if (!delay) {
2890			new.br_startblock = del_endblock;
2891			flags |= XFS_ILOG_CORE;
2892			if (cur) {
2893				if ((error = xfs_bmbt_update(cur,
2894						got.br_startoff,
2895						got.br_startblock, temp,
2896						got.br_state)))
2897					goto done;
2898				if ((error = xfs_btree_increment(cur, 0, &i)))
2899					goto done;
2900				cur->bc_rec.b = new;
2901				error = xfs_btree_insert(cur, &i);
2902				if (error && error != ENOSPC)
2903					goto done;
2904				/*
2905				 * If get no-space back from btree insert,
2906				 * it tried a split, and we have a zero
2907				 * block reservation.
2908				 * Fix up our state and return the error.
2909				 */
2910				if (error == ENOSPC) {
2911					/*
2912					 * Reset the cursor, don't trust
2913					 * it after any insert operation.
2914					 */
2915					if ((error = xfs_bmbt_lookup_eq(cur,
2916							got.br_startoff,
2917							got.br_startblock,
2918							temp, &i)))
2919						goto done;
2920					XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2921					/*
2922					 * Update the btree record back
2923					 * to the original value.
2924					 */
2925					if ((error = xfs_bmbt_update(cur,
2926							got.br_startoff,
2927							got.br_startblock,
2928							got.br_blockcount,
2929							got.br_state)))
2930						goto done;
2931					/*
2932					 * Reset the extent record back
2933					 * to the original value.
2934					 */
2935					xfs_bmbt_set_blockcount(ep,
2936						got.br_blockcount);
2937					flags = 0;
2938					error = XFS_ERROR(ENOSPC);
2939					goto done;
2940				}
2941				XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2942			} else
2943				flags |= xfs_ilog_fext(whichfork);
2944			XFS_IFORK_NEXT_SET(ip, whichfork,
2945				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2946		} else {
2947			ASSERT(whichfork == XFS_DATA_FORK);
2948			temp = xfs_bmap_worst_indlen(ip, temp);
2949			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2950			temp2 = xfs_bmap_worst_indlen(ip, temp2);
2951			new.br_startblock = nullstartblock((int)temp2);
2952			da_new = temp + temp2;
2953			while (da_new > da_old) {
2954				if (temp) {
2955					temp--;
2956					da_new--;
2957					xfs_bmbt_set_startblock(ep,
2958						nullstartblock((int)temp));
2959				}
2960				if (da_new == da_old)
2961					break;
2962				if (temp2) {
2963					temp2--;
2964					da_new--;
2965					new.br_startblock =
2966						nullstartblock((int)temp2);
2967				}
2968			}
2969		}
2970		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2971		xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2972		++*idx;
2973		break;
2974	}
2975	/*
2976	 * If we need to, add to list of extents to delete.
2977	 */
2978	if (do_fx)
2979		xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2980			mp);
2981	/*
2982	 * Adjust inode # blocks in the file.
2983	 */
2984	if (nblks)
2985		ip->i_d.di_nblocks -= nblks;
2986	/*
2987	 * Adjust quota data.
2988	 */
2989	if (qfield)
2990		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2991
2992	/*
2993	 * Account for change in delayed indirect blocks.
2994	 * Nothing to do for disk quota accounting here.
2995	 */
2996	ASSERT(da_old >= da_new);
2997	if (da_old > da_new) {
2998		xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
2999			(int64_t)(da_old - da_new), 0);
3000	}
3001done:
3002	*logflagsp = flags;
3003	return error;
3004}
3005
3006/*
3007 * Remove the entry "free" from the free item list.  Prev points to the
3008 * previous entry, unless "free" is the head of the list.
3009 */
3010STATIC void
3011xfs_bmap_del_free(
3012	xfs_bmap_free_t		*flist,	/* free item list header */
3013	xfs_bmap_free_item_t	*prev,	/* previous item on list, if any */
3014	xfs_bmap_free_item_t	*free)	/* list item to be freed */
3015{
3016	if (prev)
3017		prev->xbfi_next = free->xbfi_next;
3018	else
3019		flist->xbf_first = free->xbfi_next;
3020	flist->xbf_count--;
3021	kmem_zone_free(xfs_bmap_free_item_zone, free);
3022}
3023
3024/*
3025 * Convert an extents-format file into a btree-format file.
3026 * The new file will have a root block (in the inode) and a single child block.
3027 */
3028STATIC int					/* error */
3029xfs_bmap_extents_to_btree(
3030	xfs_trans_t		*tp,		/* transaction pointer */
3031	xfs_inode_t		*ip,		/* incore inode pointer */
3032	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
3033	xfs_bmap_free_t		*flist,		/* blocks freed in xaction */
3034	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
3035	int			wasdel,		/* converting a delayed alloc */
3036	int			*logflagsp,	/* inode logging flags */
3037	int			whichfork)	/* data or attr fork */
3038{
3039	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
3040	xfs_buf_t		*abp;		/* buffer for ablock */
3041	xfs_alloc_arg_t		args;		/* allocation arguments */
3042	xfs_bmbt_rec_t		*arp;		/* child record pointer */
3043	struct xfs_btree_block	*block;		/* btree root block */
3044	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
3045	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
3046	int			error;		/* error return value */
3047	xfs_extnum_t		i, cnt;		/* extent record index */
3048	xfs_ifork_t		*ifp;		/* inode fork pointer */
3049	xfs_bmbt_key_t		*kp;		/* root block key pointer */
3050	xfs_mount_t		*mp;		/* mount structure */
3051	xfs_extnum_t		nextents;	/* number of file extents */
3052	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
3053
3054	ifp = XFS_IFORK_PTR(ip, whichfork);
3055	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3056
3057	/*
3058	 * Make space in the inode incore.
3059	 */
3060	xfs_iroot_realloc(ip, 1, whichfork);
3061	ifp->if_flags |= XFS_IFBROOT;
3062
3063	/*
3064	 * Fill in the root.
3065	 */
3066	block = ifp->if_broot;
3067	block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3068	block->bb_level = cpu_to_be16(1);
3069	block->bb_numrecs = cpu_to_be16(1);
3070	block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3071	block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3072
3073	/*
3074	 * Need a cursor.  Can't allocate until bb_level is filled in.
3075	 */
3076	mp = ip->i_mount;
3077	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3078	cur->bc_private.b.firstblock = *firstblock;
3079	cur->bc_private.b.flist = flist;
3080	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3081	/*
3082	 * Convert to a btree with two levels, one record in root.
3083	 */
3084	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3085	args.tp = tp;
3086	args.mp = mp;
3087	args.firstblock = *firstblock;
3088	if (*firstblock == NULLFSBLOCK) {
3089		args.type = XFS_ALLOCTYPE_START_BNO;
3090		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3091	} else if (flist->xbf_low) {
3092		args.type = XFS_ALLOCTYPE_START_BNO;
3093		args.fsbno = *firstblock;
3094	} else {
3095		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3096		args.fsbno = *firstblock;
3097	}
3098	args.minlen = args.maxlen = args.prod = 1;
3099	args.total = args.minleft = args.alignment = args.mod = args.isfl =
3100		args.minalignslop = 0;
3101	args.wasdel = wasdel;
3102	*logflagsp = 0;
3103	if ((error = xfs_alloc_vextent(&args))) {
3104		xfs_iroot_realloc(ip, -1, whichfork);
3105		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3106		return error;
3107	}
3108	/*
3109	 * Allocation can't fail, the space was reserved.
3110	 */
3111	ASSERT(args.fsbno != NULLFSBLOCK);
3112	ASSERT(*firstblock == NULLFSBLOCK ||
3113	       args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3114	       (flist->xbf_low &&
3115		args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3116	*firstblock = cur->bc_private.b.firstblock = args.fsbno;
3117	cur->bc_private.b.allocated++;
3118	ip->i_d.di_nblocks++;
3119	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3120	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3121	/*
3122	 * Fill in the child block.
3123	 */
3124	ablock = XFS_BUF_TO_BLOCK(abp);
3125	ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3126	ablock->bb_level = 0;
3127	ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3128	ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3129	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3130	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3131	for (cnt = i = 0; i < nextents; i++) {
3132		ep = xfs_iext_get_ext(ifp, i);
3133		if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3134			arp->l0 = cpu_to_be64(ep->l0);
3135			arp->l1 = cpu_to_be64(ep->l1);
3136			arp++; cnt++;
3137		}
3138	}
3139	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3140	xfs_btree_set_numrecs(ablock, cnt);
3141
3142	/*
3143	 * Fill in the root key and pointer.
3144	 */
3145	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3146	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3147	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3148	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3149						be16_to_cpu(block->bb_level)));
3150	*pp = cpu_to_be64(args.fsbno);
3151
3152	/*
3153	 * Do all this logging at the end so that
3154	 * the root is at the right level.
3155	 */
3156	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3157	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3158	ASSERT(*curp == NULL);
3159	*curp = cur;
3160	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3161	return 0;
3162}
3163
3164/*
3165 * Calculate the default attribute fork offset for newly created inodes.
3166 */
3167uint
3168xfs_default_attroffset(
3169	struct xfs_inode	*ip)
3170{
3171	struct xfs_mount	*mp = ip->i_mount;
3172	uint			offset;
3173
3174	if (mp->m_sb.sb_inodesize == 256) {
3175		offset = XFS_LITINO(mp) -
3176				XFS_BMDR_SPACE_CALC(MINABTPTRS);
3177	} else {
3178		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3179	}
3180
3181	ASSERT(offset < XFS_LITINO(mp));
3182	return offset;
3183}
3184
3185/*
3186 * Helper routine to reset inode di_forkoff field when switching
3187 * attribute fork from local to extent format - we reset it where
3188 * possible to make space available for inline data fork extents.
3189 */
3190STATIC void
3191xfs_bmap_forkoff_reset(
3192	xfs_mount_t	*mp,
3193	xfs_inode_t	*ip,
3194	int		whichfork)
3195{
3196	if (whichfork == XFS_ATTR_FORK &&
3197	    ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3198	    ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3199	    ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3200		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3201
3202		if (dfl_forkoff > ip->i_d.di_forkoff)
3203			ip->i_d.di_forkoff = dfl_forkoff;
3204	}
3205}
3206
3207/*
3208 * Convert a local file to an extents file.
3209 * This code is out of bounds for data forks of regular files,
3210 * since the file data needs to get logged so things will stay consistent.
3211 * (The bmap-level manipulations are ok, though).
3212 */
3213STATIC int				/* error */
3214xfs_bmap_local_to_extents(
3215	xfs_trans_t	*tp,		/* transaction pointer */
3216	xfs_inode_t	*ip,		/* incore inode pointer */
3217	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
3218	xfs_extlen_t	total,		/* total blocks needed by transaction */
3219	int		*logflagsp,	/* inode logging flags */
3220	int		whichfork)	/* data or attr fork */
3221{
3222	int		error;		/* error return value */
3223	int		flags;		/* logging flags returned */
3224	xfs_ifork_t	*ifp;		/* inode fork pointer */
3225
3226	/*
3227	 * We don't want to deal with the case of keeping inode data inline yet.
3228	 * So sending the data fork of a regular inode is invalid.
3229	 */
3230	ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3231	ifp = XFS_IFORK_PTR(ip, whichfork);
3232	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3233	flags = 0;
3234	error = 0;
3235	if (ifp->if_bytes) {
3236		xfs_alloc_arg_t	args;	/* allocation arguments */
3237		xfs_buf_t	*bp;	/* buffer for extent block */
3238		xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3239
3240		args.tp = tp;
3241		args.mp = ip->i_mount;
3242		args.firstblock = *firstblock;
3243		ASSERT((ifp->if_flags &
3244			(XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3245		/*
3246		 * Allocate a block.  We know we need only one, since the
3247		 * file currently fits in an inode.
3248		 */
3249		if (*firstblock == NULLFSBLOCK) {
3250			args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3251			args.type = XFS_ALLOCTYPE_START_BNO;
3252		} else {
3253			args.fsbno = *firstblock;
3254			args.type = XFS_ALLOCTYPE_NEAR_BNO;
3255		}
3256		args.total = total;
3257		args.mod = args.minleft = args.alignment = args.wasdel =
3258			args.isfl = args.minalignslop = 0;
3259		args.minlen = args.maxlen = args.prod = 1;
3260		if ((error = xfs_alloc_vextent(&args)))
3261			goto done;
3262		/*
3263		 * Can't fail, the space was reserved.
3264		 */
3265		ASSERT(args.fsbno != NULLFSBLOCK);
3266		ASSERT(args.len == 1);
3267		*firstblock = args.fsbno;
3268		bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3269		memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3270		xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3271		xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3272		xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3273		xfs_iext_add(ifp, 0, 1);
3274		ep = xfs_iext_get_ext(ifp, 0);
3275		xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3276		trace_xfs_bmap_post_update(ip, 0,
3277				whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3278				_THIS_IP_);
3279		XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3280		ip->i_d.di_nblocks = 1;
3281		xfs_trans_mod_dquot_byino(tp, ip,
3282			XFS_TRANS_DQ_BCOUNT, 1L);
3283		flags |= xfs_ilog_fext(whichfork);
3284	} else {
3285		ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3286		xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3287	}
3288	ifp->if_flags &= ~XFS_IFINLINE;
3289	ifp->if_flags |= XFS_IFEXTENTS;
3290	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3291	flags |= XFS_ILOG_CORE;
3292done:
3293	*logflagsp = flags;
3294	return error;
3295}
3296
3297/*
3298 * Search the extent records for the entry containing block bno.
3299 * If bno lies in a hole, point to the next entry.  If bno lies
3300 * past eof, *eofp will be set, and *prevp will contain the last
3301 * entry (null if none).  Else, *lastxp will be set to the index
3302 * of the found entry; *gotp will contain the entry.
3303 */
3304STATIC xfs_bmbt_rec_host_t *		/* pointer to found extent entry */
3305xfs_bmap_search_multi_extents(
3306	xfs_ifork_t	*ifp,		/* inode fork pointer */
3307	xfs_fileoff_t	bno,		/* block number searched for */
3308	int		*eofp,		/* out: end of file found */
3309	xfs_extnum_t	*lastxp,	/* out: last extent index */
3310	xfs_bmbt_irec_t	*gotp,		/* out: extent entry found */
3311	xfs_bmbt_irec_t	*prevp)		/* out: previous extent entry found */
3312{
3313	xfs_bmbt_rec_host_t *ep;		/* extent record pointer */
3314	xfs_extnum_t	lastx;		/* last extent index */
3315
3316	/*
3317	 * Initialize the extent entry structure to catch access to
3318	 * uninitialized br_startblock field.
3319	 */
3320	gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3321	gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3322	gotp->br_state = XFS_EXT_INVALID;
3323#if XFS_BIG_BLKNOS
3324	gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3325#else
3326	gotp->br_startblock = 0xffffa5a5;
3327#endif
3328	prevp->br_startoff = NULLFILEOFF;
3329
3330	ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3331	if (lastx > 0) {
3332		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3333	}
3334	if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3335		xfs_bmbt_get_all(ep, gotp);
3336		*eofp = 0;
3337	} else {
3338		if (lastx > 0) {
3339			*gotp = *prevp;
3340		}
3341		*eofp = 1;
3342		ep = NULL;
3343	}
3344	*lastxp = lastx;
3345	return ep;
3346}
3347
3348/*
3349 * Search the extents list for the inode, for the extent containing bno.
3350 * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3351 * *eofp will be set, and *prevp will contain the last entry (null if none).
3352 * Else, *lastxp will be set to the index of the found
3353 * entry; *gotp will contain the entry.
3354 */
3355STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3356xfs_bmap_search_extents(
3357	xfs_inode_t     *ip,            /* incore inode pointer */
3358	xfs_fileoff_t   bno,            /* block number searched for */
3359	int             fork,      	/* data or attr fork */
3360	int             *eofp,          /* out: end of file found */
3361	xfs_extnum_t    *lastxp,        /* out: last extent index */
3362	xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3363	xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3364{
3365	xfs_ifork_t	*ifp;		/* inode fork pointer */
3366	xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3367
3368	XFS_STATS_INC(xs_look_exlist);
3369	ifp = XFS_IFORK_PTR(ip, fork);
3370
3371	ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3372
3373	if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3374		     !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3375		xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3376				"Access to block zero in inode %llu "
3377				"start_block: %llx start_off: %llx "
3378				"blkcnt: %llx extent-state: %x lastx: %x\n",
3379			(unsigned long long)ip->i_ino,
3380			(unsigned long long)gotp->br_startblock,
3381			(unsigned long long)gotp->br_startoff,
3382			(unsigned long long)gotp->br_blockcount,
3383			gotp->br_state, *lastxp);
3384		*lastxp = NULLEXTNUM;
3385		*eofp = 1;
3386		return NULL;
3387	}
3388	return ep;
3389}
3390
3391/*
3392 * Compute the worst-case number of indirect blocks that will be used
3393 * for ip's delayed extent of length "len".
3394 */
3395STATIC xfs_filblks_t
3396xfs_bmap_worst_indlen(
3397	xfs_inode_t	*ip,		/* incore inode pointer */
3398	xfs_filblks_t	len)		/* delayed extent length */
3399{
3400	int		level;		/* btree level number */
3401	int		maxrecs;	/* maximum record count at this level */
3402	xfs_mount_t	*mp;		/* mount structure */
3403	xfs_filblks_t	rval;		/* return value */
3404
3405	mp = ip->i_mount;
3406	maxrecs = mp->m_bmap_dmxr[0];
3407	for (level = 0, rval = 0;
3408	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3409	     level++) {
3410		len += maxrecs - 1;
3411		do_div(len, maxrecs);
3412		rval += len;
3413		if (len == 1)
3414			return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3415				level - 1;
3416		if (level == 0)
3417			maxrecs = mp->m_bmap_dmxr[1];
3418	}
3419	return rval;
3420}
3421
3422/*
3423 * Convert inode from non-attributed to attributed.
3424 * Must not be in a transaction, ip must not be locked.
3425 */
3426int						/* error code */
3427xfs_bmap_add_attrfork(
3428	xfs_inode_t		*ip,		/* incore inode pointer */
3429	int			size,		/* space new attribute needs */
3430	int			rsvd)		/* xact may use reserved blks */
3431{
3432	xfs_fsblock_t		firstblock;	/* 1st block/ag allocated */
3433	xfs_bmap_free_t		flist;		/* freed extent records */
3434	xfs_mount_t		*mp;		/* mount structure */
3435	xfs_trans_t		*tp;		/* transaction pointer */
3436	int			blks;		/* space reservation */
3437	int			version = 1;	/* superblock attr version */
3438	int			committed;	/* xaction was committed */
3439	int			logflags;	/* logging flags */
3440	int			error;		/* error return value */
3441
3442	ASSERT(XFS_IFORK_Q(ip) == 0);
3443
3444	mp = ip->i_mount;
3445	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3446	tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3447	blks = XFS_ADDAFORK_SPACE_RES(mp);
3448	if (rsvd)
3449		tp->t_flags |= XFS_TRANS_RESERVE;
3450	if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3451			XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3452		goto error0;
3453	xfs_ilock(ip, XFS_ILOCK_EXCL);
3454	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3455			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3456			XFS_QMOPT_RES_REGBLKS);
3457	if (error) {
3458		xfs_iunlock(ip, XFS_ILOCK_EXCL);
3459		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3460		return error;
3461	}
3462	if (XFS_IFORK_Q(ip))
3463		goto error1;
3464	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3465		/*
3466		 * For inodes coming from pre-6.2 filesystems.
3467		 */
3468		ASSERT(ip->i_d.di_aformat == 0);
3469		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3470	}
3471	ASSERT(ip->i_d.di_anextents == 0);
3472
3473	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3474	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3475
3476	switch (ip->i_d.di_format) {
3477	case XFS_DINODE_FMT_DEV:
3478		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3479		break;
3480	case XFS_DINODE_FMT_UUID:
3481		ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3482		break;
3483	case XFS_DINODE_FMT_LOCAL:
3484	case XFS_DINODE_FMT_EXTENTS:
3485	case XFS_DINODE_FMT_BTREE:
3486		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3487		if (!ip->i_d.di_forkoff)
3488			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3489		else if (mp->m_flags & XFS_MOUNT_ATTR2)
3490			version = 2;
3491		break;
3492	default:
3493		ASSERT(0);
3494		error = XFS_ERROR(EINVAL);
3495		goto error1;
3496	}
3497
3498	ASSERT(ip->i_afp == NULL);
3499	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3500	ip->i_afp->if_flags = XFS_IFEXTENTS;
3501	logflags = 0;
3502	xfs_bmap_init(&flist, &firstblock);
3503	switch (ip->i_d.di_format) {
3504	case XFS_DINODE_FMT_LOCAL:
3505		error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3506			&logflags);
3507		break;
3508	case XFS_DINODE_FMT_EXTENTS:
3509		error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3510			&flist, &logflags);
3511		break;
3512	case XFS_DINODE_FMT_BTREE:
3513		error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3514			&logflags);
3515		break;
3516	default:
3517		error = 0;
3518		break;
3519	}
3520	if (logflags)
3521		xfs_trans_log_inode(tp, ip, logflags);
3522	if (error)
3523		goto error2;
3524	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3525	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3526		__int64_t sbfields = 0;
3527
3528		spin_lock(&mp->m_sb_lock);
3529		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3530			xfs_sb_version_addattr(&mp->m_sb);
3531			sbfields |= XFS_SB_VERSIONNUM;
3532		}
3533		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3534			xfs_sb_version_addattr2(&mp->m_sb);
3535			sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3536		}
3537		if (sbfields) {
3538			spin_unlock(&mp->m_sb_lock);
3539			xfs_mod_sb(tp, sbfields);
3540		} else
3541			spin_unlock(&mp->m_sb_lock);
3542	}
3543
3544	error = xfs_bmap_finish(&tp, &flist, &committed);
3545	if (error)
3546		goto error2;
3547	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3548error2:
3549	xfs_bmap_cancel(&flist);
3550error1:
3551	xfs_iunlock(ip, XFS_ILOCK_EXCL);
3552error0:
3553	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3554	return error;
3555}
3556
3557/*
3558 * Add the extent to the list of extents to be free at transaction end.
3559 * The list is maintained sorted (by block number).
3560 */
3561/* ARGSUSED */
3562void
3563xfs_bmap_add_free(
3564	xfs_fsblock_t		bno,		/* fs block number of extent */
3565	xfs_filblks_t		len,		/* length of extent */
3566	xfs_bmap_free_t		*flist,		/* list of extents */
3567	xfs_mount_t		*mp)		/* mount point structure */
3568{
3569	xfs_bmap_free_item_t	*cur;		/* current (next) element */
3570	xfs_bmap_free_item_t	*new;		/* new element */
3571	xfs_bmap_free_item_t	*prev;		/* previous element */
3572#ifdef DEBUG
3573	xfs_agnumber_t		agno;
3574	xfs_agblock_t		agbno;
3575
3576	ASSERT(bno != NULLFSBLOCK);
3577	ASSERT(len > 0);
3578	ASSERT(len <= MAXEXTLEN);
3579	ASSERT(!isnullstartblock(bno));
3580	agno = XFS_FSB_TO_AGNO(mp, bno);
3581	agbno = XFS_FSB_TO_AGBNO(mp, bno);
3582	ASSERT(agno < mp->m_sb.sb_agcount);
3583	ASSERT(agbno < mp->m_sb.sb_agblocks);
3584	ASSERT(len < mp->m_sb.sb_agblocks);
3585	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3586#endif
3587	ASSERT(xfs_bmap_free_item_zone != NULL);
3588	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3589	new->xbfi_startblock = bno;
3590	new->xbfi_blockcount = (xfs_extlen_t)len;
3591	for (prev = NULL, cur = flist->xbf_first;
3592	     cur != NULL;
3593	     prev = cur, cur = cur->xbfi_next) {
3594		if (cur->xbfi_startblock >= bno)
3595			break;
3596	}
3597	if (prev)
3598		prev->xbfi_next = new;
3599	else
3600		flist->xbf_first = new;
3601	new->xbfi_next = cur;
3602	flist->xbf_count++;
3603}
3604
3605/*
3606 * Compute and fill in the value of the maximum depth of a bmap btree
3607 * in this filesystem.  Done once, during mount.
3608 */
3609void
3610xfs_bmap_compute_maxlevels(
3611	xfs_mount_t	*mp,		/* file system mount structure */
3612	int		whichfork)	/* data or attr fork */
3613{
3614	int		level;		/* btree level */
3615	uint		maxblocks;	/* max blocks at this level */
3616	uint		maxleafents;	/* max leaf entries possible */
3617	int		maxrootrecs;	/* max records in root block */
3618	int		minleafrecs;	/* min records in leaf block */
3619	int		minnoderecs;	/* min records in node block */
3620	int		sz;		/* root block size */
3621
3622	/*
3623	 * The maximum number of extents in a file, hence the maximum
3624	 * number of leaf entries, is controlled by the type of di_nextents
3625	 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3626	 * (a signed 16-bit number, xfs_aextnum_t).
3627	 *
3628	 * Note that we can no longer assume that if we are in ATTR1 that
3629	 * the fork offset of all the inodes will be
3630	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3631	 * with ATTR2 and then mounted back with ATTR1, keeping the
3632	 * di_forkoff's fixed but probably at various positions. Therefore,
3633	 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3634	 * of a minimum size available.
3635	 */
3636	if (whichfork == XFS_DATA_FORK) {
3637		maxleafents = MAXEXTNUM;
3638		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3639	} else {
3640		maxleafents = MAXAEXTNUM;
3641		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3642	}
3643	maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3644	minleafrecs = mp->m_bmap_dmnr[0];
3645	minnoderecs = mp->m_bmap_dmnr[1];
3646	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3647	for (level = 1; maxblocks > 1; level++) {
3648		if (maxblocks <= maxrootrecs)
3649			maxblocks = 1;
3650		else
3651			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3652	}
3653	mp->m_bm_maxlevels[whichfork] = level;
3654}
3655
3656/*
3657 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3658 * caller.  Frees all the extents that need freeing, which must be done
3659 * last due to locking considerations.  We never free any extents in
3660 * the first transaction.
3661 *
3662 * Return 1 if the given transaction was committed and a new one
3663 * started, and 0 otherwise in the committed parameter.
3664 */
3665int						/* error */
3666xfs_bmap_finish(
3667	xfs_trans_t		**tp,		/* transaction pointer addr */
3668	xfs_bmap_free_t		*flist,		/* i/o: list extents to free */
3669	int			*committed)	/* xact committed or not */
3670{
3671	xfs_efd_log_item_t	*efd;		/* extent free data */
3672	xfs_efi_log_item_t	*efi;		/* extent free intention */
3673	int			error;		/* error return value */
3674	xfs_bmap_free_item_t	*free;		/* free extent item */
3675	unsigned int		logres;		/* new log reservation */
3676	unsigned int		logcount;	/* new log count */
3677	xfs_mount_t		*mp;		/* filesystem mount structure */
3678	xfs_bmap_free_item_t	*next;		/* next item on free list */
3679	xfs_trans_t		*ntp;		/* new transaction pointer */
3680
3681	ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3682	if (flist->xbf_count == 0) {
3683		*committed = 0;
3684		return 0;
3685	}
3686	ntp = *tp;
3687	efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3688	for (free = flist->xbf_first; free; free = free->xbfi_next)
3689		xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3690			free->xbfi_blockcount);
3691	logres = ntp->t_log_res;
3692	logcount = ntp->t_log_count;
3693	ntp = xfs_trans_dup(*tp);
3694	error = xfs_trans_commit(*tp, 0);
3695	*tp = ntp;
3696	*committed = 1;
3697	/*
3698	 * We have a new transaction, so we should return committed=1,
3699	 * even though we're returning an error.
3700	 */
3701	if (error)
3702		return error;
3703
3704	/*
3705	 * transaction commit worked ok so we can drop the extra ticket
3706	 * reference that we gained in xfs_trans_dup()
3707	 */
3708	xfs_log_ticket_put(ntp->t_ticket);
3709
3710	if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3711			logcount)))
3712		return error;
3713	efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3714	for (free = flist->xbf_first; free != NULL; free = next) {
3715		next = free->xbfi_next;
3716		if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3717				free->xbfi_blockcount))) {
3718			/*
3719			 * The bmap free list will be cleaned up at a
3720			 * higher level.  The EFI will be canceled when
3721			 * this transaction is aborted.
3722			 * Need to force shutdown here to make sure it
3723			 * happens, since this transaction may not be
3724			 * dirty yet.
3725			 */
3726			mp = ntp->t_mountp;
3727			if (!XFS_FORCED_SHUTDOWN(mp))
3728				xfs_force_shutdown(mp,
3729						   (error == EFSCORRUPTED) ?
3730						   SHUTDOWN_CORRUPT_INCORE :
3731						   SHUTDOWN_META_IO_ERROR);
3732			return error;
3733		}
3734		xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3735			free->xbfi_blockcount);
3736		xfs_bmap_del_free(flist, NULL, free);
3737	}
3738	return 0;
3739}
3740
3741/*
3742 * Free up any items left in the list.
3743 */
3744void
3745xfs_bmap_cancel(
3746	xfs_bmap_free_t		*flist)	/* list of bmap_free_items */
3747{
3748	xfs_bmap_free_item_t	*free;	/* free list item */
3749	xfs_bmap_free_item_t	*next;
3750
3751	if (flist->xbf_count == 0)
3752		return;
3753	ASSERT(flist->xbf_first != NULL);
3754	for (free = flist->xbf_first; free; free = next) {
3755		next = free->xbfi_next;
3756		xfs_bmap_del_free(flist, NULL, free);
3757	}
3758	ASSERT(flist->xbf_count == 0);
3759}
3760
3761/*
3762 * Returns the file-relative block number of the first unused block(s)
3763 * in the file with at least "len" logically contiguous blocks free.
3764 * This is the lowest-address hole if the file has holes, else the first block
3765 * past the end of file.
3766 * Return 0 if the file is currently local (in-inode).
3767 */
3768int						/* error */
3769xfs_bmap_first_unused(
3770	xfs_trans_t	*tp,			/* transaction pointer */
3771	xfs_inode_t	*ip,			/* incore inode */
3772	xfs_extlen_t	len,			/* size of hole to find */
3773	xfs_fileoff_t	*first_unused,		/* unused block */
3774	int		whichfork)		/* data or attr fork */
3775{
3776	int		error;			/* error return value */
3777	int		idx;			/* extent record index */
3778	xfs_ifork_t	*ifp;			/* inode fork pointer */
3779	xfs_fileoff_t	lastaddr;		/* last block number seen */
3780	xfs_fileoff_t	lowest;			/* lowest useful block */
3781	xfs_fileoff_t	max;			/* starting useful block */
3782	xfs_fileoff_t	off;			/* offset for this block */
3783	xfs_extnum_t	nextents;		/* number of extent entries */
3784
3785	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3786	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3787	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3788	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3789		*first_unused = 0;
3790		return 0;
3791	}
3792	ifp = XFS_IFORK_PTR(ip, whichfork);
3793	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3794	    (error = xfs_iread_extents(tp, ip, whichfork)))
3795		return error;
3796	lowest = *first_unused;
3797	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3798	for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3799		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3800		off = xfs_bmbt_get_startoff(ep);
3801		/*
3802		 * See if the hole before this extent will work.
3803		 */
3804		if (off >= lowest + len && off - max >= len) {
3805			*first_unused = max;
3806			return 0;
3807		}
3808		lastaddr = off + xfs_bmbt_get_blockcount(ep);
3809		max = XFS_FILEOFF_MAX(lastaddr, lowest);
3810	}
3811	*first_unused = max;
3812	return 0;
3813}
3814
3815/*
3816 * Returns the file-relative block number of the last block + 1 before
3817 * last_block (input value) in the file.
3818 * This is not based on i_size, it is based on the extent records.
3819 * Returns 0 for local files, as they do not have extent records.
3820 */
3821int						/* error */
3822xfs_bmap_last_before(
3823	xfs_trans_t	*tp,			/* transaction pointer */
3824	xfs_inode_t	*ip,			/* incore inode */
3825	xfs_fileoff_t	*last_block,		/* last block */
3826	int		whichfork)		/* data or attr fork */
3827{
3828	xfs_fileoff_t	bno;			/* input file offset */
3829	int		eof;			/* hit end of file */
3830	xfs_bmbt_rec_host_t *ep;		/* pointer to last extent */
3831	int		error;			/* error return value */
3832	xfs_bmbt_irec_t	got;			/* current extent value */
3833	xfs_ifork_t	*ifp;			/* inode fork pointer */
3834	xfs_extnum_t	lastx;			/* last extent used */
3835	xfs_bmbt_irec_t	prev;			/* previous extent value */
3836
3837	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3838	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3839	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3840	       return XFS_ERROR(EIO);
3841	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3842		*last_block = 0;
3843		return 0;
3844	}
3845	ifp = XFS_IFORK_PTR(ip, whichfork);
3846	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3847	    (error = xfs_iread_extents(tp, ip, whichfork)))
3848		return error;
3849	bno = *last_block - 1;
3850	ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3851		&prev);
3852	if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3853		if (prev.br_startoff == NULLFILEOFF)
3854			*last_block = 0;
3855		else
3856			*last_block = prev.br_startoff + prev.br_blockcount;
3857	}
3858	/*
3859	 * Otherwise *last_block is already the right answer.
3860	 */
3861	return 0;
3862}
3863
3864STATIC int
3865xfs_bmap_last_extent(
3866	struct xfs_trans	*tp,
3867	struct xfs_inode	*ip,
3868	int			whichfork,
3869	struct xfs_bmbt_irec	*rec,
3870	int			*is_empty)
3871{
3872	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
3873	int			error;
3874	int			nextents;
3875
3876	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3877		error = xfs_iread_extents(tp, ip, whichfork);
3878		if (error)
3879			return error;
3880	}
3881
3882	nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3883	if (nextents == 0) {
3884		*is_empty = 1;
3885		return 0;
3886	}
3887
3888	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3889	*is_empty = 0;
3890	return 0;
3891}
3892
3893/*
3894 * Check the last inode extent to determine whether this allocation will result
3895 * in blocks being allocated at the end of the file. When we allocate new data
3896 * blocks at the end of the file which do not start at the previous data block,
3897 * we will try to align the new blocks at stripe unit boundaries.
3898 *
3899 * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3900 * at, or past the EOF.
3901 */
3902STATIC int
3903xfs_bmap_isaeof(
3904	struct xfs_bmalloca	*bma,
3905	int			whichfork)
3906{
3907	struct xfs_bmbt_irec	rec;
3908	int			is_empty;
3909	int			error;
3910
3911	bma->aeof = 0;
3912	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3913				     &is_empty);
3914	if (error || is_empty)
3915		return error;
3916
3917	/*
3918	 * Check if we are allocation or past the last extent, or at least into
3919	 * the last delayed allocated extent.
3920	 */
3921	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3922		(bma->offset >= rec.br_startoff &&
3923		 isnullstartblock(rec.br_startblock));
3924	return 0;
3925}
3926
3927/*
3928 * Check if the endoff is outside the last extent. If so the caller will grow
3929 * the allocation to a stripe unit boundary.  All offsets are considered outside
3930 * the end of file for an empty fork, so 1 is returned in *eof in that case.
3931 */
3932int
3933xfs_bmap_eof(
3934	struct xfs_inode	*ip,
3935	xfs_fileoff_t		endoff,
3936	int			whichfork,
3937	int			*eof)
3938{
3939	struct xfs_bmbt_irec	rec;
3940	int			error;
3941
3942	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3943	if (error || *eof)
3944		return error;
3945
3946	*eof = endoff >= rec.br_startoff + rec.br_blockcount;
3947	return 0;
3948}
3949
3950/*
3951 * Returns the file-relative block number of the first block past eof in
3952 * the file.  This is not based on i_size, it is based on the extent records.
3953 * Returns 0 for local files, as they do not have extent records.
3954 */
3955int
3956xfs_bmap_last_offset(
3957	struct xfs_trans	*tp,
3958	struct xfs_inode	*ip,
3959	xfs_fileoff_t		*last_block,
3960	int			whichfork)
3961{
3962	struct xfs_bmbt_irec	rec;
3963	int			is_empty;
3964	int			error;
3965
3966	*last_block = 0;
3967
3968	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3969		return 0;
3970
3971	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3972	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3973	       return XFS_ERROR(EIO);
3974
3975	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3976	if (error || is_empty)
3977		return error;
3978
3979	*last_block = rec.br_startoff + rec.br_blockcount;
3980	return 0;
3981}
3982
3983/*
3984 * Returns whether the selected fork of the inode has exactly one
3985 * block or not.  For the data fork we check this matches di_size,
3986 * implying the file's range is 0..bsize-1.
3987 */
3988int					/* 1=>1 block, 0=>otherwise */
3989xfs_bmap_one_block(
3990	xfs_inode_t	*ip,		/* incore inode */
3991	int		whichfork)	/* data or attr fork */
3992{
3993	xfs_bmbt_rec_host_t *ep;	/* ptr to fork's extent */
3994	xfs_ifork_t	*ifp;		/* inode fork pointer */
3995	int		rval;		/* return value */
3996	xfs_bmbt_irec_t	s;		/* internal version of extent */
3997
3998#ifndef DEBUG
3999	if (whichfork == XFS_DATA_FORK)
4000		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
4001#endif	/* !DEBUG */
4002	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4003		return 0;
4004	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4005		return 0;
4006	ifp = XFS_IFORK_PTR(ip, whichfork);
4007	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4008	ep = xfs_iext_get_ext(ifp, 0);
4009	xfs_bmbt_get_all(ep, &s);
4010	rval = s.br_startoff == 0 && s.br_blockcount == 1;
4011	if (rval && whichfork == XFS_DATA_FORK)
4012		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
4013	return rval;
4014}
4015
4016STATIC int
4017xfs_bmap_sanity_check(
4018	struct xfs_mount	*mp,
4019	struct xfs_buf		*bp,
4020	int			level)
4021{
4022	struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4023
4024	if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4025	    be16_to_cpu(block->bb_level) != level ||
4026	    be16_to_cpu(block->bb_numrecs) == 0 ||
4027	    be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4028		return 0;
4029	return 1;
4030}
4031
4032/*
4033 * Read in the extents to if_extents.
4034 * All inode fields are set up by caller, we just traverse the btree
4035 * and copy the records in. If the file system cannot contain unwritten
4036 * extents, the records are checked for no "state" flags.
4037 */
4038int					/* error */
4039xfs_bmap_read_extents(
4040	xfs_trans_t		*tp,	/* transaction pointer */
4041	xfs_inode_t		*ip,	/* incore inode */
4042	int			whichfork) /* data or attr fork */
4043{
4044	struct xfs_btree_block	*block;	/* current btree block */
4045	xfs_fsblock_t		bno;	/* block # of "block" */
4046	xfs_buf_t		*bp;	/* buffer for "block" */
4047	int			error;	/* error return value */
4048	xfs_exntfmt_t		exntf;	/* XFS_EXTFMT_NOSTATE, if checking */
4049	xfs_extnum_t		i, j;	/* index into the extents list */
4050	xfs_ifork_t		*ifp;	/* fork structure */
4051	int			level;	/* btree level, for checking */
4052	xfs_mount_t		*mp;	/* file system mount structure */
4053	__be64			*pp;	/* pointer to block address */
4054	/* REFERENCED */
4055	xfs_extnum_t		room;	/* number of entries there's room for */
4056
4057	bno = NULLFSBLOCK;
4058	mp = ip->i_mount;
4059	ifp = XFS_IFORK_PTR(ip, whichfork);
4060	exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4061					XFS_EXTFMT_INODE(ip);
4062	block = ifp->if_broot;
4063	/*
4064	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4065	 */
4066	level = be16_to_cpu(block->bb_level);
4067	ASSERT(level > 0);
4068	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4069	bno = be64_to_cpu(*pp);
4070	ASSERT(bno != NULLDFSBNO);
4071	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4072	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4073	/*
4074	 * Go down the tree until leaf level is reached, following the first
4075	 * pointer (leftmost) at each level.
4076	 */
4077	while (level-- > 0) {
4078		if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4079				XFS_BMAP_BTREE_REF)))
4080			return error;
4081		block = XFS_BUF_TO_BLOCK(bp);
4082		XFS_WANT_CORRUPTED_GOTO(
4083			xfs_bmap_sanity_check(mp, bp, level),
4084			error0);
4085		if (level == 0)
4086			break;
4087		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4088		bno = be64_to_cpu(*pp);
4089		XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4090		xfs_trans_brelse(tp, bp);
4091	}
4092	/*
4093	 * Here with bp and block set to the leftmost leaf node in the tree.
4094	 */
4095	room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4096	i = 0;
4097	/*
4098	 * Loop over all leaf nodes.  Copy information to the extent records.
4099	 */
4100	for (;;) {
4101		xfs_bmbt_rec_t	*frp;
4102		xfs_fsblock_t	nextbno;
4103		xfs_extnum_t	num_recs;
4104		xfs_extnum_t	start;
4105
4106		num_recs = xfs_btree_get_numrecs(block);
4107		if (unlikely(i + num_recs > room)) {
4108			ASSERT(i + num_recs <= room);
4109			xfs_warn(ip->i_mount,
4110				"corrupt dinode %Lu, (btree extents).",
4111				(unsigned long long) ip->i_ino);
4112			XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4113				XFS_ERRLEVEL_LOW, ip->i_mount, block);
4114			goto error0;
4115		}
4116		XFS_WANT_CORRUPTED_GOTO(
4117			xfs_bmap_sanity_check(mp, bp, 0),
4118			error0);
4119		/*
4120		 * Read-ahead the next leaf block, if any.
4121		 */
4122		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4123		if (nextbno != NULLFSBLOCK)
4124			xfs_btree_reada_bufl(mp, nextbno, 1);
4125		/*
4126		 * Copy records into the extent records.
4127		 */
4128		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4129		start = i;
4130		for (j = 0; j < num_recs; j++, i++, frp++) {
4131			xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4132			trp->l0 = be64_to_cpu(frp->l0);
4133			trp->l1 = be64_to_cpu(frp->l1);
4134		}
4135		if (exntf == XFS_EXTFMT_NOSTATE) {
4136			/*
4137			 * Check all attribute bmap btree records and
4138			 * any "older" data bmap btree records for a
4139			 * set bit in the "extent flag" position.
4140			 */
4141			if (unlikely(xfs_check_nostate_extents(ifp,
4142					start, num_recs))) {
4143				XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4144						 XFS_ERRLEVEL_LOW,
4145						 ip->i_mount);
4146				goto error0;
4147			}
4148		}
4149		xfs_trans_brelse(tp, bp);
4150		bno = nextbno;
4151		/*
4152		 * If we've reached the end, stop.
4153		 */
4154		if (bno == NULLFSBLOCK)
4155			break;
4156		if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4157				XFS_BMAP_BTREE_REF)))
4158			return error;
4159		block = XFS_BUF_TO_BLOCK(bp);
4160	}
4161	ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4162	ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4163	XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4164	return 0;
4165error0:
4166	xfs_trans_brelse(tp, bp);
4167	return XFS_ERROR(EFSCORRUPTED);
4168}
4169
4170#ifdef DEBUG
4171/*
4172 * Add bmap trace insert entries for all the contents of the extent records.
4173 */
4174void
4175xfs_bmap_trace_exlist(
4176	xfs_inode_t	*ip,		/* incore inode pointer */
4177	xfs_extnum_t	cnt,		/* count of entries in the list */
4178	int		whichfork,	/* data or attr fork */
4179	unsigned long	caller_ip)
4180{
4181	xfs_extnum_t	idx;		/* extent record index */
4182	xfs_ifork_t	*ifp;		/* inode fork pointer */
4183	int		state = 0;
4184
4185	if (whichfork == XFS_ATTR_FORK)
4186		state |= BMAP_ATTRFORK;
4187
4188	ifp = XFS_IFORK_PTR(ip, whichfork);
4189	ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4190	for (idx = 0; idx < cnt; idx++)
4191		trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4192}
4193
4194/*
4195 * Validate that the bmbt_irecs being returned from bmapi are valid
4196 * given the callers original parameters.  Specifically check the
4197 * ranges of the returned irecs to ensure that they only extent beyond
4198 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4199 */
4200STATIC void
4201xfs_bmap_validate_ret(
4202	xfs_fileoff_t		bno,
4203	xfs_filblks_t		len,
4204	int			flags,
4205	xfs_bmbt_irec_t		*mval,
4206	int			nmap,
4207	int			ret_nmap)
4208{
4209	int			i;		/* index to map values */
4210
4211	ASSERT(ret_nmap <= nmap);
4212
4213	for (i = 0; i < ret_nmap; i++) {
4214		ASSERT(mval[i].br_blockcount > 0);
4215		if (!(flags & XFS_BMAPI_ENTIRE)) {
4216			ASSERT(mval[i].br_startoff >= bno);
4217			ASSERT(mval[i].br_blockcount <= len);
4218			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4219			       bno + len);
4220		} else {
4221			ASSERT(mval[i].br_startoff < bno + len);
4222			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4223			       bno);
4224		}
4225		ASSERT(i == 0 ||
4226		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4227		       mval[i].br_startoff);
4228		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4229		       mval[i].br_startblock != HOLESTARTBLOCK);
4230		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4231		       mval[i].br_state == XFS_EXT_UNWRITTEN);
4232	}
4233}
4234#endif /* DEBUG */
4235
4236
4237/*
4238 * Trim the returned map to the required bounds
4239 */
4240STATIC void
4241xfs_bmapi_trim_map(
4242	struct xfs_bmbt_irec	*mval,
4243	struct xfs_bmbt_irec	*got,
4244	xfs_fileoff_t		*bno,
4245	xfs_filblks_t		len,
4246	xfs_fileoff_t		obno,
4247	xfs_fileoff_t		end,
4248	int			n,
4249	int			flags)
4250{
4251	if ((flags & XFS_BMAPI_ENTIRE) ||
4252	    got->br_startoff + got->br_blockcount <= obno) {
4253		*mval = *got;
4254		if (isnullstartblock(got->br_startblock))
4255			mval->br_startblock = DELAYSTARTBLOCK;
4256		return;
4257	}
4258
4259	if (obno > *bno)
4260		*bno = obno;
4261	ASSERT((*bno >= obno) || (n == 0));
4262	ASSERT(*bno < end);
4263	mval->br_startoff = *bno;
4264	if (isnullstartblock(got->br_startblock))
4265		mval->br_startblock = DELAYSTARTBLOCK;
4266	else
4267		mval->br_startblock = got->br_startblock +
4268					(*bno - got->br_startoff);
4269	/*
4270	 * Return the minimum of what we got and what we asked for for
4271	 * the length.  We can use the len variable here because it is
4272	 * modified below and we could have been there before coming
4273	 * here if the first part of the allocation didn't overlap what
4274	 * was asked for.
4275	 */
4276	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4277			got->br_blockcount - (*bno - got->br_startoff));
4278	mval->br_state = got->br_state;
4279	ASSERT(mval->br_blockcount <= len);
4280	return;
4281}
4282
4283/*
4284 * Update and validate the extent map to return
4285 */
4286STATIC void
4287xfs_bmapi_update_map(
4288	struct xfs_bmbt_irec	**map,
4289	xfs_fileoff_t		*bno,
4290	xfs_filblks_t		*len,
4291	xfs_fileoff_t		obno,
4292	xfs_fileoff_t		end,
4293	int			*n,
4294	int			flags)
4295{
4296	xfs_bmbt_irec_t	*mval = *map;
4297
4298	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4299	       ((mval->br_startoff + mval->br_blockcount) <= end));
4300	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4301	       (mval->br_startoff < obno));
4302
4303	*bno = mval->br_startoff + mval->br_blockcount;
4304	*len = end - *bno;
4305	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4306		/* update previous map with new information */
4307		ASSERT(mval->br_startblock == mval[-1].br_startblock);
4308		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4309		ASSERT(mval->br_state == mval[-1].br_state);
4310		mval[-1].br_blockcount = mval->br_blockcount;
4311		mval[-1].br_state = mval->br_state;
4312	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4313		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
4314		   mval[-1].br_startblock != HOLESTARTBLOCK &&
4315		   mval->br_startblock == mval[-1].br_startblock +
4316					  mval[-1].br_blockcount &&
4317		   ((flags & XFS_BMAPI_IGSTATE) ||
4318			mval[-1].br_state == mval->br_state)) {
4319		ASSERT(mval->br_startoff ==
4320		       mval[-1].br_startoff + mval[-1].br_blockcount);
4321		mval[-1].br_blockcount += mval->br_blockcount;
4322	} else if (*n > 0 &&
4323		   mval->br_startblock == DELAYSTARTBLOCK &&
4324		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
4325		   mval->br_startoff ==
4326		   mval[-1].br_startoff + mval[-1].br_blockcount) {
4327		mval[-1].br_blockcount += mval->br_blockcount;
4328		mval[-1].br_state = mval->br_state;
4329	} else if (!((*n == 0) &&
4330		     ((mval->br_startoff + mval->br_blockcount) <=
4331		      obno))) {
4332		mval++;
4333		(*n)++;
4334	}
4335	*map = mval;
4336}
4337
4338/*
4339 * Map file blocks to filesystem blocks without allocation.
4340 */
4341int
4342xfs_bmapi_read(
4343	struct xfs_inode	*ip,
4344	xfs_fileoff_t		bno,
4345	xfs_filblks_t		len,
4346	struct xfs_bmbt_irec	*mval,
4347	int			*nmap,
4348	int			flags)
4349{
4350	struct xfs_mount	*mp = ip->i_mount;
4351	struct xfs_ifork	*ifp;
4352	struct xfs_bmbt_irec	got;
4353	struct xfs_bmbt_irec	prev;
4354	xfs_fileoff_t		obno;
4355	xfs_fileoff_t		end;
4356	xfs_extnum_t		lastx;
4357	int			error;
4358	int			eof;
4359	int			n = 0;
4360	int			whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4361						XFS_ATTR_FORK : XFS_DATA_FORK;
4362
4363	ASSERT(*nmap >= 1);
4364	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4365			   XFS_BMAPI_IGSTATE)));
 
4366
4367	if (unlikely(XFS_TEST_ERROR(
4368	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4369	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4370	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4371		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4372		return XFS_ERROR(EFSCORRUPTED);
4373	}
4374
4375	if (XFS_FORCED_SHUTDOWN(mp))
4376		return XFS_ERROR(EIO);
4377
4378	XFS_STATS_INC(xs_blk_mapr);
4379
4380	ifp = XFS_IFORK_PTR(ip, whichfork);
4381
4382	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4383		error = xfs_iread_extents(NULL, ip, whichfork);
4384		if (error)
4385			return error;
4386	}
4387
4388	xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4389	end = bno + len;
4390	obno = bno;
4391
4392	while (bno < end && n < *nmap) {
4393		/* Reading past eof, act as though there's a hole up to end. */
4394		if (eof)
4395			got.br_startoff = end;
4396		if (got.br_startoff > bno) {
4397			/* Reading in a hole.  */
4398			mval->br_startoff = bno;
4399			mval->br_startblock = HOLESTARTBLOCK;
4400			mval->br_blockcount =
4401				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4402			mval->br_state = XFS_EXT_NORM;
4403			bno += mval->br_blockcount;
4404			len -= mval->br_blockcount;
4405			mval++;
4406			n++;
4407			continue;
4408		}
4409
4410		/* set up the extent map to return. */
4411		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4412		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4413
4414		/* If we're done, stop now. */
4415		if (bno >= end || n >= *nmap)
4416			break;
4417
4418		/* Else go on to the next record. */
4419		if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4420			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4421		else
4422			eof = 1;
4423	}
4424	*nmap = n;
4425	return 0;
4426}
4427
4428STATIC int
4429xfs_bmapi_reserve_delalloc(
4430	struct xfs_inode	*ip,
4431	xfs_fileoff_t		aoff,
4432	xfs_filblks_t		len,
4433	struct xfs_bmbt_irec	*got,
4434	struct xfs_bmbt_irec	*prev,
4435	xfs_extnum_t		*lastx,
4436	int			eof)
4437{
4438	struct xfs_mount	*mp = ip->i_mount;
4439	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4440	xfs_extlen_t		alen;
4441	xfs_extlen_t		indlen;
4442	char			rt = XFS_IS_REALTIME_INODE(ip);
4443	xfs_extlen_t		extsz;
4444	int			error;
4445
4446	alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4447	if (!eof)
4448		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4449
4450	/* Figure out the extent size, adjust alen */
4451	extsz = xfs_get_extsz_hint(ip);
4452	if (extsz) {
4453		/*
4454		 * Make sure we don't exceed a single extent length when we
4455		 * align the extent by reducing length we are going to
4456		 * allocate by the maximum amount extent size aligment may
4457		 * require.
4458		 */
4459		alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4460		error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4461					       1, 0, &aoff, &alen);
4462		ASSERT(!error);
4463	}
4464
4465	if (rt)
4466		extsz = alen / mp->m_sb.sb_rextsize;
4467
4468	/*
4469	 * Make a transaction-less quota reservation for delayed allocation
4470	 * blocks.  This number gets adjusted later.  We return if we haven't
4471	 * allocated blocks already inside this loop.
4472	 */
4473	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4474			rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4475	if (error)
4476		return error;
4477
4478	/*
4479	 * Split changing sb for alen and indlen since they could be coming
4480	 * from different places.
4481	 */
4482	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4483	ASSERT(indlen > 0);
4484
4485	if (rt) {
4486		error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4487					  -((int64_t)extsz), 0);
4488	} else {
4489		error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4490						 -((int64_t)alen), 0);
4491	}
4492
4493	if (error)
4494		goto out_unreserve_quota;
4495
4496	error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4497					 -((int64_t)indlen), 0);
4498	if (error)
4499		goto out_unreserve_blocks;
4500
4501
4502	ip->i_delayed_blks += alen;
4503
4504	got->br_startoff = aoff;
4505	got->br_startblock = nullstartblock(indlen);
4506	got->br_blockcount = alen;
4507	got->br_state = XFS_EXT_NORM;
4508	xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4509
4510	/*
4511	 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4512	 * might have merged it into one of the neighbouring ones.
4513	 */
4514	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4515
4516	ASSERT(got->br_startoff <= aoff);
4517	ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4518	ASSERT(isnullstartblock(got->br_startblock));
4519	ASSERT(got->br_state == XFS_EXT_NORM);
4520	return 0;
4521
4522out_unreserve_blocks:
4523	if (rt)
4524		xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4525	else
4526		xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4527out_unreserve_quota:
4528	if (XFS_IS_QUOTA_ON(mp))
4529		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4530				XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4531	return error;
4532}
4533
4534/*
4535 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4536 */
4537int
4538xfs_bmapi_delay(
4539	struct xfs_inode	*ip,	/* incore inode */
4540	xfs_fileoff_t		bno,	/* starting file offs. mapped */
4541	xfs_filblks_t		len,	/* length to map in file */
4542	struct xfs_bmbt_irec	*mval,	/* output: map values */
4543	int			*nmap,	/* i/o: mval size/count */
4544	int			flags)	/* XFS_BMAPI_... */
4545{
4546	struct xfs_mount	*mp = ip->i_mount;
4547	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4548	struct xfs_bmbt_irec	got;	/* current file extent record */
4549	struct xfs_bmbt_irec	prev;	/* previous file extent record */
4550	xfs_fileoff_t		obno;	/* old block number (offset) */
4551	xfs_fileoff_t		end;	/* end of mapped file region */
4552	xfs_extnum_t		lastx;	/* last useful extent number */
4553	int			eof;	/* we've hit the end of extents */
4554	int			n = 0;	/* current extent index */
4555	int			error = 0;
4556
4557	ASSERT(*nmap >= 1);
4558	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4559	ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
 
4560
4561	if (unlikely(XFS_TEST_ERROR(
4562	    (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4563	     XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4564	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4565		XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4566		return XFS_ERROR(EFSCORRUPTED);
4567	}
4568
4569	if (XFS_FORCED_SHUTDOWN(mp))
4570		return XFS_ERROR(EIO);
4571
4572	XFS_STATS_INC(xs_blk_mapw);
4573
4574	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4575		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4576		if (error)
4577			return error;
4578	}
4579
4580	xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4581	end = bno + len;
4582	obno = bno;
4583
4584	while (bno < end && n < *nmap) {
4585		if (eof || got.br_startoff > bno) {
4586			error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4587							   &prev, &lastx, eof);
4588			if (error) {
4589				if (n == 0) {
4590					*nmap = 0;
4591					return error;
4592				}
4593				break;
4594			}
4595		}
4596
4597		/* set up the extent map to return. */
4598		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4599		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4600
4601		/* If we're done, stop now. */
4602		if (bno >= end || n >= *nmap)
4603			break;
4604
4605		/* Else go on to the next record. */
4606		prev = got;
4607		if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4608			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4609		else
4610			eof = 1;
4611	}
4612
4613	*nmap = n;
4614	return 0;
4615}
4616
4617
4618STATIC int
4619xfs_bmapi_allocate(
4620	struct xfs_bmalloca	*bma,
4621	int			flags)
4622{
4623	struct xfs_mount	*mp = bma->ip->i_mount;
4624	int			whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4625						XFS_ATTR_FORK : XFS_DATA_FORK;
4626	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4627	int			tmp_logflags = 0;
4628	int			error;
4629	int			rt;
4630
4631	ASSERT(bma->length > 0);
4632
4633	rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4634
4635	/*
4636	 * For the wasdelay case, we could also just allocate the stuff asked
4637	 * for in this bmap call but that wouldn't be as good.
4638	 */
4639	if (bma->wasdel) {
4640		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4641		bma->offset = bma->got.br_startoff;
4642		if (bma->idx != NULLEXTNUM && bma->idx) {
4643			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4644					 &bma->prev);
4645		}
4646	} else {
4647		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4648		if (!bma->eof)
4649			bma->length = XFS_FILBLKS_MIN(bma->length,
4650					bma->got.br_startoff - bma->offset);
4651	}
4652
4653	/*
4654	 * Indicate if this is the first user data in the file, or just any
4655	 * user data.
4656	 */
4657	if (!(flags & XFS_BMAPI_METADATA)) {
4658		bma->userdata = (bma->offset == 0) ?
4659			XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4660	}
4661
4662	bma->minlen = (flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4663
4664	/*
4665	 * Only want to do the alignment at the eof if it is userdata and
4666	 * allocation length is larger than a stripe unit.
4667	 */
4668	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4669	    !(flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4670		error = xfs_bmap_isaeof(bma, whichfork);
4671		if (error)
4672			return error;
4673	}
4674
4675	error = xfs_bmap_alloc(bma);
4676	if (error)
4677		return error;
4678
4679	if (bma->flist->xbf_low)
4680		bma->minleft = 0;
4681	if (bma->cur)
4682		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4683	if (bma->blkno == NULLFSBLOCK)
4684		return 0;
4685	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4686		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4687		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4688		bma->cur->bc_private.b.flist = bma->flist;
4689	}
4690	/*
4691	 * Bump the number of extents we've allocated
4692	 * in this call.
4693	 */
4694	bma->nallocs++;
4695
4696	if (bma->cur)
4697		bma->cur->bc_private.b.flags =
4698			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4699
4700	bma->got.br_startoff = bma->offset;
4701	bma->got.br_startblock = bma->blkno;
4702	bma->got.br_blockcount = bma->length;
4703	bma->got.br_state = XFS_EXT_NORM;
4704
4705	/*
4706	 * A wasdelay extent has been initialized, so shouldn't be flagged
4707	 * as unwritten.
4708	 */
4709	if (!bma->wasdel && (flags & XFS_BMAPI_PREALLOC) &&
4710	    xfs_sb_version_hasextflgbit(&mp->m_sb))
4711		bma->got.br_state = XFS_EXT_UNWRITTEN;
4712
4713	if (bma->wasdel)
4714		error = xfs_bmap_add_extent_delay_real(bma);
4715	else
4716		error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4717
4718	bma->logflags |= tmp_logflags;
4719	if (error)
4720		return error;
4721
4722	/*
4723	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4724	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4725	 * the neighbouring ones.
4726	 */
4727	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4728
4729	ASSERT(bma->got.br_startoff <= bma->offset);
4730	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4731	       bma->offset + bma->length);
4732	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4733	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4734	return 0;
4735}
4736
4737STATIC int
4738xfs_bmapi_convert_unwritten(
4739	struct xfs_bmalloca	*bma,
4740	struct xfs_bmbt_irec	*mval,
4741	xfs_filblks_t		len,
4742	int			flags)
4743{
4744	int			whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4745						XFS_ATTR_FORK : XFS_DATA_FORK;
4746	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4747	int			tmp_logflags = 0;
4748	int			error;
4749
4750	/* check if we need to do unwritten->real conversion */
4751	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4752	    (flags & XFS_BMAPI_PREALLOC))
4753		return 0;
4754
4755	/* check if we need to do real->unwritten conversion */
4756	if (mval->br_state == XFS_EXT_NORM &&
4757	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4758			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4759		return 0;
4760
4761	/*
4762	 * Modify (by adding) the state flag, if writing.
4763	 */
4764	ASSERT(mval->br_blockcount <= len);
4765	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4766		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4767					bma->ip, whichfork);
4768		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4769		bma->cur->bc_private.b.flist = bma->flist;
4770	}
4771	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4772				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4773
4774	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4775			&bma->cur, mval, bma->firstblock, bma->flist,
4776			&tmp_logflags);
4777	bma->logflags |= tmp_logflags;
4778	if (error)
4779		return error;
4780
4781	/*
4782	 * Update our extent pointer, given that
4783	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4784	 * of the neighbouring ones.
4785	 */
4786	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4787
4788	/*
4789	 * We may have combined previously unwritten space with written space,
4790	 * so generate another request.
4791	 */
4792	if (mval->br_blockcount < len)
4793		return EAGAIN;
4794	return 0;
4795}
4796
4797/*
4798 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4799 * extent state if necessary.  Details behaviour is controlled by the flags
4800 * parameter.  Only allocates blocks from a single allocation group, to avoid
4801 * locking problems.
4802 *
4803 * The returned value in "firstblock" from the first call in a transaction
4804 * must be remembered and presented to subsequent calls in "firstblock".
4805 * An upper bound for the number of blocks to be allocated is supplied to
4806 * the first call in "total"; if no allocation group has that many free
4807 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4808 */
4809int
4810xfs_bmapi_write(
4811	struct xfs_trans	*tp,		/* transaction pointer */
4812	struct xfs_inode	*ip,		/* incore inode */
4813	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4814	xfs_filblks_t		len,		/* length to map in file */
4815	int			flags,		/* XFS_BMAPI_... */
4816	xfs_fsblock_t		*firstblock,	/* first allocated block
4817						   controls a.g. for allocs */
4818	xfs_extlen_t		total,		/* total blocks needed */
4819	struct xfs_bmbt_irec	*mval,		/* output: map values */
4820	int			*nmap,		/* i/o: mval size/count */
4821	struct xfs_bmap_free	*flist)		/* i/o: list extents to free */
4822{
4823	struct xfs_mount	*mp = ip->i_mount;
4824	struct xfs_ifork	*ifp;
4825	struct xfs_bmalloca	bma = { 0 };	/* args for xfs_bmap_alloc */
4826	xfs_fileoff_t		end;		/* end of mapped file region */
4827	int			eof;		/* after the end of extents */
4828	int			error;		/* error return */
4829	int			n;		/* current extent index */
4830	xfs_fileoff_t		obno;		/* old block number (offset) */
4831	int			whichfork;	/* data or attr fork */
4832	char			inhole;		/* current location is hole in file */
4833	char			wasdelay;	/* old extent was delayed */
4834
4835#ifdef DEBUG
4836	xfs_fileoff_t		orig_bno;	/* original block number value */
4837	int			orig_flags;	/* original flags arg value */
4838	xfs_filblks_t		orig_len;	/* original value of len arg */
4839	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4840	int			orig_nmap;	/* original value of *nmap */
4841
4842	orig_bno = bno;
4843	orig_len = len;
4844	orig_flags = flags;
4845	orig_mval = mval;
4846	orig_nmap = *nmap;
4847#endif
 
 
4848
4849	ASSERT(*nmap >= 1);
4850	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4851	ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4852	ASSERT(tp != NULL);
4853	ASSERT(len > 0);
4854
4855	whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4856		XFS_ATTR_FORK : XFS_DATA_FORK;
4857
4858	if (unlikely(XFS_TEST_ERROR(
4859	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4860	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4861	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4862	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4863		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4864		return XFS_ERROR(EFSCORRUPTED);
4865	}
4866
4867	if (XFS_FORCED_SHUTDOWN(mp))
4868		return XFS_ERROR(EIO);
4869
4870	ifp = XFS_IFORK_PTR(ip, whichfork);
4871
4872	XFS_STATS_INC(xs_blk_mapw);
4873
4874	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4875		error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4876						  &bma.logflags, whichfork);
4877		if (error)
4878			goto error0;
4879	}
4880
4881	if (*firstblock == NULLFSBLOCK) {
4882		if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4883			bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4884		else
4885			bma.minleft = 1;
4886	} else {
4887		bma.minleft = 0;
4888	}
4889
4890	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4891		error = xfs_iread_extents(tp, ip, whichfork);
4892		if (error)
4893			goto error0;
4894	}
4895
4896	xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4897				&bma.prev);
4898	n = 0;
4899	end = bno + len;
4900	obno = bno;
4901
4902	bma.tp = tp;
4903	bma.ip = ip;
4904	bma.total = total;
4905	bma.userdata = 0;
4906	bma.flist = flist;
4907	bma.firstblock = firstblock;
4908
 
 
 
4909	while (bno < end && n < *nmap) {
4910		inhole = eof || bma.got.br_startoff > bno;
4911		wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4912
4913		/*
4914		 * First, deal with the hole before the allocated space
4915		 * that we found, if any.
4916		 */
4917		if (inhole || wasdelay) {
4918			bma.eof = eof;
4919			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4920			bma.wasdel = wasdelay;
4921			bma.offset = bno;
 
4922
4923			/*
4924			 * There's a 32/64 bit type mismatch between the
4925			 * allocation length request (which can be 64 bits in
4926			 * length) and the bma length request, which is
4927			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4928			 * check for 32-bit overflows and handle them here.
4929			 */
4930			if (len > (xfs_filblks_t)MAXEXTLEN)
4931				bma.length = MAXEXTLEN;
4932			else
4933				bma.length = len;
4934
4935			ASSERT(len > 0);
4936			ASSERT(bma.length > 0);
4937			error = xfs_bmapi_allocate(&bma, flags);
4938			if (error)
4939				goto error0;
4940			if (bma.blkno == NULLFSBLOCK)
4941				break;
4942		}
4943
4944		/* Deal with the allocated space we found.  */
4945		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4946							end, n, flags);
4947
4948		/* Execute unwritten extent conversion if necessary */
4949		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4950		if (error == EAGAIN)
4951			continue;
4952		if (error)
4953			goto error0;
4954
4955		/* update the extent map to return */
4956		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4957
4958		/*
4959		 * If we're done, stop now.  Stop when we've allocated
4960		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4961		 * the transaction may get too big.
4962		 */
4963		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4964			break;
4965
4966		/* Else go on to the next record. */
4967		bma.prev = bma.got;
4968		if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4969			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4970					 &bma.got);
4971		} else
4972			eof = 1;
4973	}
4974	*nmap = n;
4975
4976	/*
4977	 * Transform from btree to extents, give it cur.
4978	 */
4979	if (xfs_bmap_wants_extents(ip, whichfork)) {
4980		int		tmp_logflags = 0;
4981
4982		ASSERT(bma.cur);
4983		error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4984			&tmp_logflags, whichfork);
4985		bma.logflags |= tmp_logflags;
4986		if (error)
4987			goto error0;
4988	}
4989
4990	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4991	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4992		XFS_IFORK_MAXEXT(ip, whichfork));
4993	error = 0;
4994error0:
4995	/*
4996	 * Log everything.  Do this after conversion, there's no point in
4997	 * logging the extent records if we've converted to btree format.
4998	 */
4999	if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5000	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5001		bma.logflags &= ~xfs_ilog_fext(whichfork);
5002	else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5003		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5004		bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5005	/*
5006	 * Log whatever the flags say, even if error.  Otherwise we might miss
5007	 * detecting a case where the data is changed, there's an error,
5008	 * and it's not logged so we don't shutdown when we should.
5009	 */
5010	if (bma.logflags)
5011		xfs_trans_log_inode(tp, ip, bma.logflags);
5012
5013	if (bma.cur) {
5014		if (!error) {
5015			ASSERT(*firstblock == NULLFSBLOCK ||
5016			       XFS_FSB_TO_AGNO(mp, *firstblock) ==
5017			       XFS_FSB_TO_AGNO(mp,
5018				       bma.cur->bc_private.b.firstblock) ||
5019			       (flist->xbf_low &&
5020				XFS_FSB_TO_AGNO(mp, *firstblock) <
5021				XFS_FSB_TO_AGNO(mp,
5022					bma.cur->bc_private.b.firstblock)));
5023			*firstblock = bma.cur->bc_private.b.firstblock;
5024		}
5025		xfs_btree_del_cursor(bma.cur,
5026			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5027	}
5028	if (!error)
5029		xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5030			orig_nmap, *nmap);
5031	return error;
5032}
5033
5034/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5035 * Unmap (remove) blocks from a file.
5036 * If nexts is nonzero then the number of extents to remove is limited to
5037 * that value.  If not all extents in the block range can be removed then
5038 * *done is set.
5039 */
5040int						/* error */
5041xfs_bunmapi(
5042	xfs_trans_t		*tp,		/* transaction pointer */
5043	struct xfs_inode	*ip,		/* incore inode */
5044	xfs_fileoff_t		bno,		/* starting offset to unmap */
5045	xfs_filblks_t		len,		/* length to unmap in file */
5046	int			flags,		/* misc flags */
5047	xfs_extnum_t		nexts,		/* number of extents max */
5048	xfs_fsblock_t		*firstblock,	/* first allocated block
5049						   controls a.g. for allocs */
5050	xfs_bmap_free_t		*flist,		/* i/o: list extents to free */
5051	int			*done)		/* set if not done yet */
5052{
5053	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
5054	xfs_bmbt_irec_t		del;		/* extent being deleted */
5055	int			eof;		/* is deleting at eof */
5056	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
5057	int			error;		/* error return value */
5058	xfs_extnum_t		extno;		/* extent number in list */
5059	xfs_bmbt_irec_t		got;		/* current extent record */
5060	xfs_ifork_t		*ifp;		/* inode fork pointer */
5061	int			isrt;		/* freeing in rt area */
5062	xfs_extnum_t		lastx;		/* last extent index used */
5063	int			logflags;	/* transaction logging flags */
5064	xfs_extlen_t		mod;		/* rt extent offset */
5065	xfs_mount_t		*mp;		/* mount structure */
5066	xfs_extnum_t		nextents;	/* number of file extents */
5067	xfs_bmbt_irec_t		prev;		/* previous extent record */
5068	xfs_fileoff_t		start;		/* first file offset deleted */
5069	int			tmp_logflags;	/* partial logging flags */
5070	int			wasdel;		/* was a delayed alloc extent */
5071	int			whichfork;	/* data or attribute fork */
5072	xfs_fsblock_t		sum;
5073
5074	trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5075
5076	whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5077		XFS_ATTR_FORK : XFS_DATA_FORK;
5078	ifp = XFS_IFORK_PTR(ip, whichfork);
5079	if (unlikely(
5080	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5081	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5082		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5083				 ip->i_mount);
5084		return XFS_ERROR(EFSCORRUPTED);
5085	}
5086	mp = ip->i_mount;
5087	if (XFS_FORCED_SHUTDOWN(mp))
5088		return XFS_ERROR(EIO);
5089
 
5090	ASSERT(len > 0);
5091	ASSERT(nexts >= 0);
5092
5093	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5094	    (error = xfs_iread_extents(tp, ip, whichfork)))
5095		return error;
5096	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5097	if (nextents == 0) {
5098		*done = 1;
5099		return 0;
5100	}
5101	XFS_STATS_INC(xs_blk_unmap);
5102	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5103	start = bno;
5104	bno = start + len - 1;
5105	ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5106		&prev);
5107
5108	/*
5109	 * Check to see if the given block number is past the end of the
5110	 * file, back up to the last block if so...
5111	 */
5112	if (eof) {
5113		ep = xfs_iext_get_ext(ifp, --lastx);
5114		xfs_bmbt_get_all(ep, &got);
5115		bno = got.br_startoff + got.br_blockcount - 1;
5116	}
5117	logflags = 0;
5118	if (ifp->if_flags & XFS_IFBROOT) {
5119		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5120		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5121		cur->bc_private.b.firstblock = *firstblock;
5122		cur->bc_private.b.flist = flist;
5123		cur->bc_private.b.flags = 0;
5124	} else
5125		cur = NULL;
5126
5127	if (isrt) {
5128		/*
5129		 * Synchronize by locking the bitmap inode.
5130		 */
5131		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5132		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5133	}
5134
5135	extno = 0;
5136	while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5137	       (nexts == 0 || extno < nexts)) {
5138		/*
5139		 * Is the found extent after a hole in which bno lives?
5140		 * Just back up to the previous extent, if so.
5141		 */
5142		if (got.br_startoff > bno) {
5143			if (--lastx < 0)
5144				break;
5145			ep = xfs_iext_get_ext(ifp, lastx);
5146			xfs_bmbt_get_all(ep, &got);
5147		}
5148		/*
5149		 * Is the last block of this extent before the range
5150		 * we're supposed to delete?  If so, we're done.
5151		 */
5152		bno = XFS_FILEOFF_MIN(bno,
5153			got.br_startoff + got.br_blockcount - 1);
5154		if (bno < start)
5155			break;
5156		/*
5157		 * Then deal with the (possibly delayed) allocated space
5158		 * we found.
5159		 */
5160		ASSERT(ep != NULL);
5161		del = got;
5162		wasdel = isnullstartblock(del.br_startblock);
5163		if (got.br_startoff < start) {
5164			del.br_startoff = start;
5165			del.br_blockcount -= start - got.br_startoff;
5166			if (!wasdel)
5167				del.br_startblock += start - got.br_startoff;
5168		}
5169		if (del.br_startoff + del.br_blockcount > bno + 1)
5170			del.br_blockcount = bno + 1 - del.br_startoff;
5171		sum = del.br_startblock + del.br_blockcount;
5172		if (isrt &&
5173		    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5174			/*
5175			 * Realtime extent not lined up at the end.
5176			 * The extent could have been split into written
5177			 * and unwritten pieces, or we could just be
5178			 * unmapping part of it.  But we can't really
5179			 * get rid of part of a realtime extent.
5180			 */
5181			if (del.br_state == XFS_EXT_UNWRITTEN ||
5182			    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5183				/*
5184				 * This piece is unwritten, or we're not
5185				 * using unwritten extents.  Skip over it.
5186				 */
5187				ASSERT(bno >= mod);
5188				bno -= mod > del.br_blockcount ?
5189					del.br_blockcount : mod;
5190				if (bno < got.br_startoff) {
5191					if (--lastx >= 0)
5192						xfs_bmbt_get_all(xfs_iext_get_ext(
5193							ifp, lastx), &got);
5194				}
5195				continue;
5196			}
5197			/*
5198			 * It's written, turn it unwritten.
5199			 * This is better than zeroing it.
5200			 */
5201			ASSERT(del.br_state == XFS_EXT_NORM);
5202			ASSERT(xfs_trans_get_block_res(tp) > 0);
5203			/*
5204			 * If this spans a realtime extent boundary,
5205			 * chop it back to the start of the one we end at.
5206			 */
5207			if (del.br_blockcount > mod) {
5208				del.br_startoff += del.br_blockcount - mod;
5209				del.br_startblock += del.br_blockcount - mod;
5210				del.br_blockcount = mod;
5211			}
5212			del.br_state = XFS_EXT_UNWRITTEN;
5213			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5214					&lastx, &cur, &del, firstblock, flist,
5215					&logflags);
5216			if (error)
5217				goto error0;
5218			goto nodelete;
5219		}
5220		if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5221			/*
5222			 * Realtime extent is lined up at the end but not
5223			 * at the front.  We'll get rid of full extents if
5224			 * we can.
5225			 */
5226			mod = mp->m_sb.sb_rextsize - mod;
5227			if (del.br_blockcount > mod) {
5228				del.br_blockcount -= mod;
5229				del.br_startoff += mod;
5230				del.br_startblock += mod;
5231			} else if ((del.br_startoff == start &&
5232				    (del.br_state == XFS_EXT_UNWRITTEN ||
5233				     xfs_trans_get_block_res(tp) == 0)) ||
5234				   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5235				/*
5236				 * Can't make it unwritten.  There isn't
5237				 * a full extent here so just skip it.
5238				 */
5239				ASSERT(bno >= del.br_blockcount);
5240				bno -= del.br_blockcount;
5241				if (got.br_startoff > bno) {
5242					if (--lastx >= 0) {
5243						ep = xfs_iext_get_ext(ifp,
5244								      lastx);
5245						xfs_bmbt_get_all(ep, &got);
5246					}
5247				}
5248				continue;
5249			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5250				/*
5251				 * This one is already unwritten.
5252				 * It must have a written left neighbor.
5253				 * Unwrite the killed part of that one and
5254				 * try again.
5255				 */
5256				ASSERT(lastx > 0);
5257				xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5258						lastx - 1), &prev);
5259				ASSERT(prev.br_state == XFS_EXT_NORM);
5260				ASSERT(!isnullstartblock(prev.br_startblock));
5261				ASSERT(del.br_startblock ==
5262				       prev.br_startblock + prev.br_blockcount);
5263				if (prev.br_startoff < start) {
5264					mod = start - prev.br_startoff;
5265					prev.br_blockcount -= mod;
5266					prev.br_startblock += mod;
5267					prev.br_startoff = start;
5268				}
5269				prev.br_state = XFS_EXT_UNWRITTEN;
5270				lastx--;
5271				error = xfs_bmap_add_extent_unwritten_real(tp,
5272						ip, &lastx, &cur, &prev,
5273						firstblock, flist, &logflags);
5274				if (error)
5275					goto error0;
5276				goto nodelete;
5277			} else {
5278				ASSERT(del.br_state == XFS_EXT_NORM);
5279				del.br_state = XFS_EXT_UNWRITTEN;
5280				error = xfs_bmap_add_extent_unwritten_real(tp,
5281						ip, &lastx, &cur, &del,
5282						firstblock, flist, &logflags);
5283				if (error)
5284					goto error0;
5285				goto nodelete;
5286			}
5287		}
5288		if (wasdel) {
5289			ASSERT(startblockval(del.br_startblock) > 0);
5290			/* Update realtime/data freespace, unreserve quota */
5291			if (isrt) {
5292				xfs_filblks_t rtexts;
5293
5294				rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5295				do_div(rtexts, mp->m_sb.sb_rextsize);
5296				xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5297						(int64_t)rtexts, 0);
5298				(void)xfs_trans_reserve_quota_nblks(NULL,
5299					ip, -((long)del.br_blockcount), 0,
5300					XFS_QMOPT_RES_RTBLKS);
5301			} else {
5302				xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5303						(int64_t)del.br_blockcount, 0);
5304				(void)xfs_trans_reserve_quota_nblks(NULL,
5305					ip, -((long)del.br_blockcount), 0,
5306					XFS_QMOPT_RES_REGBLKS);
5307			}
5308			ip->i_delayed_blks -= del.br_blockcount;
5309			if (cur)
5310				cur->bc_private.b.flags |=
5311					XFS_BTCUR_BPRV_WASDEL;
5312		} else if (cur)
5313			cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5314		/*
5315		 * If it's the case where the directory code is running
5316		 * with no block reservation, and the deleted block is in
5317		 * the middle of its extent, and the resulting insert
5318		 * of an extent would cause transformation to btree format,
5319		 * then reject it.  The calling code will then swap
5320		 * blocks around instead.
5321		 * We have to do this now, rather than waiting for the
5322		 * conversion to btree format, since the transaction
5323		 * will be dirty.
5324		 */
5325		if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5326		    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5327		    XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5328			XFS_IFORK_MAXEXT(ip, whichfork) &&
5329		    del.br_startoff > got.br_startoff &&
5330		    del.br_startoff + del.br_blockcount <
5331		    got.br_startoff + got.br_blockcount) {
5332			error = XFS_ERROR(ENOSPC);
5333			goto error0;
5334		}
5335		error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5336				&tmp_logflags, whichfork);
5337		logflags |= tmp_logflags;
5338		if (error)
5339			goto error0;
5340		bno = del.br_startoff - 1;
5341nodelete:
5342		/*
5343		 * If not done go on to the next (previous) record.
5344		 */
5345		if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5346			if (lastx >= 0) {
5347				ep = xfs_iext_get_ext(ifp, lastx);
5348				if (xfs_bmbt_get_startoff(ep) > bno) {
5349					if (--lastx >= 0)
5350						ep = xfs_iext_get_ext(ifp,
5351								      lastx);
5352				}
5353				xfs_bmbt_get_all(ep, &got);
5354			}
5355			extno++;
5356		}
5357	}
5358	*done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5359
5360	/*
5361	 * Convert to a btree if necessary.
5362	 */
5363	if (xfs_bmap_needs_btree(ip, whichfork)) {
5364		ASSERT(cur == NULL);
5365		error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5366			&cur, 0, &tmp_logflags, whichfork);
5367		logflags |= tmp_logflags;
5368		if (error)
5369			goto error0;
5370	}
5371	/*
5372	 * transform from btree to extents, give it cur
5373	 */
5374	else if (xfs_bmap_wants_extents(ip, whichfork)) {
5375		ASSERT(cur != NULL);
5376		error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5377			whichfork);
5378		logflags |= tmp_logflags;
5379		if (error)
5380			goto error0;
5381	}
5382	/*
5383	 * transform from extents to local?
5384	 */
5385	error = 0;
5386error0:
5387	/*
5388	 * Log everything.  Do this after conversion, there's no point in
5389	 * logging the extent records if we've converted to btree format.
5390	 */
5391	if ((logflags & xfs_ilog_fext(whichfork)) &&
5392	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5393		logflags &= ~xfs_ilog_fext(whichfork);
5394	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5395		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5396		logflags &= ~xfs_ilog_fbroot(whichfork);
5397	/*
5398	 * Log inode even in the error case, if the transaction
5399	 * is dirty we'll need to shut down the filesystem.
5400	 */
5401	if (logflags)
5402		xfs_trans_log_inode(tp, ip, logflags);
5403	if (cur) {
5404		if (!error) {
5405			*firstblock = cur->bc_private.b.firstblock;
5406			cur->bc_private.b.allocated = 0;
5407		}
5408		xfs_btree_del_cursor(cur,
5409			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5410	}
5411	return error;
5412}
5413
5414/*
5415 * returns 1 for success, 0 if we failed to map the extent.
 
 
 
 
 
 
5416 */
5417STATIC int
5418xfs_getbmapx_fix_eof_hole(
5419	xfs_inode_t		*ip,		/* xfs incore inode pointer */
5420	struct getbmapx		*out,		/* output structure */
5421	int			prealloced,	/* this is a file with
5422						 * preallocated data space */
5423	__int64_t		end,		/* last block requested */
5424	xfs_fsblock_t		startblock)
5425{
5426	__int64_t		fixlen;
5427	xfs_mount_t		*mp;		/* file system mount point */
5428	xfs_ifork_t		*ifp;		/* inode fork pointer */
5429	xfs_extnum_t		lastx;		/* last extent pointer */
5430	xfs_fileoff_t		fileblock;
 
 
 
 
 
 
 
 
 
 
 
 
5431
5432	if (startblock == HOLESTARTBLOCK) {
5433		mp = ip->i_mount;
5434		out->bmv_block = -1;
5435		fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5436		fixlen -= out->bmv_offset;
5437		if (prealloced && out->bmv_offset + out->bmv_length == end) {
5438			/* Came to hole at EOF. Trim it. */
5439			if (fixlen <= 0)
5440				return 0;
5441			out->bmv_length = fixlen;
5442		}
5443	} else {
5444		if (startblock == DELAYSTARTBLOCK)
5445			out->bmv_block = -2;
5446		else
5447			out->bmv_block = xfs_fsb_to_db(ip, startblock);
5448		fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5449		ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5450		if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5451		   (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5452			out->bmv_oflags |= BMV_OF_LAST;
5453	}
5454
5455	return 1;
5456}
5457
5458/*
5459 * Get inode's extents as described in bmv, and format for output.
5460 * Calls formatter to fill the user's buffer until all extents
5461 * are mapped, until the passed-in bmv->bmv_count slots have
5462 * been filled, or until the formatter short-circuits the loop,
5463 * if it is tracking filled-in extents on its own.
5464 */
5465int						/* error code */
5466xfs_getbmap(
5467	xfs_inode_t		*ip,
5468	struct getbmapx		*bmv,		/* user bmap structure */
5469	xfs_bmap_format_t	formatter,	/* format to user */
5470	void			*arg)		/* formatter arg */
5471{
5472	__int64_t		bmvend;		/* last block requested */
5473	int			error = 0;	/* return value */
5474	__int64_t		fixlen;		/* length for -1 case */
5475	int			i;		/* extent number */
5476	int			lock;		/* lock state */
5477	xfs_bmbt_irec_t		*map;		/* buffer for user's data */
5478	xfs_mount_t		*mp;		/* file system mount point */
5479	int			nex;		/* # of user extents can do */
5480	int			nexleft;	/* # of user extents left */
5481	int			subnex;		/* # of bmapi's can do */
5482	int			nmap;		/* number of map entries */
5483	struct getbmapx		*out;		/* output structure */
5484	int			whichfork;	/* data or attr fork */
5485	int			prealloced;	/* this is a file with
5486						 * preallocated data space */
5487	int			iflags;		/* interface flags */
5488	int			bmapi_flags;	/* flags for xfs_bmapi */
5489	int			cur_ext = 0;
5490
5491	mp = ip->i_mount;
5492	iflags = bmv->bmv_iflags;
5493	whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5494
5495	if (whichfork == XFS_ATTR_FORK) {
5496		if (XFS_IFORK_Q(ip)) {
5497			if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5498			    ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5499			    ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5500				return XFS_ERROR(EINVAL);
5501		} else if (unlikely(
5502			   ip->i_d.di_aformat != 0 &&
5503			   ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5504			XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5505					 ip->i_mount);
5506			return XFS_ERROR(EFSCORRUPTED);
5507		}
5508
5509		prealloced = 0;
5510		fixlen = 1LL << 32;
5511	} else {
5512		if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5513		    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5514		    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5515			return XFS_ERROR(EINVAL);
5516
5517		if (xfs_get_extsz_hint(ip) ||
5518		    ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5519			prealloced = 1;
5520			fixlen = XFS_MAXIOFFSET(mp);
5521		} else {
5522			prealloced = 0;
5523			fixlen = XFS_ISIZE(ip);
5524		}
5525	}
5526
5527	if (bmv->bmv_length == -1) {
5528		fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5529		bmv->bmv_length =
5530			max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5531	} else if (bmv->bmv_length == 0) {
5532		bmv->bmv_entries = 0;
5533		return 0;
5534	} else if (bmv->bmv_length < 0) {
5535		return XFS_ERROR(EINVAL);
5536	}
5537
5538	nex = bmv->bmv_count - 1;
5539	if (nex <= 0)
5540		return XFS_ERROR(EINVAL);
5541	bmvend = bmv->bmv_offset + bmv->bmv_length;
5542
5543
5544	if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5545		return XFS_ERROR(ENOMEM);
5546	out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5547	if (!out) {
5548		out = kmem_zalloc_large(bmv->bmv_count *
5549					sizeof(struct getbmapx));
5550		if (!out)
5551			return XFS_ERROR(ENOMEM);
5552	}
5553
5554	xfs_ilock(ip, XFS_IOLOCK_SHARED);
5555	if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5556		if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5557			error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5558			if (error)
5559				goto out_unlock_iolock;
5560		}
5561		/*
5562		 * even after flushing the inode, there can still be delalloc
5563		 * blocks on the inode beyond EOF due to speculative
5564		 * preallocation. These are not removed until the release
5565		 * function is called or the inode is inactivated. Hence we
5566		 * cannot assert here that ip->i_delayed_blks == 0.
5567		 */
5568	}
5569
5570	lock = xfs_ilock_map_shared(ip);
5571
5572	/*
5573	 * Don't let nex be bigger than the number of extents
5574	 * we can have assuming alternating holes and real extents.
5575	 */
5576	if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5577		nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5578
5579	bmapi_flags = xfs_bmapi_aflag(whichfork);
5580	if (!(iflags & BMV_IF_PREALLOC))
5581		bmapi_flags |= XFS_BMAPI_IGSTATE;
5582
5583	/*
5584	 * Allocate enough space to handle "subnex" maps at a time.
 
5585	 */
5586	error = ENOMEM;
5587	subnex = 16;
5588	map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5589	if (!map)
5590		goto out_unlock_ilock;
5591
5592	bmv->bmv_entries = 0;
5593
5594	if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5595	    (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5596		error = 0;
5597		goto out_free_map;
5598	}
5599
5600	nexleft = nex;
5601
5602	do {
5603		nmap = (nexleft > subnex) ? subnex : nexleft;
5604		error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5605				       XFS_BB_TO_FSB(mp, bmv->bmv_length),
5606				       map, &nmap, bmapi_flags);
5607		if (error)
5608			goto out_free_map;
5609		ASSERT(nmap <= subnex);
5610
5611		for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5612			out[cur_ext].bmv_oflags = 0;
5613			if (map[i].br_state == XFS_EXT_UNWRITTEN)
5614				out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5615			else if (map[i].br_startblock == DELAYSTARTBLOCK)
5616				out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5617			out[cur_ext].bmv_offset =
5618				XFS_FSB_TO_BB(mp, map[i].br_startoff);
5619			out[cur_ext].bmv_length =
5620				XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5621			out[cur_ext].bmv_unused1 = 0;
5622			out[cur_ext].bmv_unused2 = 0;
5623
5624			/*
5625			 * delayed allocation extents that start beyond EOF can
5626			 * occur due to speculative EOF allocation when the
5627			 * delalloc extent is larger than the largest freespace
5628			 * extent at conversion time. These extents cannot be
5629			 * converted by data writeback, so can exist here even
5630			 * if we are not supposed to be finding delalloc
5631			 * extents.
5632			 */
5633			if (map[i].br_startblock == DELAYSTARTBLOCK &&
5634			    map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
5635				ASSERT((iflags & BMV_IF_DELALLOC) != 0);
5636
5637                        if (map[i].br_startblock == HOLESTARTBLOCK &&
5638			    whichfork == XFS_ATTR_FORK) {
5639				/* came to the end of attribute fork */
5640				out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5641				goto out_free_map;
5642			}
5643
5644			if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5645					prealloced, bmvend,
5646					map[i].br_startblock))
5647				goto out_free_map;
5648
5649			bmv->bmv_offset =
5650				out[cur_ext].bmv_offset +
5651				out[cur_ext].bmv_length;
5652			bmv->bmv_length =
5653				max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5654
5655			/*
5656			 * In case we don't want to return the hole,
5657			 * don't increase cur_ext so that we can reuse
5658			 * it in the next loop.
5659			 */
5660			if ((iflags & BMV_IF_NO_HOLES) &&
5661			    map[i].br_startblock == HOLESTARTBLOCK) {
5662				memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5663				continue;
5664			}
5665
5666			nexleft--;
5667			bmv->bmv_entries++;
5668			cur_ext++;
5669		}
5670	} while (nmap && nexleft && bmv->bmv_length);
5671
5672 out_free_map:
5673	kmem_free(map);
5674 out_unlock_ilock:
5675	xfs_iunlock_map_shared(ip, lock);
5676 out_unlock_iolock:
5677	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5678
5679	for (i = 0; i < cur_ext; i++) {
5680		int full = 0;	/* user array is full */
5681
5682		/* format results & advance arg */
5683		error = formatter(&arg, &out[i], &full);
5684		if (error || full)
5685			break;
5686	}
5687
5688	if (is_vmalloc_addr(out))
5689		kmem_free_large(out);
5690	else
5691		kmem_free(out);
5692	return error;
5693}
5694
5695#ifdef DEBUG
5696STATIC struct xfs_buf *
5697xfs_bmap_get_bp(
5698	struct xfs_btree_cur	*cur,
5699	xfs_fsblock_t		bno)
5700{
5701	struct xfs_log_item_desc *lidp;
5702	int			i;
5703
5704	if (!cur)
5705		return NULL;
5706
5707	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5708		if (!cur->bc_bufs[i])
5709			break;
5710		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5711			return cur->bc_bufs[i];
5712	}
5713
5714	/* Chase down all the log items to see if the bp is there */
5715	list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5716		struct xfs_buf_log_item	*bip;
5717		bip = (struct xfs_buf_log_item *)lidp->lid_item;
5718		if (bip->bli_item.li_type == XFS_LI_BUF &&
5719		    XFS_BUF_ADDR(bip->bli_buf) == bno)
5720			return bip->bli_buf;
5721	}
5722
5723	return NULL;
5724}
5725
5726STATIC void
5727xfs_check_block(
5728	struct xfs_btree_block	*block,
5729	xfs_mount_t		*mp,
5730	int			root,
5731	short			sz)
5732{
5733	int			i, j, dmxr;
5734	__be64			*pp, *thispa;	/* pointer to block address */
5735	xfs_bmbt_key_t		*prevp, *keyp;
5736
5737	ASSERT(be16_to_cpu(block->bb_level) > 0);
5738
5739	prevp = NULL;
5740	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5741		dmxr = mp->m_bmap_dmxr[0];
5742		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5743
5744		if (prevp) {
5745			ASSERT(be64_to_cpu(prevp->br_startoff) <
5746			       be64_to_cpu(keyp->br_startoff));
5747		}
5748		prevp = keyp;
5749
5750		/*
5751		 * Compare the block numbers to see if there are dups.
 
 
5752		 */
5753		if (root)
5754			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5755		else
5756			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5757
5758		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5759			if (root)
5760				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5761			else
5762				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5763			if (*thispa == *pp) {
5764				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5765					__func__, j, i,
5766					(unsigned long long)be64_to_cpu(*thispa));
5767				panic("%s: ptrs are equal in node\n",
5768					__func__);
5769			}
5770		}
5771	}
5772}
5773
5774/*
5775 * Check that the extents for the inode ip are in the right order in all
5776 * btree leaves.
5777 */
5778
5779STATIC void
5780xfs_bmap_check_leaf_extents(
5781	xfs_btree_cur_t		*cur,	/* btree cursor or null */
5782	xfs_inode_t		*ip,		/* incore inode pointer */
5783	int			whichfork)	/* data or attr fork */
5784{
5785	struct xfs_btree_block	*block;	/* current btree block */
5786	xfs_fsblock_t		bno;	/* block # of "block" */
5787	xfs_buf_t		*bp;	/* buffer for "block" */
5788	int			error;	/* error return value */
5789	xfs_extnum_t		i=0, j;	/* index into the extents list */
5790	xfs_ifork_t		*ifp;	/* fork structure */
5791	int			level;	/* btree level, for checking */
5792	xfs_mount_t		*mp;	/* file system mount structure */
5793	__be64			*pp;	/* pointer to block address */
5794	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
5795	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
5796	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
5797	int			bp_release = 0;
5798
5799	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5800		return;
5801	}
5802
5803	bno = NULLFSBLOCK;
5804	mp = ip->i_mount;
5805	ifp = XFS_IFORK_PTR(ip, whichfork);
5806	block = ifp->if_broot;
5807	/*
5808	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
 
 
 
5809	 */
5810	level = be16_to_cpu(block->bb_level);
5811	ASSERT(level > 0);
5812	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5813	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5814	bno = be64_to_cpu(*pp);
5815
5816	ASSERT(bno != NULLDFSBNO);
5817	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5818	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5819
5820	/*
5821	 * Go down the tree until leaf level is reached, following the first
5822	 * pointer (leftmost) at each level.
5823	 */
5824	while (level-- > 0) {
5825		/* See if buf is in cur first */
5826		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5827		if (bp) {
5828			bp_release = 0;
5829		} else {
5830			bp_release = 1;
5831		}
5832		if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5833				XFS_BMAP_BTREE_REF)))
5834			goto error_norelse;
5835		block = XFS_BUF_TO_BLOCK(bp);
5836		XFS_WANT_CORRUPTED_GOTO(
5837			xfs_bmap_sanity_check(mp, bp, level),
5838			error0);
5839		if (level == 0)
5840			break;
5841
5842		/*
5843		 * Check this block for basic sanity (increasing keys and
5844		 * no duplicate blocks).
5845		 */
 
 
 
5846
5847		xfs_check_block(block, mp, 0, 0);
5848		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5849		bno = be64_to_cpu(*pp);
5850		XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5851		if (bp_release) {
5852			bp_release = 0;
5853			xfs_trans_brelse(NULL, bp);
 
 
5854		}
5855	}
5856
5857	/*
5858	 * Here with bp and block set to the leftmost leaf node in the tree.
5859	 */
5860	i = 0;
5861
5862	/*
5863	 * Loop over all leaf nodes checking that all extents are in the right order.
5864	 */
5865	for (;;) {
5866		xfs_fsblock_t	nextbno;
5867		xfs_extnum_t	num_recs;
 
 
 
5868
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5869
5870		num_recs = xfs_btree_get_numrecs(block);
 
 
 
 
 
 
 
 
 
5871
5872		/*
5873		 * Read-ahead the next leaf block, if any.
5874		 */
5875
5876		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5877
5878		/*
5879		 * Check all the extents to make sure they are OK.
5880		 * If we had a previous block, the last entry should
5881		 * conform with the first entry in this one.
5882		 */
5883
5884		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5885		if (i) {
5886			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5887			       xfs_bmbt_disk_get_blockcount(&last) <=
5888			       xfs_bmbt_disk_get_startoff(ep));
5889		}
5890		for (j = 1; j < num_recs; j++) {
5891			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5892			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5893			       xfs_bmbt_disk_get_blockcount(ep) <=
5894			       xfs_bmbt_disk_get_startoff(nextp));
5895			ep = nextp;
5896		}
5897
5898		last = *ep;
5899		i += num_recs;
5900		if (bp_release) {
5901			bp_release = 0;
5902			xfs_trans_brelse(NULL, bp);
5903		}
5904		bno = nextbno;
5905		/*
5906		 * If we've reached the end, stop.
5907		 */
5908		if (bno == NULLFSBLOCK)
5909			break;
5910
5911		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5912		if (bp) {
5913			bp_release = 0;
5914		} else {
5915			bp_release = 1;
5916		}
5917		if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5918				XFS_BMAP_BTREE_REF)))
5919			goto error_norelse;
5920		block = XFS_BUF_TO_BLOCK(bp);
5921	}
5922	if (bp_release) {
5923		bp_release = 0;
5924		xfs_trans_brelse(NULL, bp);
5925	}
5926	return;
5927
5928error0:
5929	xfs_warn(mp, "%s: at error0", __func__);
5930	if (bp_release)
5931		xfs_trans_brelse(NULL, bp);
5932error_norelse:
5933	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5934		__func__, i);
5935	panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5936	return;
5937}
5938#endif
5939
5940/*
5941 * Count fsblocks of the given fork.
5942 */
5943int						/* error */
5944xfs_bmap_count_blocks(
5945	xfs_trans_t		*tp,		/* transaction pointer */
5946	xfs_inode_t		*ip,		/* incore inode */
5947	int			whichfork,	/* data or attr fork */
5948	int			*count)		/* out: count of blocks */
5949{
5950	struct xfs_btree_block	*block;	/* current btree block */
5951	xfs_fsblock_t		bno;	/* block # of "block" */
5952	xfs_ifork_t		*ifp;	/* fork structure */
5953	int			level;	/* btree level, for checking */
5954	xfs_mount_t		*mp;	/* file system mount structure */
5955	__be64			*pp;	/* pointer to block address */
5956
5957	bno = NULLFSBLOCK;
5958	mp = ip->i_mount;
5959	ifp = XFS_IFORK_PTR(ip, whichfork);
5960	if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5961		xfs_bmap_count_leaves(ifp, 0,
5962			ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5963			count);
5964		return 0;
5965	}
5966
5967	/*
5968	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5969	 */
5970	block = ifp->if_broot;
5971	level = be16_to_cpu(block->bb_level);
5972	ASSERT(level > 0);
5973	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5974	bno = be64_to_cpu(*pp);
5975	ASSERT(bno != NULLDFSBNO);
5976	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5977	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5978
5979	if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5980		XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5981				 mp);
5982		return XFS_ERROR(EFSCORRUPTED);
5983	}
5984
5985	return 0;
5986}
5987
5988/*
5989 * Recursively walks each level of a btree
5990 * to count total fsblocks is use.
5991 */
5992STATIC int                                     /* error */
5993xfs_bmap_count_tree(
5994	xfs_mount_t     *mp,            /* file system mount point */
5995	xfs_trans_t     *tp,            /* transaction pointer */
5996	xfs_ifork_t	*ifp,		/* inode fork pointer */
5997	xfs_fsblock_t   blockno,	/* file system block number */
5998	int             levelin,	/* level in btree */
5999	int		*count)		/* Count of blocks */
6000{
6001	int			error;
6002	xfs_buf_t		*bp, *nbp;
6003	int			level = levelin;
6004	__be64			*pp;
6005	xfs_fsblock_t           bno = blockno;
6006	xfs_fsblock_t		nextbno;
6007	struct xfs_btree_block	*block, *nextblock;
6008	int			numrecs;
6009
6010	if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6011		return error;
6012	*count += 1;
6013	block = XFS_BUF_TO_BLOCK(bp);
6014
6015	if (--level) {
6016		/* Not at node above leaves, count this level of nodes */
6017		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6018		while (nextbno != NULLFSBLOCK) {
6019			if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6020				0, &nbp, XFS_BMAP_BTREE_REF)))
6021				return error;
6022			*count += 1;
6023			nextblock = XFS_BUF_TO_BLOCK(nbp);
6024			nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6025			xfs_trans_brelse(tp, nbp);
6026		}
6027
6028		/* Dive to the next level */
6029		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6030		bno = be64_to_cpu(*pp);
6031		if (unlikely((error =
6032		     xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6033			xfs_trans_brelse(tp, bp);
6034			XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6035					 XFS_ERRLEVEL_LOW, mp);
6036			return XFS_ERROR(EFSCORRUPTED);
6037		}
6038		xfs_trans_brelse(tp, bp);
6039	} else {
6040		/* count all level 1 nodes and their leaves */
6041		for (;;) {
6042			nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6043			numrecs = be16_to_cpu(block->bb_numrecs);
6044			xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6045			xfs_trans_brelse(tp, bp);
6046			if (nextbno == NULLFSBLOCK)
6047				break;
6048			bno = nextbno;
6049			if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6050				XFS_BMAP_BTREE_REF)))
6051				return error;
6052			*count += 1;
6053			block = XFS_BUF_TO_BLOCK(bp);
6054		}
6055	}
6056	return 0;
6057}
6058
6059/*
6060 * Count leaf blocks given a range of extent records.
6061 */
6062STATIC void
6063xfs_bmap_count_leaves(
6064	xfs_ifork_t		*ifp,
6065	xfs_extnum_t		idx,
6066	int			numrecs,
6067	int			*count)
6068{
6069	int		b;
6070
6071	for (b = 0; b < numrecs; b++) {
6072		xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6073		*count += xfs_bmbt_get_blockcount(frp);
6074	}
6075}
6076
6077/*
6078 * Count leaf blocks given a range of extent records originally
6079 * in btree format.
6080 */
6081STATIC void
6082xfs_bmap_disk_count_leaves(
6083	struct xfs_mount	*mp,
6084	struct xfs_btree_block	*block,
6085	int			numrecs,
6086	int			*count)
6087{
6088	int		b;
6089	xfs_bmbt_rec_t	*frp;
6090
6091	for (b = 1; b <= numrecs; b++) {
6092		frp = XFS_BMBT_REC_ADDR(mp, block, b);
6093		*count += xfs_bmbt_disk_get_blockcount(frp);
6094	}
6095}
6096
6097/*
6098 * dead simple method of punching delalyed allocation blocks from a range in
6099 * the inode. Walks a block at a time so will be slow, but is only executed in
6100 * rare error cases so the overhead is not critical. This will alays punch out
6101 * both the start and end blocks, even if the ranges only partially overlap
6102 * them, so it is up to the caller to ensure that partial blocks are not
6103 * passed in.
6104 */
6105int
6106xfs_bmap_punch_delalloc_range(
6107	struct xfs_inode	*ip,
6108	xfs_fileoff_t		start_fsb,
6109	xfs_fileoff_t		length)
6110{
6111	xfs_fileoff_t		remaining = length;
6112	int			error = 0;
6113
6114	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6115
6116	do {
6117		int		done;
6118		xfs_bmbt_irec_t	imap;
6119		int		nimaps = 1;
6120		xfs_fsblock_t	firstblock;
6121		xfs_bmap_free_t flist;
6122
6123		/*
6124		 * Map the range first and check that it is a delalloc extent
6125		 * before trying to unmap the range. Otherwise we will be
6126		 * trying to remove a real extent (which requires a
6127		 * transaction) or a hole, which is probably a bad idea...
6128		 */
6129		error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6130				       XFS_BMAPI_ENTIRE);
6131
6132		if (error) {
6133			/* something screwed, just bail */
6134			if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6135				xfs_alert(ip->i_mount,
6136			"Failed delalloc mapping lookup ino %lld fsb %lld.",
6137						ip->i_ino, start_fsb);
6138			}
6139			break;
6140		}
6141		if (!nimaps) {
6142			/* nothing there */
6143			goto next_block;
6144		}
6145		if (imap.br_startblock != DELAYSTARTBLOCK) {
6146			/* been converted, ignore */
6147			goto next_block;
6148		}
6149		WARN_ON(imap.br_blockcount == 0);
6150
6151		/*
6152		 * Note: while we initialise the firstblock/flist pair, they
6153		 * should never be used because blocks should never be
6154		 * allocated or freed for a delalloc extent and hence we need
6155		 * don't cancel or finish them after the xfs_bunmapi() call.
6156		 */
6157		xfs_bmap_init(&flist, &firstblock);
6158		error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6159					&flist, &done);
6160		if (error)
6161			break;
6162
6163		ASSERT(!flist.xbf_count && !flist.xbf_first);
6164next_block:
6165		start_fsb++;
6166		remaining--;
6167	} while(remaining > 0);
6168
 
6169	return error;
6170}
6171
6172/*
6173 * Convert the given file system block to a disk block.  We have to treat it
6174 * differently based on whether the file is a real time file or not, because the
6175 * bmap code does.
6176 */
6177xfs_daddr_t
6178xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
6179{
6180	return (XFS_IS_REALTIME_INODE(ip) ? \
6181		 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
6182		 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
6183}